Datasets:

nickypro

/

minipile-split

like 0

--- author: - Hao Lin bibliography: - 'cite.bib' - 'growthrefs.bib' title: Properties of the limit shape for some last passage growth models in random environments --- abs ack

ArXiv

--- author: - | ZhongYing Fan\ Department of Physics, Beijing Normal University, 100875 Beijing, China\ title: Dynamic Mott gap from holographic fermions in geometries with hyperscaling violation --- Introduction ============ In recent years, AdS/CFT correspondence has been widely used to study condensed-matter theory (AdS/CMT). The strongly coupled conformal theory in the boundary is mapped to weakly coupled gravity theory in the bulk. With this great advantage, people have successfully constructed holographic models of Fermi and non-Fermi liquids in kinds of geometries[@1; @2; @3; @4; @5; @6; @7; @8; @9] and analytically investigated the liquids properties, showing the dispersion relation and the width of the quasiparticle-like excitation. Since condensed-matter systems are usually described by non-relativistic field theories, in order to search more proper gravity duals people have further generalized the correspondence to non-relativistic holography, which describes anisotropic scaling behaviors for temporal and spatial coordinates[@10; @11; @12; @13] i.e. Lifshitz-like geometry with dynamical exponent. For realistic systems, another important exponent i.e. hyperscaling violation will emerge and play a crucial role in low energy physics. It is certainly necessary to extend holography to this non-trivial case. This is realized by employing the standard Einstein-Maxwell-dilaton action in the bulk[@4; @14; @15; @16; @17]. The metric behaves like: $$ds^2=-\frac{dt^2}{r^{2m}}+r^{2n}dr^2+\frac{dx_i^2}{r^2} \label{1}\ ,$$ where $i=1,2,...,d$ is the space index, $m$ and $n$ are related to dynamical exponent $z$ and hyperscaling violation $\theta$ by $$z =\frac{m+n+1}{n+2}\ ,\quad \theta =\frac{n+1}{n+2}\cdot d\ . \label{2}$$ Note that when $n=-1$, the metric reduces to the pure Lifshitz spacetime and $n=-2$ corresponds to a class of spacetime conformally related to $AdS_2\times R_d$ with locally critical limit $z\rightarrow \infty,\ \theta\rightarrow -\infty, \mbox{while}\ z/\theta$ is fixed to be a constant[@18]. The metric transforms as $$t \rightarrow \lambda^z t,\ x_i\rightarrow \lambda x_i,\ r\rightarrow \lambda^{(d-\theta)/d} r,\ ds\rightarrow \lambda ^{\theta/d} ds\ .\label{3}$$ Clearly, the metric is not scale invariant. In general, the dual boundary theory exhibits this peculiar behavior below some non-trivial dimensional scale. However, we will not consider this complication. Instead we assume that the metric is asymptotical geometry with hyperscaling violation in this paper. We found remarkable influence of hyscaling violation on the dynamical gap by introducing a magnetic dipole coupling for bulk fermions. It was first proposed in [@19; @20] and further studied in [@21; @22; @23; @shangyu]. A gap in the spectral function was opened when the dipole coupling strength $p$ exceeds some critical value, which behaves like a Mott insulator. The gap becomes wider when $p$ increases. The coupling strength $p$ plays a role similar to the dimensionless interaction strength $U/t$ in the Hubbard model of fermions. The novel feature we find is that two bands exist in the spectral function, an upper band on the positive frequency axis and a lower band on the negative side, respectively. These two bands behave qualitatively different with the increasing of the interaction $p$ and hyperscaling violation $\theta$. The remainder of this paper is organized as follows: In section 2, we briefly review the effective gravity model, i.e, Einstein-Maxwell-dilaton theory for geometries with hyperscaling violation. In section 3, we study the bulk fermions with a dipole interaction, deriving the equations of motion for the retarded correlator. In section 4, we numerically solve the equations of motion under proper boundary conditions and extract the main results of the emergence of the gap. Finally, we present a conclusion in section 5. Effective Gravity Model ======================= The standard Einstein-Maxwell-dilaton (EMD) action reads $$S=\int \mathrm{d}^{d+2}x \sqrt{-g}\ [R-2(\partial{\phi})^2-V(\phi)-\frac{\kappa^2}{2}Z(\phi)F^2-\frac{\kappa^2}{2}H^2]\ , \label{action}$$ where the AdS radius has been set to 1. The solutions with hyperscaling violation are listed in the following: $$ds^2=-r^{-2m}h(r)dt^2+r^{2n}h^{-1}(r)dr^2+\frac{dx^2_i}{r^2}\ ,\qquad h(r)=1-(\frac{r}{r_h})^{\delta}\ , \label{ds2}$$ $$F^{rt}=F_0r^{(m-n+d)}Z^{-1}(\phi)\ ,\qquad H^{rt}=H_0r^{(m-n+d)}\ , \label{fh}$$ $$\phi=k_0\log{r}\ ,\qquad \quad k_0=\sqrt{\frac d2 (m-n-2)}\ , \label{phi}$$ $$V(\phi)=-V_0 e^{-\beta \phi}\ ,\qquad V_0=\delta(m+d-1)\ ,\qquad \beta=\frac{2(n+1)}{k_0}\ ,$$ $$Z^{-1}(\phi)=Z_0 e^{-\alpha \phi}+Z_1\ ,\ \alpha=\frac{2(n+d+1)}{k_0}\ ,\ Z_0=\frac{\delta(m-1)}{\kappa^2F_0^2}\ ,\ Z_1=-\frac{H_0^2}{F_0^2}\ .$$ where $\delta=m+n+d+1$, $r_h$ is the location of the horizon, $F_0$, $H_0$ are constants which are proportional to the conserved charges carried by the black brane. The Hawking temperature and the entropy density of the black brane are give by $$T=\frac{\delta}{4\pi } \frac{1}{r_h^{(m+n+1)}}\quad ,\quad s=\frac{1}{8\kappa^2 r_h^d}\ . \label{t}$$ In the zero temperature limit ($r_h\rightarrow \infty$), the entropy density approaches to zero. This is important for realistic systems with degenerate ground states. In order to admit a stable theory, the dilaton solution is required to be real, leading to $m \geq n+2$ or equivalently $z \geq 1+\theta/d,\ \theta < d$. Moreover, in the asymptotic limit $r\rightarrow 0$, the field strength $F^{\mu\nu}$ diverges such that the dual chemical potential cannot be well defined. Therefore we introduce another gauge field $H=dB$ to obtain a proper definition for the finite density $$B(r)=\mu (1-\frac{r^{(d-m+n+1)}}{r_h^{(d-m+n+1)}})\ dt\ , \label{b}$$ where $\mu$ is the chemical potential. The constraint condition which makes $B \ \mbox{and}\ H $ finite in the UV limit is $$2\leq m-n \leq d\ ,\qquad d\geq 3\ . \label{mn}$$ The divergent behavior of the field $F^{\mu\nu}$ certainly needs to be treated properly in a holographic renormalization procedure which we will not discuss in this paper. Since the bulk fermions we consider don’t couple to the dilaton and $F$ fields directly, the results we obtain are still credible, in the absence of a full treatment of the holographic EMD theory. Holographic fermion with magnetic dipole coupling ================================================= In order to explore the effects of magnetic dipole coupling on the spectral function of fermions, we start from the following action $$S_f[\Psi]=i \int \mathrm{d}^{d+2}x\sqrt{-g}\ \overline{\Psi}(\Gamma^a\mathcal{D}_a-M-i p \mathcal{H})\Psi+S_{bdy}[\Psi]\ , \label{faction}$$ $$S_{bdy}[\Psi]=i\int_\epsilon \mathrm{d}^{d+1}x\sqrt{-g_\epsilon}\sqrt{g^{rr}}\overline{\Psi}_+\Psi_-\ ,\label{bdy}$$ where $S_{bdy}$ is a boundary action to ensure a well defined variational principle[@24] for the total fermion action. $\overline{\Psi}=\Psi \Gamma^t$, $\mathcal{D}_a=(e_a)^\mu D_\mu$, with $D_\mu=\partial_\mu-i q B_\mu+\frac 14 \omega_{\mu ab}\Gamma^{ab}$, $\Gamma^{ab}=\frac 12 [\Gamma^a, \Gamma^b]$. $\omega_{\mu ab}$ is the spin connection 1-form and $\mathcal{H}=\frac 12 \Gamma^{ab} (e_a)^\mu(e_b)^\nu H_{\mu\nu}$. $\Gamma^a$ are the $d+2$ dimensional gamma matrices; $(e_a)^\mu$ are vielbeins and $M$ is the fermion mass. Furthermore, $g_\epsilon$ is the determinant of the induced metric on the constant $r$ slice, $r=\epsilon$. $\Psi_\pm$ is defined by $$\Psi_\pm=\frac 12 (1\pm \Gamma^r) \Psi\ ,\qquad \Gamma^r \Psi_\pm=\pm \Psi_\pm\ .$$ The Dirac equation derived from the action reads $$(\Gamma^a\mathcal{D}_a-M-i p \mathcal{H})\Psi=0 \label{dirac}\ .$$ Taking a Fourier transformation $$\Psi(r,x_\mu)=(-gg^{rr})^{-\frac 14}e^{-i\omega t+ik_ix^i} \psi(r,k_\mu)\ ,\quad k_\mu=(-\omega,\ \vec{k})\ ,$$ where the prefactor was introduced to remove the spin connection in the equations of motion. Since the theory is rotational invariant, we can choose the momentum along $x_1$ direction. The Gamma matrices are chosen as follows $$\Gamma^r= \left( \begin{array} {ccc} -\sigma^3 & 0\\ 0 & -\sigma^3 \end{array}\right) \ ,\quad \Gamma^t= \left( \begin{array}{ccc} i\sigma^1 & 0 \\ 0 & i\sigma^1 \end{array}\right) \ ,\quad \Gamma^{x_1}= \left( \begin{array}{ccc} -\sigma^2 & 0 \\ 0 & -\sigma^2 \end{array}\right)\ ,$$ where $\sigma$ are Pauli matrices. We further set $$\psi= \left(\begin{array}{ccc} \psi_+ \\ \psi_-\end{array}\right)\ ,\qquad \psi_\pm=\left(\begin{array}{ccc}u_\pm \\ d_\pm \end{array}\right)\ .$$ Since the Dirac equation is first order, there exists some relation between $\psi_+$ and $\psi_-$. Assuming $\psi_+(r,k_\mu)=- i \xi(r,k_\mu)\psi_-(r,k_\mu)$, we can derive an elegant equation to extract correlators $$\label{xi} \sqrt{g^{rr}}\partial_r \xi_\pm +2M \xi_\pm=(v_- \pm k\sqrt{g^{x_1x_1}})\xi^2_\pm+(v_+ \mp k\sqrt{g^{x_1x_1}})\ ,$$ where $\xi_+=iu_-/u_+,\ \xi_-=id_-/d_+$, $\xi_\pm$ are the eigenvalues of the matrices $\xi$. $v_\pm$ are defined as follows: $$v_{\pm}=\sqrt{-g^{tt}}(\omega+q B_t \pm p \sqrt{g^{rr}}\partial_r B_t)\ .$$ The corresponding retarded functions can be readily obtained as follows[@25] $$G_O(k_\mu)=\lim_{r\rightarrow 0} \xi(r,k_\mu)\ . \label{twopointfunction}$$ At the event horizon, we impose in-falling boundary conditions $$\xi(r_h,k_\mu)=i\ ,\qquad \mbox{for}\ \omega\neq 0\ . \label{eventcondition}$$ We emphasize that the dimension of the fermionic operator $O$ is $\Delta=(m+d)/2$ which leads to the fact that the unitarity bound was automatically satisfied with $m\geq 0$ given by the null energy conditions[@25]. The fermion mass decouples from the operator UV dimension, contributing only to the IR physics which is peculiar in the asymptotical geometries with hyperscaling violation. Numerical results and Emergence of the gap ========================================== To extract the effects of bulk dipole coupling on the spectral function, we need to numerically solve the flow equation (\[xi\]) with initial conditions (\[eventcondition\]). The spectral function is proportional to $ImG(\omega,k)$, up to normalization. Due to the relation $G_{11}(\omega,k)=G_{22}(\omega,-k)$, we will only consider $G_{22}(\omega,k)$ and omit the subscript in the following. For convenience, we set $M=0, \mu=1, q=2, z=2, d=3$. The dipole interaction strength $p$ and hyperscaling violation $\theta\ (\mbox{or}\ n)$ remain to be free. First, we fix the hyperscaling violation, considering $n=0$ case. From the plots above in figure \[f1\], we find that a sharp quasi-particle like peak occurs at $k_F\approx 1.2044$ for $p=0$, indicating the existence of a Fermi surface. The dual liquid is of Fermi type with linear dispersion relation at the maximum height of the spectral function. For larger charge $q$, it has been investigated with great detail in [@25] that more branches of Fermi surfaces appear. Moreover, when $q$ is sufficiently large, there exists a peculiar Fermi shell-like structure, which contains many sharp and singular peaks in some narrow interval of the momentum space. When the interaction strength was turned on at $p=4$, a gap emerges as is shown in the plots below of figure 1. There are two bands, located at positive frequency (we call it upper band)and negative frequency (called lower band) regions respectively. Evidently, the lower band is stronger than the upper one, occupying the main intensity of the spectral function. More interestingly, the upper band appears very sharp. In the big momentum region, the lower band is also as sharp as the upper one but disperses for relatively small momentum. From plots in figure \[f2\][^1], we can see that when $p$ increases further, the gap becomes larger. The upper band still keeps sharp for all momentum, translationally moving to the higher frequency region. However, the lower band is deformed much by transfer of the spectral weight to relatively higher momentum space. In order to show the emergence of the gap in detail, we present the plots of spectral function in figure \[f3\]. For very small $p$, the spectral function still has a sharp peak at $\omega=0$, showing the main feature of a Fermi surface. As $p$ increases, the intensity of the peak degrades and the spectral density begins to appear at the negative frequency axis. As $p$ increases further, the spectral density is transferred to the positive frequency region. Finally, at some critical interaction strength $p_{crit}$ the original sharp peak at $\omega=0$ disappears and two stable bands emerge in both frequency regions. To further explore the properties of the spectral function, we show the spectral function as a function of $\omega$ for $p=0,\ 4,\ 6,\ 8$ for sample values of momentum. From the left plot above in figure \[f7\], some peaks appear at both frequency regions. Around $\omega=0$, the peaks become sharper with its height tending to infinity, indicating that a Fermi surface exists at $k=k_F$. When $p$ is amplified, the quasiparticle-like peaks around $\omega=0$ degrade and vanish when $p$ exceeds some critical value $p_{crit}$. A gap will be opened for all momentum as the cases in literatures [@19; @20; @21; @22]. Evidently, the upper band appears sharper than the lower one. For the lower band, the height of the spectral function increases monotonically with the increasing of momentum. As $p$ increases further, the gap widens. Both of the bands appear robust. Notice that in the right plot below of figure \[f7\], the lower band disappears (the green and blue lines) when momentum exceeds some critical value, implying that a redistribution and deformation happens. All of these results are consistent with our 3D and density plots in figure \[f2\]. In order to determine the critical strength $p_{crit}$, we plot the density of states $A(\omega)$, the total spectral weight, which is defined by the integral of the spectral function $ImG(\omega,k)$ over $k$. We find that the onset of the gap is at $p_{crit}\approx 1.2$. Notice that for small $p$ ($p< p_{crit}$) the total spectral weight mainly distributes at the negative frequency region. As $p$ increases, it transfers to the positive region to open a gap. When the value of $p$ is large enough, the spectral weight will redistribute and backtrack to the negative region again. These results are compatible with our previous observations (figure 1, figure 2, figure 3 and figure 4). Finally, in figure \[f4\] we find that the width of the gap $\Delta$ increases with the increasing of the interaction strength $p$. We now vary the hyperscaling violation $\theta$ with fixed $p$. Without loss of generality, we set $p=6$. In figure \[f5\] and \[f6\], we show plots of the spectral function for $n=0.5$ and $n=1$. Clearly, the bands are highly suppressed as $n$ increases. The upper band disappears first. The lower band also becomes smooth gradually. As $n$ is amplified further, we may argue that in the $\theta\rightarrow d$ limit, any sharp peak of the spectral function will be completely smoothed out. The spectral density may transfer and redistribute to all frequency-momentum space homogeneously, with no explicit gap and band structure. This probably indicates some unknown critical phase. Conclusions =========== In this paper, we have studied the novel features of fermions in the presence of bulk dipole coupling in the geometries with hyperscaling violation. For a finite hyperscaling violation $\theta=d/2$, we observe that when the dipole interaction strength $p$=0, a sharp quasi-particle like peak occurs near $k_F\approx 1.2044$ at zero frequency, showing the existence of a Fermi surface. As $p$ increases, the intensity of the sharp peak degrades and the spectral weight begins to appear at the negative frequency region but is soon transferred to the positive frequency space. When $p$ crosses a critical value $p_{crit}$, the Fermi sea disappears. Instead, a stable gap and two bands emerge for all momentums. The upper band appears sharper than the lower one which however occupies the main intensity of the spectral function. When $p$ increases further, the gap becomes wider. The upper band keeps sharp. In contrast, the lower band is deformed much by redistributing the intensity to small momentum space. We also find that the width of the gap increases with the increasing of $p$. When we fix $p=6$ and turn on larger hyperscaling violation at $n=0.5, 1$, the peaks and bands are substantially suppressed. More interestingly, the upper band disappears first while the lower band becomes smooth gradually. Thus, the strength of the spectral density might distribute homogeneously in all frequency-momentum space in the $\theta\rightarrow d$ limit. It is of certain interests to explore this postulated critical phase in this limit. We will address it in the near future. Acknowledgments =============== I would like to thank Professor Sije Gao for his useful suggestions and encouragement. This work is supported by NSFC Grants NO.10975016, NO.11235003 and NCET-12-0054. [99]{} T.Faulkner, H.Liu, J.McGreevy, and D.Vegh, [*Emergent quantum criticality, Fermi surfaces and $\mbox{AdS}_2$*]{}, Phys. Rev. D [**83**]{},125002 (2011) \[arXiv:0907.2694v2 \[hep-th\]\]. N.Iqbal, H.Liu, and M.Mezei, [*Lectures on holographic non-Fermi liquids and quantum phase transitions*]{}, arXiv:1110.3814v1 \[hep-th\]. H.Liu, J.McGreevy, and D.Vegh, [*Non-Fermi liquids from holography*]{}, Phys. Rev. D [**83**]{},065029 (2011) \[arXiv:0903.2477 \[hep-th\]\]. N.Iizuka, N.Kundu, P.Narayan and S.P.Trivedi, [*Holographic Fermi and non-Fermi liquids with transitions in dilaton gravity*]{}, JHEP [**01**]{} (2012) 094 \[arXiv:1105.1162 \[hep-th\]\]. H.Lü, Zhao-Long Wang, [*Exact Green’s function and Fermi surfaces from conformal gravity*]{}, arXiv:1210.4560 \[hep-th\]. Jun Li, Hai-Shan Liu, H.Lü, and Zhao-Long Wang, [*Fermi surfaces and Analytic Green’s functions from conformal gravity*]{}, JHEP [**02**]{} (2013) 109 \[arXiv:1210.5000 \[hep-th\]\]. S.A.Hartnoll, D.M.Hofman, D.Vegh, [*Stellar spectroscopy: Fermions and holographic Lifshitz criticality*]{}, JHEP [**1108**]{}, 096 (2011) \[arXiv:1105.3197\[hep-th\]\]. S.A.Hartnoll and Alireza Tavanfar, [*Electron stars for holographic metallic criticality*]{}, Phys. Rev. D [**83**]{},046003 (2011) \[arXiv:1008.2828 \[hep-th\]\]. Steven S.Gubser, Jie Ren, [*Analytic Green’s functions from holography*]{}, Phys. Rev. D [**86**]{}, 046004 (2012) \[arXiv:1204.6315 \[hep-th\]\]. S.Kachru, X.Liu, M.Mulligan, [*Gravity duals of Lifshitz-like Fixed points*]{}, Phys. Rev. D [**78**]{}, 106005 (2008) \[arXiv:0808.1725 \[hep-th\]\]. M.H.Dehghani, R.B.Mann, R.Pourhasan, [*Charged Lifshitz Black Holes*]{}, Phys. Rev. D [**84**]{}, 046002 (2011) \[arXiv:1102.0578 \[hep-th\]\]. J.Tarrĺo, S.Vandoren, [*Black holes and black branes in Lifshitz spacetimes*]{}, JHEP [**09**]{} (2011) 017 \[arXiv:1105.6335 \[hep-th\]\]. M.Taylor, [*Non-relativistic holography*]{}, arXiv:0812.0530 \[hep-th\]. Christos Charmousis, Blaise Gouteraux, Bom Soo Kim, Elias Kiritsis, Rene Meyer, [*Effective Holographic Theories for low-temperature condensed matter systems*]{}, JHEP [**11**]{} (2010) 151 \[arXiv:1005.4690 \[hep-th\]\]. B.Goutéraux, E.Kiritsis, [*Generalized holographic quantum criticality at finite density* ]{}, JHEP [**12**]{} (2011) 036 \[arXiv:1107.2116 \[hep-th\]\]. Xi Dong, S.Harrison, S.Kachru, G.Torroba and H.Wang, [*Aspects of holography for theories with hyperscaling violation*]{}, JHEP [**06**]{} (2012) 041 \[arXiv:1201.1905v4 \[hep-th\]\]. M.Edalati, Juan F.Pedraza, W.T.Garcia, [**]{}, Phys. Rev. D [**87**]{}, 046001 (2013) \[arXiv:1210.6993 \[hep-th\]\]. S.A.Hartnoll and E.Shaghoulian, [*Spectral weight in holographic scaling geometries*]{}, JHEP [**07**]{} (2012) 078 \[arXiv:1203.4236\[hep-th\]\]. Mohammad Edalati, Robert G.Leigh, and Philip W.Phillips, [*Dynamically Generated Mott Gap from Holography*]{}, Phys. Rev. Lett. [**106**]{} (2011) 091602 \[arXiv:1010.3238\]. M.Edalati, R.G.Leigh, K.W.Lo and P.W.Phillips, [*Dynamical gap and cuprate-like physics from holography*]{}, Phys. Rev. D [**83**]{} (2011) 046012 \[arXiv:1012.3751\]. Jian-Pin Wu and Hua-Bi Zeng, [*Dynamic gap from holographic fermions in charged dilaton black branes*]{}, JHEP [**04**]{} (2012) 068 \[arXiv:1201.2485\]. Xiao-Mei Kuang, Bin Wang, and Jian-Pin Wu, [*Dynamic gap from holography in charged dilaton black hole*]{}, arXiv:1210.5735. Wei-Jia Li, Hongbao Zhang, [*Holographic non-relativistic fermionic fixed point and bulk dipole coupling*]{} JHEP 11 (2011) 018 \[arXiv:1110.4559 \[hep-th\]\]. Wen-Yu Wen, Shang-Yu Wu, [*Dipole coupling effect of holographic fermion in charged dilatonic gravity*]{}, Physic Letters B [**712**]{} (2012) 266-271 \[arXiv:1202.6539\]. M.Henneaux, [*Boundary terms in the AdS/CFT correspondence for spinor fields*]{}, arXiv:hep-th/9902137. ZhongYing Fan, [*Holographic fermions in asymptotically scaling geometries with hyperscaling violation*]{}, arXiv:1303.6053 \[hep-th\]. [^1]: The black part in the 3D plots are purely numerical noise.

ArXiv

--- author: - | Efrain J. Ferrer,\ Western Illinois University, USA\ E-mail: , - | Cristina Manuel\ Instituto de Fisica Corpuscular (CSIC-U. de Valencia), Spain\ E-mail: title: Dense Quark Matter in a Magnetic Field --- Introduction ============ It is well established that at high baryon density the combination of asymptotic freedom and the existence of attractive channels in the color interaction between the quarks lying in the large Fermi surface come together to promote the formation of quark-quark pairs, which in turn break the color gauge symmetry giving rise to the phenomenon of color superconductivity. At densities much higher than the masses of the u, d, and s quarks, one can assume the three quarks as massless and the favored state results to be the so-called Color-Flavor-Locking (CFL) phase [@alf-raj-wil-99/537], characterized by a spin zero diquark condensate antisymmetric in both color and flavor. The conditions of extremely large density and very low temperature required for color superconductivity cannot be recreated in Earth’s labs. Fortunately, nature provides us with a laboratory to probe color superconductivity, the cores of celestial compact objects. These compact stars typically have very large magnetic fields. Neutron stars can have magnetic fields as large as $B \sim 10^{12} - 10^{14}$ G in their surfaces, while in magnetars they are in the range $B \sim 10^{14} - 10^{15}$ G, and perhaps as high as $10^{16}$ G [@magnetars] (for a recent review of magnetic fields in dense stars see [@lugones/0504454]). Even though we do not know yet of any suitable mechanism to produce more intense fields, the virial theorem [@virial] allows the field magnitude to reach values as large as $10^{18}-10^{19}$ G. If quark stars are self-bound rather than gravitational-bound objects, the upper limit that has been obtained by comparing the magnetic and gravitational energies, could go even higher. A natural question to ask is: What is the effect, if any, of the huge star’s magnetic field in the color superconducting core? A complete answer to this question would require a rather involved study of quark matter at the intermediate range of densities proper of neutron stars, where the strange quark mass cannot be ignored, with the additional complication of an extra parameter, the magnetic field. However, as a first, more tractable approach to this question, one can ignore the strange quark mass effects and look for the consequences of an external magnetic field on the superconducting phase, assuming that the quark matter is formed by three massless flavors. This was the strategy followed in our recent paper [@MCFL], whose main results will be described in what follows. In this talk I will show the way a magnetic field affects the pairing structure and hence its symmetry, ultimately producing a different superconducting phase that we have called Magnetic Color-Flavor-Locking (MCFL) phase. In a conventional superconductor, since Cooper pairs are electrically charged, the electromagnetic gauge invariance is spontaneously broken, thus the photon acquires a Meissner mass that can screen a weak magnetic field, the phenomenon of Meissner effect. In spin-zero color superconductivity, although the color condensate has non-zero electric charge, there is a linear combination of the photon and a gluon that remains massless [@alf-raj-wil-99/537]. This new field plays the role of the “in-medium” photon in the color superconductor, so the propagation of light in the color superconductor is different from that in an electric superconductor. Because of the long-range “rotated” electromagnetic field, a spin-zero color superconductor may be penetrated by a rotated magnetic field $\widetilde{B}$. Although a few works [@oldCS-B] had previously addressed the problem of the interaction of an external magnetic field with dense quark matter, none of these studies considered the modification produced by the field on the gap itself. However, as we have recently shown [@MCFL], the gap structure gets modified due to the penetrating field. To understand this, notice that, although the condensate is $\widetilde{Q}$-neutral, some of the quarks participating in the pairing are $\widetilde{Q}$-charged and hence can couple to the background field, which in turn affects the gap equations through the Green functions of these $\widetilde{Q}$-charged quarks. Due to the coupling of the charged quarks with the external field, the color-flavor space is augmented by the $\widetilde{Q}$-charge color-flavor operator, and consequently the CFL order parameter splits in new independent pieces giving rise to a new phase, the MCFL phase. The MCFL Gap Structure {#MCFL Gap} ====================== The linear combination of the photon $A_{\mu}$ and a gluon $G^{8}_{\mu}$ that behaves as a long-range field in the spin-zero color superconductor is given by [@alf-raj-wil-99/537; @alf-berg-raj-NPB-02], $$\widetilde{A}_{\mu}=\cos{\theta}A_{\mu}-\sin{\theta}G^{8}_{\mu},\label{1}$$ while the orthogonal combination $\widetilde{G}_{\mu}^8=\sin{\theta}A_{\mu}+\cos{\theta}G^{8}_{\mu}$ is massive. In the CFL phase the mixing angle $\theta$ is sufficiently small ($\sin{\theta}\sim e/g\sim1/40$). Thus, the penetrating field in the color superconductor is mostly formed by the photon with only a small gluon admixture. The unbroken $U(1)$ group corresponding to the long-range rotated photon (i.e. the $\widetilde {U}(1)_{\rm e.m.}$) is generated, in flavor-color space, by $\widetilde {Q} = Q \times 1 - 1 \times Q$, where $Q$ is the electromagnetic charge generator. We use the conventions $Q = -\lambda_8/\sqrt{3}$, where $\lambda_8$ is the 8th Gell-Mann matrix. Thus our flavor-space ordering is $(s,d,u)$. In the 9-dimensional flavor-color representation that we will use in this paper (the color indexes we are using are (1,2,3)=(b,g,r)), the $\widetilde{Q}$ charges of the different quarks, in units of $\widetilde{e} = e \cos{\theta}$, are $$\label{q-charges} \begin{tabular}{|c|c|c|c|c|c|c|c|c|} \hline % after \\: \hline or \cline{col1-col2} \cline{col3-col4} ... $s_{1}$ & $s_{2}$ & $s_{3}$ & $d_{1}$ & $d_{2}$ & $d_{3}$ & $u_{1}$ & $u_{2}$ & $u_{3}$ \\ \hline 0 & 0 & - & 0 & 0 & - & + & + & 0 \\ \hline \end{tabular}$$ In the presence of an external rotated magnetic field the kinetic part of the quarks’ Lagrangian density must be rewritten using the covariant derivative $$L_{quarks}^{em}=\overline{\psi }(i\Pi_{\mu }\gamma ^{\mu })\psi \ , \label{quark-free-lag}$$ with $$\Pi _{\mu }=i\partial _{\mu }+\widetilde{e}\widetilde{Q}\widetilde{A}_{\mu} \ . \label{pi-operator}$$ where $$\widetilde{Q}=\Omega _{+}-\Omega _{-} \ . \label{Q-omeg-relat}$$ is the rotated charge operator. The charge projectors $$\Omega _{+}={\rm diag}(0,0,0,0,0,0,1,1,0) \ , \label{pos-omeg}$$ $$\Omega _{-}={\rm diag}(0,0,1,0,0,1,0,0,0) \ , \label{neg-omeg}$$ and $$\Omega _{0}={\rm diag}(1,1,0,1,1,0,0,0,1) \ , \label{neut-omeg}$$ obey the algebra $$\Omega _{\eta }\Omega _{\eta ^{\prime }}=\delta _{\eta \eta ^{\prime }}\Omega _{\eta },\qquad \eta ,\eta ^{\prime }=0,+,- \ . \label{omeg-algeb}$$ $$\Omega _{0}+\Omega _{+}+\Omega _{-}=1 \ . \label{omeg-sum}$$ The rotated magnetic field naturally separates the quark fields according to their $\tilde{Q}$ charge. The fermion field in the $9\times9$ representation used above, $\psi^{T}=(s_{1},s_{2},s_{3},d_{1},d_{2},d_{3},u_{1},u_{2},u_{3})$, can then be written as the sum of three fields with zero, positive and negative rotated electromagnetic charges, $$\psi =\psi _{(0)}+\psi _{(+)}+\psi _{(-)} \ , \label{ferm-sum}$$ where the $(0)$-, ($+/-)$-charged fields can be respectively written in terms of the charge projectors as $$\psi _{(0)}=\Omega _{0}\psi \ ,\qquad \psi _{(+)}=\Omega _{+}\psi \ ,\qquad \psi _{(-)}=\Omega _{-}\psi \ . \label{fields-def}$$ A strong magnetic field affects the flavor symmetries of QCD, as different quark flavors have different electromagnetic charges. For three light quark flavors, only the subgroup of $SU(3)_L \times SU(3)_R$ that commutes with $Q$, the electromagnetic generator, is a symmetry of the theory. Similarly, in the CFL phase a strong $\widetilde{B}$ field should affect the symmetries of the theory, as $\widetilde{Q}$ does not commute with the whole locked $SU(3)$ group. Based on the above considerations, and imposing that the condensate should retain the highest degree of symmetry, we proposed [@MCFL] the following ansatz for the gap structure in the presence of a magnetic field $$%\begin{eqnarray} %\label{gapMCFL} \Delta=\left( \begin{array}{ccccccccc} 2\Delta^{'}_{S} & 0 & 0 & 0 & \Delta_{A}+\Delta_{S} & 0 & 0 & 0 & \Delta^{B}_{A}+\Delta^{B}_{S}\\ 0 & 0 & 0 & \Delta_{S}-\Delta_{A} & 0 & 0 & 0 & 0& 0 \\ 0 & 0 & 0 & 0 & 0 & 0 & \Delta^{B}_{S}-\Delta^{B}_{A} & 0 & 0 \\ 0 & \Delta_{S}-\Delta_{A} & 0 & 0 & 0 & 0 & 0 & 0 & 0 \\ \Delta_{A}+\Delta_{S} & 0 & 0 & 0 & 2\Delta^{'}_{S} & 0 & 0 & 0 & \Delta^{B}_{A}+\Delta^{B}_{S}\\ 0 & 0 & 0 & 0 & 0 & 0& 0 & \Delta^{B}_{S}-\Delta^{B}_{A} & 0 \\ 0 & 0 & \Delta^{B}_{S}-\Delta^{B}_{A} & 0 & 0 & 0 & 0 & 0 & 0 \\ 0 & 0 & 0 & 0 & 0 & \Delta^{B}_{S}-\Delta^{B}_{A} & 0 & 0&0\\ \Delta^{B}_{A}+\Delta^{B}_{S} & 0 & 0 & 0 & \Delta^{B}_{A}+\Delta^{B}_{S} & 0 & 0 &0&2\Delta^{''}_{S} \label{order-parameter} \end{array} \right)$$ We call the reader’s attention to the fact that despite the $\widetilde{Q}$-neutrality of all the condensates, they can be composed either by neutral or by charged quarks. Condensates formed by $\widetilde{Q}$-charged quarks feel the field directly through the minimal coupling of the background field $\widetilde{B}$ with the quarks in the pair. A subset of the condensates formed by $\widetilde{Q}$-neutral quarks, can feel the presence of the field via tree-level vertices that couple them to charged quarks. The gaps $\Delta^B_{A/S}$ are antisymmetric/symmetric combinations of condensates composed by charged quarks and condensates formed by this kind of neutral quarks. The gaps $\Delta_{A}$, as well as $\Delta_{S}$, $\Delta^{'}_{S}$ and $\Delta^{''}_{S}$, on the other hand, are antisymmetric and symmetric combinations of condensates formed by neutral quarks that do not belong to the above subset. The only way the field can affect them is through the system of highly non-linear coupled gap equations. At zero field the CFL gap matrix is recovered since in that case $\Delta^{B}_{A}=\Delta_{A}$ and $\Delta^{B}_{S}=\Delta_{S}=\Delta^{'}_{S}=\Delta^{''}_{S}$. Although the symmetry of the problem allows for four independent symmetric gaps, the condensates $\Delta^{'}_{S}$ and $\Delta^{''}_{S}$ are only due to subleading color symmetric interactions, and as explained in the previous paragraph, they are formed by neutral quarks that are not coupled to charged quarks, so they belong to the same class as $\Delta_{S}$. Therefore, there is no reason to expect that they will differ much from $\Delta_{S}$. Hence, in a first approach to the problem, we will consider $\Delta_S\simeq \Delta^{'}_{S} \simeq \Delta^{''}_{S}$. The order parameter (\[order-parameter\]) implies the following symmetry breaking pattern: $SU(3)_{\rm color} \times SU(2)_L \times SU(2)_R \times U(1)_B \times U^{(-)}(1)_A \times U(1)_{\rm e.m.} \rightarrow SU(2)_{{\rm color}+L+R} \times {\widetilde {U}(1)}_{\rm e.m.}$. The $U^{(-)}(1)_A$ symmetry is connected with the current which is an anomaly-free linear combination of $s,d$ and $u$ axial currents [@miransky-shovkovy-02]. The locked $SU(2)$ corresponds to the maximal unbroken symmetry, and as such it maximizes the condensation energy. Notice that it commutes with the rotated electromagnetic group ${\widetilde {U}(1)}_{\rm e.m.}$. The counting of broken generators, after taking into account the Anderson-Higgs mechanism, tells us that there are only five Nambu-Goldstone bosons. One is associated to the breaking of the baryon symmetry; three Goldstone bosons are associated to the breaking of $SU(2)_A$, and another one associated to the breaking of $U^{(-)}(1)_A$. All the Nambu-Goldstone bosons are $\widetilde{Q}$-neutral. The number and properties of the lightest particles in the MCFL have implications for the low-energy physics of the phase. Since in her talk Cristina Manuel will address the low-energy physics of the MCFL phase, I will not extend on this topic in mine. Effective Action in the Presence of a Magnetic Field ==================================================== Let us construct the effective action of the system in the presence of a magnetic field. With this aim, we will use a Nambu-Jona-Lasinio (NJL) four-fermion interaction abstracted from one-gluon exchange [@alf-raj-wil-99/537]. Although this simplified treatment disregards the effect of the $\widetilde {B}$-field on the gluon dynamics and assumes the same NJL couplings for both the situation with and without magnetic field, it keeps the main attributes of the theory, thereby providing the correct qualitative physics. We start from the mean-field effective action $$\begin{aligned} I_{B}(\overline{\psi},\psi ) =\int\limits_{x,y}\{\frac{1}{2}[\overline{\psi } _{(0)}(x)[G_{(0)0}^{+}]^{-1}(x,y)\psi _{(0)}(y)+\overline{ \psi }_{(+)}(x)[G_{(+)0}^{+}]^{-1}(x,y)\psi _{(+)}(y) \nonumber \\ +\overline{\psi}_{(-)}(x)[G_{(-)0}^{+}]^{-1}(x,y)\psi _{(-)}(y)+ \overline{\psi } _{(0)C}(x)[G_{(0)0}^{-}]^{-1}(x,y)\psi _{(0)C}(y)\nonumber \\ +\overline{ \psi }_{(+)C}(x)[G_{(+)0}^{-}]^{-1}(x,y)\psi _{(+)C}(y) +\overline{\psi}_{(-)C}(x)[G_{(-)0}^{-}]^{-1}(x,y)\psi _{(-)C}(y)] \nonumber \\ +\frac{1}{2}[\overline{\psi }_{(0)C}(x)\Delta ^{+}(x,y)\psi _{(0)}(y)+h.c.] + \frac{1}{2}[\overline{\psi }_{(+)C}(x)\Delta ^{+}(x,y)\psi _{(-)}(y)\nonumber \\ +\overline{\psi }_{(-)C}(x)\Delta ^{+}(x,y)\psi _{(+)}(y)+h.c.]\} \ , \label{b-coord-action}\end{aligned}$$ where the external magnetic field has been explicitly introduced through minimal coupling with the $\widetilde{Q}-$charged fermions. The presence of the field is also taken into account in the diquark condensate $\Delta^{+}=\gamma_{5}\Delta$, whose color-flavor structure is given by Eq.(\[order-parameter\]). In (\[b-coord-action\]) symbols in parentheses indicate neutral $(0)$, positive $(+)$ or negative $(-)$ $\tilde{Q}-$charged quarks. Supra-indexes $+$ or $-$ in the propagators indicate, as it is customary, whether it is the inverse propagator of a field or conjugated field respectively. Then, for example, $[G_{(+)0}^{-}]^{-1}$ corresponds to the bare inverse propagator of positively charged conjugate fields, and so on. The explicit expressions of the inverse propagators are $$\lbrack G_{(0)0}^{\pm}]^{-1}(x,y)=[i\gamma ^{\mu }\partial _{\mu }-m\pm \mu \gamma ^{0}]\delta ^{4}(x-y) \ , \label{neut-x-inv-prop}$$ $$\lbrack G_{(+)0}^{\pm }]^{-1}(x,y)=[i\gamma ^{\mu }\Pi ^{(+)}_{\mu }-m\pm \mu \gamma ^{0}]\delta ^{4}(x-y) \ , \label{B-x-inv-prop+}$$ $$\lbrack G_{(-)0}^{\pm }]^{-1}(x,y)=[i\gamma ^{\mu }\Pi ^{(-)}_{\mu }-m\pm \mu \gamma ^{0}]\delta ^{4}(x-y) \ , \label{B-x-inv-prop-}$$ with $$\Pi ^{(\pm)}_{\mu }=i\partial _{\mu }\pm\widetilde{e}\widetilde{A}_{\mu } \ . \label{piplusminus}$$ Transforming the field-dependent quark propagators to momentum space can be performed with the use of the Ritus’ method, originally developed for charged fermions [@Ritus:1978cj] and recently extended to charged vector fields [@efi-ext]. In Ritus’ approach the diagonalization in momentum space of charged fermion Green’s functions in the presence of a background magnetic field is carried out using the eigenfunction matrices $E_p(x)$. These are the wave functions of the asymptotic states of charged fermions in a uniform magnetic field and play the role in the magnetized medium of the usual plane-wave (Fourier) functions $e^{i px}$ at zero field. Below we introduce the basic properties of this transformation. The transformation functions $E^{(\pm)}_{q}(x)$ for positively ($+$), and negatively ($-$) charged fermion fields are obtained as the solutions of the field dependent eigenvalue equation $$(\Pi^{(\pm)}\cdot\gamma)E^{(\pm)}_{q}(x)=E^{(\pm)}_{q}(x)(\gamma\cdot\overline{p}^{(\pm)}) \ , \label{eigenproblem}$$ with $\overline{p}^{(\pm )}$ given by $$\label{pbar+} \overline{p}^{(\pm )}=(p_{0},0,\pm \sqrt{2|\widetilde{e}\widetilde{B}|k},p_{3}) \ ,$$ and $$E^{(\pm)}_{q}(x)=\sum\limits_{\sigma }E^{(\pm)}_{q\sigma }(x)\delta (\sigma ) \ , \label{9}$$ with eigenfunctions $${E}^{(\pm)}_{p\sigma }(x)=\mathcal{N}% _{n_{(\pm)}}e^{-i(p_{0}x^{0}+p_{2}x^{2}+p_{3}x^{3})}D_{n_{(\pm)}}(\varrho _{(\pm)}) \ , \label{Epsigma+}$$ where $D_{n_{(\pm)}}(\varrho _{(\pm)})$ are the parabolic cylinder functions with argument $\varrho _{(\pm)}$ defined by $$\varrho _{(\pm)}=\sqrt{2|\widetilde{e}\widetilde{B}|}(x_{1}\pm p_{2}/\widetilde{e}% \widetilde{B}) \ , \label{rho+}$$ and index $n_{(\pm)}$ given by $$\label{normaliz-const+} n_{(\pm)}\equiv n_{(\pm)}(k,\sigma)= k \pm \frac{\widetilde{e}\widetilde{B}}{2|% \widetilde{e}\widetilde{B}|}\sigma-\frac{1}{2} \ , \qquad\qquad n_{(\pm)}=0,1,2,...$$ $k=0,1,2,3,...$ is the Landau level, and $\sigma$ is the spin projection that can take values $\pm 1$ only. Notice that in the lowest Landau level, $k=0$, only particles with one of the two spin projections, namely, $\sigma=1$ for positively charged particles, are allowed. The normalization constant $% \mathcal{N}_{n_{(\pm)}}$ is $$\label{normaliz-const} \mathcal{N}_{n_{(\pm)}}=(4\pi| \widetilde{e}\widetilde{B}|)^{\frac{1}{4}}/% \sqrt{n_{(\pm)}!} \ .$$ In (\[9\]) the spin matrices $\delta(\sigma )$ are defined as $$\delta(\sigma )= {\rm diag}(\delta _{\sigma 1},\delta _{\sigma -1},\delta _{\sigma 1},\delta _{\sigma -1}),\qquad \sigma =\pm 1 \ , \label{10}$$ and satisfy the following relations $$\delta \left( \pm \right) ^{\dagger }=\delta \left( \pm \right) \ ,\qquad \delta \left( \pm \right) \delta \left( \pm \right) =\delta \left( \pm \right) \ ,\qquad \delta \left( \pm \right) \delta \left( \mp \right) =0 \ ,$$ $$\gamma ^{\shortparallel }\delta \left( \pm \right) =\delta \left( \pm \right) \gamma ^{\shortparallel },\quad \gamma ^{\bot }\delta \left( \pm \right) =\delta \left( \mp \right) \gamma ^{\bot } \ . \label{24}$$ In Eq. (\[24\]) the notation $\gamma ^{\shortparallel }=(\gamma ^{0},\gamma ^{3})$ and $\gamma ^{\bot }=(\gamma ^{1},\gamma ^{2})$ was used. The functions $E^{(\pm)}_{p}$ are complete $$\label{complete-Ep} \sum_{k}\int dp_{0}dp_{2}dp_{3}{E}^{(\pm)}_{p}(x){\overline{E}}% ^{(\pm)}_{p}(y)=(2\pi)^{4}\delta^{(4)}(x-y) \ ,$$ and orthonormal, $$\label{ortho-Ep} \int_{x}{\overline{E}}^{(\pm)}_{p^{\prime}}(x){E}^{(\pm)}_{p}(x)=(2\pi)^{4}% \Lambda_{k}\delta_{kk^{\prime}}\delta(p_{0}-p^{\prime}_{0}) \delta(p_{2}-p^{\prime}_{2})\delta(p_{3}-p^{\prime}_{3})$$ with the $(4\times4)$ matrix $\Lambda_{k}$ given by $$\begin{aligned} \Lambda_{k}= \left\{ \begin{array}{cc} \delta(\sigma= {\rm sgn}[eB]) \qquad\qquad for \qquad k=0,\\ \qquad I \qquad\qquad\qquad\qquad for \qquad k>0 . \end{array} \right.\end{aligned}$$ The matrix structure $\Lambda_{k}$ was recently introduced in Ref. [@Leung05]. It had been previously omitted in the orthonormal condition of the $E_p(x)$ functions given in Refs. [@Ritus:1978cj; @efi-ext; @orthonormality]. Nevertheless, it should be underlined that this matrix only appears in the zero Landau level contribution, and consequently it enters as an irrelevant multiplicative factor in the Schwinger-Dyson equations in the lowest Landau level approximation. Thus, all the results obtained in the works [@Ritus:1978cj; @efi-ext; @orthonormality] remain valid. In Eqs. (\[complete-Ep\])-(\[ortho-Ep\]) we introduced the notation $\overline{E}_{p}^{(\pm)}(x)=\gamma _{0}({E}_{p}^{(\pm)}(x))^{\dag }\gamma _{0}$. Under the $E_p(x)$ functions, positively ($\psi _{(+)}$), negatively ($\psi _{(-)}$) charged fields transform according to $$\psi _{(\pm)}(x)=\sum_{k}\int dp_{0}dp_{2}dp_{3}E_{p}^{(\pm)}(x)\psi _{(\pm)}(p) \ ,\label{psi+-transf}$$ $$\overline{\psi }_{(\pm)}(x)=\sum_{k}\int dp_{0}dp_{2}dp_{3}\overline{\psi }% _{(\pm)}(p)\overline{E}_{p}^{(\pm)}(x) \ . \label{psibar+-transf}$$ One can show that $$\label{gamma-piplus-propert} [\gamma_{\mu}(\Pi_{(+)\mu}\pm \mu\delta_{\mu0})- m]{E}^{(+)}_{p}(x)={E}% ^{(+)}_{p}(x)[\gamma_{\mu}(\overline{p}^{(+)}_{\mu}\pm \mu\delta_{\mu0})- m] \ ,$$ and $$\label{gamma-piminus-propert} [\gamma_{\mu}(\Pi_{(-)\mu}\pm \mu\delta_{\mu0})- m]{E}^{(-)}_{p}(x)={E}% ^{(-)}_{p}(x)[\gamma_{\mu}(\overline{p}^{(-)}_{\mu}\pm \mu\delta_{\mu0})- m] \ .$$ The conjugate fields transform according to, $$\label{conjpsiplustransf} \psi_{(+)C}(x)=\sum_{k}\int dp_{0}dp_{2}dp_{3}E^{(-)}_{p}(x)\psi_{(+)C}(p),$$ $$\label{conjpsiminustransf} \psi_{(-)C}(x)=\sum_{k}\int dp_{0}dp_{2}dp_{3}E^{(+)}_{p}(x)\psi_{(-)C}(p) \ .$$ After transforming to momentum space one can introduce Nambu-Gorkov fermion fields of different $\tilde{Q}$ charges. They are the $\tilde{Q}$-neutral Gorkov field $$\Psi _{(0)}=\left( \begin{array}{c} \psi _{(0)} \\ \psi _{(0)C} \end{array} \right) \ ,$$ the positive $$\Psi _{(+)}=\left( \begin{array}{c} \psi _{(+)} \\ \psi _{(-)C} \end{array} \right) \ ,$$ and the negative one $$\Psi _{(-)}=\left( \begin{array}{c} \psi _{(-)} \\ \psi _{(+)C} \end{array} \right) \ .$$ Using them, the Nambu-Gorkov effective action in the presence of a constant magnetic field $\widetilde{B}$ can be written as $$\begin{aligned} \label{b-action} I^{B}(\overline{\psi},\psi) =\frac{1}{2}\int\frac{d^{4}p}{(2\pi)^{4}} \overline{\Psi}% _{(0)}(p){\cal S}^{-1}_{(0)}(p)\Psi_{(0)}(p)\nonumber\\ +\frac{1}{2}\int\frac{d^{4}p}{(2\pi)^{4}} \overline{\Psi}_{(+)}(p){\cal S}^{-1}_{(+)}(p)\Psi_{(+)}(p)+\frac{1}{2}\int\frac{d^{4}p}{% (2\pi)^{4}} \overline{\Psi}_{(-)}(p){\cal S}^{-1}_{(-)}(p)\Psi_{(-)}(p) \ ,\end{aligned}$$ where $$\begin{aligned} \label{p-neutr-inv-propg} {\cal S}^{-1}_{(0)}(p)=\left( \begin{array}{cc} [G_{(0)0}^{+}]^{-1}(p) & \Delta_{(0)}^{-} \\ & \\ \Delta_{(0)}^{+} & [G_{(0)0}^{-}]^{-1}(p) \end{array} \right) \ ,\end{aligned}$$ $$\begin{aligned} \label{p-posit-inv-propg} {\cal S}^{-1}_{(+)}(p)=\left( \begin{array}{cc} [G_{(+)0}^{+}]^{-1}(p) & \Delta_{(+)}^{-} \\ & \\ \Delta_{(+)}^{+} & [G_{(+)0}^{-}]^{-1}(p) \end{array} \right) \ ,\end{aligned}$$ $$\begin{aligned} \label{p-negat-inv-propg} {\cal S}^{-1}_{(-)}(p)=\left( \begin{array}{cc} [G_{(-)0}^{+}]^{-1}(p) & \Delta_{(-)}^{-} \\ & \\ \Delta_{(-)}^{+} & [G_{(-)0}^{-}]^{-1}(p) \end{array} \right) \ ,\end{aligned}$$ with $$\Delta_{(+)}^{+}= \Omega_{-}\Delta^{+}\Omega_{+},$$ $$\Delta_{(-)}^{+}= \Omega_{+}\Delta^{+}\Omega_{-},$$ $$\Delta_{(0)}^{+}= \Omega_{0}\Delta^{+}\Omega_{0},$$ Notice that to form the positive (negative) Nambu-Gorkov field we used the positive (negative) fermion field and the charge conjugate of the negative (positive) field. This is done so that the rotated charge of the up and down components in a given Nambu-Gorkov field be the same. This way to form the Nambu-Gorkov fields is mandated by what kind of field enters in a given condensate term, which in turn is related to the neutrality of the fermion condensate $\langle \overline{\psi}_{C}\psi\rangle$ with respect to the rotated $\tilde{Q}$-charge. In momentum space the bare inverse propagator for the neutral field is $$\lbrack G_{(0)0}^{\pm }]^{-1}(p)=[\gamma _{\mu }(p_{\mu }\pm \mu \delta _{\mu 0})-m] \ , \label{neut-bareprop+-}$$ where the momentum is the usual $p=(p_{0},p_{1},p_{2},p_{3})$ of the case with no background field. For positively and negatively charged fields the bare inverse propagators are $$\lbrack G_{(+)0}^{\pm }]^{-1}(p)=[\gamma _{\mu }(\overline{p}% _{\mu }^{(+)}\pm \mu \delta _{\mu 0})-m] \ , \label{pos-bareprop}$$ and $$\lbrack G_{(-)0}^{\pm }]^{-1}(p)=[\gamma _{\mu }(\overline{p}% _{\mu }^{(-)}\pm \mu \delta _{\mu 0})-m] \ \label{neg-bareprop}$$ respectively. Gap Solutions ============= The main question we would like to address now is: Can we find a region of magnetic fields where the gaps $\Delta_{A}$ and $\Delta^B_{A}$, (or $\Delta_{S}$ and $\Delta^B_{S}$), differ enough from each other that the system is not in the CFL phase anymore, but in the MCFL phase? To explore the possible answer to this question we need to solve the gap equations derived from the Nambu-Gorkov effective action (\[b-action\]). In coordinate space the QCD gap equation reads $$\label{gap-eq} \Delta^+(x,y) = i\frac{g^2}{4} \lambda_A^T\, \gamma^\mu\, S_{21}(x,y) \gamma^\nu\, \lambda_B D^{AB}_{\mu \nu} (x,y) \ ,$$ where $S_{21}(x,y)$ is the off-diagonal part of the Nambu-Gorkov fermion propagator in coordinate space and, for simplicity, we have omitted explicit color and flavor indices in the gap and fermion propagator. Here $D^{AB }_{\mu \nu}$ is the gluon propagator. In a NJL model the gap equation can be obtained from Eq. (\[gap-eq\]) simply by substituting the gluon propagator by $$D^{AB}_{\mu \nu} (x,y) = \frac{1}{\Lambda^2} \,g_{\mu \nu}\, \delta^{AB} \,\delta^{(4)}(x -y) \ .$$ The NJL model is characterized by a coupling constant $g$ and an ultraviolet cutoff $\Lambda$. The ultraviolet cutoff should be much larger than any of the energy scales of the system, typically the chemical potential. In the presence of a magnetic field we should also assume that $\Lambda$ is larger than the magnetic energy $\sqrt{{\tilde e}{\tilde B} }$. In other studies of color superconductivity within the NJL model, the values of $g$ and $\Lambda$ are chosen to match some QCD vacuum properties, thus hoping to get in such a way correct approximated quantitative results of the gaps. We will follow the same philosophy here, noticing however that this completely ignores the effect of the magnetic field on the gluon dynamics. To solve the gap equation (\[gap-eq\]) for the whole range of magnetic-field strengths we need to use numerical methods. We have found, however, a situation where an analytical solution is possible. This corresponds to the case $\widetilde{e}\widetilde{B} \gtrsim \mu^2/2$. Taking into account that the leading contribution to the gap solution comes from quark energies near the Fermi level, it follows that for fields in this range only the LLL ($l=0$) contributes. Using the approximation $\Delta^B_A \gg \Delta^B_S, \Delta_A$, and $\Delta_A \gg \Delta_S$, the gap equations decouple and the equation for $\Delta^B_A$ is $$\begin{aligned} \label{maingeq} \Delta^B_A & \approx & \frac{g^2}{3 \Lambda^2} \int_{\Lambda} \frac{d^3 q}{(2 \pi)^3} \frac{ \Delta^B_A}{\sqrt{(q-\mu)^2 + 2 (\Delta^B_A)^2 }} \nonumber \\ & + & \frac{g^2 \widetilde{e}\widetilde{B}}{3 \Lambda^2} \int_{-\Lambda}^{\Lambda} \frac{d q}{(2 \pi)^2} \frac{ \Delta^B_A}{\sqrt{(q-\mu)^2 + (\Delta^B_A)^2 }} ,\end{aligned}$$ where the first/second term in the r.h.s. of Eq.(\[maingeq\]) corresponds to the contribution of $\widetilde{Q}$-neutral/charged quark propagators, respectively. For the last one, we dropped all Landau levels but the lowest, as we are interested in the leading term. The solution of Eq. (\[maingeq\]) reads $$\label{gapBA} \Delta^B_A \sim 2 \sqrt{\delta \mu} \, \exp{\Big( - \frac{3 \Lambda^2 \pi^2} {g^2 \left(\mu^2 + \widetilde{e} \widetilde{B} \right)} \Big) } \ ,$$ with $\delta \equiv \Lambda - \mu$. It can be compared with the antisymmetric CFL gap [@review] $$\label{gapCFL} \Delta^{\rm CFL}_A \sim 2 \sqrt{\delta \mu} \, \exp{\Big( -\frac{3 \Lambda^2 \pi^2} {2 g^2 \mu^2} \Big) } \ .$$ In this approximation the remaining gap equations read $$\begin{aligned} \label{symgeq} \Delta^B_S & \approx & -\frac{g^2}{6 \Lambda^2} \int_{\Lambda} \frac{d^3 q}{(2 \pi)^3} \frac{ \Delta^B_A}{\sqrt{(q-\mu)^2 + 2 (\Delta^B_A)^2 }} \nonumber\\ & + & \frac{g^2 \widetilde{e}\widetilde{B}}{6 \Lambda^2} \int_{-\Lambda}^{\Lambda} \frac{d q}{(2 \pi)^2} \frac{ \Delta^B_A}{\sqrt{(q-\mu)^2 + (\Delta^B_A)^2 }} \ ,\end{aligned}$$ $$\begin{aligned} \label{antisymme-gapeqapp} \Delta_A & \approx & \frac {g^2}{4 \Lambda^2} \int_{\Lambda} \frac{d^3q}{(2 \pi)^3} \Big( \frac {17}{9} \frac{\Delta_A}{ \sqrt{ (q - \mu)^2 +\Delta_A^2 }} \nonumber \\ & + & \frac{7}{9} \frac{\Delta_A}{ \sqrt{(q-\mu)^2 + 2 (\Delta^B_A)^2 } } \Big) \ , \end{aligned}$$ and $$\begin{aligned} \label{symme-gapeqapp} \Delta_S & \approx & \frac {g^2}{18 \Lambda^2} \int_{\Lambda} \frac{d^3q}{(2 \pi)^3} \Big( \frac{\Delta_A}{ \sqrt{ (q - \mu)^2 +\Delta_A^2 }} \nonumber \\ & - & \frac{\Delta_A}{ \sqrt{(q-\mu)^2 + 2 (\Delta^B_A)^2 } } \Big) \ . \end{aligned}$$ We express below the solution of these gap equations as ratios over the CFL antisymmetric and symmetric gaps $$\frac{\Delta_A}{\Delta^{\rm CFL}_A} \sim \frac{1}{2^{(7/34)}} \exp{\Big(-\frac{36}{17 x} + \frac{21}{17}\frac{1}{x (1 + y)} + \frac{3}{2x} \Big) } \ ,$$ where $x \equiv g^2 \mu^2/\Lambda^2 \pi^2$, and $y \equiv \widetilde{e}\widetilde{B}/\mu^2$, and $$\begin{aligned} \frac{\Delta^B_S}{\Delta^{\rm CFL}_S} & \sim & \frac{\Delta^B_A}{\Delta^{\rm CFL}_A} \left( \frac{3}{4} + \frac{9}{2 x \ln{2}} \frac{y -1}{y+1} \right) \ ,\\ \frac{\Delta_S}{\Delta^{\rm CFL}_S} & \sim & \frac{\Delta_A}{\Delta^{\rm CFL}_A} \frac 32 \left(1 - \frac{4}{1+y} \right) \ .\end{aligned}$$ Note that our analytic solutions are only valid at strong magnetic fields. The lower value $\widetilde{e}\widetilde{B} \sim \mu^2/2$ corresponds to $\widetilde{e}\widetilde{B} \sim (0.8 - 1.1) \cdot 10^{18}$G, for $\mu \sim 350- 400$ MeV. For fields of this order and larger the $\Delta^B_A$ gap is larger than $\Delta^{\rm CFL}_A$ at the same density values. Nevertheless, we have estimated (see the details in Cristina Manuel’s talk in the proceedings) that the separation between CFL and MCFL will take place already at fields $\sim$ $10^{16}G$. All the gaps feel the presence of the external magnetic field. The effect of the magnetic field in $\Delta^{B}_{A}$ is to increase the density of states, which enters in the argument of the exponential as typical of a BCS solution. The density of states appearing in (\[gapBA\]) is just the sum of those of neutral and charged particles participating in the given gap equation (for each Landau level, the density of states around the Fermi surface for a charged quark is $\widetilde{e}\widetilde{B}/2 \pi^2$). All the $\widetilde{Q}$-charged quarks have common gap $\Delta^{B}_{A}$. Hence, the densities of the charged quarks are all equal. As two of these quarks have positive $\widetilde{Q}$ charge, while the other two have it negative, the $\widetilde{Q}$ neutrality of the medium is guaranteed without having to introduce any electron density. Conclusions =========== In this paper, we have shown that a magnetic field leads to the formation of a new color-flavor locking phase, characterized by a smaller vector symmetry than the CFL phase. The essential role of the penetrating magnetic field is to modify the density of states of charged quarks on the Fermi surface. To better understand the relevance of this new phase in astrophysics we need to explore the region of moderately strong magnetic fields $\widetilde{e}\widetilde{B}< \mu^2/2$, which requires to carry out a numerical study of the gap equations including the effect of higher Landau levels. Because the total density of states around the Fermi surface for charged particles does not vary monotonically with the number of Landau levels, we still expect to find a meaningful splitting of the gaps at these fields and therefore a qualitative separation between the CFL and MCFL phases. **Acknowledgments**\ The work of E.J.F. and V.I. was supported in part by NSF grant PHY-0070986, and C.M. was supported by MEC under grant FPA2004-00996. [10]{} M. Alford , K. Rajagopal and F. Wilczek, *“Color Flavor Locking and Chiral Symmetry Breaking ,”* *Nucl. Phys. B* **537**, 443 (1999). C. Thompson and R. C. Duncan, *“The Soft Gamma Repeaters as Very Strongly Magnetized Neutron Stars. 2. Quiescent Neutrino, X-ray, and Alfven Wave Emission,”* *Astrophys. J.* [**473**]{}, 322 (1996). German Lugones, *“Magnetic Fields in High-Density Stellar Matter, ”* \[[astro-ph/0504454]{}\]. L. Dong and S.L. Shapiro, *“Cold equation of state in a strong magnetic field - Effects of inverse beta-decay,”* *Astrophys. J.* [**383**]{}, 745 (1991). E. J. Ferrer ,V. de la Incera and C. Manuel, *"Magnetic Color Flavor Locking Phase in High Density QCD,”* *PRL* [**95**]{}, 152002 (2005). E. V. Gorbar,*“Color superconductivity in an external magnetic field,”* *Phys. Rev. D* [**62**]{}, 014007 (2000); K. Iida and G. Baym, *“Superfluid phases of quark matter. III. Supercurrents and vortices,”* *Phys. Rev. D* [**66**]{}, 014015 (2002); I. Giannakis and H-C Ren,*“The Ginzburg-Landau theory and the surface energy of a colour superconductor,”* *Nucl. Phys. B* [**669**]{}, 462 (2003). Alford M, Berges J, and Rajagopal K, *“Magnetic fields within color superconducting neutron star cores,”* *Nucl. Phys. B* **571**, 269 (2000). V. A. Miransky, and I. A. Shovkovy, *“Magnetic catalysis and anisotropic confinement in QCD,”* *Phys. Rev. D* [**66**]{}, 045006 (2002). V.I. Ritus,*“Radiative Corrections in Quantum Elctrodynamics with Intense Field and their Analytical Properties,”* *Ann.Phys.* [**69**]{}, 555 (1972). E. Elizalde, E. J. Ferrer, and V. de la Incera, *“Neutrino Self-Energy and Index of Refraction in Strong Magnetic Field: A New Approach,”* *Ann. of Phys.* [**295**]{}, 33 (2002); *“Neutrino Propagation in a Strongly Magnetized Medium,”* *Phys. Rev. D* [**70**]{}, 043012 (2004). C. N. Leung, and S.-Y. Wang, *“Gauge independent approach to chiral symmetry breaking in a strong magnetic field,”* hep-ph/0510066. D.-S Lee, C. N. Leung and Y. J. Ng, *“Chiral symmetry breaking in a uniform external magnetic field,”* *Phys. Rev. D* [**55**]{}, 6504 (1997); E. J. Ferrer, and V. de la Incera, *“Ward-Takahashi Identity with External Field in Ladder QED,”* *Phys. Rev. D* [**58**]{}, 065008 (1998); *"Magnetic Catalysis in the Presence of Scalar Fields* *Phys. Lett. B* [**481**]{}, 287 (2000); E. Elizalde, E. J. Ferrer, and V. de la Incera, *“Beyond-Constant-Mass-Approximation Magnetic Catalysis in the Gauge Higgs-Yukawa Model,”* *Phys. Rev. D* [**68**]{}, 096004 (2003). K. Rajagopal and F. Wilczek,*“The condensed matter physics of QCD,”* hep-ph/0011333.

ArXiv

--- abstract: '[The connection between molecular resonance phenomena in light heavy-ion collisions, alpha-clustering and extremely deformed states in light $\alpha$-like nuclei is discussed. For example, the superdeformed bands recently discovered in light N=Z nuclei such as $^{36}$Ar, $^{40}$Ca, $^{48}$Cr, and $^{56}$Ni by $\gamma$-ray spectroscopy may have a special link with resonant states in collisions with $\alpha$-like nuclei. The resonant reactions involving identical bosons such as $^{12}$C+$^{12}$C, $^{16}$O+$^{16}$O $^{24}$Mg+$^{24}$Mg and $^{28}$Si+$^{28}$Si are of interest. For instance, a butterfly mode of vibration of the J$^{\pi}$ = 38$^{+}$ resonance of $^{28}$Si+$^{28}$Si has been discovered in recent particle $\gamma$-ray angular correlations measurements. The search for signatures of strongly deformed shapes and clustering in light N=Z nuclei is also the domain of charged particle spectroscopy. The investigation of $\gamma$-decays in $^{24}$Mg has been undertaken for excitation energies where previously nuclear molecular resonances were found in $^{12}$C+$^{12}$C collisions. In this case the $^{12}$C-$^{12}$C scattering states can be related to the breakup resonance and, tentatively, to the resonant radiative capture $^{12}$C+$^{12}$C reaction. ]{}' address: 'Institut de Recherches Subatomiques, UMR7500, IN2P3-CNRS/Université Louis Pasteur, B.P. 28, F-67037 Strasbourg Cedex 2, France' author: - 'C. Beck' title: 'Molecular resonance phenomena and alpha-clustering: recent progress and perspectives' --- Introduction ============ The recent discovery of highly deformed shapes and superdeformed (SD) rotational bands in the N=Z nuclei $^{36}$Ar [@Svensson00], $^{40}$Ca [@Ideguchi01], $^{48}$Cr [@Lenzi96; @Thummerer01] and $^{56}$Ni [@Rudolph99] has renewed the interest in theoretical calculations for [*sd*]{}-shell nuclei around $^{40}$Ca [@Inakura02; @Sakuda02; @Kanada02]. Therefore the A$_{CN}$ $\approx$ 30-60 mass region becomes of particular interest [@Beck04] since quasimolecular resonances have also been observed for these $\alpha$-like nuclei, in particular, in the $^{28}$Si+$^{28}$Si reaction [@Betts81; @Beck01; @Nouicer99]. Although there is no experimental evidence to link the SD bands with the higher lying rotational bands formed by known quasimolecular resonances [@Bromley60], both phenomena are believed to originate from highly deformed configurations of these systems. The interpretation of resonant structures observed in the excitation functions in various combinations of light $\alpha$-cluster nuclei in the energy regime from the barrier up to regions with excitation energies of 30-50 MeV remains a subject of contemporary debate. In particular, in collisions between two $^{12}{\rm C}$ nuclei, these resonances have been interpreted in terms of nuclear molecules [@Bromley60]. However, in many cases these structures have been connected to strongly deformed shapes and to the alpha-clustering phenomena [@Marsh86; @Flocard84; @Leander75]. There has been a continuing discussion as to whether these molecular resonances represent true cluster states in the $^{24}$Mg compound system, or whether they simply reflect scattering states in the ion-ion potential. In this paper, a few examples will be given showing the close connection between molecular resonance phenomena, alpha-clustering, and nuclear deformation. Molecular resonances in $^{28}$Si+$^{28}$Si =========================================== [r]{}\[0cm\][8cm]{} {width="8cm" height="12cm"} The molecule-like sequences of resonances observed in $^{28}$Si + $^{28}$Si with measured angular momenta up to L = 42$\hbar$ represented some nuclear excitations with the highest spins ever observed [@Betts81]. In the number of open channels model [@Beck94; @Beck95], highly successful in selecting the systems showing resonance behavior, the main condition for observing a resonance behavior is associated with surface transparency. The $^{28}$Si + $^{28}$Si reaction is a particularly favorable case [@Beck95], where the corresponding optical model (OM) potentials have small imaginary components at distances corresponding to peripheral collisions. Therefore, the well established J$^{\pi}$ = 38$^{+}$ molecular resonance observed in $^{28}$Si + $^{28}$Si data at E$_{lab}$ = 112 MeV has been studied at the [Vivitron]{} Tandem facility of the IReS by both fragment-fragment and fragment-fragment-$\gamma$ coincidence measurements [@Beck01; @Nouicer99]. A subsequent experiment [@Chandana02] using charged particle spectroscopy techniques with the [Icare]{} multidetector array [@Rousseau02] has indicated the occurence of strongly deformed shapes at high spin (with axis ratios consistent with SD bands) for the $^{56}$Ni composite system at the resonant energy in agreement with very recent $\gamma$-ray spectroscopy data obtained at much lower spins [@Rudolph99]. From the analysis of the particle angular distributions of the mass-symmetric $^{28}$Si + $^{28}$Si exit-channel [@Nouicer99], it could be concluded that, at the resonance energy, the spin vectors of the $^{28}$Si fragments do not couple with the orbital angular momentum, leading to $m=0$. The fragment-fragment-$\gamma$ coincidence data [@Beck01] demonstrate that for the $^{28}$Si fragments, the mutually excited states are the most strongly populated. The resonance behavior appears to involve preferentially the low-lying states of the mass-symmetric channel. In Fig. 1 the results of the $\gamma$-ray angular correlations for the mutual excitation exit-channel (2$_{1}^{+}$,2$_{1}^{+}$) are shown. The distributions are presented in terms of the polar angles with respect to the three different quantization axes defined as: (a) the beam axis, (b) the axis normal to the scattering plane, and (c) the axis perpendicular to both axes defined in (a) and (b). The (c) axis corresponds approximately to the molecular axis of the outgoing binary fragments. the strong minimum in Fig. 1(b) at 90$^{\circ}$ implies that the magnetic substate m is equal to zero ($m=0$), and, thus, that the intrinsic spin vectors of the 2$^{+}$ states are oriented in the reaction plane perpendicularly to the orbital angular momentum. The value of the total angular momentum, therefore, remains close to ${\bf L} = 38\,\hbar$, in good agreement with the angular distributions results [@Nouicer99]. These observations do favor the calculations of the molecular model of Uegaki and Abe [@Uegaki94], which predict a vanishing spin alignment for $^{28}$Si + $^{28}$Si arising from a butterfly mode of vibration of the di-nuclear system at the resonance energy. Spin-alignment measurements of molecular states =============================================== The present $^{28}$Si + $^{28}$Si data show, for the first time in a heavy-ion collision, a vanishing spin alignment. Therefore, the comparison between the three symmetric systems $^{12}$C + $^{12}$C, $^{24}$Mg + $^{24}$Mg and $^{28}$Si + $^{28}$Si in Fig. 2 shows an interesting contrast in the spin orientation at resonance energies. [r]{}\[0cm\][8cm]{} {width="8cm" height="12cm"} The results indicate that the $^{28}$Si + $^{28}$Si oblate-oblate system, illustrated by Fig. 2(b), is characterised by spin disalignment in contrast to the spin alignment observed for both the $^{12}$C + $^{12}$C system [@Konnerth85; @Wuosmaa03] of Fig. 2(c) (oblate-oblate) and the $^{24}$Mg + $^{24}$Mg system [@Wuosmaa87] of Fig. 2(a) (prolate-prolate). Molecular-model calculations [@Uegaki94] are capable to explain the vanishing spin alignment in the oblate-oblate $^{28}$Si + $^{28}$Si system [@Nouicer99], where both nuclear spin vectors are perpendicular to the orbital angular momentum lying in the reaction plane. A new butterfly mode of vibration for the $^{28}$Si-$^{28}$Si triaxial molecule can then be speculated. However, the question of why the spin orientations in the two oblate-oblate systems $^{28}$Si + $^{28}$Si and $^{12}$C + $^{12}$C are so different is still unclear. Differences in the interactions between the constituent nuclei may play a key role. For example, there is a remarkable difference in the available molecular configurations in the excitation spectra of the $^{12}$C and $^{28}$Si nuclei. In $^{12}$C + $^{12}$C there are few molecular configurations, located at the energies associated with the observed resonances, while in $^{28}$Si + $^{28}$Si there are many more configurations. It is possible that the large number of available configurations [@Beck95] allows for the formation of coherent or collective states. This explains the sharp resonances which decay into many inelastic channels. In the $^{12}$C + $^{12}$C system, such coherent effects may not be allowed to develop. In this case the individual configurations may be observed. Overall, the results of these experiments provide further support for recent theoretical investigations that view the resonances in terms of shape-isomeric states stabilized in hyperdeformed secondary minima [@Marsh86; @Flocard84; @Leander75]. However, a more global understanding will surely require a more fundamental synthesis of the theories which describe reaction mechanisms with those describing nuclear structure at high excitation energy and angular momentum. $^{24}$Mg breakup states and the $^{12}$C+$^{12}$C molecule =========================================================== In the framework of the search for nuclear molecules the most spectacular results have often been obtained for the $^{12}$C+$^{12}$C reaction [@Beck94]. However, the question whether $^{12}$C+$^{12}$C molecular resonances represent true cluster states in the $^{24}$Mg compound system, or whether they simply reflect scattering states in the ion-ion potential is still unresolved [@Bromley60]. In many cases these structures have been connected to strongly deformed shapes and to the alpha-clustering phenomena, predicted from the $\alpha$-cluster model [@Marsh86], Hartree-Fock calculations [@Flocard84], the Nilsson-Strutinsky approach [@Leander75]. Various decay branches from the highly excited $^{24}$Mg$^*$ nucleus, including the emission of $\alpha$ particles or heavier fragments such as $^{8}$Be and $^{12}$C, are possibly available. However, $\gamma$-decays have not been observed so far. Actually the $\gamma$-ray branches are predicted to be rather small at these excitation energies, although some experiments have been reported [@McGrath81; @Metag82; @Haas97], which have searched for these very small branches expected in the range of 10$^{-4}$-10$^{-5}$ fractions of the total width [@Beck04; @Beck03]. The rotational bands built on the knowledge of the measured spins and excitation energies can be extended to rather small angular momenta, where finally the $\gamma$-decay becomes a larger part of the total width. The population of such states in $\alpha$-cluster nuclei, which are lying below the threshold for fission decays and for other particle decays, is favored in binary reactions, where at a fixed incident energy the composite nucleus is formed with an excitation energy range governed by the two-body reaction kinematics. These states may be coupled to intrinsic states of $^{24}$Mg$^{*}$ as populated by a breakup process (via resonances) as shown in previous works [@Fulton86; @Curtis95; @Singer00]. The $^{24}$Mg+$^{12}$C reaction has been extensively investigated by several measurements of the $^{12}$C($^{24}$Mg,$^{12}$C$^{12}$C)$^{12}$C breakup channel [@Fulton86; @Curtis95; @Singer00]. Sequential breakups are found to occur from specific states in $^{24}$Mg at excitation energies ranging from 20 to 35 MeV, which are linked to the ground state and also have an appreciable overlap with the $^{12}$C+$^{12}$C quasi-molecular configuration. Several attempts [@Curtis95] were made to link the $^{12}$C+$^{12}$C barrier resonances [@Bromley60] with the breakup states. The underlying reaction mechanism is now fairly well established [@Singer00] and many of the barrier resonances appear to be correlated indicating that a common structure may exist in both instances. This is another indication of the possible link between barrier resonances and secondary minima in the compound nucleus. The study of particle-$\gamma$ coincidences in binary reactions in reverse kinematics is probably a unique tool for the search for extreme shapes related to clustering. In this way the $^{24}$Mg+$^{12}$C reaction has been investigated with high selectivity at E$_{lab}$($^{24}$Mg) = 130 MeV with the Binary Reaction Spectrometer (BRS) in coincidence with [Euroball IV]{} installed at the [Vivitron]{} [@Beck04; @Beck03]. The choice of the $^{12}{\rm C}(^{24}{\rm Mg},^{12}{\rm C})^{24}{\rm Mg^{*}}$ reaction implies that for an incident energy of E$_{lab}$ = 130 MeV an excitation energy range up to E$^{*}$ = 30 MeV in $^{24}$Mg is covered [@Curtis95]. The BRS gives access to a novel approach to the study of nuclei at large deformations [@Beck04; @Beck03]. The excellent channel selection capability of binary and/or ternary fragments gives a powerful identification among the reaction channels, implying that [Euroball IV]{} is used mostly with one or two-fold multiplicities, for which the total $\gamma$-ray efficiency is very high. The BRS trigger consists of a kinematical coincidence set-up combining two large-area heavy-ion telescopes. Both detector telescopes comprise each a two-dimensional position sensitive low-pressure multiwire chamber in conjunction with a Bragg-curve ionization chamber. All detection planes are four-fold subdivided in order to improve the resolution and to increase the counting rate capability (100 k-events/s). The two-body Q-value has been reconstructed using events for which both fragments are in well selected states chosen for spectroscopy purposes as well as to determine the reaction mechanism responsible for the population of these peculiar states. The inverse kinematics of the $^{24}$Mg+$^{12}$C reaction and the negative Q-values give ideal conditions for the trigger on the BRS, because the chosen angular range is optimum and because the solid angle transformation gives a factor 10 for the detection of the heavy fragments. Thus we have been able to cover a large part of the angular distribution of the binary process with high efficiency, and a selection of events in particular angular ranges has been achieved. In binary exit-channels the exclusive detection of both ejectiles allows precise Q-value determination, Z-resolution and simultaneously optimal Doppler-shift correction.\ ![*Doppler corrected $\gamma$-ray spectrum for $^{24}$Mg, using particle-particle-$\gamma$ coincidences, measured in the $^{24}$Mg(130 MeV)+$^{12}$C reaction with the BRS/[Euroball IV]{} detection system (see text).* ](figcluster3.eps){width="68.00000%"} \[Fig.3\] Fig. 3 displays a Doppler-corrected $\gamma$-ray spectrum in coincidence with $^{24}$Mg events identified in the Bragg-Peak vs energy spectra of the BRS. All known transitions of $^{24}$Mg [@Beck01; @Wiedenhover01] can be identified in the energy range depicted. As expected we see decays feeding the yrast line of $^{24}$Mg up to the 8$^{+}_{2}$ level. The population of some of the observed states, in particular, the 2$^{+}$, 3$^{+}$ and 4$^{+}$ members of the K$^{\pi}$ = 2$^{+}$ rotational band, appears to be selectively enhanced. The strong population of the K$^{\pi}$ = 2$^{+}$ band and his 4$^{+}$ member at E$_x$ = 6.01 MeV has also been observed in the $^{12}$C($^{12}$C,$\gamma$) radiative capture reaction [@Jenkins03]. Furthermore, there is an indication of a $\gamma$-ray around 5.95 MeV which may be identified with the 10$^{+}_{1}$ $\rightarrow$ 8$^{+}_{2}$ transition as proposed in Ref. [@Wiedenhover01]. It has been checked in the $\gamma$-$\gamma$ coincidences that most of the states of Fig. 3 belong to cascades which contain the characteristic 1368 keV $\gamma$-ray and pass through the lowest 2$^{+}$ state in $^{24}$Mg. Still a number of transitions in the high-energy part of the spectrum (6-8 MeV) have not been clearly identified. Even at higher energies, $^{24}$Mg states appear to show up around 10 MeV (not shown) with very poor statistics and of unknown structure. Similar states were also observed in the radiative capture reaction [@Jenkins03]. Their occurence may be in qualitative agreement with a decay scenario of radiative capture states proposed by Baye and Descouvemont [@Baye84; @Descouvemont86] in the framework of a microscopic study of the $^{12}$C+$^{12}$C system with the Coordinate Generator Method. The reason why the search for a $\gamma$-decay in $^{12}$C+$^{12}$C has not been conclusive so far [@McGrath81; @Metag82; @Haas97] is due to the excitation energy in $^{24}$Mg as well as the spin region (8$\hbar$-12$\hbar$) which were chosen too high. The next step of the analysis will be the use of the BRS trigger in order to select the excitation energy range by the two-body Q-value (in the $^{12}$C+$^{24}$Mg channel), and thus we will be able to study the region around the decay barriers, where $\gamma$-decay becomes observable. According to recent predictions $\gamma$-rays from 6$^{+}$ $\rightarrow$ 4$^{+}$ should have measurable branching ratios. Work is currently in progress to analyse the $\gamma$ rays from the $^{12}$C($^{24}$Mg,$^{12}$C $^{12}$C)$^{12}$C ternary breakup reaction. Summary and conclusions ======================= We have discussed the possible link between resonant states in collisions with identical bosons such as $^{12}$C+$^{12}$C, $^{24}$Mg+$^{24}$Mg and $^{28}$Si+$^{28}$Si and the SD bands recently discovered in light N=Z nuclei such as $^{36}$Ar, $^{40}$Ca, $^{48}$Cr, and $^{56}$Ni. A new butterfly mode of vibration of the well established J$^{\pi}$ = 38$^{+}$ resonance of the $^{28}$Si-$^{28}$Si triaxial molecule has been discovered experimentally. The connection of alpha-clustering and quasimolecular resonances has been discussed with the search for the $^{12}$C+$^{12}$C molecule populated by the $^{24}$Mg+$^{12}$C breakup reaction. The most spectacular result is the strong population of the K$^{\pi}$ = 2$^{+}$ band of the $^{24}$Mg nucleus that has also been observed in an exploratory investigation of the $^{12}$C($^{12}$C,$\gamma$) radiative capture reaction [@Jenkins03]. Subsequent radiative capture experiments are planned in the near future with highly efficient spectrometers (the [Dragon]{} separator at [Triumf]{} and the [Fma]{} at Argonne) to investigate the overlap of $^{24}$Mg states observed in the present work with radiative capture states. As far as the $\gamma$-ray spectroscopy is concerned, the coexistence of $\alpha$-cluster states and SD states predicted in $^{32}$S by recent antisymmetrized molecular dynamics (AMD) calculations [@Horiuchi03] is still an experimental challenge. This kind of experiments designed to measure very small $\Gamma$$_\gamma$/$\Gamma$$_{total}$ branching ratios is extremely difficult since it requires not only high-efficient fragment detection in conjunction with a high-resolution Ge multidetector such as the [Gammasphere]{} and [Euroball]{} 4$\pi$ $\gamma$ arrays but also a large amount of beam time. [ I am pleased to acknowledge the physicists of both the [Icare]{} and the BRS/[Euroball IV]{} collaborations, with special thanks to M. Rousseau, P. Papka, A. Sànchez i Zafra, C. Bhattacharya, and S. Thummerer. We thank the staff of the [Vivitron]{} for providing us with good stable beams, M.A. Saettel for preparing targets, and J. Devin and C. Fuchs for their excellent support during the experiments. Parts of this work was supported by the french IN2P3/CNRS and the EC Euroviv contract HPRI-CT-1999-00078.]{} [9]{} C.E. Svensson [*et al.*]{}, Phys. Rev. Lett. [**85**]{}, 2693 (2000). E. Ideguchi [*et al.*]{}, Phys. Rev. Lett. [**87**]{}, 222501 (2001). S.M. Lenzi [*et al.*]{}, Z. Phys. [**A354**]{}, 117 (1996). S. Thummerer [*et al.*]{}, J. Phys. G: Nucl. Part. Phys. [**27**]{}, 1405 (2001). D. Rudolph [*et al.*]{}, Phys. Rev. Lett. [**82**]{}, 3763 (1999). T. Inakura [*et al.*]{}, Nucl. Phys. A [**710**]{}, 261 (2002). T. Sakuda and S. Ohkubo, Nucl. Phys. A [**712**]{}, 59 (2002). Y. Kanada-En’yo, K. Kimura, and H. Horiuchi, AIP Conf. Proc. [**644**]{}, 188 (2003). C. Beck, International Journal of Modern Physics **A, (2004) to be published** R. R. Betts [*et al.*]{}, Phys. Rev. Lett. **47, 23 (1981).** C. Beck [*et al.*]{}, Phys. Rev. C [**63**]{}, 014607 (2001). R. Nouicer [*et al.*]{}, Phys. Rev. C **60, 041303 (1999).** K.A. Erb and D.A. Bromley, [*Treatise on Heavy Ion Science*]{}, Vol. [**3**]{}, 201 (1985). S. Marsh and W.D. Rae, Phys. Lett. B [**180**]{}, 185 (1986). H. Flocard [*et al.*]{}, Prog. Theor. Phys. [**72**]{}, 1000 (1984). G. Leander and S.E. Larsson, Nucl. Phys. A [**239**]{}, 93 (1975). C. Beck [*et al.*]{}, Phys. Rev. C **49, 2618 (1994).** C. Beck [*et al.*]{}, Nucl. Phys. A **583, 269 (1995).** C. Bhattacharya [*et al.*]{}, Phys. Rev. C [**65**]{}, 014611 (2002). M. Rousseau [*et al.*]{}, Phys. Rev. C [**66**]{}, 034612 (2002). E. Uegaki and Y. Abe, Phys. Lett. **B340, 143 (1994).** D. Konnerth [*et al.*]{}, Phys. Rev. Lett. **55, 588 (1985).** A. H. Wuosmaa [*et al.*]{}, Phys. Lett. **B571, 155 (2003).** A. H. Wuosmaa [*et al.*]{}, Phys. Rev. Lett. **58, 1312 (1987).** R.L. McGrath [*et al.*]{}, Phys. Rev. C [**7**]{}, 1280 (1981). V. Metag [*et al.*]{}, Phys. Rev. C [**25**]{}, 1486 (1982). F. Haas [*et al.*]{}, Il Nuovo Cimento [**110A**]{}, 989 (1997). C. Beck [*et al.*]{}, Nucl. Phys. A (to be published) and arXiv:nucl-ex/0309007 (2003). B.R. Fulton [*et al.*]{}, Phys. Lett. B [**267**]{}, 325 (1991). N. Curtis [*et al.*]{}, Phys. Rev. C [**51**]{}, 1554 (1995). S.M. Singer [*et al.*]{}, Phys. Rev. C [**62**]{}, 054609 (2000). I. Wiedenhöver [*et al.*]{}, Phys. Rev. Lett. [**87**]{}, 142502 (2001). D.G. Jenkins [*et al.*]{}, AIP Conf. Proc. [**656**]{}, 329 (2003); and private communications. D. Baye and P. Descouvemont, Nucl. Phys. A **419, 397 (1984).** P. Descouvemont and D. Baye, Phys. Lett. **196B, 143 (1986).** H. Horiuchi, Y. Kanada-En’yo, and K. Kimura, Nucl. Phys. A [**722**]{}, 80c (2003).

ArXiv

--- abstract: 'In this work, we consider an initial problem for second order partial differential equations with Caputo fractional derivatives in the time-variable and Bessel operator in the space-variable. For non-local boundary conditions, we present a solution of this problem in an explicit form representing it by the Fourier-Bessel series. The obtained solution is written in terms of multinomial Mittag-Leffler functions and first kind Bessel functions.' address: - ' Erkinjon Karimov: Institute of Mathematics National University of Uzbekistan Tashkent, 100125 Uzbekistan ' - ' Murat Mamchuev: Department of Theoretical and Mathematical Physics Institute of Applied Mathematics and Automation Shortanova str. 89-A, Nalchik, 360000 Kabardino-Balkar Republic Russia ' - ' Michael Ruzhansky: Department of Mathematics Imperial College London 180 Queen’s Gate, London, SW7 2AZ United Kingdom ' author: - Erkinjon Karimov - Murat Mamchuev - Michael Ruzhansky title: 'Non-local initial problem for second order time-fractional and space-singular equation' --- [^1] Introduction and formulation of a problem ========================================= It is well-known that partial differential equations are playing a key role in constructing mathematical models for many real-life processes. Especially, during the last decades, many applications of various kinds of fractional differential equations became target of intensive research due to both theoretical and practical reasons, see e.g. [@KST06] for an overview. Many kinds of boundary problems, including direct [@nakh] and inverse problems [@isak], were formulated for different type of PDEs of integer order including several differential operators of fractional order. We note works [@gejbhal]-[@gorluch] devoted to studying partial differential equations with multiple Caputo derivatives. Precisely, in [@gorluch] the authors studied fractional differential equations with Caputo fractional derivatives and using the operational method, solutions of initial boundary problem for those equations were obtained in an explicit form involving a multinomial Mittag-Leffler function. Certain properties of this function were obtained by Li, Liu and Yamamoto [@yamli] and applied to studying initial-boundary problems for time-fractional diffusion equations with positive constant coefficients. Later, Liu [@liu] established strong maximum principle for fractional diffusion equations with multiple Caputo derivatives and investigated a related inverse problem. Daftardar-Gejji and Bhalikar [@gejbhal], using the method of separation of variables, solved some boundary-value problems for multi-term fractional diffusion-wave equation. We also note works related to the Bessel operator. In [@masmes], the initial inverse problem for the heat equation with Bessel operator was investigated. Inverse initial and inverse source problems for time-fractional diffusion equation with zero order Bessel operator were recently studied in [@fatma]. Direct and inverse problems for PDEs containing two-term time fractional Caputo derivatives of orders up to 1, and Bessel operator of order $\nu$ were investigated in [@akmr]. The consideration of non-local initial conditions is often justified by practical usage in certain real-life processes. For instance, when the initial temperature for the heat equation is not given instantly, but there is an information related with the temperature on a certain time interval that can be described by a non-local initial condition in a simple form. Boundary-value problems with non-local initial conditions were considered in works [@pao] for reaction-diffusion equations, in [@shop] for heat equation, in [@rk1]-[@rk2] for degenerate parabolic equations, and in [@ker] for a mixed parabolic equation. For integer orders much more is known, and for a review of different questions of time decay of solutions for hyperbolic equations with integer order derivatives we can refer to [@RS]. In the present work we deal with the non-local initial boundary problem for multi-term time fractional PDE with Bessel operator of order $\nu$. We use Fourier-Bessel series expansion in order to find the explicit solution for the considered problem, yielding also its existence. Because of the singularities in the Bessel operator such conditions appear naturally in space variables. We note that most of the current literature deals with diffusion type equations considering time-derivatives of orders up to 1, see e.g. [@yamli]. One of the novelties of the present paper is that we consider wave type equations allowing fractional derivatives of order up to 2, with additional fractional dissipation type terms. If there are multiple fractional time-derivative terms, a multinomial Mittag-Leffler function appears in the representation of solutions. Let us now describe the problem in more detail. We consider the equation $$\label{eq1} L(u)-B_\nu(u)=f(t,x)$$ in a rectangular domain $D=\left\{(x,t):\, 0<x<1,\,\,0<t<T\right\}$, $T>0$, where $f(t,x)$ is a given function, $$\label{eq2-3} L(u)=\partial_{0t}^{\alpha}u(t,x)-\sum\limits_{i=1}^n \lambda_i \partial_{0t}^{\alpha_i}u(t,x)$$ is the time component of the equation, with orders $$0<\alpha_i\leq1,\quad \alpha_i\leq \alpha\le 2,\quad n\in \mathbb{N}, \quad \lambda_i\in \mathbb{R},$$ and $$\label{eq2-3b} B_\nu(u)=u_{xx}(t,x)+\frac{1}{x}u_x(t,x)-\frac{\nu^2}{x^2}u(t,x)$$ is the Bessel part of the equation with $\nu>0$. Here $$\partial_{0t}^{\beta}g(t)=\left\{ \begin{aligned} & \frac{1}{\Gamma \left( k-\alpha \right)}\int\limits_{0}^{t}{\frac{{g}^{(k)}\left( z \right)}{{{\left( t-z \right)}^{\alpha-k+1 }}}dz,\,}\,\,\alpha\notin \mathbb{N}_0, \\ & \frac{d^kg(t)}{dt^k},\quad\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\alpha \in\mathbb{N}, \\ & g(t), \quad\quad\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\alpha=0, \end{aligned} \right.$$ is a fractional differential operator of Caputo type, where $k=[\alpha]+1$, and $[\alpha]$ is the integer part of $\alpha$. We can refer to [@KST06] for further details on the Caputo fractional derivative operators. The non-local initial boundary problem for equation - is formulated as follows: Let $M\in\mathbb R$. To find a solution $u(t,x)$ of equation - in $D$, which satisfies - regularity conditions $u\in W$ with $$\label{eq4} W=\left\{u(t,x):\, u\in C(\overline{D}), \,\,u_{xx},\,\, \partial_{0t}^{\alpha}u\in C(D),\,\,\int\limits_0^1 \sqrt{x} |u(t,x)|dx<+\infty\right\};$$ - boundary and non-local initial conditions $$\begin{aligned} \label{eq5-6} \lim\limits_{x\rightarrow 0} x u_x(t,x)=0,\,\, u(t,1)=0, \\ \label{eq5-6b} u(0,x)+M u(T,x)=0,\,\, 0\le x\le 1,\,\,\,\, [\alpha]\cdot u_t(0,t)=0,\,\,0<x<1. \end{aligned}$$ Main result =========== The main result of this note is the following well-posedness theorem for the initial problem -. The interesting part are the conditions on $f$ allowing one to handle the singularities in the coefficients of the Bessel operator, and the non-resonance conditions relating the parameter $M$ with the length $T$ of the time interval, coefficients and fractional orders of time-derivatives, through the multinomial Mittag-Leffler function. \[THM:main\] Assume that - $f(x,t)$ is differentiable four times with respect to $x$; - $f(0,t)=f'(0,t)=f''(0,t)=f'''(0,t)=0,\,\,f(1,t)=f'(1,t)=f''(1,t)=0$; - $\frac{\partial^4 f(x,t)}{\partial x}$ is bounded; - $f(x,t)$ is continuous and continuously differentiable with respect to $t$, and non-resonance conditions $$\label{EQ:M} M\neq -\frac{1}{E_{(\alpha-\alpha_1,\alpha-\alpha_2,...,\alpha),1}(\lambda_1 T^{\alpha-\alpha_1},...,\lambda_n T^{\alpha-\alpha_n},-\gamma_k^2T^\alpha)}$$ hold for all $k=1,2,\ldots$. Then there exists a unique solution of the problem -, and it can be written in the following form: $$\label{EQ:sol} \begin{aligned} &u(t,x)=\sum\limits_{k=1}^\infty \left[\int\limits_0^t z^{\alpha-1}E_{(\alpha-\alpha_1,\alpha-\alpha_2,...,\alpha-\alpha_n,\alpha),\alpha}(\lambda_1 z^{\alpha-\alpha_1},...,\lambda_n z^{\alpha-\alpha_n},-\gamma_k^2 z^\alpha)f_k(t-z)dz-\right.\\ &-\frac{M}{1+ME_{(\alpha-\alpha_1,\alpha-\alpha_2,...,\alpha),1}(\lambda_1 T^{\alpha-\alpha_1},...,\lambda_n T^{\alpha-\alpha_n},-\gamma_k^2T^\alpha)}\times\\ &\times \int\limits_0^T z^{\alpha-1}E_{(\alpha-\alpha_1,\alpha-\alpha_2,...,\alpha-\alpha_n,\alpha),\alpha}(\lambda_1 z^{\alpha-\alpha_1},...,\lambda_n z^{\alpha-\alpha_n},-\gamma_k^2 z^\alpha)f_k(T-z)dz+\\ &\left.+E_{(\alpha-\alpha_1,\alpha-\alpha_2,...,\alpha),1}(\lambda_1 t^{\alpha-\alpha_1},...,\lambda_n t^{\alpha-\alpha_n},-\gamma_k^2t^\alpha)\right]J_\nu(\gamma_k x). \end{aligned}$$ The numbers $\gamma_k$ and the functions appearing in the formula are explained in Section \[SEC:series\]. Briefly, here $E_{(\alpha-\alpha_1,\alpha-\alpha_2,...,\alpha-\alpha_n,\alpha),\alpha}(\cdot)$ is the multinomial Mittag-Leffler function, $J_\nu$ are the first kind Bessel functions, $\gamma_k$ are their positive zeros, $\lambda_k$ are coefficients in the operator , and $f_k$ are Bessel expansions of $f$. For the proof of Theorem \[THM:main\] we start finding a formal solution in a series form and the convergence of the appearing series will be shown in Section \[SEC:proof\]. Representation of a solution {#SEC:series} ---------------------------- We start by a formal discussion of the representation of solutions in . Let $$J_\nu(z)=\sum\limits_{i=0}^\infty \frac{(-1)^i(z/2)^{2i+\nu}}{i!(i+\nu)!}$$ be the Bessel function of the first kind (see e.g. [@Wats]). It is known that for $\nu>0$, the Bessel function $J_\nu(z)$ has countably many zeros, moreover, they are real and have pairwise opposite signs. Denote the $k^{th}$ positive root of the equation $J_\nu(z)=0$ by $\gamma_k$, $k=1,2,\ldots$. For large $k$, we have (see [@tols]) $$\gamma_k\simeq k\pi+\frac{\nu\pi}{2}-\frac{\pi}{4}.$$ We now expand functions $u(t,x)$ and $f(t,x)$ in the Fourier-Bessel series (see e.g. [@tols]), writing them in the form $$\label{eq7} u(t,x)=\sum\limits_{k=1}^\infty U_k(t) J_\nu\left(\gamma_k x\right),$$ $$\label{eq8} f(t,x)=\sum\limits_{k=1}^\infty f_k(t) J_\nu\left(\gamma_k x\right),$$ where $$\label{eq9} U_k(t)=\frac{2}{J_{\nu+1}^2(\gamma_k)}\int\limits_0^1 u(x,t)\, xJ_\nu\left(\gamma_k x\right)dx,$$ $$\label{eq10} f_k(t)=\frac{2}{J_{\nu+1}^2(\gamma_k)}\int\limits_0^1 f(x,t)\, xJ_\nu\left(\gamma_k x\right)dx.$$ It is known that if a function $g=g(x)$ is piecewise continuous on $[0,l]$ and satisfies $$\int\limits_0^l\sqrt{x}|g(x)|dx<+\infty,$$ then for $\nu>-1/2$, the Fourier-Bessel series converges at every point $x_0\in (0,l)$, see e.g. [@tols]. Since we are looking for a function $u(t,x)\in W$, it satisfies these required conditions in order to be represented by a Fourier-Bessel series. We substitute - into equation and obtain the eigenvalue problem $$\label{eq11} L(U_k)+\gamma_k^2U_k(t)=f_k(t).$$ According to [@gorluch], the solution for equation satisfying initial conditions $$\label{eq12} U_k(0)=A,\,\,[\alpha] U_k'(0)=0,$$ can be represented in the form $$\begin{gathered} \label{eq13} U_k(t)= \int\limits_0^t z^{\alpha-1}E_{(\alpha-\alpha_1,\alpha-\alpha_2,...,\alpha-\alpha_n,\alpha),\alpha}(\lambda_1 z^{\alpha-\alpha_1},...,\lambda_n z^{\alpha-\alpha_n},-\gamma_k^2 z^\alpha)\times \\ \times f_k(t-z)dz+A \overline{U_0}(t),\end{gathered}$$ where $$\begin{gathered} \label{eq14} \overline{U_0}(t)=1+\sum\limits_{i=1}^n \lambda_it^{\alpha-\alpha_i}E_{(\alpha-\alpha_1,\alpha-\alpha_2,...,\alpha-\alpha_n,\alpha),1+\alpha-\alpha_i}(\lambda_1 t^{\alpha-\alpha_1},...,\lambda_n t^{\alpha-\alpha_n},-\gamma_k^2 t^\alpha)-\\ -\gamma_k^2t^\alpha E_{(\alpha-\alpha_1,\alpha-\alpha_2,...,\alpha-\alpha_n,\alpha),1+\alpha}(\lambda_1 t^{\alpha-\alpha_1},...,\lambda_n t^{\alpha-\alpha_n},-\gamma_k^2 t^\alpha),\end{gathered}$$ and $$\label{eq15} E_{(a_1,a_2,...,a_n),b}(z_1,z_2,...,z_n)=\sum\limits_{k=0}^\infty \sum\limits_{\begin{array}{l}l_1+l_2+...+l_n=k\\ l_1\geq0,...,l_n\geq 0\\ \end{array}}\frac{k!}{l_1!...l_n!}\frac{\prod\limits_{i=1}^n z_i^{l_i}}{\Gamma(b+\sum\limits_{i=1}^n a_i l_i)}$$ is the multinomial Mittag-Leffler function ([@gorluch]). From we find that $$\begin{gathered} \label{eq16} U_k(T)= \int\limits_0^T z^{\alpha-1}E_{(\alpha-\alpha_1,\alpha-\alpha_2,...,\alpha-\alpha_n,\alpha),\alpha}(\lambda_1 z^{\alpha-\alpha_1},...,\lambda_n z^{\alpha-\alpha_n},-\gamma_k^2 z^\alpha)\times \\ \times f_k(T-z)dz+A \overline{U_0}(T).\end{gathered}$$ Considering $U_k(0)=A$, from the first condition in , we get the relation $$A+MU_k(T)=0.$$ Using we find $A$ to be $$\begin{gathered} A=-\frac{M}{1+M\overline{U_0}(T)}\int\limits_0^T z^{\alpha-1}E_{(\alpha-\alpha_1,\alpha-\alpha_2,...,\alpha-\alpha_n,\alpha),\alpha}(\lambda_1 z^{\alpha-\alpha_1},...,\lambda_n z^{\alpha-\alpha_n},-\gamma_k^2 z^\alpha)\times\\ \times f_k(T-z)dz.\end{gathered}$$ Substituting this value of $A$ into , we rewrite it as $$\begin{gathered} \label{eq17} U_k(t)=\int\limits_0^t z^{\alpha-1}E_{(\alpha-\alpha_1,\alpha-\alpha_2,...,\alpha-\alpha_n,\alpha),\alpha}(\lambda_1 z^{\alpha-\alpha_1},...,\lambda_n z^{\alpha-\alpha_n},-\gamma_k^2 z^\alpha)f_k(t-z)dz-\\ -\frac{M\overline{U_0}(t)}{1+M\overline{U_0}(T)}\int\limits_0^T z^{\alpha-1}E_{(\alpha-\alpha_1,\alpha-\alpha_2,...,\alpha-\alpha_n,\alpha),\alpha}(\lambda_1 z^{\alpha-\alpha_1},...,\lambda_n z^{\alpha-\alpha_n},-\gamma_k^2 z^\alpha)\times \\ \times f_k(T-z)dz.\end{gathered}$$ If we use the formula (see [@liu]) $$\begin{aligned} &1+\sum\limits_{j=1}^{n+1}\lambda_jt^{\alpha-\alpha_j}E_{(\alpha-\alpha_1,\alpha-\alpha_2,...,\alpha-\alpha_{n+1}),1+\alpha-\alpha_j}(\lambda_1 t^{\alpha-\alpha_1},...,\lambda_{n+1} t^{\alpha-\alpha_{n+1}})\\ &=E_{(\alpha-\alpha_1,\alpha-\alpha_2,...,\alpha-\alpha_{n+1}),1}(\lambda_1 t^{\alpha-\alpha_1},...,\lambda_{n+1} t^{\alpha-\alpha_{n+1}}), \end{aligned}$$ representation can be rewritten as $$\overline{U_0}(t)=E_{(\alpha-\alpha_1,\alpha-\alpha_2,...,\alpha-\alpha_{n+1}),1}(\lambda_1 t^{\alpha-\alpha_1},...,\lambda_n t^{\alpha-\alpha_n},-\gamma_k^2 t^\alpha).$$ Denoting $$\begin{gathered} \label{eq18-19} F_k(t)= \int\limits_0^t z^{\alpha-1}E_{(\alpha-\alpha_1,\alpha-\alpha_2,...,\alpha-\alpha_n,\alpha),\alpha}(\lambda_1 z^{\alpha-\alpha_1},...,\lambda_n z^{\alpha-\alpha_n},-\gamma_k^2 z^\alpha)f_k(t-z)dz, \\ \overline{U_0}(t)=E_{(\alpha-\alpha_1,\alpha-\alpha_2,...,\alpha),1}(\lambda_1 t^{\alpha-\alpha_1},...,\lambda_n t^{\alpha-\alpha_n},-\gamma_k^2t^\alpha),\end{gathered}$$ we rewrite the function as $$\label{eq20} U_k(t)=F_k(t)-\frac{M}{1+M\overline{U_0}(T)}F_k(T)\overline{U_0}(t).$$ We note that the above expression is well-defined in view of the non-resonance conditions , that is $$M\neq -\frac{1}{\overline{U_0}(T)}=-\frac{1}{E_{(\alpha-\alpha_1,\alpha-\alpha_2,...,\alpha),1}(\lambda_1 T^{\alpha-\alpha_1},...,\lambda_n T^{\alpha-\alpha_n},-\gamma_k^2T^\alpha)}$$ holds for all $k$. Finally, based on , we rewrite our formal solution as $$\label{eq21} u(t,x)=\sum\limits_{k=1}^\infty \left[F_k(t)-\frac{M}{1+M\overline{U_0}(T)}F_k(T)\overline{U_0}(t)\right]J_\nu(\gamma_kx).$$ Justification of formal solution {#SEC:proof} -------------------------------- In this section we prove convergence of the obtained infinite series corresponding to functions $u(t,x),\,u_{xx}(t,x)$ and $\partial_{0t}^\alpha u(t,x)$. In order to prove the convergence of these series, we use the estimate for the Mittag-Leffler function , obtained in [@yamli Lemma 3.2], of the form $$\label{eq22} |E_{(\alpha-\alpha_1,\alpha-\alpha_2,...,\alpha-\alpha_n),\rho}(z_1,z_2,...,z_n)|\leq \frac{C}{1+|z_1|}.$$ Let us first prove the convergence of series . For this, we collect several other known estimates. First, we use the following theorem on the estimate of the Fourier-Bessel coefficient: \[THM:1\] Let $f(x)$ be a function defined on the interval $[0,1]$ such that $f(x)$ is differentiable $2s$ times $(s\in\mathbb N)$ and such that \(1) $f(0)=f'(0)=...=f^{(2s-1)}(0)=0$; \(2) $f^{(2s)}(x)$ is bounded (this derivative may not exist at certain points); \(3) $f(1)=f'(1)=...=f^{(2s-2)}(1)=0$. Then the following inequality is satisfied by the Fourier-Bessel coefficients of $f(x)$: $$\label{eq23} |f_k|\leq \frac{c}{\gamma_k^{2s-1/2}}\,\,\,\,(c=const),$$ where $\gamma_k$ is the $k^{th}$ positive zero of of the function $J_\nu(x)$. In our case we have $$f_k(t)=F_k(t)-\frac{M}{1+M\overline{U_0}(T)}F_k(T)\overline{U_0}(t).$$ According to , using estimate and imposing conditions (1)-(3) of Theorem in the case $s=1$ on the function $f(t,x)$, we get $$\label{eq24} \left|F_k(t)-\frac{M}{1+M\overline{U_0}(T)}F_k(T)\overline{U_0}(t)\right|\leq \frac{C_1}{\gamma_k^{3/2}}.$$ The series then converges absolutely and uniformly on $[0,1]$ in view of the following theorem: \[THM:2\] If $\nu\geq 0$, $C>0$, and if the constants $c_k$ satisfy $$|c_k|\leq \frac{C}{\gamma_k^{1+\varepsilon}},$$ for some $\varepsilon>0$, then the series $$c_1J_\nu(\gamma_1x)+c_2J_\nu(\gamma_2x)+c_3J_\nu(\gamma_3x)+...$$ converges absolutely and uniformly on $[0,1]$. According to [@tols Theorem 2, p. 236], sufficient condition for differentiating the series twice term by term, i.e. for the validity of the relation $$\label{EQ:diff} u_{xx}(t,x)=\sum\limits_{k=1}^\infty U_k(t)\gamma_k^2 J_\nu''(\gamma_kx)$$ will be $$\label{EQ:quest} |U_k(t)|\leq \frac{C_2}{\gamma_k^{5/2+\epsilon}},$$ where $U_k$ are as in , and $C_2$ and $\epsilon$ are positive constants. Based on this estimation and considering also (see [@tols p. 233]) $$|J_\nu(\gamma_kx)|\leq \frac{C_3}{\sqrt{\gamma_k x}},\quad (C_3=const),$$ we deduce (see e.g. [@tols p. 236]) $$\left|U_k(t)\gamma_k^2J_\nu''(\gamma_kx)\right|\leq \frac{C_2}{\gamma_k^{2+\epsilon}x^2}+\frac{C_3(|\nu|+\nu^2)}{\gamma_k^{3+\epsilon}x^2\sqrt{x}}+\frac{C_4}{\gamma_k^{1+\epsilon}\sqrt{x}},$$ which provides the convergence of the series . We note that if we impose conditions on the given function $f(x,t)$ of the form - $f(x,t)$ is differentiable four times with respect to $x$; - $f(0,t)=f'(0,t)=f''(0,t)=f'''(0,t)=0,\,\,f(1,t)=f'(1,t)=f''(1,t)=0$; - $\frac{\partial^4 f(x,t)}{\partial x}$ is bounded, one can see that Theorem \[THM:2\] implies estimation . The convergence of series corresponding to $\partial _{0t}^\alpha u(x,t)$, $u_x(x,t)$ can be shown in similar ways, completing the proof of Theorem \[THM:main\]. [99]{} P. Agarwal, E. Karimov, M. Mamchuev and M. Ruzhansky, On boundary-value problems for a partial differential equation with Caputo and Bessel operators, arXiv:1611.01624 F. Al-Musalhi, N. Al-Salti and S. Kerbal, Inverse problems of a fractional differential equation with Bessel operator, arXiv: 1609.04587v1 V. Daftardar-Gejji, S. Bhalekar, Boundary value problems for multi-term fractional differential equations, J. Math. Anal. Appl., 345 (2008), 754-765. V. Isakov, Inverse problems for partial differential equations (Second edition), Springer, New York, 2006. A. A. Kerefov. The Gevrey problem for a certain mixed-parabolic equation. (Russian) Diff. Uravn., 13(1) (1977), 76-83. A. A. Kilbas, H. M. Srivastava, J. J. Trujillo, Theory and applications of fractional differential equations, Elsevier, Amsterdam, 2006. Z. Li, Y. Liu and M. Yamamoto, Initial-boundary value problems for multi-term time-fractional diffusion equations with positive constant coefficients, Appl. Math. Comput. 257 (2015), 381-397. Y. Liu, Strong maximum principle for multi-term time-fractional diffusion equations and its application to an inverse source problem. arXiv:1510.06878v1 Y. Luchko, R. Gorenflo, An operational method for solving fractional differential equations with the Caputo derivatives, Acta Math. Vietnam., 24 (1999), 207-233. K. Masood, S. Messaoudi and F. D. Zaman, Initial inverse problem in heat equation with Bessel operator, International Journal of Heat and Mass Transfer, 45 (2012), 2959-2965. A. M. Nakhushev, [Drobniye ischisleniya i yego primenenie, in: Fractional Calculus and its Applications]{}. Fizmatlit, Moscow, 2003 (in Russian). C. V. Pao, Reaction diffusion equations with nonlocal boundary and nonlocal initial conditions, J. Math. Anal. Appl., 195 (1995), 702-718. J. M. Rassias, E. T. Karimov, Boundary-value problems with non-local initial condition for degenerate parabolic equations, Contemp. Anal. Appl. Math., 1(1) (2013), 42-48. J. M. Rassias, E. T. Karimov, Boundary-value problems with non-local initial condition for parabolic equations with parameter, European J. Pure Appl. Math., 3(6) (2010), 948-957. M. Ruzhansky, J. Smith, Dispersive and Strichartz estimates for hyperbolic equations with constant coefficients, MSJ Memoirs, 22, Mathematical Society of Japan, Tokyo, 2010. N. N. Shopolov, Mixed problem with non-local initial condition for a heat conduction equation, Reports of Bulgarian Academy of Sciences, 3(7) (1981), 935-936. G. P. Tolstov, Fourier series, Prentice Hall, Inc., Englewood Cliffs, N.J., 1962. G. N. Watson, A treatise on the theory of Bessel functions, Cambridge University Press, Second Edition, 1966. [^1]: The last author was supported in parts by the EPSRC grant EP/K039407/1 and by the Leverhulme Grant RPG-2014-02. No new data was collected or generated during the course of research.

ArXiv

--- abstract: 'We study generalized cluster algebras introduced by Chekhov and Shapiro. When the coefficients satisfy the normalization and quasi-reciprocity conditions, one can naturally extend the structure theory of seeds in the ordinary cluster algebras by Fomin and Zelevinsky to generalized cluster algebras. As the main result, we obtain formulas expressing cluster variables and coefficients in terms of $c$-vectors, $g$-vectors, and $F$-polynomials.' address: 'Graduate School of Mathematics, Nagoya University, Chikusa-ku, Nagoya, 464-8604, Japan' author: - Tomoki Nakanishi bibliography: - '../../biblist/biblist.bib' title: Structure of seeds in generalized cluster algebras --- Introduction ============ In [@Chekhov11] Chekhov and Shapiro introduced [*generalized cluster algebras*]{}, which naturally generalize the ordinary cluster algebras by Fomin and Zelevinsky [@Fomin02]. In generalized cluster algebras, the celebrated [*binomial*]{} exchange relation for cluster variables of ordinary cluster algebras $$\begin{aligned} \begin{split} x'_k x_k &= p_k^- \prod_{j=1}^n x_j^{[-b_{jk}]_+ } + p_k^+ \prod_{j=1}^n x_j^{[b_{jk}]_+ } \\ &= \Biggl(\prod_{j=1}^n x_j^{[-b_{jk}]_+ }\Biggr) (p_k^{-} + p_k^+ w_k), \quad w_k=\prod_{j=1}^n x_j^{b_{jk}} \end{split}\end{aligned}$$ is replaced by the [*polynomial*]{} one of arbitrary degree $d_k\geq 1$, $$\begin{aligned} \label{eq:rel1} x'_k x_k = \Biggl(\prod_{j=1}^n x_j^{[-\beta_{jk}]_+ }\Biggr)^{d_k} \sum_{s=0}^{d_k} p_{k,s} w_k^s, \quad w_k=\prod_{j=1}^n x_j^{\beta_{jk}},\end{aligned}$$ where $\beta_{jk}=b_{jk}/d_k$ are assumed to be integers and the coefficients $p_{k,s}$ should be also mutated appropriately. This generalization is expected to be natural, since it originates in the transformations preserving the associated Poisson bracket [@Gekhtman05]. In fact, it was shown in [@Chekhov11] that the generalized cluster algebras have the [*Laurent property*]{}, which is regarded as the most characteristic feature of the ordinary cluster algebras. Furthermore, it was also shown in [@Chekhov11] that the [*finite-type classification*]{} of the generalized cluster algebras reduces to the one for the ordinary case. These results already imply that, despite the apparent complexity of their exchange relations , generalized cluster algebras may be well-controlled like the ordinary ones. See also [@Rupel13] for the result on greedy bases in rank 2 generalized cluster algebras. Besides the above cluster-algebra-theoretic interest, the generalized cluster algebra structure naturally appears for the Teichmüller spaces of Riemann surfaces with orbifold points [@Chekhov11]. More recently, it also appears in representation theory of quantum affine algebras [@Gleitz14] and also in the study of WKB analysis [@Iwaki14b]. In view of these developments, and also for potentially more versatility of polynomial exchange relations than the binomial one, it is not only natural but also necessary to develop a structure theory of seeds in generalized cluster algebras which is parallel to the one for the ordinary cluster algebras by [@Fomin07]. The core notion of the theory of [@Fomin07] is a cluster pattern with [*principal coefficients*]{}, from which other important notions such as [*$c$-vectors*]{}, [*$g$-vectors*]{}, and [*$F$-polynomials*]{} are also induced. Then, the main result of [@Fomin07] is the formulas expressing cluster variables and coefficients in terms of $c$-vectors, $g$-vectors, and $F$-polynomials. These formulas are especially important in view of the categorification of cluster algebras by (generalized) cluster categories (see [@Plamondon10b] and reference therein). The purpose of this paper is to provide results parallel to the above ones for generalized cluster algebras. To be more precise, we consider a class of generalized cluster algebras whose coefficients satisfy the [*normalization condition*]{} and what we call the [*quasi-reciprocity condition*]{}. For this class of generalized cluster algebras, we introduce the notions of a cluster pattern with principal coefficients, $c$-vectors, $g$-vectors, and $F$-polynomials. Then, as a main result, we obtain the formulas expressing cluster variables and coefficients in terms of $c$-vectors, $g$-vectors, and $F$-polynomials, which are parallel to the ones in [@Fomin07]. To summarize, [*generalized cluster algebras preserve essentially every feature of the ordinary ones*]{}, and this is the main message of the paper. We thank Anne-Sophie Gleitz, Kohei Iwaki, and Michael Shapiro for useful discussions and communications. Generalized cluster algebras ============================ In this section we recall basic notions of generalized cluster algebras following [@Chekhov11]. However, we slightly modify the setting of [@Chekhov11] to match the setting of (ordinary) cluster algebras in [@Fomin07]. Generalized seed mutations -------------------------- Throughout the paper we always assume that any matrix is an [*integer*]{} matrix. Recall that a matrix $B=(b_{ij})_{i.j=1}^n$ is said to be [*skew-symmetrizable*]{} if there is an $n$-tuple of positive integers $\bfd=(d_1,\dots,d_n)$ such that $d_ib_{ij}=-d_jb_{ji}$. We start by fixing a semifield $\bbP$, whose addition is denoted by $\oplus$. Let $\bbZ\bbP$ be the group ring of $\bbP$, and let $\bbQ\bbP$ be the field of the fractions of $\bbZ\bbP$. Let $w_1,\dots,w_n$ be any algebraic independent variables, and let $\calF=\bbQ\bbP(w)$ be the field of the rational functions in $w=(w_1$,…, $w_n)$ with coefficients in $\bbQ\bbP$. The following definition is the usual one [@Fomin07]. A (labeled) seed in $\bbP$ is a triplet $(\bfx,\bfy,B)$ such that - $B$ is a skew-symmetrizable matrix, called an [*exchange matrix*]{}, - $\bfx=(x_1,\dots,x_n)$ is an $n$-tuple of elements in $\calF$, called [*cluster variables*]{} or [*$x$-variables*]{}, - $\bfy=(y_1,\dots,y_n)$ is an $n$-tuple of elements in $\bbP$, called [*coefficients*]{} or [*$y$-variables*]{}. Next we introduce a pair $(\bfd,\bfz)$ of data for generalized seed mutations we consider. Firstly, $\bfd=(d_1,\dots,d_n)$ is an $n$-tuple of positive integers, and we call these integers the [*mutation degrees*]{}. We stress that we do [*not*]{} impose the skew-symmetric condition $d_ib_{ij}=-d_jb_{ji}$. Secondly, $\bfz$ is a family of elements in $\bbP$, $$\begin{aligned} \bfz=(z_{i,s})_{ i=1,\dots,n; s=1,\dots,d_i-1}\end{aligned}$$ satisfying the following condition: (reciprocity) $$\begin{aligned} \label{eq:p1} z_{i,s}=z_{i,d_i-s} \quad (s=1,\dots,d_i-1).\end{aligned}$$ We call them the [*frozen coefficients*]{}, since they are not “mutated", or simply the [*$z$-variables*]{}. We also set $$\begin{aligned} z_{i,0}=z_{i,d_i}=1.\end{aligned}$$ For $\bfd=(1,\dots,1)$, $\bfz$ is empty, and it reduces to the ordinary case. (Here and below, “ordinary" means the case of ordinary cluster algebras). \[defn:mut1\] Let $(\bfd,\bfz)$ be given as above. For any seed $(\bfx,\bfy,B)$ in $\bbP$ and $k=1,\dots,n$, the [*$(\bfd,\bfz)$-mutation of $(\bfx,\bfy,B)$ at $k$*]{} is another seed $(\bfx',\bfy',B')=\mu_k(\bfx,\bfy,B)$ in $\bbP$ defined by the following rule: $$\begin{aligned} \label{eq:bmut1} b'_{ij}&= \begin{cases} -b_{ij}& \mbox{$i=k$ or $j=k$}\\ b_{ij}+d_k \left([- b_{ik}]_+ b_{kj} + b_{ik}[ b_{kj}]_+\right) & \mbox{$i,j\neq k$,}\\ \end{cases} \\ \label{eq:ymut1} {y}'_i&= \begin{cases} \displaystyle {y}_k^{-1} & i=k\\ \displaystyle {y}_i \Biggl( {y}_k^{[\ve {b}_{ki}]_+} \Biggr)^{d_k} \Biggl( \bigoplus_{s=0}^{d_k} z_{k,s} {y}_k^{\ve s} \Biggr)^{-{b}_{ki}} & i\neq k,\\ \end{cases} \\ \label{eq:xmut1} x'_i&= \begin{cases} \displaystyle x_k^{-1} \Biggl(\prod_{j=1}^n x_j^{[-\ve {b}_{jk}]_+} \Biggr)^{d_k} \frac{\displaystyle \sum_{s=0}^{d_k} z_{k,s} \hat{y}_k^{\ve s} } {\displaystyle \bigoplus_{s=0}^{d_k} z_{k,s} {y}_k^{\ve s} } & i=k\\ x_i& i\neq k,\\ \end{cases}\end{aligned}$$ where $\ve=\pm 1$, $[a]_+=\max(a,0)$, and we set $$\begin{aligned} \label{eq:yhat1} \hat{y}_i &= y_i \prod_{j=1}^n x_j^{{b}_{ji}}.\end{aligned}$$ When the data $(\bfd,\bfz)$ is clearly assumed, we may drop the prefix and simply call it the [*(generalized) mutation*]{}. Let $D=(d_i\delta_{ij})_{i,j=1}^n$ be the diagonal matrix with diagonal entries $\bfd$. It is important to note that the mutation is equivalent to the [*ordinary*]{} mutation of exchange matrices between $DB$ and $DB'$, and also between $BD$ and $B'D$ in [@Fomin07]. The following properties are easy to confirm: - The formulas and are [*independent*]{} of the choice of the sign $\ve$ due to . - The mutation $\mu_k$ is involutive, i.e., $\mu_k(\mu_k(\bfx,\bfy,B))=(\bfx,\bfy,B)$. \[rem:CS1\] Here we transposed every matrix in [@Chekhov11]. Also the matrix $B$ therein is the matrix $DB^{T}$ here, and $\beta_{ij}$ therein is $b_{ji}$ here. In this paper we do not use the freedom of the choice of sign $\ve$ in and , and it can be safely set as $\ve=1$ throughout. Nevertheless, we keep it in all formulas involved since it is useful for several purposes, for example, to consider [*signed mutations*]{} which appeared in [@Iwaki14b]. Under the mutation $\mu_k$, the $\hat{y}$-variables mutate in the same way as the $y$-variables, namely, $$\begin{aligned} \label{eq:yhatmut1} \hat{y}'_i&= \begin{cases} \displaystyle \hat{y}_k^{-1} & i=k\\ \displaystyle \hat{y}_i \Biggl( \hat{y}_k^{[\ve {b}_{ki}]_+} \Biggl)^{d_k} \Biggl( \sum_{s=0}^{d_k} z_{k,s} \hat{y}_k^{\ve s} \Biggr)^{-{b}_{ki}} & i\neq k.\\ \end{cases}\end{aligned}$$ This is proved using the technique in [@Fomin07 Proposition 3.9]. Next let us explain how our setting is regarded as a specialization of the setting of [@Chekhov11]. In [@Chekhov11] a seed in $\bbP$ is defined as a triplet $(\bfx,\bfp,B)$, where $\bfx$ and $B$ are the same ones here (up to the identification of $B$ as in Remark \[rem:CS1\]), but $\bfp$ is a family of elements in $\bbP$, $$\begin{aligned} \label{eq:p1a} \bfp=(p_{i,s})_{i=1,\dots,n; s=0,\dots,d_i}.\end{aligned}$$ Then, for the mutation $(\bfx',\bfp',B')=\mu_k(\bfx,\bfp,B)$, the following formulas replace and : $$\begin{aligned} \label{eq:pmut1} \begin{split} p'_{k,s}&=p_{k,d_k-s},\\ \frac{p'_{i,s}}{p'_{i,0}}&= \begin{cases} \displaystyle \frac{p_{i,s}}{p_{i,0}} (p_{k,d_k}^{ {b}_{ki}} )^s & i\neq k,\ b_{ki}\geq 0\\ \displaystyle \frac{p_{i,s}}{p_{i,0}} (p_{k,0}^{ {b}_{ki}} )^s & i\neq k,\ b_{ki}\leq 0,\\ \end{cases} \end{split} \\ \label{eq:xmut2} x'_i&= \begin{cases} \displaystyle x_k^{-1} \Biggl(\prod_{j=1}^n x_j^{[- {b}_{jk}]_+} \Biggr)^{d_k} \Biggl(\sum_{s=0}^{d_k} p_{k,s} {u}_k^{ s} \Biggr) & i=k\\ x_i& i\neq k,\\ \end{cases}\end{aligned}$$ where $$\begin{aligned} \label{eq:z1} u_i &= \prod_{j=1}^n x_j^{{b}_{ji}}.\end{aligned}$$ Now, let us start from a seed $(\bfx,\bfy,B)$ in our setting. Comparing and , we naturally identify $$\begin{aligned} \label{eq:pu1} p_{i,s}=\frac{z_{i,s} {y}_i^{ s}}{\bigoplus_{r=0}^{d_i} z_{i,r} {y}_i^{ r}}.\end{aligned}$$ Then, it is easy to check that the mutation follows from and . Moreover, the specialization satisfies the following properties: (normalization) $$\begin{aligned} \label{eq:p2} \bigoplus_{s=0}^{d_i} p_{i,s} &= 1,\end{aligned}$$ (quasi-reciprocity) for each $i=1,\dots,n$, there is some $y_i\in \bbP$ such that $$\begin{aligned} \label{eq:p3} \frac{p_{i,s}} {p_{i,0}} \frac{p_{i,d_i}} {p_{i,d_i-s}}=y_i^{2s}, \qquad s=1,\dots, d_i.\end{aligned}$$ Conversely, suppose that a family $\bfp$ in satisfies properties and . First we note that such a $y_i$ is unique, since any semifield $\bbP$ is torsion-free [@Fomin02 Section 5]. Next we define $z_{i,s}\in \bbP$ ($i=1,\dots,n; s=0,\dots,d_i$) by $$\begin{aligned} \label{eq:u1} \frac{p_{i,s}} {p_{i,0}}=y_i^s z_{i,s}.\end{aligned}$$ In particular, we have $z_{i,0}=1$. Then, substituting in , we obtain $$\begin{aligned} \label{eq:u2} z_{i,s}z_{i,d_i}z_{i,d_i-s}^{-1}=1, \qquad s=1,\dots, d_i.\end{aligned}$$ In particular, by setting $s=d_i$, we have $z_{i,d_i}^2=1$. Once again, since $\bbP$ is torsion-free, we have $z_{i,d_i}=1$. Then, again by , we have the reciprocity $z_{i,s}=z_{i,d_i-s}$ ($s=1,\dots,d_i-1$). Meanwhile, by and , we have $$\begin{aligned} p_{i,0}= \frac{1}{\bigoplus_{s=0}^{d_i} z_{i,s} {y}_i^{ s}}.\end{aligned}$$ Then, by again, we recover the specialization . Finally, it is straightforward to recover the mutation from and . Furthermore, by , one can also confirm that the coefficients $z_{i,s}$ do not mutate. Generalized cluster algebras and Laurent property {#subsec:generalized} ------------------------------------------------- Let $\bbT_n$ be the $n$-regular tree whose edges are labeled by the numbers $1,\dots,n$. Following [@Fomin02], let us write $t {\buildrel k \over -} t'$ if the vertices $t$ and $t'$ of $\bbT_n$ are connected by the edge labeled by $k$. A [*$(\bfd,\bfz)$-cluster pattern $\Sigma$ in $\bbP$*]{} is an assignment of a seed $\Sigma_t$ in $\bbP$ to each vertex $t$ of $\bbT$ such that, if $t {\buildrel k \over -} t'$ then the assigned seeds $\Sigma_t$ and $\Sigma_{t'}$ are obtained from each other by the $(\bfd,\bfz)$-mutation at $k$. We fix a vertex $t_0$ of $\bbT_n$ and call it the [*initial vertex*]{}. Accordingly, the assigned seed $\Sigma_{t_0}=(\bfx_{t_0},\bfy_{t_0},B_{t_0})$ at $t_0$ is called the [*initial seed*]{}. Let us write, for simplicity, $$\begin{aligned} \bfx_{t_0}=\bfx=(x_1,\dots,x_n), \quad \bfy_{t_0}=\bfy=(y_1,\dots,y_n), \quad B_{t_0}=B=(b_{ij})_{i,j=1}^n.\end{aligned}$$ On the other hand, for the seed $\Sigma_{t}=(\bfx_{t},\bfy_{t},B_{t})$ assigned to a general vertex $t$ of $\bbT_n$, we write $$\begin{aligned} \bfx_{t}=(x^t_1,\dots,x^t_n), \quad \bfy_{t}=(y^t_1,\dots,y^t_n), \quad B_{t}=(b^t_{ij})_{i,j=1}^n.\end{aligned}$$ The [*generalized cluster algebra $\calA$ associated with a $(\bfd,\bfz)$-cluster pattern $\Sigma$ in $\bbP$*]{} is a $\bbZ\bbP$-subalgebra of $\calF$ generated by all $x$-variables $x^t_i$ ($t\in \bbT, i=1,\dots,n$) occurring in $\Sigma$. It is denoted by $\calA=\calA(\bfx,\bfy,B;\bfd,\bfz)$, where $(\bfx,\bfy,B)$ is the initial seed of $\Sigma$. For any $(\bfd,\bfz)$-cluster pattern in $\bbP$, each $x$-variable $x^t_i$ is expressed as a subtraction-free rational function of $\bfx$ with coefficients in $\bbQ\bbP$. The following stronger property due to [@Chekhov11] is of fundamental importance. \[thm:Laurent1\] For any $(\bfd,\bfz)$-cluster pattern in $\bbP$, each $x$-variable $x^t_i$ is expressed as a Laurent polynomial of $\bfx$ with coefficients in $\bbZ\bbP$. Example {#subsec:ex1} ------- As the simplest nontrivial example, we consider $\bfd=(2,1)$, $\bfz=(z_{1,1})$, and an initial seed $(\bfx,\bfy,B)$ in $\bbP$ such that $$\begin{aligned} B= \begin{pmatrix} 0 & -1 \\ 1 & 0\\ \end{pmatrix}.\end{aligned}$$ (This example also appears in the proof of [@Chekhov11 Theorem 2.7].) Accordingly, $$\begin{aligned} \hat{y}_1=y_1x_2, \quad \hat{y}_2=y_2x_1^{-1}.\end{aligned}$$ We note that $$\begin{aligned} DB= \begin{pmatrix} 0 & -2 \\ 1 & 0\\ \end{pmatrix}, \quad BD= \begin{pmatrix} 0 & -1 \\ 2 & 0\\ \end{pmatrix},\end{aligned}$$ which are the initial exchange matrices for ordinary cluster algebras of type $B_2=C_2$. Set $\Sigma(1)=(\bfx(1),\bfy(1),B(1))$ to be the initial seed $(\bfx,\bfy,B)$, and consider the seeds $\Sigma(t)=(\bfx(t), \bfy(t),B(t))$ ($t=2,\dots,7$) obtained by the following sequence of alternative mutations of $\mu_1$ and $\mu_2$. $$\begin{aligned} \label{eq:seedmutseq3} &\Sigma(1) \ \mathop{\leftrightarrow}^{\mu_{1}} \ \Sigma(2) \ \mathop{\leftrightarrow}^{\mu_{2}} \ \Sigma(3) \ \mathop{\leftrightarrow}^{\mu_{1}} \ \Sigma(4) \ \mathop{\leftrightarrow}^{\mu_{2}} \ \Sigma(5) \ \mathop{\leftrightarrow}^{\mu_{1}} \ \Sigma(6) \ \mathop{\leftrightarrow}^{\mu_{2}} \ \Sigma(7).\end{aligned}$$ By , we have $$\begin{aligned} B(t)=(-1)^{t+1}B.\end{aligned}$$ Then, using the exchange relations and , we obtain the explicit expressions of $x$- and $y$-variables in Table \[tab:data1\], where we set $z_{1,1}=z$ for simplicity. We observe the same periodicity of mutations of seeds for the ordinary cluster algebras of type $B_2=C_2$. $$\begin{aligned} {3} & \begin{cases} x_1(1)=x_1\\ x_2(1)=x_2,\\ \end{cases} && \hskip-70pt \begin{cases} y_1(1)=y_1\\ y_2(1)=y_2,\\ \end{cases} \\ \allowbreak & \begin{cases} \displaystyle x_1(2)=x_1^{-1}\frac{1+z\hat{y}_1+\hat{y}_1^2}{1\oplus z y_1\oplus y_1^2}\\ x_2(2)=x_2,\\ \end{cases} && \hskip-70pt \begin{cases} y_1(2)=y_1^{-1}\\ y_2(2)=y_2(1\oplus z y_1\oplus y_1^2),\\ \end{cases} \notag \\ & \begin{cases} \displaystyle x_1(3)=x_1^{-1}\frac{1+z\hat{y}_1+\hat{y}_1^2}{1\oplus z y_1\oplus y_1^2}\\ \displaystyle x_2(3)=x_2^{-1} \frac{1 + \hat{y}_2 +z \hat{y}_1\hat{y}_2+ \hat{y}_1^2 \hat{y}_2} {1\oplus y_2\oplus z y_1y_2 \oplus y_1^2y_2},\\ \end{cases} && \hskip-70pt \begin{cases} y_1(3)=y_1^{-1}(1\oplus y_2 \oplus z y_1y_2\oplus y_1^2 y_2)\\ y_2(3)=y_2^{-1}(1\oplus z y_1\oplus y_1^2)^{-1},\\ \end{cases} \notag \\ \allowbreak & \begin{cases} \displaystyle x_1(4)=x_1x_2^{-2} \frac{ 1+ 2 \hat{y}_2+ \hat{y}_2^2 + z\hat{y}_1\hat{y}_2+ z\hat{y}_1\hat{y}_2^2 + \hat{y}_1^2\hat{y}_2^2} {1\oplus 2 y_2\oplus y_2^2 \oplus zy_1y_2\oplus zy_1y_2^2 \oplus y_1^2y_2^2} \\ \displaystyle x_2(4)=x_2^{-1} \frac{1 + \hat{y}_2 +z \hat{y}_1\hat{y}_2+ \hat{y}_1^2 \hat{y}_2} {1\oplus y_2\oplus z y_1y_2 \oplus y_1^2y_2},\\ \end{cases} \notag \\ \allowbreak & && \hskip-70pt \begin{cases} y_1(4)=y_1(1\oplus y_2 \oplus z y_1y_2\oplus y_1^2 y_2)^{-1}\\ y_2(4)=y_1^{-2}y_2^{-1}(1\oplus 2 y_2\oplus y_2^2\\ \qquad \qquad \oplus zy_1y_2\oplus zy_1y_2^2 \oplus y_1^2y_2^2 ),\\ \end{cases} \notag \\ & \begin{cases} \displaystyle x_1(5)=x_1x_2^{-2} \frac{ 1+ 2 \hat{y}_2+ \hat{y}_2^2 + z\hat{y}_1\hat{y}_2+ z\hat{y}_1\hat{y}_2^2 + \hat{y}_1^2\hat{y}_2^2} {1\oplus 2 y_2\oplus y_2^2 \oplus zy_1y_2\oplus zy_1y_2^2 \oplus y_1^2y_2^2} \\ \displaystyle x_2(5)= x_1x_2^{-1} \frac{1 + \hat{y}_2 }{1\oplus y_2},\\ \end{cases} \notag \\ \allowbreak & && \hskip-70pt \begin{cases} y_1(5)=y_1^{-1}y_2^{-1}(1\oplus y_2)\\ y_2(5)=y_1^{2}y_2(1\oplus 2 y_2\oplus y_2^2\\ \qquad \qquad \oplus zy_1y_2\oplus zy_1y_2^2 \oplus y_1^2y_2^2 )^{-1},\\ \end{cases} \notag \\ \allowbreak & \begin{cases} \displaystyle x_1(6)=x_1\\ \displaystyle x_2(6)=x_1x_2^{-1}\frac{1+\hat{y}_2}{1\oplus y_2},\\ \end{cases} && \hskip-70pt \begin{cases} \displaystyle y_1(6)=y_1y_2(1\oplus y_2)^{-1}\\ y_2(6)=y_2^{-1},\\ \end{cases} \notag \\ \allowbreak & \begin{cases} x_1(7)=x_1\\ x_2(7)=x_2,\\ \end{cases} && \hskip-70pt \begin{cases} y_1(7)=y_1\\ y_2(7)=y_2.\\ \end{cases} \notag\end{aligned}$$ Structure of seeds in generalized cluster patterns ================================================== The goal of this section is to establish some basic structural results on seeds in a $(\bfd,\bfz)$-cluster pattern which are parallel to the ones in [@Fomin07]. $X$-functions and $Y$-functions ------------------------------- Let us temporarily regard $\bfy=(y_i)_{i=1}^n$, and $\bfz =(z_{i,s})_{i=1,\dots,n;s=1,\dots,d_i-1}$ with $z_{i,s}=z_{i,d_i-s}$ as formal variables. Let $\bbQ_{\mathrm{sf}}(\bfy, \bfz)$ be the [*universal semifield*]{} of $\bfy$ and $\bfz$, which consists of the rational functions in $\bfy$ and $\bfz$ with subtraction-free expressions [@Fomin07]. Let $\mathrm{Trop}(\bfy, \bfz)$ be the [*tropical semifield*]{} of $\bfy$ and $\bfz$, which is the multiplicative abelian group freely generated by $\bfy$ and $\bfz$ with [*tropical sum*]{} $\oplus$ defined by $$\begin{aligned} \label{eq:tsum1} \left( \prod_i y_i^{a_i} \prod_{i,s} z_{i,s}^{a_{i,s}} \right) \oplus \left( \prod_i y_i^{b_i} \prod_{i,s} z_{i,s}^{b_{i,s}} \right) = \prod_i y_i^{\min(a_i,b_i)} \prod_{i,s} z_{i,s}^{\min(a_{i,s},b_{i,s})}.\end{aligned}$$ A $(\bfd,\bfz)$-cluster pattern with [*principal coefficients*]{} is a $(\bfd,\bfz)$-cluster pattern in $\bbP=\mathrm{Trop}(\bfy, \bfz)$ with initial seed $(\bfx,\bfy,B)$, where $\bfx$ and $B$ are arbitrary. Let $\Sigma$ be the $(\bfd,\bfz)$-cluster pattern with principal coefficients and initial seed $(\bfx,\bfy,B)$. By the Laurent property in Theorem \[thm:Laurent1\], each $x$-variable $x^t_i$ in $\Sigma$ is expressed as $X^t_i(\bfx,\bfy,\bfz)\in \bbZ\bbP[\bfx^{\pm1}]$ with $\bbP=\mathrm{Trop}(\bfy, \bfz)$. We call them the [*$X$-functions*]{} of $\Sigma$. For principal coefficients, actually we have the following stronger result than Theorem \[thm:Laurent1\], which is parallel to [@Fomin03a Proposition 11.2] and [@Fomin07 Proposition 3.6]. \[prop:poly1\] We have $$\begin{aligned} \label{eq:X1} X^t_i(\bfx,\bfy,\bfz)\in \mathbb{Z}[\bfx^{\pm1},\bfy,\bfz].\end{aligned}$$ We follow the argument in the proof of [@Fomin03a Proposition 11.2]. Let $p$ be any variable in $\bfy$ or $\bfz$. Let us view $X^t_i(\bfx,\bfy,\bfz)$ as a Laurent polynomial in $p$, say $h(p)$, with coefficients of Laurent polynomials in the rest of the variables in $\bfx$, $\bfy$, and $\bfz$. We show that $h(p)$ is a polynomial in $p$ with nonzero constant term having subtraction-free rational expression by the induction on the distance between $t$ and $t_0$ in $\bbT_n$. The crucial point is that the coefficients $p_{k,s}=z_{k,s}y_k^s/ \bigoplus_{r=0}^{d_k} z_{k,r}y_k^r$ in the mutation are normalized as . Since $\bbP=\mathrm{Trop}(\bfy, \bfz)$, this means that $p_{k,s}$ ($s=0,\dots, d_r$) are polynomials in $p$, and there is no common factor in $p$. Thus, the right hand side of is a polynomial in $p$ with nonzero constant term having subtraction-free rational expression by the induction hypothesis and the “trivial lemma" in [@Fomin03a Lemma 5.2]. Let $\Sigma$ be the $(\bfd,\bfz)$-cluster pattern in the universal semifield $\bbQ_{\mathrm{sf}}(\bfy,\bfz)$ with initial seed $(\bfx,\bfy,B)$. Each $y$-variable $y^t_i$ in $\Sigma$ is expressed as a subtraction-free rational function $Y^t_i(\bfy,\bfz)\in \bbQ_{\mathrm{sf}}(\bfy, \bfz)$. We call them the [*$Y$-functions*]{} of $\Sigma$. Due to the universal property of the semifield $\bbQ_{\mathrm{sf}}(\bfy,\bfz)$ [@Fomin07 Definition 2.1], the following fact holds. \[lem:yeval1\] For any $(\bfd,\bfz)$-cluster pattern in $\bbP$ with the same initial exchange matrix $B$ as above, we have $$\begin{aligned} y^t_{i} = Y^t_i\vert_{\bbP}(\bfy,\bfz),\end{aligned}$$ where the right hand side stands for the evaluation of $Y^t_i(\bfy,\bfz)$ in $\bbP$. $c$-vectors, $F$-polynomials, and $g$-vectors --------------------------------------------- Let us extend the notions of $c$-vectors, $F$-polynomials, and $g$-vectors in [@Fomin07] to a $(\bfd,\bfz)$-cluster pattern with principal coefficients. ### $C$-matrices and $c$-vectors For a $(\bfd,\bfz)$-cluster pattern with principal coefficients, each $y$-variable $y^t_i\in \mathrm{Trop}(\bfy, \bfz)$ is, by definition, a Laurent monomial of $\bfy$ and $\bfz$ with coefficient 1. The following simple fact was observed in [@Iwaki14b] in the special case. \[lem:yev1\] Each $y$-variable $y^t_i$ is actually a Laurent monomial of $\bfy$ with coefficient 1. This is equivalent to saying that the frozen coefficients $\bfz$ never enter in $y^t_i$. This is true for the initial $y$-variables. Then, the claim can be shown by induction on the distance between $t$ and $t_0$ in $\bbT_n$, by inspecting the mutation and the definition of the tropical sum . Let $\Sigma$ be a $(\bfd,\bfz)$-cluster pattern with principal coefficients. Let us express each $y$-variable $y^t_j$ in $\Sigma$ as $$\begin{aligned} \label{eq:yc1} y^t_j =Y^t_i\vert_{\mathrm{Trop}(\bfy, \bfz)}(\bfy,\bfz)= \prod_{i=1}^n y_i^{c_{ij}^t}.\end{aligned}$$ The resulting matrices $C^t=(c^t_{ij})_{i,j=1}^n$ and their column vectors $c^t_{j}=(c^t_{ij})_{i=1}^n$ are called the [*$C$-matrices*]{} and the [*$c$-vectors*]{} of $\Sigma$, respectively. The following mutation/recurrence formula provides a combinatorial description of $c$-vectors. The $c$-vectors of a $(\bfd,\bfz)$-cluster pattern with principal coefficients satisfy the following recurrence relation: (initial condition) $$\begin{aligned} c_{ij}^{t_0} = \delta_{ij},\end{aligned}$$ (recurrence relation) for $t {\buildrel k \over -} t'$, $$\begin{aligned} \label{eq:cmut1} c^{t'}_{ij}= \begin{cases} -c^{t}_{ik} & j=k\\ c^{t}_{ij} +c^t_{ik}[\ve d_kb^t_{kj}]_+ +[-\ve c^t_{ik}]_+ d_kb^t_{kj} &j\neq k, \end{cases}\end{aligned}$$ where $\ve=\pm1$ and it is independent of the choice of the sign $\ve$. As already remarked in the proof of Lemma \[lem:yev1\], for a $(\bfd,\bfz)$-cluster pattern with principal coefficients, the mutation is simplified as $$\begin{aligned} \label{eq:ymut2} {y}^{t'}_i&= \begin{cases} \displaystyle {y}^{t}_k{}^{-1} & i=k\\ \displaystyle {y}^t_i \Biggl( {y}^t_k{}^{[\ve {b}^t_{ki}]_+} \Biggr)^{d_k} \Biggl( \bigoplus_{s=0}^{d_k} {y}^t_k{}^{\ve s} \Biggr)^{-{b}^t_{ki}} & i\neq k.\\ \end{cases}\end{aligned}$$ This is equivalent to due to the following formula in $\mathrm{Trop}(\bfy, \bfz)$: $$\begin{aligned} \label{eq:trop1} \frac{1 } {\displaystyle \bigoplus_{s=0}^{d_k} \Biggl(\prod_{j=1}^n y_j^{\ve c^t_{jk}} \Biggr)^{ s} } = \Biggl(\prod_{j=1}^n y_j^{[-\ve c^t_{jk}]_+} \Biggr)^{d_k} .\end{aligned}$$ We observe that the above relation coincides with the one for the $c$-vectors of the ordinary cluster pattern with principal coefficients and initial seed $(\bfx,\bfy,DB)$ in [@Fomin02 Proposition 5.8]. Therefore, we have the following result. \[prop:c1\] The $c$-vectors of the $(\bfd,\bfz)$-cluster pattern with principal coefficients and initial seed $(\bfx,\bfy,B)$ coincide with the $c$-vectors of the ordinary cluster pattern with principal coefficients and initial seed $(\bfx,\bfy,DB)$. Alternatively, one can relate these $c$-vectors with the $c$-vectors of the ordinary cluster pattern with principal coefficients and initial seed $(\bfx,\bfy,BD)$ as follows. Let us introduce $$\begin{aligned} \label{eq:ct1} \tilde{c}^t_{ij} = d^{-1}_i c^t_{ij} d_j. \end{aligned}$$ Then, $ \tilde{c}_{ij}^{t_0} = \delta_{ij}$, and is rewritten as $$\begin{aligned} \label{eq:ctmut1} \tilde{c}^{t'}_{ij}= \begin{cases} -\tilde{c}^{t}_{ik} & j=k\\ \tilde{c}^{t}_{ij} +\tilde{c}^t_{ik}[\ve b^t_{kj}d_j]_+ +[-\ve \tilde{c}^t_{ik}]_+ b^t_{kj}d_j &j\neq k. \end{cases}\end{aligned}$$ Therefore, we have the following result. \[prop:ct1\] The $\tilde{c}$-vectors, which are the column vectors in , of the $(\bfd,\bfz)$-cluster pattern with principal coefficients and initial seed $(\bfx,\bfy,B)$ coincide with the $c$-vectors of the ordinary cluster pattern with principal coefficients and initial seed $(\bfx,\bfy,BD)$. We need this alternative description for the description of the $g$-vectors below. ### $F$-polynomials Thanks to Proposition \[prop:poly1\], the following definition makes sense. Let $\Sigma$ be a $(\bfd,\bfz)$-cluster pattern with principal coefficients. For each $t\in \bbT_n$ and $i=1,\dots,n$, a polynomial $F^t_i(\bfy,\bfz)\in \mathbb{Z}[\bfy,\bfz]$ is defined by the specialization of the $X$-function $X^t_i(\bfx,\bfy,\bfz)$ of $\Sigma$ with $x_1=\cdots = x_n =1$. They are called the [*$F$-polynomials*]{} of $\Sigma$. The following mutation/recurrence formula provides a combinatorial description of $F$-polynomials. The $F$-polynomials for a $(\bfd,\bfz)$-cluster pattern with principal coefficients satisfy the following recurrence relation: [(initial condition)]{} $$\begin{aligned} F_i^{t_0}=1,\end{aligned}$$ (recurrence relation) for $t {\buildrel k \over -} t'$, $$\begin{aligned} \label{eq:Fmut1} F^{t'}_{i}= \begin{cases} \displaystyle F^t_k{}^{-1} \Biggl(\prod_{j=1}^n y_j^{[-\ve c^t_{jk}]_+} F^{t}_j{}^{[-\ve {b}^t_{jk}]_+} \Biggr)^{d_k} \displaystyle \sum_{s=0}^{d_k} z_{k,s} \Biggl(\prod_{j=1}^n y_j^{\ve c^t_{jk}} F^{t}_j{}^{\ve{b}^t_{jk}}\Biggr)^{ s} & i=k\\ F^t_i&i\neq k, \end{cases}\end{aligned}$$ where $\ve=\pm1$ and it is independent of the choice of the sign $\ve$. By specializing the mutation with $\bbP=\mathrm{Trop}(\bfy, \bfz)$, we obtain $$\begin{aligned} \label{eq:Xmut1} X^{t'}_i&= \begin{cases} \displaystyle X^t_k{}^{-1} \Biggl(\prod_{j=1}^n X^{t}_j{}^{[-\ve {b}^t_{jk}]_+} \Biggr)^{d_k} \frac{\displaystyle \sum_{s=0}^{d_k} z_{k,s} \Biggl(\prod_{j=1}^n y_j^{\ve c^t_{jk}} X^{t}_j{}^{\ve {b}^t_{jk}}\Biggr)^{ s} } {\displaystyle \bigoplus_{s=0}^{d_k} \Biggl(\prod_{j=1}^n y_j^{\ve c^t_{jk}} \Biggr)^{ s} } & i=k\\ X^t_i& i\neq k.\\ \end{cases}\end{aligned}$$ Then, specializing it with $x_1=\dots x_n=1$, and using , we obtain . ### $G$-matrices and $g$-vectors Let $\Sigma$ be the $(\bfd,\bfz)$-cluster pattern with principal coefficients and initial seed $(\bfx,\bfy,B)$. Let $\mathbb{Z}[\bfx^{\pm1},\bfy,\bfz]$ be the one in Proposition \[prop:poly1\]. Following [@Fomin07], we introduce a $\bbZ^n$-grading in $\mathbb{Z}[\bfx^{\pm1},\bfy,\bfz]$ as follows: $$\begin{aligned} \deg(x_i) = \bfe_i, \quad \deg(y_i) = -{\bfb}_j, \quad \deg(z_{i,r})=0.\end{aligned}$$ Here, $\bfe_i$ is the $i$th unit vector of $\bbZ^n$, and ${\bfb}_j=\sum_{i=1}^n {b}_{ij} \bfe_i$ is the $j$th column of the initial matrix $B=(b_{ij})_{i,j=1}^n$. Note that $\deg(\hat{y}_i)=0$ by . \[prop:gvec1\] The $X$-functions are homogeneous with respect to the $\bbZ^n$-grading. We repeat the same argument in [@Fomin07 Proposition 6.1] using the induction on the distance between $t$ and $t_0$ in $\bbT_n$. Using and Lemma \[lem:yeval1\] specialized to a $(\bfd,\bfz)$-cluster pattern with principal coefficients, we have $$\begin{aligned} \label{eq:Xmut2} X^{t'}_i&= \begin{cases} \displaystyle X^t_k{}^{-1} \Biggl(\prod_{j=1}^n X^{t}_j{}^{[-\ve {b}^t_{jk}]_+} \Biggr)^{d_k} \frac{\displaystyle \sum_{s=0}^{d_k} z_{k,s} Y^t_k{}^{\ve s}\vert_{\calF} (\hat{\bfy},\bfz) } {\displaystyle \bigoplus_{s=0}^{d_k} z_{k,s} {Y}^t_k{}^{\ve s}\vert_{\mathrm{Trop}(\bfy,\bfz)} (\bfy,\bfz) } & i=k\\ X^t_i& i\neq k.\\ \end{cases}\end{aligned}$$ Then, the right hand side is homogeneous due to the induction hypothesis. Let $\Sigma$ be the $(\bfd,\bfz)$-cluster pattern with principal coefficients and initial matrix $(\bfx,\bfy,B)$. Thanks to Proposition \[prop:gvec1\], the degree vector $\deg(X^t_i)$ of each $X$-function $X^t_i$ of $\Sigma$ is defined. Let us express it as $$\begin{aligned} \deg(X^t_j)=\sum_{i=1}^n g^t_{ij}\bfe_i.\end{aligned}$$ The resulting matrices $G^t=(g^t_{ij})_{i,j=1}^n$ and their column vectors $g^t_{j}=(g^t_{ij})_{i=1}^n$ are called the [*$G$-matrices*]{} and the [*$g$-vectors*]{} of $\Sigma$, respectively. The following mutation/recurrence formula provides a combinatorial description of $g$-vectors. The $g$-vectors of the $(\bfd,\bfz)$-cluster pattern with principal coefficients and initial seed $(\bfx,\bfy,B)$ satisfy the following recurrence relation: (initial condition) $$\begin{aligned} g_{ij}^{t_0} = \delta_{ij},\end{aligned}$$ (recurrence relation) for $t {\buildrel k \over -} t'$, $$\begin{aligned} \label{eq:gmut1} g^{t'}_{ij}= \begin{cases} \displaystyle -g^{t}_{ik} +\sum_{\ell=1}^n g^t_{i\ell} [-\ve {b}^t_{\ell k}d_k]_+ -\sum_{\ell=1}^n {b}_{i\ell}[-\ve c^t_{\ell k}d_k]_+ & j=k\\ g^{t}_{ij} &j\neq k, \end{cases}\end{aligned}$$ where $\ve=\pm1$ and it is independent of the choice of the sign $\ve$. This is obtained by comparing the degrees of both sides of . By using the $\tilde{c}$-vectors in , the relation is rewritten as follows. $$\begin{aligned} \label{eq:gmut2} g^{t'}_{ij}= \begin{cases} \displaystyle -g^{t}_{ik} +\sum_{\ell=1}^n g^t_{i\ell} [-\ve {b}^t_{\ell k}d_k]_+ -\sum_{\ell=1}^n {b}_{i\ell}d_{\ell}[-\ve \tilde{c}^t_{\ell k}]_+ & j=k\\ g^{t}_{ij} &j\neq k. \end{cases}\end{aligned}$$ Having Proposition \[prop:ct1\] in mind, we observe that this relation coincides with the one for the $g$-vectors of the ordinary cluster pattern with principal coefficients and initial seed $(\bfx,\bfy,BD)$ in [@Fomin07 Proposition 6.6]. Therefore, we have the following result. \[prop:g1\] The $g$-vectors of the $(\bfd,\bfz)$-cluster pattern with principal coefficients and initial seed $(\bfx,\bfy,B)$ coincide with the $g$-vectors of the ordinary cluster pattern with principal coefficients and initial seed $(\bfx,\bfy,BD)$. For the sake of completeness, let us also present the counterpart of Proposition \[prop:ct1\]. Let us introduce $$\begin{aligned} \label{eq:gt1} \tilde{g}^t_{ij}= d_i g^t_{ij}d_j^{-1}.\end{aligned}$$ Then, the relation is also rewritten as $$\begin{aligned} \label{eq:gmut3} \tilde{g}^{t'}_{ij}= \begin{cases} \displaystyle -\tilde{g}^{t}_{ik} +\sum_{\ell=1}^n \tilde{g}^t_{i\ell} [-\ve d_{\ell}b^t_{\ell k}]_+ -\sum_{\ell=1}^n d_i b_{i\ell}[-\ve c^t_{\ell k}]_+ & j=k\\ \tilde{g}^{t}_{ij} &j\neq k. \end{cases}\end{aligned}$$ Having Proposition \[prop:c1\] in mind, we observe that this relation coincides with the one for the $g$-vectors of the ordinary cluster pattern with principal coefficients and initial seed $(\bfx,\bfy,DB)$. Therefore, we have the following result. \[prop:gt1\] The $\tilde{g}$-vectors, which are the column vectors in , of the $(\bfd,\bfz)$-cluster pattern with principal coefficients and initial seed $(\bfx,\bfy,B)$ coincide with the $g$-vectors of the ordinary cluster pattern with principal coefficients and initial seed $(\bfx,\bfy,DB)$. We see a duality between the $c$-vectors and the $g$-vectors in Propositions \[prop:c1\], \[prop:ct1\], \[prop:g1\], and \[prop:gt1\]. In particular, the $c$-vectors are associated with the matrix $DB$, while the $g$-vectors are associated with the matrix $BD$. This is somewhat suggested from the beginning in the monomial parts in the relations and . ### Sign-coherence Let $\Sigma$ be a $(\bfd,\bfz)$-cluster pattern with principal coefficients. A $c$-vector $c^t_j$ of $\Sigma$ is said to be [*sign-coherent*]{} if it is nonzero and all components are either nonnegative or nonpositive. The following proposition is parallel to [@Fomin07 Proposition 5.6]. For any $(\bfd,\bfz)$-cluster pattern with principal coefficients, the following two conditions are equivalent. - Any $F$-polynomial $F^t_i(\bfy,\bfz)$ has constant term 1. - Any $c$-vector $c^t_i$ is sign-coherent. This is proved by a parallel argument to the one for [@Fomin07 Proposition 5.6] by using the recursion relation for the $F$-polynomials. We omit the detail. In the ordinary case it was conjectured in [@Fomin07 Conjecture 5.6] that the sign-coherence holds for any $c$-vector of any cluster pattern with principal coefficients. This was proved by [@Derksen10 Theorem 1.7] when the initial exchange matrix $B$ is skew-symmetric, and very recently it was proved in full generality by [@Gross14 Corollary 5.5]. Since our $c$-vectors are identified with the $c$-vectors of some ordinary cluster pattern with principal coefficients by Proposition \[prop:c1\], we obtain the following theorem as a corollary of [@Gross14 Corollary 5.5]. Any $c$-vector of any $(\bfd,\bfz)$-cluster pattern with principal coefficients is sign-coherent. As a consequence of the sign-coherence, we also obtain the following duality between the $C$- and $G$-matrices by applying [@Nakanishi11a Eq. (3.11)] (see also [@Nakanishi11c Proposition 3.2]), which is valid under the sign-coherence property. Recall that for a skew-symmetrizable matrix $B$ the matrix $DB$ is still skew-symmetrizable. Let $C^t$ and $G^t$ be the $C$- and $G$-matrices at $t\in \bbT_n$ of any $(\bfd,\bfz)$-cluster pattern $\Sigma$ with principal coefficients. Let $R=(r_i\delta_{ij})_{i,j=1}^n$ be a diagonal matrix with positive diagonal entries such that $RDB$ is skew-symmetric. Then, the following relation holds. $$\begin{aligned} R^{-1}D^{-1} (G^t)^T DRC^t = I.\end{aligned}$$ This is obtained by combining [@Nakanishi11a Eq. (3.11)] with Propositions \[prop:c1\] and \[prop:gt1\]. Main formulas ------------- Finally, we present the main formulas, which express the $x$- and $y$-variables of any $(\bfd,\bfz)$-cluster pattern $\Sigma$ in any semifield $\bbP$ in terms of $F$-polynomials, $c$-vectors, and $g$-vectors defined for the same initial exchange matrix of $\Sigma$. For any $(\bfd,\bfz)$-cluster pattern in $\bbP$, the following formula holds. $$\begin{aligned} \label{eq:ycf1} y^t_i= \prod_{j=1}^n y_j^{c^t_{ji}} \prod_{j=1}^n F^t_j\vert_{\bbP}(\bfy,\bfz)^{{b}^t_{ji}}.\end{aligned}$$ The derivation is parallel to [@Fomin07 Proposition 3.13]. We apply Lemma \[lem:yeval1\] to a $(\bfd,\bfz)$-cluster pattern with principal coefficients, and we obtain $$\begin{aligned} \hat{y}^t_i = Y^t_i(\hat{\bfy},\bfz).\end{aligned}$$ On the other hand, specializing to the same $(\bfd,\bfz)$-cluster pattern with principal coefficients, we have $$\begin{aligned} \begin{split} \hat{y}^t_i &= Y^t_i \vert_{\mathrm{Trop}(\bfy,\bfz)}(\bfy,\bfz)\prod_{j=1}^n X^t_j(\bfx,\bfy,\bfz)^{{b}^t_{ji}} \\ &=\prod_{j=1}^n y_j^{c^t_{ji}}\prod_{j=1}^n X^t_j(\bfx,\bfy,\bfz)^{{b}^t_{ji}} , \end{split}\end{aligned}$$ where we used in the second equality. Thus, we have $$\begin{aligned} Y^t_i(\hat{\bfy},\bfz)=\prod_{j=1}^n y_j^{c^t_{ji}} \prod_{j=1}^n X^t_j(\bfx,\bfy,\bfz)^{{b}^t_{ji}}.\end{aligned}$$ Now, we set $x_1=\dots=x_n=1$. Then, $\hat{\bfy}=\bfy$, and we obtain $$\begin{aligned} \label{eq:ycf2} Y^t_i (\bfy,\bfz)= \prod_{j=1}^n y_j^{c^t_{ji}} \prod_{j=1}^n F^t_j(\bfy,\bfz)^{{b}^t_{ji}}.\end{aligned}$$ Finally, evaluating it in $\bbP$, we obtain . For any $(\bfd,\bfz)$-cluster pattern in $\bbP$, the following formula holds. $$\begin{aligned} \label{eq:xf1} x^t_i & = \Biggl(\prod_{j=1}^n x_j^{g^t_{ji}}\Biggr) \frac{ F^t_i\vert_{\calF}(\hat{\bfy},\bfz) } { F^t_i\vert_{\bbP}({\bfy},\bfz) }.\end{aligned}$$ The derivation is parallel to [@Fomin07 Corollary 6.3]. First, we obtain the following equality exactly in the same way as [@Fomin07 Theorem 3.7], and we skip its derivation: $$\begin{aligned} \label{eq:xf2} x^t_i &= \frac{ X^t_i\vert_{\calF}(\bfx,{\bfy},\bfz) } { F^t_i\vert_{\bbP}({\bfy},\bfz) }.\end{aligned}$$ On the other hand, by the definition of the $g$-vectors, we have $$\begin{aligned} X^{t}_i (\dots,\gamma_ix_i,\dots; \dots, \prod_{j=1}^n \gamma_k ^{-{b}_{ki}}y_i, \dots; \dots, z_{i,r},\dots) = \Biggl(\prod_{j=1} \gamma_j^{g^{t}_{ji}}\Biggr) X^t_i(\bfx,\bfy,\bfz).\end{aligned}$$ By setting $\gamma_i = x_i^{-1}$, we have $$\begin{aligned} F^{t}_i (\hat{\bfy},\bfz) = \Biggl(\prod_{j=1} x_j^{-g^{t}_{ji}}\Biggr) X^t_i(\bfx,\bfy,\bfz).\end{aligned}$$ Combining it with , we obtain . Example {#example} ------- Let us consider the example in Section \[subsec:ex1\] again. From the data in Table \[tab:data1\], one can read off the following data for the $C$-matrix $C(t)$, the $G$-matrix $G(t)$ and the $F$-polynomials $F_i(t)$ for the seed $\Sigma(t)$ with principal coefficients therein. $$\begin{aligned} {3} C(1)&=\begin{pmatrix} 1 & 0 \\ 0 & 1\\ \end{pmatrix}, \quad & G(1)&=\begin{pmatrix} 1 & 0 \\ 0 & 1\\ \end{pmatrix}, \quad && \begin{cases} F_1(1)=1\\ F_2(1) = 1, \end{cases} \\ C(2)&=\begin{pmatrix} -1 & 0 \\ 0 & 1\\ \end{pmatrix}, \quad & G(2)&=\begin{pmatrix} -1 & 0 \\ 0 & 1\\ \end{pmatrix}, \quad && \begin{cases} F_1(2)=1+ zy_1 + y_1^2\\ F_2(2) = 1, \end{cases} \notag \\ C(3)&=\begin{pmatrix} -1 & 0 \\ 0 & -1\\ \end{pmatrix}, \quad & G(3)&=\begin{pmatrix} -1 & 0 \\ 0 & -1\\ \end{pmatrix}, \quad && \begin{cases} F_1(3)=1+ zy_1 + y_1^2\\ F_2(3) = 1+y_2+zy_1y_2+y_1^2y_2, \end{cases} \notag \\ C(4)&=\begin{pmatrix} 1 & -2 \\ 0 & -1\\ \end{pmatrix}, \quad & G(4)&=\begin{pmatrix} 1 & 0 \\ -2 & -1\\ \end{pmatrix}, \quad && \begin{cases} F_1(4)=1+ 2y_2 + y_2^2\\ \qquad\qquad+zy_1y_2+zy_1y_2^2+y_1^2y_2^2\\ F_2(4) = 1+y_2+zy_1y_2+y_1^2y_2, \end{cases} \notag \\ C(5)&=\begin{pmatrix} -1 & 2 \\ -1 & 1\\ \end{pmatrix}, \quad & G(5)&=\begin{pmatrix} 1 & 1 \\ -2 & -1\\ \end{pmatrix}, \quad && \begin{cases} F_1(5)=1+ 2y_2 + y_2^2\\ \qquad\qquad+zy_1y_2+zy_1y_2^2+y_1^2y_2^2\\ F_2(5) = 1+y_2, \end{cases} \notag \\ C(6)&=\begin{pmatrix} 1 & 0 \\ 1 & -1\\ \end{pmatrix}, \quad & G(6)&=\begin{pmatrix} 1 & 1 \\ 0 & -1\\ \end{pmatrix}, \quad && \begin{cases} F_1(6)=1\\ F_2(6) = 1+y_2, \end{cases} \notag \\ C(7)&=\begin{pmatrix} 1 & 0 \\ 0 & 1\\ \end{pmatrix}, \quad & G(7)&=\begin{pmatrix} 1 & 0 \\ 0 & 1\\ \end{pmatrix}, \quad && \begin{cases} F_1(7)=1\\ F_2(7) = 1. \end{cases} \notag\end{aligned}$$

ArXiv

--- abstract: 'The Dynkin algebras are the hereditary artin algebras of finite representation type. The paper exhibits the number of support-tilting modules for any Dynkin algebra. Since the support-tilting modules for a Dynkin algebra of Dynkin type $\Delta$ correspond bijectively to the generalized non-crossing partitions of type $\Delta$, the calculations presented here may also be considered as a categorification of results concerning the generalized non-crossing partitions. In the Dynkin case $\mathbb A$, we obtain the Catalan triangle, in the cases $\mathbb B$ and $\mathbb C$ the increasing part of the Pascal triangle, and finally in the case $\mathbb D$ an expansion of the increasing part of the Lucas triangle.' author: - | Mustafa A. A. Obaid\ `drmobaid@yahoo.com`\ \ S. Khalid Nauman\ `snauman@kau.edu.sa`\ \ Wafaa M. Fakieh\ `wafaa.fakieh@hotmail.com`\ \ Claus Michael Ringel\ `ringel@math.uni-bielefeld.de`\ \ King Abdulaziz University, P O Box 80200\ Jeddah, Saudi Arabia\ title: | The Numbers of Support-Tilting Modules\ for a Dynkin Algebra --- Introduction ============ Let $\Lambda$ be a hereditary artin algebra. Here we consider left $\Lambda$-modules of finite length and call them just modules. The category of all modules will be denoted by $\operatorname{mod}\Lambda$. We let $n = n(\Lambda)$ denote the [*rank*]{} of $\Lambda$; this is by definition the number of simple modules (always, when counting numbers of modules of a certain kind, we actually mean the number of isomorphism classes). Following earlier considerations of Brenner and Butler, tilting modules were defined in [@[HR1]]. In the present setting, a [*tilting module*]{} $M$ is a module without self-extensions with precisely $n$ isomorphism classes of indecomposable direct summands, and we will assume, in addition, that $M$ is multiplicity-free. The endomorphism ring of a tilting module is said to be a tilted algebra. There is a wealth of papers devoted to tilted algebras, and the [*Handbook of Tilting Theory*]{} [@[AHK]] can be consulted for references. The present paper deals with the Dynkin algebras: these are the connected hereditary artin algebras which are representation-finite, thus their valued quivers are of Dynkin type $\Delta_n = \mathbb A_n, \mathbb B_n, \dots, \mathbb G_2$ (see [@[DR1]]). Its aim is to discuss the number of tilting modules for such an algebra. The corresponding tilted algebras were classified by various authors in the eighties. It seems to be clear that a first step of such a classification result was the determination of all tilting modules, however there are only few traces in the literature (also the Handbook [@[AHK]] is of no help). Apparently, the relevance of the number of tilting modules was seen at that time only in special cases. The tilting modules for a linearly ordered quiver of type $\mathbb A_n$ were exhibited in [@[HR2]] and Gabriel [@[G]] pointed out that here we encounter one of the numerous appearances of the Catalan numbers $\frac1{n+1}\binom{2n}n$. For the cases $\mathbb D_n$, the number of tilting modules was determined by Bretscher-Läser-Riedtmann [@[BLR]] in their study of self-injective representation-finite algebras. Given a module $M$, we let $\Lambda(M)$ denote its [*support algebra;*]{} this is the factor algebra of $\Lambda$ modulo the ideal which is generated by all idempotents $e$ with $eM = 0$ and is again a hereditary artin algebra (but usually not connected, even if $\Lambda$ is connected). The rank of the support algebra of $M$ will be called the [*support-rank*]{} of $M$. A module $T$ is said to be [*support-tilting*]{} provided $M$ considered as a $\Lambda(M)$-module is a tilting module. It may be well-known that the number of tilting modules of a Dynkin algebra depends only on its Dynkin type; at least for path algebras of quivers we can refer to Ladkani [@[L]]. Section \[algebras\] of the present paper provides a proof in general. It follows that the number of support-tilting modules with support-rank $s$ also depends only on the type $\Delta_n$; we let $a_s(\Delta_n)$ denote the number of support-tilting $\Lambda$-modules with support-rank $s$, where $\Lambda$ is of type $\Delta_n$. Of course, $a_n(\Delta_n)$ is just the number of tilting modules, and we denote by $a(\Delta_n)$ the number of all support-tilting modules; thus $a(\Delta_n) = \sum_{s=0}^n a_s(\Delta_n)$. The present paper presents the numbers $a(\Delta_n)$ and $a_s(\Delta_n)$ for $0\le s \le n$ in a unified way. Of course, the exceptional cases $\mathbb E_6, \mathbb E_7,\mathbb E_8, \mathbb F_4, \mathbb G_2$ can be treated with a computer (but actually, also by hand); thus our main interest lies in the series $\mathbb A, \mathbb B, \mathbb C, \mathbb D$. In the case $\mathbb A$, we obtain in this way the [**Catalan triangle**]{} , in the case $\mathbb B$ and $\mathbb C$ the increasing part of the [**Pascal triangle,**]{} and finally in the case $\mathbb D$ an [**expansion**]{} of the increasing part of the [**Lucas triangle**]{} (see Section \[triangles\]; an outline will be given later in the introduction). **The numbers** --------------- All the numbers which are presented here for the cases $\mathbb A, \mathbb B, \mathbb C, \mathbb D$ are related to the binomial coefficients $\binom st$ and they coincide for $\mathbb B_n$ and $\mathbb C_n$ (as we will show in Section \[algebras\]); thus it is sufficient to deal with the cases $\mathbb A, \mathbb B, \mathbb D$. For $\mathbb B$, the binomial coefficients themselves will play a dominant role. For the cases $\mathbb A$ and $\mathbb D$, suitable multiples are relevant. In case $\mathbb A$, these are the Catalan numbers $C_n = \frac1{n+1}\binom{2n}n$, as well as related numbers. For the case $\mathbb D$, it will be convenient to use the notation $\left[\smallmatrix t\cr s \endsmallmatrix\right] = \frac{s+t}t\binom t s$ as proposed by Bailey [@[B]], since the relevant numbers in case $\mathbb D$ can be written in this way. Hubery and Krause [@[HK]] have pointed out that the numbers $a(\Delta)$ for the simply laced diagrams $\Delta$ were discussed already in 1987 by Gabriel and de la Peña [@[GP]], but let us quote “although they have the correct number for $\mathbb E_8$, their numbers for $\mathbb E_6$ and $\mathbb E_7$ are slightly wrong”. The numbers $a(\Delta_n)$ and $a_s(\Delta_n)$ for $0\le s \le n$: $$\hbox{\beginpicture \setcoordinatesystem units <1.8cm,1.2cm> \plot -.3 -.5 7.25 -.5 / \plot 0.4 0.3 0.4 -3.4 / \put{$\Delta_n$} at 0 0 \put{$\mathbb A_n$} at 1 0 \put{$\mathbb B_n, \mathbb C_n$} at 2 0 \put{$\mathbb D_n$} at 3 0 \put{$\mathbb E_6$} at 4 0 \put{$\mathbb E_7$} at 4.8 0 \put{$\mathbb E_8$} at 5.6 0 \put{$\mathbb F_4$} at 6.4 0 \put{$\mathbb G_2$} at 7 0 \put{$a_n(\Delta_n)$} at 0 -1 \put{$\frac1{n+1}\binom{2n}n$} at 1 -1 \put{$\binom{2n-1}{n-1}$} at 2 -1 \put{$\left[{\smallmatrix 2n-2\cr n-2\endsmallmatrix}\right]$} at 3 -1 \put{$418$} at 4 -1 \put{$2\,431$} at 4.8 -1 \put{$17\,342$} at 5.6 -1 \put{$66$} at 6.4 -1 \put{$5$} at 7 -1 \put{$a_s(\Delta_n)$} at 0 -1.9 \put{$\smallmatrix 0\le s < n \endsmallmatrix$} at 0 -2.25 \put{$\frac{n-s+1}{n+1}\binom{n+s}s$} at 1 -2 \put{$\binom{n+s-1}s$} at 2 -2 \put{$\left[{\smallmatrix n+s-2\cr s\endsmallmatrix}\right]$} at 3 -2 \put{\rm see Section \ref{exceptional}} at 5.5 -2 \setdots <1mm> \plot 3.8 -2.05 4.8 -2.05 / \plot 6.2 -2.05 7.1 -2.05 / \put{$a(\Delta_n)$} at 0 -3 \put{$\frac1{n+2}\binom{2n+2}{n+1}$} at 1 -3 \put{$\binom{2n}n$} at 2 -3 \put{$\left[{\smallmatrix 2n-1\cr n-1\endsmallmatrix}\right]$} at 3 -3 \put{$833$} at 4 -3 \put{$4\,160$} at 4.8 -3 \put{$25\,080$} at 5.6 -3 \put{$105$} at 6.4 -3 \put{$8$} at 7 -3 \endpicture}$$ By analogy with the Bailey notation $\left[\smallmatrix t\cr s\endsmallmatrix\right]$ one may be tempted to introduce the following notation for the Catalan triangle: $\left]\smallmatrix t\cr s\endsmallmatrix\right[ = \frac{t-2s+1}{t-s+1}\binom t s$. Then the numbers for the case $\mathbb A$ are written as follows: $$a_n(\mathbb A_n) = \left]\smallmatrix 2n\cr n\endsmallmatrix\right[,\ a_s(\mathbb A_n) = \left]\smallmatrix n+s\cr s\endsmallmatrix\right[,\ a(\mathbb A) = \left]\smallmatrix 2n+2\cr n+1\endsmallmatrix\right[.$$ The reader should observe that for $\mathbb A_n$ and $\mathbb B_n$, the formula given for $a_s(\Delta_n)$ and $0\le s < n$ works also for $s = n$. This is not the case for $\mathbb D_n$: whereas $\binom{2n-2}{n-2} = \binom{2n-2}n$, the numbers $\left[{\smallmatrix 2n-2\cr n-2\endsmallmatrix}\right]$ and $\left[{\smallmatrix 2n-2\cr n\endsmallmatrix}\right]$ are different (the difference will be highlighted at the end of Section \[triangles\]). The Lucas triangle consists of the numbers $\left[{\smallmatrix t\cr s\endsmallmatrix}\right]$ for all $0 \le s \le t$; it therefore uses the numbers $\left[{\smallmatrix 2n-2\cr n\endsmallmatrix}\right]$ at the positions, whereas the $\mathbb D$-triangle (which we will now consider) uses the numbers $\left[{\smallmatrix 2n-2\cr n-2\endsmallmatrix}\right]$. The triangles $\mathbb A, \mathbb B, \mathbb D$ ----------------------------------------------- The non-zero numbers $a_s(\Delta_n)$ for $\Delta = \mathbb A, \mathbb B, \mathbb D$ yield three triangles having similar properties. We will exhibit them in Section \[triangles\]; see the triangles \[triangle1\], \[triangle2\], \[triangle3\]. The triangle \[triangle1\] of type $\mathbb A$ is the Catalan triangle itself; this is in Sloane’s OEIS [@[S]]. The triangle \[triangle2\] of type $\mathbb B$ is the triangle , corresponding to the increasing part of the Pascal triangle (thus it consists of the binomial coefficients $\binom ts$ with $2s\le t+1$). The triangle \[triangle3\] of type $\mathbb D$ is an expansion of the increasing part of the Lucas triangle . Taking the increasing part of the rows in the Lucas triangle (thus the numbers $\left[\smallmatrix t\cr s \endsmallmatrix\right]$ with $2s\le t+1$), we obtain numbers which occur in the triangle of type $\mathbb D$, namely the numbers $a_s(\mathbb D_n)$ with $0 \le s < n$; the numbers $a_n(\mathbb D_n)$ on the diagonal however are given by a similar, but deviating formula (they are listed as the sequence ). The Lucas triangle is , but the triangle $\mathbb D$ itself was, at the time of the writing, not yet recorded in OEIS; now it is . We see that the entries $a_s(n)$ of the triangles $\mathbb A$ and $\mathbb B$, as well as those of the lower triangular part of the triangle $\mathbb D$ can be obtained in a unified way from three triangles with entries $z_s(t)$ which satisfy the following recursion formula $$z_s(t) = z_{s-1}(t-1)+z_s(t-1)$$ (they are exhibited in Section \[triangles\] as triangles S \[triangle1-S\], S \[triangle2-S\], S \[triangle3-S\] using the shearing $a_s(n) = z_s(n+s-1)$). The recursion formula can be rewritten as $z_s(t) = \sum_{i= 0}^{s}z_i(t-s+i+1)$ (sometimes called the hockey stick formula). A consequence of the hockey stick formula is the fact that summing up the rows of any of the three triangles $\mathbb A, \mathbb B, \mathbb D$, we again obtain numbers which appear in the triangle. Let us provide further details on the triangles to be sheared. Consider first the case $\mathbb B$. Here we start with the Pascal triangle; thus we deal with the triangle with numbers $z_s(t) = \binom ts$ and the initial conditions are $z_0(t) = z_t(t) = 1$ for all $t\ge 0$. In case $\mathbb D$, we start with the Lucas triangle with numbers $z_s(t) = \left[\smallmatrix t\cr s \endsmallmatrix\right]$, and the initial conditions are $z_0(t) = 1,\ z_t(t) = 2$ for all $t\ge 1$ (these initial conditions are the reason for calling the Lucas triangle also the $(1,2)$-triangle). In the case $\mathbb A$ we start with a sheared Catalan triangle, and here the initial conditions are $z_0(t) = 1$ and $z_{t+1}(2t) = 0$ for all $t\ge 0$. Related results --------------- Let us repeat that in this paper $a_n(\Delta_n)$ denotes the number of tilting modules, $a(\Delta_n)$ the number of support-tilting modules, for $\Lambda$ of Dynkin type $\Delta_n$. As we have mentioned, the relevance of the numbers $a_n(\Delta_n)$ and $a(\Delta_n)$ was not fully realized in the eighties. It became apparent through the work of Fomin and Zelevinksy when dealing with cluster algebras and the corresponding cluster complexes (see in particular [@[FZ]] and [@[FR]]): the numbers $a_n(\Delta_n)$ and $a(\Delta_n)$ appear in [@[FZ]] as the numbers $N(\Delta_n)$ of clusters and $N^+(\Delta_n)$ of positive clusters, respectively (see Propositions 3.8 and 3.9 of [@[FZ]]). For the numbers $a_s(\Delta_n)$ in general, see Chapoton [@[C]] in case $\mathbb A$ and $\mathbb B$ and Krattentaler [@[Kt]] in case $\mathbb D$. A conceptual proof of the equalities $a(\Delta_n) = N(\Delta_n)$ and $a_n(\Delta_n) = N^+(\Delta_n)$ has been given by Ingalls and Thomas [@[IT]] in case $\Delta_n$ is simply laced (thus of type $\mathbb A, \mathbb D$ of $\mathbb E$). The considerations of Ingalls and Thomas have been extended by the authors [@[ONFR]] to the non-simply laced cases. The papers [@[IT]] and [@[ONFR]] show in which way the representation theory of hereditary artin algebras can be used in order to categorify the cluster complex of Fomin and Zelevinsky: this is the reason for the equalities. Another method to relate clusters and support tilting modules is due to Marsh, Reineke and Zelevinsky [@[MRZ]]. Finally, let us stress that also the Coxeter diagrams $\mathbb H_3$ and $\mathbb H_4$ can be treated in a similar way, using hereditary artinian rings which are not artin algebras; this will be shown in [@[FR]]. The main result of the present paper is the direct calculation of the numbers $a_n(\Delta_n)$ in the case $\Delta = \mathbb B$; see Section \[tilting\]. Of course, using [@[ONFR]], this calculation can be replaced by referring to the determination of the corresponding cluster numbers by Fomin and Zelevinsky in [@[FZ]]. On the other hand, we hope that our proof is of interest in itself. There is an independent development which has to be mentioned, namely the theory of generalized non-crossing partitions (see for example [@[A]]). It is the Ingalls-Thomas paper [@[IT]] (and [@[ONFR]]; see also the survey [@[R2]]) which provides the basic setting for using the representation theory of a hereditary artin algebra $\Lambda$ in order to deal with non-crossing partitions. It turns out that there is a large number of counting problems for $\operatorname{mod}\Lambda$ which yield the same answer, namely the numbers $a(\Delta_n)$ and $a_s(\Delta_n)$. For example, $a_s(\Delta)$ is also the number of antichains in $\operatorname{mod}\Lambda$ of size $s$: an [*antichain*]{} $A = \{A_1,\dots,A_t\}$ in $\operatorname{mod}\Lambda$ is a set of pairwise orthogonal bricks (a brick is a module whose endomorphism ring is a division ring, and two bricks $A_1, A_2$ are said to be orthogonal provided $\operatorname{Hom}(A_1,A_2) = 0 = \operatorname{Hom}(A_2,A_1)$; antichains are called discrete subsets in [@[GP]] and $\operatorname{Hom}$-free subsets in [@[HK]]). Since the support-tilting modules for a Dynkin algebra of Dynkin type $\Delta$ correspond bijectively to the non-crossing partitions of type $\Delta$, the calculations presented here may be considered as a categorification of results concerning non-crossing partitions (for a general outline see Hubery-Krause [@[HK]]). Finally, let us mention that there is a corresponding discussion of the number of ad-nilpotent ideals of a Borel subalgebra of a simple Lie algebra; see Panyushev [@[P]]. **Outline of the paper** ------------------------ Let us stress again that there is an inductive procedure using the hook formula (Proposition \[hook-formula\]) and a modified hook formula (Proposition \[modified\]) in order to obtain the numbers $a_s(\mathbb A_n)$ for $0\le s \le n$, as well as the numbers $a_s(\Delta_n)$ for $\Delta = \mathbb B, \mathbb D$ for $0 \le s < n$, provided we know the numbers $a_n(\Delta_n)$. As we have mentioned, for the numbers $a_n(\mathbb D_n)$ we may refer to [@[BLR]]. In Section \[algebras\], we will show that the numbers $a_n(\mathbb B_n)$ and $a_n(\mathbb C_n)$ coincide; thus it remains to determine the numbers $a_n(\mathbb B_n)$. This will be done in Section \[tilting\]. In Section \[summation\], we calculate $a(\Delta_n)$ for $\Delta = \mathbb A, \mathbb B, \mathbb D$. Section \[triangles\] presents the triangles $\mathbb A, \mathbb B, \mathbb D$ as well as the corresponding Catalan, Pascal, and Lucas triangles, and some observations concerning repetition of numbers in the triangles are recorded. Section \[exceptional\] provides the numbers $a_s(\Delta_n)$ for the exceptional cases $\Delta_n = \mathbb E_6, \mathbb E_7, \mathbb E_8, \mathbb F_4, \mathbb G_2$. The triangles {#triangles} ============= \[triangle1\][**The triangle of type $\mathbb A$; this is** ]{} $$\hbox{\beginpicture \setcoordinatesystem units <1cm,.45cm> \put{$a_s(\mathbb A_n) = \dfrac{n-s+1}{n+1}\dbinom{n+s}s$} at 7 -1.5 \multiput{1} at 0 0 0 -1 0 -2 0 -3 0 -4 0 -5 0 -6 0 -7 0 -8 0 -9 / \put{1} at 1 -1 \put{2} at 1 -2 \put{2} at 2 -2 \put{3} at 1 -3 \put{5} at 2 -3 \put{5} at 3 -3 \put{4} at 1 -4 \put{9} at 2 -4 \put{14} at 3 -4 \put{14} at 4 -4 \put{5} at 1 -5 \put{14} at 2 -5 \put{28} at 3 -5 \put{42} at 4 -5 \put{42} at 5 -5 \put{6} at 1 -6 \put{20} at 2 -6 \put{48} at 3 -6 \put{90} at 4 -6 \put{132} at 5 -6 \put{132} at 6 -6 \put{7} at 1 -7 \put{27} at 2 -7 \put{75} at 3 -7 \put{165} at 4 -7 \put{297} at 5 -7 \put{429} at 6 -7 \put{429} at 7 -7 \put{8} at 1 -8 \put{35} at 2 -8 \put{110} at 3 -8 \put{275} at 4 -8 \put{572} at 5 -8 \put{1001} at 6 -8 \put{1430} at 7 -8 \put{1430} at 8 -8 \put{9} at 1 -9 \put{44} at 2 -9 \put{154} at 3 -9 \put{429} at 4 -9 \put{1001} at 5 -9 \put{2002} at 6 -9 \put{3432} at 7 -9 \put{4862} at 8 -9 \put{4862} at 9 -9 \put{${\smallmatrix n\endsmallmatrix}$} at -2 1 \put{${\smallmatrix 0\endsmallmatrix}$} at -2 0 \put{${\smallmatrix 1\endsmallmatrix}$} at -2 -1 \put{${\smallmatrix 2\endsmallmatrix}$} at -2 -2 \put{${\smallmatrix 3\endsmallmatrix}$} at -2 -3 \put{${\smallmatrix 4\endsmallmatrix}$} at -2 -4 \put{${\smallmatrix 5\endsmallmatrix}$} at -2 -5 \put{${\smallmatrix 6\endsmallmatrix}$} at -2 -6 \put{${\smallmatrix 7\endsmallmatrix}$} at -2 -7 \put{${\smallmatrix 8\endsmallmatrix}$} at -2 -8 \put{${\smallmatrix 9\endsmallmatrix}$} at -2 -9 \plot -2 1.5 -1.5 1 / \put{${\smallmatrix s\endsmallmatrix}$} at -1.5 1.5 \put{${\smallmatrix 0\endsmallmatrix}$} at 0 1.5 \put{${\smallmatrix 1\endsmallmatrix}$} at 1 1.5 \put{${\smallmatrix 2\endsmallmatrix}$} at 2 1.5 \put{${\smallmatrix 3\endsmallmatrix}$} at 3 1.5 \put{${\smallmatrix 4\endsmallmatrix}$} at 4 1.5 \put{${\smallmatrix 5\endsmallmatrix}$} at 5 1.5 \put{${\smallmatrix 6\endsmallmatrix}$} at 6 1.6 \put{${\smallmatrix 7\endsmallmatrix}$} at 7 1.5 \put{${\smallmatrix 8\endsmallmatrix}$} at 8 1.5 \put{${\smallmatrix 9\endsmallmatrix}$} at 9 1.5 \put{} at 0 -10 \put{sum} [r] at 11.5 1.5 \put{1} [r] at 11.5 0 \put{2} [r] at 11.5 -1 \put{5} [r] at 11.5 -2 \put{14} [r] at 11.5 -3 \put{42} [r] at 11.5 -4 \put{132} [r] at 11.5 -5 \put{429} [r] at 11.5 -6 \put{1430} [r] at 11.5 -7 \put{4862} [r] at 11.5 -8 \put{16796} [r] at 11.5 -9 \endpicture}$$ \[triangle2\][**The triangle of type $\mathbb B$; this is** ]{} $$\hbox{\beginpicture \setcoordinatesystem units <1cm,.45cm> \put{$a_s(\mathbb B_n) = \dbinom{n+s-1}s$} at 7 -1.5 \multiput{1} at 0 -1 0 -2 0 -3 0 -4 0 -5 0 -6 0 -7 0 -8 0 -9 / \put{1} at 0 0 \put{1} at 11.4 0 \put{1} at 1 -1 \put{2} at 1 -2 \put{3} at 2 -2 \put{3} at 1 -3 \put{6} at 2 -3 \put{10} at 3 -3 \put{4} at 1 -4 \put{10} at 2 -4 \put{20} at 3 -4 \put{35} at 4 -4 \put{5} at 1 -5 \put{15} at 2 -5 \put{35} at 3 -5 \put{70} at 4 -5 \put{126} at 5 -5 \put{6} at 1 -6 \put{21} at 2 -6 \put{56} at 3 -6 \put{126} at 4 -6 \put{252} at 5 -6 \put{462} at 6 -6 \put{7} at 1 -7 \put{28} at 2 -7 \put{84} at 3 -7 \put{210} at 4 -7 \put{462} at 5 -7 \put{924} at 6 -7 \put{1716} at 7 -7 \put{8} at 1 -8 \put{36} at 2 -8 \put{120} at 3 -8 \put{330} at 4 -8 \put{792} at 5 -8 \put{1716} at 6 -8 \put{3432} at 7 -8 \put{6435} at 8 -8 \put{9} at 1 -9 \put{45} at 2 -9 \put{165} at 3 -9 \put{495} at 4 -9 \put{1287} at 5 -9 \put{3003} at 6 -9 \put{6435} at 7 -9 \put{12870} at 8 -9 \put{24310} at 9 -9 \put{${\smallmatrix n\endsmallmatrix}$} at -2 1 \put{${\smallmatrix 0\endsmallmatrix}$} at -2 0 \put{${\smallmatrix 1\endsmallmatrix}$} at -2 -1 \put{${\smallmatrix 2\endsmallmatrix}$} at -2 -2 \put{${\smallmatrix 3\endsmallmatrix}$} at -2 -3 \put{${\smallmatrix 4\endsmallmatrix}$} at -2 -4 \put{${\smallmatrix 5\endsmallmatrix}$} at -2 -5 \put{${\smallmatrix 6\endsmallmatrix}$} at -2 -6 \put{${\smallmatrix 7\endsmallmatrix}$} at -2 -7 \put{${\smallmatrix 8\endsmallmatrix}$} at -2 -8 \put{${\smallmatrix 9\endsmallmatrix}$} at -2 -9 \plot -2 1.5 -1.5 1 / \put{${\smallmatrix s\endsmallmatrix}$} at -1.5 1.5 \put{${\smallmatrix 0\endsmallmatrix}$} at 0 1.5 \put{${\smallmatrix 1\endsmallmatrix}$} at 1 1.5 \put{${\smallmatrix 2\endsmallmatrix}$} at 2 1.5 \put{${\smallmatrix 3\endsmallmatrix}$} at 3 1.5 \put{${\smallmatrix 4\endsmallmatrix}$} at 4 1.5 \put{${\smallmatrix 5\endsmallmatrix}$} at 5 1.5 \put{${\smallmatrix 6\endsmallmatrix}$} at 6 1.6 \put{${\smallmatrix 7\endsmallmatrix}$} at 7 1.5 \put{${\smallmatrix 8\endsmallmatrix}$} at 8 1.5 \put{${\smallmatrix 9\endsmallmatrix}$} at 9 1.5 \put{} at 0 -10 \put{sum} [r] at 11.5 1.5 \put{} [r] at 11.5 0 \put{2} [r] at 11.5 -1 \put{6} [r] at 11.5 -2 \put{20} [r] at 11.5 -3 \put{70} [r] at 11.5 -4 \put{252} [r] at 11.5 -5 \put{924} [r] at 11.5 -6 \put{3432} [r] at 11.5 -7 \put{12870} [r] at 11.5 -8 \put{48620} [r] at 11.5 -9 \endpicture}$$ \[triangle3\][**The triangle of type $\mathbb D$; this is now** ]{} $$\hbox{\beginpicture \setcoordinatesystem units <1cm,.45cm> \put{$a_s(\mathbb D_n) \,= \left\{\begin{matrix} \cr\cr\cr\end{matrix}\right.$} [l] at 4.2 -1.4 \put{$\left[ \smallmatrix{n+s-2}\cr s \endsmallmatrix\right]$ \ for $0\le s <n; $} [l] at 6.4 -.3 \put{$ \left[ \smallmatrix{2n-2}\cr n-2 \endsmallmatrix\right]$\qquad for $s=n$.} [l] at 6.4 -2.5 \multiput{1} at 0 -2 0 -3 0 -4 0 -5 0 -6 0 -7 0 -8 0 -9 / \multiput{$\cdot$} at 0 0 0 -1 1 -1 / \multiput{$\cdot$} at 11.4 0 11.4 -1 / \put{} at 1 -1 \put{2} at 1 -2 \put{1} at 2 -2 \put{3} at 1 -3 \put{5} at 2 -3 \put{5} at 3 -3 \put{4} at 1 -4 \put{9} at 2 -4 \put{16} at 3 -4 \put{20} at 4 -4 \put{5} at 1 -5 \put{14} at 2 -5 \put{30} at 3 -5 \put{55} at 4 -5 \put{77} at 5 -5 \put{6} at 1 -6 \put{20} at 2 -6 \put{50} at 3 -6 \put{105} at 4 -6 \put{196} at 5 -6 \put{294} at 6 -6 \put{7} at 1 -7 \put{27} at 2 -7 \put{77} at 3 -7 \put{182} at 4 -7 \put{378} at 5 -7 \put{714} at 6 -7 \put{1122} at 7 -7 \put{8} at 1 -8 \put{35} at 2 -8 \put{112} at 3 -8 \put{294} at 4 -8 \put{672} at 5 -8 \put{1386} at 6 -8 \put{2640} at 7 -8 \put{4290} at 8 -8 \put{9} at 1 -9 \put{44} at 2 -9 \put{156} at 3 -9 \put{450} at 4 -9 \put{1122} at 5 -9 \put{2508} at 6 -9 \put{5148} at 7 -9 \put{9867} at 8 -9 \put{16445} at 9 -9 \put{${\smallmatrix n\endsmallmatrix}$} at -2 1 \put{${\smallmatrix 0\endsmallmatrix}$} at -2 0 \put{${\smallmatrix 1\endsmallmatrix}$} at -2 -1 \put{${\smallmatrix 2\endsmallmatrix}$} at -2 -2 \put{${\smallmatrix 3\endsmallmatrix}$} at -2 -3 \put{${\smallmatrix 4\endsmallmatrix}$} at -2 -4 \put{${\smallmatrix 5\endsmallmatrix}$} at -2 -5 \put{${\smallmatrix 6\endsmallmatrix}$} at -2 -6 \put{${\smallmatrix 7\endsmallmatrix}$} at -2 -7 \put{${\smallmatrix 8\endsmallmatrix}$} at -2 -8 \put{${\smallmatrix 9\endsmallmatrix}$} at -2 -9 \plot -2 1.5 -1.5 1 / \put{${\smallmatrix s\endsmallmatrix}$} at -1.5 1.5 \put{${\smallmatrix 0\endsmallmatrix}$} at 0 1.5 \put{${\smallmatrix 1\endsmallmatrix}$} at 1 1.5 \put{${\smallmatrix 2\endsmallmatrix}$} at 2 1.5 \put{${\smallmatrix 3\endsmallmatrix}$} at 3 1.5 \put{${\smallmatrix 4\endsmallmatrix}$} at 4 1.5 \put{${\smallmatrix 5\endsmallmatrix}$} at 5 1.5 \put{${\smallmatrix 6\endsmallmatrix}$} at 6 1.6 \put{${\smallmatrix 7\endsmallmatrix}$} at 7 1.5 \put{${\smallmatrix 8\endsmallmatrix}$} at 8 1.5 \put{${\smallmatrix 9\endsmallmatrix}$} at 9 1.5 \put{} at 0 -10 \put{sum} [r] at 11.5 1.5 \put{} [r] at 11.5 0 \put{} [r] at 11.5 -1 \put{4} [r] at 11.5 -2 \put{14} [r] at 11.5 -3 \put{50} [r] at 11.5 -4 \put{182} [r] at 11.5 -5 \put{672} [r] at 11.5 -6 \put{2508} [r] at 11.5 -7 \put{9438} [r] at 11.5 -8 \put{35750} [r] at 11.5 -9 \setdashes <1mm> \plot 1.5 -1.5 1.5 -2.5 2.5 -2.5 2.5 -3.5 3.5 -3.5 3.5 -4.5 4.5 -4.5 4.5 -5.5 5.5 -5.5 5.5 -6.5 6.5 -6.5 6.5 -7.5 7.5 -7.5 7.5 -8.5 8.5 -8.5 8.5 -9.5 9.5 -9.5 9.5 -10 / \endpicture}$$ \[triangle1-S\][**The sheared Catalan triangle** ]{} $$\hbox{\beginpicture \setcoordinatesystem units <.9cm,.45cm> \put{$\dbinom t s- \dbinom t {s-1} = \dfrac{t-2s+1}{t-s-1}\dbinom ts$} at 7 -2 \multiput{1} at 0 0 0 -1 0 -2 0 -3 0 -4 0 -5 0 -6 0 -7 0 -8 0 -9 / \put{1} at 1 -1 \put{2} at 1 -2 \put{} at 2 -2 \put{3} at 1 -3 \put{2} at 2 -3 \put{} at 3 -3 \put{4} at 1 -4 \put{5} at 2 -4 \put{} at 3 -4 \put{} at 4 -4 \put{5} at 1 -5 \put{9} at 2 -5 \put{5} at 3 -5 \put{} at 4 -5 \put{} at 5 -5 \put{6} at 1 -6 \put{14} at 2 -6 \put{14} at 3 -6 \put{} at 4 -6 \put{} at 5 -6 \put{} at 6 -6 \put{7} at 1 -7 \put{20} at 2 -7 \put{28} at 3 -7 \put{14} at 4 -7 \put{} at 5 -7 \put{} at 6 -7 \put{} at 7 -7 \put{8} at 1 -8 \put{27} at 2 -8 \put{48} at 3 -8 \put{42} at 4 -8 \put{} at 5 -8 \put{} at 6 -8 \put{} at 7 -8 \put{} at 8 -8 \put{9} at 1 -9 \put{35} at 2 -9 \put{75} at 3 -9 \put{90} at 4 -9 \put{42} at 5 -9 \put{} at 6 -9 \put{} at 7 -9 \put{} at 8 -9 \put{} at 9 -9 \put{${\smallmatrix t\endsmallmatrix}$} at -2 1 \put{${\smallmatrix 0\endsmallmatrix}$} at -2 0 \put{${\smallmatrix 1\endsmallmatrix}$} at -2 -1 \put{${\smallmatrix 2\endsmallmatrix}$} at -2 -2 \put{${\smallmatrix 3\endsmallmatrix}$} at -2 -3 \put{${\smallmatrix 4\endsmallmatrix}$} at -2 -4 \put{${\smallmatrix 5\endsmallmatrix}$} at -2 -5 \put{${\smallmatrix 6\endsmallmatrix}$} at -2 -6 \put{${\smallmatrix 7\endsmallmatrix}$} at -2 -7 \put{${\smallmatrix 8\endsmallmatrix}$} at -2 -8 \put{${\smallmatrix 9\endsmallmatrix}$} at -2 -9 \plot -2 1.5 -1.5 1 / \put{${\smallmatrix s\endsmallmatrix}$} at -1.5 1.5 \put{${\smallmatrix 0\endsmallmatrix}$} at 0 1.5 \put{${\smallmatrix 1\endsmallmatrix}$} at 1 1.5 \put{${\smallmatrix 2\endsmallmatrix}$} at 2 1.5 \put{${\smallmatrix 3\endsmallmatrix}$} at 3 1.5 \put{${\smallmatrix 4\endsmallmatrix}$} at 4 1.5 \put{${\smallmatrix 5\endsmallmatrix}$} at 5 1.5 \put{${\smallmatrix 6\endsmallmatrix}$} at 6 1.6 \put{${\smallmatrix 7\endsmallmatrix}$} at 7 1.5 \put{${\smallmatrix 8\endsmallmatrix}$} at 8 1.5 \put{${\smallmatrix 9\endsmallmatrix}$} at 9 1.5 \plot 5.5 -10 5.5 -8.5 4.5 -8.5 4.5 -6.5 3.5 -6.5 3.5 -4.5 2.5 -4.5 2.5 -2.5 1.5 -2.5 1.5 -0.5 0.5 -0.5 0.5 .5 / \setdots <.5mm> \plot 0 0 1 -1 / \plot 0 -1 2 -3 / \plot 0 -2 3 -5 / \plot 0 -3 4 -7 / \plot 0 -4 5 -9 / \plot 0 -5 5 -10 / \plot 0 -6 4 -10 / \plot 0 -7 3 -10 / \plot 0 -8 2 -10 / \plot 0 -9 1 -10 / \put{} at 0 -10 \endpicture}$$ \[triangle2-S\][**The Pascal triangle , left of the staircase line is the increasing part**]{} $$\hbox{\beginpicture \setcoordinatesystem units <.9cm,.45cm> \put{$\dbinom t s$} at 7 -2 \multiput{1} at 0 0 0 -1 0 -2 0 -3 0 -4 0 -5 0 -6 0 -7 0 -8 0 -9 / \put{1} at 1 -1 \put{2} at 1 -2 \put{1} at 2 -2 \put{3} at 1 -3 \put{3} at 2 -3 \put{1} at 3 -3 \put{4} at 1 -4 \put{6} at 2 -4 \put{4} at 3 -4 \put{1} at 4 -4 \put{5} at 1 -5 \put{10} at 2 -5 \put{10} at 3 -5 \put{5} at 4 -5 \put{1} at 5 -5 \put{6} at 1 -6 \put{15} at 2 -6 \put{20} at 3 -6 \put{15} at 4 -6 \put{6} at 5 -6 \put{1} at 6 -6 \put{7} at 1 -7 \put{21} at 2 -7 \put{35} at 3 -7 \put{35} at 4 -7 \put{21} at 5 -7 \put{7} at 6 -7 \put{1} at 7 -7 \put{8} at 1 -8 \put{28} at 2 -8 \put{56} at 3 -8 \put{70} at 4 -8 \put{56} at 5 -8 \put{28} at 6 -8 \put{8} at 7 -8 \put{1} at 8 -8 \put{9} at 1 -9 \put{36} at 2 -9 \put{84} at 3 -9 \put{126} at 4 -9 \put{126} at 5 -9 \put{84} at 6 -9 \put{36} at 7 -9 \put{9} at 8 -9 \put{1} at 9 -9 \put{${\smallmatrix t\endsmallmatrix}$} at -2 1 \put{${\smallmatrix 0\endsmallmatrix}$} at -2 0 \put{${\smallmatrix 1\endsmallmatrix}$} at -2 -1 \put{${\smallmatrix 2\endsmallmatrix}$} at -2 -2 \put{${\smallmatrix 3\endsmallmatrix}$} at -2 -3 \put{${\smallmatrix 4\endsmallmatrix}$} at -2 -4 \put{${\smallmatrix 5\endsmallmatrix}$} at -2 -5 \put{${\smallmatrix 6\endsmallmatrix}$} at -2 -6 \put{${\smallmatrix 7\endsmallmatrix}$} at -2 -7 \put{${\smallmatrix 8\endsmallmatrix}$} at -2 -8 \put{${\smallmatrix 9\endsmallmatrix}$} at -2 -9 \plot -2 1.5 -1.5 1 / \put{${\smallmatrix s\endsmallmatrix}$} at -1.5 1.5 \put{${\smallmatrix 0\endsmallmatrix}$} at 0 1.5 \put{${\smallmatrix 1\endsmallmatrix}$} at 1 1.5 \put{${\smallmatrix 2\endsmallmatrix}$} at 2 1.5 \put{${\smallmatrix 3\endsmallmatrix}$} at 3 1.5 \put{${\smallmatrix 4\endsmallmatrix}$} at 4 1.5 \put{${\smallmatrix 5\endsmallmatrix}$} at 5 1.5 \put{${\smallmatrix 6\endsmallmatrix}$} at 6 1.6 \put{${\smallmatrix 7\endsmallmatrix}$} at 7 1.5 \put{${\smallmatrix 8\endsmallmatrix}$} at 8 1.5 \put{${\smallmatrix 9\endsmallmatrix}$} at 9 1.5 \plot 5.5 -10 5.5 -8.5 4.5 -8.5 4.5 -6.5 3.5 -6.5 3.5 -4.5 2.5 -4.5 2.5 -2.5 1.5 -2.5 1.5 -0.5 0.5 -0.5 0.5 0.5 / \setdots <.5mm> \plot 0 0 1 -1 / \plot 0 -1 2 -3 / \plot 0 -2 3 -5 / \plot 0 -3 4 -7 / \plot 0 -4 5 -9 / \plot 0 -5 5 -10 / \plot 0 -6 4 -10 / \plot 0 -7 3 -10 / \plot 0 -8 2 -10 / \plot 0 -9 1 -10 / \put{} at 0 -10 \endpicture}$$ \[triangle3-S\][**The Lucas triangle , left of the staircase line is the increasing part**]{} $$\hbox{\beginpicture \setcoordinatesystem units <.9cm,.45cm> \put{$\left[\begin{matrix} t\cr s \end{matrix}\right]$} at 7 -2 \multiput{1} at 0 -1 0 -2 0 -3 0 -4 0 -5 0 -6 0 -7 0 -8 0 -9 / \put{$\cdot$} at 0 0 \put{2} at 1 -1 \put{3} at 1 -2 \put{2} at 2 -2 \put{4} at 1 -3 \put{5} at 2 -3 \put{2} at 3 -3 \put{5} at 1 -4 \put{9} at 2 -4 \put{7} at 3 -4 \put{2} at 4 -4 \put{6} at 1 -5 \put{14} at 2 -5 \put{16} at 3 -5 \put{9} at 4 -5 \put{2} at 5 -5 \put{7} at 1 -6 \put{20} at 2 -6 \put{30} at 3 -6 \put{25} at 4 -6 \put{11} at 5 -6 \put{2} at 6 -6 \put{8} at 1 -7 \put{27} at 2 -7 \put{50} at 3 -7 \put{55} at 4 -7 \put{36} at 5 -7 \put{13} at 6 -7 \put{2} at 7 -7 \put{9} at 1 -8 \put{35} at 2 -8 \put{77} at 3 -8 \put{105} at 4 -8 \put{91} at 5 -8 \put{49} at 6 -8 \put{15} at 7 -8 \put{2} at 8 -8 \put{10} at 1 -9 \put{44} at 2 -9 \put{112} at 3 -9 \put{182} at 4 -9 \put{196} at 5 -9 \put{140} at 6 -9 \put{64} at 7 -9 \put{17} at 8 -9 \put{2} at 9 -9 \put{${\smallmatrix t\endsmallmatrix}$} at -2 1 \put{${\smallmatrix 0\endsmallmatrix}$} at -2 0 \put{${\smallmatrix 1\endsmallmatrix}$} at -2 -1 \put{${\smallmatrix 2\endsmallmatrix}$} at -2 -2 \put{${\smallmatrix 3\endsmallmatrix}$} at -2 -3 \put{${\smallmatrix 4\endsmallmatrix}$} at -2 -4 \put{${\smallmatrix 5\endsmallmatrix}$} at -2 -5 \put{${\smallmatrix 6\endsmallmatrix}$} at -2 -6 \put{${\smallmatrix 7\endsmallmatrix}$} at -2 -7 \put{${\smallmatrix 8\endsmallmatrix}$} at -2 -8 \put{${\smallmatrix 9\endsmallmatrix}$} at -2 -9 \plot -2 1.5 -1.5 1 / \put{${\smallmatrix s\endsmallmatrix}$} at -1.5 1.5 \put{${\smallmatrix 0\endsmallmatrix}$} at 0 1.5 \put{${\smallmatrix 1\endsmallmatrix}$} at 1 1.5 \put{${\smallmatrix 2\endsmallmatrix}$} at 2 1.5 \put{${\smallmatrix 3\endsmallmatrix}$} at 3 1.5 \put{${\smallmatrix 4\endsmallmatrix}$} at 4 1.5 \put{${\smallmatrix 5\endsmallmatrix}$} at 5 1.5 \put{${\smallmatrix 6\endsmallmatrix}$} at 6 1.6 \put{${\smallmatrix 7\endsmallmatrix}$} at 7 1.5 \put{${\smallmatrix 8\endsmallmatrix}$} at 8 1.5 \put{${\smallmatrix 9\endsmallmatrix}$} at 9 1.5 \plot 5.5 -10 5.5 -8.5 4.5 -8.5 4.5 -6.5 3.5 -6.5 3.5 -4.5 2.5 -4.5 2.5 -2.5 1.5 -2.5 1.5 -0.5 0.5 -0.5 0.5 0 / \setdots <.5mm> \plot 0 -1 2 -3 / \plot 0 -2 3 -5 / \plot 0 -3 4 -7 / \plot 0 -4 5 -9 / \plot 0 -5 5 -10 / \plot 0 -6 4 -10 / \plot 0 -7 3 -10 / \plot 0 -8 2 -10 / \plot 0 -9 1 -10 / \put{} at 0 -10 \endpicture}$$ In the triangle \[triangle3\] of type $\mathbb D$ and in the corresponding Lucas triangle S \[triangle3-S\] some values are left open (this is indicated by a dot). In the Lucas triangle S \[triangle3-S\], this concerns the value at the position $(0,0)$ which could be denoted as $\left[\smallmatrix 0\cr 0 \endsmallmatrix\right]$. The value should be one of the numbers $1$ or $2$ (in OEIS , the number is chosen to be $2$). Note that here we deal with the product $\frac00\binom00$: whereas $\binom 0 0 = 1$ is well-defined, there is the ambiguous fraction $\frac 0 0$. In the triangle \[triangle3\] of type $\mathbb D$, the positions $(0,0), (0,1), (1,1)$ are left open, since the series of Dynkin diagrams $\mathbb D_n$ starts with $n = 2$ (but see ); by definition $\mathbb D_2 = \mathbb A_1\sqcup \mathbb A_1$ and $\mathbb D_3 = \mathbb A_3$. As a consequence, also the corresponding entries in the summation sequence are missing. **Some observations concerning the triangles $\mathbb A, \mathbb B, \mathbb D$** -------------------------------------------------------------------------------- The sum sequence occurs as a diagonal. In the $\mathbb A$-triangle, the sum sequence is the same sequence as the main diagonal (and these are just the Catalan numbers): $$a(\mathbb A_n) = a_{n+1}(\mathbb A_{n+1}).$$ In the $\mathbb B$-triangle, the sum sequence is the same sequence as the second diagonal $$a(\mathbb B_n) = a_{n}(\mathbb B_{n+1}). \ $$ In the $\mathbb D$-triangle, the sum sequence is the same sequence as the fourth diagonal $$a(\mathbb D_n) = a_{n-1}(\mathbb D_{n+2}).$$ The main diagonal uses the same sequence as one of the other diagonals. In the $\mathbb A$-triangle, this concerns the main diagonal and the second diagonal: $$a_n(\mathbb A_n) = a_{n-1}(\mathbb A_{n}). \ $$ In the $\mathbb B$-triangle, this concerns the main diagonal and the second diagonal: $$a_n(\mathbb B_n) = a_{n-1}(\mathbb B_{n+1}).$$ In the $\mathbb D$-triangle, this concerns the main diagonal and the fifth diagonal: $$a_n(\mathbb D_n) = a_{n-2}(\mathbb D_{n+2}).$$ It may be of interest to exhibit explicit bijections between the corresponding sets of support-tilting modules. It seems that only in the case $\mathbb A$, this can be done easily (see Remark \[type-a\]). **Comparison between the Lucas triangle and the $\mathbb D$-triangle** ---------------------------------------------------------------------- The difference between the number $\left[{\begin{matrix} 2n-2\cr n\end{matrix}}\right]$ and $\left[{\begin{matrix} 2n-2\cr n-2\end{matrix}}\right]$ seems to be of interest: $$\left[{\begin{matrix} 2n-2\cr n\end{matrix}}\right] - \left[{\begin{matrix} 2n-2\cr n-2\end{matrix}}\right] = \frac{1}{n}\binom{2n-2}{n-1}.$$ This means the following: \[comparision\] $$\left[\begin{matrix} 2n-2\cr n \end{matrix}\right] - a_n(\mathbb D_n) = a_{n-1}(\mathbb A_{n-1}).$$ We show that $$\frac{3n-4}{n}\binom{2n-2}{n-2} +\frac{1}{n}\binom{2n-2}{n-1} \ =\ \frac{3n-2}{2n-2}\binom{2n-2}{n}$$ We rewrite $$\begin{aligned} \binom{2n-2}{n-2} &=& \frac{n}{2n-2} \binom{2n-2}{n}, \\ \binom{2n-2}{n-1} &=& \frac{n}{n-1}\binom{2n-2}{n}. \end{aligned}$$ The assertion now follows from the equality $$\frac{3n-4}{n}\cdot \frac{n}{2n-2}\ +\ \frac{1}{n}\cdot \frac{n}{n-1}\ =\ \frac{3n-2}{2n-2}.$$ Here is a table of these numbers $$\hbox{\beginpicture \setcoordinatesystem units <1cm,.45cm> \put{${\smallmatrix n\endsmallmatrix}$} at -2 1 \put{$\left[{\begin{matrix} 2n-2\cr n\end{matrix}}\right]$} [r] at 0.3 1 \put{$\left[{\begin{matrix} 2n-2\cr n-2\end{matrix}}\right]$} [r] at 2.4 1 \put{$\dfrac{1}{n}\dbinom{2n-2}{n-1}$} [r] at 4.9 1 \put{${\smallmatrix 2\endsmallmatrix}$} at -2 -2 \put{${\smallmatrix 3\endsmallmatrix}$} at -2 -3 \put{${\smallmatrix 4\endsmallmatrix}$} at -2 -4 \put{${\smallmatrix 5\endsmallmatrix}$} at -2 -5 \put{${\smallmatrix 6\endsmallmatrix}$} at -2 -6 \put{${\smallmatrix 7\endsmallmatrix}$} at -2 -7 \put{${\smallmatrix 8\endsmallmatrix}$} at -2 -8 \put{${\smallmatrix 9\endsmallmatrix}$} at -2 -9 \put{$2$} [r] at 0 -2 \put{$7$} [r] at 0 -3 \put{$25$} [r] at 0 -4 \put{$91$} [r] at 0 -5 \put{$336$} [r] at 0 -6 \put{$1254$} [r] at 0 -7 \put{$4719$} [r] at 0 -8 \put{$17875$} [r] at 0 -9 \put{$1$} [r] at 2 -2 \put{$5$} [r] at 2 -3 \put{$20$} [r] at 2 -4 \put{$77$} [r] at 2 -5 \put{$294$} [r] at 2 -6 \put{$1122$} [r] at 2 -7 \put{$4290$} [r] at 2 -8 \put{$16445$} [r] at 2 -9 \put{$1$} [r] at 4 -2 \put{$2$} [r] at 4 -3 \put{$5$} [r] at 4 -4 \put{$14$} [r] at 4 -5 \put{$42$} [r] at 4 -6 \put{$132$} [r] at 4 -7 \put{$429$} [r] at 4 -8 \put{$1430$} [r] at 4 -9 \put{} at 0 -10 \endpicture}$$ Proposition \[comparision\] is essentially the modified hook formula for type $\mathbb D$ which will be presented in Proposition \[modified\]: the Lucas triangle uses the hook formula for the whole triangle, whereas the triangle $\mathbb D$ uses the modified hook formula on the subdiagonal. The exceptional cases {#exceptional} ===================== Here are the numbers $a_s(\Delta_n)$ and $a(\Delta_n)$ in the exceptional cases $\mathbb E_6, \mathbb E_7, \mathbb E_8, \mathbb F_4,$ and $\mathbb G_2$ (we add some suitable additional rows in order to stress the induction scheme): $$\hbox{\beginpicture \setcoordinatesystem units <.97cm,.45cm> \put{1} at 0 0 \put{3} at 1 0 \put{4} at 2 0 \put{2} at 3 0 \put{1} at 0 -1 \put{4} at 1 -1 \put{9} at 2 -1 \put{14} at 3 -1 \put{14} at 4 -1 \put{1} at 0 -2 \put{5} at 1 -2 \put{14} at 2 -2 \put{30} at 3 -2 \put{55} at 4 -2 \put{77} at 5 -2 \put{1} at 0 -3 \put{6} at 1 -3 \put{20} at 2 -3 \put{50} at 3 -3 \put{110} at 4 -3 \put{228} at 5 -3 \put{418} at 6 -3 \put{1} at 0 -4 \put{7} at 1 -4 \put{27} at 2 -4 \put{77} at 3 -4 \put{187} at 4 -4 \put{429} at 5 -4 \put{1001} at 6 -4 \put{2431} at 7 -4 \put{1} at 0 -5 \put{8} at 1 -5 \put{35} at 2 -5 \put{112} at 3 -5 \put{299} at 4 -5 \put{728} at 5 -5 \put{1771} at 6 -5 \put{4784} at 7 -5 \put{17342} at 8.1 -5 \put{1} at 0 -7 \put{3} at 1 -7 \put{6} at 2 -7 \put{10} at 3 -7 \put{1} at 0 -8 \put{4} at 1 -8 \put{10} at 2 -8 \put{24} at 3 -8 \put{66} at 4 -8 \put{1} at 0 -10 \put{2} at 1 -10 \put{5} at 2 -10 \put{$$} at -2 2 \put{$\mathbb E_3 = \mathbb A_2\sqcup\mathbb A_1$} [l] at -3 -0 \put{$\mathbb E_4 = \mathbb A_4$} [l] at -3 -1 \put{$\mathbb E_5 = \mathbb D_5$} [l] at -3 -2 \put{$\mathbb E_6$} [l] at -3 -3 \put{$\mathbb E_7$} [l] at -3 -4 \put{$\mathbb E_8$} [l] at -3 -5 \put{$$} [l] at -2.2 -6 \put{$\mathbb B_3$} [l] at -3 -7 \put{$\mathbb F_4$} [l] at -3 -8 \put{$$} [l] at -2.2 -9 \put{$\mathbb G_2$} [l] at -3 -10 \plot -2 1.5 -1.5 1 / \put{${\smallmatrix s\endsmallmatrix}$} at -1.5 1.5 \put{${\smallmatrix 0\endsmallmatrix}$} at 0 1.5 \put{${\smallmatrix 1\endsmallmatrix}$} at 1 1.5 \put{${\smallmatrix 2\endsmallmatrix}$} at 2 1.5 \put{${\smallmatrix 3\endsmallmatrix}$} at 3 1.5 \put{${\smallmatrix 4\endsmallmatrix}$} at 4 1.5 \put{${\smallmatrix 5\endsmallmatrix}$} at 5 1.5 \put{${\smallmatrix 6\endsmallmatrix}$} at 6 1.6 \put{${\smallmatrix 7\endsmallmatrix}$} at 7 1.5 \put{${\smallmatrix 8\endsmallmatrix}$} at 8 1.5 \put{} at 0 -10 \put{sum} [r] at 11 1.5 \put{10} [r] at 11 0 \put{42} [r] at 11 -1 \put{182} [r] at 11 -2 \put{833} [r] at 11 -3 \put{4160} [r] at 11 -4 \put{25080} [r] at 11 -5 \put{} [r] at 11 -6 \put{20} [r] at 11 -7 \put{105} [r] at 11 -8 \put{} [r] at 11 -9 \put{8} [r] at 11 -10 \endpicture}$$ Hereditary artin algebras {#algebras} ========================= The basic setting ----------------- Let $\Lambda$ be a hereditary artin algebra. Since by assumption $\operatorname{Ext}_\Lambda^i = 0$ for $i \ge 2$, we write $\operatorname{Ext}(M,M')$ instead of $\operatorname{Ext}_\Lambda^1(M,M')$. The vertices of the quiver $Q(\Lambda)$ are the isomorphism classes $[S]$ of the simple $\Lambda$-modules $S$ and there is an arrow $[S] \to [S']$ provided $\operatorname{Ext}(S,S') \neq 0$. Note that $Q(\Lambda)$ is finite and directed (the latter means that the simple modules can be labeled $S(i)$ such that the existence of an arrow $[S(i)]\to [S(j)]$ implies that $i > j$). We endow $Q(\Lambda)$ with a valuation as follows: given an arrow $[S] \to [S']$, consider $\operatorname{Ext}(S,S')$ as a left $\operatorname{End}(S)^{\text{op}}$-module and also as a left $\operatorname{End}(S')$-module and put $$v([S],[S']) = (\dim {}_{\operatorname{End}(S)}\operatorname{Ext}(S,S'))(\dim {}_{\operatorname{End}(S')^{\text{op}}}\operatorname{Ext}(S,S'))$$ provided $v([S],[S']) > 1$. Given a vertex $i$ of $Q(\Lambda)$, we let $S(i)=S_\Lambda (i), P(i) = P_\Lambda(i), I(i)=I_\Lambda(i)$, respectively, denote a simple, an indecomposable projective or injective module corresponding to the vertex $i$. If $M$ is a module, the set of vertices of the quiver $Q(\Lambda(M))$ will be called the [*support*]{} of $M$ and $M$ is said to be [*sincere*]{} provided any vertex of $Q(\Lambda)$ belongs to the support of $M$ (thus provided the only idempotent $e \in \Lambda$ with $eM = 0$ is $e=0$). We also will be interested in the corresponding valued graph $\overline Q(\Lambda)$ which is obtained from the valued quiver $Q(\Lambda)$ by replacing the arrows by edges: one says that one [*forgets the orientation*]{} of the quiver. In the special case where $v([S],[S']) = v$ with $v = 2$ or $v = 3$, it is usual to replace the arrow $[S] \longrightarrow [S']$ by a double arrow $[S] \Longrightarrow [S']$ (if $v = 2$) or a similar triple arrow (if $v=3$). Using the bimodule $\operatorname{Ext}(S,S')$ one obtains an embedding either of $\operatorname{End}S$ into $\operatorname{End}S'$, or of $\operatorname{End}S'$ into $\operatorname{End}S$; thus one of the division rings is a subring of the other, with index equal to $v$. One marks the relative size of the endomorphism rings by an additional arrowhead drawn in the middle of the edge, pointing from the larger endomorphism ring to the smaller one (it should be stressed that these inner arrowheads must not be confused with the outer ones). For example, in case there are two simple modules labeled $1$ and $2$ with an arrow $1 \leftarrow 2$ and $v(1,2) = 2$, there are the following two possibilities: $$\hbox{\beginpicture \setcoordinatesystem units <1cm,1cm> \put{\beginpicture \multiput{$\circ$} at 5 0 6 0 / \put{${\smallmatrix 1\endsmallmatrix}$} at 5 -.3 \put{${\smallmatrix 2\endsmallmatrix}$} at 6 -.3 \plot 5.8 0.03 5.2 0.03 / \plot 5.8 -.03 5.2 -.03 / \plot 5.3 0.1 5.1 0 5.3 -.1 / \plot 5.6 0.1 5.4 0 5.6 -.1 / \endpicture} at 0 0 \put{\beginpicture \multiput{$\circ$} at 5 0 6 0 / \put{${\smallmatrix 1\endsmallmatrix}$} at 5 -.3 \put{${\smallmatrix 2\endsmallmatrix}$} at 6 -.3 \plot 5.8 0.03 5.2 0.03 / \plot 5.8 -.03 5.2 -.03 / \plot 5.3 0.1 5.1 0 5.3 -.1 / \plot 5.4 0.1 5.6 0 5.4 -.1 / \endpicture} at 3 0 \endpicture}$$ On the left we see that $\operatorname{End}S(1)$ is a division subring of $\operatorname{End}S(2)$. On the right, $\operatorname{End}S(2)$ is a division subring of $\operatorname{End}S(1)$. (Let us exhibit corresponding algebras: let $K:k$ be a field extension of degree $2$ and consider the algebras $\Lambda = \left[\smallmatrix k & K \cr 0 & K \endsmallmatrix\right]$ and $\Lambda' = \left[\smallmatrix K & K \cr 0 & k \endsmallmatrix\right]$; the left quiver shown above is $Q(\Lambda)$, and the right quiver is $Q(\Lambda')$.) Here are the corresponding valued graphs, which are obtained by forgetting the orientation (thus deleting the outer arrowheads, but not the inner ones): $$\hbox{\beginpicture \setcoordinatesystem units <1cm,1cm> \put{\beginpicture \multiput{$\circ$} at 5 0 6 0 / \put{${\smallmatrix 1\endsmallmatrix}$} at 5 -.3 \put{${\smallmatrix 2\endsmallmatrix}$} at 6 -.3 \plot 5.8 0.03 5.2 0.03 / \plot 5.8 -.03 5.2 -.03 / \plot 5.6 0.1 5.4 0 5.6 -.1 / \endpicture} at 0 0 \put{\beginpicture \multiput{$\circ$} at 5 0 6 0 / \put{${\smallmatrix 1\endsmallmatrix}$} at 5 -.3 \put{${\smallmatrix 2\endsmallmatrix}$} at 6 -.3 \plot 5.8 0.03 5.2 0.03 / \plot 5.8 -.03 5.2 -.03 / \plot 5.4 0.1 5.6 0 5.4 -.1 / \endpicture} at 3 0 \endpicture}$$ They are called $\mathbb B_2$ and $\mathbb C_2$, respectively (observe that there is a difference between $\mathbb B_2$ and $\mathbb C_2$ only if they occur as subgraphs of larger graphs). We recall the following [@[DR1]]. [*A connected hereditary artin algebra $\Lambda$ is representation-finite if and only if $\overline Q(\Lambda)$ is one of the Dynkin diagrams $$\mathbb A_n, \mathbb B_n, \mathbb C_n, \mathbb D_n, \mathbb E_6, \mathbb E_7, \mathbb E_8, \mathbb F_4, \mathbb G_2$$ and in this case the indecomposable $\Lambda$-modules correspond bijectively to the positive roots.*]{} Change of orientation --------------------- We want to show that the number of basic tilting modules is independent of the orientation. We recall that a module is said to be [*basic*]{} provided it is a direct sum of pairwise non-isomorphic indecomposable modules; an artin algebra $\Lambda$ is [*basic*]{} provided the regular representation ${}_\Lambda\Lambda$ is basic. In case $\Lambda$ is the path algebra of a quiver, we may refer to Ladkani [@[L]]. In the case of the tensor algebra of a species (in particular in the case of the path algebra of a quiver), any change of orientation is obtained by applying a sequence of BGP-reflection functors; see [@[DR2]]. For a general hereditary artin algebra $\Lambda$, we have to deal with APR-tilting functors as defined by Auslander, Platzeck and Reiten [@[APR]]. In order to do so, we may assume that $\Lambda$ is basic. We start with a simple projective module $S$, write ${}_\Lambda\Lambda = S\oplus P$ with a projective module $P$, and consider $W = P\oplus \tau^-S$ (where $\tau = \tau_\Lambda$ is the Auslander-Reiten translation in $\operatorname{mod}\Lambda$) and $\Lambda' = (\operatorname{End}W)^{\text{op}}$. Note that $W$ is a tilting module (called an APR-tilting module) and the quiver $Q(\Lambda')$ is obtained from the quiver $Q(\Lambda)$ by changing the orientation of all the arrows which involve the vertex $\omega = [S]$. We let $\Lambda'' = (\operatorname{End}P)^{\text{op}}$ denote the restriction of $\Lambda$ to the quiver $Q''$ obtained from $Q(\Lambda)$ by deleting the vertex $\omega$ and the arrows ending in $\omega$. Of course, $Q''$ is also a subquiver of $Q(\Lambda')$ and $\Lambda''$ is the restriction of $\Lambda'$ to $Q''$ (thus $\Lambda$ is a one-point coextension of $\Lambda''$, whereas $\Lambda'$ is a one-point extension of $\Lambda''$). We let $S'$ denote the simple $\Lambda'$-module with support $\omega$. \[APR\] Let $\Lambda$ be a hereditary artin algebra and $S$ a simple projective module. Let $W$ be the APR-tilting module defined by $S$ and $\Lambda' = (\operatorname{End}W)^{\text{op}}$. Then there is a canonical bijection $\eta$ between the basic tilting $\Lambda$-modules and the basic tilting $\Lambda'$-modules. In order to define $\eta$, we distinguish two cases. First, if $T$ is a basic tilting module such that $S$ is not a direct summand of $T$, let $\eta(T) = \operatorname{Hom}(W,T)$; this is a basic tilting $\Lambda'$-module and $S'$ is not a direct summand of $\eta(T)$. Second, consider a basic tilting $\Lambda$-module of the form $S\oplus T$. Let $T'' = T/U$, where $U$ is the sum of the images of all the maps $S \to T$. Obviously, $T''$ is a basic tilting $\Lambda''$-module which we may consider as a $\Lambda'$-module. We form the universal extension $T'$ of $T''$ using copies of $S'$. Then $T'\oplus S'$ is a basic tilting $\Lambda'$-module (and $S'$ is a direct summand). We may identify the Grothendieck groups $K_0(\Lambda)$ and $K_0(\Lambda')$, using the common factor algebra $\Lambda''$ and identifying the dimension vectors of $S$ and $S'$. Then, in the first case, the dimension vector of $\eta(T)$ is obtained from the dimension vector of $T$ by applying the reflection $\sigma$ defined by $S$. In the second case, the dimension vectors of $T$ and $\eta(T)$ coincide. Actually, here we use twice the internal reflection defined by $S$ in [@[R]], first in the category $\operatorname{mod}\Lambda$, second in the category $\operatorname{mod}\Lambda'$. The combinatorial backbone -------------------------- Let $\Lambda$ be a Dynkin algebra and assume that the vertices of $Q(\Lambda)$ are labeled $1\le i \le n$. Let $P(i) = P_\Lambda(i)$ be indecomposable projective. Since we assume that $\Lambda$ is a Dynkin algebra, there is a natural number $q(i) = q(P(i))$ such that $\tau^{-q(i)}P(i)$ is indecomposable injective; the modules $M(i,u) = \tau^{-u}P(i)$ with $0 \le u \le q(i)$ and $1\le i \le n$ furnish a complete list of the indecomposable $\Lambda$-modules. \[hammock\] Let $\Lambda, \Lambda'$ be Dynkin algebras and assume that the simple modules of both algebras are indexed by $1\le i \le n$. Assume that $q(P_\Lambda(i)) = q(P_{\Lambda'}(i)) = q(i)$ for all $1\le i \le n$. If the support of $M(u,i) = \tau_\Lambda^{-u}P_\Lambda(i)$ and $M'(i,u) = \tau_{\Lambda'}^{-u}P_{\Lambda}(i)$ coincide for all $0 \le u \le q(i)$ and $1\le i \le n$, then $a_s(\Lambda) = a_s(\Lambda')$ for all $s$. We may interpret the numbers $a_s(\Lambda)$ and $a_s(\Lambda')$ as the number of antichains in $\operatorname{mod}\Lambda$ and $\operatorname{mod}\Lambda'$, respectively, which have support-rank $s$. Note that the support of a module $M$ is the set of numbers $1\le i \le n$ such that $\operatorname{Hom}(P(i),M) \neq 0$. Note that $\operatorname{Hom}(M(i,u),M(j,v)) = 0$ if and only if $\operatorname{Hom}(M'(i,u),M'(j,v))$ $ = 0$. Namely, if $u\le v$, the Auslander-Reiten translation (see for example [@[ARS]]) furnishes a group isomorphism $$\begin{aligned} \operatorname{Hom}(M(i,u),M(j,v)) &\simeq& \operatorname{Hom}(M(i,0),M(j,v-u)) \cr &=&\operatorname{Hom}(P_{\Lambda}(i),M(j,v-u)),\end{aligned}$$ and similarly we have $ \operatorname{Hom}(M'(i,u),M'(j,v)) \simeq \operatorname{Hom}(P_{\Lambda'}(i),M'(j,v-u))$. It follows that $\operatorname{Hom}(M(i,u),M(j,v)) = 0$ if and only if $i$ is not in the support of $M(j,v-u)$ if and only if $i$ is not in the support of $M'(j,v-u)$ if and only if $\operatorname{Hom}(M'(i,u),M'(j,v)) = 0$. If $u > v$, then $$\operatorname{Hom}(M(i,0),M(j,v)) \simeq \operatorname{Hom}(M(i,u-v),M(j,0)) = 0,$$ since $M(i,u-v)$ is indecomposable and non-projective, whereas $M(j,0)$ is projective. Similarly, we also have $ \operatorname{Hom}(M'(i,0),M'(j,v)) = 0$. As a consequence we see that given an antichain $A = \{A_1,\dots,A_t\}$ in $\operatorname{mod}\Lambda$, the function $M(i,u)\mapsto M'(i,u)$ yields an antichain $A' = \{A'_1,\dots,A'_t\}$ in $\operatorname{mod}\Lambda'$. Of course, the support-rank of $A$ and $A'$ are the same. This completes the proof. The numbers $a_s(\Lambda)$ depend only on the Dynkin type of $Q(\Lambda)$, not on $\Lambda$ itself. According to Proposition \[hammock\], the numbers $a_s(\Lambda)$ depend only on $Q(\Lambda)$. According to Proposition \[APR\], the orientation of $Q(\Lambda)$ does not play a role. Thus, is $\Lambda$ is of Dynkin type $\Delta$, we write $a_s(\Delta)$ instead of $a_s(\Lambda)$. For all $0\le s \le n$, we have $a_s(\mathbb B_n) = a_s(\mathbb C_n)$. Apply the Proposition \[hammock\] to the algebras $\Lambda$ and $\Lambda'$ with valued quivers $$\hbox{\beginpicture \setcoordinatesystem units <1cm,1cm> \put{\beginpicture \multiput{$\circ$} at 0 0 1 0 2 0 4 0 5 0 / \put{$\cdots$} at 3 0 {\arrow <1.5mm> [0.25,0.75] from 0.8 0 to 0.2 0} {\arrow <1.5mm> [0.25,0.75] from 1.8 0 to 1.2 0} {\arrow <1.5mm> [0.25,0.75] from 2.5 0 to 2.2 0} {\arrow <1.5mm> [0.25,0.75] from 4.8 0 to 4.2 0} \plot 3.8 0 3.5 0 / \put{${\smallmatrix 1\endsmallmatrix}$} at 0 -.3 \put{${\smallmatrix 2\endsmallmatrix}$} at 1 -.3 \put{${\smallmatrix 3\endsmallmatrix}$} at 2 -.3 \put{${\smallmatrix n-2\endsmallmatrix}$} at 4 -.3 \put{${\smallmatrix n-1\endsmallmatrix}$} at 5 -.3 \put{${\smallmatrix n\endsmallmatrix}$} at 6 -.3 \plot 5.8 0.03 5.2 0.03 / \plot 5.8 -.03 5.2 -.03 / \plot 5.3 0.1 5.1 0 5.3 -.1 / \plot 5.4 0.1 5.6 0 5.4 -.1 / \multiput{$\circ$} at 6 0 / \endpicture} at 0 0 \put{\beginpicture \multiput{$\circ$} at 0 0 1 0 2 0 4 0 5 0 / \put{$\cdots$} at 3 0 {\arrow <1.5mm> [0.25,0.75] from 0.8 0 to 0.2 0} {\arrow <1.5mm> [0.25,0.75] from 1.8 0 to 1.2 0} {\arrow <1.5mm> [0.25,0.75] from 2.5 0 to 2.2 0} {\arrow <1.5mm> [0.25,0.75] from 4.8 0 to 4.2 0} \plot 3.8 0 3.5 0 / \put{${\smallmatrix 1\endsmallmatrix}$} at 0 -.3 \put{${\smallmatrix 2\endsmallmatrix}$} at 1 -.3 \put{${\smallmatrix 3\endsmallmatrix}$} at 2 -.3 \put{${\smallmatrix n-2\endsmallmatrix}$} at 4 -.3 \put{${\smallmatrix n-1\endsmallmatrix}$} at 5 -.3 \put{${\smallmatrix n\endsmallmatrix}$} at 6 -.3 \plot 5.8 0.03 5.2 0.03 / \plot 5.8 -.03 5.2 -.03 / \plot 5.3 0.1 5.1 0 5.3 -.1 / \plot 5.6 0.1 5.4 0 5.6 -.1 / \multiput{$\circ$} at 6 0 / \endpicture} at 0 -1 \endpicture}$$ respectively; the first valued quiver is of type $\mathbb B_n$, and the second is of type $\mathbb C_n$. It is well-known (and easy to see) that $q(P_\Lambda(i)) = n\!-\!1 = q(P_{\Lambda'}(i))$ for all $1\le i \le n$ and that the modules $M(i,u)$ and $M'(i,u)$ for $1\le i\le n$ and $0\le u \le n-1$ have the same support. The tilting modules for $\mathbb B_n$ {#tilting} ===================================== We are going to determine the number of tilting modules for the Dynkin algebras of type $\mathbb B_n$; namely we will show that $a_n(\mathbb B_n) = \binom{2n-1}{n-1}$. By induction, we assume knowledge about the representation theory of $\mathbb B_i$ with $i < n$, as well as the calculation of $a_s(\mathbb B_n)$ for $s < n$ as shown in Section \[hook\]. We consider a Dynkin algebra $\Lambda$ with quiver $$\hbox{\beginpicture \setcoordinatesystem units <1cm,1cm> \multiput{$\circ$} at 0 0 1 0 2 0 4 0 5 0 / \put{$\cdots$} at 3 0 {\arrow <1.5mm> [0.25,0.75] from 0.8 0 to 0.2 0} {\arrow <1.5mm> [0.25,0.75] from 1.8 0 to 1.2 0} {\arrow <1.5mm> [0.25,0.75] from 2.5 0 to 2.2 0} {\arrow <1.5mm> [0.25,0.75] from 4.8 0 to 4.2 0} \plot 3.8 0 3.5 0 / \put{${\smallmatrix 1\endsmallmatrix}$} at 0 -.3 \put{${\smallmatrix 2\endsmallmatrix}$} at 1 -.3 \put{${\smallmatrix 3\endsmallmatrix}$} at 2 -.3 \put{${\smallmatrix n-2\endsmallmatrix}$} at 4 -.3 \put{${\smallmatrix n-1\endsmallmatrix}$} at 5 -.3 \put{${\smallmatrix n\endsmallmatrix}$} at 6 -.3 \plot 5.8 0.03 5.2 0.03 / \plot 5.8 -.03 5.2 -.03 / \plot 5.3 0.1 5.1 0 5.3 -.1 / \plot 5.4 0.1 5.6 0 5.4 -.1 / \multiput{$\circ$} at 6 0 / \endpicture}$$ We interpret $a_n(\mathbb B_n)$ as the number of sincere antichains (by definition, an antichain $A = \{A_1,\dots, A_t\}$ is [*sincere*]{} provided the module $\bigoplus A_i$ is sincere) and write it as the sum $$a_n(\mathbb B_n) = u(\mathbb B_n) + v(\mathbb B_n)$$ where $u(\mathbb B_n)$ is the number of antichains with a sincere element, whereas $v(\mathbb B_n)$ is the number of sincere antichains without a sincere element. These two numbers will be calculated separately. The calculation of $u(\mathbb B_n)$ ----------------------------------- We let $w(\mathbb B_n)$ denote the number of antichains which do not contain any injective module. $$w(\mathbb B_n) = a_n(\mathbb B_n).$$ Let $\mathcal W$ be the set of antichains without injective modules and $\mathcal S$ the set of sincere antichains. We want to construct a bijection $\eta:\mathcal S \to \mathcal W$. Note that an element of $\mathcal S$ contains at most one injective module, since the injective modules are pairwise comparable with respect to $\operatorname{Hom}$. If $A\in \mathcal S$ contains no injective module, then let $\eta(A) = A$. If $A\in \mathcal S$ contains the injective module $I(i)$, let $\eta(A)$ be obtained from $A$ by deleting $I(i)$ and note that $\eta(A)$ is no longer sincere (since all the modules $A_j$ in $\eta(A)$ satisfy $\operatorname{Hom}(A_j,I(i)) = 0$). It follows that $\eta$ is an injective map. In order to show that $\eta$ is surjective, assume that $B$ is an antichain in $\mathcal W$. If $B$ is sincere, then it belongs to $\mathcal S$ and by definition $\eta(B) = B$. If $B$ is not sincere, let $i$ be the smallest number such that $i$ is not in the support of $B$. Let $A$ be obtained from $B$ by adding $I(i)$. Then clearly $A$ is sincere and $\eta(A) = B$. \[type-a\] A similar proof applies to the linearly oriented quiver of type $\mathbb A_n$. It yields the formula $$a(\mathbb A_{n-1}) = a_n(\mathbb A_n).$$ Also, instead of looking at antichains which do not contain any injective module, we may consider antichains which do not contain any projective module. Now we are able to determine $u(\mathbb B_n)$. $$u(\mathbb B_n) = a_{n-1}(\mathbb B_n) = \binom{2n-2}{n-1}.$$ Note that the sincere indecomposable representations of $\Lambda$ are the modules $X(i) = \tau^{-n+i}P(i)$ with $1\le i \le n$. The dimension vector of $X(n)$ is $(1,\dots,1)$, whereas for $1\le i <n$, the length of $X(n)$ is $n+i$ and its dimension vector is of the form $(1,\dots,1)+(0,\dots,0,1,\dots,1)$. It is easy to see that $\operatorname{Hom}(X(i),X(j)) \neq 0$ for $i\ge j$, thus any antichain contains at most one $X(i)$. Let $u_{i}(\mathbb B_n)$ be the antichains which contain $X(i)$, thus $$u(\mathbb B_n) = \sum\nolimits_{i=1}^n u_{i}(\mathbb B_n).$$ Let $\mathcal X_i$ be the set of indecomposable modules $M$ such that $\operatorname{Hom}(X(i),M) = 0 = \operatorname{Hom}(M,X(i))$. Thus, the antichains which contain $X(i)$ correspond bijectively to the antichains in $\mathcal X_i$. In general, the set $\mathcal X_i$ consists of three triangles I, II, III: $$\hbox{\beginpicture \setcoordinatesystem units <.4cm,.4cm> \multiput{} at 0 0 27 9.5 / \plot 0 0 9 9 18 0 27 9 / \setdots <1mm> \plot 0 0 18 0 / \plot 9 9 27 9 / \setsolid \plot 4 4 8 0 17 9 / \plot 2 0 4 2 6 0 / \plot 11 9 13 7 15 9 / \plot 10 0 13 3 16 0 / \setshadegrid span <.5mm> \vshade -0.3 -.5 .3 <,z,,> 4 3.7 4.3 <z,z,,> 8 -.3 8.3 <z,z,,> 9 0.7 9.2 <z,z,,> 13 4.7 5.3 <z,z,,> 16.8 .7 9.2 <z,z,,> 18 -.3 9.2 <z,,,> 27.3 9 9.2 / \put{$X(i)$} [l] at 13.5 5 \put{I} at 4 .7 \put{II} at 13 8.3 \put{III} at 13 1.1 \endpicture}$$ The triangle I is the wing at the vertex $\tau^{-1}P(n-i-1)$, the triangle II is the wing at the vertex $\tau^{-n+i-1}P(i+2)$, and the triangle III is the wing at the vertex $\tau^{-n+i+1}P(i-2)$. We also are interested in a larger triangle II$'$ which contains the triangle II as well as $n-i$ additional modules (all being successors of $X(i)$), namely the wing at the vertex $\tau^{-n+i}P(i+1)$. The full subcategory $\mathcal X'$ of all direct sums of indecomposable modules in the wings I, II$'$, III is the thick subcategory with simple objects $$S(2), S(3),\dots, S(n-i+1); \quad \tau^{n-i}P(n); \quad S(n-i+3), \dots, S(n-1).$$ The position of these modules is indicated here by bullets: $$\hbox{\beginpicture \setcoordinatesystem units <.4cm,.4cm> \multiput{} at 0 0 27 9.5 / \plot 0 0 9 9 18 0 27 9 / \setdots <1mm> \plot 0 0 18 0 / \plot 9 9 27 9 / \setsolid \plot 4 4 8 0 17 9 / \plot 2 0 4 2 6 0 / \plot 11 9 14 6 / \setsolid \plot 10 0 13 3 16 0 / \setshadegrid span <.5mm> \vshade 2 0 0 <,z,,> 4 0 2 <z,,,> 6 0 0 / \vshade 11 9 9 <,z,,> 14 6 9 <z,,,> 17 9 9 / \vshade 10 0 0 <,z,,> 13 0 3 <z,,,> 16 0 0 / \put{$X(i)$} [l] at 13.5 5 \put{I} at 4 .9 \put{II$'$} at 14 8 \put{III} at 13 1.1 \multiput{$\bullet$} at 2 0 6 0 10 0 16 0 17 9 / \endpicture}$$ (A full subcategory $\mathcal A$ of $\operatorname{mod}\Lambda$ is called a [*thick*]{} subcategory provided it is closed under kernels, cokernels and extensions (see for example [@[K]]). Note that a thick subcategory is an abelian category, and the inclusion functor $\mathcal A \to \operatorname{mod}\Lambda$ is exact.) The category $\mathcal X'$ is of type $\mathbb B_{n-i} \sqcup \mathbb A_{i-2}$ (the $\mathbb A_{i-2}$-part is given by the triangle III, whereas the $\mathbb B_{n-i}$-part is given by the triangles I and II$'$). Note that the indecomposables in I and II just correspond to the non-injective indecomposables in the $\mathbb B_{n-i}$-part. This shows that $$u_i(\mathbb B_n) = w(\mathbb B_{n-i})a(\mathbb A_{i-2}) = a_{n-i}(\mathbb B_{n-i})a_{i-1}(\mathbb A_{i-1}).$$ In the special cases $i = 1,2,n-1,n$, the same formula holds. Namely, for $i = 1$ and $i=2$, the triangle III is empty, whereas the triangles I and II$'$ together yield a category of type $\mathbb B_{n-i}$. In the cases $i=n-1$ and $i=n$, the triangles I and II are empty, whereas the triangle III yields a category of type $\mathbb A_{i-2}$. Thus we see $$u(\mathbb B_n) = \sum\nolimits_{i=1}^n u_i(\mathbb B_n) = \sum\nolimits_{i=1}^n a_{i-1}(\mathbb A_{i-1})a_{n-i}(\mathbb B_{n-i}).$$ But the latter expression is the recursion formula for $a_{n-1}(\mathbb B_n)$, since the number of support-tilting modules $T$ with support $\{1,2,\dots,n\}\setminus\{i\}$ is just $a_{i-1}(\mathbb A_{i-1})a_{n-i}(\mathbb B_{n-i})$. The calculation of $v(\mathbb B_n)$ ----------------------------------- $$v(\mathbb B_n) = a_{n-2}(\mathbb B_{n+1}) = \binom{2n-2}{n-2}.$$ Let $\mathcal V$ be the set of sincere antichains of $\Lambda$-modules without a sincere element. Let $A = (A_1,\dots, A_r)$ be in $\mathcal V$. Since $A$ is sincere, we may assume that $\operatorname{Hom}(P(1),A_1)\neq 0$. Since $A_1$ is not sincere, we must have $\operatorname{Hom}(P(n),A_1) = 0$, thus $A_1$ is a representation of a Dynkin algebra of type $\mathbb A_{n-1}$ and actually an indecomposable projective representation (also as a $\Lambda$-module), thus $A_1 = P(i)$ for some $i$ with $1\le i < n$. Since an antichain can contain only one indecomposable projective module, we see that $A_1$ is uniquely determined. We let $\mathcal V_i$ denote the sincere antichains $A$ such that $A_1 = P(i)$. For $2\le j \le r$, we have $\operatorname{Hom}(P(i),A_j) = \operatorname{Hom}(A_1,A_j) = 0$. It follows that $(A_2,\dots, A_r)$ is an antichain with support in $[1,i-1]\cup[i+1,n]$. Altogether, we see that any element of $A$ has support either in $[1,i]$ or in $[i+1,n]$. The elements of $A$ with support in $[1,i]$ but different from $A_1$ form an arbitrary antichain with support in $[2,i-1]$, thus the number of elements is $a(\mathbb A_{i-2})$, at least if $i\ge 2$. Note that $a(\mathbb A_{i-2}) = a_{i-1}(\mathbb A_{i-1})$. The elements of $A$ with support in $[i+1,n]$ form a sincere antichain for $\mathbb B_{n-i}$. Thus the number of such antichains is $a_{n-i}(\mathbb B_{n-i})$. This shows that for $i\ge 2$, the set $\mathcal V_i$ has cardinality $a_{i-1}(\mathbb A_{i-1})a_{n-i}(\mathbb B_{n-i})$. This formula holds true also for $i=1$, since the number of elements of $\mathcal V_1$ is $a_{n-1}(\mathbb B_{n-1})$ and $a_0(\mathbb A_0) = 1$. Thus we see that $$\begin{aligned} v(\mathbb B_n) &=& \sum\nolimits_{i=1}^{n-1} a_{i-1}(\mathbb A_{i-1})a_{n-i}(\mathbb B_{n-i}) \cr &=& -a_{n-1}(\mathbb A_{n-1}) +\sum\nolimits_{i=1}^{n} a_{i-1}(\mathbb A_{i-1})a_{n-i}(\mathbb B_{n-i}) \cr &=& -\frac1n\binom{2n-2}{n-1} + \binom{2n-2}{n-1} = \binom{2n-2}{n-2}.\end{aligned}$$ Altogether we see $$u(\mathbb B_n) + v(\mathbb B_n) = \binom{2n-2}{n-1} + \binom{2n-2}{n-2}= \binom{2n-1}{n-1}.$$ The calculation of $v(\mathbb B_n)$ shows the following relationship between the cases $\mathbb A$ and $\mathbb B$: $$a_{n-1}(\mathbb B_n) = a_{n-2}(\mathbb B_{n+1}) + a_{n-1}(\mathbb A_{n-1}).$$ Support-tilting modules: the hook formula {#hook} ========================================= The hook formula ---------------- \[hook-formula\] Let $\Delta = \mathbb A, \mathbb B, \mathbb D, \mathbb E$. Then $$a_s(\Delta_n) = a_s(\Delta_{n-1}) + a_{s-1}(\Delta_n)$$ for all $n\ge m$ and $1\le s \le n-c$, where $m = 1,2,3,4$ and $c=0,1,2,3$ for $\Delta = \mathbb A, \mathbb B, \mathbb D, \mathbb E$, respectively. Here we use the convention that $\mathbb B_1 = \mathbb A_1, \mathbb D_2 = \mathbb A_1\sqcup \mathbb A_1, \mathbb E_3 = \mathbb A_2\sqcup \mathbb A_1, \mathbb E_4 = \mathbb A_4, \mathbb E_5 = \mathbb D_5$. In the triangles \[triangle1\], \[triangle2\], \[triangle3\], as well as in Section \[exceptional\], this equality concerns the following kind of hooks: $$\hbox{\beginpicture \setcoordinatesystem units <.9cm,.9cm> \multiput{} at 0 0 3 3 / \plot 0 0 0 3 3 0 / \plot 0.7 0.7 1.3 0.7 1.3 1.3 1 1.3 1 1 0.7 1 0.7 0.7 / \multiput{$\bullet$} at 0.85 0.85 1.15 1.15 / \put{$\circ$} at 1.15 0.85 \endpicture}$$ The hook formula asserts that the sum of the values at the positions marked by bullets is the value at the position marked by the circle. The various assertions concern the following general situation: up to the choice of an orientation, we deal with an artin algebra $\Lambda$ with the following valued quiver with $n$ vertices: $$\hbox{\beginpicture \setcoordinatesystem units <1cm,1cm> \multiput{} at 0 1 8 -1 / \multiput{$\circ$} at 0 0 1 0 2 0 4 0 5 0 / \put{$\cdots$} at 3 0 {\arrow <1.5mm> [0.25,0.75] from 0.8 0 to 0.2 0} {\arrow <1.5mm> [0.25,0.75] from 1.8 0 to 1.2 0} {\arrow <1.5mm> [0.25,0.75] from 2.5 0 to 2.2 0} {\arrow <1.5mm> [0.25,0.75] from 4.8 0 to 4.2 0} \plot 3.8 0 3.5 0 / \put{${\smallmatrix 1\endsmallmatrix}$} at 0 -.3 \put{${\smallmatrix 2\endsmallmatrix}$} at 1 -.3 \put{${\smallmatrix 3\endsmallmatrix}$} at 2 -.3 \put{${\smallmatrix n-c-1\endsmallmatrix}$} at 4 -.3 \put{${\smallmatrix n-c\endsmallmatrix}$} at 5 -.3 {\arrow <1.5mm> [0.25,0.75] from 5.6 0.3 to 5.2 0.1} {\arrow <1.5mm> [0.25,0.75] from 5.6 -.3 to 5.2 -.1} \setdots <.5mm> \setquadratic \plot 5.4 0 6 .5 7 .5 7.5 0 7 -.5 6 -.5 5.4 0 / \endpicture}$$ on the left, we have a quiver of type $\mathbb A_{n-c}$ with arrows $i \leftarrow i\!+\!1$. The remaining $c$ vertices are in the dotted “cloud” to the right. All arrows between the cloud and the $\mathbb A_{n-c}$–quiver end in the vertex $n-c$. We let $Q'$ denote the valued quiver obtained by deleting the vertex $1$ and the arrow ending in $1$; let $\Lambda'$ be the corresponding factor algebra of $\Lambda$. Here are the cases we are interested in. $$\hbox{\beginpicture \setcoordinatesystem units <1cm,1cm> \put{\beginpicture \multiput{$\circ$} at 0 0 1 0 2 0 4 0 5 0 / \put{$\cdots$} at 3 0 {\arrow <1.5mm> [0.25,0.75] from 0.8 0 to 0.2 0} {\arrow <1.5mm> [0.25,0.75] from 1.8 0 to 1.2 0} {\arrow <1.5mm> [0.25,0.75] from 2.5 0 to 2.2 0} {\arrow <1.5mm> [0.25,0.75] from 4.8 0 to 4.2 0} \plot 3.8 0 3.5 0 / \put{${\smallmatrix 1\endsmallmatrix}$} at 0 -.3 \put{${\smallmatrix 2\endsmallmatrix}$} at 1 -.3 \put{${\smallmatrix 3\endsmallmatrix}$} at 2 -.3 \put{${\smallmatrix n-1\endsmallmatrix}$} at 4 -.3 \put{${\smallmatrix n\endsmallmatrix}$} at 5 -.3 \setdots <1mm> \setquadratic \plot 5.2 0.1 6 .2 7 .2 7.5 0 7 -.2 6 -.2 5.2 -.1 / \put{$\mathbb A_n$} at -1 0 \put{$c = 0$} at 9 0 \endpicture} at 0 1 \put{\beginpicture \multiput{$\circ$} at 0 0 1 0 2 0 4 0 5 0 / \put{$\cdots$} at 3 0 {\arrow <1.5mm> [0.25,0.75] from 0.8 0 to 0.2 0} {\arrow <1.5mm> [0.25,0.75] from 1.8 0 to 1.2 0} {\arrow <1.5mm> [0.25,0.75] from 2.5 0 to 2.2 0} {\arrow <1.5mm> [0.25,0.75] from 4.8 0 to 4.2 0} \plot 3.8 0 3.5 0 / \put{${\smallmatrix 1\endsmallmatrix}$} at 0 -.3 \put{${\smallmatrix 2\endsmallmatrix}$} at 1 -.3 \put{${\smallmatrix 3\endsmallmatrix}$} at 2 -.3 \put{${\smallmatrix n-2\endsmallmatrix}$} at 4 -.3 \put{${\smallmatrix n-1\endsmallmatrix}$} at 5 -.3 \plot 5.8 0.03 5.2 0.03 / \plot 5.8 -.03 5.2 -.03 / \plot 5.3 0.1 5.1 0 5.3 -.1 / \plot 5.4 0.1 5.6 0 5.4 -.1 / \multiput{$\circ$} at 6 0 / \setdots <1mm> \setquadratic \plot 5.2 0.1 6 .2 7 .2 7.5 0 7 -.2 6 -.2 5.2 -.1 / \put{$\mathbb B_n$} at -1 0 \put{$c = 1$} at 9 0 \endpicture} at 0 0 \put{\beginpicture \multiput{$\circ$} at 0 0 1 0 2 0 4 0 5 0 / \put{$\cdots$} at 3 0 {\arrow <1.5mm> [0.25,0.75] from 0.8 0 to 0.2 0} {\arrow <1.5mm> [0.25,0.75] from 1.8 0 to 1.2 0} {\arrow <1.5mm> [0.25,0.75] from 2.5 0 to 2.2 0} {\arrow <1.5mm> [0.25,0.75] from 4.8 0 to 4.2 0} \plot 3.8 0 3.5 0 / \put{${\smallmatrix 1\endsmallmatrix}$} at 0 -.3 \put{${\smallmatrix 2\endsmallmatrix}$} at 1 -.3 \put{${\smallmatrix 3\endsmallmatrix}$} at 2 -.3 \put{${\smallmatrix n-3\endsmallmatrix}$} at 4 -.3 \put{${\smallmatrix n-2\endsmallmatrix}$} at 4.9 -.3 {\arrow <1.5mm> [0.25,0.75] from 5.8 0.5 to 5.2 0.1} {\arrow <1.5mm> [0.25,0.75] from 5.8 -.5 to 5.2 -.1} \multiput{$\circ$} at 6 .5 6 -.5 / \setdots <1mm> \setquadratic \plot 5.2 0.1 6 .8 7 .8 7.5 0 7 -.8 6 -.8 5.2 -.1 / \put{$\mathbb D_n$} at -1 0 \put{$c = 2$} at 9 0 \endpicture} at 0 -1.5 \put{\beginpicture \multiput{$\circ$} at 0 0 1 0 2 0 4 0 5 0 / \put{$\cdots$} at 3 0 {\arrow <1.5mm> [0.25,0.75] from 0.8 0 to 0.2 0} {\arrow <1.5mm> [0.25,0.75] from 1.8 0 to 1.2 0} {\arrow <1.5mm> [0.25,0.75] from 2.5 0 to 2.2 0} {\arrow <1.5mm> [0.25,0.75] from 4.8 0 to 4.2 0} \plot 3.8 0 3.5 0 / \put{${\smallmatrix 1\endsmallmatrix}$} at 0 -.3 \put{${\smallmatrix 2\endsmallmatrix}$} at 1 -.3 \put{${\smallmatrix 3\endsmallmatrix}$} at 2 -.3 \put{${\smallmatrix n-4\endsmallmatrix}$} at 4 -.3 \put{${\smallmatrix n-3\endsmallmatrix}$} at 4.9 -.3 {\arrow <1.5mm> [0.25,0.75] from 5.8 0.5 to 5.2 0.1} {\arrow <1.5mm> [0.25,0.75] from 5.8 -.5 to 5.2 -.1} {\arrow <1.5mm> [0.25,0.75] from 6.8 -.5 to 6.2 -.5} \multiput{$\circ$} at 6 .5 6 -.5 7 -.5 / \setdots <1mm> \setquadratic \plot 5.2 0.1 6 .8 7 .8 7.5 0 7 -.8 6 -.8 5.2 -.1 / \put{} at -1 0 \put{$\mathbb E_n$} at -1 0 \put{$c = 3$} at 9 0 \endpicture} at 0 -3.5 \put{} at 0 -4.7 \endpicture}$$ Let $1\le s \le n-c$. Then $$a_s(\Lambda) = a_s(\Lambda') + a_{s-1}(\Lambda).$$ The support-tilting modules $T$ for $\Lambda$ with $1$ not in the support are just the support-tilting modules for $\Lambda'$. Let $\mathcal S_s(\Lambda;1)$ be the set of the basic support-tilting $\Lambda$-modules $T$ with support-rank $s$ and $\operatorname{Hom}(P(1),T) \neq 0$. Let $\mathcal S_{s-1}(\Lambda)$ be the set of basic support-tilting $\Lambda$-modules $T$ with support-rank $s-1$. We construct a bijection $$\alpha:\mathcal S_s(\Lambda;1) \longrightarrow \mathcal S_{s-1}(\Lambda).$$ This will establish the formula. Let $X$ be an indecomposable representation with support-rank $s \le n-c$ and $\operatorname{Hom}(P(1),X) \neq 0$. Then the support of $X$ is contained in the $\mathbb A_{n-c}$-subquiver, so $X$ is thin and its support is an interval of the form $[1,v]$ with $1\le v \le n-c$ (a module is said to be [*thin*]{} provided the composition factors are pairwise non-isomorphic; in our setting thin indecomposable modules are uniquely determined by the support, thus we may just write $X = [1,v]$). Let $T$ be a module in $\mathcal S_s(\Lambda;1)$. At least one of the indecomposable direct summand of $T$, say $X$, satisfies $\operatorname{Hom}(P(1),X)\neq 0$ and we choose $X = [1,v]$ of largest possible length. We claim that $\operatorname{Hom}(P(w),T) = 0$ for any arrow $v \leftarrow w$. Assume, to the contrary, that there is an indecomposable direct summand $Y$ of $T$ with $\operatorname{Hom}(P(w),Y) \neq 0$. The maximality of $X$ shows that $\operatorname{Hom}(P(1),Y) = 0$. But then $\operatorname{Ext}(Y,X) \neq 0$ contradicts the fact that $T$ has no self-extensions. (Namely, if the support of $X$ and $Y$ is disjoint, then the arrow $v\leftarrow w$ yields directly a non-trivial extension of $X$ by $Y$; if the support of $X$ and $Y$ is not disjoint, then there is a proper non-zero factor module of $X$ which is a proper submodule of $Y$, thus there is a non-zero map $X \to Y$ which is neither injective nor surjective — again we obtain a non-trivial extension of $X$ by $Y$.) Thus the support of $T$ is the disjoint union of the set $\{1,2,\dots,v\}$ and a set $S''$ which does not contain a vertex $w$ with an arrow $v\leftarrow w$. The indecomposable direct summands of $T$ with support in $\{1,2,\dots,v\}$ yield a tilting module for this $\mathbb A_v$-quiver, and $X$ is the indecomposable projective-injective representation of this $\mathbb A_v$-quiver. Deleting $X$ from this tilting module, we obtain a support-tilting representation of $\mathbb A_v$ with support-rank $v-1$. Thus if we write $T = X\oplus T'$, then $T'$ is a support-tilting $\Lambda$-module with support-rank $s-1$ (namely, it is the direct sum of a support-tilting module with support properly contained in $\{1,2,\dots,v\}$ and a support-tilting module with support $S''$). We define $\alpha(T) = T'$; this yields the map $$\alpha:\mathcal S_s(\Lambda;1) \longrightarrow \mathcal S_{s-1}(\Lambda)$$ we are looking for. It remains to be shown that $\alpha$ is surjective and that we can recover $T$ from $\alpha(T)$. Thus, let $T'$ be in $\mathcal S_{s-1}(\Lambda)$. Then there are at least $c+1$ vertices outside of the support of $T'$. Case 1: These are the vertices in the cloud and precisely one additional vertex, say $i$ (with $1\le i \le n-c$). Note that in this case $s = n-c$. Let $T = T'\oplus [1,n-c]$. Since $T'$ is a support-tilting module of $\mathbb A_{n-c}$ with support-rank $n-c-1$ and $[1,n-c]$ is the indecomposable projective-injective representation of $\mathbb A_{n-c}$, we see that $T = T'\oplus[1,n-c]$ is a tilting module for $\mathbb A_{n-c}$. Case 2: At least two vertices between $1$ and $n-c$ do not belong to $\operatorname{Supp}T'$, say let $i<j$ be the smallest such numbers. Then let $T = T'\oplus [1,j-1]$. The modified hook formula ------------------------- \[modified\] $$\begin{aligned} a_{n-1}(\mathbb D_n) &=& a_{n-1}(\mathbb D_{n-1}) + a_{n-2}(\mathbb D_n) + a_{n-2}(\mathbb A_{n-2}), \\ a_{n-2}(\mathbb E_n) &=& a_{n-2}(\mathbb E_{n-1}) + a_{n-3}(\mathbb E_n) + a_{n-3}(\mathbb A_{n-3}).\end{aligned}$$ Again, we consider a general setting, namely we consider an artin algebra $\Lambda$ with the following valued quiver with $n$ vertices and we assume that $c \ge 2$: $$\hbox{\beginpicture \setcoordinatesystem units <1cm,1cm> \multiput{} at 0 1 8 -1 / \multiput{$\circ$} at 0 0 1 0 2 0 4 0 5 0 6 0.5 6 -0.5 / \put{$\cdots$} at 3 0 {\arrow <1.5mm> [0.25,0.75] from 0.8 0 to 0.2 0} {\arrow <1.5mm> [0.25,0.75] from 1.8 0 to 1.2 0} {\arrow <1.5mm> [0.25,0.75] from 2.5 0 to 2.2 0} {\arrow <1.5mm> [0.25,0.75] from 4.8 0 to 4.2 0} \plot 3.8 0 3.5 0 / \put{${\smallmatrix 1\endsmallmatrix}$} at 0 -.3 \put{${\smallmatrix 2\endsmallmatrix}$} at 1 -.3 \put{${\smallmatrix 3\endsmallmatrix}$} at 2 -.3 \put{${\smallmatrix n-c-1\endsmallmatrix}$} at 3.9 -.3 \put{${\smallmatrix n-c\endsmallmatrix}$} at 4.9 -.3 \put{${\smallmatrix n-c+1\endsmallmatrix}$} at 6 .8 \put{${\smallmatrix n-c+2\endsmallmatrix}$} at 6 -.8 {\arrow <1.5mm> [0.25,0.75] from 5.8 0.4 to 5.2 0.1} {\arrow <1.5mm> [0.25,0.75] from 5.8 -.4 to 5.2 -.1} \setdots <.5mm> \setquadratic \plot 5.4 0 6 .5 7 .5 7.5 0 7 -.5 6 -.5 5.4 0 / \endpicture}$$ On the left, we have a quiver of type $\mathbb A_{n-c}$ with arrows $i \leftarrow i+1$, and the remaining $c$ vertices are in the dotted “cloud” to the right. There are precisely two vertices in the cloud, namely $n-c+1$ and $n-c+2$ with arrows $n-c \leftarrow n-c+1$ and $n-c \leftarrow n-c+2$ and there is no other arrows between the cloud and the $\mathbb A_{n-c}$ quiver. Again, we let $Q'$ denote the valued quiver obtained by deleting the vertex $1$ and the arrow ending in $1$ and by $\Lambda'$ the corresponding factor algebra of $\Lambda$ and we show the following: \[mod-lemma\] $$a_{n-c+1}(\Lambda) = a_{n-c+1}(\Lambda') + a_{n-c}(\Lambda) + a_{n-c}(\mathbb A_{n-c}).$$ The proof follows closely the proof of Proposition \[hook-formula\]. The support-tilting modules $T$ for $\Lambda$ with $1$ not in the support are just the support-tilting modules for $\Lambda'$. We construct a surjection $\alpha$ from the set $\mathcal S_{n-c+1}(\Lambda;1)$ of the support-tilting $\Lambda$-modules $T$ with support-rank $n-c+1$ and $\operatorname{Hom}(P(1),T) \neq 0$ onto the set $\mathcal S_{n-c}(\Lambda)$ of support-tilting $\Lambda$-modules $T$ with support-rank $n-c$. In the present setting, $\alpha$ will not be injective, but will be a double cover: pairs in $\mathcal S(\Lambda;1)$ are identified by $\alpha$; the number of such pairs will be just $a_{n-c}(\mathbb A_{n-c})$. As above, one shows that any module $T$ in $\mathcal S_{n-c+1}(\Lambda;1)$ is of the form $T = X\oplus T'$ where $X$ is indecomposable, $\operatorname{Hom}(P(1),X)\neq 0$ and $X$ is of maximal possible length. Note that the support of $X$ is contained either in $\{1,2,\dots,n-c+1\}$ or in $\{1,2,\dots,n-c,n-c+2\}$. In particular, $X$ is uniquely determined (since the support of $T$ cannot contain all the vertices $1,2,\dots,n-c+2$). As above, the mapping $\alpha$ will be the deletion of the summand $X$. Let $Z$ be the indecomposable module with support $\{1,2,\dots,n-c+1\}$ and $Z'$ the indecomposable module with support $\{1,2,\dots,n-c,n-c+2\}$. Starting with a tilting module $T'$ for $\mathbb A_{n-c}$, we may form the direct sums $Z\oplus T'$ and $Z'\oplus T'$. Then these are elements of $\mathcal S_{n-c+1}(\Lambda;1)$, both of which are mapped under $\alpha$ to the same module $T'$. These are the $a_{n-c}(\mathbb A_{n-c})$ pairs of elements of $\mathcal S(\Lambda;1)$ which are identified by $\alpha$. It follows that $\mathcal S(\Lambda;1)$ has cardinality $a_{n-c}(\Lambda) + a_{n-c}(\mathbb A_{n-c})$. The two assertions of Proposition \[modified\] are special cases of Lemma \[mod-lemma\]. For the first assertion, $\Lambda$ is of type $\mathbb D_n$, $\Lambda'$ of type $\mathbb D_{n-1}$, and $c = 2$. For the second assertion, $\Lambda$ is of type $\mathbb E_n$, $\Lambda'$ of type $\mathbb E_{n-1}$ and $c = 3$. For another proof of the modified hook formula, see Hubery [@[H]]. \[hook-cor\] $$a_{n-1}(\mathbb D_n) = \left[ \begin{matrix} 2n-3\cr n-1\end{matrix} \right].$$ We start with the previous observation $$\begin{aligned} a_{n-1}(\mathbb D_n) &=&a_{n-1}(\mathbb D_{n-1}) + a_{n-2}(\mathbb D_n) + a_{n-2}(\mathbb A_{n-2}) \\ &=& \frac{3n-7}{2n-4}\binom{2n-4}{n-3} +\frac{3n-6}{2n-4}\binom{2n-4}{n-2} +\frac{1}{n-1}\binom{2n-4}{n-2}.\end{aligned}$$ Write $$\begin{aligned} \binom{2n-4}{n-3} &=& \frac{n-2}{2n-3}\binom{2n-3}{n-1}, \\ \binom{2n-4}{n-2} &=& \frac{n-1}{2n-3}\binom{2n-3}{n-1}.\end{aligned}$$ One easily shows that $$\frac{3n-7}{2n-4}\cdot \frac{n-2}{2n-3} + \frac{3n-6}{2n-4}\cdot\frac{n-1}{2n-3} + \frac1{n-1}\cdot\frac{n-1}{2n-3}\ =\ \frac{3n-4}{2n-3}.$$ As a consequence, we get $$\begin{aligned} a_{n-1}(\mathbb D_n) &=&\frac{3n-4}{2n-3}\binom{2n-3}{n-1} \\ &=& \left[\begin{matrix} 2n-3\cr n-1\end{matrix} \right].\end{aligned}$$ Summation formulas {#summation} ================== An immediate consequence of the previous section is the following assertion: Let $\Delta = \mathbb A$, or $\mathbb B$ and $n\ge 0$, or $\Delta = \mathbb D$ and $n\ge 2$. If $1\le s \le n-1$, then $$\sum\nolimits_{i=0}^s a_i(\Delta_n) = a_s(\Delta_{n+1}).$$ We use induction. For $s = 0$ both sides are equal to $1$. For $s \ge 1$ we have $$\begin{aligned} \sum\nolimits_{i=0}^s a_i(\Delta_n) &=& a_s(\Delta_n) + \sum\nolimits_{i=0}^{s-1} a_i(\Delta_n) \\ &=& a_s(\Delta_n) + a_{s-1}(\Delta_{n+1}) \\ &=& a_s(\Delta_{n+1}),\end{aligned}$$ the last equality being the hook formula. Let $\Delta = \mathbb A, \mathbb B$ or $\mathbb D$. Then $$a(\Delta_n) = a_n(\Delta_n)+a_{n-1}(\Delta_{n+1})$$ [**Case $\mathbb A_n$**]{} $$a(\mathbb A_n) = \frac 1{n+1}\binom{2n}{n}+ \frac 3{n+2}\binom{2n}{n-1} = \frac 1{n+2}\binom{2n+2}{n+1}.$$ [**Case $\mathbb B_n$**]{} $$a(\mathbb B_n) = \binom{2n-2}{n-1} + \binom{2n-1}n = \binom{2n}n.$$ [**Case $\mathbb D_n$**]{} $$a(\mathbb D_n) = \left[\begin{matrix} 2n-2\cr n-2 \end{matrix}\right] + \left[\begin{matrix} 2n-2\cr n-1 \end{matrix}\right] = \left[\begin{matrix} 2n-1\cr n-1 \end{matrix}\right].$$ **Acknowledgment** ================== The authors are indebted to Henning Krause and Dieter Vossieck for providing the references [@[GP]] and [@[BLR]], and to Lutz Hille for helpful discussions concerning the problem of determining the number of tilting modules. They thank the referee for pointing out mistakes in the proof of Corollary \[hook-cor\]. Andrew Hubery has to be praised for his careful reading of the manuscript. His detailed comments have improved the presentation considerably. This work is funded by the Deanship of Scientific Research, King Abdulaziz University, under grant No. 2-130/1434/HiCi. The authors, therefore, acknowledge technical and financial support of KAU. [99]{} L. Angeleri-Hügel, D. Happel, and H. Krause, [*Handbook of Tilting Theory,*]{} London Math. Soc. Lecture Note Series, Vol. 332, Cambridge University Press, 2007. D. Armstrong, [*Generalized Noncrossing Partitions and Combinatorics of Coxeter Groups.*]{} Mem. Amer. Math. Soc., Vol. 949, 2009. M. Auslander, M. I. Platzeck, and I. Reiten, Coxeter functors without diagrams, [*Trans. Amer. Math. Soc.*]{} [**250**]{} (1979), 1–46. M. Auslander, I. Reiten, and S. Smalø, [*Representation Theory of Artin Algebras,*]{} Cambridge Studies in Advanced Mathematics, Vol. 36, Cambridge University Press, 1997. D. F. Bailey, Counting arrangements of $1$’s and $-1$’s, [*Math. Mag.*]{} [**69**]{} (1996), 128–131. O. Bretscher, Chr. Läser, and Chr. Riedtmann, Selfinjective and simply connected algebras. [*Manuscripta Math.,*]{} [**36**]{} (1981), 253–307. F. Chapoton, Enumerative properties of generalized associahedra, [*Sém. Lothar. Combin.*]{} [**51**]{} (2004). V. Dlab and C. M. Ringel, On algebras of finite representation type, [*J. Algebra*]{} [**33**]{} (1975), 306-394 V. Dlab and C. M. Ringel, [*Indecomposable Representations of Graphs and Algebras,*]{} Mem. Amer. Math. Soc., Vol. 173, 1976. W. Fakieh and C. M. Ringel, The hereditary artinian rings of type $\mathbb H_3$ and $\mathbb H_4$, in preparation. S. Fomin and N. Reading, Root systems and generalized associahedra, in [*Geometric combinatorics.*]{} IAS/Park City Math. Ser., 13, Amer. Math. Soc., 2007, 63–131. S. Fomin and A. Zelevinsky, Y-systems and generalized associahedra. [*Ann. of Math.*]{} (2) [**158**]{} (2003), 977–1018. P. Gabriel, Un jeu? Les nombres de Catalan, [*Uni Zürich, Mitteilungsblatt des Rektorats,*]{} 12. Jahrgang, Heft [**6**]{} (1981), 4–5. P. Gabriel and J. A. de la Peña, Quotients of representation-finite algebras, [*Comm. Algebra*]{} [**15**]{} (1987), 279–307 D. Happel and C. M. Ringel, Tilted algebras. [*Trans. Amer. Math. Soc.*]{} [**274**]{} (1982), 399–443. D. Happel and C. M. Ringel, Construction of tilted algebras, in [*Representations of Algebras: Proceedings ICRA 3.*]{} Lecture Notes in Math., Vol. 903, Springer, 1981, pp. 125–167. A. Hubery, On the modified hook formula, in preparation. A. Hubery and H. Krause, A categorification of non-crossing partitions, [http://arxiv.org/abs/1310.1907,]{} to appear in [*J. Eur. Math. Soc.*]{} C. Ingalls and H. Thomas, Noncrossing partitions and representations of quivers. [*Compos. Math.*]{} [**145**]{} (2009), 1533–1562. Chr. Krattenthaler, The F-triangle of the generalised cluster complex. [*Topics in discrete mathematics,*]{} Algorithms Combin. 26, Springer, 2006, pp.  93–126. H. Krause, Thick subcategories of modules over commutative noetherian rings (with an appendix by Srikanth Iyengar), [*Math. Ann.*]{} [**340**]{} (2008), 733–747. S. Ladkani, Universal derived equivalences of posets of tilting modules. [http://arxiv.org/abs/0708.1287.]{} R. Marsh, M. Reineke, and A. Zelevinsky, Generalized associahedra via quiver representations, [*Trans. Amer. Math. Soc.*]{} [**355**]{} (2003), 4171–4186. A. A. Obaid, S. K. Nauman, W. S. Al Shammakh, W. M. Fakieh, and C. M. Ringel, The number of complete exceptional sequences for a Dynkin algebra. [*Colloq. Math.*]{} [**133**]{} (2013), 197–210 A. A. Obaid, S. K. Nauman, W. M. Fakieh, and C. M. Ringel, The Ingalls-Thomas bijections, preprint. D. I. Panyushev, Ad-nilpotent ideals of a Borel subalgebra: generators and duality. [*J. Algebra*]{} [**274**]{} (2004), 822–846 C. M. Ringel, Reflection functors for hereditary algebras. [*J. Lond. Math. Soc.*]{} (2) [**21**]{} (1980), 465–479. C. M. Ringel, The Catalan combinatorics of the hereditary artin algebras. [http://arxiv.org/abs/1502.06553.]{} N. J. A. Sloane, Online Encyclopedia of Integer Sequences. [http://oeis.org/]{} ------------------------------------------------------------------------ 2010 [*Mathematics Subject Classification*]{}: Primary: 05E10. Secondary: 16G20, 16G60, 05A19, 16D90, 16G70. *Keywords:* Dynkin algebra, Dynkin diagram, tilting module, support-tilting module, lattice of non-crossing partitions, cluster combinatorics, generalized Catalan number, Catalan triangle, Pascal triangle, Lucas triangle, categorification. ------------------------------------------------------------------------ (Concerned with sequences , , , , , , .)

ArXiv

--- abstract: | Ces notes correspondent à un cours donné lors de l’école thématique de théorie ergodique au C.I.R.M. à Marseille en avril 2006. Nous présentons et démontrons un théorème de Kerckhoff, Masur et Smillie sur l’unique ergodicité du flot directionnel sur une surface de translation dans presque toutes les directions.\ La preuve suit essentiellement celle présentée dans un survol de Masur et Tabachnikov. Nous donnons une preuve complète et élémentaire du théorème. address: - ' IRMAR, Université de Rennes 1, Campus de Beaulieu, Bâtiment 22, 35042 Rennes Cedex, France. ' - ' Centre de Physique Théorique de Marseille (CPT), UMR CNRS 6207 Université du Sud Toulon-Var et Fédération de Recherches des Unités de Mathématiques de Marseille Luminy, Case 907, F-13288 Marseille Cedex 9, France. ' author: - Sébastien Gouëzel - Erwan Lanneau bibliography: - 'biblio.bib' date: 27 novembre 2006 title: | Un théorème de Kerckhoff, Masur et Smillie :\ Unique ergodicité sur les surfaces plates --- Introduction ============ Le but de ce texte est de démontrer un théorème de Steven Kerckhoff, Howard Masur et John Smillie concernant un résultat de théorie ergodique sur les surfaces de translation ([@Kerckhoff:Masur:Smillie]). Le prototype d’une surface de translation est donné par le tore plat standard $\mathbb T^2={\mathbb R}^2 /{\mathbb Z}^2$ muni de sa métrique plate (ou euclidienne) ${{\rm d}\hspace{-0.5pt}}z$. On peut construire géométriquement $\mathbb T^2$ de la manière suivante. Un domaine fondamental pour l’action de $\mathbb Z^2$ sur $\mathbb R^2$ par translation est donné par le carré unité $]0,1[^2$. Le tore s’obtient alors en identifiant les bords opposés de ce carré à l’aide de translations (voir figure \[fig:tore\]a). On peut définir sur $\mathbb T^2$ un flot directionnel $\mathcal F_\theta$, pour $\theta \in \mathbb S^1$. C’est le flot linéaire de pente constante égale à $\theta$ : si $x\in \mathbb T^2$ et $t>0$ alors $\mathcal F_\theta^t(x)$ est l’unique point de $\mathbb T^2$ à distance $t$ de $x$ et situé sur la géodésique (orientée positivement) de pente $\theta$ passant par $x$ (voir figure \[fig:tore\]b). On définit de manière équivalente ${\mathcal F}_\theta^t(x)$ pour $t<0$. Par construction il est clair que le flot $\mathcal F_\theta$ laisse invariant la mesure de Lebesgue notée $\operatorname{Leb}$. On peut donc considérer le système dynamique $(\mathbb T^2,\operatorname{Leb},\mathcal F_\theta)$ ; il vérifie le théorème classique suivant, dû à Hermann Weyl : Pour Lebesgue presque tout $\theta\in\mathbb S^1$, le système dynamique $(\mathbb T^2,\operatorname{Leb},\mathcal F_\theta)$ est uniquement ergodique. Rappelons que cela signifie que $\mathcal F_\theta$ admet une seule mesure de probabilité invariante. De manière équivalente, toutes les orbites du flot sont denses dans le tore et uniformément distribuées par rapport à la mesure de Lebesgue. On dira aussi souvent dans la suite que le flot directionnel est uniquement ergodique dans presque toutes les directions. En fait, le théorème de Weyl est même plus précis que cela, puisqu’il donne l’unique ergodicité dans toutes les directions *irrationnelles* ; les directions exceptionnelles forment donc un ensemble dénombrable. Une surface de translation est un triplet $({\mathcal S},\Sigma,\omega)$ où ${\mathcal S}$ est une surface compacte, connexe, sans bord, orientée, $\Sigma=\{P_1,\dots,P_n\}$ est un ensemble fini de points $P_i\in {\mathcal S}$ et $\omega=\left\{ (U_i,z_i) \right\}_i$ est un atlas de translation sur ${\mathcal S}\setminus \Sigma$. Par atlas de translation, nous demandons que les changements de cartes soient du type $z_i=z_j+cst$. On demandera de plus que, au voisinage de chaque singularité $P_i$, $\Sigma$ soit isomorphe à un revêtement de ${\mathbb R}^2 {-}\{0\}$ avec un nombre fini $k_i+1$ de feuillets. Par abus de langage nous dénoterons souvent $({\mathcal S},\Sigma,\omega)$ simplement par $({\mathcal S},\omega)$ ou juste ${\mathcal S}$ lorsque le contexte sera clair. La structure euclidienne sur ${\mathcal S}$ induit naturellement une mesure que l’on appellera encore mesure de Lebesgue $\operatorname{Leb}$. Comme précédemment, nous pouvons définir un flot directionnel $\mathcal F_\theta$ sur ${\mathcal S}$ laissant invariant la métrique euclidienne définie sur ${\mathcal S}$, via les cartes. Ce flot n’est en fait pas correctement défini sur $\Sigma$, ni même sur les trajectoires qui arrivent ou partent de $\Sigma$ (il y en a un nombre fini). Ainsi, $\mathcal F_\theta$ est défini sur un sous-ensemble dense et de mesure pleine de ${\mathcal S}$ mais, par abus de langage, nous parlerons néanmoins du flot ${\mathcal F}_\theta$ sur ${\mathcal S}$. Comme le flot directionnel dans ${\mathbb R}^2$ préserve la mesure de Lebesgue, ${\mathcal F}_\theta$ préserve la mesure $\operatorname{Leb}$ sur ${\mathcal S}$. On peut alors énoncer le théorème que nous allons démontrer dans ce texte et qui généralise le théorème de Weyl énoncé ci-dessus. Pour toute surface de translation ${\mathcal S}$ et pour Lebesgue presque tout $\theta\in\mathbb S^1$, le flot directionnel $\mathcal F_\theta$ sur ${\mathcal S}$ est uniquement ergodique. Le théorème ci-dessus reste vrai dans un cadre plus général. Nous pouvons étendre la définition de surfaces de translation aux surfaces de demi-translation en imposant que les changements de cartes soient de la forme $z_i=\pm z_j + cst$. L’objet naturel venant avec un atlas de translation est une forme différentielle abélienne. L’analogue pour les surfaces de demi-translation sont les formes différentielles quadratiques (voir [@Hubbard:Masur] pour un exposé plus détaillé). La preuve du théorème de Kerckhoff, Masur et Smillie dans ce cadre étendu est simplement plus technique et ne nécessite pas d’idée nouvelle fondamentale. Par conséquent, nous nous restreindrons aux surfaces de translation. Remarquons aussi que, contrairement au tore, il y a des exemples de surfaces de translation où il y a un nombre non dénombrable de directions minimales et non uniquement ergodiques ([@Veech:68; @Keane]). Une des différences essentielles avec le tore est que le flot linéaire sur un tore est une isométrie globale, i.e. la métrique plate ne possède pas de singularités. Enfin on peut noter que le théorème KMS est un renforcement d’un théorème de H. Masur et W. Veech ([@Masur:82; @Veech:82]). Une section de Poincaré du flot directionnel sur un intervalle dans ${\mathcal S}$ produit un [*échange d’intervalles*]{}. Masur et Veech ont démontré que presque tout échange d’intervalles est uniquement ergodique. Dans cette note, nous allons suivre une démarche un peu atypique : après avoir donné quelques exemples et propriétés basiques des surfaces de translation, et démontré une version faible du théorème KMS, nous exposerons une preuve complète et élémentaire du théorème KMS.\ Ce n’est qu’ensuite que nous introduirons des outils conceptuels supplémentaires qui apporteront un autre éclairage sur la preuve. Nous espérons que la démonstration du théorème pourra être vue comme une justification *a priori* de l’introduction de ces outils. Quelques définitions et exemples ================================ Surfaces de translation ----------------------- Avant d’aller plus loin, donnons quelques propriétés utiles des surfaces de translation. Comme nous l’avons vu précédemment, ${\mathcal S}\setminus \Sigma$ possède un atlas de translation. Ceci permet de définir une métrique euclidienne via la forme différentielle globale $\omega={{\rm d}\hspace{-0.5pt}}z_i$ sur $U_i$. Ainsi ${\mathcal S}\setminus \Sigma$ est localement isométrique à un plan $\mathbb R^2$, la courbure de la métrique étant nulle en tout point. La formule de Gau[ß]{}-Bonnet implique alors que la courbure de la métrique sur ${\mathcal S}$ est concentrée dans les points $P_i$. Par définition, il existe pour tout $i$ une identification entre un voisinage épointé de $P_i$ dans ${\mathcal S}$ et un revêtement de degré $k_i+1$ de ${\mathbb R}^2 {-}\{0\}$. La surface ${\mathcal S}$ est ainsi isométrique à un cône d’angle $2 (k_i+1)\pi$. Nous parlerons ainsi pour $P_i$ de singularité conique d’angle $2 (k_i+1)\pi$. Notons ici qu’au voisinage d’un point régulier, la surface est bien isométrique à un “cône plat” d’angle $2 (k+1)\pi = 2\pi$ avec $k=0$. La courbure de Gau[ß]{} est donnée par $\kappa=-k_i\pi$ et la formule de Gau[ß]{}-Bonnet s’écrit ici : $$\sum_{i=1}^n k_i = 2g-2 \qquad \textrm{o\`{u} } g=\textrm{genre}({\mathcal S}).$$ Le théorème de Riemann-Roch implique que pour toute partition entière $(k_1,\dots,k_n)$ de $2g-2$, il existe une surface de translation avec exactement $n$ singularités coniques d’angles $2 (k_i+1)\pi$ pour $i=1,\dots,n$ ; mais nous n’utiliserons pas ce fait ici. On peut définir un feuilletage vertical (respectivement horizontal) sur ${\mathcal S}\setminus \Sigma$ par les lignes de niveaux des cartes locales : $z_i^{-1}(x=cst)$ (respectivement $z_i^{-1}(y=cst)$). Cela permet alors, dans ces coordonnées, d’introduire le flot directionnel $\mathcal F_\theta$ pour $\theta\in \mathbb S^1$. La métrique euclidienne est invariante sous l’action de ce flot. Les orbites du flot directionnel ne sont bien sûr pas toutes bien définies : certaines feuilles rencontrent des singularités en temps fini. Néanmoins ${\mathcal S}$ ne possède qu’un nombre fini de singularités et par chaque singularité ne passe qu’un nombre fini de feuilles dans la direction $\theta$. Ceci implique alors que pour tout $\theta$, le flot ${\mathcal F}_\theta$ est presque partout bien défini sur ${\mathcal S}$. La structure euclidienne sur ${\mathcal S}$ est très facile à visualiser localement ; dans les coordonnées fournies par le feuilletage horizontal et le feuilletage vertical, une géodésique est localement une droite de pente constante. Bien sûr globalement une feuille peut avoir des comportements baroques. Dans toute la suite, sauf mention du contraire, nous allons nous restreindre aux surfaces d’aire $1$, c’est-à-dire $\operatorname{Aire}(S) := \int_S \omega \wedge \overline{\omega} = 1$. Nous utiliserons les terminologies suivantes pour le flot directionnel : - Une feuille passant par une singularité est une *séparatrice*. - Une feuille connectant deux singularités (éventuellement les mêmes) est une *connexion de selles* ou segment géodésique. - Une feuille ne passant pas par une singularité est dite *régulière*. \[par:RecollePolygones\] La géométrie des surfaces de translation peut se comprendre en voyant une telle surface comme recollement de polygones. Plus précisément, considérons des polygones $P_1,\dots, P_n$ du plan, d’adhérences disjointes, tels que les côtés de ces polygones soient regroupés par paires de côtés parallèles et de même longueur. On peut alors former une surface ${\mathcal S}$ en identifiant les côtés correspondants. Nous ajoutons la conditions que l’angle conique autour des sommets des polygones (vu dans ${\mathcal S}$) soit de la forme $2c\pi$ où $c$ est un entier (sans cette condition, la notion d’angle est mal définie, voir par exemple la définition analogue dans [@Masur:06]). La surface ${\mathcal S}$ devient alors une surface de translation. C’est une surface de translation, dont les singularités forment un sous-ensemble des sommets des polygones $P_j$. Réciproquement, toute surface de translation peut s’obtenir ainsi, comme nous allons le voir. Notons aussi que pour des raisons évidentes, nous ne considérerons que des surfaces connexes. \[def:Triangulation\] On appelle triangulation d’une surface de translation une triangulation dont les sommets sont les singularités et les arêtes des connexions de selles. \[ExisteTriangulation\] Si on part d’un ensemble de connexions de selles dont les intérieurs sont deux à deux disjoints, on peut les compléter pour obtenir une triangulation de la surface (ce qui se vérifie aisément en ajoutant des arêtes tant que c’est possible et en remarquant que quand ce n’est plus possible la surface est nécessairement triangulée). Ainsi, toute surface admet une triangulation. On en déduit aisément que toute surface de translation s’obtient en recollant des triangles suivant la procédure décrite en \[par:RecollePolygones\]. Considérons une triangulation d’une surface de translation ${\mathcal S}$ de genre $g$, et notons respectivement $s,e,f$ le nombre de sommets, côtés et faces de cette triangulation. Alors $s-e+f=2-2g$ par la formule d’Euler. Comme chaque face a trois côtés et chaque côté est sur le bord de deux faces, nécessairement $2e=3f$. Comme $s$ est le nombre de singularités, on obtient $$\label{FormuleFaces} f=2(s+2g-2).$$ Exemples -------- Le tore plat fournit beaucoup d’exemples dérivés. Soit $f : {\mathcal S}\rightarrow \mathbb T^2$ un revêtement ramifié au dessus de l’origine. Si on note ${{\rm d}\hspace{-0.5pt}}z$ la métrique plate sur $\mathbb T^2$ alors la métrique $f^\ast{{\rm d}\hspace{-0.5pt}}z$ équipe ${\mathcal S}\setminus f^{-1}(0)$ d’une structure de surface plate (voir figure \[fig:exemple:3:carreaux\]). ![ \[fig:exemple:3:carreaux\] Un revêtement ramifié de degré $3$ au dessus du tore. Le revêtement est ramifié au dessus de l’origine et l’unique point critique est une singularité pour la métrique, d’angle $6\pi$. On a donc $2\cdot \textrm{genre}({\mathcal S})-2=2$, soit $\textrm{genre}({\mathcal S})=2$. ](3carreaux.eps){width="9cm"} Un exemple beaucoup plus “générique” peut être construit à partir d’une surface plate quelconque, obtenue en recollant des polygones : il suffit de bouger les côtés des polygones de manière à préserver la propriété des paires de côtés (la figure \[fig:exemple:3:carreaux:modif\] est un exemple “déformé” de la figure \[fig:exemple:3:carreaux\]). Le terme générique employé ci-dessus sera expliqué dans la section \[sec:espace:module\]. ![ \[fig:exemple:3:carreaux:modif\] Une surface de translation de genre $2$. L’angle conique de l’unique singularité (point blanc) est $6\pi$.](3carreauxmodif.eps){width="4.5cm"} Nous donnons dans la figure \[fig:exemple:4:slit\] un dernier exemple, toujours de genre $2$, mais cette fois-ci avec deux singularités coniques, nécessairement chacune d’angle $4\pi$. ![ \[fig:exemple:4:slit\] Deux tores unité recollés le long du segment géodésique vertical $P_1P_2$ centré et de hauteur $\lambda \in ]0,1[$ (on identifie les côtés de même étiquette par translation). La surface ${\mathcal S}_\lambda$ correspondante est de genre $2$. Elle possède deux singularités coniques (points blanc et noir) chacune d’angle $4\pi$. ](slit.eps){width="6.5cm"} Billards rationnels et surfaces de translation ---------------------------------------------- Soit $P \subset {\mathbb R}^2$ un polygone. Dans tout ce paragraphe, on notera $\Gamma$ le sous groupe des isométries linéaires de ${\mathbb R}^2$ engendré par les parties linéaires des réflexions par rapport aux côtés de $P$. On note $\tilde {\mathcal F}$ le flot du billard sur $P\times {\mathbb{S}}^1$ : le point $\tilde {\mathcal F}_t(x,\theta)$ est obtenu en partant de $x$, en suivant une ligne droite dans la direction $\theta$ jusqu’à rencontrer un côté du polygone $P$. On “rebondit” alors suivant les lois de Descartes de l’optique géométrique et on continue ce processus pendant un temps $t$. Ce flot préserve la mesure $\operatorname{Leb}_P \otimes \operatorname{Leb}_{{\mathbb{S}}^1}$, et il est défini pour toutes les trajectoires qui n’arrivent pas dans un des coins du polygone. Lorsque les angles du billard sont des multiples rationnels de $\pi$, ce flot n’est évidemment pas ergodique : si $\theta \in {\mathbb{S}}^1$, l’ensemble $P\times \bigcup_{\gamma\in \Gamma} \{\gamma \cdot \theta\}$ est invariant par $\tilde {\mathcal F}$ (notons que $\Gamma$ est alors fini). On notera alors $\tilde {\mathcal F}_\theta$ la restriction de $\tilde {\mathcal F}$ à cet ensemble ; ce flot préserve la mesure $\operatorname{Leb}\otimes \sum_{\gamma \in \Gamma} \delta_{\gamma\cdot\theta}$. Une construction classique (voir [@FoKe; @KaZa]) permet d’associer à chaque polygone $P$ une surface de translation (non nécessairement compacte) ${\mathcal S}(P)$ telle que le flot du billard et le flot directionnel commutent avec la construction. Nous résumons ici la construction en question. Pour $\gamma \in \Gamma$ on notera $P_\gamma=\gamma(P)$. Nous allons décrire une relation d’équivalence sur la réunion disjointe $\bigsqcup_{\gamma \in \Gamma} P_\gamma$ des polygones $P_\gamma$. Si $c$ est un côté de $P_\gamma$ soit $\delta_c\in\Gamma$ la partie linéaire de la réflexion par rapport à $c$. On identifie alors $c\subset P_\gamma$ avec $\delta_c(c)\subset P_{\delta \circ \gamma}$ par translation. L’ensemble quotient pour cette relation est noté ${\mathcal S}(P)$. La surface ${\mathcal S}(P) \setminus \{ \textrm{sommets de } P \}$ hérite naturellement d’un atlas de translation via les polygones $P_\gamma$. Il est facile de vérifier que cet atlas s’étend globalement en une structure de translation sur ${\mathcal S}$ tout entier. Les singularités de la métrique forment un sous-ensemble (éventuellement strict) des sommets des polygones $P_\gamma$. Par ailleurs, la surface ${\mathcal S}$ est compacte si et seulement si le polygone $P$ est rationnel, c’est à dire si et seulement si les angles de $P$ sont des multiples rationnels de $\pi$. Le polygone $P$ et ses images par les éléments de $\Gamma$ pavent alors la surface ${\mathcal S}$ et on obtient une application $\pi : {\mathcal S}(P) \rightarrow P$. Par construction les orbites du flot ${\mathcal F}$ se projettent sur les orbites du flot $\tilde {\mathcal F}$. Lorsque ${\mathcal S}$ est compacte, les orbites de ${\mathcal F}_\theta$ se projettent même sur celles de $\tilde{\mathcal F}_\theta$, de manière $\pi$-équivariante. Ainsi comme corollaire direct du théorème KMS on obtient le résultat suivant. \[cor:KMS\_billards\] Pour tout billard rationnel, l’ensemble $$\{\theta \in \mathbb S^1,\ \tilde {\mathcal F}_\theta \textrm{ est uniquement ergodique} \}$$ est de mesure pleine. Notons que ce résultat a des conséquences pour les billards non nécessairement rationnels. En effet, en approchant un billard (polygonal) arbitraire par des billards rationnels, grâce à une remarque de A. Katok, il permet de démontrer le résultat suivant. \[cor:ergodique\] L’ensemble $\{ Q\ ; \ Q \textrm{ est un polygone \`{a} } k \textrm{ c\^{o}t\'{e}s et } \tilde {\mathcal F}\textrm{ est ergodique} \}$ forme un $G_\delta$ dense dans l’ensemble des polygones à $k$ côtés (pour la topologie produit, c’est à dire en identifiant l’ensemble des polygones à $k$ côtés avec un ouvert de ${\mathbb R}^{2k}$). Si l’on est principalement intéressé par les billards, et en particulier par le corollaire \[cor:KMS\_billards\], on pourrait essayer de ne pas mentionner du tout les surfaces de translation, et de transcrire directement la preuve du théorème KMS que nous allons présenter dans le langage des billards. Cependant, un ingrédient essentiel de la preuve du théorème KMS est la *déformation* des surfaces de translation, par l’action du groupe ${\textrm{SL}_2(\mathbb R)}$ (et plus précisément par son sous-groupe constitué des matrices diagonales). Ce processus de déformation est aisé à mettre en place dans le cadre des surfaces de translation (voir la section \[sec:action\]), alors qu’il est impossible dans l’espace des billards rationnels ! Moralement, cela s’explique par le fait qu’un billard rationnel et son flot sont des notions euclidiennes, alors qu’une surface de translation est une notion affine. C’est ce changement de groupe structurel qui permet de faire marcher la preuve. Ainsi, les surfaces de translation sont un outil indispensable dans l’étude des billards rationnels. Nous terminons cette section par quelques remarques. Dans l’étude des surfaces de translation, il existe trois grandes classes de résultats. 1. Tout d’abord, certains types de résultats sont valables pour *toutes* les surfaces de translation. C’est par exemple le cas du théorème KMS. Les techniques utilisées pour démontrer ce genre de résultats ne sont en fait pas très nombreuses (elles sont essentiellement géométriques et combinatoires). La preuve du théorème KMS que nous allons donner est assez représentative de ce type d’arguments. 2. Nous verrons plus loin que l’on peut mettre une mesure sur l’ensemble des surfaces de translation lorsque l’on fixe le type topologique (genre et singularités). Certains résultats sont démontrés pour *presque toutes* les surfaces de translation, au sens de cette mesure. Les résultats obtenus ainsi sont souvent beaucoup plus précis que ceux qui sont valables en toute généralité. Néanmoins, le prix à payer est que ces résultats ne s’appliquent pas aux billards puisque l’ensemble $\{{\mathcal S}(P),\ P \textrm{ polygone rationnel} \}$ est de mesure nulle dans l’espace des surfaces de translation ! 3. Enfin, on peut se restreindre à certaines classes de surfaces (surfaces de Veech, ou surfaces de genre 2 par exemple) et chercher à obtenir des résultats de classification, presque de nature algébrique. Les techniques utilisées pour démontrer des résultats du deuxième ou du troisième type sont très variées, et ne seront pas du tout abordées dans ce texte. Le théorème KMS faible ====================== \[theo:kms:faible\] Pour toute surface de translation ${\mathcal S}$ et pour Lebesgue presque tout $\theta\in\mathbb S^1$, le flot directionnel $\mathcal F_\theta$ sur ${\mathcal S}$ est minimal. Ce théorème est dû à A. Katok et A. Zemljakov ([@KaZa]). Il implique le corollaire suivant, qui est une version “transitive” du corollaire \[cor:ergodique\]. L’ensemble $\{ Q\ ; \ Q \textrm{ est un polygone \`{a} } k \textrm{ c\^{o}t\'{e}s et } \tilde {\mathcal F}\textrm{ est transitif} \}$ forme un $G_\delta$ dense dans l’ensemble des polygones à $k$ côtés (pour la topologie produit, c’est à dire en identifiant l’ensemble des polygones à $k$ côtés avec un ouvert de ${\mathbb R}^{2k}$). Le résultat ci-dessus est bien plus faible que le théorème de Kerckhoff, Masur et Smillie (théorème KMS fort). En particulier, nous allons montrer ici que l’ensemble des directions non-minimales est dénombrable, donc de mesure nulle. Dans le cas général, il peut arriver que l’ensemble des directions non uniquement ergodiques soit bien plus “gros”. Par exemple, pour les surfaces ${\mathcal S}_\lambda$ de la figure \[fig:exemple:4:slit\], on peut montrer que si $\lambda$ est diophantien alors l’ensemble des directions non-uniquement ergodiques sur la surface ${\mathcal S}_\lambda$ est de dimension de Hausdorff $1/2$ ([@Cheung], voir aussi section \[sec:raffinements\]). Le théorème \[theo:kms:faible\] repose sur les propositions \[prop:denombrable\] et \[prop:minimal\] suivantes. \[prop:denombrable\] Sur une surface de translation, l’ensemble des directions de connexions de selles est au plus dénombrable. Il suffit de passer au revêtement universel. \[prop:minimal\] Soit ${\mathcal S}$ une surface de translation. Si le flot directionnel $\mathcal F_\theta$ ne possède pas de connexion de selles alors il est minimal : c’est à dire toutes les feuilles de $\mathcal F_\theta$ sont denses. On peut sans perte de généralité se restreindre au cas du feuilletage vertical. Avant de démontrer cette proposition, nous aurons besoin d’un lemme géométrique et d’un lemme technique. \[lm:geo\] Soit ${\mathcal S}$ une surface de translation de genre $g\geq 2$. Supposons que le flot vertical possède une feuille verticale régulière fermée. Alors le flot vertical possède aussi une connexion de selles. Soit $\alpha$ une feuille verticale régulière fermée sur ${\mathcal S}$. Comme ${\mathcal S}$ ne possède qu’un nombre fini de singularités, il existe une autre feuille verticale fermée parallèle à $\alpha $ et proche de $\alpha$. De proche en proche, cela fournit un plongement isométrique du cylindre $]0,r[ \times \mathbb S^1$ de périmètre $|\alpha|$ et de largeur $r$ dans ${\mathcal S}$. Ce cylindre est feuilleté par des feuilles régulières verticales homologues à $\alpha$. Par ailleurs $\operatorname{Aire}({\mathcal S})=1$ donc on ne peut pas “épaissir” ce cylindre à l’infini. Les deux seules obstructions topologiques sont les suivantes. Ou bien il existe une connexion de selles verticale, située sur le bord du cylindre, ou bien ${\mathcal S}$ s’obtient en recollant les deux bords du cylindre ensemble. Ce dernier cas implique alors que $g({\mathcal S}) = 1$ ce qui est une contradiction. Le lemme est ainsi démontré. \[lm:technique\] Soit $\alpha^{+}$ une feuille verticale (orientée positivement) non périodique (i.e. non fermée). Soient $P\in \alpha$ un point et $I=[P,Q]$ un segment géodésique transverse au flot vertical. Alors $]P,Q[ \ \cap\ \alpha^{+} \not = \emptyset$. La surface ${\mathcal S}$ ne possède qu’un nombre fini de séparatrices verticales. Ainsi on peut choisir $I'=[P,Q'] \subset I$ tel que toutes les feuilles verticales issues de $I'$ ne rencontrent pas de singularités avant de revenir à $I$. Considérons alors un petit rectangle vertical $R_h$ [*plongé*]{} dans ${\mathcal S}$, de base $I'$ et de hauteur $h$, avec $h$ petit. Le bord vertical gauche de $R_h$ est inclus dans $\alpha^{+}$. Notons par $\beta^{+}$ le bord vertical droit. Lorsque $h \to \infty$ le rectangle $R_h$ ne reste pas plongé dans ${\mathcal S}$ : en effet l’aire de ${\mathcal S}$ est finie et l’aire de $R_h$ tends vers l’infini. Donc (rappelons que $R_h$ ne rencontre pas de singularité avant de revenir à $I$) deux possibilités peuvent arriver. Ou bien $\alpha^{+}$ intersecte $]P,Q'[$ et le lemme est démontré (figure \[fig:lemme\]a), ou bien c’est $\beta^{+}$ qui intersecte $]P,Q'[$ (figure \[fig:lemme\]b). Dans ce dernier cas, notons $Q'' \in I'$ l’unique point tel que la feuille verticale issue de $Q''$ dans $R_h$ arrive à $P$. Considérons alors $I''=[P,Q''] \subset I'$ et appliquons le raisonnement analogue à celui précédent. On obtient alors un petit rectangle de base $I''$. Le même argument amène à la dichotomie précédente. Il est facile de voir alors que la deuxième conclusion dans cette dichotomie ne peut-être satisfaite. Le lemme est démontré. On peut maintenant conclure. Supposons que ${\mathcal S}$ ne contienne pas de connexion de selles verticale. Soit $\beta$ une feuille verticale. Si $\beta$ est fermée alors on obtient une contradiction grâce au lemme \[lm:geo\]. Notons $A$ l’adhérence de $\beta$ dans ${\mathcal S}$. Nous allons montrer que $A={\mathcal S}$. Sinon prenons $P \in \partial A$. Soit $\alpha$ la feuille verticale issue de $P$. Si $\alpha$ est fermée alors le lemme \[lm:geo\] fournit de nouveau une contradiction. Sinon on note $I=[P,Q]$ un petit segment géodésique, transverse au flot vertical, passant par $P$ et tel que son intérieur $\operatorname{Int}(I)$ soit inclus dans ${\mathcal S}\setminus A$. Notons aussi que $A$ est invariant par le flot vertical donc $\alpha \subset A$. Le lemme \[lm:technique\] implique alors que $\operatorname{Int}(I) \cap \alpha \not =\emptyset$ ce qui est évidemment incompatible avec les deux dernières assertions. Donc $A={\mathcal S}$ et ainsi toutes les feuilles verticales sont denses. La proposition est démontrée. Pour aller plus loin que le théorème \[theo:kms:faible\], on peut chercher à caractériser les directions exceptionnelles, c’est-à-dire les directions pour lesquelles il y a une connexion de selles. C’est un problème très difficile en général. Par exemple en genre $2$, pour les surfaces de Veech, cet ensemble est toujours (quitte à normaliser convenablement) un corps de nombres union l’infini (plus précisément, soit $\mathbb Q \cup \{\infty \}$, soit $K \cup \{\infty \}$ oú $K$ est un corps quadratique réel). Preuve du théorème KMS ====================== \[def:PartirInfini\] Soit ${\mathcal S}$ une surface de translation. On dira que ${\mathcal S}$ part à l’infini dans la direction $\pi/2$, ou simplement que ${\mathcal S}$ part à l’infini, si pour tout ${\varepsilon}>0$, il existe $T>0$ tel que, pour tout $t\geq T$, il existe une connexion de selles dans ${\mathcal S}$ (de composantes horizontales et verticales $h$ et $v$) telle que $e^t |h| + e^{-t}|v| \leq {\varepsilon}$. On parlera de même de “partir à l’infini dans la direction $\theta$” en considérant les composantes des connexions de selle dans la direction $\theta$ et la direction orthogonale à $\theta$. Les deux ingrédients essentiels de la preuve du théorème KMS sont les résultats suivants. Soit ${\mathcal S}$ une surface de translation qui ne part pas à l’infini dans la direction $\theta$. Alors le flot ${\mathcal F}_\theta$ est uniquement ergodique. Soit ${\mathcal S}$ une surface de translation. Alors l’ensemble $$\{\theta\in {\mathbb{S}}^1,\ {\mathcal S}\textrm{ part \`{a} l'infini dans la direction }\theta\}$$ est de mesure nulle. Ces deux théorèmes impliquent manifestement le théorème KMS. Leurs démonstrations sont complètement différentes et indépendantes, et seront présentées dans les sections \[PreuveThmA\] et \[PreuveThmB\] respectivement. Remarquons aussi tout de suite que le théorème $A$ n’est pas une équivalence, contrairement au cas du genre $1$ (voir section \[sec:raffinements\]). Le théorème $A$ est aussi connu sous le nom de critère de Masur. Notons qu’il existe une version combinatoire de ce critère, due à M. Boshernitzan (voir [@Boshernitzan]). Nous reviendrons section \[sec:raffinements\] sur ce critère. Avant de démontrer les théorèmes $A$ et $B$, on va introduire l’action de ${\textrm{SL}_2(\mathbb R)}$ sur les surfaces de translation, ce qui permet de reformuler de manière un peu plus confortable la notion de départ à l’infini définie en \[def:PartirInfini\]. Action de ${\textrm{SL}_2(\mathbb R)}$ sur les surfaces de translation {#sec:action} ====================================================================== Le groupe ${\textrm{SL}_2(\mathbb R)}$ agit linéairement sur ${\mathbb R}^2$ ; il agit donc aussi naturellement sur les surfaces de translation via les cartes locales. Si $({\mathcal S},\Sigma,\omega)$ est une surface de translation avec $\omega=\left\{ (U_i,z_i) \right\}_i$ et si $A\in {\textrm{SL}_2(\mathbb R)}$ est une matrice alors on définit l’action comme suit. $$A \cdot ({\mathcal S},\Sigma,\omega) := ({\mathcal S},\Sigma,A\omega)$$ où $A\omega$ est, par définition, le nouvel atlas plat $A \omega:=\left\{ (U_i, A \circ z_i) \right\}_i$, $A \circ z_i$ désignant l’action linéaire de $A$ sur la coordonnée $z_i$. Remarquons que l’action préserve les surfaces d’aire $1$. Nous serons particulièrement intéressé par les $3$ sous-groupes à $1$ paramètre suivants. $$g_t = \left( \begin{array}{ll} e^t & 0 \\ 0 & e^{-t} \end{array} \right), \qquad h_s = \left( \begin{array}{ll} 1 & s \\ 0 & 1 \end{array} \right) \qquad \textrm{ et } \qquad R_\theta = \left( \begin{array}{lr} \cos(\theta) & -\sin(\theta) \\ \sin(\theta) & \cos(\theta) \end{array} \right).$$ L’action de $g_t$ sera appelée [*flot géodésique de Teichmüller*]{} ou simplement flot géodésique pour une raison qui deviendra claire dans les sections suivantes. L’action de $g_t$ “écrase” les feuilles verticales par $e^t$ et “dilate” les feuilles horizontales par $e^t$. Le flot $h_s$ est le flot horocyclique et le flot $R_\theta$ est le flot circulaire. L’action de $R_\theta$ consiste à faire tourner le feuilletage vertical et le feuilletage horizontal de ${\mathcal S}$ d’un angle de $-\theta$. Nous noterons pour la suite ${\mathcal S}_\theta = R_\theta {\mathcal S}$ et ${\mathcal S}_{\theta,t} = g_t {\mathcal S}_\theta$. On vérifie sans peine que le flot vertical sur ${\mathcal S}_{\pi/2-\theta}$ correspond au flot dans la direction $\theta$ sur ${\mathcal S}$. Si ${\mathcal S}$ est une surface de translation, on appelle *systole* de ${\mathcal S}$ la longueur (dans la métrique euclidienne) de la plus petite connexion de selles de ${\mathcal S}$. On la notera $\operatorname{sys}({\mathcal S})$. Avec ces définitions, on peut reformuler aisément la définition \[def:PartirInfini\] comme suit : une surface part à l’infini si et seulement si $\operatorname{sys}(g_t {\mathcal S}) \to 0$ quand $t \to \infty$. Elle part à l’infini dans la direction $\theta$ si et seulement si $\operatorname{sys}(g_t R_{\pi/2-\theta} {\mathcal S})\to 0$ quand $t \to \infty$. Preuve du théorème A {#PreuveThmA} ==================== Comme d’habitude, il suffit de le prouver pour le flot vertical $\theta=\pi/2$, puisque le résultat général s’en déduit par rotation. L’idée essentielle de la preuve est la suivante : si la surface ne part pas à l’infini, alors on va pouvoir extraire une “sous-suite convergente” (en un certain sens à expliquer) par un argument de compacité. On conclura ensuite en utilisant la connexité de la surface limite. \[def:Convergence\] Soient $s,g\in {\mathbb{N}}$. On dira qu’une suite de surfaces de translation ${\mathcal S}_n$ d’aire $1$ (de genre $g$, avec $s$ singularités) converge s’il existe des triangulations de ${\mathcal S}_n$ par des triangles $T_n^1,\dots, T_n^{f}$ (avec $f$ ne dépendant que de $g$ et $s$, d’après ) telles que 1. chaque suite de triangle $(T_n^i)_{n\in {\mathbb{N}}}$ converge, comme suite de triangles dans ${\mathbb R}^2$, vers un triangle (peut-être dégénéré i.e. aplati) $T^i$. 2. le motif de recollement des $T_n^i$ pour former ${\mathcal S}_n$ est constant à partir d’un certain rang. On impose aussi dans ce type de convergence que la systole $\operatorname{sys}({\mathcal S}_n)$ de ${\mathcal S}_n$ soit uniformément minorée par une constante non-nulle. Ceci implique en particulier que la surface limite est non “dégénérée”. Il est alors possible de recoller les triangles $T^i$ suivant les mêmes motifs que les $T_n^i$ pour $n$ assez grand, pour obtenir une surface de translation, que l’on notera ${\mathcal S}_\infty$, de genre $g$ avec $s$ singularités. On dira alors que ${\mathcal S}_n$ converge vers ${\mathcal S}_\infty$. La topologie de ${\mathcal S}_\infty$ est la même que celle des surfaces ${\mathcal S}_n$ (elle est déterminée par le motif combinatoire de recollement des triangles). Notons aussi que si l’on n’imposait pas la condition sur la systole de ${\mathcal S}_n$, il serait *a priori* possible que des singularités fusionnent dans la surface limite ! Par exemple la surface limite pourrait ne plus être connexe. \[par:PossibilitesConvergence\] Ce type de convergence est assez faible, mais il permet de réaliser les opérations suivantes. 1. Si $x_n \in {\mathcal S}_n$, cela a un sens de parler de la convergence de $x_n$ vers $x\in {\mathcal S}_\infty$, en regardant ce qui se passe dans le triangle contenant $x_n$. 2. Si on se donne une partie de ${\mathcal S}_\infty$, on peut l’approcher par une partie de ${\mathcal S}_n$ si $n$ est assez grand. Nous n’aurons pas besoin d’autre chose pour démontrer le théorème $A$. \[lem:BonneTriangulation\] Soit ${\varepsilon}_0>0$. Il existe une constante $C_0$ ne dépendant que de ${\varepsilon}_0$ telle que toute surface de translation d’aire $1$ et de systole au moins ${\varepsilon}_0$ admette une triangulation dont tous les côtés ont une longueur au plus $C_0$. Par triangulation, on entend ici une triangulation dont les sommets sont les singularités et les arêtes des connexions de selles, comme en \[def:Triangulation\]. Toute surface de translation admet une triangulation, d’après \[ExisteTriangulation\]. Partons d’une telle triangulation, et considérons une arête de longueur $K$ maximale. Notons-la $AC$, et considérons les deux triangles qui bordent cette arête, comme sur la figure \[triangle\]. ![\[triangle\]](triangle.eps){width="8.5cm"} Comme l’aire de la surface est au plus $1$, ces deux triangles sont d’aire au plus $1$. Ainsi l’aire de $AHB$ est $1/2(AC.BH) \leq 1$ ce qui donne $BH\leq 2/K$. Comme $AB$ et $BC$ sont de longueur au moins ${\varepsilon}_0$, Pythagore dans le triangle $ABH$ donne $AH^2 = AB^2-BH^2 \geq {\varepsilon}^2_0 - 4/K^2 \geq {\varepsilon}^2_0/4$ si $K$ est grand (c’est à dire si $K \geq \cfrac{4}{\sqrt{3}{\varepsilon}_0}$). Donc $AH\geq {\varepsilon}_0/2$. De la même manière on tire $CH\geq {\varepsilon}_0/2$. Les mêmes inégalités ont lieu pour $B'H'$, $AH'$ et $CH'$. On retriangule alors la surface en retirant l’arête $AC$ et en la remplaçant par une arête $BB'$. Cette arête est de longueur au plus $BH+B'H'+HH' \leq 4/K +(K-{\varepsilon}_0)$, qui est majoré par $K-{\varepsilon}_0/2$ si $K$ est très grand (c’est à dire si $K \geq 2/{\varepsilon}_0$). Ainsi, on peut faire décroître la longueur maximale des arêtes de la triangulation d’une quantité fixée. En répétant ce processus, on arrive alors à une triangulation satisfaisant la borne requise. \[cor:Compacite\] Soit ${\mathcal S}_n$ une suite de surfaces de translation d’aire $1$, avec un genre et un nombre de singularités fixés. On suppose que la systole de ${\mathcal S}_n$ ne tend pas vers $0$ quand $n$ tend vers l’infini. Alors ${\mathcal S}_n$ admet une sous-suite qui converge vers ${\mathcal S}_\infty$ au sens de la définition \[def:Convergence\]. De plus le genre et le nombre de singularités de ${\mathcal S}_\infty$ sont les mêmes que ${\mathcal S}_n$. Grâce au lemme \[lem:BonneTriangulation\], on obtient une sous-suite de ${\mathcal S}_n$ ayant une triangulation avec des côtés de taille bornée supérieurement et inférieurement, par des triangles $T_n^i$. Les suites de tels triangles dans ${\mathbb R}^2$ forment un ensemble compact, on peut donc choisir une sous-suite de ${\mathcal S}_n$ de telle sorte que les $T_n^i$ convergent. Il y a aussi un nombre fini de motifs de recollement possibles, on obtient donc une sous-suite convergente quitte à extraire encore une fois. Notons aussi que dans le lemme précédent, lorsque l’on retriangule la surface, la nombre de faces $f$ et le genre $g$ de la surface restent inchangés. Donc d’après la formule  le nombre de singularités $s$ reste aussi inchangé. Le corollaire est démontré. On peut commencer la preuve du théorème $A$ proprement dite. Soit ${\mathcal S}$ une surface de translation. On note ${\mathcal S}_t$ la surface obtenue en dilatant les distances horizontales d’un facteur $e^t$, et en contractant les distances verticales de ce même facteur. Formellement, ${\mathcal S}_t= g_t {\mathcal S}$. On suppose que ${\mathcal S}$ ne part pas à l’infini. Ceci garantit alors l’existence d’une suite $t_n\to \infty$ et de ${\varepsilon}_0>0$ tels que toute connexion de selles de la surface ${\mathcal S}_{t_n}$ soit de longueur au moins ${\varepsilon}_0$. D’après le corollaire \[cor:Compacite\], on peut extraire de la suite ${\mathcal S}_{t_n}$ une sous-suite convergente, que nous noterons simplement ${\mathcal S}_n$. Les surfaces ${\mathcal S}_n$ et ${\mathcal S}$ sont des structures de translation sur la même surface topologique sous-jacente. Pour bien les distinguer, nous noterons $f_n$ l’application $f_n : {\mathcal S}\rightarrow {\mathcal S}_n$ qui est l’identité sur cette surface topologique sous-jacente. Soit $\mu$ une mesure invariante ergodique pour le flot vertical sur ${\mathcal S}$. On note $B(\mu) \subset {\mathcal S}$ l’ensemble des points $x$ qui sont typiques pour la mesure $\mu$, i.e., tels que $\frac{1}{T} \int_{0}^T \delta_{{\mathcal F}_{\pi/2}^t(x)}$ converge faiblement vers $\mu$ quand $T\to \pm \infty$, où $\delta_y$ désigne la masse de Dirac au point $y$. On définit un sous-ensemble $A(\mu) \subset {\mathcal S}_\infty$ comme suit. C’est la réunion, pour $x\in B(\mu)$, de l’ensemble des valeurs d’adhérence possibles dans ${\mathcal S}_\infty$ de la suite $f_n(x)\in {\mathcal S}_n$ (comme défini en \[par:PossibilitesConvergence\]). Pour démontrer qu’il y a une seule mesure invariante $\mu$, on va en fait montrer qu’il y a un seul ensemble $A(\mu)$. On appellera *rectangle* un rectangle dont les côtés sont horizontaux et verticaux, et *rectangle plongé* un rectangle qui ne rencontre pas les singularités. \[lem:MuEgaleNu\] Considérons dans ${\mathcal S}_\infty$ un rectangle plongé. On suppose qu’il existe $x\in B(\mu)$ et $y\in B(\nu)$ (où $\mu$ et $\nu$ sont deux mesures invariantes) tels que $f_n(x)$ et $f_n(y)$ convergent respectivement vers deux sommets opposés de ce rectangle. Alors $\mu=\nu$. Supposons que $\mu$ et $\nu$ ne coïncident pas. Comme les rectangles plongés de hauteur $1$ engendrent la tribu des boréliens, il existe un rectangle $U=I\times J$, plongé dans ${\mathcal S}$, tel que $\mu(U) \not=\nu(U)$, et de hauteur $1$. Si $n$ est assez grand, on peut tracer dans ${\mathcal S}_n$ un rectangle $R_n$ à côtés horizontaux et verticaux, ne contenant pas de singularité, et dont $f_n(x)$ et $f_n(y)$ sont deux sommets opposés. Cela résulte de l’hypothèse dans ${\mathcal S}_\infty$, et de la convergence de ${\mathcal S}_n$ vers ${\mathcal S}_\infty$. De plus, la taille verticale $c_n$ de $R_n$ converge vers $c>0$. Le rectangle $f_n^{-1}(R_n)$ dans ${\mathcal S}$ est un rectangle de hauteur $T_n=c_n e^{t_n}\to \infty$, il est très haut et très fin, et possède $x$ et $y$ pour sommets opposés. Notons $C_n(x)$ et $C_n(y)$ les côtés verticaux de $f_n^{-1}(R_n)$ contenant respectivement $x$ et $y$. Comme $x$ est typique pour $\mu$, le théorème de Birkhoff donne $\operatorname{Card}( I \cap C_n(x)) \sim T_n \mu(U)$ quand $n\rightarrow \infty$. De même, $\operatorname{Card}( I \cap C_n(y)) \sim T_n \nu(U)$ quand $n\rightarrow \infty$. Comme le rectangle $f_n^{-1}(R_n)$ ne contient pas de singularité, tout segment horizontal traversant $C_n(x)$ doit traverser aussi $C_n(y)$ (sauf éventuellement une fois près de chaque bord du segment horizontal, voir figure \[fig:rectangle\]). ![ \[fig:rectangle\] Le rectangle $f_n^{-1}(R_n)$ vu sur la surface ${\mathcal S}$. Un segment horizontal traversant $C_n(x)$ traverse aussi $C_n(y)$ (sauf éventuellement une fois près de chaque bord du segment horizontal) donc $|\operatorname{Card}(I\cap C_n(x))- \operatorname{Card}( I\cap C_n(y))| \leq 2$. ](rectangle.eps){width="6.5cm"} Ceci implique alors que $|\operatorname{Card}(I\cap C_n(x))- \operatorname{Card}( I\cap C_n(y))| \leq 2$. Comme $T_n \to \infty$, on obtient $T_n \mu(U) \sim T_n \nu(U)$ et donc $\mu(U)=\nu(U)$, ce qui est absurde et conclut. \[lem:AmuPartout\] Tout ouvert de ${\mathcal S}_\infty$ contient un point appartenant à un ensemble $A(\mu)$ pour une certaine mesure invariante ergodique $\mu$. Soit $U$ un ouvert de ${\mathcal S}_\infty$ ; c’est la limite d’une suite d’ouverts $U_n \in {\mathcal S}_n$. De plus, $\operatorname{Leb}(U_n) \to \operatorname{Leb}(U)>0$. Notons $V_n=f_n^{-1}(U_n) \subset {\mathcal S}$. Comme l’application $f_n : {\mathcal S}\to {\mathcal S}_n$ envoie la mesure de Lebesgue de ${\mathcal S}$ sur la mesure de Lebesgue de ${\mathcal S}_n$, $\operatorname{Leb}(V_n)$ ne tend pas vers $0$. Montrons qu’il existe une mesure $\mu$ invariante ergodique telle que $\mu(V_n)$ ne tende pas vers $0$, par l’absurde. On écrit la désintégration ergodique de $\operatorname{Leb}$, comme $\operatorname{Leb}= \int_{\Omega} \mu_\xi {{\rm d}\hspace{-0.5pt}}P(\xi)$ où $(\Omega, P)$ est un espace probabilisé et $(\mu_\xi)_{\xi\in \Omega}$ parcourt l’ensemble des mesures invariantes ergodiques. Si tous les $\mu_\xi(V_n)$ tendaient vers $0$, alors par convergence dominée $\operatorname{Leb}(V_n)$ tendrait également vers $0$, ce qui est absurde. Fixons donc $\mu$ telle que $\mu(V_n)$ ne tende pas vers $0$. Alors $$\mu\{ x{\; : \; }x\text{ appartient \`{a} une infinit\'{e} de }V_n\}>0.$$ En particulier, il existe $x\in B(\mu)$ qui appartient à une infinité de $V_n$. La suite $f_n(x) \in U_n$ admet alors une valeur d’adhérence dans $\overline{U}$. On peut maintenant conclure. On fixe des petits ouverts $U_1,\dots, U_N$ de la surface limite ${\mathcal S}_\infty$, satisfaisant la propriété suivante : pour tous points $x_i \in U_i$ : 1. pour tous $i,j$, il existe une suite de rectangles plongés ayant des sommets opposés parmi les $x_k$, tels que le premier rectangle ait $x_i$ pour sommet et le dernier ait $x_j$ pour sommet. 2. pour tout $x\in {\mathcal S}_\infty$ qui n’est pas une singularité, il existe un rectangle plongé ayant $x$ et l’un des $x_j$ pour sommets opposés. De tels ouverts existent bien, par un argument de connexité sur la surface ${\mathcal S}_\infty$. L’ouvert $U_1$ contient un point $x_1$ appartenant à un ensemble $A(\mu_1)$, par le lemme \[lem:AmuPartout\]. Quitte à considérer une sous-suite de ${\mathcal S}_n$, on peut même supposer qu’il existe $y_1 \in B(\mu)$ tel que $f_n(y_1)$ converge vers $x_1$. On se restreint à cette sous-suite d’indices. L’ouvert $U_2$ contient un point $x_2$ appartenant à un ensemble $A(\mu_2)$, encore par le lemme \[lem:AmuPartout\], et quitte à extraire on peut encore supposer qu’on a convergence le long de la suite ${\mathcal S}_n$. On répète ainsi ce processus d’extraction $N$ fois, et on obtient $N$ points $x_1,\dots,x_N$ dans $U_1,\dots, U_N$, limites de suites $f_n(y_i)$ avec $y_i \in B(\mu_i)$ pour une certaine mesure invariante ergodique $\mu_i$. La première propriété dans le choix des ouverts $U_i$, combinée avec le lemme \[lem:MuEgaleNu\], montre que toutes les mesures $\mu_i$ sont égales à une même mesure $\mu$. Soit maintenant $\nu$ une autre mesure ergodique invariante. Soit $y\in B(\mu)$, considérons $x$ une valeur d’adhérence dans ${\mathcal S}_\infty$ de la suite $f_n(y)$, disons que $x$ est la limite d’une suite $f_{j(n)}(y)$. Si $x$ n’est pas une singularité, il existe un rectangle plongé dans ${\mathcal S}_\infty$ ayant $x$ et l’un des $x_i$ comme sommets. En appliquant le lemme \[lem:MuEgaleNu\] à la suite ${\mathcal S}_{j(n)}$, on obtient $\mu=\nu$. Si $x$ est une singularité, on ne peut pas appliquer le lemme \[lem:MuEgaleNu\], mais on peut encore conclure en reprenant sa preuve et en utilisant le fait que $f_{j(n)}(y)$ n’est pas une singularité (les détails géométriques sont laissés au lecteur). Dans tous les cas, on obtient $\mu=\nu$. On a ainsi montré qu’il y avait une seule mesure invariante, ce qui conclut donc la preuve du théorème $A$. Preuve du théorème B {#PreuveThmB} ==================== Rappelons que nous devons démontrer que, si ${\mathcal S}$ est une surface de translation, alors l’ensemble $\{\theta\in {\mathbb{S}}^1,\ {\mathcal S}\textrm{ part \`{a} l'infini dans la direction }\theta\}$ est de mesure nulle. On va raisonner par l’absurde pour démontrer le théorème B. Donnons tout d’abord une traduction “naïve” de la négation du théorème B. Si ce théorème est faux pour une surface ${\mathcal S}$, alors il existe un sous-ensemble $S$ de ${\mathbb{S}}^1$ de mesure de Lebesgue non nulle tel que la surface ${\mathcal S}$ tend vers l’infini dans la direction $\theta$, pour tout $\theta \in S$. Autrement dit, pour tout ${\varepsilon}>0$ et pour tout $\theta \in S$, il existe $T({\varepsilon},\theta)$ tel que, pour tout $t>T({\varepsilon},\theta)$, la surface ${\mathcal S}_{\theta,t}$ admet une connexion de selles de longueur au plus ${\varepsilon}$. On choisit $T({\varepsilon},\theta)$ minimal, de telle sorte que la fonction $\theta \mapsto T({\varepsilon},\theta)$ est mesurable. Par conséquent, il existe un ensemble $S_{\varepsilon}\subset S$ de mesure au moins $\operatorname{Leb}(S)/2$ tel que $T_{\varepsilon}:=\sup_{\theta\in S_{\varepsilon}} T({\varepsilon},\theta)$ soit fini. Toutes les surfaces ${\mathcal S}_{\theta, T_{\varepsilon}+1}$ pour $\theta \in S_{\varepsilon}$ ont alors une connexion de selles de longueur au plus ${\varepsilon}$. En appliquant un procédé diagonal à la suite ${\varepsilon}_n=1/n$, on obtient la conclusion suivante. \[prop:RencontrePTpetit\] Soit ${\mathcal S}$ une surface de translation qui ne satisfait pas la conclusion du théorème B. Alors il existe des sous-ensembles $S_n$ de ${\mathbb{S}}^1$ avec $\inf_{n\in{\mathbb{N}}} \operatorname{Leb}(S_n)>0$, une suite $T_n \to \infty$ et une suite ${\varepsilon}_n \to 0$ tels que : pour tout $n\in {\mathbb{N}}$, pour tout $\theta \in S_n$, la surface ${\mathcal S}_{\theta, T_n}$ a une connexion de selles de longueur au plus ${\varepsilon}_n$. Malheureusement, cette négation du théorème B est trop naïve pour amener directement à une contradiction. On aura besoin de la version forte suivante : \[prop:RencontrePasPetit\] Soit ${\mathcal S}$ une surface de translation ne satisfaisant pas la conclusion du théorème B. Alors il existe $c>0$, des sous-ensembles $S_n$ de ${\mathbb{S}}^1$ avec $\inf_{n\in {\mathbb{N}}} \operatorname{Leb}(S_n)>0$, une suite $T_n \to \infty$ et une suite ${\varepsilon}_n \to 0$ tels que : pour tout $n\in {\mathbb{N}}$, pour tout $\theta \in S_n$, la surface ${\mathcal S}_{\theta, T_n}$ a une connexion de selles de longueur au plus ${\varepsilon}_n$, qui ne rencontre pas d’autre connexion de selles de longueur $\leq c$. Pour démontrer cette proposition, on devra travailler avec des ensembles de connexions de selles courtes et deux à deux disjointes (par disjointes, on entend que les intérieurs des connexions de selles ne se rencontrent pas). Pour ce faire, on utilisera l’outil géométrique suivant : Soit ${\mathcal S}$ une surface de translation. Un *complexe* de ${\mathcal S}$ est une partie $K$ de ${\mathcal S}$ qui est triangulable et dont les bords sont des connexions de selles. On suppose de plus que, si trois connexions de selles appartiennent à $K$ et bordent un triangle ne contenant pas de singularité, alors ce triangle est inclus dans $K$. Si $K$ est un complexe, on note $\operatorname{Aire}(K)$ sa surface, $|\partial K|$ le maximum des longueurs des connexions de selles formant le bord de $K$, et $\operatorname{Comp}(K)$ la complexité de $K$, i.e., le nombre de connexions de selles nécessaire pour trianguler $K$. Notons que $\operatorname{Comp}(K)$ est bien défini puisque deux triangulations différentes de $K$ utilisent le même nombre de connexions de selles, par la formule d’Euler. Comme toute triangulation de $K$ peut être complétée en une triangulation de ${\mathcal S}$ (par \[ExisteTriangulation\]), et comme le nombre d’arêtes d’une triangulation de ${\mathcal S}$ est constant, on en déduit que la complexité d’un complexe est uniformément bornée par une constante ne dépendant que du genre de ${\mathcal S}$ et du nombre de singularités. La proposition suivante permet de construire des complexes de plus en plus grands (pour l’inclusion). \[prop:ConstruitComplexe\] Soient $K$ un complexe, et $\sigma$ une connexion de selles qui traverse $\partial K$ ou qui ne rencontre pas $K$. Alors il existe un complexe $K'$ contenant $K$ avec 1. $\operatorname{Comp}(K')> \operatorname{Comp}(K)$. 2. $|\partial K'| \leq 2|\partial K| + |\sigma|$. 3. $\operatorname{Aire}(K') \leq \operatorname{Aire}(K)+|\partial K|^2 + |\partial K|\cdot |\sigma|$. Notons que la proposition n’affirme *pas* que $K'$ contient $\sigma$ ! C’est de la géométrie à l’ancienne que n’aurait pas désavouée Pythagore. On traite plusieurs cas. Dans tous les cas, on construira $K'$ qui contient strictement $K$, la borne sur la complexité est donc triviale. Les autres bornes seront également des conséquences directes de la construction géométrique. \(1) Si $\sigma$ ne rencontre pas $K$ dans son intérieur. On ajoute simplement $\sigma$ à $K$ pour former $K'$ (et on remplit les triangles éventuels bordés par $\sigma$ et deux connexions de selles déjà dans $K$). \(2) Si $\sigma$ rencontre $K$, mais une des extrémités de $\sigma$ n’est pas dans $K$, appelons-la $P$. Partant de $P$, on parcourt $\sigma$ jusqu’à rencontrer $K$, en un point $H$ qui appartient à une connexion de selles $[AB]$ du bord de $K$, comme sur la figure \[Augmente1\]. ![\[Augmente1\]](augmente1.eps){width="11cm"} On redresse le chemin $AHP$ petit à petit, jusqu’à rencontrer une singularité, puis on continue le redressement pour la portion restante. On obtient une ligne brisée reliant $A$ à $P$ qui délimite, avec $[AH]$ et $[HP]$, une zone de ${\mathcal S}$ ne contenant pas de singularité. Si l’une des connexions de selles de la ligne brisée n’est pas dans $K$, on l’ajoute à $K$ pour former $K'$. Comme la ligne brisée est de longueur au plus $|AH|+|AP|$, $K'$ ainsi formé vérifie la conclusion de la proposition, i.e. $|\partial K'| \leq |\partial K| + |\sigma|$. Si toutes les connexions de selles de la ligne brisée sont dans $K$, on répète le processus de l’autre côté, entre $B$ et $P$. Le cas qui pose problème est lorsque cette nouvelle ligne brisée ne contient que des connexions de selles dans $K$. Si l’une des lignes brisées entre $A$ et $P$, ou $B$ et $P$, n’est pas un segment, elle contient un point intermédiaire $Q$. En reliant $Q$ à $A$ ou $B$, on obtient une connexion de selles qui n’est pas dans $K$ (puisque $K$ ne rencontre pas $]HP[$), et en l’ajoutant à $K$ on obtient un complexe $K'$ qui vérifie les propriétés requises. Il reste le cas où $[AP]$ et $[BP]$ sont deux connexions de selles appartenant à $K$. Mais alors, les connexions de selles $[AP]$, $[BP]$ et $[AB]$ appartiennent à $K$ et bordent un triangle sans singularité dans son intérieur. La condition que nous avons imposée dans la définition des complexes implique que $K$ contient le triangle $ABP$, ce qui est une contradiction. \(3) Sinon, une partie de l’intérieur de $\sigma$ n’est pas contenue dans $K$, mais traverse de part et d’autre le bord de $K$, comme sur la figure \[Augmente2\]. ![\[Augmente2\]](augmente2.eps){width="13cm"} On déplace alors $\sigma$ parallèlement à elle-même, dans une direction, jusqu’à rencontrer une singularité. Si on s’est ramené au cas (1) ou (2), on a gagné (le nouveau $\sigma$, translaté, est de longueur au plus $|\sigma|+|\partial K|$, et on vérifie que le $K'$ qu’on forme alors vérifie bien les bornes requises). Sinon, on a rencontré une ligne parallèle à $\sigma$, appartenant au bord de $K$ (contenant éventuellement des singularités dans son intérieur, et éventuellement réduite à un point). On essaie alors de translater $\sigma$ dans l’autre direction, et soit on se ramène à (1) ou (2), soit on a finalement une situation comme dans le dernier cas de la figure \[Augmente2\]. De plus, le quadrilatère ainsi formé n’est pas un triangle sans singularité au bord (il appartiendrait alors déjà à $K$, par définition d’un complexe). On va donc pouvoir relier une singularité du bord supérieur à une du bord inférieur, pour former $K'$, qui convient. On peut utiliser cette construction géométrique pour démontrer la proposition \[prop:RencontrePasPetit\]. Le c[œ]{}ur de la preuve va être de faire une récurrence sur la complexité de certains complexes. Comme la complexité des complexes est uniformément bornée, cette récurrence s’arrêtera à un certain moment. On part de ${\mathcal S}$ une surface de translation qui ne satisfait pas la conclusion du théorème B. Pour $i\in {\mathbb{N}}$, on note $(P_i)$ la propriété suivante : il existe ${\varepsilon}_n \to 0$, $T_n \to \infty$ et $S_n \subset {\mathbb{S}}^1$ avec $\inf_{n\in {\mathbb{N}}}\operatorname{Leb}( S_n) >0$ tels que, pour tout $\theta \in S_n$, la surface ${\mathcal S}_{\theta, T_n}$ admet un complexe $K(n,\theta)$ de complexité $\geq i$, avec $\operatorname{Aire}(K)\leq {\varepsilon}_n$ et $|\partial K|\leq {\varepsilon}_n$. On sait déjà que ${\mathcal S}$ satisfait la propriété $(P_1)$ : un complexe de complexité $1$ est simplement une connexion de selles, et la proposition \[prop:RencontrePTpetit\] donne donc la conclusion de $(P_1)$. Supposons maintenant que $(P_i)$ est satisfait, et considérons les suites ${\varepsilon}_n, T_n, S_n$ et $K(n,\theta)$ correspondantes. Soit $c>0$. Si la mesure de “l’ensemble des $\theta$ tels que $\partial K(n,\theta)$ ne rencontre pas de connexion de selles de longueur $\leq c$” ne tend pas vers $0$, on prend une connexion de selles dans le bord de $K(n,\theta)$, elle vérifie la conclusion de la proposition \[prop:RencontrePasPetit\] et on a gagné. Notons que $\partial K(n,\theta)$ est non vide pour $n$ assez grand puisque l’aire de $K(n,\theta)$ tend vers $0$, ce qui garantit que $K(n,\theta)$ ne recouvre pas toute la surface. Sinon, démontrons $(P_{i+1})$. Par un procédé diagonal (en prenant une suite de valeurs de $c$ qui tend vers $0$), on obtient une suite $\delta_n$ qui tend vers $0$, une sous-suite $T'_n$ de $T_n$ (disons $T'_n=T_{j(n)}$) et des ensembles $S'_n$ de mesure uniformément minorée (inclus dans $S_{j(n)}$) tels que, pour tout $\theta \in S'_n$, on ait un complexe $L(n,\theta)=K(j(n),\theta)$ de complexité au moins $i$ avec $|\partial L|\leq \delta_n$ et $|\operatorname{Aire}(L)|\leq \delta_n$, ainsi qu’une connexion de selles $\sigma(n,\theta)$ de longueur au plus $\delta_n$ et qui rencontre le bord de $L$. En appliquant la proposition \[prop:ConstruitComplexe\] à $L$ et $\sigma$, on obtient un complexe $K'$ de complexité $\geq i+1$ d’aire et de complexité petites, bornées disons par ${\varepsilon}_n'=3\delta_n$. On a établi la propriété $(P_{i+1})$. Si la surface ${\mathcal S}$ ne vérifiait pas la conclusion de la proposition \[prop:RencontrePasPetit\], on démontrerait ainsi $(P_1)$, puis $(P_2)$, puis par récurrence $(P_i)$ pour tout $i$. Ceci est impossible puisqu’il existe une borne uniforme sur la complexité des complexes de ${\mathcal S}$. Ainsi, la proposition est démontrée. La contradiction viendra de la proposition suivante. \[prop:PetiteMesure\] Soient ${\mathcal S}$ une surface de translation et $c>0$. Notons $A(t,{\varepsilon},c)$ l’ensemble des $\theta \in {\mathbb{S}}^1$ tels que ${\mathcal S}_{\theta,t}$ a une connexion de selles de longueur au plus ${\varepsilon}$ qui ne rencontre pas d’autre connexion de selles de longueur au plus $c$. Il existe alors des constantes $\bar {\varepsilon}>0, T>0$ et $K>0$ telles que, pour tout $t\geq T$, pour tout ${\varepsilon}\leq \bar{\varepsilon}$, $\operatorname{Leb}(A(t,{\varepsilon},c)) \leq K {\varepsilon}$. Posons $A=A(t,{\varepsilon},c)$, on veut estimer $\operatorname{Leb}(A)$. Soit $\theta \in A$, on considère une connexion de selles $\alpha$ (dans ${\mathcal S}$) de longueur au plus ${\varepsilon}$ et ne rencontrant pas de connexion de selles de longueur au plus $c$ (dans ${\mathcal S}_{\theta,t}$). On note $I_\alpha$ l’ensemble des angles $\theta'$ tels que $\alpha$ est de longueur au plus ${\varepsilon}$ dans ${\mathcal S}_{\theta',t}$, et $J_\alpha$ l’ensemble des angles $\theta'$ tels que $\alpha$ est de longueur au plus $c$ dans ${\mathcal S}_{\theta',t}$. Ces ensembles sont des réunions de deux intervalles opposés sur le cercle ${\mathbb{S}}^1$. Quitte à considérer une seule des deux composantes (ou à projectiviser), on fera comme si $I_\alpha$ et $J_\alpha$ étaient des intervalles. Soit $\hat J_\alpha$ l’intervalle moitié de $J_\alpha$ (i.e., l’intervalle de même centre que $J_\alpha$ et de longueur moitié). Finalement, on note $K_\alpha$ l’ensemble des angles $\theta'$ tels que $\alpha$ soit de longueur au plus ${\varepsilon}$ dans ${\mathcal S}_{\theta',t}$, et ne rencontre pas de connexion de selles de longueur $\leq c$. Si ${\varepsilon}$ est assez petit et $t$ est assez grand, on a $$\{ \theta\} \subset K_\alpha \subset I_\alpha \subset \hat J_\alpha \subset J_\alpha.$$ Un petit calcul trigonométrique montre que, pour une certaine constante $K$, $\operatorname{Leb}(I_\alpha) \leq K{\varepsilon}\operatorname{Leb}(J_\alpha)$. On en déduit $\operatorname{Leb}(I_\alpha) \leq 2K {\varepsilon}\operatorname{Leb}( \hat J_\alpha)$. Notons $B$ l’ensemble des connexions de selles $\alpha$ telles que $K_\alpha \not=\emptyset$. Ce sont les connexions de selles qui servent à définir $A$. Autrement dit, $A=\bigcup_{\alpha \in B} K_\alpha$. Soient $\alpha,\beta \in B$ deux connexions de selles distinctes. Si $\hat J_\alpha$ et $\hat J_\beta$ s’intersectent, alors $J_\alpha$ recouvre $\hat J_\beta$ ou inversement. Dans le premier cas, comme $K_\beta \subset \hat J_\beta$, on obtient un angle $\theta \in K_\beta \cap J_\alpha$. Dans la surface de translation ${\mathcal S}_{\theta,t}$, $\alpha$ est donc de longueur au plus $c$. Par définition de $K_\beta$, on en déduit que $\alpha$ et $\beta$ ne s’intersectent pas. Plus généralement, si $\alpha_1,\dots, \alpha_k \in B$ sont deux à deux distinctes et $\bigcap \hat J_{\alpha_i} \not=\emptyset$, on montre de même que les connexions de selles $\alpha_i$ sont deux à deux disjointes. Mais le nombre de connexions de selles deux à deux disjointes est uniformément borné en fonction du genre de la surface ${\mathcal S}$ et du nombre de singularités (puisqu’on peut compléter tout ensemble de connexions de selles disjointes pour former une triangulation de la surface, qui a un nombre d’arêtes fixé). Il existe donc $p<\infty$ tel que chaque angle de ${\mathbb{S}}^1$ appartienne à au plus $p$ intervalles $\hat J_\alpha$. Finalement, $$\begin{aligned} \operatorname{Leb}(A) & = \operatorname{Leb}( \bigcup_{\alpha\in B} K_\alpha) \leq \sum_{\alpha \in B} \operatorname{Leb}(I_\alpha) \leq 2 K {\varepsilon}\sum_{\alpha\in B} \operatorname{Leb}( \hat J_\alpha) \\& \leq 2Kp {\varepsilon}\operatorname{Leb}( \bigcup_{\alpha \in B} \hat J_\alpha) \leq 2Kp {\varepsilon}\operatorname{Leb}({\mathbb{S}}^1). \qedhere \end{aligned}$$ On peut maintenant conclure : Les propositions \[prop:RencontrePasPetit\] et \[prop:PetiteMesure\] sont manifestement incompatibles l’une avec l’autre, puisque l’ensemble $S_n$ donné par la première proposition a une mesure bornée inférieurement, alors que la seconde proposition montre que $\operatorname{Leb}(S_n)=O({\varepsilon}_n) \to 0$. Traduction de la preuve du théorème KMS en genre $1$ ==================================================== Dans ce paragraphe, on explicite différemment, avec un point de vue plus géométrique, les objets qui sont apparus dans la preuve du théorème KMS, dans le cas des tores. La notion essentielle de “départ à l’infini” utilisée dans la preuve va se reformuler en termes de “l’espace des tores plats”. Notons que, pour pouvoir parler de systole, il faut avoir au moins une singularité (éventuellement artificielle, i.e., avec $k_i=0$). On travaillera donc sur des tores ayant un point marqué (plus ou moins arbitraire). Soit $(e_1,e_2)$ une base directe de ${\mathbb R}^2$. On peut alors former une surface de translation de genre $1$, avec un point marqué, comme suit : on part du polygone $P=\{x e_1 +y e_2,\ 0 \leq x \leq 1, 0 \leq y \leq 1\}$ puis on recolle ses côtés opposés parallèles, comme en \[par:RecollePolygones\]. On obtient ainsi une surface de translation ${\mathcal S}$, sur laquelle on marque l’image de l’origine. De plus, tout tore plat avec un point marqué peut s’obtenir de cette manière. Soit ${\mathcal T}_1$ l’ensemble des bases directes de ${\mathbb R}^2$, cet ensemble fournit donc une paramétrisation de “l’espace des tores plats avec un point marqué”. Notons $e_1=\left(\begin{smallmatrix}a \\ b \end{smallmatrix}\right)$ et $e_2=\left( \begin{smallmatrix}c \\ d \end{smallmatrix}\right)$. On a une identification entre ${\mathcal T}_1$ et $\textrm{GL}^{+}_2({\mathbb R})$ donnée par $(e_1,e_2) \mapsto A_{(e_1,e_2)}= \left(\begin{smallmatrix}a & b \\ c & d \end{smallmatrix}\right)$ (le fait d’avoir transposé dans cette identification nous servira plus loin). Si ${\mathcal T}^{(1)}_1$ désigne l’ensemble des tores d’aire $1$ alors ${\mathcal T}^{(1)}_1 = {\textrm{SL}_2(\mathbb R)}$ via l’identification précédente. Analysons l’action de ${\textrm{SL}_2(\mathbb R)}$, décrite dans la section \[sec:action\], sur ${\mathcal T}_1$. Si $M\in {\textrm{SL}_2(\mathbb R)}$ alors l’image par $M$ du tore plat donné par $(e_1,e_2)$ est le tore plat donné par $(Me_1,Me_2)$. Via l’identification ci-dessus, $M$ agit donc sur $\textrm{GL}^{+}_2({\mathbb R})$ comme la multiplication à droite par la matrice transposée $M^{t}$. En particulier, l’action du flot géodésique sur ${\mathcal T}^{(1)}_1$ correspond à l’action par multiplication à droite de $g_t$ sur ${\textrm{SL}_2(\mathbb R)}$. L’ensemble ${\mathcal T}_1$ décrit ci-dessus n’est pas très satisfaisant. En effet, deux bases directes $(e_1,e_2)$ et $(e'_1,e'_2)$ peuvent engendrer deux surfaces de translation ${\mathcal S}$ et ${\mathcal S}'$ isomorphes (i.e., il existe un difféomorphisme de ${\mathcal S}$ dans ${\mathcal S}'$ dont la différentielle dans l’atlas de translation est partout égale à l’identité). Notons $\mathcal M_1$ le quotient de ${\mathcal T}_1$ par cette relation d’équivalence. On va décrire $\mathcal M_1$ via l’identification ci-dessus. Deux bases directes $(e_1,e_2)$ et $(e'_1,e'_2)$ sont équivalentes si et seulement si il existe une matrice $B=\left(\begin{smallmatrix}x & y\\ z & t \end{smallmatrix}\right)$ dans $\textrm{SL}_2({\mathbb Z})$ telle que $e'_1=xe_1+ye_2$ et $e'_2=ze_1+te_2$, i.e., si $(e_1,e_2)$ et $(e'_1,e'_2)$ sont deux bases d’un même réseau. Cette condition se lit encore $A_{(e'_1,e'_2)}=B A_{(e_1,e_2)}$. Ainsi, $\mathcal M_1$ est identifié au quotient (à gauche) $\textrm{SL}_2({\mathbb Z}) \setminus \textrm{GL}^{+}_2({\mathbb R})$. Soit $\mathcal M_1^{(1)}$ le sous-ensemble de $\mathcal M_1$ formé des surfaces d’aire $1$, il est identifié à $\textrm{SL}_2({\mathbb Z})\setminus {\textrm{SL}_2(\mathbb R)}$. Notons que l’action de $g_t$ passe au quotient sur $\mathcal M^{(1)}_1$ (une action à gauche et une action à droite commutent toujours) et s’interprète algébriquement comme la multiplication à droite de $g_t$ sur $\textrm{SL}_2({\mathbb Z}) \setminus {\textrm{SL}_2(\mathbb R)}$. \[prop:recurrent\] On peut alors relier la notion naïve de départ à l’infini donnée en \[def:PartirInfini\] avec la topologie de $\textrm{SL}_2({\mathbb Z}) \setminus {\textrm{SL}_2(\mathbb R)}$. \[DepartInfiniGenre1\] Un tore $\mathbb T^2 \in \textrm{SL}_2({\mathbb Z}) \setminus {\textrm{SL}_2(\mathbb R)}$ part à l’infini (au sens de la définition \[def:PartirInfini\]) si et seulement si $g_t \mathbb T^2$ quitte tout compact de $\textrm{SL}_2({\mathbb Z}) \setminus {\textrm{SL}_2(\mathbb R)}$ quand $t\to \infty$. Une suite de tores dans $\textrm{SL}_2({\mathbb Z}) \setminus {\textrm{SL}_2(\mathbb R)}$ quitte tout compact si et seulement si la systole tend vers $0$ le long de cette suite. La proposition en découle immédiatement. Ainsi, le théorème A se comprend bien en termes d’espace des réseaux. On aurait aussi pu décrire ${\mathcal T}_1$ comme l’ensemble des atlas plats sur un tore, quotienté par l’ensemble des difféomorphismes qui sont isotopes à l’identité. De ce point de vue géométrique $\mathcal M^{(1)}_1$ devient alors l’ensemble des atlas plats d’aire $1$ sur un tore, quotienté par l’ensemble des difféomorphismes. Cela revient aussi à prendre ${\mathcal T}^{(1)}_1$ modulo le groupe modulaire, i.e., le groupe de tous les difféomorphismes (préservant l’orientation) modulo ceux isotopes à l’identité. En effet si $\phi$ est un difféomorphisme du tore $\mathbb T^2$ et $\{(U_i,z_i)\}_i$ est un atlas plat, alors $\{(\phi(U_i),z_i\circ \phi^{-1})\}_i$ est un autre atlas plat représentant le même tore plat. Ici le groupe modulaire n’est autre que $\textrm{SL}_2({\mathbb Z})$. Ce point de vue va être utile en genre supérieur. Nous allons maintenant redémontrer le théorème B, en genre $1$. Pour cela, il est utile d’introduire un autre point de vue sur l’espace $\textrm{SL}_2({\mathbb Z}) \setminus {\textrm{SL}_2(\mathbb R)}$, celui de la géométrie hyperbolique. Le groupe ${\textrm{SL}_2(\mathbb R)}$ agit sur le demi-plan de Poincaré $\mathbb H=\{z\in {\mathbb C},\ \textrm{Im}(z) > 0 \}$, par $$\left(\begin{matrix} a & b\\c&d \end{matrix}\right)\cdot z= \frac{az+b}{cz+d}.$$ Cette action préserve la métrique hyperbolique. De plus, cette action est transitive et le stabilisateur du point $i\in \mathbb H$ est $\textrm{SO}(2)$, donc $\mathbb H \simeq {\textrm{SL}_2(\mathbb R)}/ \textrm{SO}(2)$, et ${\textrm{SL}_2(\mathbb R)}$ est identifié au fibré unitaire tangent à $\mathbb H$. Un domaine fondamental pour l’action (à gauche) de $\textrm{SL}_2({\mathbb Z})$ sur $\mathbb H$ est présenté figure \[fig:surface:modulaire\]. La surface modulaire est par définition le quotient $\textrm{SL}_2({\mathbb Z})\setminus \mathbb H$. Cette surface est isomorphe à une sphère privée d’un point avec deux singularités pour la métrique hyperbolique. Notons que les deux singularités correspondent au tore plat carré et au tore plat hexagonal. L’espace des surfaces de translation de genre $1$ avec un point marqué, i.e. $\textrm{SL}_2({\mathbb Z}) \setminus {\textrm{SL}_2(\mathbb R)}$, est donc identifié au fibré unitaire tangent à la surface modulaire. De plus, l’action de $g_t$ correspond au flot géodésique sur la surface modulaire. ![ \[fig:surface:modulaire\] Un domaine fondamental pour l’action de $\textrm{SL}_2({\mathbb Z})$ sur $\mathbb H$. ](modulaire.eps){width="8cm"} \[surface:veech\] On peut maintenant redémontrer en genre $1$ une version forte du théorème B, en utilisant des propriétés bien connues du flot géodésique sur la surface modulaire. Soit ${\mathcal S}$ une surface de translation de genre $1$ avec un point marqué. Alors $$\label{eq:PartInfiniGenre1} \{\theta\in {\mathbb{S}}^1,\ {\mathcal S}\textrm{ part \`{a} l'infini dans la direction }\theta\}$$ est au plus dénombrable. On utilisera la propriété suivante du flot géodésique sur la surface modulaire : il existe un compact $K$ tel que toute géodésique qui ne part pas verticalement dans le cusp revient une infinité de fois dans $K$ (i.e. les géodésiques qui ne partent pas vers le cusp sont récurrentes, c’est la proposition du point \[prop:recurrent\]). Soit ${\mathcal S}$ une surface de translation avec un point marqué, considérons $x\in \textrm{SL}_2({\mathbb Z})\setminus {\textrm{SL}_2(\mathbb R)}$ le point correspondant, et $y=\pi(x)$ son image dans la surface modulaire $\textrm{SL}_2({\mathbb Z})\setminus \mathbb H$. L’ensemble est alors en bijection avec les géodésiques issues de $y$ qui partent à l’infini dans la surface modulaire. Pour tout $n>0$, l’ensemble $G_n$ des géodésiques partant de $y$ et qui sont verticales dans le cusp après le temps $n$ est discret, donc fini par compacité de ${\mathbb{S}}^1$. L’ensemble , en bijection avec $\bigcup G_n$, est donc dénombrable, ce qui conclut la preuve. Cette preuve ne s’adapte pas en genre supérieur (et le résultat de dénombrabilité devient faux !) car les propriétés hyperboliques du flot $g_t$ deviennent beaucoup plus faibles qu’en genre $1$. La morale du théorème B est qu’il reste cependant suffisamment d’hyperbolicité pour obtenir des résultats précis. L’espace des modules des surfaces de translation {#sec:espace:module} ================================================ Comme nous venons de le voir dans la section précédente, le bon cadre conceptuel pour démontrer le théorème KMS est d’introduire une structure de “variété” (plus précisément d’orbifold) sur “l’espace des surfaces de translation” (en genre fixé et avec une combinatoire également fixée). Pour que cette notion soit utile, on voudrait avoir un analogue de la proposition \[DepartInfiniGenre1\], i.e., le départ à l’infini au sens de la définition \[def:PartirInfini\] doit correspondre au fait de quitter tout compact de l’espace des surfaces de translation. Dans cette section, sans rien démontrer, on va décrire une telle construction. Fixons ${\mathcal S}$ une surface compacte connexe, de genre $g$. Fixons aussi $\Sigma = \{P_1,\dots,P_n\}$ un sous-ensemble fini de ${\mathcal S}$, et $\kappa=(k_1,\dots, k_n)$ des entiers tels que $\sum (k_i+1)=2g-2$. On note ${\mathcal{A}}({\mathcal S}, \kappa)$ l’ensemble des structures de translation sur $X$ ayant des singularités coniques d’angles $2\pi(k_i+1)$ en $P_i$. C’est un ensemble d’atlas, il est énorme (et en particulier beaucoup trop gros pour qu’on puisse travailler avec lui). On peut néanmoins le munir naturellement d’une topologie : si $\omega \in {\mathcal{A}}({\mathcal S}, \kappa)$ et ${\varepsilon}>0$, on définit le ${\varepsilon}$-voisinage de $\omega$ comme étant l’ensemble des atlas de translation $\omega'$ sur ${\mathcal S}$ tels que, pour tout $x\in {\mathcal S}{-}\Sigma$, les composées $z\circ {z'}^{-1}$ et $z'\circ z^{-1}$ de cartes de translation autour de $x$ (correspondant à $\omega$ et $\omega'$) ont des dérivées ${\varepsilon}$-proches de l’identité. Intuitivement, $\omega'$ est dans le ${\varepsilon}$-voisinage de $\omega$ si $\omega'$ et $\omega$ sont d’accord à ${\varepsilon}$ près sur la direction des vecteurs dans ${\mathbb R}^2$, leur norme, les horizontales, les verticales, etc. Soit ${\mathcal T}_g({\mathcal S},\kappa)$ obtenu en identifiant deux éléments $\omega$ et $\omega'$ de ${\mathcal{A}}({\mathcal S},\kappa)$ s’il existe un homéomorphisme de ${\mathcal S}$ fixant les $P_i$, isotope à l’identité (relativement aux $P_i$) et envoyant $\omega$ sur $\omega'$. On munit ${\mathcal T}_g({\mathcal S},\kappa)$ de la topologie quotient. C’est “l’espace de Teichmüller” des surfaces de translation. En général, il faut se méfier de la topologie quotient, mais ici tout va bien : l’espace ${\mathcal T}_g(\kappa)$ est séparé, métrisable, localement compact, et même localement homéomorphe à un certain espace ${\mathbb C}^d$ (ici $d$ est $2g+n-1$ comme nous allons le voir). On peut même aller plus loin et munir ${\mathcal T}_g({\mathcal S}, \kappa)$ d’une structure de variété (analytique), comme suit. Si $\omega\in {\mathcal{A}}({\mathcal S},\kappa)$, on peut utiliser la structure de translation pour relever les chemins sur ${\mathcal S}$, en partant de $0$. L’extrémité du relèvement est invariante par homotopie, et on obtient donc une application de $H_1({\mathcal S}, \Sigma ; {\mathbb Z})$ dans ${\mathbb R}^2$, i.e., un élément du groupe de cohomologie $H^1({\mathcal S},\Sigma; {\mathbb R}^2)$. Cet élément ne change pas si l’on remplace $\omega$ par une structure de translation qui lui est isotope. En passant au quotient, on obtient donc une application canonique $$\Theta : {\mathcal T}_g({\mathcal S},\kappa) \to H^1({\mathcal S}, \Sigma ; {\mathbb R}^2).$$ Pour la topologie définie ci-dessus sur ${\mathcal T}_g({\mathcal S}, \kappa)$, l’application $\Theta$ est un homéomorphisme local. On peut donc utiliser $\Theta$ comme carte locale, pour mettre une structure de variété sur ${\mathcal T}_g(\kappa)$. Dans le cas du genre $1$, ${\mathcal T}_1$ correspond simplement à $\textrm{GL}^{+}_2({\mathbb R})$ (avec sa structure de variété usuelle). Si $g \geq 2$, on obtient bien que ${\mathcal T}_g(\kappa)$ est une varitété de dimension (réelle) $2(2g+n-1)$. Pour obtenir un espace plus petit, avec de meilleures propriétés de compacité, et donc de récurrence pour la dynamique, il faut quotienter encore plus. Soit ${\mathcal{M}}_g(\kappa)$ le quotient de ${\mathcal T}_g(\kappa)$ par l’action du groupe modulaire. Autrement dit, on identifie deux éléments $\omega$ et $\omega'$ de ${\mathcal{A}}$ si $\omega'$ s’obtient à partir de $\omega$ par un difféomorphisme de ${\mathcal S}$ fixant les $P_i$ (mais pas nécessairement isotope à l’identité). L’action du groupe modulaire sur ${\mathcal T}_g(\kappa)$ n’est pas libre, mais presque. Ainsi, ${\mathcal{M}}_g(\kappa)$ n’est pas naturellement muni d’une structure de variété (il a des singularités), mais presque : c’est un *orbifold*, i.e., il est localement difféomorphe à un espace ${\mathbb C}^d$ sauf en un nombre fini de points. On notera aussi ${\mathcal{M}}_g^{(1)}(\kappa)$ l’ensemble des surfaces dans ${\mathcal{M}}_g(\kappa)$ d’aire $1$. Dans ce cadre, le résultat de compacité naïf qu’on a démontré dans le corollaire \[cor:Compacite\] se traduit de la manière suivante. Notons $\operatorname{sys}(\omega)$ la systole de $\omega\in {\mathcal{M}}_g^{(1)}(\kappa)$. Pour tout ${\varepsilon}>0$, l’ensemble des $\omega \in {\mathcal{M}}_g^{(1)}(\kappa)$ tels que $\operatorname{sys}(\omega) \geq {\varepsilon}$ est compact. Ainsi, une suite $\omega_n$ sort de tout compact si et seulement si $\operatorname{sys}(\omega_n) \to 0$. La définition de départ à l’infini donnée en \[def:Convergence\] coïncide donc avec celle découlant de la topologie naturelle sur l’espace des modules. L’espace $H^1({\mathcal S}, \Sigma ; {\mathbb R}^2)$ est muni d’une mesure de Lebesgue canonique (donnant covolume $1$ au réseau $H^1({\mathcal S}, \Sigma ; {\mathbb Z}^2)$). En la tirant en arrière par $\Theta$, on obtient une mesure canonique sur ${\mathcal T}_g(\kappa)$. Cette mesure est invariante sous l’action du groupe modulaire, et passe donc au quotient sur ${\mathcal{M}}_g(\kappa)$. Elle induit même une mesure sur ${\mathcal{M}}_g^{(1)}(\kappa)$. Un résultat important de Masur et Veech (voir [@Masur:82; @Veech:82]) affirme que cette mesure est de masse finie sur ${\mathcal{M}}_g^{(1)}(\kappa)$. De plus elle préservée par le flot $g_t$. Masur et Veech montrent même que le flot $g_t$ est ergodique sur toute composante connexe de ${\mathcal{M}}_g^{(1)}(\kappa)$ (qui a au plus trois composantes connexes).\ Quand on parle de “presque toute surface de translation”, on fait référence à cette mesure canonique. \[surface:reseau\] Un cas particulièrement intéressant est fourni lorsque le stabilisateur $$\textrm{SL}({\mathcal S},\omega):=\{A\in {\textrm{SL}_2(\mathbb R)},\ A\cdot ({\mathcal S},\omega) \simeq ({\mathcal S},\omega)\}$$ d’une surface de translation $({\mathcal S},\omega)$ est un réseau de $\textrm{SL}_2({\mathbb R})$. On dit alors que $({\mathcal S},\omega)$ est une [*surface de Veech*]{}. L’idée de la preuve donnée en \[surface:veech\] peut alors s’adapter (la propriété essentielle que nous avons utilisée est le fait que $\textrm{SL}_2({\mathbb Z})$ est un réseau dans $\textrm{SL}_2({\mathbb R})$). On peut alors obtenir des propriétés remarquables sur le flot directionnel sur ${\mathcal S}$ (voir [@Veech:89]). Notamment, pour une surface de Veech, les directions minimales pour le flot directionnel sont [*exactement*]{} les directions uniquement ergodiques. Raffinements, résultats supplémentaires {#sec:raffinements} ======================================= Dans ce paragraphe, on discute les raffinements possibles aux théorèmes $A$ et $B$. Tout d’abord, pour le théorème A, on a équivalence entre non-unique ergodicité et départ à l’infini, en genre $1$. Cela reste vrai pour les surfaces de Veech (voir \[surface:reseau\]) ([@Veech:89]). Mais ce n’est plus vrai en général. Mentionnons les résultats suivants. 1. Pour toute fonction $f(t)$ qui tend vers $0$ quand $t\to \infty$, il existe une surface ${\mathcal S}$ telle que $\operatorname{sys}(g_t {\mathcal S})$ tend vers $0$ quand $t \to \infty$, mais $f(t)=o ( \operatorname{sys}(g_t {\mathcal S}))$ (voir [@Cheung:slow]). Autrement dit, on peut converger arbitrairement lentement vers l’infini. 2. Pour toute composante connexe de l’espace des modules, il existe une constante $c>0$ telle que, si $t^{-c}=o( \operatorname{sys}(g_t {\mathcal S}))$ alors le flot vertical sur ${\mathcal S}$ est uniquement ergodique. Autrement dit, si on tend vers l’infini mais suffisamment lentement, alors la conclusion du théorème $A$ reste valable. Ce théorème n’est pas vide puisqu’il existe effectivement des géodésiques qui partent lentement vers l’infini d’après le point précédent. 3. Pour toute composante connexe de l’espace des modules, il existe une constante $c' >c$ et une surface ${\mathcal S}$ telle que $t^{-c'}=o(\operatorname{sys}(g_t{\mathcal S}))$, mais le flot vertical sur ${\mathcal S}$ n’est pas uniquement ergodique. Ainsi, la vitesse de décroissance en $t^{-c}$ est vraiment la “vitesse critique”. Ces deux derniers points sont dus à Cheung et Eskin (voir [@Cheung:Eskin]). On pourra aussi consulter [@Cheung:Masur]. 4. Il existe une version combinatoire du théorème A (voir [@Boshernitzan]).\ Une section de Poincaré du flot $\mathcal F_\theta$ sur un intervalle fournit un échange d’intervalles $T$. Les points de discontinuités correspondent aux séparatrices. Notons $m_n$ la longueur du plus petit intervalle échangé par $T^n$. Alors, si $T$ n’est pas uniquement ergodique, $nm_n \rightarrow \infty$ quand $n\rightarrow \infty$. Pour le théorème $B$, on peut se demander quelle est la taille précise de l’ensemble $$\{\theta\in {\mathbb{S}}^1,\ {\mathcal S}\textrm{ part \`{a} l'infini dans la direction }\theta\}.$$ On sait déjà qu’il est de mesure nulle. En fait on peut dire un peu plus sur le sous-ensemble suivant : $$\begin{gathered} \label{equation:veech} \{\theta\in {\mathbb{S}}^1,\textrm{ le flot } \mathcal F_\theta \textrm{ sur } {\mathcal S}\textrm{ n'est pas uniquement ergodique} \} \subset \\ \{\theta\in {\mathbb{S}}^1,\ {\mathcal S}\textrm{ part \`{a} l'infini dans la direction }\theta\}.\end{gathered}$$ On peut chercher à estimer sa dimension de Hausdorff. On sait déjà qu’en genre $1$, l’ensemble des directions exceptionnelles coïncide avec les rationnels et est donc dénombrable. Ce n’est plus vrai en genre plus grand, mais : 1. Pour toute surface de translation, la dimension de Hausdorff de cet ensemble est au plus $1/2$ (voir [@Masur:HD]). 2. Il existe des surfaces de translation pour lesquelles cette dimension est exactement $1/2$ [@Cheung] (ces exemples proviennent de billards carrés avec un mur, voir figure \[fig:exemple:4:slit\]). 3. Pour toute composante connexe de l’espace des modules ${\mathcal{M}}_g(\kappa)$, il existe une constante $\delta \in ]0,1/2]$ telle que, pour presque toute surface de translation dans cette composante connexe, la dimension en question est égale à $\delta$ [@Masur:Smillie:1]. 4. Pour les surfaces de Veech (voir \[surface:reseau\]) l’inclusion (\[equation:veech\]) est une égalité ([@Veech:89]).

ArXiv

--- abstract: 'With its excellent spatial resolution, low background, and hard-band response, the [*Chandra*]{} X-ray Observatory is ideal for performing exploratory surveys. These efficient, sensitive observations can place constraints on fundamental properties of a quasar continuum including the X-ray luminosity, the ratio of X-ray to UV power, and the X-ray spectral shape. To demonstrate the power of such surveys to provide significant insight, we consider two examples, a Large Bright Quasar Survey sample of broad absorption line quasars and a sample of Sloan Digital Sky Survey (SDSS) quasars with extreme C [iv]{} blueshifts. In both cases, exploratory [*Chandra*]{} observations provide important information for a physical understanding of UV spectroscopic differences in quasars.' author: - 'Sarah C. Gallagher' - 'Gordon T. Richards' - 'W. Nielsen Brandt & George Chartas' title: 'The Power of Exploratory *Chandra* Observations' --- \#1[[*\#1*]{}]{} \#1[[*\#1*]{}]{} = \#1 1.25in .125in .25in Introduction ============ X-ray emission appears to be a universal signature of quasar spectral energy distributions, confirming expectations from accretion physics. Based on rapid variability of soft X-rays in conjunction with the standard black-hole paradigm, these photons are believed to be emitted from the region immediately surrounding the black hole. Energetically, X-rays are significant, contributing 2–20$\%$ of the bolometric luminosity. X-ray observations are thus an important component of any multi-wavelength campaign to probe quasar populations. The excellent spatial resolution of the [*Chandra*]{} High Resolution Mirror Assembly and the effective background rejection of the ACIS instrument make this combination uniquely powerful for quasar surveys. For reference, during a 5 ks observation, the 0.5–8.0 keV background within a $2\arcsec$-radius source region is typically $\sim$0.1 ct. Because ACIS is photon-limited even beyond $100$ ks (Alexander et al. 2003), the point-source detection limit scales [*linearly*]{} with exposure time, unlike the $\sqrt{t} $ dependence common in other wavelength bands. In conjunction with sub-arcsec positional accuracy, known optical point sources can be robustly detected with 3–5 photons. In 5 ks, this corresponds to a flux of $\sim7\times10^{-15}$ erg cm$^{-2}$ s$^{-1}$ for a typical quasar X-ray spectrum. A 3–7 ks [*Chandra*]{} exposure, the regime of exploratory observations, is generally insufficient for gathering enough X-rays for spectral analysis of a quasar. However, the strategy of exploratory observations enables the extension of results from spectroscopic observations of individual targets to larger, well-defined samples, and the investigation of connections between X-ray properties and other wavelength regimes. From these datasets, standard X-ray observables are flux, hardness ratio,[^1] and $\alpha_{\rm ox}$.[^2] We briefly describe the initial results from two exploratory [*Chandra*]{} quasar surveys to illustrate the utility of this observing strategy. Other examples in the literature of successful applications of this approach to quasar studies include surveys of high-$z$ (e.g., Brandt et al. 2002; Vignali et al. 2003), red (Wilkes et al. 2002), and X-ray weak (Risaliti et al. 2003) quasars. X-ray Insights from the LBQS BAL Quasar [*Chandra*]{} Survey ============================================================ We are in the process of performing the largest exploratory survey to date of a well-defined sample of broad absorption line (BAL) quasars drawn from the Large Bright Quasar Survey (LBQS). Since BAL quasars are known to be very faint X-ray sources (e.g., Green & Mathur 1996; Gallagher et al. 1999), exploratory observations are the only means of observing sufficient numbers to determine the X-ray properties of the population as a whole. The [*Chandra*]{} data alone are revealing. As seen in Figure 1a, the hardness ratio appears to be anti-correlated with $\alpha_{\rm ox}$. This indicates that the X-ray weakest BAL quasars have the hardest spectra, consistent with the understanding from spectroscopic observations of a handful of objects (e.g., Gallagher et al. 2002) that the X-ray spectra are heavily absorbed. Examining the connection between the X-ray and UV absorption properties of the quasars has also placed observational contraints on quasar disk-wind models (Gallagher et al. 2003). In addition to exploratory [*Chandra*]{} observations, this sample is also being targeted by both SCUBA (PI Priddey) and [ *SIRTF*]{} to characterize the submm through hard X-ray spectral energy distributions of BAL quasars as a whole. The Connection Between C [iv]{} Blueshift and X-ray Properties ============================================================== High-ionization broad emission lines such as C [iv]{} have been known to yield redshifts systematically lower than those measured from Mg [ii]{} (e.g., Tytler & Fan 1992); i.e., these C [iv]{} lines are blueshifted relative to the systemic velocity. In a study of $\sim 800$ SDSS quasars with $1.5 \le z \le 2.2$, Richards et al. (2002) found that the C [iv]{}–Mg [ ii]{} velocity shifts ranges over $\ge2000$ km s$^{-1}$. Furthermore, the C [iv]{} blueshift (hereafter C4B) was correlated with UV properties, most notably the relative $\Delta(g-i)$ color (Richards et al. 2003). That is, the bluest quasars typically exhibited the largest C4Bs. Positing that the C4B might result from the orientation of the accretion disk or the opening angle of the disk wind, we proposed an exploratory [ *Chandra*]{} survey to investigate this hypothesis. Six targets, three each from the extreme ends of the C4B distribution, were approved for Cycle 4 observations. While any trends based on six data points need verification, the initial results from this small survey are intriguing. Figure 1b shows that hardness appears to increase with $\alpha_{\rm ox}$ for the SDSS C4B quasar sample. Though this trend is not statistically significant (Spearman’s rank-order correlation coefficient, $r_{s}$, is 0.67 for a significance level, $p_{rs}$, of 0.15[^3]), the fact that the hardest sources are not X-ray weaker is relevant. This suggests that the hardness of the spectra may not be due to intrinsic absorption in the same way as we see with the BAL quasars. Extending the study to the UV properties, we tested $\alpha_{\rm ox}$ versus C4B and hardness ratio versus $\Delta(g-i)$ (see Figure 2). Though $\alpha_{\rm ox}$ and C4B are consistent with being uncorrelated ($r_{s}$=$-$0.77, $p_{rs}$=0.07), the hardness ratio is significantly correlated with $\Delta(g-i)$ ($r_{s}$=$-$0.99, $p_{rs}=3\times10^{-4}$). As shown in Figure 2b, the bluer quasars appear to have softer X-ray spectra, i.e., more negative values of the hardness ratio. This is in line with expectation if UV continuum color is solely related to intrinsic obscuration. However, the lack of connection with $\alpha_{\sc ox}$ makes this interpretation uncertain. The connection of UV spectroscopic properties to X-ray emission in these objects implies a physical connection, and more data to investigate this claim are certainly warranted. This experiment also illustrates the value of the SDSS to multiwavelength quasar studies. Given the large number and uniform data quality of the available SDSS quasars, samples can be chosen with precision. Since hardness ratio can vary with $z$ due to absorption and $\alpha_{\rm ox}$ is a function of $l_{\rm 2500}$, the luminosity density at rest-frame 2500 Å (Vignali et al. 2003), sample tuning significantly reduces potential selection biases. In this C4B quasar survey, the redshifts range from $z$=1.65–1.89 and $l_{\rm 2500}$ spans only a factor of $\sim5$. The properties of interest, in this case the C [iv]{} blueshift and $\Delta(g-i)$, are thus more reliably isolated for comparison with the X-ray emission. We acknowledge the support of [*Chandra*]{} X-ray center grants GO1–2105X (SCG, WNB) and GO3–4144A (SCG, GTR, WNB). WNB thanks NASA LTSA grant NAG5–13035. Alexander, D. M., et al. 2003, , 126, 539 Brandt, W. N., et al. 2002, , 569, L5 Green, P. J., & Mathur, S. 1996, , 462, 637 Gallagher, S. C., et al. 1999, , 519, 549 Gallagher, S. C., et al. 2002, , 567, 37 Gallagher, S. C., et al. 2003, AdvSpRes, in press (astro-ph/0212304) Richards, G. T., et al. 2002, , 124, 1 Richards, G. T., et al. 2003, , 126, 1131 Risaliti, G., et al. 2003, , 587, 9 Tytler, D., & Fan, X. 1992, , 79, 1 Vignali, C., et al. 2003, , 125, 2876 Wilkes, B. J., et al. 2002, , 564, 65 [^1]: The hardness ratio is defined to be $(h-s)/f$, where $h$=2–8 keV ct, $s$=0.5–2.0 keV ct, and $f$=0.5–8.0 keV ct. [^2]: The quantity $\alpha_{\rm ox}$ equals $0.384\log(f_{\rm X}/f_{\rm 2500})$ where $f_{\rm X}$ and $f_{\rm 2500}$ are the flux densities at rest-frame 2 keV and 2500 Å, respectively. [^3]: The significance level ranges from 0.0–1.0 with a small value indicating a significant correlation.

ArXiv

--- abstract: 'Recommendation Systems (SR) suggest items exploring user preferences, helping them with the information overload problem. Two approaches to SR have received more prominence, Collaborative Filtering, and Content-Based Filtering. Moreover, even though studies are indicating their advantages and disadvantages, few results empirically prove their characteristics, similarities, and differences. In this work, an experimental methodology is proposed to perform comparisons between recommendation algorithms for different approaches going beyond the “precision of the predictions”. For the experiments, three algorithms of recommendation were tested: a baseline for Collaborative Filtration and two algorithms for Content-based Filtering that were developed for this evaluation. The experiments demonstrate the behavior of these systems in different data sets, its main characteristics and especially the complementary aspect of the two main approaches.' author: - Rafael Glauber and Angelo Loula bibliography: - 'sigproc.bib' title: | Collaborative Filtering vs. Content-Based Filtering:\ differences and similarities --- &lt;ccs2012&gt; &lt;concept&gt; &lt;concept\_id&gt;10002951.10003260.10003261.10003269&lt;/concept\_id&gt; &lt;concept\_desc&gt;Information systems Collaborative filtering&lt;/concept\_desc&gt; &lt;concept\_significance&gt;500&lt;/concept\_significance&gt; &lt;/concept&gt; &lt;concept&gt; &lt;concept\_id&gt;10002951.10003317.10003347.10003350&lt;/concept\_id&gt; &lt;concept\_desc&gt;Information systems Recommender systems&lt;/concept\_desc&gt; &lt;concept\_significance&gt;500&lt;/concept\_significance&gt; &lt;/concept&gt; &lt;concept&gt; &lt;concept\_id&gt;10002951.10003260&lt;/concept\_id&gt; &lt;concept\_desc&gt;Information systems World Wide Web&lt;/concept\_desc&gt; &lt;concept\_significance&gt;300&lt;/concept\_significance&gt; &lt;/concept&gt; &lt;concept&gt; &lt;concept\_id&gt;10002951.10003260.10003261.10003270&lt;/concept\_id&gt; &lt;concept\_desc&gt;Information systems Social recommendation&lt;/concept\_desc&gt; &lt;concept\_significance&gt;300&lt;/concept\_significance&gt; &lt;/concept&gt; &lt;/ccs2012&gt;

ArXiv

--- abstract: 'In recent years, the availability of massive data sets and improved computing power have driven the advent of cutting-edge machine learning algorithms. However, this trend has triggered growing concerns associated with its ethical issues. In response to such a phenomenon, this study proposes a feasible solution that combines ethics and computer science materials in artificial intelligent classrooms. In addition, the paper presents several arguments and evidence in favor of the necessity and effectiveness of this integrated approach. $\newline$' author: - | Tai Vu\ Department of Computer Science\ Stanford University\ taivu@stanford.edu bibliography: - 'pwr1.bib' title: 'Combating The Machine Ethics Crisis: An Educational Approach' --- The Age of Intelligent Machines =============================== $\newline$ Machine learning, a primary branch of Artificial Intelligence (AI), is regarded as one defining technological frontier shaping the 21st century. The advent of intelligent machines that learn and evolve through experience has pervaded a variety of sectors, including science, economics, finance, laws, transportation, and medicine.\ In accordance with this trend, tech giants like Google, Facebook, Microsoft, and Amazon strive to take the lead in the machine learning race. According to McKinsey Global Institute’s report, investment in AI-aided projects in 2017 was between $\$18$ billion and $\$27$ billion [@mckinsey:2017]. Cutting-edge machine learning programs such as spam filtering, image recognition, recommender systems, and text translation are becoming dominant in a whole host of business industries.\ However, as machine learning advances, it has aroused growing concerns over its ethical and societal implications, ranging from unfairness to transparency. To address these existing problems, some renowned universities have incorporated ethics contents into their AI curricula. However, more academic institutions across the world need to adopt this ethics training approach to equip machine learning developers with the knowledge needed for handling ethical issues that arise from their AI-enabled products.\ Machine Learning and Its Beneficial Aspects =========================================== $\newline$ Before delving into the discussions of machine ethics, it is essential to comprehend machine learning’s mechanics. In simple terms, machine learning is a state-of-the-art technology that extracts patterns and trends from datasets. By analyzing an array of observations, the systems employ analytical, probabilistic, and statistical models to devise an estimated mathematical function mapping the attributes of these samples to the needed output. Afterwards, the machine uses this correlation to make predictions on new observations. For instance, a medical diagnosis algorithm trained on gigantic records of patients can forecast the likelihood of individuals having cancer cells based on their age, occupation, daily activities, eating habits, and their families’ medical history.\ The key advantage of machine learning over conventional computer programs is its “learning” nature. Traditional algorithms rely on sets of hand-coded instructions to execute tasks step by step. In contrast, machine learning engineers remove these task-specific procedures and instead show the systems large collections of data points as problem-solving examples. Subsequently, these programs can teach themselves to infer hidden correlations amongst the data and predict outcomes on unseen inputs.\ Machine learning’s predictive power plays a pivotal role in business and economic settings. In fact, governments leverage machine learning to foresee economic downturns and then enact fiscal and monetary policies accordingly. In addition, many companies employ these intelligent programs to analyze their customer data, gaining invaluable insights into customer preferences and market demands. This information supports managers to make informed sales and marketing decisions under time constraints.\ Ethical Consideration:\ What Could Go Wrong? ======================= $\newline$ Despite the aforementioned promises, over-reliance on machine learning comes with a number of ethical problems. These existing limitations demonstrate an urgent need for practical approaches and solutions.\ One pressing issue regarding machine-driven applications is the cost of fairness. In fact, many intelligent systems exhibit signs of unfairness and bias against certain human groups. This phenomenon is illustrated by the case of Amazon’s autonomous recruitment tool, which favored male candidates over their female counterparts. Specifically, this system “reportedly downgraded resumes containing the words “women’s” and filtered out candidates who had attended two women-only colleges” (Hamilton, 2018). In addition, racial discrimination occurred with a risk-assessment product called COMPAS, which assisted judges in Broward County, Florida to select prisoners to let out on bail. A 2016 report by The ProPublica highlighted “significant racial disparities” against black people: “The formula was particularly likely to falsely flag black defendants as future criminals, wrongly labeling them this way at almost twice the rate as white defendants” [@cohen:nd].\ Furthermore, machine-driven data analytics is apt to violate citizens’ privacy and liberty. For example, Facebook’s photo-tagging system is called into question, since it identifies the identity of individuals in any images on Facebook, collecting them to build massive datasets of photos. The facial recognition program draws privacy concerns, as Facebook can track its users’ daily activities via their pictures. Additionally, this photo-scanning feature may be exploited for ungraceful purposes like harassment or bullying. A harasser can simply upload an image of their target, and Facebook “will recognize their face and ping that user, doing the harasser’s work for them” [@vincent:2017].\ Hence, an ethical lens urges caution against the unintended moral consequences of applied machine learning such as unfairness and the violation of privacy. It is necessary to examine how these challenges disrupt people’s daily lives and how to effectively handle such ethical repercussions. The search for viable solutions to these questions demands a collaboration of experts from a wide range of branches, including educators and academic institutions.\ Reforming AI Education with Ethics Training =========================================== $\newline$ It is commonly believed that legislative regulation is the most effective remedy to resolve the machine ethics crisis. Elon Musk, co-founder and CEO of SpaceX, Tesla and OpenAI, urged US governors to quickly regulate AI software before it threatens human civilization [@condliffe:2017]. From this angle, governments should propound moral codes for assessing machine learning programs and outlaw the use of systems that do not conform to these standards. It is true that this approach may partially eradicate unethical machine-driven tools, thereby protecting users’ benefits and human rights.\ Nonetheless, this proposal seems rather impractical, given the technological complexity and abstraction of machine learning. It is observed that the majority of policy-makers lack technical backgrounds in computer science in general and artificial intelligence in particular. US Congress’ questions to Facebook’s CEO Mark Zuckerberg regarding the firm’s data management suggested a lack of fundamental knowledge of modern technologies and straightforward business models derived from them [@stewart:2018]. Meanwhile, machine learning is an advanced field that requires sophisticated understandings of mathematics, statistics and computer science. These fields take adept software developers years to master. There are more than thirty distinct machine learning models, each of which possesses different natures and requires different policy approaches. Without essential expertise, governors face enormous difficulty devising legal frameworks for the use of intelligent machines. These regulations may not be applicable to real-world AI-enabled products. What is more, artificial intelligence is developing at a pace that policy-makers can hardly keep up with. Over the past five years, machine learning has progressed by leaps and bounds, beating humans in multiple spheres, including language translation, image recognition, and cancer detection, and this trend is projected to continue at an accelerating rate [@shoham:2018]. In contrast, the Congress often needs a long time to discuss, evaluate and enact a legislative proposal. Technology would escalate to a new level of sophistication and advance, making the policy obsolete.\ On top of that, while the enforcement of these regulations filters out improper algorithms, it can limit machine learning applications and hinder the power of artificial intelligence. The strict control of data use, for example, can result in a scarcity of massive datasets for the training and validation of intelligent machines. The process of testing and assessing AI programs in government agencies is also highly time-consuming. According to a 2016 government report on AI policy, excessive or inappropriate regulatory responses can create “bottlenecks” that slow down the adoption and development of AI innovations [@president:2016]. Furthermore, the policy approach is insufficient, as it fails to address the root of the problem. Eventually, it is machine learning engineers who design, implement and execute the automated programs. Therefore, it is ideal to equip AI developers with proper ethical foundations, which allow them to produce highly accurate machine learning software that complies with ethical and social standards.\ With growing preferences for AI-driven systems comes great demand for competent machine learning engineers, which further calls for ethics training in current AI-focused curricula. An employment report by Indeed reveals that, “with an average salary base of $\$146,085$ and a whopping $344\%$ growth in job postings, machine learning engineer is an extremely promising position” [@indeed:2019]. These career prospects accompany a surge in machine learning education: “2017 introductory AI enrollment was $3.6\times$ that of 2012, while 2017 introductory ML course enrollment was $5\times$ that of 2012” [@shoham:2018]. However, the ethical components of these technical classes have been underappreciated. A survey on machine learning engineers shows that roughly $12\%$ of them regard ethics as important, while only $5\%$ employ ethical knowledge when pursuing their AI careers [@wollowski:2016]. If this phenomenon remains unaddressed, a myriad of machine-aided products will be created without moral consideration, thereby exacerbating the current situation. Thus, these statistics illustrate the role of ethics contents in universities’ AI curricula, as such a practice would transform the minds of millions of machine learning developers who drive future artificial intelligence advances.\ Promoting Transparency ====================== $\newline$ A growing ethical worry amongst governments and citizens is a lack of algorithmic transparency in AI-enabled systems. An ethics-oriented AI curriculum would provide machine learning engineers with an ethical mindset, which encourages them to make their programs transparent to the public’s eyes and resolve potential moral concerns.\ The need for explainable machine learning is worthy of consideration, since opacity is a common problem amongst many learning models. Since a multitude of currently-used algorithms are derived from complex statistical and mathematical principles, it is tough for their developers to explain their mechanics to the general public. The notion of such “black box” algorithms exceptionally holds for deep learning, an upgraded version of machine learning. This model is implemented on neural networks, which encompass thousands of interconnected neurons distributed into a chain of sequential layers. This structure enables deep learning to incorporate many machine learning algorithms into a single framework, handle massive datasets, and uplevel its predictive accuracy. However, this productivity comes at the cost of transparency. “Once \[the network\] becomes very large, and it has thousands of units per layer and maybe hundreds of layers, then it becomes quite un-understandable,” suggests Jaakkola, an MIT computer science professor [@knight:2017]. In other words, deep learning is a “dark black box”, even for capable software engineers and computer scientists. Recognizing such complexities, many AI practitioners currently employ available open-source deep learning libraries, such as Keras, Pytorch, TensorFlow, Scikit-learn, MXNet, and Caffe, without paying attention to their inner workings [@lorica:2019]. What they frequently do is to plug in inputs, wait for their computers to run and then get desired outputs.\ The “black box” nature of machine learning algorithms has several detrimental effects. Although they seem to perform competently in the meantime, there is no guarantee that an unethical machine-driven decision would not occur in the future. When that happens, AI engineers will have a tough time understanding its underlying causes and alleviating its consequences. Additionally, AI’s opaque property draws moral criticisms from governors and the public. As an illustration, job candidates who get rejected by machine-based recruitment systems and customers who are denied loans by banks’ automated profiling processes may question whether these programs treat them unfairly and contain biases. In response to such matters, the European Union introduced the General Data Protection Regulation, censoring the use of inscrutable artificial intelligence techniques. A paper by Bryce Goodman and Seth Flaxman, two researchers at the University of Oxford, illustrates that the “right to explanation” in this new framework, which allows users to ask for explicit explanations of algorithmic decisions, poses a challenge to many “black box” algorithms like random forests, support vector machines, and neural networks [@goodman:2017]. Even though deep learning offers high accuracy and efficiency, it cannot be put into practice.\ The case of opaque machine learning programs drives the necessity of educated AI developers who has the ability to explain the functioning behind training algorithms. That is one learning goal of “CS181: Computers, Ethics, and Public Policy”, a popular course at Stanford University. This type of computer science classes raises students’ awareness of ethical issues stemming from non-transparent machine learning [@cs181:2020]. Such a practice motivates young computer scientists to put efforts into understanding factors that constitute the interpretability of a model and aiming for the explanation of these automated processes. With these toolkits, they can design and utilize complicated architectures like deep learning with great confidence, gain users’ trust, and lessen public ethical worries.\ A large number of machine learning practitioners can reap certain benefits from the widespread teaching of these algorithmic transparency contents. Specifically, they are equipped with practical ways of purposefully developing and deploying interpretable models. Engineers may limit the number of variables and parameters in the training phase, which diminishes the number of uncovered correlations. Additionally, they can opt for applied models with comprehensible learning outputs like decisions tree, which allows users to easily walk through branches of a tree structure without getting confused by obscure parameters [@selbst:2018]. In addition, the ethics knowledge acts as an incentive for AI developers to delve into the internal process of “black box” architectures like deep neural networks rather than ignoring them and focusing entirely on inputs and outputs. By acquiring in-depth understandings of the models, they are able to simplify them and study practical tools to deliver the outcomes of these programs to unfamiliar stakeholders. As the opacity is alleviated, engineers also face fewer hurdles searching for and fixing problems like unfair selections and biases, thereby mitigating existing moral concerns over these systems.\ What is more, the teaching of machine-related transparency benefits policy-making practices. By demystifying how intelligent machines reach final answers, developers can aid governmental agencies to make policy-related choices on the validity and applicability of training models and assess their ethical impacts on society. In this way, high-performance programs like deep learning can be permitted, while their functionality is kept under the government’s control. One may propound an alternative idea of instructing policy-makers the mechanics of machine learning, so that they can make decisions without AI engineers’ support. However, this measure is infeasible, since governors with no technical background would struggle with extensive mathematics, computer science and statistics know-how. The proposal is also unnecessary, since policy-makers only need to comprehend the rationale behind machine-driven decisions instead of in-depth mathematical formulas and concepts. Hence, it seems practical to introduce the knowledge of transparency to software developers. Having undertaken years of intensive training, they have fewer obstacles when digging deep into the functioning of machine learning models and translating it into simple words.\ Heightening Developers’ Data Integrity ====================================== $\newline$ In addition to resolving the question of algorithmic transparency, teaching ethics in AI classes can stimulate data integrity amongst machine learning developers. This improvement plays an indispensable role in combating ethical problems of intelligent machines, because training data is a principal part of the learning process.\ One critical pattern that AI practitioners need to be aware of is unintended biases in the datasets. In fact, majority groups often dominate the datasets, and this dominance can be amplified in the training process, which contributes to highly biased outcomes. One such case is indicated in the research “Gender Shades” by MIT Media Lab. By evaluating facial recognition products by IBM, Microsoft, and Face++, they pointed out key discriminations in terms of gender, skin type, and ethnicity. In particular, “$95.9\%$ of the faces misgendered by Face++ were those of female subjects,” while “$93.6\%$ of faces misgendered by Microsoft were those of darker subjects.” Such flaws arise as the training data is “overwhelmingly composed of lighter-skinned subjects.” Those “substantial disparities” in the computer vision systems demand the “urgent attention” of AI developers to the importance of cleaning hidden biases in the input data [@buolamwini:2018].\ To handle such issues, machine learning instructors should teach their students a code of ethics and fairness in processing data. This approach helps learners accumulate hands-on understandings of the primary sources of biases in training datasets. Specifically, the data may lack information about minority groups like African-American residents or contain prejudices coming from the history and culture of human beings. Without learning these concepts, engineers would be likely to neglect potential biases and plug the raw datasets directly into AI-enabled systems in the face of the pressure of quickly extracting useful insights from their industries. Consequently, their algorithms would inherit the discriminations and produce flawed outcomes, which can scale uncontrollably over computer systems. By contrast, research has shown that possessing ethical knowledge allows developers to withstand pressure from business concerns and appreciate the significance of cleaning data and removing biases [@dodig:2003].\ Several skeptics may believe that the awareness of data biases is of no help, as these characteristics and preferences are the nature of real-world data and are challenging to address. However, these critics overlook the fact that students in ethics-oriented AI courses grasp typical types of discrimination, including gender, race, skin type, and ethnicity. In many cases, they can calculate simple statistics to target these kinds of biases. Besides that, the teaching of ethics content sharpens machine learning engineers’ competence in high-level data preparation and bias detection. A fruitful model is “CS294: Fairness in Machine Learning” at the University of California Berkeley. This class facilitates their students’ discussion of data and unfairness in social contexts and enables them to harness advanced statistical modeling and causal inference methods, such as Simpson’s paradox, measurement theory, sampling, and unsupervised learning, to attack the core of these flaws [@cs294:2017]. If such a format is popularized, all AI students will develop a penchant for handling the data preparation process and uncovering biases properly. This way, machine neutrality is guaranteed and a major element of machine ethics crisis is mitigated.\ Ethical Designs of Machine Learning =================================== $\newline$ Asides from hidden data biases, ethical training in AI courses can guide ethical decisions throughout the development stages of computer programs, so that machine learning engineers have the ability to make moral algorithmic designs and implementation choices.\ Ethical dilemma facing developers’ programming process is an important subject that AI courses in some US universities start to cover. For instance, the class “The Ethics and Governance of Artificial Intelligence” at Massachusetts Institute of Technology allows students to scrutinize “The Moral Machine experiment”, an online platform collecting public viewpoints on how self-driving cars should distribute the harms amongst distinct stakeholders in car crashes [@mit:2017]. Such techniques grant students access to global preferences including sparing more lives and sparing young lives, which are “essential building blocks” when they construct their algorithms. It is also important for them to obtain key variations by gender, religiosity, geographic features, and economic strengths: “both male and female respondents indicated preference for sparing females, but the latter group showed a stronger preference ... the preference to spare younger characters rather than older characters is much less pronounced for countries in the Eastern cluster, and much higher for countries in the Southern cluster” [@awad:2018].\ Some critics may argue that knowledge of diverging ethical perspectives is unnecessary for machine learning engineers, since their job is solely to devise high-performance models. This stance is superficial, because without understandings of these diverse standpoints, developers are inclined to make unreasonable coding assumptions that match their own sets of moral values, and the final outputs may contrast with others’ social norms and rules. They may also follow traditional optimization purposes, and even though their products like driverless cars maximize the number of protected people, they do not take into account varied ethical standards and thus trigger ethical concerns in different regions. Therefore, AI courses should instruct students to grasp conflicting viewpoints about ethical codes for machine learning, which is attributable to marked discrepancies in genders, cultural values, and living standards. Research in ethics training demonstrates that this expertise “\[enables\] the AI \[practitioners\] to reach a more ethically comprehensive position, allowing \[them\] to deploy familiar modes of reasoning while challenging \[them\] to look beyond \[their\] own utility and personal concerns" [@goldsmith:2017].\ Furthermore, AI professors ought to provide their students with ethical frameworks to reason about their algorithmic designs. This way, students are motivated to delve into machine ethics, looking at various schools of ethical thought from philosophical, historical, cultural, economic and legal standpoints. Therefore, young machine learning engineers are able to develop a comprehensive roadmap of ethical theory and determine ethical choices introduced in their machine learning projects. In fact, the effectiveness of in-depth moral reasoning is substantiated in a paper on ethics education: “By understanding the reasoning structure of the different theories, a practitioner is better equipped to follow the ramifications of her own values and judgments, and - once she has seen their implications - to reconsider those judgments and values” [@goldsmith:2017]. For that reason, Stanford University offers the interdisciplinary course “CS122: Artificial Intelligence - Philosophy, Ethics, and Impact” that examines ethical analysis in relation to autonomous machines. This is a striking example to follow, since it arms students with intellectual tools and ethical foundation to think critically and analytically about their implementation decisions and “successfully navigate the coming age of intelligent machines” [@cs122:2014].\ The Merits of an Integrated Approach ==================================== $\newline$ While the introduction of ethics in AI classrooms is needed, the way of delivering such knowledge to machine learning students is equally important. Some educators may believe that universities should force AI students to take separate ethics courses as part of their major requirements instead of mixing computer science and ethics. These specifically designed courses allow learners to interact with prominent, knowledgeable experts and develop specialized understandings of ethical reasoning from the ground up. On the other hand, these special courses seem conceptually demanding and less appealing to science-oriented students who want to get a taste of how ethics is correlated to machine learning products. A case study has illustrated that this approach may be counterproductive. As computer science students often experience intense workload, “even the most conscientious students would begin skipping classes and not studying if they thought they could get away with it” [@unger:2005].\ In contrast, blending ethics into machine learning classes is a more captivating and effective measure. One clear benefit is that when AI professors teach computer ethics, they serve as role models who employ moral reasoning to enhance their machine learning algorithms. Additionally, the course staff can invite renowned ethics faculty members to instruct in-depth ethical theories, thereby guaranteeing the quality and depth of moral contents for their students. Meanwhile, since these classes do not emphasize the pure modes of ethical thought and the evolution of human morals, learners are not obligated to read wordy textbooks and monotonous papers on history, humanities, and philosophy. Instead, students’ work focus on real-world ethical and technological implications on different parties, including their jobs and lives, so this emphasis cultivates their interest in learning the subject.\ On top of that, these integrated classes can pick a discussion-based format, which facilitates students’ collaborative learning experience. For example, in “CS181: Computers, Ethics, and Public Policy” at Stanford University, instructors often pose engaging questions and provide guidance for learners to formulate answers in groups. Students can play the roles of distinct stakeholders like computer scientists, policy-makers, businesses, and ordinary citizens and discuss with their peers about real case studies in ethical codes of machine learning forecasts [@cs181:2020]. Empirical research has indicated that these instructor-guided discussions and role-playing activities in ethics classrooms act as catalysts to “productive conversations”, “cogent arguments”, and the acquisition of diverse standpoints, which positively correlate to students’ efforts and academic performances [@quinn:2006].\ What is more, homework assignments should combine technical problem-solving with ethical reasoning to optimize learners’ scholastic competence. For illustration, CS181 offers coding challenges that require students to incorporate algorithmic fairness into their AI-aided software [@cs181:2020]. For computer science majors, these practical problems are more engaging than reading and analyzing lengthy research papers, while still meet the goal of ethical training. Besides that, students gain a fresh outlook on designing actual machine learning algorithms under ethical consideration. A case study has suggested the advantage of combining coding skills with ethical analysis, which supports students to “connect their roles as professionals with their roles as moral agents” [@alenskis:1997]. Thus, this is a promising model for other colleges to follow.\ Another strategy to boost the effectiveness of machine ethics education is called Embedded EthiCS, currently experimented at Harvard University. This program embeds ethics mini-modules into AI courses, giving students practical competence in thinking through ethical challenges [@grosz:2018]. One merit of this approach is that students’ grasp of new machine learning algorithms is accompanied by their ethical implications, which give learners comprehensive insights into the use cases of the technology. In addition, by partitioning ethical thoughts into small components in different courses, this curriculum design removes the barriers of theoretical ethics for computer science undergrads, because it allows for gradual, slow-paced acquisition of ethics contents, as suggested by education research [@alenskis:1997]. This practice ultimately increases the amount of ethical analysis gained in the long run as students fulfill their AI specialization. Moreover, this strategy brings ethics to a wider range of machine learning students, since computer science courses often have large capacities, compared to limited spots in small ethics-focused classrooms.\ Guaranteeing Social Responsibility ================================== $\newline$ In addition to dealing with ethical issues, the inclusion of ethical topics in AI courses helps evoke social responsibility amongst young researchers. With machine learning skills and an awareness of moral and social challenges facing humanity, developers are motivated to put their knowledge into practice and take the initiative to better their community. To fuel that inspiration, Stanford University offers “CS21SI: AI for Social Good”, which dives into fundamental machine learning techniques and their ethical and societal implications [@cs21si:2019]. Panel discussions with guest speakers from academia and industry broaden students’ understanding of the peril of biased and immoral machine learning and the merits of using AI in a socially conscious manner. Afterwards, students work on hands-on programming projects addressing severe ethical and social issues in various branches, not limited to the tech sector. Not only does this approach challenge students to think about their roles, missions and impacts on society, but it also inspires them to apply their machine learning expertise in social good domains, including education, government, and healthcare. Thus, this class is an effective model that every AI curriculum should adopt.\ Outside the traditional classroom setting, academic institutions should promote socially conscious mindsets by supporting student-run groups focused on moral computer science in general and ethical AI in particular. Research has suggested that student organizations substantially contribute to students’ involvement in undergraduate education and reinforce their academic knowledge [@nadler:1997]. To that end, Stanford has invested in several clubs like CS + Social Good and EthiCS ([@cssg:nd], [@wagner:2017]). These groups create open conversations where their members voice their viewpoints about core ethical issues of intelligent machines. CS + Social Good also partners with research labs and companies, enabling students to conduct real-world ethical AI projects in social good spaces. Such clubs complement formal AI and ethics education, since they sharpen students’ ethical reasoning skills and encourage students’ ethical behaviors and responsibilities of returning back to their communities.\ The Way Forward =============== $\newline$ In summary, incorporating ethics into machine learning curricula is of paramount importance to addressing current AI-related ethical issues. This model proves fruitful in some top-notch universities, and it needs to be widely spread to other academic institutions. By fostering ethical education in the AI sector, universities and colleges can produce socially minded machine learning engineers, who can drive ethically sound machine-aided products and mitigate moral concerns. In the long run, they will shape technological innovations, open the door to new data science revolutions, and bolster the progress of artificial intelligence and human intelligence.

ArXiv

--- author: - | Diaaeldin Taha\ University of Washintgon in Seattle, <dtaha@uw.edu> title: 'The Boca-Cobeli-Zaharescu Map Analogue for the Hecke Triangle Groups $G_q$' --- Introduction ============ For any integer $Q \geq 1$, the *Farey sequence at level $Q$* is the set $$\mathcal{F}(Q) = \{a/q \mid a,q \in {\mathbb{Z}},\ 0 \leq a \leq q \leq Q,\ \gcd(a, q) = 1\}$$ of irreducible fractions in the interval $[0, 1]$ with denominators not exceeding $Q$, arranged in increasing order. The Farey sequence is one of the famous enumerations of the rationals, and its applications permeate mathematics. Some of the fundamental properties of $\mathcal{F}(Q)$ are the following: 1. If $a_1/q_1, a_2/q_2 \in \mathcal{F}(Q)$ are two consecutive fractions, then $0 < q_1, q_2 \leq Q$, and $q_1 + q_2 \geq Q$. 2. If $a_1/q_1, a_2/q_2 \in \mathcal{F}(Q)$ are two consecutive fractions, then they satisfy the *Farey neighbor* identity $a_2 q_1 - a_1 q_2 = 1$. 3. If $a_1/q_1, a_2/q_2, a_3/q_3 \in \mathcal{F}(Q)$ are three consecutive fractions, then they satisfy the *next-term* identities $$a_3 = k a_2 - a_1,$$ and $$q_3 = k q_2 - q_1$$ where $k = \left\lfloor\frac{Q + q_1}{q_2}\right\rfloor$. Around the turn of the new millennium, F. P. Boca, C. Cobeli, and A. Zaharescu [@Boca2001-he] encoded the above properties of the Farey sequence as the *Farey triangle* $$\mathscr{T} := \{(a, b) \mid 0 < a, b \leq 1,\ a + b > 1\},$$ and what is now increasingly known as the Boca-Cobeli-Zaharescu (BCZ) map $T : \mathscr{T} \to \mathscr{T}$[^1] $$T(a, b) := \left(b, -a + \left\lfloor\frac{1 + a}{b}\right\rfloor b \right)$$ which satisfies the property that $$T\left(\frac{q_1}{Q}, \frac{q_2}{Q}\right) = \left(\frac{q_2}{Q}, \frac{q_3}{Q}\right)$$ for any three consecutive fractions $a_1/q_1, a_2/q_2, a_3/q_3 \in \mathcal{F}(Q)$. Since then, quoting R. R. Hall, and P. Shiu [@Hall2003-it], the aforementioned trio have “made some very interesting applications” of $\mathscr{T}$ and $T$ (and the weak convergence of particular measures on $\mathscr{T}$ supported on the orbits of $T$ to the Lebesgue probability measure $dm = 2 da db$) to the study of distributions related to Farey fractions. Earlier in the current decade, J. Athreya, and Y. Cheung [@Athreya2013-ql] showed that the Farey triangle $\mathscr{T}$, the BCZ map $T : \mathscr{T} \to \mathscr{T}$, and the Lebesgue probability measure $dm = 2\,da\,db$ on $\mathscr{T}$ form a Poincaré section with roof function $R(a,b) = \frac{1}{ab}$ to the horocycle flow $h_s = \begin{pmatrix}1 & 0\\-s & 1\end{pmatrix}$, $s \in {\mathbb{R}}$, on $X_2 = {\operatorname{SL}}(2,{\mathbb{R}})/{\operatorname{SL}}(2,{\mathbb{Z}})$, with (a scalar multiple of) the Haar probability measure $\mu_2$ inherited from ${\operatorname{SL}}(2,{\mathbb{R}})$. Following that, analogues of the BCZ map have been computed for the golden L translation surface (whose ${\operatorname{SL}}(2, {\mathbb{R}})$ orbit corresponds to ${\operatorname{SL}}(2, {\mathbb{R}})/G_5$, where $G_5$ is the Hecke triangle group $(2, 5, \infty)$) by J. Athreya, J. Chaika, and S. Lelievre[^2] in [@Athreya2015-nq], and later for the regular octagon by C. Uyanick, and G. Work in [@Uyanik2016-gx]. In both cases, the sought for application was determining the slope gap distributions for the holonomy vectors of the golden L and the regular octagon. Soon after, B. Heersink [@Heersink2016-hg] computed the BCZ map analogues for finite covers of ${\operatorname{SL}}(2, {\mathbb{R}})/{\operatorname{SL}}(2, {\mathbb{Z}})$ using a process developed by A. M. Fisher, and T. A. Schmidt [@Fisher2014-ke] for lifting Poincaré sections of the geodesic flow on ${\operatorname{SL}}(2, {\mathbb{R}})/SL(2, {\mathbb{Z}})$ to covers of thereof. In that case, the sought for application was studying statistics of various subsets of the Farey sequence. In this paper, we derive the BCZ map analogue for the Hecke triangle groups $G_q$, $q \geq 3$, which are the subgroups of ${\operatorname{SL}}(2, {\mathbb{R}})$ with generators $$S := \begin{pmatrix}0 & -1 \\ 1 & 0\end{pmatrix}, \text{ and } T_q := \begin{pmatrix}1 & \lambda_q \\ 0 & 1\end{pmatrix},$$ where $\lambda_q := 2 \cos\left(\frac{\pi}{q}\right) \geq 1$. Along the way, we investigate the discrete orbits $$\Lambda_q = G_q(1, 0)^T$$ of the linear action of $G_q$ on the plane ${\mathbb{R}}^2$, and present some results on the geometry of numbers, Diophantine properties, and statistics of $\Lambda_q$. Our starting point is showing that the orbits $\Lambda_q$ have a tree structure that extend the famous Stern-Brocot trees for the rationals. The said trees were studied a bit earlier by C. L. Lang and M. L. Lang in [@Lang2016-qs], though their focus was on the Möbius action of $G_q$ on the hyperbolic plane. ![The elements of $\Lambda_5$ in the square $[0, 100]^2$ generated using \[theorem: next term algorithm\] and \[remark: practical next term algorithm\].[]{data-label="figure: G_5 vectors"}](G_5_vectors.pdf) An earlier version of this paper was announced in October 2018 under the title “The Golden L Ford Circles”, which only considered $G_5$ and its Ford circles. An excellent paper [@Davis2018-al] by D. Davis and S. Lelievre that investigates the $G_5$-Stern-Brocot tree as a tool for studying the periodic paths on the pentagon, double pentagon, and golden L surfaces was announced at the same time. We strongly recommend the aforementioned paper as a more geometrically flavored application of the said trees. Organization ------------ This paper is organized as follows: - In \[section: discrete orbits\], we characterize and study the discrete orbits of the linear action of $G_q$ on the plane ${\mathbb{R}}^2$ (\[proposition: column - unimodular pair identification\]), show that those discrete orbits have a tree structure analogous to the Stern-Brocot trees for the rationals (\[theorem: G\_q Stern Brocot process is well-defined and exhaustive\]), and derive the Boca-Cobeli-Zaharescu map analogues for $G_q$ (\[theorem: G\_q BCZ maps\]). We also characterize the periodic points for the $G_q$-BCZ map analogues (\[corollary: characterizing BCZ periodic points\]), and present an algorithm for generating the elements of $\Lambda_q$ in increasing order of slope (\[theorem: next term algorithm\]). We also collect some consequences of the existence of $G_q$-Stern-Brocot trees that we use throughout the paper in \[corollary: odds and ends\]. - In \[section: cross section\], we give the Poincaré cross sections to the horocycle flow on the quotients ${\operatorname{SL}}(2, {\mathbb{R}})/G_q$ corresponding to the $G_q$ BCZ map analogues we have in section 2 (\[theorem: cross section\]). As a consequence, we get an equidistribution result (\[theorem: weak limit and asymptotic growth\]) that we use for the applications in \[section: applications\]. - In \[section: applications\], we present a number of applications of the results in this paper to the statistics of subsets of $\Lambda_q$. In particular, we give the main asymptotic term for the number of elements of $\Lambda_q$ in homothetic dilations of triangles (\[proposition: counting in triangles\]), equidistribution of homothetic dilations $\frac{1}{\tau}\Lambda_q$ in the square $[-1, 1]^2$ as $\tau \to \infty$ (\[corollary: equidistribution in the square\]), the slope gap distribution for the elements of $\Lambda_q$ (\[corollary: limiting distribution of slopegap\]), and the distribution of the Euclidean distance between successive $G_q$-Ford circles (\[corollary: limiting distribution of centdist\]). We also get a weak form of the Dirichelet approximation theorem for $\Lambda_q$ for free (\[proposition: weak Dirichelet approximation\]). Notation -------- As is customary when working with the groups $G_q$, we write $$U_q := T_q S = \begin{pmatrix}\lambda_q & -1 \\ 1 & 0\end{pmatrix}.$$ The matrix $U_q$ is conjugate to a rotation with angle $\pi/q$, and can be easily seen to preserve the quadratic form $$Q_q((x,y)^T) = x^2 - \lambda_q xy + y^2$$ when $U_q$ acts linearly on the plane ${\mathbb{R}}^2$. The main object that we study in this paper is the orbit of the vector $(1, 0)^T \in {\mathbb{R}}^2$ under the linear action of $G_q$ on the plane $$\Lambda_q = G_q(1, 0)^T.$$ The set $\Lambda_q$ is symmetric against the lines $y = \pm x$, $x = 0$, and $y = 0$ since $G_q$ contains $S^3 T_q^{-1} S = T_q^T$, $S^3 = S^T$, and $(T_qS)^q = -\operatorname{Id}_2$. Of special significance to us are the elements $$\mathfrak{w}_i^q = (x_i^q, y_i^q) = U_q^i (1, 0)^T,$$ where $i = 0 , 1, \cdots, 2 q - 1$. Note that $\mathfrak{w}_0^q = (1, 0)^T$, $\mathfrak{w}_1^q = (\lambda_q, 1)^T$, $\mathfrak{w}_{q-2}^q = (1, \lambda_q)^T$, $\mathfrak{w}_{q-1}^q = (0, 1)^T$, and $\mathfrak{w}_q^q = (-1, 0)^T$. (Since $U_q$ is conjugate to a $\pi/q$-rotation, $U_q^q = -\operatorname{Id}_2$. This gives the last equality.) Moreover, the vectors $\{\mathfrak{w}_i^q\}_{i = 0}^{2q - 1}$ lie on the ellipse $Q_q((x, y)^T) = x^2 - \lambda_q x y + y^2 = 1$. (-2,0)–(2,0) node\[right\][$\scriptstyle x$]{}; (0,-2)–(0,2) node\[above\][$\scriptstyle y$]{}; (0,0) ellipse (2.289cm and 0.743cm); (1,0) circle\[radius=1pt\]; (1.618,1) circle\[radius=1pt\]; (1.618,1.618) circle\[radius=1pt\]; (1,1.618) circle\[radius=1pt\]; (0,1) circle\[radius=1pt\]; (-1,0) circle\[radius=1pt\]; (-1.618,-1) circle\[radius=1pt\]; (-1.618,-1.618) circle\[radius=1pt\]; (-1,-1.618) circle\[radius=1pt\]; (0,-1) circle\[radius=1pt\]; at (1+0.2,0-0.2) [$\scriptstyle \mathfrak{w}_0^5$]{}; at (1.618+0.3,1) [$\scriptstyle \mathfrak{w}_1^5$]{}; at (1.618+0.2,1.618+0.2) [$\scriptstyle \mathfrak{w}_2^5$]{}; at (1,1.618+0.275) [$\scriptstyle \mathfrak{w}_3^5$]{}; at (0-0.2,1+0.2) [$\scriptstyle \mathfrak{w}_4^5$]{}; at (-1-0.2,0+0.2) [$\scriptstyle \mathfrak{w}_5^5$]{}; at (-1.618-0.3,-1) [$\scriptstyle \mathfrak{w}_6^5$]{}; at (-1.618-0.2,-1.618-0.2) [$\scriptstyle \mathfrak{w}_7^5$]{}; at (-1,-1.618-0.275) [$\scriptstyle \mathfrak{w}_8^5$]{}; at (0+0.2,-1-0.2) [$\scriptstyle \mathfrak{w}_9^5$]{}; Given two vectors $\mathbf{u}_0 = (x_0, y_0)^T, \mathbf{u}_1 = (x_1, y_1)^T \in {\mathbb{R}}^2$, we denote their *(scalar) wedge product* by $$\mathbf{u}_0 \wedge \mathbf{u}_1 = x_0 y_1 - x_1 y_0,$$ and their *dot product* by $$\mathbf{u}_0 \cdot \mathbf{u}_1 = x_0 x_1 + y_0 y_1.$$ One useful inequality that we use more than once in this paper is that if $\mathbf{u}_0, \mathbf{u}_1, \mathbf{v}$ are non-zero vectors in ${\mathbb{R}}^2$, with the angle $\angle \mathbf{u}_0 \mathbf{u}_1$ not exceeding $\pi/2$, and $\mathbf{v}$ belonging to the sector $(0, \infty)\mathbf{u}_0 + (0, \infty)\mathbf{u}_1 = \{\alpha \mathbf{u}_0 + \beta \mathbf{u}_1 \mid \alpha, \beta > 0\}$, then $$\label{equation: sin inequalities} 0 < \frac{\mathbf{u}_0 \wedge \mathbf{v}}{\|\mathbf{u}_0\| \|\mathbf{v}\| }, \frac{\mathbf{v} \wedge \mathbf{u}_1}{\|\mathbf{v}\| \|\mathbf{u}_1\|} < \frac{\mathbf{u}_0 \wedge \mathbf{u}_1}{\|\mathbf{u}_0\| \|\mathbf{u}_1\|}.$$ This follows from the identities $\mathbf{u}_0 \wedge \mathbf{v} = \|u_0\| \|\mathbf{v}\| \sin(\angle \mathbf{u}_0 \mathbf{v})$, $\mathbf{v} \wedge \mathbf{u}_1 = \|\mathbf{v}\| \|\mathbf{u}_1\| \sin(\angle\mathbf{v}\mathbf{u}_1)$, and $\mathbf{u}_0 \wedge \mathbf{u}_1 = \|\mathbf{u}_0\| \|\mathbf{u}_1\| \sin(\angle\mathbf{u}_0\mathbf{u}_1)$, in addition to the inequalities $\sin(\angle\mathbf{u}_0\mathbf{v}), \sin(\angle\mathbf{v}\mathbf{u}_1) < \sin(\angle\mathbf{u}_0\mathbf{u}_1)$. Finally, we say that the two vectors $\mathbf{u}_0, \mathbf{u}_1 \in {\mathbb{R}}^2$ are *unimodular* if $\mathbf{u}_0 \wedge \mathbf{u}_1 = 1$. For readability, we sometimes will denote the usual product on ${\mathbb{R}}$ by $\times$. So $2 \times 3 = 6$, and so on. Finally, we write $$h_s := \begin{pmatrix}1 & 0 \\ -s & 1\end{pmatrix},$$ for $s \in {\mathbb{R}}$, $$s_\tau = \begin{pmatrix}\tau & 0 \\ 0 & \tau^{-1}\end{pmatrix},$$ for $\tau > 0$, and $$g_{a, b} = \begin{pmatrix}a & b \\ 0 & a^{-1}\end{pmatrix},$$ for $a > 0$, and $b \in {\mathbb{R}}$. The above matrices satisfy the identities $g_{\tau, 0} = s_\tau$, $h_s h_t = h_{s + t}$, and $h_s s_\tau = s_\tau h_{s\tau^2}$. The Discrete Orbits, Stern-Brocot Trees, and Boca-Cobeli-Zaharescu Map Analogue for $\Lambda_q = G_q(1, 0)^T$ {#section: discrete orbits} ============================================================================================================= The Discrete Orbits of the Linear Action of $G_q$ on the Plane ${\mathbb{R}}^2$ ------------------------------------------------------------------------------- \[proposition: column - unimodular pair identification\] The following are true. 1. If the orbit of $\mathbf{u} \in {\mathbb{R}}^2$ under the linear action of $G_q$ is a discrete subset of ${\mathbb{R}}^2$, then either $\mathbf{u} = (0, 0)^T$, or $G_q \mathbf{u}$ is a homothetic dilation of $\Lambda_q = G_q(1,0)^T$. 2. The ellipse $Q_q((x, y)^T) = x^2 - \lambda_q x y + y^2 = 1$ does not contain any elements of $\Lambda_q$ in its interior. 3. The elements $\mathfrak{w}_0^q, \mathfrak{w}_1^q, \cdots, \mathfrak{w}_{q-1}^q$ of $\Lambda_q$ satisfy the *Farey neighbor identities* $$\mathfrak{w}_i^q \wedge \mathfrak{w}_{i+1}^q = 1,$$ for $i = 0, 1, \cdots, q - 2$, in addition to $$\mathfrak{w}_0^q \wedge \mathfrak{w}_{q-1}^q = 1.$$ 4. If $\mathbf{u}_0, \mathbf{u}_1 \in \Lambda_q$ are two unimodular vectors (i.e $\mathbf{u}_0 \wedge \mathbf{u}_1 = 1$), then there exists $A \in G_q$ such that $A\mathbf{u}_0 = \mathfrak{w}_0^q = (1, 0)^T$ and $A\mathbf{u}_1 = \mathfrak{w}_{q-1}^q = (0, 1)^T$. That is, the pairs of unimodular vectors of $\Lambda_q$ are in a one-to-one correspondence with the columns of the matrices in $G_q$. For the first claim: Assume without loss of generality that $\mathbf{u} \neq (0, 0)^T$. Let $\Sigma_i^q = (0,\infty) \mathfrak{w}_i^q + [0,\infty) \mathfrak{w}_{i+1 \mod 2q}^q = \{\alpha \mathfrak{w}_i^q + \beta \mathfrak{w}_{i+1 \mod 2q}^q \mid \alpha, \beta > 0\}$, $i = 0, 1, \cdots, 2q - 1$, be the radial sectors of ${\mathbb{R}}^2 \setminus \{(0, 0)^T\}$ defined by the directions $\{\mathfrak{w}_i^q\}_{i=0}^{2q-1}$. Note that the matrix $U_q$ bijectively maps each sector $\Sigma_i^q$ to the sector $\Sigma_{i + 1 \mod 2q}^q$ for $i = 0, 1, \cdots, 2q-1$, and maintains the values of the quadratic form $Q_q$ at each point. Also, $T_q^{-1}$ maps the sector $\Sigma_0^q$ to $\cup_{i=0}^{q-2} \Sigma_i = [0,\infty)(1,0)^T + (0, \infty)(1,0)^T$, decreasing the $Q_q$-values of all the points in the interior of $\Sigma_0$, and fixing all the points on the ray in the direction of $\mathfrak{w}_0^q = (1,0)^T$. (This follows from $T_q^{-1} \mathfrak{w}_0^q = \mathfrak{w}_0^q = (1, 0)^T$, and $T_q^{-1}\mathfrak{w}_1^q = \mathfrak{w}_{q-1}^q = (0,1)^T$.) Starting with the vector $\mathbf{u}$ whose $G_q$-orbit is being considered, we repeatedly apply the following process: 1. If $\mathbf{u} \in \Sigma_i^q$ for some $1 \leq i \leq 2q - 1$, then replace $\mathbf{u}$ with $U_q^{-i} \mathbf{u} \in \Sigma_0^q$. This maintains the $Q_q$-value of $\mathbf{u}$. 2. Replace $\mathbf{u} \in \Sigma_0^q$ with $T_q^{-1}\mathbf{u} \in \cup_{i=0}^{q-2}\Sigma_i^q$. This fixes $\mathbf{u}$ if it lies on the ray in the direction of $\mathfrak{w}_0^q = (1, 0)^T$, and otherwise reduces the $Q_q$-value of $\mathbf{u}$. After each iteration of this process, either the point $\mathbf{u}$ lands on the line $y = 0$ and is fixed by further applications of the process, or is mapped to another point in $G_q\mathbf{u}$ with a strictly smaller $Q_q$-value. By the discreteness of $G_q\mathbf{u}$, the point $\mathbf{u}$ will eventually land on the line $y = 0$. This implies that there exists a non-zero $\alpha \in {\mathbb{R}}$ such that $\mathbf{u} \in \alpha G_q (1, 0)^T$, from which follows that $G_q\mathbf{u} = \alpha \Lambda_q$. This proves the first claim. The second claim follows from the fact that $(1, 0)^T \in \Lambda_q$ lies on the ellipse $Q_q((x, y)^T) = x^2 - \lambda_q x y + y^2 = 1$. No point in $\Lambda_q$ can have a $Q_q$-value smaller than $1$, as the iterative process used above will produce an element of $\Lambda_q$ that is parallel to $(1, 0)$ and shorter than it, which cannot happen by the discreteness of $\Lambda_q$. For the third claim: we have for all $i = 0, 1, \cdots, q - 2$ that $$\mathfrak{w}_i^q \wedge \mathfrak{w}_{i+1}^q = (U_q^i \mathfrak{w}_0^q) \wedge (U_q^i \mathfrak{w}_1^q) = \det(U_q^i) \times \mathfrak{w}_0^q \wedge \mathfrak{w}_1^q = 1 \times (1,0)^T \wedge (\lambda_q, 1)^T = 1.$$ We also have that $$\mathfrak{w}_0^q \wedge \mathfrak{w}_{q-1}^q = (1,0)^T \wedge (0, 1)^T = 1.$$ For the fourth claim: By definition, there exists $B \in G_q$ such that $B\mathbf{u}_0 = (1, 0)^T$. Acting by $B^{-1}$, the two vectors $\widetilde{\mathbf{u}}_0 = B^{-1}\mathbf{u}_0 = (1, 0)^T$ and $\widetilde{\mathbf{u}}_1 = B^{-1}\mathbf{u}_1$ satisfy $$\widetilde{\mathbf{u}}_0 \wedge \widetilde{\mathbf{u}}_1 = \det(B^{-1}) \times \mathbf{u}_0 \wedge \mathbf{u}_1 = 1.$$ If $\widetilde{\mathbf{u}}_1 = (x, y)$, then $y = 1$. Shearing by $T_q$, we have $T_q^n\widetilde{\mathbf{u}}_0 = \widetilde{\mathbf{u}}_0$, and $T_q^n \widetilde{\mathbf{u}}_1 = (x + n \lambda_q, 1)^T$ for all $n \in {\mathbb{Z}}$. Since $\mathfrak{w}_1^q = (\lambda_q, 1)^T$ and $\mathfrak{w}_{q-1}^q = (0,1)^T$ are two elements of $\Lambda_q$ on the ellipse $Q_q = 1$, are at height $y = 1$, are a horizontal distance $\lambda_q$ away from each other, and $T_q^{-1}\mathfrak{w}_0^q = \mathfrak{w}_{q-1}^q$, then there exists $n_0 \in {\mathbb{Z}}$ such that $T_q^{n_0} \widetilde{\mathbf{u}}_1 = \mathfrak{w}_{q-1}^q$. Now, taking $A = T_q^{n_0}B^{-1}$ proves the claim. The Stern-Brocot Trees for $\Lambda_q = G_q(1,0)^T$ --------------------------------------------------- We refer to the process of iteratively replacing a pair of vectors $\mathbf{u}_0, \mathbf{u}_1 \in \Lambda_q$ that are unimodular (i.e. $\mathbf{u}_0 \wedge \mathbf{u}_1 = 1$) with the vectors $$x_0^q \mathbf{u}_0 + y_0^q \mathbf{u}_1 = \mathbf{u}_0, x_1^q \mathbf{u}_0 + y_1^q \mathbf{u}_1, \cdots, x_{q-2}^q \mathbf{u}_0 + y_{q-2}^q \mathbf{u}_1, x_{q-1}^q \mathbf{u}_0 + y_{q-1}^q \mathbf{u}_1 = \mathbf{u}_1$$ as the *$G_q$-Stern-Brocot process*. We refer to the vectors $\{x_i^q \mathbf{u}_0 + y_i^q \mathbf{u}_1\}_{i=1}^{q-2}$ as the *($G_q$-Stern-Brocot) children* of $\mathbf{u}_0, \mathbf{u}_1$, and successive children of the children of $\mathbf{u}_0, \mathbf{u}_1$ as the *($G_q$-Stern-Brocot) grandchildren* of $\mathbf{u}_0, \mathbf{u}_1$. \[theorem: G\_q Stern Brocot process is well-defined and exhaustive\] Let $\mathbf{u}_0, \mathbf{u}_1 \in \Lambda_q$ be two unimodular vectors (i.e. $\mathbf{u}_0 \wedge \mathbf{u}_1 = 1$). The $G_q$-Stern-Brocot process applied to $\mathbf{u}_0$ and $\mathbf{u}_1$ generates a well-defined tree of elements of $\Lambda_q$, and exhausts the elements of $\Lambda_q$ in the sector $[0,\infty)\mathbf{u}_0 + [0,\infty)\mathbf{u}_1 = \{\alpha \mathbf{u}_0 + \beta \mathbf{u}_1 \mid \alpha, \beta \geq 0\}$. That the Stern-Brocot process is well-defined for any two unimodular elements $\mathbf{u}_0$ and $\mathbf{u}_1$ of $\Lambda_q$ follows from \[proposition: column - unimodular pair identification\]. In particular, since $\mathbf{u}_0$ and $\mathbf{u}_1$ are unimodular, then there exists $A \in G_q$ whose columns are $\mathbf{u}_0$ and $\mathbf{u}_1$ (i.e. $A(1,0)^T = \mathbf{u}_0$ and $A(0,1)^T = \mathbf{u}_1$). The vectors $\mathfrak{w}_i^q = (x_i^q, y_i^q)^T = x_i^q (1, 0)^T + y_i^q (0, 1)^T$, with $i = 0, 1, \cdots, q - 1$, are unimodular in pairs (by the Farey neighbor identities from \[proposition: column - unimodular pair identification\]), and so their images $A\mathfrak{w}_i^q = x_i^q \mathbf{u}_0 + y_i^q \mathbf{u}_1$, $i = 0, 1, \cdots, q - 1$, satisfy the same Farey neighbor identities, are all elements of $\Lambda_q$, and all belong to the sector $[0,\infty)\mathbf{u}_0 + [0,\infty)\mathbf{u}_1$. It remains to prove that the Stern-Brocot process is exhaustive, and our proof is similar to that of the classical proof for Farey fractions. We first need to show that the wedge products of pairs of non-parallel elements of $\Lambda_q$ are bounded away from zero.[^3] Given two elements $\mathbf{w}_0, \mathbf{w}_1$ of $\Lambda_q$, we assume that if $0 < \mathbf{w}_0 \wedge \mathbf{w}_1 < \epsilon$, then $\epsilon$ cannot be arbitrarily small. Pick any $A \in G_q$ with $A\mathbf{u}_0 = (1, 0)^T$. Writing $A\mathbf{u}_1 = (x, y)^T$, then $0 < A\mathbf{u}_0 \wedge A\mathbf{u}_1 = y < \epsilon$. Shearing by $T_q^{\pm} = \begin{pmatrix}1 & \pm\lambda_q \\ 0 & 1\end{pmatrix}$, we can find $n \in \mathbb{Z}$ such that $T_q^n A\mathbf{u}_1 = (x + n\lambda_qy, y)^T$ has an $x$-component $0 \leq x + n \lambda_q y < \lambda_q \epsilon$. From this follows that $\|T_q^n A \mathbf{u}_1\| \leq \epsilon\sqrt{1 + \lambda_q^2}$, and so $\epsilon$ cannot be arbitrarily small by the discreteness of $\Lambda_q$. It thus follows that for all $q \geq 3$, there exists $\epsilon_q$ such that the wedge product of any non-parallel pair of elements of $\Lambda_q$ is bounded below by $\epsilon_q$ in absolute value. Now, we write $\mathbf{u}_0 = (q_0,a_0)^T$, and $\mathbf{u}_1 = (q_1,a_1)^T$, and assume that $\mathbf{u}_0, \mathbf{u}_1$ belong to the first quadrant. (We can safely do that by the last claim of \[proposition: column - unimodular pair identification\].) If $(x,y)^T \in \Lambda_q$ belongs to the sector $(0,\infty)\mathbf{u}_0 + (0,\infty)\mathbf{u}_1$, the orientation of the vectors gives $\mathbf{u}_0 \wedge (x,y)^T, (x,y)^T \wedge \mathbf{u}_1 > 0$, and so $\mathbf{u}_0 \wedge (x,y)^T, (x,y)^T \wedge \mathbf{u}_1 \geq \epsilon_q$. We define the component sum function $\varsigma : {\mathbb{R}}^2 \to {\mathbb{R}}$ by $\varsigma(r,s)^T = r + s$ for all $(r,s)^T \in {\mathbb{R}}^2$. We thus get $$\begin{aligned} \varsigma(\mathbf{u}_1) \left(\mathbf{u}_0 \wedge (x,y)^T\right) + \varsigma(\mathbf{u}_0) \left((x,y)^T \wedge \mathbf{u}_1\right) &=& \begin{aligned} & (a_1 + q_1)(y q_0 - x a_0) \\ & \ + (a_0 + q_0)(a_1 x - q_1 y) \end{aligned} \\ &=& (a_1 q_0 - a_0 q_1) (x + y) \\ &=& \mathbf{u}_0 \wedge \mathbf{u}_1 \times \varsigma(x,y)^T \\ &=& \varsigma(x,y)^T,\end{aligned}$$ and so $$\label{equation: varsigma bound} \varsigma(x,y)^T \geq \epsilon_q \left(\varsigma(\mathbf{u}_0) + \varsigma(\mathbf{u}_1)\right).$$ Assuming without loss of generality that we are starting the Stern-Brocot process with $(1,0)^T$ and $(0,1)^T$, we have that the $\varsigma$ value of any vector that is generated at the $n$th step, $n \geq 0$, is bounded below by $n + 1$. (We demonstrate this fact at the end of this proof.) At any step, if $(x,y)^T$ is not one of the $q-2$ Stern-Brocot children of $\mathbf{u}_0$ and $\mathbf{u}_1$, then it belongs to a sector defined by one of the $q-1$ pairs of successive unimodular vectors that have been generated at this step. This cannot take place forever as each step of Stern-Brocot increases the right hand side of \[equation: varsigma bound\] by at least $\epsilon_q$. This implies that $(x,y)^T$ eventually shows up as a child, and we are done. Now we prove the lower bound on the $\varsigma$ value. If $\mathbf{c}$ is the $G_q$-Stern-Brocot child of two vectors $\mathbf{p}_1, \mathbf{p}_2$ in the first quadrant, then $\mathbf{c} = x_{i_0}^q \mathbf{p}_1 + y_{i_0}^q \mathbf{p}_2$ for some $1 \leq i_0 \leq q - 2$, and so $\varsigma(\mathbf{c}) = x_{i_0}^q \varsigma(\mathbf{p}_1) + y_{i_0}^q \varsigma(\mathbf{p}_2) \geq \varsigma(\mathbf{p}_1) + \varsigma(\mathbf{p}_2)$, since $x_{i_0}^q, y_{i_0}^q \geq 1$. It is easy to see that each of the vectors that are generated at one stage must have at least one parent that was generated at the previous stage. Since $\varsigma((1, 0)^T), \varsigma((0, 1)^T) = 1$, it now follows by induction that the $\varsigma \geq n + 1$ for all the vectors that are generated at the $n$th stage for $n \geq 0$. In the following corollary, we collect some consequences of the existence of Stern-Brocot tree for $\Lambda_q$ that we use in the remainder of this paper. \[corollary: odds and ends\] The following are true. 1. If $\mathbf{v}_0, \mathbf{v}_1 \in \Lambda_q$ are such that $\mathbf{v}_0 \neq \pm \mathbf{v}_1$, then $|\mathbf{v}_0 \wedge \mathbf{v}_1| \geq 1$. 2. Let $\mathbf{v} \in {\mathbb{R}}^2 \setminus \{(0, 0)^T\}$ be an arbitrary non-zero vector in the plane. Then either $\mathbf{v}$ is parallel to a vector in $\Lambda_q$, or for any unimodular pair $\mathbf{u}_0, \mathbf{u}_1 \in \Lambda_q$, if $\mathbf{v}$ belongs to the sector $(0,\infty)\mathbf{u}_0 + (0, \infty)\mathbf{u}_1$, then there exists a pair of unimodular $G_q$-Stern-Brocot grandchildren $\mathbf{w}_0, \mathbf{w}_1$ of $\mathbf{u}_0, \mathbf{u}_1$ such that $\mathbf{v}$ belongs to the sector $(0,\infty)\mathbf{w}_0 + (0, \infty)\mathbf{w}_1$, and $\mathbf{w}_0, \mathbf{w}_1$ are different from $\mathbf{u}_0, \mathbf{u}_1$. 3. Let $\mathbf{u}_0, \mathbf{u}_1 \in \Lambda_q$ be two unimodular vectors, and $\{\mathbf{w}_n\}_{n=1}^\infty$ be any sequence of elements of $\Lambda_q$ such that for each $n \geq 1$, $\mathbf{w}_n$ is generated at the $n$th iteration of the $G_q$-Stern-Brocot process applied to the two unimodular vectors $\mathbf{u}_0, \mathbf{u}_1$. Then $\lim_{n \to \infty} \|\mathbf{w}_n\| = \infty$. 4. The slopes of the non-vertical vectors in $\Lambda_q$ are dense in ${\mathbb{R}}$. We first prove the following: If $\mathbf{u}_0, \mathbf{u}_1 \in \Lambda_q$ are unimodular (i.e. $\mathbf{u}_0 \wedge \mathbf{u}_1 = 1$), then after $n \geq 1$ applications of the $G_q$-Stern-Brocot process, the two vectors $\mathbf{w}_n^r = n \lambda_q \mathbf{u}_0 + \mathbf{u}_1 = (n\lambda_q, 1)^T$ and $\mathbf{w}_n^l = \mathbf{u}_0 + n \lambda_q \mathbf{u}_1 = (1, n\lambda_q)^T$ are $G_q$-Stern-Brocot grandchildren of $\mathbf{u}_0$ and $\mathbf{u}_1$, and all the grandchildren of $\mathbf{u}_0$ and $\mathbf{u}_1$ that have been generated by the $n$th step belong to the sector $[0,\infty)\mathbf{w}_n^r + [0,\infty)\mathbf{w}_n^l$. Now, since $(x_1^q, y_1^q)^T = \mathfrak{w}_1^q = U_q(1, 0)^T = (\lambda_q, 1)^T$, and $(x_{q-2}^q, y_{q-2}^q)^T = \mathfrak{w}_{q-2}^q = U_q^{-1}(0,1)^T = (1, \lambda_q)^T$, it follows from \[theorem: G\_q Stern Brocot process is well-defined and exhaustive\] that the two vectors $\mathbf{w}_1^r = x_1^q \mathbf{u}_0 + y_1^q \mathbf{u}_1 = \lambda_q \mathbf{u}_0 + \mathbf{u}_1$ and $\mathbf{w}_1^l = x_{q-2}^q \mathbf{u}_0 + y_{q-2}^q \mathbf{u}_1 = \mathbf{u}_0 + \lambda_q \mathbf{u}_1$ are Stern-Brocot children of $\mathbf{u}_0$ and $\mathbf{u}_1$, and that all the children of $\mathbf{u}_0$ and $\mathbf{u}_1$ that were generated after one iteration are contained in the sector corresponding to $\mathbf{w}_1^r$ and $\mathbf{w}_1^l$. The remainder of the claim follows by repeatedly applying the Stern-Brocot process to the unimodular pair $\mathbf{u}_0$ and $\mathbf{w}_n^r$, and the unimodular pair $\mathbf{w}_n^l$ and $\mathbf{u}_1$, for all $n \geq 2$. For the first claim: Since $-\operatorname{Id}_2 = U_q^q$ is in $G_q$, we can assume that the angle between $\mathbf{v}_0$ and $\mathbf{v}_1$ does not exceed $\pi/2$. We also permute $\mathbf{v}_0$ and $\mathbf{v}_1$ if need be so that $\mathbf{v}_0 \wedge \mathbf{v}_1 > 0$. Furthermore, we can assume that $\mathbf{v}_0 = (1, 0)^T$. (There exists $A \in G_q$ such that $\mathbf{v}_0 = A(1, 0)^T$, and so we can replace $\mathbf{v}_0$ and $\mathbf{v}_1$ with $\widetilde{\mathbf{v}}_0 = A^{-1}\mathbf{v}_0$ and $\widetilde{\mathbf{v}}_1 = A^{-1}\mathbf{v}_1$, and preserve the wedge product $\widetilde{\mathbf{v}}_0 \wedge \widetilde{\mathbf{v}}_1 = \det(A^{-1}) \times \mathbf{v}_0 \wedge \mathbf{v}_1 = \mathbf{v}_0 \wedge \mathbf{v}_1$.) We now have that $\mathbf{v}_0 = (1, 0)^T$, and that $\mathbf{v}_1$ is in the first quadrant. That is, $\mathbf{v}_1$ is either $\mathfrak{w}_{q-1}^q = (0, 1)^T$, or a $G_q$-Stern-Brocot grandchild of $\mathfrak{w}_0^q = (1, 0)^T$ and $\mathfrak{w}_{q-1}^q = (0, 1)^T$. Writing $\mathbf{v}_1 = (x_{\mathbf{v}_1}, y_{\mathbf{v}_1})^T$, we have $\mathbf{v}_0 \wedge \mathbf{v}_1 = y_{\mathbf{v}_1}$. The $y$ components of the vectors $\{\mathbf{w}_n^r\}_{n=1}^\infty$ from the first claim in the corollary all are all $y = 1$, and so $y_{\mathbf{v}_1} = \mathbf{v}_0 \wedge \mathbf{v}_1 \geq 1$ as required. For the second claim: The unit vectors in the directions of $\{\mathbf{w}_n^r\}_{n=1}^\infty$ and $\{\mathbf{w}_n^l\}_{n=1}^\infty$ converge to $\mathbf{u}_0$ and $\mathbf{u}_1$ as $n \to \infty$. As such, if the vector $\mathbf{v}$ is not in $\Lambda_q$, then it will eventually be contained in the sector bounded by $\mathbf{w}_{n_0}^l$ and $\mathbf{w}_{n_0}^l$ for some $n_0 \geq 1$, and consequently belongs to the sector bounded by a pair of unimodular grandchildren of $\mathbf{u}_0$ and $\mathbf{u}_1$. For the third claim: By the fourth claim in \[proposition: column - unimodular pair identification\], and the boundedness of the elements of $G_q$ as linear operators on ${\mathbb{R}}^2$, we can assume without loss of generality that $\mathbf{u}_0 = (1, 0)^T$ and $\mathbf{u}_1 = (0, 1)^T$. At the end of the proof of \[theorem: G\_q Stern Brocot process is well-defined and exhaustive\], we showed that if $\mathbf{w}_n$ is generated at the $n$th stage of the $G_q$-Stern-Brocot process applies to $(1, 0)^T$ and $(0, 1)^T$, then $\mathbf{w}_n \geq n + 1$. If $\mathbf{w}_n = (r, s)^T$, then $\varsigma(\mathbf{w}_n) = r + s \leq \sqrt{2} \sqrt{r^2 + s^2} \leq \sqrt{2} \|\mathbf{w}_n\|$, which proves the claim. For the fourth claim: It suffices to show that if $\alpha \geq 0$ is not the slope of a vector in $\Lambda_q$, then $\alpha$ can be approximated by slopes of vectors in $\Lambda_q$. Writing $\mathbf{v} = (1, \alpha)^T$, we note that if $\mathbf{u}_0, \mathbf{u}_1 \in \Lambda_q$ are two unimodular vectors in the first quadrant whose sector contains $\mathbf{v}$, then by \[equation: sin inequalities\] we have $$0 < \mathbf{u}_0 \wedge \mathbf{v} < \frac{\|\mathbf{v}\|}{\|\mathbf{u}_1\|} \mathbf{u}_0 \wedge \mathbf{u}_1 = \mathbf{v} < \frac{\|\mathbf{v}\|}{\|\mathbf{u}_1\|}.$$ Writing $\mathbf{u}_0 = (x, y)^T$, and assuming that $x > 0$, we thus get $$\label{equation: Diophantine approximation} 0 \leq \alpha - \frac{y}{x} \leq \frac{\sqrt{1+\alpha^2}}{x\|\mathbf{u}_1\|}.$$ Now, we can start with $\mathbf{u}_0 = (1, 0)^T$ and $\mathbf{u}_1 = (0, 1)^T$ as two vectors in the first quadrant whose sector contains $\mathbf{v}$, and by the second claim in this corollary, we can repeatedly replace $\mathbf{u}_0$ and $\mathbf{u}_1$ with unimodular pairs that are generated at later stages of the Stern-Brocot process. In \[equation: Diophantine approximation\], $x \geq 1$, and $\lim_{n \to \infty} \|\mathbf{u}_1\| = \infty$, and we are done. The Boca-Cobeli-Zaharescu Map Analogue for $\Lambda_q = G_q(1,0)^T$ ------------------------------------------------------------------- In the following theorem, we present the BCZ map analogue for $\Lambda_q$. In essence, this theorem along with the next-term algorithm (\[theorem: next term algorithm\]) extend the properties of the Farey sequence alluded to in the introduction using the BCZ map formalization. \[theorem: G\_q BCZ maps\] The following are true. 1. For any $A \in {\operatorname{SL}}(2, {\mathbb{R}})$, if $A \Lambda_q$ has a horizontal vector of length not exceeding $1$ (i.e. a horizontal vector in $A \Lambda_q \cap S_1$), then $A \Lambda_q$ can be uniquely identified with a point $(a_A, b_A)$ in the *$G_q$-Farey triangle* $$\mathscr{T}^q = \{(a, b) \in {\mathbb{R}}^2 \mid 0 < a \leq 1,\ 1 - \lambda_q a < b \leq 1\}$$ through $B\Lambda_q = g_{a_A,b_A}\Lambda_q$. Moreover, the value $a_A$ agrees with the length of the horizontal vector in $A\Lambda_q \cap S_1$. 2. Let $(a, b) \in \mathscr{T}^q$ be any point in the $G_q$-Farey triangle. The set $g_{a, b} \Lambda_q \cap S_1$ has a vector with smallest positive slope. Consequently, there exists a smallest $s = R_q(a, b) > 0$ such that $h_s g_{a,b} \Lambda_q$ has a horizontal vector of length not exceeding $1$, and hence $h_s g_{a, b} \Lambda_q$ corresponds to a unique point ${\operatorname{BCZ}}_q(a, b) \in \mathscr{T}^q$ in the $G_q$-Farey triangle. The function $R_q : \mathscr{T}^q \to {\mathbb{R}}_+$ is referred to as the *$G_q$-roof function*, and the map ${\operatorname{BCZ}}_q(a, b) : \mathscr{T}^q \to \mathscr{T}^q$ is referred to as the *$G_q$-BCZ map*. 3. The $G_q$-Farey triangle $\mathscr{T}^q$ can be partitioned into the union of $$\mathscr{T}_i^q := \{(a, b) \in \mathscr{T}^q \mid (a, b)^T \cdot \mathfrak{w}_{i - 1} > 1,\ (a, b)^T\cdot\mathfrak{w}_i \leq 1\},$$ with $i = 2, 3, \cdots, q - 1$, such that if $(a, b) \in \mathscr{T}_i^q$, then $g_{a, b}\mathfrak{w}_i^q$ is the vector of least positive slope in $g_{a,b}\Lambda_q \cap S_1$, and - the value of the roof function $R_q(a, b)$ is given by $$R_q(a, b) = \frac{y_i^q}{a \times (a,b)^T \cdot \mathfrak{w}_i^q}, \text{ and }$$ - the value of the BCZ map ${\operatorname{BCZ}}_q(a, b)$ is given by $${\operatorname{BCZ}}_q(a, b) = \left((a,b)^T \cdot \mathfrak{w}_i^q, (a,b)^T \cdot \mathfrak{w}_{i+1}^q + k_i^q(a, b) \times \lambda_q \times (a, b)^T \cdot \mathfrak{w}_i^q\right),$$ where the *$G_q$-index* $k_i^q(a,b)$ is given by $$k_i^q(a,b) = \left\lfloor \frac{1 - (a,b)^T \cdot \mathfrak{w}_{i+1}^q}{\lambda_q \times (a,b)^T \cdot \mathfrak{w}_i^q} \right\rfloor.$$ (-0.5,0)–(1.25,0) node\[right\][$\scriptstyle a$]{}; (0,-0.5)–(0,1.25) node\[above\][$\scriptstyle b$]{}; (0, 1) – (1, -0.618) – (1, 1) – (0, 1); (0.382, 0.382) – (1, 0); (0.618, 0) – (1, -0.382); at (1.2, 0.5) [$\mathscr{T}_4^5$]{}; at (1.2, -0.191) [$\mathscr{T}_3^5$]{}; at (1.2, -0.5) [$\mathscr{T}_2^5$]{}; at (1.1, 0.191) \[rotate=90\] [$a = 1$]{}; at (0.5, 1.09) [$\mathcal{L}_4^{5, \text{top}}$]{}; at (0.69, 0.3) [$\mathcal{L}_3^{5, \text{top}}$]{}; at (0.81, -0.05) [$\mathcal{L}_2^{5, \text{top}}$]{}; at (0.4, 0.1) [$\mathcal{L}_1^{5, \text{top}}$]{}; We first need the following lemma. \[lemma: orbit-stabilizer lemma\] Given $A \in {\operatorname{SL}}(2, {\mathbb{R}})$, if $A\Lambda_q$ contains both $(1, 0)^T$ and $(0, 1)^T$, then $A\Lambda_q = \Lambda_q$. In particular, the following are true. 1. For any $B \in {\operatorname{SL}}(2, {\mathbb{R}})$, $B \Lambda_q = \Lambda_q$ if and only if $B \in G_q$. From this follows that the sets $C\Lambda_q$, with $C$ varying over ${\operatorname{SL}}(2, {\mathbb{R}})$, can be identified with the elements of ${\operatorname{SL}}(2, {\mathbb{R}})/G_q$. 2. For any $B \in {\operatorname{SL}}(2, {\mathbb{R}})$, if $B \Lambda_q$ contains a horizontal vector of length $a > 0$, then there exists $b \in {\mathbb{R}}$ such that $B \Lambda_q = g_{a,b} \Lambda_q$. We first prove the main claim. Let $A \in {\operatorname{SL}}(2, {\mathbb{R}})$ be such that $A \Lambda_q$ contains both $(1, 0)^T$ and $(0, 1)^T$. Then there exists $\mathbf{u}_0, \mathbf{u}_1 \in \Lambda_q$ such that $\mathbf{u}_0 = A^{-1}(1, 0)^T$ and $\mathbf{u}_1 = A^{-1}(0, 1)^T$, and $\mathbf{u}_0 \wedge \mathbf{u}_1 = \det(A^{-1}) \times (1, 0)^T \times (0, 1)^T = 1$. The columns of $A^{-1}$ thus form a unimodular pair of elements of $\Lambda_q$, and so by the last claim of \[proposition: column - unimodular pair identification\], the matrix $A^{-1}$, and by necessity $A$, belong to the group $G_q$. The first claim now follows from the fact that if $B \in {\operatorname{SL}}(2, {\mathbb{R}})$ is such that $B\Lambda_q = \Lambda_q$, then $B\Lambda_q$ contains both $(1, 0)^T$ and $(0, 1)^T$. We now prove the second claim. Let $\mathbf{u}_0 \in \Lambda_q$ be such that $B\mathbf{u}_0 = (a, 0)^T \in B\Lambda_q$ is parallel to the horizontal vector in question. If $A \in {\operatorname{SL}}(2, {\mathbb{R}})$ is such that $\mathbf{u}_0 = A(1, 0)^T$, then $\mathbf{u}_1 = A(0, 1)^T$ is an element of $\Lambda_q$ with $\mathbf{u}_0 \wedge \mathbf{u}_1 = 1$. Writing $\widetilde{\mathbf{u}}_0 = \begin{pmatrix}a^{-1} & 0 \\ 0 & a\end{pmatrix}B\mathbf{u}_0 = (1, 0)^T$, and $\widetilde{\mathbf{u}}_1 = \begin{pmatrix}a^{-1} & 0 \\ 0 & a\end{pmatrix}B\mathbf{u}_1 = (x, y)^T$, we have that $\widetilde{\mathbf{u}}_0 \wedge \widetilde{\mathbf{u}}_1 = 1$, and so $y = 1$. Shearing by $T_{-x} = \begin{pmatrix}1 & -x \\ 0 & 1\end{pmatrix}$, we have that $T_{-x} \widetilde{\mathbf{u}}_0 = (1, 0)^T$, and $T_{-x} \widetilde{\mathbf{u}}_1 = (0, 1)^T$. That is, the set $\begin{pmatrix}1 & -x \\ 0 & 1\end{pmatrix} \begin{pmatrix}a^{-1} & 0 \\ 0 & a\end{pmatrix} B \Lambda_q$ contains both $(1, 0)^T$, and $(0,1)^T$, and so is equal to $\Lambda_q$. From this follows that $$\begin{aligned} B\Lambda_q &=& \begin{pmatrix}a^{-1} & 0 \\ 0 & a\end{pmatrix}^{-1} \begin{pmatrix}1 & -x \\ 0 & 1\end{pmatrix}^{-1} \Lambda_q \\ &=& \begin{pmatrix}a & ax \\ 0 & a^{-1}\end{pmatrix} \Lambda_q \\ &=& g_{a, ax} \Lambda_q,\end{aligned}$$ and taking $b = ax$ proves the claim. We now proceed to prove \[theorem: G\_q BCZ maps\]. We first derive the explicit values of the roof function $R_q(a, b)$ and BCZ map ${\operatorname{BCZ}}_q(a, b)$ in the second half of the third claim for a given point $(a, b) \in \mathscr{T}_i^q$, $i = 2, 3, \cdots, q - 1$, assuming the remainder of the theorem, and then resume the proof of the theorem from the beginning. If $(a, b) \in \mathscr{T}_i^q$, with $2 \leq i \leq q - 1$, then $g_{a, b} \mathfrak{w}_i^q$ has the smallest positive slope in $g_{a, b} \Lambda_q \cap S_1$ by our (yet to be proven) assumption. This gives $$\begin{aligned} R_q(a, b) &=& {\operatorname{slope}}\left(g_{a,b} \mathfrak{w}_i^q\right) \\ &=& {\operatorname{slope}}\left(\begin{pmatrix}a & b \\ 0 & a^{-1}\end{pmatrix} \begin{pmatrix}x_i^q \\ y_i^q\end{pmatrix}\right) \\ &=& \frac{y_i^q}{a(a x_i^q + b y_i^q)} \\ &=& \frac{y_i^q}{a \times (a, b)^T \cdot \mathfrak{w}_i^q}. \end{aligned}$$ Now, let $[\mathfrak{w}_i^q\ \mathfrak{w}_{i+1}^q]$ be the matrix whose columns are $\mathfrak{w}_i^q$ and $\mathfrak{w}_{i + 1}^q$. The matrix $[\mathfrak{w}_i^q\ \mathfrak{w}_{i+1}^q]$ is in $G_q$ by \[proposition: column - unimodular pair identification\] since its columns are two unimodular elements of $\Lambda_q$. We show that $h_{R_q(a, b)} g_{a, b} [\mathfrak{w}_i^q\ \mathfrak{w}_{i+1}^q] = g_{(a, b)^T \cdot \mathfrak{w}_i^q, (a, b)^T \cdot \mathfrak{w}_{i + 1}^q}$, and follow that by finding the representative of $g_{(a, b)^T \cdot \mathfrak{w}_i^q, (a, b)^T \cdot \mathfrak{w}_{i + 1}^q}\Lambda_q$ in $\mathscr{T}^q$. (Note that $h_{R_q(a,b)}g_{a,b}\Lambda_q = g_{(a, b)^T \cdot \mathfrak{w}_i^q, (a, b)^T \cdot \mathfrak{w}_{i + 1}^q}\Lambda_q$ by \[lemma: orbit-stabilizer lemma\].) Keeping the Farey neighbor identity $\mathfrak{w}_i^q \wedge \mathfrak{w}_{i + 1}^q = x_i^q y_{i + 1}^q - x_{i + 1}^q y_i^q = 1$ in mind, we have $$\begin{aligned} h_{R_q(a, b)} g_{a, b} [\mathfrak{w}_i^q\ \mathfrak{w}_{i+1}^q] &=& \begin{pmatrix}1 & 0 \\ -\frac{y_i^q}{a(a x_i^q + b y_i^q)} & 1\end{pmatrix} \begin{pmatrix} a x_i^q + b y_i^q & a x_{i + 1}^q + b y_{i + 1}^q \\ a^{-1}y_i^q & a^{-1}y_{i + 1}^q\end{pmatrix} \\ &=& \begin{pmatrix}a x_i^q + b y_i^q & a x_{i + 1}^q + b y_{i + 1}^q \\ 0 & \frac{1}{a x_i^q + b y_i^q}\end{pmatrix} \\ &=& g_{a x_i^q + b y_i^q, a x_{i + 1}^q + b y_{i + 1}^q} \\ &=& g_{(a, b)^T \cdot \mathfrak{w}_i^q, (a, b)^T \cdot \mathfrak{w}_{i + 1}^q}.\end{aligned}$$ Write $\alpha = (a, b)^T \cdot \mathfrak{w}_i^q$, and $\beta = (a, b)^T \cdot \mathfrak{w}_{i + 1}^a$. Since $T_q = \begin{pmatrix}1 & \lambda_q \\ 0 & 1\end{pmatrix}$ is in $G_q$, and $g_{\alpha, \beta} T_q^k = g_{\alpha, \beta + k \lambda_q \alpha}$ for all $k \in {\mathbb{Z}}$, then $g_{\alpha, \beta} \Lambda_q = g_{\alpha, \beta + k \lambda_q \alpha} \Lambda$ for all $k \in {\mathbb{Z}}$ by \[lemma: orbit-stabilizer lemma\]. Taking $k_0 = \left\lfloor \frac{1 - \beta}{\lambda_q \alpha} \right\rfloor$, we get $1 - \lambda_q \alpha < \beta + k_0 \lambda_q \alpha \leq 1$. We will also see in a bit that $0 < \alpha \leq 1$ (which is equivalent to $(a, b)$ lying between the lines $\mathcal{L}_i^{q,\text{bot}}$ and $\mathcal{L}_i^{q, \text{top}}$ that we will be working with for the remainder of the proof). We thus have $h_{R_q(a, b)} g_{a, b} \Lambda_q = g_{\alpha, \beta + k_0 \lambda_q \alpha} \Lambda_q$, with $(\alpha, \beta + k_0 \lambda_q \alpha) \in \mathscr{T}^q$, and so $$\begin{aligned} {\operatorname{BCZ}}_q(a, b) &=& (\alpha, \beta + k_0 \lambda_q \alpha) \\ &=& \left((a,b)^T \cdot \mathfrak{w}_i^q, (a,b)^T \cdot \mathfrak{w}_{i+1}^q + k_i^q(a, b) \times \lambda_q \times (a, b)^T \cdot \mathfrak{w}_i^q\right)\end{aligned}$$ as required. Now, for the first claim of the theorem: If $B \Lambda_q$ has a horizontal vector of length $a \in (0, 1]$, then there exists $b \in {\mathbb{R}}$ such that $B \Lambda_q = g_{a, b} \Lambda_q$ by \[lemma: orbit-stabilizer lemma\]. Since $T_q \in G_q$, and $g_{a,b}T_q^{\pm 1} = g_{a, b \pm \lambda_q a}$, then $B\Lambda_q = g_{a, b}\Lambda_q = g_{a, b + n \lambda_q a} \Lambda_q$ for all $n \in {\mathbb{Z}}$. From this follows that $B\Lambda_q = g_{(a_B, b_B)}\Lambda_q$, with $(a_B, b_B) = \left(a, b + \left\lfloor\frac{1-b}{\lambda_q a}\right\rfloor \lambda_q a\right) \in \mathscr{T}^q$ as required. It now remains to show that this identification is unique. That is, given $(a, b), (c, d) \in \mathscr{T}^q$, if $g_{a,b}\Lambda_q = g_{c, d}\Lambda_q$, then $(a, b) = (c, d)$. Now, $$g_{c,d}^{-1} g_{a,b} = \begin{pmatrix}a/c & b/c - d/a \\ 0 & c/a\end{pmatrix} \in G_q.$$ By the identification in \[proposition: column - unimodular pair identification\], we thus have $(a/c, 0)^T = g_{c,d}^{-1} g_{a,b}(1, 0)^T \in \Lambda_q$, and so $a/c = \pm 1$, from which $a = c$. We also have $(b/c - d/a, 1)^T = g_{c,d}^{-1} g_{a,b}(0, 1)^T \in \Lambda_q$. It can be easily seen from the second claim in \[proposition: column - unimodular pair identification\] that all the points in $\Lambda_q$ at height $y = 1$ are of the form $(n\lambda_q, 1)^T = T_q^n (0, 1)^T$ with $n \in {\mathbb{Z}}$, and so $b/c - d/a = n\lambda_q$ for some $n_0 \in {\mathbb{Z}}$. That is, $b - d = n_0 \lambda_q a$. At the same time, $b, d \in (1 - \lambda_q a, 1]$, and so, since $a > 0$, we get that $b - d \in (-\lambda_q, \lambda_q)$. It now follows that $n_0 = 0$, and $b = d$. Finally, for the second claim, and the beginning of the third claim of the theorem, we consider the lines $$\mathcal{L}_i^{q, \text{bot}} := \{(a, b) \in {\mathbb{R}}^2 \mid (a, b)^T \cdot \mathfrak{w}_i^q = 0\},$$ and $$\mathcal{L}_i^{q, \text{top}} := \{(a, b) \in {\mathbb{R}}^2 \mid (a, b)^T \cdot \mathfrak{w}_i^q = 1\}$$ for $i = 1, 2, \cdots, q - 1$. Note that the lines $\mathcal{L}_1^{q, \text{top}}$ and $\mathcal{L}_{q - 1}^{q, \text{top}}$ agree with the sides $\lambda_q a + b = 1$ and $b = 1$ of $\mathscr{T}^q$. We now show that for $i = 2, 3, \cdots, q - 1$, if $(a, b) \in \mathscr{T}^q$ is in $\mathscr{T}_i^q$ (i.e. above the line $\mathcal{L}_{i-1}^{q, \text{top}}$ and below, or on the line $\mathcal{L}_i^{q, \text{top}}$), then $g_{a, b} \mathfrak{w}_i^q$ belongs to the strip $S_1$, and has the smallest positive slope among the elements of $g_{a, b} \Lambda_q \cap S_1$. For any $i = 2, 3, \cdots, q- 1$, if $(a, b) \in \mathscr{T}^q$ lies in the region above the line $\mathcal{L}_i^{q, \text{bot}}$, and below or on the line $\mathcal{L}_i^{q, \text{top}}$, then the $x$-component $(a, b)^T \cdot \mathfrak{w}_i^q$ of $g_{a, b} \mathfrak{w}_i^q$ satisfies $0 < (a, b)^T \cdot \mathfrak{w}_i^q \leq 1$, and so $g_{a, b} \mathfrak{w}_i^q$ belongs to $g_{a, b}\Lambda_q \cap S_1$. As we will see in a bit, the regions $\mathscr{T}_i^q$, $i = 2, 3, \cdots, q - 1$, cover $\mathscr{T}^q$, and so $g_{a, b} \Lambda_q \cap S_1 \neq \emptyset$ for all $(a, b) \in \mathscr{T}^q$. Moreover, for any $(a, b) \in \mathscr{T}^q$, the elements of $g_{a, b}\Lambda_q \cap S_1$ do not accumulate by the discreteness of $\Lambda_q$, and so there must exist an element of $g_{a, b}\Lambda_q \cap S_1$ with smallest positive slope. This proves the second claim. Finally, we prove that the regions in question cover the triangle $\mathscr{T}^q$, along with the first half of the third claim of the theorem. I.e., that for $i = 2, 3, \cdots, q - 1$, if $(a, b) \in \mathscr{T}^q$, then $g_{a, b} \mathfrak{w}_i^q$ has the smallest positive slope in $g_{a, b}\Lambda_q$. We break this down into three steps: 1. For $i = 1, 2, \cdots, q - 1$, the line segments $\mathcal{L}_i^{q, \text{top}} \cap \mathscr{T}^q$ lie above each other, and have increasing (non-positive) slopes. (That is, if $1 \leq i_1 < i_2 \leq q - 1$, then the line segment $\mathcal{L}_{i_1}^{q, \text{top}} \cap \mathscr{T}^q$ lies below the line segment $\mathcal{L}_{i_2}^{q, \text{top}} \cap \mathscr{T}^q$, and ${\operatorname{slope}}(\mathcal{L}_{i_2}^{q, \text{top}}) > {\operatorname{slope}}(\mathcal{L}_{i_1}^{q, \text{top}})$.) 2. For each $i = 2, 3, \cdots, q - 1$, the line segment $\mathcal{L}_i^{q, \text{bot}} \cap \mathscr{T}^q$ lies below the line segment $\mathcal{L}_{i - 1}^{q, \text{top}}$. (This proves the claim that the regions $\mathscr{T}_i^q$, $i = 2, 3, \cdots, q - 1$, cover $\mathscr{T}^q$.) 3. For $i = 1, 2, \cdots, q - 2$, if $(a, b) \in \mathscr{T}^q$ lies above the line $\mathcal{L}_i^{q, \text{top}}$, the the $g_{a,b}$ images of $\mathfrak{w}_i^q$ along with its $G_q$-Stern-Brocot children with $\mathfrak{w}_{i+1}^q$ have $x$-components that exceed $1$, and so are not in $g_{a, b} \Lambda_q \cap S_1$. The third step follows immediately from the fact that the $G_q$-Stern-Brocot children of $\mathfrak{w}_i^q$ and $\mathfrak{w}_{i+1}^q$, $i = 1, 2, \cdots, q - 2$, are all linear combinations of $\mathfrak{w}_i^q$ and $\mathfrak{w}_{i+1}^q$ with coefficients that are at least $1$. It thus remains to prove the first two steps. For the first step: The lines $\mathcal{L}_i^{q, \text{top}}$, $i = 2, 3, \cdots, q - 1$, intersect the right side $a = 1$ of the Farey triangle $\mathscr{T}^q$ at $(1, b_i)^T$, where $b_i = \frac{1 - x_i}{y_i}$ (recall that $\mathfrak{w}_i^q = (x_i, y_i)^T$). It is easy to see that the heights $b_i$ increase as $i$ increases. (For instance, by acting on the vectors $\{\mathfrak{w}_i^q\}_{i=2}^{q-1}$, which go around the ellipse $Q_q((x, y)^T) = x^2 - \lambda_q x y + y^2 = 1$, by the linear function $T : (x, y)^T \mapsto (1 - x, y)^T$, and considering the inverse slopes of the images.) It now suffices to show that for $i = 2, 3, \cdots, q - 2$, the lines $\mathcal{L}_i^{q, \text{top}}$ and $\mathcal{L}_{i+1}^{q, \text{top}}$ intersect below on the left side $\lambda_q a + b = 1$ of the triangle $\mathscr{T}^q$ to show that the segment $\mathcal{L}_{i + 1}^{q, \text{top}} \cap \mathscr{T}^q$ lies entirely above the segment $\mathcal{L}_i^{q, \text{top}} \cap \mathscr{T}^q$, and that the former has a bigger slope than the latter. (Recall that the side $\lambda_q a + b = 1$ does *not* belong to the set $\mathscr{T}^q$.) To find the sought for intersection, we solve the simultaneous system of equations $(a_0, b_0) \cdot \mathfrak{w}_i^q = 1$ and $(a_0, b_0) \cdot \mathfrak{w}_{i+1}^q = 1$, or equivalently $\begin{pmatrix}x_i^q & y_i^q \\ x_{i + 1}^q & y_{i + 1}^q\end{pmatrix} \begin{pmatrix}a_0 \\ b_0\end{pmatrix} = 1$, for $(a_0, b_0)^T$. Since $\mathfrak{w}_i^q \wedge \mathfrak{w}_{i+1}^q = x_i^q y_{i+1}^q - x_{i+1}^q y_i^q = 1$, we have $\begin{pmatrix}a_o \\ b_0\end{pmatrix} = \begin{pmatrix}y_{i+1}^q & -y_i^q \\ -x_{i+1}^q & x_i^q\end{pmatrix} \begin{pmatrix}1 \\ 1\end{pmatrix} = \begin{pmatrix}y_{i + 1}^q - y_i^q \\ -x_{i+1}^q + x_i^q\end{pmatrix}$. Recalling that $(x_{i+1}^q, y_{i+1}^q)^T = \mathfrak{w}_{i+1}^q = U_q \mathfrak{w}_i^q = (\lambda_q x_i^q - y_i^q, x_i^q)^T$, we have $$\begin{aligned} \lambda_q a_0 + b_0 &=& \lambda_q (y_{i+1}^q - y_i^q) + (-x_{i+1}^q + x_i^q) \\ &=& \lambda_q (x_i^q - y_i^q) + (-\lambda_q x_i^q + y_i^q + x_i^q) \\ &=& (1 - \lambda_q) y_i^q + x_i^q.\end{aligned}$$ The intersection $(a_0, b_0)^T$ thus lies on or below $\lambda_q a + b = 1$ if $(1 - \lambda_q)y_i^q + x_i^q \leq 1$, or equivalently $\frac{x_i^q - 1}{\lambda_q - 1} \leq y_i^q$. (Recall that $\lambda_q = 2 \cos(\pi/q) \geq 1$ for $q \geq 3$.) Now we consider the ellipse $x^2 - \lambda_q x y + y^2 = 1$ and the line $y = \frac{x - 1}{\lambda_q - 1}$. The two points $\mathfrak{w}_0^q = (1, 0)^T$ and $\mathfrak{w}_1^q = (\lambda_q, 1)^T$ lie at the intersection of the aforementioned ellipse and line, and so the remaining points $\{\mathfrak{w}_i^q\}_{i=2}^{q-1}$ lie above the line $y = \frac{x - 1}{\lambda_q - 1}$, thus proving the inequality $y_i \geq \frac{x_i - 1}{\lambda_q - 1}$ for all $i = 2, 3, \cdots, q - 2$. For the second step: If $i = 2, 3, \cdots, q - 1$, the slope of the line segment $\mathcal{L}_{i}^{q, \text{bot}}$ agrees with that of $\mathcal{L}_{i}^{q, \text{top}}$, and so exceeds that of $\mathcal{L}_{i-1}^{q, \text{top}}$. It thus suffices to show that for $i = 2, 3, \cdots, q - 1$, the lines $\mathcal{L}_i^{q, \text{bot}}$ and $\mathcal{L}_{i - 1}^{q, \text{top}}$ intersect at a point on the right of the side $a = 1$ of the triangle $\mathscr{T}^q$. Towards that end, we compare the heights $b_{i-1} = \frac{1-x_{i-1}^q}{y_{i-1}^q}$ and $b_i^\prime = \frac{-x_1^q}{y_1^q}$ at which the lines $\mathcal{L}_{i-1}^{q, \text{top}}$ and $\mathcal{L}_i^{q, \text{bot}}$ intersect the side $a = 1$ of $\mathcal{T}^q$. We have that $b_{i-1} \geq b_i^\prime$ if and only if $y_i^q \geq x_{i-1}^q y_i^q - x_i^q y_{i-1}^q = \mathfrak{w}_{i-1}^q \wedge \mathfrak{w}_i^q = 1$, which is true for all $i = 2, 3, \cdots, q - 1$ by \[proposition: column - unimodular pair identification\]. This ends the proof. The $h_\cdot$-Periodic Points in ${\operatorname{SL}}(2, {\mathbb{R}})/G_q$, and the ${\operatorname{BCZ}}_q$-Periodic Points in $\mathscr{T}^q$ ------------------------------------------------------------------------------------------------------------------------------------------------ \[lemma: vertical vectors and periodicity\] For any $A \in {\operatorname{SL}}(2, {\mathbb{R}})$, the following are equivalent. 1. The set $A\Lambda_q$ contains a vertical vector. 2. There exists $s_0 > 0$ such that $h_{s_0}(A\Lambda_q) = A\Lambda_q$. That is, $A\Lambda_q$ is $h_\cdot$-periodic. 3. There exists $\tau_0 > 0$ such that $A\Lambda_q \cap S_{\tau_0} = \emptyset$. Moreover, if $A\Lambda_q$ contains a vertical vector of length $a$, then the $h_\cdot$-period of $A\Lambda_q$ is $\lambda_q a^2$. We prove $(1) \Rightarrow (2) \Rightarrow (3)$ directly, and $(3) \Rightarrow (1)$ by contradiction. First, we note that $\Lambda_q$ is $h_\cdot$-periodic since $h_\lambda \in G_q$, and so $h_\lambda \Lambda_q = \Lambda_q$. We also note that for any $s, t \in {\mathbb{R}}$, $\tau > 0$, and $B \in {\operatorname{SL}}(2, {\mathbb{R}})$ we have $h_t (s_\tau h_s B\Lambda_q) =s_\tau h_s (h_{t\tau^2} B\Lambda_q)$, and so $B\Lambda_q$ is $h_\cdot$-periodic iff $s_\tau h_s B\Lambda_q$ is $h_\cdot$-periodic. For $(1) \Rightarrow (2)$: Let $A\Lambda_q$ contain a vertical vector $(0, a)^T$, with $a > 0$. Then $s_a A\Lambda_q$ contains the vertical vector $(0, 1)^T$. Pick any vector $\mathbf{u}_0 \in s_a A\Lambda_q$ such that $\mathbf{u}_0 \wedge (0, 1)^T = 1$ (and so the $x$-component of $\mathbf{u}_0$ is $1$). If $s = {\operatorname{slope}}(\mathbf{u}_0) \neq 0$, then $h_s\mathbf{u}_0$ is a horizontal vector with the same $x$-component as $\mathbf{u}_0$, i.e. $1$, and $h_s (0, 1)^T = (0, 1)^T$. By \[lemma: orbit-stabilizer lemma\], $h_s s_a A\Lambda_q = \Lambda_q$, and so $A\Lambda_q = s_\frac{1}{a} h_{-s} \Lambda_q$, from which $A\Lambda_q$ is $h_\cdot$-periodic. For $(2) \Rightarrow (3)$: Let $\tau_1 > 0$ be such that $A\Lambda_q \cap S_{\tau_1} \neq \emptyset$. For any vector $\mathbf{u}_0 \in A\Lambda_q \cap S_{\tau_1}$, and any $s > 0$, the vector $h_s \mathbf{u}_0$ has the same $x$-component as $\mathbf{u}_0$, and ${\operatorname{slope}}(h_s \mathbf{u}_0) = {\operatorname{slope}}(\mathbf{u}_0) - s$. If $s_0$ is an $h_\cdot$-period of $A\Lambda_q$, then the set of lengths of the finitely many horizontal vectors that appear in $h_s(A\Lambda_q \cap S_{\tau_1})$ as $s$ goes from $0$ to $s_0$ agrees with the set of $x$-components of the vectors in $A\Lambda_q \cap S_{\tau_1}$. This implies that the $X$-components of vectors in $A\Lambda_q$ are bounded from below, and so there must exist a $\tau_0 > 0$ such that $A\Lambda_q \cap S_{\tau_0} = \emptyset$. Finally, we prove $(3) \Rightarrow (1)$ by contradiction. If $A\Lambda_q$ contains no vertical vectors, then $\mathbf{v} = (0, 1)^T$ is not parallel to any vector in $A\Lambda_q$. By \[corollary: odds and ends\], there exists sequences of unimodular pairs $\{\mathbf{u}_{0, n}, \mathbf{u}_{1, n}\}_{n=1}^\infty$ such that for each $n \geq 2$, the vector $\mathbf{v}$ belongs to the sector $(0, \infty)\mathbf{u}_{0, n} + (0, \infty)\mathbf{u}_{1, n}$, and $\mathbf{u}_{0, n}, \mathbf{u}_{1, n}$ are $G_q$-Stern-Brocot children of $\mathbf{u}_{0, n - 1}, \mathbf{u}_{1, n - 1}$. From \[equation: sin inequalities\], we get $$0 < \mathbf{u}_{0, n} \wedge (0, 1)^T < \frac{1}{\|\mathbf{u}_{1, n}\|} \to 0.$$ That is, $A\Lambda_q$ contains vectors with arbitrarily small positive $x$-components. Finally, if $A\Lambda_q$ contains a vertical vector of length $a$, we showed earlier in this proof that $A\Lambda_q$ must be of the form $s_\frac{1}{a} h_{-s} \Lambda_q$ for some $s \in {\mathbb{R}}$. For any $t \in {\mathbb{R}}$, we have $h_t\left(s_\frac{1}{a} h_{-s} \Lambda_q\right) = s_\frac{1}{a} h_{-s} \left(h_\frac{t}{a^2} \Lambda_q\right)$, which implies that the $h_\cdot$-period of $A\Lambda_q$ is $a^2$ times that of $\Lambda_q$. \[corollary: characterizing BCZ periodic points\] For any $(a, b) \in \mathscr{T}^q$, the following are equivalent. 1. The point $(a, b)$ is ${\operatorname{BCZ}}_q$-periodic. 2. The set $g_{a, b}\Lambda_q$ is $h_\cdot$-periodic. 3. The ratio $b/a$ is the (inverse) slope of a vector in $\Lambda_q$. That the first two claims are equivalent is obvious, and so we proceed to characterize the points $(a, b) \in \mathscr{T}^q$ for which $g_{a, b}\Lambda_q$ is $h_\cdot$-periodic. Note that $g_{a, b} = s_a g_{1, b/a}$, and that for any $s \in {\mathbb{R}}$, $h_s (s_a g_{1, b/a}) = s_a (h_{a^2 s} g_{1, b/a})$. That is, $g_{a, b} \Lambda_q$ is $h_\cdot$-periodic if $g_{1, b/a}\Lambda_q$ is $h_\cdot$-periodic. By \[lemma: vertical vectors and periodicity\], $g_{1,b/a}\Lambda_q$ is $h_\cdot$-periodic iff it contains a vertical vector. Since $g_{1, b/a}$ is a horizontal shear, the set $g_{1,b/a}\Lambda_q$ contains a vertical vector exactly when $b/a$ is the inverse slope of a vector in $\Lambda_q$. The claim now follows from the symmetry of $\Lambda_q$ against the line $y = x$. The $G_q$-Next-Term Algorithm {#subsection: G_q next term algorithm} ----------------------------- \[theorem: next term algorithm\] Let $A \in {\operatorname{SL}}(2, {\mathbb{R}})$, $\tau > 0$ be such that $A \Lambda_q \cap S_\tau \neq \emptyset$, and $\{\mathbf{u}_n = (q_n, a_n)^T\}_{n=0}^\infty$ be elements of $A \Lambda_q \cap S_\tau$ with successive slopes. The set $s_\frac{1}{\tau} h_{{\operatorname{slope}}(\mathbf{u}_0)} A \Lambda_q$ has a horizontal vector $(q_0/\tau, 0)^T \in s_\frac{1}{\tau} h_{{\operatorname{slope}}(\mathbf{u}_0)} A \Lambda_q \cap S_1$, and hence corresponds to a unique point $(a, b) \in \mathscr{T}^q$ (i.e. $s_\frac{1}{\tau} h_{{\operatorname{slope}}(\mathbf{u}_0)} A \Lambda_q = g_{a,b}\Lambda_q$). The following are then true. 1. For each $n \geq 0$, the set $s_\frac{1}{\tau} h_{{\operatorname{slope}}(\mathbf{u}_n)} A \Lambda_q$ has a horizontal vector $(q_n/\tau, 0)^T \in s_\frac{1}{\tau} h_{{\operatorname{slope}}(\mathbf{u}_n)} A \Lambda_q \cap S_1$, and corresponds to ${\operatorname{BCZ}}_q^n(a, b)$. 2. If we denote the $x$-component of ${\operatorname{BCZ}}^n(a, b)$ by $L_n^q(a, b)$ for all $n \geq 0$, then the $x$-components of the vectors $\{\mathbf{u}_n\}_{n=0}^\infty$ are equal to $$q_n = \tau L_n^q(a, b) = \tau L_0^q({\operatorname{BCZ}}_q^n(a, b)).$$ Moreover, the $y$-components of the vectors $\{\mathbf{u}_n\}_{n=0}^\infty$ can be recursively generated using the formula $$a_{n + 1} = q_{n + 1} \left(\frac{a_n}{q_n} + \frac{1}{\tau^2} R_q({\operatorname{BCZ}}_q^n(a, b))\right)$$ for all $n \geq 0$. This motivates the following definition. For any $A \in {\operatorname{SL}}(2, {\mathbb{R}})$, $\tau > 0$ with $A\Lambda_q \cap S_\tau \neq \emptyset$, and $\mathbf{u} \in A \Lambda_q \cap S_\tau$, we refer to the unique point in the Farey triangle $\mathscr{T}^q$ corresponding to $s_\frac{1}{\tau} h_{{\operatorname{slope}}(\mathbf{u})} A \Lambda_q$ from \[theorem: next term algorithm\] as the *$G_q$-Farey triangle representatitve* of the triple $(A, \tau, \mathbf{u})$, and denote it by ${\operatorname{FTR}}_q(A, \tau, \mathbf{u})$. Using this notation, we can succinctly rewrite the first claim in \[theorem: next term algorithm\] as $${\operatorname{FTR}}_q(A, \tau, \mathbf{u}_n) = {\operatorname{BCZ}}_q^n\left({\operatorname{FTR}}_q(A, \tau, \mathbf{u}_0)\right)$$ for all $n \geq 0$. \[remark: practical next term algorithm\] For any $\tau \geq 1$, the vector $\mathbf{u}_0 = (1, 0)^T$ belongs to $\Lambda_q \cap S_\tau$, and so ${\operatorname{FTR}}_q(I_2, \tau, \mathbf{u}_0)$ is well defined. We have $$s_\frac{1}{\tau} h_{{\operatorname{slope}}(\mathbf{u_0})} \Lambda_q = s_\frac{1}{\tau} \Lambda_q = s_\frac{1}{\tau} T_q^{\left\lfloor\frac{\tau}{\lambda_q}\right\rfloor} \Lambda_q = g_{\frac{1}{\tau}, \left\lfloor\frac{\tau}{\lambda_q}\right\rfloor\frac{\lambda_q}{\tau}}\Lambda_q,$$ with $0 < \frac{1}{\tau} \leq 1$, and $1 - \lambda_q \left( \frac{1}{\tau}\right) < \left\lfloor\frac{\tau}{\lambda_q}\right\rfloor\frac{\lambda_q}{\tau} \leq 1$, from which $\left(\frac{1}{\tau}, \left\lfloor\frac{\tau}{\lambda_q}\right\rfloor\frac{\lambda_q}{\tau}\right) \in \mathscr{T}^q$ is the $G_q$-Farey triangle representative of the triple $(\Lambda_q, \tau, \mathbf{u}_0)$. By the symmetry of $\Lambda_q$ against the lines $y = \pm x$, $x = 0$, and $y = 0$, it suffices to generate the vectors in $\Lambda_q \cap S_\tau$ with slopes in $[0, 1]$ to get all the vectors in $\Lambda_q \cap [-\tau, \tau]^2$. For each $n \geq 0$, a direct calculation gives $s_\frac{1}{\tau} h_{{\operatorname{slope}}(\mathbf{u}_n)} \mathbf{u}_n = (q_n/\tau, 0)$, which is a horizontal vector of length not exceeding $1$ in $s_\frac{1}{\tau} h_{{\operatorname{slope}}(\mathbf{u}_n)} A\Lambda_q$. By the first claim in \[theorem: G\_q BCZ maps\], the set $s_\frac{1}{\tau} h_{{\operatorname{slope}}(\mathbf{u}_n)} A\Lambda_q$ corresponds to a unique point $(c_n, d_n) \in \mathscr{T}^q$ with $q_n/\tau = a_n$, and $(c_0, d_0) = (a, b)$. The vectors $\mathbf{u}_n$ and $\mathbf{u}_{n+1}$ have consecutive slopes in $A\Lambda_q \cap S_\tau$, and so the two vectors $s_\frac{1}{\tau} h_{{\operatorname{slope}}(\mathbf{u}_n)} \mathbf{u}_n$ and $s_\frac{1}{\tau} h_{{\operatorname{slope}}(\mathbf{u}_n)} \mathbf{u}_{n+1}$ have consecutive slopes in $s_\frac{1}{\tau} h_{{\operatorname{slope}}(\mathbf{u}_n)} A\Lambda_q \cap S_1$. In other words, the vector $s_\frac{1}{\tau} h_{{\operatorname{slope}}(\mathbf{u}_n)} \mathbf{u}_{n+1}$ is the vector of smallest positive slope in $g_{c_n, d_n} \Lambda_q \cap S_1$, from which $$\begin{aligned} R_q(c_n, d_n) &=& {\operatorname{slope}}\left(s_\frac{1}{\tau} h_{{\operatorname{slope}}(\mathbf{u}_n)} \mathbf{u}_{n+1}\right) \\ &=& \tau^2 \left({\operatorname{slope}}(\mathbf{u}_{n+1}) - {\operatorname{slope}}(\mathbf{u}_n)\right) \\ &=& \tau^2 \left(\frac{a_{n+1}}{q_{n+1}} - \frac{a_n}{q_n}\right),\end{aligned}$$ and $$h_{R_q(c_n, d_n)} g_{c_n, d_n}\Lambda_q = s_\frac{1}{\tau} h_{{\operatorname{slope}}(\mathbf{u}_{n+1})} A\Lambda_q = g_{c_{n+1}, d_{n+1}}\Lambda_q,$$ and so $${\operatorname{BCZ}}_q(c_n, d_n) = (c_{n+1}, d_{n+1})$$ by the second claim in \[theorem: G\_q BCZ maps\]. By induction, we get $(c_n, d_n) = {\operatorname{BCZ}}_q^n(c_0, d_0)$, $q_n = \tau c_n = \tau L_n^q(c_0, d_0)$, and the sought for recursive expression for $a_{n+1}$. A Poincaré Cross Section for the Horocycle Flow on the Quotient ${\operatorname{SL}}(2, {\mathbb{R}})/G_q$ {#section: cross section} ========================================================================================================== Let $X_q$ be the homogeneous space ${\operatorname{SL}}(2, {\mathbb{R}})/G_q$, $\mu_q$ be the probability Haar measure on $X_q$ (i.e. $\mu_q(X_q) = 1$), and $\Omega_q$ be the subset of $X_q$ corresponding to sets $A\Lambda_q$, $A \in {\operatorname{SL}}(2, {\mathbb{R}})$, with a horizontal vector of length not exceeding $1$. Note that $\Omega_q$ can be identified with the Farey triangle $\mathscr{T}^q$ via $\left((a, b) \in \mathscr{T}^q\right) \mapsto (g_{a,b}G_q \in \Omega_q)$ by \[lemma: orbit-stabilizer lemma\] and \[theorem: G\_q BCZ maps\]. Finally, let $m_q = \frac{2}{\lambda_q}dadb$ be the Lebesgue probability measure on $\mathscr{T}^q$. Following [@Athreya2013-ql], we have the following. \[theorem: cross section\] The triple $(\mathscr{T}^q, m_q, {\operatorname{BCZ}}_q)$, with $\mathscr{T}^q$ identified with $\Omega_q$, is a cross section to $(X_q, \mu_q, h_\cdot)$, with roof function $R_q$. Consider the suspension space $$S_{R_q}\mathscr{T}^q := \{\left((a, b), s\right) \in \mathscr{T}^q \times {\mathbb{R}}\mid 0 \leq s \leq R_q(a, b)\} / \sim_q,$$ with $\left((a, b), R_q(a, b)\right) \sim_q \left({\operatorname{BCZ}}_q(a, b), 0\right)$ for all $(a, b) \in \mathscr{T}^q$, as a subset of $X_q$. The suspension flow of $S_{R_q}\mathscr{T}^q$ can be identified with the horocycle flow $h_\cdot$ on $S_{R_q}\mathscr{T}^q$ as a subset of $X_q$ by \[theorem: G\_q BCZ maps\]. The probability measure $dm_q^{R_q} = \frac{1}{m_q(R_q)} dm_qds$ is $h_\cdot$-invariant, and the suspension space $S_{R_q}\mathscr{T}^q$ contains non-closed horocycles (e.g by \[lemma: vertical vectors and periodicity\] and \[corollary: characterizing BCZ periodic points\]). By Dani-Smillie [@Dani1984-ji], the subset $S_{R_q}\mathscr{T}^q$ has full $\mu_q$ measure in $X_q$, and the probability measures $dm_q^{R_q}$ and $\mu_q$ can be identified. This proves the claim. Limiting Distributions of Farey Triangle Representatives, and Equidistribution of the Slopes of $\Lambda_q$ {#subsection: limiting distributions of FTR} ----------------------------------------------------------------------------------------------------------- (-0.5,0)–(4.5,0) node\[right\][$\scriptstyle x$]{}; (0,-0.5)–(0,4.5) node\[above\][$\scriptstyle y$]{}; (0,0) – (3,1.5) – (3,3); (1, 0) circle \[radius=0.1em\]; (3.236, 1) circle \[radius=0.1em\]; (4.236, 1.618) circle \[radius=0.1em\]; (3.618, 1.618) circle \[radius=0.1em\]; (1.618, 1) circle \[radius=0.1em\]; (4.236, 3.236) circle \[radius=0.1em\]; (4.236, 3.618) circle \[radius=0.1em\]; (1.618, 1.618) circle \[radius=0.1em\]; (0, 1) circle \[radius=0.1em\]; (1, 3.236) circle \[radius=0.1em\]; (1.618, 4.236) circle \[radius=0.1em\]; (1.618, 3.618) circle \[radius=0.1em\]; (1, 1.618) circle \[radius=0.1em\]; (3.236, 4.236) circle \[radius=0.1em\]; (3.618, 4.236) circle \[radius=0.1em\]; (3, -0.25) – (3, 4.5); (0, 0) – (4.5, 2.25); (0, 0) – (4.5, 4.5); at (3, -0.5) [$\scriptstyle x = 3$]{}; at (5.1, 2.25) [$\scriptstyle y = 0.5x$]{}; at (5.1, 4.5) [$\scriptstyle y = x$]{}; For any $A \in {\operatorname{SL}}(2, {\mathbb{R}})$, $\tau > 0$, and interval $I \subseteq {\mathbb{R}}$, we denote by $$\mathcal{F}_I(A\Lambda_q, \tau) := \{\mathbf{u} \in A\Lambda_q \cap S_\tau \mid {\operatorname{slope}}(\mathbf{u}) \in I\}$$ the set of vectors in $A\Lambda_q$ with positive $x$-components not exceeding $\tau$, and slopes in $I$. If $I \subset {\mathbb{R}}$ is a finite interval, we write $$N_I(A\Lambda_q, \tau) := \# \mathcal{F}_I(A\Lambda_q, \tau)$$ for the number of elements of $\mathcal{F}_I(A\Lambda_q, \tau)$. Note that if $I$ is a non-degenerate interval, then $\lim_{\tau \to \infty} N_I(A\Lambda_q, \tau) = \infty$ by the density of the slopes of $A\Lambda_q$ from \[corollary: odds and ends\]. For any $A \in {\operatorname{SL}}(2, {\mathbb{R}})$, finite, non-empty, non-degenerate interval $I \subset {\mathbb{R}}$, and $\tau > 0$ with $\mathcal{F}_I(A\Lambda_q, \tau) = \{\mathbf{u}_i\}_{i=0}^{N_I(A\Lambda_q, \tau) - 1} \neq \emptyset$, we define the following probability measure on the Farey triangle $\mathscr{T}^q$ $$\begin{aligned} \rho_{A\Lambda_q, I, \tau} &:=& \frac{1}{N_I(A\Lambda_q, \tau)} \sum_{i = 0}^{N_I(A\Lambda_q, \tau) - 1} \delta_{{\operatorname{FTR}}_q(A, \tau, \mathbf{u}_i)} \\ &=& \frac{1}{N_I(A\Lambda_q, \tau)} \sum_{i = 0}^{N_I(A\Lambda_q, \tau) - 1} \delta_{{\operatorname{BCZ}}_q^i\left({\operatorname{FTR}}_q(A, \tau, \mathbf{u}_0)\right)}.\end{aligned}$$ \[theorem: weak limit and asymptotic growth\] Let $A \in {\operatorname{SL}}(2, {\mathbb{R}})$, and $I \subset {\mathbb{R}}$ be a finite, non-empty, non-degenerate interval in ${\mathbb{R}}$. Then as $\tau \to \infty$, the number of elements of $\mathcal{F}_I(A\Lambda_q, \tau)$ has the asymptotic growth $$N_I(A\Lambda_q, \tau) \sim \frac{|I|}{m_q(R_q)} \tau^2,$$ and the measures $\rho_{A\Lambda_q, I, \tau}$ converge weakly $$\rho_{A\Lambda_q, I, \tau} \rightharpoonup m_q$$ to the probability Lebesgue measure $m_q$. \[corollary: equidistribution of slopes\] For any $A \in {\operatorname{SL}}(2, {\mathbb{R}})$, and any finite interval $\emptyset \neq I \subset {\mathbb{R}}$, the slopes of the vectors $\mathcal{F}_I(A\Lambda_q, \tau)$ equidistribute in $I$ as $\tau \to \infty$. For $\tau > 0$ with $\mathcal{F}_I(A\Lambda_q, \tau) \neq \emptyset$, we define the measures $$\sigma_{A\Lambda_q, I, \tau} = \frac{N_I(A\Lambda_q, \tau)}{\tau^2} \rho_{A\Lambda_q, I, \tau}$$ on the Farey triangle $\mathscr{T}^q$. Denote by $d\sigma_{A\Lambda_q, I, \tau}^{R_q}$ the measure $d\sigma_{g\Lambda_q, I, \tau}ds$ on the suspension space $S_{R_q}\mathscr{T}^q$ (which can identified with $X_q$ by \[theorem: cross section\]). In what follows, we denote the elements of $\mathcal{F}_I(A\Lambda_q, \tau)$ by $\{\mathbf{u}_i = \mathbf{u}_i(A\Lambda_q, I, \tau)\}_{i=0}^{N_I(A\Lambda_q, \tau) - 1}$, and write $\mathbf{u}_{N_I(A\Lambda_q, \tau)} = \mathbf{u}_{N_I(A\Lambda_q, \tau)}(A\Lambda_q, I, \tau)$ for the element of $A\Lambda_q \cap S_\tau$ of smallest slope bigger than any value in $I$. By the density of the slopes of $A\Lambda_q$ from \[corollary: odds and ends\], we have that ${\operatorname{slope}}(\mathbf{u}_0)$ and ${\operatorname{slope}}(\mathbf{u}_{N_I(A\Lambda_q, \tau)})$ converge to the end points of the interval $I$, which we denote $\alpha$ and $\beta$ (i.e. $|I| = \beta - \alpha$). We show the convergence $\sigma_{A\Lambda_q, I, \tau} \rightharpoonup |I|/m_q(R_q) m_q$ by proving the convergence $\sigma_{A\Lambda_q, I, \tau}^{R_q} \rightharpoonup |I| \mu_q$. Given any continuous, bounded function $f : X_q \to {\mathbb{R}}$, we have $$\begin{aligned} \sigma_{g\Lambda_q, I, \tau}^{R_q}(f) &=& \frac{1}{\tau^2} \int_{\tau^2 {\operatorname{slope}}(\mathbf{u}_0)}^{\tau^2 {\operatorname{slope}}(\mathbf{u}_{N_I(A\Lambda_q, \tau)})} f\left(h_s \left(s_\frac{1}{\tau}AG_q\right)\right)\,ds \\ &=& \frac{1}{\tau^2} \int_{\tau^2 {\operatorname{slope}}(\mathbf{u}_0)}^{\tau^2 {\operatorname{slope}}(\mathbf{u}_{N_I(A\Lambda_q, \tau)})} f\left(s_\frac{1}{\tau} h_\frac{s}{\tau^2}AG_q\right)\,ds \\ &=& \int_{{\operatorname{slope}}(\mathbf{u}_0)}^{{\operatorname{slope}}(\mathbf{u}_{N_I(A\Lambda_q, \tau)})} f\left(s_\frac{1}{\tau} h_t AG_q\right)\,dt \\ &=& \int_\alpha^\beta f\left(s_\frac{1}{\tau} h_t AG_q\right)\,dt + o(1) \\ &\to& (b - a) \mu_q(f)\end{aligned}$$ as $\tau \to \infty$ (with the convergence of the measures supported on horocycles following from, for example, [@Kleinbock1996-ag 2.2.1]). This proves the weak convergence $\sigma_{g\Lambda_q, I, \tau}^{R_q} \rightharpoonup |I| \mu_q$. Denoting by $\pi_q : S_{R_q}\mathscr{T}^q \to \mathscr{T}^q$ the projection map $\left(\left((a, b), s\right) \in S_{R_q}\mathscr{T}^q\right) \mapsto \left((a, b) \in \mathscr{T}^q\right)$, we thus have $$\sigma_{g\Lambda_q, I, \tau} = \frac{1}{R_q} (\pi_q)_\ast \sigma_{g\Lambda_q, I, \tau}^{R_q} \rightharpoonup \frac{|I|}{R_q} (\pi_q)_\ast \mu_q = \frac{|I|}{m_q(R_q)} m_q.$$ From $\rho_{A\Lambda_q, I, \tau}(\mathscr{T}^q) = 1$, we get $$\lim_{\tau \to \infty} \frac{N_I(A\Lambda_q, \tau)}{\tau^2} = \lim_{\tau \to \infty} \sigma_{A\Lambda_q, I, \tau}(\mathscr{T}^q) = \frac{|I|}{m_q(R_q)} m_q(\mathscr{T}^q) = \frac{|I|}{m_q(R_q)},$$ which is the asymptotic growth from \[theorem: weak limit and asymptotic growth\]. This also gives the weak limit $\rho_{A\Lambda_q, I, \tau} \rightharpoonup m_q$. As for \[corollary: equidistribution of slopes\], if $\emptyset \neq J \subseteq I$ is any non-empty subinterval of $I$, we have $$\lim_{\tau \to \infty} \frac{N_J(A\Lambda_q, \tau)}{N_I(A\Lambda_q, \tau)} = \frac{|J|}{|I|},$$ which proves the sought for equidistribution. Applications {#section: applications} ============ In this section, we give a few applications of the $G_q$-BCZ maps to the statistics of subsets of $\Lambda_q$. In \[subsection: counting in triangles\], we derive the main asymptotic term for the number of vectors of $\Lambda_q$ in homothetic dilations of triangles. In \[subsubsection: slopegap distribution\], we derive the distribution of the slope gaps of $\Lambda_q$. Finally, in \[subsubsection: centdist distribution\], we derive the distribution of the Euclidean distances between the centers of $G_q$ -Ford circles. Several other applications of the $G_3$-BCZ map to the statistics of the visible lattice points $\Lambda_3 = {\mathbb{Z}}_\text{prim}^2 = \{(x, y) \in {\mathbb{Z}}^2 \mid \gcd(x, y) = 1\}$ can be similarly extended–almost verbatim–to general $\Lambda_q$. This list includes, but is not limited to, an old Diophantine approximation problem of [@Erdos1958-xi] Erdös, P., Szüsz, P., & Turán solved independently by Xiong and Zaharescu [@Xiong2006-ef], and Boca [@Boca2008-bu] for $G_3 = {\operatorname{SL}}(2, {\mathbb{Z}})$, and Heersink [@Heersink2016-hg] for finite index subgroups of $G_3 = {\operatorname{SL}}(2, {\mathbb{Z}})$; the average depth of cusp excursions of the horocycle flow on $X_2 = {\operatorname{SL}}(2, {\mathbb{R}})/G_3$ by Athreya and Cheung [@Athreya2013-ql]; and the statistics of weighted Farey sequences by Panti [@Panti2015-jz]. Asymptotic Growth of the Number of Elements of $\Lambda_q$ in Homothetic Dilations of Triangles {#subsection: counting in triangles} ----------------------------------------------------------------------------------------------- For any $A \in {\operatorname{SL}}(2, {\mathbb{R}})$, $\tau > 0$, and finite interval $I \subset {\mathbb{R}}$, the set $\mathcal{F}_I(A\Lambda_q, \tau)$ introduced in \[subsection: limiting distributions of FTR\] is the collection of points of $A\Lambda_q$ which belong to the triangle $\{(x, y)^T \in {\mathbb{R}}^2 \mid y/x \in I, 0 < x \leq \tau\}$. We have the main term for the asymptotic growth rate of the number of aforementioned vectors $N_I(A\Lambda_q, \tau)$ as $\tau \to \infty$, which can be immediately interpreted as a statement on the asymptotic growth of the number of vectors of $A\Lambda_q$ in homothetic dilations of triangles that have a vertex at the origin as we do in \[proposition: counting in triangles\]. In \[corollary: equidistribution in the square\], we show the equidistribution of the homothetic dilations $\frac{1}{\tau}\Lambda_q$ in the square $[-1, 1]^2$ as $\tau \to \infty$. In what follows, we write $f(\tau) \sim g(\tau)$ as $\tau \to \infty$ for any two functions $f, g$ to indicate that $\lim_{\tau \to \infty} f(\tau)/g(\tau) = 1$. \[proposition: counting in triangles\] Let $\Delta$ be a triangle in the plane ${\mathbb{R}}^2$ with one vertex at the origin. Then for any $A \in {\operatorname{SL}}(2, {\mathbb{R}})$, and any $\tau > 0$ the number of elements $\#\left(A\Lambda_q \cap \tau \Delta \right)$ has the asymptotic growth rate $$\#\left(A\Lambda_q \cap \tau \Delta\right) \sim \left(\frac{2}{m_q(R_q)} {\operatorname{area}}(\Delta)\right) \tau^2$$ as $\tau \to \infty$. We also get the following. \[corollary: equidistribution in the square\] For any $A \in {\operatorname{SL}}(2, {\mathbb{R}})$, and $\tau \geq 1$, let $\lambda_\tau^{A\Lambda_q}$ be the probability measure defined for any Borel subset $C$ of the square $[-1, 1]^2$ by $$\lambda_\tau^{A\Lambda_q}(C) = \frac{\#(\frac{1}{\tau}A\Lambda_q \cap C)}{\#\left(\frac{1}{\tau}A\Lambda_q \cap [-1, 1]^2\right)}.$$ Then the measures $\lambda_\tau^{A\Lambda_q}$ converge weakly to the Lebesgue probability measure ${\operatorname{Unif}}_{[-1, 1]^2}$ on $[-1, 1]^2$ as $\tau \to \infty$. We first prove the theorem assuming that the side $L$ of the triangle $\Delta$ opposite to the origin is included in $\Delta$. Let $\operatorname{rot}_\Delta \in {\operatorname{SL}}(2, {\mathbb{R}})$ be the rotation that rotates the side of $\Delta$ opposite to the vertex at the origin onto a vertical line segment. (That is, the side of $\operatorname{rot}_\Delta \Delta$ opposite to the vertex at the origin is vertical.) Denote by $d_\Delta > 0$ the perpendicular distance from the vertex at the origin to the side of $\operatorname{rot}_\Delta \Delta$ opposite to the aforementioned vertex, and by $I_\Delta \subset {\mathbb{R}}$ the interval of slopes of the points in $\operatorname{rot}_\Delta \Delta$. For any $\tau > 0$, we have that $\tau (\operatorname{rot}_\Delta \Delta) \cap (\operatorname{rot}_\Delta A \Lambda_q) = \mathcal{F}_I(\operatorname{rot}_\Delta A \Lambda_q, \tau d)$, and that the rotation $\operatorname{rot}_\Delta$ is a bijection from $\tau \Delta \cap A \Lambda_q$ to $\tau (\operatorname{rot}_\Delta \Delta) \cap (\operatorname{rot}_\Delta A \Lambda_q)$. From this and \[theorem: weak limit and asymptotic growth\] follows that $$\#\left(A \Lambda_q \cap \tau \Delta\right) = N_{I_\Delta}(\operatorname{rot}_\Delta A \Lambda_q, \tau d) \sim \frac{|I_\Delta|}{m_q(R_q)} (\tau d_\Delta)^2 = \frac{2}{m_q(R_q)} {\operatorname{area}}(\Delta) \tau^2$$ which proves the claim. Including or excluding any of the two sides of the triangle $\Delta$ that pass through the origin does not change $|I_\Delta|$, and hence the main term for the asymptotic growth in question remains the same. We now show that the main term does not change when the side $L$ of $\Delta$ opposite to the origin is removed as well. For any $\delta > 0$, denote by $\Delta^\prime = \Delta^\prime(\Delta, \delta)$ the homothetic dilation of $\Delta$ such that $0 < {\operatorname{area}}(\Delta) - {\operatorname{area}}(\Delta^\prime) \leq \delta$. The line segment $L$ belongs to $\Delta \setminus \Delta^\prime$. By the above, $\lim_{\tau \to \infty} \left(\#(A\Lambda_q \cap \tau \Delta) - \#(A\Lambda_q \cap \tau \Delta^\prime)\right)/\tau^2 = 2\left({\operatorname{area}}(\Delta) - {\operatorname{area}}(\Delta^\prime)\right)/m_q(R_q) \leq 2\delta/m_q(R_q)$. It thus follows that for all $\epsilon > 0$, there exists $\tau_0 = \tau_0(A\Lambda_q, \Delta, \delta, \epsilon)$ such that for all $\tau > \tau_0$ we have $$\frac{\#\left(A\Lambda_q \cap \tau L\right)}{\tau^2} \leq \frac{\#(A\Lambda_q \cap \tau \Delta) - \#(A\Lambda_q \cap \tau \Delta^\prime)}{\tau^2} \leq \frac{2\delta}{m_q(R_q)} + \epsilon.$$ By the arbitrariness of $\delta$ and $\epsilon$, we get $\lim_{\tau \to \infty} \frac{\#\left(A\Lambda_q \cap \tau L\right)}{\tau^2} = 0$. This proves that adding or removing a finite number of line segments does not affect the main term for the asymptotic growth of the number of elements of $A\Lambda_q$ in homothetic dilations of triangles. That the set functions $\lambda_\tau^{A\Lambda_q}$ are probability measures on $[-1, 1]^2$ is clear. We proceed to prove that they converge weakly to ${\operatorname{Unif}}_{[-1, 1]}$. First, we note that given any rectangle $\mathcal{R}$ in the plane, we can express $\mathcal{R}$ using the union and/or difference of four triangles each having a vertex at the origin. From this follows that $\lim_{\tau \to \infty} \frac{\#(A\Lambda_q \cap \tau \mathcal{R})}{\tau^2} = \frac{2}{m_q(R_q)} {\operatorname{area}}(\mathcal{R})$. Consequently, if $\mathcal{R}$ belongs to $[-1,1]^2$, then $\lim_{\tau \to \infty} \lambda_\tau^{A\Lambda_q}(\mathcal{R}) = {\operatorname{Unif}}_{[-1, 1]^2}(\mathcal{R})$. Fix a continuous function $f : [-1, 1]^2 \to {\mathbb{R}}$. Given a $\delta > 0$, there exists a finite partition $\mathscr{P} = \mathscr{P}(f, \delta)$ of the square $[-1, 1]^2$ into rectangles such that the difference between the supremum and infimum of $f$ over each of the rectangles in the partition does not exceed $\delta$. That is, $\sup_\mathcal{R}(f) \leq \inf_\mathcal{R}(f) + \delta$ for all $\mathcal{R} \in \mathscr{P}$. (This is possible by the uniform continuity of $f$ over $[-1, 1]^2$.) Given $\epsilon > 0$, there exists $\tau_0 = \tau_0(A\Lambda_q, \mathscr{P}(f, \delta), \epsilon)$ such that $\left|\lambda_\tau^{A\Lambda_q}(\mathcal{R}) - {\operatorname{Unif}}_{[-1, 1]^2}(\mathcal{R})\right| \leq \epsilon$ for all $\tau > \tau_0$, and all $\mathcal{R} \in \mathscr{P}$. We thus have $$\begin{aligned} \lambda_\tau^{A\Lambda_q}(f) &\leq& \sum_{\mathcal{R} \in \mathscr{P}} \sup_\mathcal{R}(f)\ \lambda_\tau^{A\Lambda_q}(\mathcal{R}) \\ &\leq& \epsilon \max_{[-1,1]^2}(|f|) \#\mathscr{P} + \sum_{\mathcal{R} \in \mathscr{P}} \sup_{\mathcal{R}}(f) {\operatorname{Unif}}_{[-1, 1]^2}(\mathcal{R}) \\ &\leq& \epsilon \max_{[-1,1]^2}(|f|) \#\mathscr{P} + \delta + \sum_{\mathcal{R} \in \mathscr{P}} \inf_\mathcal{R}(f) {\operatorname{Unif}}_{[-1, 1]^2}(\mathcal{R}) \\ &\leq& \epsilon \max_{[-1,1]^2}(|f|) \#\mathscr{P}(f, \delta) + \delta + {\operatorname{Unif}}_{[-1, 1]^2}(f).\end{aligned}$$ Similarly $$\begin{aligned} \lambda_\tau^{A\Lambda_q}(f) &\geq& \sum_{\mathcal{R} \in \mathscr{P}} \inf_\mathcal{R}(f)\ \lambda_\tau^{A\Lambda_q}(\mathcal{R}) \\ &\geq& -\epsilon \max_{[-1,1]^2}(|f|) \#\mathscr{P} + \sum_{\mathcal{R} \in \mathscr{P}} \inf_\mathcal{R}(f){\operatorname{Unif}}_{[-1,1]^2}(\mathcal{R}) \\ &\geq& -\epsilon \max_{[-1,1]^2}(|f|) \#\mathscr{P} - \delta + \sum_{\mathcal{R} \in \mathscr{P}} \sup_{\mathcal{R}}(f){\operatorname{Unif}}_{[-1,1]^2}(\mathcal{R}) \\ &\geq& -\epsilon \max_{[-1,1]^2}(|f|) \#\mathscr{P}(f, \delta) - \delta + {\operatorname{Unif}}_{[-1, 1]^2}(f).\end{aligned}$$ That is, $\left|\lambda_\tau^{A\Lambda_q}(f) - {\operatorname{Unif}}_{[-1,1]^2}(f)\right| \leq \epsilon \max_{[-1,1]^2}(|f|) \#\mathscr{P}(f, \delta) + \delta$. By the arbitrariness of $\delta$ and $\epsilon$, we get $\lim_{\tau \to \infty} \lambda_\tau^{A\Lambda_q}(f) = {\operatorname{Unif}}_{[-1, 1]^2}(f)$. This proves the claim. $G_q$-Farey Statistics {#subsection: G_q Farey statistics} ---------------------- In \[proposition: limiting distributions of Farey statistics\] below we derive the limiting distribution of quantities that can be expressed as functions in the $G_q$-Farey triangle representatives (\[subsection: G\_q next term algorithm\]) of the elements of the sets $\mathcal{F}_I(A\Lambda_q, \tau)$ (\[subsection: limiting distributions of FTR\]) as $\tau \to \infty$. As examples of said distributions, we consider the slope gap distribution of $\Lambda_q$ in \[subsubsection: slopegap distribution\], and the distribution of the Euclidean distance between $G_q$-Ford circles in \[subsubsection: centdist distribution\]. \[proposition: limiting distributions of Farey statistics\] Let $F : \mathscr{T}^q \to {\mathbb{R}}$ be a function, continuous on the Farey triangle $\mathscr{T}^q$ except perhaps on the image of finitely many $C^1$ curves $\{c_i : I_i \to \mathscr{T}^q\}_{i=1}^m$, with $I_i \subset {\mathbb{R}}$ being finite, closed intervals of ${\mathbb{R}}$. For any $A \in {\operatorname{SL}}(2, {\mathbb{R}})$, and any finite interval $I \subset {\mathbb{R}}$, the limit of the distribution $${\operatorname{FareyStat}}_{F,\tau}^{A\Lambda_q,I}(t) := \frac{\# \left\{\mathbf{u} \in \mathcal{F}_I(A\Lambda_q, \tau) \mid F\left({\operatorname{FTR}}_q(A \Lambda_q, \tau, \mathbf{u})\right) \geq t\right\}}{N_I(A \Lambda_q, \tau)}$$ as $\tau \to \infty$ exists for all $t \in {\mathbb{R}}$, and is equal to $${\operatorname{FareyStat}}_F(t) = m_q\left({\mathds}{1}_{F \geq t}\right),$$ where ${\mathds}{1}_{F \geq t}$ is the indicator function of the subset $$\left\{(a, b) \in \mathscr{T}^q \mid F(a,b) \geq t\right\}$$ of $\mathscr{T}^q$, and $m_q$ is the Lebesgue probability measure $dm_q = \frac{2}{\lambda_q} dadb$ on $\mathscr{T}^q$. Fix $t \in {\mathbb{R}}$. We then have $$\begin{aligned} {\operatorname{FareyStat}}_{F,\tau}^{A\Lambda_q,I}(t) &=& \frac{\# \left\{\mathbf{u} \in \mathcal{F}_I(A \Lambda_q, \tau) \mid F \left({\operatorname{FTR}}_q(A, \tau, \mathbf{u})\right) \geq t\right\}}{N_I(A\Lambda_q, \tau)} \\ &=& \frac{1}{N_I(A \Lambda_q, \tau)} \sum_{i=0}^{N_I(A \Lambda_q, \tau) - 1} {\mathds}{1}_{F \geq t}\left({\operatorname{FTR}}_q(A, \tau, \mathbf{u}_i\right) \\ &=& \rho_{A \Lambda_q, I, \tau}\left({\mathds}{1}_{F \geq t}\right),\end{aligned}$$ and so we proceed to show that $\lim_{\tau \to \infty} \rho_{A \Lambda_q, I, \tau} \left({\mathds}{1}_{F \geq t}\right) = m_q\left({\mathds}{1}_{F \geq t}\right)$. Consider the following sets $$\begin{aligned} A_t &=& \{(a, b) \in \mathscr{T}^q \mid F(a,b) \geq t\}, \\ B_t &=& \{(a, b) \in \mathscr{T}^q \mid F(a, b) \geq t\} \cup \bigcup_{i=1}^m c_i(I_i), \text{ and} \\ C &=& \bigcup_{i=1}^m c_i(I_i).\end{aligned}$$ The set $C_t$ is null with respect to the measure $m_q$, and $A_t \Delta B_t \subseteq C$, and so $m_q({\mathds}{1}_{A_t}) = m_q({\mathds}{1}_{B_t})$. The sets $B_t$ and $C$ are closed, and so their indicator functions ${\mathds}{1}_{B_t}$ and ${\mathds}{1}_C$ are bounded, and upper semi-continuous. Theorem \[theorem: weak limit and asymptotic growth\] gives $\lim_{\tau \to \infty} \rho_{A\Lambda_q, I, \tau}({\mathds}{1}_{B_t}) = m_q({\mathds}{1}_{B_t})$, and $\lim_{\tau \to \infty} \rho_{A\Lambda_q, I, \tau}({\mathds}{1}_{C_t}) = m_q({\mathds}{1}_{C_t}) = 0$. Since $|{\mathds}{1}_{A_t} - {\mathds}{1}_{B_t}| \leq {\mathds}{1}_C$ on all of $\mathscr{T}^q$, we get $\lim_{\tau \to \infty} \rho_{A\Lambda_q, I, \tau}({\mathds}{1}_{A_t}) = m_q({\mathds}{1}_{A_t})$. ### Slope Gap Distribution {#subsubsection: slopegap distribution} Let $A \in {\operatorname{SL}}(2, {\mathbb{R}})$, and $\tau > 0$ be such that $\mathcal{F}(A\Lambda_q, \tau) \neq \emptyset$. Given two vectors $\mathbf{u}_0, \mathbf{u}_1 \in \mathcal{F}(A\Lambda_q, \tau)$ with consecutive slopes, we denote the difference between the slopes of $\mathbf{u}_0$ and $\mathbf{u}_1$ by ${\operatorname{slopegap}}(A\Lambda_q, \tau, \mathbf{u}_0) = {\operatorname{slope}}(\mathbf{u}_1) - {\operatorname{slope}}(\mathbf{u}_0)$. We have the following on the limiting distribution of ${\operatorname{slopegap}}$. \[corollary: limiting distribution of slopegap\] Let $A \in {\operatorname{SL}}(2, {\mathbb{R}})$, $I \subset {\mathbb{R}}$ be a finite interval. The limit of $${\operatorname{SlopeGap}}_\tau^{A\Lambda_q, I}(t) := \frac{\# \left\{\mathbf{u} \in \mathcal{F}_I(A\Lambda_q, \tau) \mid \tau^2 {\operatorname{slopegap}}(A\Lambda_q, \tau, \mathbf{u}) \geq t\right\}}{N_I(A\Lambda_q, \tau)}$$ as $\tau \to \infty$ exists for all $t \in {\mathbb{R}}$, and is equal to $m_q({\mathds}{1}_{R_q \geq t})$, where $m_q = \frac{2}{\lambda_q} da db$ is the Lebesgue probability measure on the $G_q$-Farey triangle $\mathscr{T}^q$. Let $\tau > 0$ be such that $\mathcal{F}_I(A\Lambda_q, \tau) = \{\mathbf{u}_n = (q_n, a_n)^T\}_{n=0}^{N_I(A\Lambda_q, \tau) - 1} \neq \emptyset$. For any $0 \leq n \leq N_I(A\Lambda_q, \tau) - 2$, we have by \[theorem: next term algorithm\] that $${\operatorname{slopegap}}(A\Lambda_q, \tau, \mathbf{u}_n) = \frac{a_{n+1}}{q_{n+1}} - \frac{a_n}{q_n} = \frac{1}{\tau^2} R_q({\operatorname{FTR}}_q(A\Lambda_q, \tau, \mathbf{u}_n)).$$ This implies that ${\operatorname{SlopeGap}}_\tau^{A\Lambda_q,I}(t) = {\operatorname{FareyStat}}_{R_q,\tau}^{A\Lambda_q,I}(t)$, and the proposition then follows from \[proposition: limiting distributions of Farey statistics\]. ### The $G_q$-Ford Circles, and Their Geometric Statistics {#subsubsection: centdist distribution} ![The $G_5$-Ford circles corresponding to the vectors in $\mathcal{F}_{[0, \varphi]}(\Lambda_5, \infty)$ along with the circle at infinity. The circles tangent to the line $y=0$ at $\overline{\varphi}$, $1$, and $\varphi$ are the $G_5$-Stern-Brocot children of the circles at $0$ and infinity.[]{data-label="figure: G_5 Ford circles"}](ford_circles.pdf) For any point $\mathbf{w} = (r,s)^T \in {\mathbb{R}}^2$, the *Ford circle* $C[\mathbf{w}]$ [@Ford1938-ya] corresponding to $\mathbf{w}$ is defined to be either - the circle with radius $\frac{1}{2r^2}$, and center at $\left(\frac{s}{r}, \frac{1}{2r^2}\right)$, if $r \neq 0$, or - the straight line $y = s^2$, if $r = 0$. It is well-known that for any two vectors $\mathbf{w}_1, \mathbf{w}_2 \in {\mathbb{R}}^2$, the Ford circles $C[\mathbf{w}_1]$ and $C[\mathbf{w}_2]$ intersect if $|\mathbf{w}_0 \wedge \mathbf{w}_1| < 1$, are tangent if $|\mathbf{w}_0 \wedge \mathbf{w}_1| = 1$, and are wholly external if $|\mathbf{w}_0 \wedge \mathbf{w}_1| > 1$. It follows from \[theorem: G\_q Stern Brocot process is well-defined and exhaustive\] and \[corollary: odds and ends\] that for any $A \in {\operatorname{SL}}(2, {\mathbb{R}})$, the Ford circles corresponding to any two distinct elements of $A\Lambda_q \cap S_\infty$ are either tangent or wholly external, and that the $G_q$-Stern-Brocot children of any two unimodular vectors of $A\Lambda_q \cap S_\infty$ correspond to a chain of $q - 2$ tangent circles between the two circles corresponding to the “parents”. Let $A \in SL(2, {\mathbb{R}})$, and $\tau > 0$ be such that $\mathcal{F}(A \Lambda_q, \tau) \neq \emptyset$. Given two vectors $\mathbf{u}_0, \mathbf{u}_1 \in \mathcal{F}(A \Lambda_q, \tau)$ with consecutive slopes, we denote the distance between the centers of $C[\mathbf{u}_0]$ and $C[\mathbf{u}_1]$ by ${\operatorname{centdist}}(A \Lambda_q, \tau, \mathbf{u}_0)$. We have the following on the limiting distribution of ${\operatorname{centdist}}$, extending a result from [@Athreya2015-gk] for $G_3$-Ford circles. \[corollary: limiting distribution of centdist\] Let $A \in {\operatorname{SL}}(2, {\mathbb{R}})$, and $I \subset {\mathbb{R}}$ be a finite interval. The limit of $${\operatorname{CentDist}}_\tau^{A\Lambda_q,I}(t) := \frac{\# \left\{\mathbf{u} \in \mathcal{F}_I(A\Lambda_q, \tau) \mid \tau^2 {\operatorname{centdist}}(A\Lambda_q, \tau, \mathbf{u}) \geq t\right\}}{N_I(A\Lambda_q, \tau)}$$ as $\tau \to \infty$ exists for all $t \in {\mathbb{R}}$, and is equal to $m_q({\mathds}{1}_{F_q \geq t})$, where $m_q = \frac{2}{\lambda_q}dadb$ is the Lebesgue probability measure on the $G_q$-Farey triangle $\mathscr{T}^q$, and $F_q : \mathscr{T}^q \to {\mathbb{R}}$ is the function defined by $$F_q(a, b) = \sqrt{R_q(a, b)^2 + \frac{1}{4} \left(\frac{1}{L_1^q(a, b)^2} - \frac{1}{L_0^q(a, b)^2}\right)^2},$$ where $L_0^q$ and $L_1^q$ are as in \[theorem: next term algorithm\]. As an immediate consequence of the second claim in \[corollary: odds and ends\], we get the following weak form of Dirichelet’s approximation theorem for $\Lambda_q$. \[proposition: weak Dirichelet approximation\] Let $A \in {\operatorname{SL}}(2, {\mathbb{R}})$, and $\alpha \in {\mathbb{R}}$. The line $x = \alpha$ either passes through the center of a $G_q$-Ford circle corresponding to a vector in $A \Lambda_q$, or there exist infinitely many vectors in $A \Lambda_q$ whose Ford circles intersect $x = \alpha$. In particular, $\alpha$ is either the slope of a vector in $A \Lambda_q$, or there exist infinitely many $(q, a)^T \in A \Lambda_q$ such that $$\left|\alpha - \frac{a}{q}\right| \leq \frac{1}{2q^2}.$$ \[figure: Euclidean distance distribution\] ![The graph of the limiting distribution $\lim_{t \to \infty} {\operatorname{CentDist}}_\tau^{\Lambda_5,I}(t) = m_5({\mathds}{1}_{F_5 \geq t})$ of the Euclidean distance between successive $G_5$-Ford circles from \[corollary: limiting distribution of centdist\].](G_5_Euclidean_distance_distribution.pdf "fig:") Let $\tau > 0$ be such that $\mathcal{F}_I(A\Lambda_q, \tau) = \{\mathbf{u}_n = (q_n, a_n)^T\}_{n=0}^{N_I(A\Lambda_q, \tau) - 1} \neq \emptyset$. For any $0 \leq n \leq N_I(A\Lambda_q, \tau) - 2$, we have by \[theorem: next term algorithm\] that $$\frac{a_{n+1}}{q_{n+1}} - \frac{a_n}{q_n} = \frac{1}{\tau^2} R_q({\operatorname{FTR}}_q(A\Lambda_q, \tau, \mathbf{u}_n)),$$ and $$\frac{1}{2q_{n+1}^2} - \frac{1}{2q_n^2} = \frac{1}{2 \tau^2 L_1^q({\operatorname{FTR}}_q(A\Lambda_q, \tau, \mathbf{u}_n))^2} - \frac{1}{2 \tau^2 L_0^q({\operatorname{FTR}}_q(A\Lambda_q, \tau, \mathbf{u}_n))^2}.$$ From this follows that the distance between the centers of $C[\mathbf{u}_n]$ and $C[\mathbf{u}_{n+1}]$ is given by $$\begin{aligned} {\operatorname{centdist}}(A\Lambda_q, \tau, \mathbf{u}_n) &=& \sqrt{\left(\frac{a_{n+1}}{q_{n+1}} - \frac{a_n}{q_n}\right)^2 + \left(\frac{1}{2q_{n+1}^2} - \frac{1}{2q_n^2}\right)^2} \\ &=& \frac{1}{\tau^2} F_q({\operatorname{FTR}}_q(A\Lambda_q, \tau, \mathbf{u}_n)).\end{aligned}$$ This implies that ${\operatorname{CentDist}}_\tau^{A\Lambda_q,I}(t) = {\operatorname{FareyStat}}_{F_q,\tau}^{A\Lambda_q,I}(t)$, and the proposition then follows from \[proposition: limiting distributions of Farey statistics\]. [9]{} Athreya, J. S., Chaika, J., & Lelievre, S. (2015). The gap distribution of slopes on the golden L. Recent trends in ergodic theory and dynamical systems, 631, 47–62. Athreya, J., Chaubey, S., Malik, A., & Zaharescu, A. (2015). Geometry of Farey–Ford polygons. New York Journal of Mathematics, 21, 637–656. Athreya, J. S., & Cheung, Y. (2013). A Poincaré Section for the Horocycle Flow on the Space of Lattices. International Mathematics Research Notices. Augustin, V., Boca, F. P., Cobeli, C., & Zaharescu, A. (2001). The h-spacing distribution between Farey points. Mathematical Proceedings of the Cambridge Philosophical Society, 131(1), 23–38. Boca, F. P. (2008). A Problem of Erdös, Szöz, and Turán Concerning Diophantine Approximations. International Journal of Number Theory, 04(04), 691–708. Boca, F. P., Cobeli, C., & Zaharescu, A. (2001). A conjecture of R. R. Hall on Farey points. Journal fur die Reine und Angewandte Mathematik, 2001(535). Dani, S. G., & Smillie, J. (1984). Uniform distribution of horocycle orbits for Fuchsian groups. Duke Mathematical Journal, 51(1), 185–194. Davis, D., & Lelievre, S. (2018, October 26). Periodic paths on the pentagon, double pentagon and golden L. arXiv \[math.DS\]. http://arxiv.org/abs/1810.11310 Erdös, P., Szüsz, P., & Turán, P. (1958). Remarks on the theory of diophantine approximation. Colloquium Mathematicum, 6(1), 119–126. Fisher, A. M., & Schmidt, T. A. (2014). Distribution of approximants and geodesic flows. Ergodic Theory and Dynamical Systems, 34(6), 1832–1848. Ford, L. R. (1938). Fractions. The American mathematical monthly: the official journal of the Mathematical Association of America, 45(9), 586–601. Hall, R. R., & Shiu, P. (2003). The index of a Farey sequence. Michigan Mathematical Journal, 51(1), 209–223. Hardy, G. H., Wright, E. M., & (Edward Maitland), E. (1979). An Introduction to the Theory of Numbers. Clarendon Press. Heersink, B. (2016). Poincaré sections for the horocycle flow in covers of ${\operatorname{SL}}(2,{\mathbb{R}})/{\operatorname{SL}}(2,{\mathbb{Z}})$ and applications to Farey fraction statistics. Monatshefte für Mathematik, 179(3), 389–420. Kleinbock, D. Y., & Margulis, G. A. (1996). Bounded orbits of nonquasiunipotent flows on homogeneous spaces. American Mathematical Society Translations, Series 2, 141–172. Lang, C. L., & Lang, M. L. (2016). Arithmetic and geometry of the Hecke groups. Journal of Algebra, 460, 392–417. Panti, G. (2015, March 9). The weighted Farey sequence and a sliding section for the horocycle flow. arXiv \[math.DS\]. http://arxiv.org/abs/1503.02539 Uyanik, C., & Work, G. (2016). The Distribution of Gaps for Saddle Connections on the Octagon. International Mathematics Research Notices, 2016(18), 5569–5602. Smillie, J., & Weiss, B. (2010). Characterizations of lattice surfaces. Inventiones Mathematicae, 180(3), 535–557. Xiong, M. S., & Zaharescu, A. (2006). A problem of Erdös-Szüsz-Turán on diophantine approximation. Acta Arithmetica, 125(2), 163. [^1]: In the remainder of this paper, we will denote the BCZ maps we compute for the Hecke triangle groups $G_q$ by ${\operatorname{BCZ}}_q$, reserving the symbols $T_q$ for particular generators of $G_q$. [^2]: By \[theorem: G\_q BCZ maps\], we get for $q = 5$ the indices $k_2(a, b) = \left\lfloor\frac{1-(a+\varphi b)}{\varphi^2(a + b)}\right\rfloor$, $k_3(a, b) = \left\lfloor \frac{1-b}{\varphi(a+\varphi b)} \right\rfloor$, and $k_4(a, b) = \left\lfloor \frac{1+a}{\varphi b} \right\rfloor$. This corrects the indices given in theorem 3.1 of [@Athreya2015-nq]. [^3]: This can be trivially extended into a proof that the set of wedge products of the elements of $\Lambda_q$ is discrete, similar to a characterization of lattice surfaces from [@Smillie2010-nb].

ArXiv

--- abstract: 'We investigate the interplay between magnetic and structural dynamics in ferromagnetic atomic point contacts. In particular, we look at the effect of the atomic relaxation on the energy barrier for magnetic domain wall migration and, reversely, at the effect of the magnetic state on the mechanical forces and structural relaxation. We observe changes of the barrier height due to the atomic relaxation up to 200%, suggesting a very strong coupling between the structural and the magnetic degrees of freedom. The reverse interplay is weak, i.e. the magnetic state has little effect on the structural relaxation at equilibrium or under non-equilibrium, current-carrying conditions.' author: - Maria Stamenova - Sudhakar Sahoo - 'Cristián G. Sánchez' - 'Tchavdar N. Todorov' - Stefano Sanvito title: 'Magneto-mechanical interplay in spin-polarized point contacts' --- Existing experimental techniques are capable of constructing low-dimensional magnetic constrictions consisting of a tiny number of atoms and measuring their transport properties [@chopra; @exp1D; @viret]. In this rapidly evolving field experiments and theoretical simulations are closely entangled and the latter can provide important information on the fundamental dynamics at the atomic scale. The modeling of atomic-sized ferromagnetic devices under bias requires the combined description of electron transport and of the local magnetization and structural dynamics at the atomic level [@stepanyuk]. In fact, since the spin-polarized current transfers both spin and charge, it exerts a twofold effect on the current-carrying structure, interlinking its structural and magnetic degrees of freedom. In order to investigate the mutual interplay between magnetic and structural properties, we have developed a computational scheme for evaluating spin-polarized currents and the associated current-induced forces and torques. On one hand, we are able to examine the energetics of magnetization dynamical processes as a function of the atomic rearrangements in magnetic point contacts under bias. On the other hand, we can ravel out the effect of the magnetic configuration itself on the structural relaxation. Our main finding is that this interplay is strong only in one direction. While the atomic rearrangements can modify drastically the point contact magneto-dynamics, the magnetic configuration has little effect on the atomic configuration. Our computational approach combines the non-equilibrium Green’s function (NEGF) method for evaluating the charge density and the current [@alex; @alex2] with a description of the magnetization dynamics in terms of quasistatic thermally-activated transitions between stationary configurations [@nie]. This scheme is general and is conceptually transferable [@nie] to first-principles Hamiltonians, for instance [@alex2], within density functional theory. However, it is currently implemented in an empirical single-orbital tight-binding (TB) model [@ch_noble] with parameters chosen to simulate the mechanical properties of noble metals and the basic electronic structure of a generic magnetic metal. This has the benefit of being reasonably realistic while keeping the computational overheads to a minimum. The magnetic and structural degrees of freedom both enter our model as classical variables. We associate a classical magnetic dipole moment (MM) and a set of Cartesian coordinates to each atom in the system. The magnetic state (MS) is defined by a set of angular coordinates $\Phi \equiv \{\phi_i\}$. These represent the angles of the MMs of the “live" atoms (those contributing to the dynamics) with respect to a given direction. The structural state (SS) is defined by a set of spatial coordinates $\mathcal R \equiv \{\pmb {R}_i\}$ for these atoms. The interplay between MS and SS is then investigated by either fixing $\Phi$ and evolving $\mathcal R$ or by fixing $\mathcal R$ and evolving $\Phi$. In the first case we study how the MS affects the structural relaxation, and in the second how the SS modifies the magnetic dynamics. Our mixed quantum-classical Hamiltonian for this system reads $$\begin{aligned} \label{fullH} H(\Phi,\mathcal R)& = &\sum_{i,j,\sigma} [H^{TB}_{ij}(\mathcal {R}) + V^{\sigma}_{ij}(\Phi)] \, c_{i\sigma}^{\dagger} c_{j\sigma} + \nonumber \\ & & +\, \Omega(\mathcal{R})+W(\Phi)\end{aligned}$$ where $c_{i\sigma}^{\dagger}$ and $c_{i\sigma}$ are creation and annihilation operators for electrons with spin $\sigma$ at atomic site $i$. $H(\Phi,\mathcal R)$ is separated into several quantum and classical terms. The first term reads $$\label{TBHam} H^{TB}_{ij}(\mathcal R)=\left[ \mathcal {E}_0 +V^{TB}_{i} (\mathcal R) \right] \delta_{ij} -\frac{\epsilon c}{2} \left. \left( \frac{a_f}{R_{ij}} \right)^{\!q} \right|_{j\ne i}$$ where $\mathcal{E}_0$ is the on-site energy for an isolated atom, $V^{TB}_{i}$ represents an empirical approximation to the second-order Coulomb energy at the $i$-th site $$V^{TB}_{i}(\mathcal R)=\sum_{k} f_{ik}(\mathcal R) \Delta q_k=\sum_{k} \frac{\kappa_{el} \Delta q_k}{\sqrt{R_{ik}^2 + \kappa_{el}^2/U^2}}$$ where we have defined $f_{ik}(\mathcal R)$, $\kappa_{el}=e^{2}/4\pi\varepsilon_0=14.4$ eVÅ, $\Delta q_j=\sum_{\sigma} \rho_{jj}^{\sigma}-(\rho_0)_{jj}$ is the deviation of the electron occupancy of site $j$ ($\rho_{jj}^{\sigma}$ being a diagonal term of the density matrix for spin $\sigma$) from its equilibrium non-interacting value $\rho_0$. The last term in Eq. (\[TBHam\]) is the intersite hopping integral, depending as an inverse power $q=4$ on the interatomic distance $R_{ij}=|\pmb{R}_i-\pmb{R}_j|$. Here we use $\epsilon=7.8680$ meV, $c=139.07$, $a_f=4.08$ Å, which are parameters fitted to elastic properties of bulk gold [@ch_noble]. The value of the on-site Coulomb repulsion is $U=7$ eV and the band filling is $\rho_0=0.7$ electrons per atom. This non-integer band filling effectively accounts for the *s-p-d* hybridization. The structure of our atomic point-contact system is presented in Figure \[contact\]. It consists of a linear chain of three atoms attached to two semi-infinite leads with a simple cubic structure and $3 \times 3$ atom cross-section. The lead magnetizations are polarized along the $z$-axis and they are anti-parallel (AP) to each other. The MMs of the atoms in the chain are allowed to rotate in the $(x-z)$-plane, and are described by the set of polar coordinates $\Phi=\{\phi_i\}$ ($i=1,2,3$) about the $z$-axis. In this setup an abrupt domain wall (DW) is formed in the constriction. Note that our model does not include spatial spin anisotropy, therefore the direction of the spin-quantization axis $z$ is arbitrary. In this sense a Neél and a Bloch wall are physically identical within our model. {width="1.0\columnwidth" height="0.6\columnwidth"} \[contact\] The spin-dependent term in the electronic part of Eq. (\[fullH\]) is approximated by a classical Heisenberg-type interaction as follows $$V^{\sigma}_{ij}(\Phi)=-J \pmb{\sigma}\cdot \pmb{S}_i \delta_{ij}=-\sigma_z \frac{J}{2}\cos{(\phi_i)} \delta_{ij} \,,$$ where $\pmb{\sigma}=(0,0,\frac{1}{2}\sigma_z)$ is the spin of the current-carrying electrons, collinear and polarized along the $z$ axis ($\sigma_z=\pm 1$) and $J>0$ is the exchange integral. We assume that the local MMs, associated with the local spins, are constant in magnitude [@stepanyuk], and impose $|\pmb{S}_i|$=1. The remaining terms in Eq. (\[fullH\]) involve only classical variables. We have adopted a repulsive pair potential which decays as an inverse power law [@ch_noble], [^1] $$\Omega(\mathcal R)=\frac{1}{2}\sum_{i,j\neq i} \Omega_{ij}(\mathcal R) = \frac{\epsilon}{2}\sum_{i,j\neq i} \left( \frac{a_f}{R_{ij}}\right)^{11}\, ,$$ and a Heisenberg-type nearest-neighbors spin-spin interaction between the localized spins $$W(\Phi)=-\frac{J_{dd}}{2}\sum_{i,j\neq i} \pmb{S}_i \cdot \pmb{S}_j = -\frac{J_{dd}}{2}\sum_{i,j\neq i}\cos{(\phi_i-\phi_j)} \, ,$$ where $J_{dd}>0$ is a parameter describing the strength of the intersite exchange interaction. Typical values for the exchange parameters are $J=1$ eV and $J_{dd}=50$ meV, which are of the same order of magnitude as those for bulk ferromagnetic metals within the $s-d$ model [@zener], and those derived from the Curie temperature within the Heisenberg model [@rushbrooke]. Clearly, the low coordination alters the values for the exchange parameters [@exp1D], though this effect was not explicitly included here. However, we have repeated the calculation for $J$ from 0.8 eV upto 2 eV as well as for other force parameterizations (for Cu [@ch_noble]) and found no qualitative differences in the main physical results. Forces and torques, associated with the classical degrees of freedom, are derived from the Hellmann-Feynman theorem, generalized for non-equilibrium systems. [@ths; @ventra] In our model the corresponding forces read[@ch_tdtb] $$\label{frcs} F_{i}=-\! \sum_{j\neq i} \left[ 2 (\nabla_i H^{TB}_{ij}) \mathrm{Re}\! \left[ \rho_{ij} \right] + \Delta q_i \Delta q_j \nabla_i f_{ij} + \nabla_i \Omega_{ij} \right] ,$$ where the live atoms $(i=0,\ldots,4)$ are those in the chain and their leftmost and rightmost neighbors in the leads $(i=0,4)$. Index $j$ runs over all the atoms in the self-consistent region, which apart from the chain includes two atomic planes from each lead (see Fig. \[contact\]). The torques $T$ read [@nie] $$\label{trqs} T_{i}= -\frac{J}{2} s_i \sin\phi_i-\frac{J_{dd}}{2} \left[\sin(\phi_i-\phi_{i-1}) +\sin(\phi_{i+1}-\phi_i)\right] \, ,$$ where $s_i=\rho_{ii}^{\uparrow}-\rho_{ii}^{\downarrow}$ is the on-site spin-polarization $(i=1,2,3)$, $\phi_0=0$ and $\phi_4=\pi$ are considered frozen and aligned along the same direction as in the corresponding lead. The quantum transport problem is solved by using the NEGF method [@alex; @alex2; @nie]. An analytical expression is used for the surface Green’s function of the leads and a positive external bias $V$ is introduced as a rigid shift of $V/2$ of the on-site energies of the left-hand side electron reservoir and of $-V/2$ for the right-hand side with respect to their equilibrium value. The non-integer band-filling of the spin-split *s* band in the leads generates an asymmetry in the transport properties of the spin-up and spin-down conduction channels upon left-to-right inversion of the system in its AP state. The relative spin-polarization $P^{\sigma} = (n^{\uparrow}-n^{\downarrow}) / (n^{\uparrow}+n^{\downarrow})$ of the density-of-states at the Fermi level, $n^{\sigma}(E_f)$, of the left lead in equilibrium is about -19% (-17%) for $J=1$ eV (2 eV). The net current is calculated as in Ref. [@nie] . The self-consistent density-matrix is used into Eqs. (\[frcs\]) and (\[trqs\]), and both forces and torques are determined. {width="1.0\columnwidth" height="0.9\columnwidth"} \[Pos\] The “live" atoms in the constriction are relaxed at a given finite bias for a fixed MS. For symmetry reasons the atomic relaxation always results in displacements only along the longitudinal direction ($y$-axis). Thus, we denote the relaxed atomic positions at a bias voltage of $V$ and a magnetic state $\phi$ during relaxation by $y_V (\phi)$. The initial geometry, denoted by $y_{\rm uni}$, is that of equidistant atoms with a nearest-neighbor distance of $a=2.5$ Å(see Fig. \[contact\]). This is near the equilibrium bond length of a periodic 1D chain within our model [@ch_frc]. However, such a bond length produces a compressive stress in the bulk leads, as a result of which the leftmost and rightmost atoms are pushed slightly out of the leads and the whole chain at equilibrium shrinks by about 2% \[see Fig. \[Pos\](a)\]. First we investigate how the MS of the constriction affects its structural relaxation. We relax the live atoms in $\{\phi^{(1)}\}=(0,0,\pi)$ or $\{\phi^{(2)}\}=(0,\pi,\pi)$ configuration, which represent two possible spatial positions of the DW inside the constriction, as well as for intermediate MSs with $\phi_2 \in [0,\pi]$ and $\phi_{1,3}$ such that $T_{1,3}=0$. The DW-motion induced displacements $\Delta y_{\phi} = y_V (\phi_2)-y_V(00\pi)$ are monotonic functions of $\phi_2$ \[see Fig. \[Pos\](b), where $V=1V$\]. They represent nearly rigid translations of the whole atomic chain in the direction of the electron flow. However, these displacements are very small and constitute about 3% of the displacements from the uniform structure $\Delta y_{\rm uni}=y_0 (\{\phi^{(1,2)}\})-y_{\rm uni}$ \[Fig. \[Pos\](a)\]. The former depend weakly on the bias \[Fig. \[Pos\](d)\]. The effect of the bias voltage on the atomic relaxation $\Delta y_V = y_V (\phi) - y_0(\phi)$ for $\phi=\{\phi^{(1,2)}\}$ is demonstrated in Figure \[Pos\](c). The relative current-induced displacements are smooth functions of $V$ and promote a tendency towards dimerization. $ \left| \Delta y_V \right| /a < 0.7 \% $ (upto $V=2$ V) are of the same order of in magnitude to the displacements $\Delta y_{\phi}$. Interestingly, the former are almost insensitive to the MS $\{\phi^{(1,2)}\}$, which is a result of the weak bias-dependence of $\Delta y_{\{\phi^{(2)}\}}$ \[see Fig. \[Pos\](d)\]. Other MS, having multiple abrupt DWs \[e.g. $(\pi,0,\pi)$\], are found to produce an effect of similar magnitude on the atomic relaxation (not shown in this work). This establishes that the structural properties of a magnetic nano-device under bias are to a large extent independent of the magnetic state. Next we address the converse, that is whether the structural relaxation affects the energy barrier for DW migration. This is calculated from the torque $T_2$ as the MM of atom “2" is quasistatically rotated. The other two live MMs are continually relaxed so that their associated torques are kept at zero [@nie]. The net work for this rotation is then $$\label{work} W(\phi_2)=-\int_0^{\phi_2} T_2 (\phi '_2) \,\rm{d}\phi '_2$$ Our calculated DW migration barriers (DWMB) for a few values of $J$ are shown in Figure \[barriers\]. A key feature of these profiles is the tilt of the barrier at any finite bias even for a uniform atomic arrangement. This tilt results from the fact that in the absence of the special symmetry in the density of states that characterize the present simple cubic structure in the case of a half-filled band [@nie], the leading contributions to current-induced forces and torques are linear in the bias. This asymmetry results in a preferential spatial localization of the DW at a given bias: under the present (positive) bias the $(0,0,\pi)$ configuration is stable, while the $(0,\pi,\pi)$ is at most metastable. The DW can then be driven back and forth in the constriction by an alternating current. That is an explicit realization of current-driven DW motion. ![(Color online) DW-migration energy barriers at different bias $V$ for $J_{dd}=50$ meV and $J=1, 1.5, 2$ eV (panels from left to right). The solid (dashed) lines are for structure relaxed at $(0,0,\pi)$ \[$(0,\pi,\pi)$\]. $W(V,\phi_2)$ corresponds to geometry relaxed at the given bias $V$; $\Delta W_0=W-W_0$ and $\Delta W_{\rm uni}=W-W_{\rm uni}$, where $W_0$ refers to geometry relaxed at $V=0$ and $W_{\rm {uni}}$ to uniform geometry.]({./}/figure03.eps){width="1.0\columnwidth" height="0.9\columnwidth"} \[barriers\] We have then investigated the contribution of atomic relaxation to the DW migration barrier. The bottom panels of Figure \[barriers\] show $\Delta W_0=W-W_0$ and $\Delta W_{\rm uni}=W-W_{\rm {uni}}$ as functions of $\phi_2$ at different bias voltages $V$, where $W(\phi_2)$ is the DW migration work (see Eq. (\[work\])) for atomic structure relaxed at the given bias and MS, while $W_0 (\phi_2)$ and $W_{\rm uni}(\phi_2)$ correspond to atomic structure relaxed at $V=0$ and to a uniform (unrelaxed) structure respectively. The small current-induced atomic displacements (see Fig. \[Pos\]) systematically increase the DWMB height ($\Delta W_0>0$) by a few percent. $\Delta W_0$ does not show a dependence on the MS. However, the relaxation at $V=0$ from the uniform arrangement, which shortens the interatomic distance between the live MMs by about 4% \[Fig. \[Pos\] (a)\], reduces the height of DWMB approximately by $\Delta W_{\rm uni}/W (\phi_2)=25-30\%$ (see Fig. \[barriers\]) for $J=1.5, 2$ eV. This effect can increase dramatically (to about 200%) for exchange parameters close to magnetic phase transitions. [@nie], [^2] Thus the magnetic properties are strongly affected by the geometry of the contact, especially in the region of parameters where the $J$ coupling mechanism starts competing with the direct exchange mechanism. ![(Color online) $I-V$ curves for a geometry relaxed at $V$. Solid (dashed) lines represent $(0,0,\pi)$ $[(0,\pi,\pi)]$ state. Top inset: dependence of net current for $V=1$ V on the DW-migration reaction coordinate $\phi_2$, circles represent structure, relaxed at $\phi_2$ \[see Fig. \[Pos\](b)\]. Bottom inset: $\Delta I=I-I_{0/\rm{uni}}$, where $I_{0}$ ($I_{\rm uni}$) refer to relaxed at $V=0$ (uniform) structure.]({./}/figure04.eps){width="1.0\columnwidth" height="0.9\columnwidth"} \[IVs\] Finally, the effect of the structural relaxation on the conductivity of the system is found to be small (Fig. \[IVs\]). The current is clearly insensitive to the DW migration within the constriction. The overall variation of the net current for the rotation of $\phi_2$ for fixed geometry, which is in itself a small quantity, is further substantially compensated by structural rearrangement, i.e. the structure is found to respond to the magnetic rearrangement by structural adjustment, which minimizes the variation in the conductivity (see inset of Fig. \[IVs\]). We have also found a decrease in conductivity due to relaxation of the structure from the uniform geometry. This agrees qualitatively with the findings in Ref. [@lmto], although the effect we observe is much smaller in magnitude. In conclusion we have developed a method to investigate the interplay between the magnetic and structural degrees of freedom of a ferromagnetic atomic point contact under bias. We have used it to assess the effects of the structural relaxation on the magnetic DWMB and reversely the effect of the magnetic configuration on the structural relaxation. Our main finding is that the interplay is only in one direction, that is the structural relaxation strongly modifies the DW migration barrier. In particular, we have found that the DWMB shows a substantial asymmetry, which increases with the external bias even for a spatially symmetric system. That opens the possibility of voltage-controlled DW motion in such systems. The current-induced displacements ($\Delta y_{V} /a<0.7\%$) from the relaxed at $V=0$ structure produce a small positive shift in the DWMB height ($\sim 3\%$). This is small compared to the effect of the relaxation from the initial uniform atomic configuration at the given bias. The latter is $\Delta y_{\rm uni} /a< 4\%$ and has a much more dramatic effect on the DWMB profile, reducing the barrier height by upto 2/3 or even making the alternative MS unstable, i.e. blocking the DW migration, for some exchange parameters. That is a signiture of the strong non-linear dependence of spin-polarized transport properties on structural rearrangements. However, structure is not affected by the DW migration under bias. This is so because interatomic forces depend on the total charge density of the current-carrying electrons, not on their spin polarization. This work has been sponsored by the Irish Higher Educational Authority under the North South Programme for Collaborative Research. The authors wish to acknowledge the SFI/HEA Irish Centre for High-End Computing (ICHEC) for the provision of computational facilities and support. [hbp]{} Chopra H. D., Sullivan M. R., Armstrong J. N., et al., Nature Materials **11**, 832 (2005) P. Gambardella, A. Dallmeyer, K. Maiti, M. C. Malagoli, W. Eberhardt, K. Kem, C. Carbone, Nature **416**, 301 (2002) M. Viret et al., Phys. Rev. B **66**, 220401(R) (2002) V. S. Stepanyuk, A. L. Klavsyuk, W. Hergert, A. M. Saletsky, P. Bruno, I. Mertig, Phys. Rev B **70**, 195420 (2004) A. R. Rocha, S. Sanvito, Phys. Rev. B **70**, 094406 (2004) A. R. Rocha, V.M. Garcia Suarez, S.W. Bailey, C.J. Lambert, J. Ferrer, and S. Sanvito, Nature Materials **4**, 335 (2005) M. Stamenova, S. Sanvito, T. N. Todorov, Phys. Rev. B **72**, 134407 (2005) A. P. Sutton, T. N. Todorov, M. J. Cawkwell and J. Hoekstra, Phil. Mag. A **81**, 1833 (2001) T. N. Todorov, J. Hoekstra and A. P. Sutton, Phil. Mag. B **80**, 421 (2000) M. Di Ventra, S. T. Pantelides, Phys. Rev. B **61**, 16207 (2000) T. N. Todorov, J. Phys.: Condens. Matter **13**, 10125 (2001) T. N. Todorov, J. Hoekstra and A. P. Sutton, Phys. Rev. Lett. **86**, 3606 (2001) C. Zener, Phys. Rev. **81**, 440 (1951) G. S. Rushbrooke and P. J. Wood, J. Mol. Phys. **1**, 257 (1958) A. K. Solanki, R. F. Sabriryanov, E. Y. Tsymbal, S. S. Jaswal, J. Magn. Magn. Mat. **272-276**, 1730 (2004) [^1]: The pair potential and the hopping integrals are taken to zero, between first and second nearest neighbors in the simple cubic lattice, via a smooth tail in the range 2.8-3.3 Å. [^2]: This occurs for instance when $J$ and $J_{dd}$ are such that the zero-bias magnetic energy landscape over the “reaction coordinate" ($\phi_2$) changes from having two stable MSs with colinear local MM to only one stable MS with $\phi_2=\pi/2$. For $J_{dd}=50$ meV and this parameterization this occurs for $J$ between 0.8 eV and 0.9 eV.

ArXiv

hep-ph/9710444\ UFIFT-HEP-97-22\ CRETE-97-13 **The Quantum Gravitationally Induced Stress Tensor\ During Inflation** N. C. Tsamis$^{\dagger}$ *Department of Physics\ University of Crete\ Heraklion, CR-71003 GREECE* R. P. Woodard$^*$ *Department of Physics\ University of Florida\ Gainesville, FL 32611 USA* ABSTRACT We derive non-perturbative relations between the expectation value of the invariant element in a homogeneous and isotropic state and the quantum gravitationally induced pressure and energy density. By exploiting previously obtained bounds for the maximum possible growth of perturbative corrections to a locally de Sitter background we show that the two loop result dominates all higher orders. We also show that the quantum gravitational slowing of inflation becomes non-perturbatively strong earlier than previously expected. PACS numbers: 04.60.-m, 98.80.Cq $^{\dagger}$ e-mail: tsamis@physics.uch.gr\ $^*$ e-mail: woodard@phys.ufl.edu Introduction ============ Gauge-fixed perturbation theory is by far the simplest method for computing quantum corrections to a classical geometry. Even when the state of interest is not stationary this can be done using Schwinger’s formalism for expectation values [@schw; @rj]. The procedure is first to compute the expectation value of the invariant element in the presence of the desired state: $$\left\langle \psi \left\vert g_{\mu\nu}(t,{\vec x}) dx^{\mu} dx^{\nu} \right\vert \psi \right\rangle = {\widehat g}_{\mu\nu}(t,{\vec x}) dx^{\mu} dx^{\nu} \; .$$ One then forms ${\widehat g}_{\mu\nu}$ into gauge invariant and gauge independent observables to infer how quantum effects distort the geometry. Geometrically significant [*differences*]{} between the classical and quantum backgrounds can be ascribed to a quantum-induced stress tensor. In pure gravity this is defined from the deficit by which ${\widehat g}_{\mu\nu}$ fails to obey the classical Einstein equation: $$8 \pi G {\widehat T}_{\mu\nu} \equiv {\widehat R}_{\mu\nu} - \frac12 {\widehat g}_{\mu\nu} {\widehat R} + {\widehat g}_{\mu\nu} \Lambda \; . \label{eq:stress}$$ Here ${\widehat R}_{\mu\nu}$ and ${\widehat R}$ are the Ricci tensor and Ricci scalar constructed from ${\widehat g}_{\mu\nu}$ and it should be noted that we have included a cosmological constant $\Lambda$ in Einstein’s equation. Note also that the relation between the induced stress tensor and the quantum background ${\widehat g}_{\mu\nu}$ is, in principle, non-perturbative, even though the only practical way of computing ${\widehat g}_{\mu\nu}$ is perturbatively. The purpose of this paper is to derive the leading late time dependence, [*to all orders*]{}, for the induced stress tensor appropriate to a recent calculation of the quantum gravitational back-reaction on an initially empty and inflating universe [@tw1]. That we can obtain an all-orders result arises from the conjunction of the non-perturbative relation (\[eq:stress\]) and explicit bounds on the maximum late time growth of perturbative corrections to a rather technical variant of the amputated 1-point function. Section 2 reviews the definition of this quantity and the procedure through which it is used to compute ${\widehat T}_{\mu\nu}$. Section 3 shows how the perturbative bounds on the former imply an all-orders result for the latter. We discuss the consequences of this result in Section 4. In what remains of this Introduction we review the theoretical context of our previous work and its physical motivation. Because the late time behavior is dominated by ultraviolet finite, non-local terms, we were able to use the Lagrangian of general relativity with a positive cosmological constant: $${\cal L} = {1 \over 16 \pi G} \left(R - 2 \Lambda\right) \sqrt{-g} + {\rm counterterms} \; ,$$ absorbing ultraviolet divergences with local counterterms as required. We worked on the manifold $T^3 \times R$ in the presence of a homogeneous and isotropic state for which the invariant element takes the following form in co-moving coordinates: $${\widehat g}_{\mu\nu}(t,{\vec x}) dx^{\mu} x^{\nu} = - dt^2 + \exp[2 b(t)] d{\vec x} \cdot d{\vec x} \; . \label{eq:element}$$ Our state is free de Sitter vacuum at $t=0$ in these coordinates, corresponding to the following classical background: $$b_{\rm class}(t) = H t \qquad , \qquad H^2 \equiv \frac13 \Lambda \; . \label{eq:class}$$ The physical motivation for our work is the possibility that the cosmological constant only appears to be unnaturally small today because it is screened by an infrared process in quantum gravity. This process is the buildup of gravitational interaction energy between virtual gravitons that are pulled apart by the inflationary expansion of the classical background (\[eq:element\]-\[eq:class\]). The effect acts to slow inflation because gravity is attractive. It requires a enormous time to become significant because gravity is a weak interaction, even for inflation on the GUT scale.[^1] However, inflation must eventually be ended because the effect adds coherently for as long as exponential expansion persists. The effect is also unique to gravitons. Only massless particles can give a coherent effect, and the other phenomenologically viable quanta of zero mass are prevented from doing so by conformal invariance. Our mechanism offers a natural explanation for how inflation can have lasted a long time, without fine tuning and without the need for fundamental scalars. Indeed, it results in such a long period of inflation that all energetically favorable phase transitions may have time to occur during this period, even if some are subsequently reversed by re-heating. If so, the cosmological constant which is finally screened would be that of the true vacuum, and the evolution after inflation would be almost that which is usually obtained by keeping gravity classical and fine tuning this parameter to zero. Although perturbation theory must break down at the end of inflation, one can use the technique to partially verify our proposal. For example, the presence of infrared divergences in in-out matrix elements [@tw2] and scattering amplitudes [@tw3] invalidates the null hypothesis that inflation persists to asymptotically late times with only perturbatively small corrections. One can also use Schwinger’s formalism to follow the evolution of the background until quantum corrections become non-perturbatively large [@tw1]. It was previously believed that this occurred at the same time for all orders. The burden of this paper is to show that in fact the two loop effect becomes strong at a time when all higher orders are still insignificant. Of course one cannot extend past the breakdown of perturbation theory by using perturbation theory, but we now have precise information about how the breakdown occurs. Perturbation Theory Revisited ============================= The purpose of this section is to explain the connection between the induced stress tensor of co-moving coordinates and the quantities we actually computed. We begin with the coordinate system of the classical background. For a variety of reasons, it is simplest to formulate perturbation theory in conformal coordinates: $$-dt^2 + \exp[2 H t] \; d{\vec x} \cdot d{\vec x} = \Omega^2 \left(-du^2 + d{\vec x} \cdot d{\vec x}\right) \; ,$$ $$\Omega \equiv {1 \over H u} = \exp(H t) \; .$$ Note the temporal inversion and the fact that the onset of inflation at $t=0$ corresponds to $u = H^{-1}$. The infinite future is $u \rightarrow 0^+$. Perturbation theory is organized most conveniently in terms of a “pseudo-graviton” field, $\psi_{\mu \nu}$, obtained by conformally re-scaling the metric: $$g_{\mu \nu} \equiv \Omega^2 {\widetilde g}_{\mu \nu} \equiv \Omega^2 \left(\eta_{\mu \nu} + \kappa \psi_{\mu \nu}\right) \; .$$ Our notation is that pseudo-graviton indices are raised and lowered with the Lorentz metric, and that the loop counting parameter is $\kappa^2 \equiv 16 \pi G$. After some judicious partial integrations the invariant part of the bare Lagrangian takes the following form [@tw4]: $$\begin{aligned} {\cal L}_{\rm inv} & = & \sqrt{-{\widetilde g}} {\widetilde g}^{\alpha \beta} {\widetilde g}^{\rho \sigma} {\widetilde g}^{\mu \nu} \left(\frac12 \psi_{\alpha \rho , \mu} \psi_{\nu \sigma , \beta} - \frac12 \psi_{\alpha \beta , \rho} \psi_{\sigma \mu , \nu} + \frac14 \psi_{\alpha \beta , \rho} \psi_{\mu \nu , \sigma} \right. \nonumber \\ & & \mbox{} \left. - \frac14 \psi_{\alpha \rho , \mu} \psi_{\beta \sigma , \nu}\right) \Omega^2 -\frac12 \sqrt{-{\widetilde g}} {\widetilde g}^{ \rho \sigma} {\widetilde g}^{\mu \nu} \psi_{\rho \sigma , \mu} \psi_{\nu}^{~\alpha} (\Omega^{2})_{,\alpha} \; . \label{eq:Lagrangian}\end{aligned}$$ Since $\Omega \sim u^{-1}$, it might seem as if the final term is stronger at late times than the others. In reality it is only comparable because its undifferentiated pseudo-graviton field must always contain a “$0$” index — $\psi_{\nu}^{~\alpha} (\Omega^2)_{,\alpha} = 2 u^{-1} \psi_{\nu 0} \Omega^2$ — and “$0$” components of the pseudo-graviton propagator are suppressed by a factor of $u$ [@tw3]. Gauge fixing is accomplished through the addition of $-\frac12 \eta^{\mu \nu} F_{\mu} F_{\nu}$ where: $$F_{\mu} \equiv \left(\psi^{\rho}_{~\mu , \rho} - \frac12 \psi^{\rho}_{~\rho , \mu} + 2 \psi^{\rho}_{~\mu} \; {(\ln \Omega)}_{,\rho}\right) \Omega \; .$$ The resulting gauge fixed kinetic operator has the form: $$\begin{aligned} D_{\mu \nu}^{~~ \rho \sigma} & \equiv & \left(\frac12 {\overline \delta}_{ \mu}^{~(\rho} \; {\overline \delta}_{\nu}^{~\sigma)} - \frac14 \eta_{\mu \nu} \; \eta^{\rho\sigma} - \frac12 \delta_{\mu}^{~0} \; \delta_{\nu}^{~0} \; \delta_0^{~\rho} \; \delta_0^{~\sigma} \right){\rm D}_A \nonumber \\ & & \mbox{} + \delta_{(\mu}^{~~0} \; {\overline \delta}_ {\nu)}^{~~(\rho} \; \delta_0^{~\sigma)} \; {\rm D}_B + \delta_{\mu}^{~0} \; \delta_{\nu}^{~0} \; \delta_0^{~\rho} \; \delta_0^{~\sigma} \; {\rm D}_C \; . \label{eq:kinetic}\end{aligned}$$ A variety of conventions in this relation deserve comment. First, indices enclosed in a parenthesis are symmetrized. Second, the presence of a bar over a Kronecker delta or a Lorentz metric indicates that the temporal components of these tensors are deleted: $${\overline \delta}^{\mu}_{~\nu} \equiv \delta^{\mu}_{~\nu} - \delta^{\mu}_{~0} \delta^0_{~\nu} \qquad , \qquad {\overline \eta}_{\mu\nu} \equiv \eta_{\mu\nu} + \delta^0_{~\mu} \delta^0_{~\nu} \; .$$ The symbol $D_A$ stands for the kinetic operator of a massless, minimally coupled scalar: $${\rm D}_A \equiv \Omega \left(\partial^2 + \frac2{u^2}\right) \Omega \; ,$$ while $D_B = D_C$ denote the kinetic operator of a conformally coupled scalar: $${\rm D}_B = {\rm D}_C \equiv \Omega \> \partial^2 \Omega \; .$$ What we actually computed was the amputated expectation value of $\kappa \psi_ {\mu \nu} (u,{\vec x})$ which, on general grounds, must have the following form: $$D_{\mu \nu}^{~~\rho \sigma} \; \left\langle 0 \left\vert \; \kappa \psi_{\rho \sigma}(x) \; \right\vert 0 \right\rangle = a(u) \; {\overline \eta}_{\mu \nu} + c(u) \; \delta^0_{~\mu} \delta^0_{~\nu} \; . \label{eq:amputated}$$ Attaching the external leg gives the invariant element, but in a perturbatively corrected version of conformal coordinates: $${\widehat g}_{\mu \nu}(t,{\vec x}) \; dx^{\mu} dx^{\nu} = - \Omega^2 \left[1 - C(u)\right] \; du^2 + \Omega^2 \left[1 + A(u)\right] \; d{\vec x} \cdot d{\vec x} \; . \label{eq:conformal}$$ The external leg of the 1-point function is a retarded Green’s function in Schwinger’s formalism. From the gauge fixed kinetic operator (\[eq:kinetic\]) we see that the coefficient functions $A(u)$ and $C(u)$ have the following expressions in terms of the scalar retarded propagators acting on $a(u)$ and $c(u)$ [@tw5]: $$\begin{aligned} A(u) & = & -4 G^{\rm ret}_A[a](u) + G^{\rm ret}_C[3a + c](u) \; ,\label{eq:A}\\ C(u) & = & G^{\rm ret}_C[3a + c](u) \; . \label{eq:C}\end{aligned}$$ It is simple to work out what the retarded propagators of $D_A$ and $D_C$ give when acting on any power of the conformal time: $$\begin{aligned} \lefteqn{G^{\rm ret}_A[u^{-4} (Hu)^{\varepsilon}] = {H^2 \over \varepsilon (3 - \varepsilon)} \left\{ (Hu)^{\varepsilon} - 1 + {1 \over 3} \varepsilon - {1 \over 3} \varepsilon (H u)^3\right\} \; , } \label{eq:GA} \\ \lefteqn{G^{\rm ret}_C[u^{-4} (Hu)^{\varepsilon}] = } \nonumber \\ & & {H^2 \over (1 - \varepsilon) (2 - \epsilon)} \left\{- (Hu)^{\varepsilon} + (2 - \varepsilon) H u - (1 - \varepsilon) (H u)^2\right\} \; . \label{eq:GC}\end{aligned}$$ One-particle-irreducible diagrams containing $\ell$ loops can be shown to contribute to $a(u)$ and $c(u)$ at late times no more strongly than some number times [@tw1]: $$\kappa^{2\ell} H^{2\ell-2} u^{-4} \ln^{\ell}(Hu) \; .$$ The action of the scalar retarded propagators on such a term is obtained by differentiating (\[eq:GA\]-\[eq:GC\]) $\ell$ times with respect to $\varepsilon$ and then taking the limit $\varepsilon \rightarrow 0$. Because of the factor of $\varepsilon^{-1}$ on the right hand side of (\[eq:GA\]), $G^{\rm ret}_A$ acquires an extra logarithm whereas $G^{\rm ret}_C$ does not. For example, the leading contributions at two loops give: $$\begin{aligned} \lefteqn{G^{\rm ret}_A\left[\kappa^4 H^2 u^{-4} \ln^2(Hu)\right] =} \nonumber \\ & & (\kappa H)^4 \left\{{1 \over 9} \ln^3(Hu) + {1 \over 9} \ln^2(H u) + {2 \over 27} \ln(H u) + {2 \over 81} - {2 \over 81} (H u)^3\right\} \; ,\\ \lefteqn{G^{\rm ret}_C\left[\kappa^4 H^2 u^{-4} \ln^2(Hu)\right] =} \nonumber \\ & & (\kappa H)^4 \left\{-{1 \over 2} \ln^2(H u) - {3 \over 2} \ln(H u) - {7 \over 4} + 2 H u - {1 \over 4} (H u)^2\right\} \; .\end{aligned}$$ This phenomenon has great significance. Its physical origin is the fact that $A$-type Green’s functions receive contributions from throughout the timelike region inside the past lightcone while the $C$-type Green’s functions have support only on the lightlike surface of the past lightcone [@tw4]. Comparison between (\[eq:element\]) and (\[eq:conformal\]) results in the following formulae for the conversion to co-moving time: $$\begin{aligned} d(Ht) & = & - \sqrt{1 - C(u)} \;\; d\left[\ln(Hu)\right] \; , \label{eq:con1} \\ b(t) & = & -\ln(Hu) + \frac12 \ln\left[1 + A(u)\right] \; . \label{eq:con2}\end{aligned}$$ It is then straightforward to work out the relation between physically interesting quantities defined in co-moving coordinates and the things one actually computes in perturbation theory. For example, the effective Hubble constant is:[^2] $$H_{\rm eff}(t) \equiv {db(t) \over dt} = {H \over \sqrt{1 - C(u)}} \; \left\{ 1 - \frac12 u {d \over du} \ln\left[1 + A(u)\right]\right\} \; . \label{eq:heff}$$ Two particularly interesting quantities come from the induced stress tensor: the energy density $T_{00} = \rho(t)$ and the pressure $T_{ij} = p(t) g_{ij}$. The task of this section is completed by first using (\[eq:stress\]) to express these in terms of $b(t)$ and then converting to the coefficient functions $A(u)$ and $C(u)$: $$\begin{aligned} \rho(t) & = & {1 \over 8 \pi G} \left(3 \dot{b}^2(t) - 3 H^2\right) \; , \nonumber \\ \label{eq:rho} & = & {1 \over 8 \pi G} {3 H^2 \over 1 - C} \left\{C - {u A' \over 1 + A} + {1 \over 4} \left({u A' \over 1 + A}\right)^2\right\} \; , \\ p(t) & = & - {2 \ddot{b}(t) \over 8 \pi G} - \rho(t) \; , \nonumber \\ \label{eq:pressure} & = & {1 \over 8 \pi G} {H^2 \over 1 - C} \left\{ {u C' \over 1 - C} \left[1 - {1 \over 2} {u A' \over 1 + A}\right] - {u \left(u A'\right)' \over 1 + A} + \left({u A' \over 1 + A}\right)^2\right\} \nonumber \\ & & \mbox{} - \rho(t) \; .\end{aligned}$$ A dot in these formulae indicates differentiation with respect to $t$, while a prime denotes differentiation with respect to $u$. Two Loop Dominance ================== Our perturbative work [@tw1] produced explicit results for the late time ($u \rightarrow 0^+$) behavior of the coefficient functions $a(u)$ and $c(u)$ at two loops: $$\begin{aligned} a(u) & = & H^{-2} \left({\kappa H \over 4 \pi u}\right)^4 \left\{ -43 \ln^2(Hu) + {\rm (subleading)}\right\} + O(\kappa^6) \label{eq:a} \\ c(u) & = & H^{-2} \left({\kappa H \over 4 \pi u}\right)^4 \left\{ 15 \ln^2(Hu) + {\rm (subleading)}\right\} + O(\kappa^6) \; . \label{eq:c}\end{aligned}$$ We also obtained the following limit on the maximum possible late time correction to $a(u)$ and $c(u)$ from one-particle-irreducible (1PI) graphs containing $\ell$ loops: $$\kappa^{2\ell} H^{2\ell-2} u^{-4} \ln^{\ell}(Hu) \; .$$ This bound seemed to suggest that all orders become strong at the same time: $$-\ln(H u) \sim {1 \over \kappa^2 H^2} = {3 \over 8 \pi} {1 \over G \Lambda} \gg 1 \; .$$ That conclusion is valid for the non-invariant quantities $a(u)$ and $c(u)$, but not for invariants such as the effective Hubble constant, the induced energy density and the induced pressure. The purpose of this section is to show that, for these quantities, two loop effects become strong at a time when higher loop corrections are still insignificant. We will also use the two loop results to derive explicit formulae for the physical invariants which are valid until perturbation theory breaks down. The key is the extra logarithm which the spatial trace coefficient $A(u)$ acquires from the $A$-type Green’s function. The 1PI amputated coefficient functions have the following form: $$\begin{aligned} a_{1PI}(u) & = & \sum_{\ell=2}^{\infty} a_{\ell} \kappa^{2\ell} H^{2\ell-2} u^{-4} \ln^{\ell}(Hu) + {\rm subdominant} \; , \\ c_{1PI}(u) & = & \sum_{\ell=2}^{\infty} c_{\ell} \kappa^{2\ell} H^{2\ell-2} u^{-4} \ln^{\ell}(Hu) + {\rm subdominant} \; ,\end{aligned}$$ where $a_{\ell}$ and $c_{\ell}$ are pure numbers. The non-amputated coefficient functions are defined by acting retarded Green’s functions according to relations (\[eq:A\]-\[eq:C\]). From the general action of the retarded propagators (\[eq:GA\]-\[eq:GC\]), we obtain expansions for the leading terms induced by 1PI graphs:[^3] $$\begin{aligned} A_{1PI}(u) & = & - {4 \over 3} \ln(H u) \sum_{\ell=2}^{\infty} {a_{\ell} \over \ell + 1} \left( \kappa^2 H^2 \ln(H u)\right)^{\ell} + {\rm subdominant} \, \\ C_{1PI}(u) & = & - {1 \over 2} \sum_{\ell=2}^{\infty} (3 a_{\ell} + c_{\ell}) \left( \kappa^2 H^2 \ln(H u)\right)^{\ell} + {\rm subdominant} \; .\end{aligned}$$ From expression (\[eq:con2\]) for $b(t)$ we see that inflation stops when $A(u)$ approaches $-1$. The $\ell=2$ term in $A(u)$ passes through $-1$ when: $$-\ln(H u) = \left({-9 \over 4 a_2}\right)^{\frac13} \left({1 \over \kappa H} \right)^{\frac43} \; .$$ At this time the higher $\ell$ effects in $A(u)$ are of strength: $$\ln^{\ell+1}(H u) (\kappa H)^{2\ell} \sim (\kappa H)^{\frac23 \ell - \frac43} \; .$$ This is insignificant when one recalls that $\kappa H \sim 10^{-5}$, even for GUT scale inflation. And [*all*]{} the terms in $C(u)$ are insignificant because they have one fewer power of the large logarithm. We have still to account for tadpoles coming from the shift of the background. One does this by shifting the fields of the interaction Lagrangian (\[eq:Lagrangian\]) and studying the effect of the induced interactions. For example, most 3-point vertices have the generic form: $\psi \partial \psi \partial \psi$. Suppose that the two differentiated fields are taken by the lowest order $A(u)$ terms. This gives a 1-point interaction whose coefficient is: $$\kappa \Omega^2 {d \over du} \left(\kappa^3 H^4 \ln^3(Hu)\right) {d \over du} \left(\kappa^3 H^4 \ln^3(Hu)\right) \sim \kappa^7 H^6 u^{-4} \ln^4(Hu) \; .$$ When one accounts for the extra factor of $\kappa$ in our definition (\[eq:amputated\]) the result is no stronger than the 1PI terms already allowed for at $\ell=4$ loops. In fact one can do considerably better at higher order, but never good enough to catch up with the two loop effect. The fastest possible growth for either $A(u)$ or $C(u)$ is: $$(\kappa H)^{4N+8} \ln^{3N+5}(Hu) \; ,$$ starting at $N=0$. (The order $(\kappa H)^4$ and $(\kappa H)^6$ terms are purely 1PI.) When the two loop term becomes of order one these contributions are still suppressed by a factor of the small number $(\kappa H)^{4/3} \ltwid 10^{-7}$. It remains to obtain the promised all-orders results for the dominant late time behavior of $H_{\rm eff}(t)$, $\rho(t)$ and $p(t)$. To simplify the formulae we define the following small parameter: $$\epsilon \equiv \left({\kappa H \over 4 \pi}\right)^2 = {G \Lambda \over 3 \pi} = {8 \over 3} \left({M \over M_P}\right)^4 \; .$$ Our explicit two loops results (\[eq:a\]-\[eq:c\]) imply: $$\begin{aligned} A(u) & = & \epsilon^2 \left\{ {172 \over 9} \ln^3(Hu) + {\rm (subleading)}\right\} + O(\epsilon^3) \; , \\ C(u) & = & \epsilon^2 \left\{ 57 \ln^2(Hu) + {\rm (subleading)}\right\} + O(\epsilon^3) \; .\end{aligned}$$ From $C(u)$ and relation (\[eq:con1\]) we infer the transformation to co-moving time: $$Ht = - \left\{1 - {19 \over 2} \epsilon^2 \ln^2(H u) + \dots \right\} \ln(H u) \; .$$ This can be inverted to give: $$\ln(H u) = - \left\{1 + {19 \over 2} (\epsilon H t)^2 + \dots \right\} H t \; .$$ It follows that we may set $\ln(H u)$ to $- H t$, to a very good approximation, for as long as perturbation theory remains valid. We can now write $A(u)$ as a function of the co-moving time: $$A(u) = - {172 \over 9} \epsilon^2 (H t)^3 + \dots \; .$$ The higher corrections are again insignificant when the first term becomes of order unity. We can also obtain $b(t)$ as an explicit function of time: $$b(t) \approx H t + {1 \over 2} \ln(1 + A) \; .$$ Substituting into (\[eq:heff\]) gives the effective Hubble constant: $$\begin{aligned} H_{\rm eff}(t) & \approx & H + {1 \over 2} {d \over dt} \ln(1 + A) \; , \\ & \approx & H \left\{1 - {\frac{86}3 \epsilon^2 (H t)^2 \over 1 - \frac{172}9 \epsilon^2 (H t)^3}\right\} \; . \label{eq:Hexplicit}\end{aligned}$$ Note that the numerator of the correction term is still quite small when the denominator blows up. This is why we are justified in neglecting other terms — from $C(u)$ — which are also of order $(\epsilon H t)^2$. Going through the same exercise for the induced energy density gives: $$\begin{aligned} \rho(t) & \approx & {\Lambda \over 8 \pi G} \left\{- {1 \over H} {\dot{A} \over 1 + A} + {1 \over 4 H^2} \left({\dot{A} \over 1 + A}\right)^2\right\} \; , \\ & \approx & {\Lambda \over 8 \pi G} \left\{- {\frac{172}3 \epsilon^2 (H t)^2 \over 1 - \frac{172}9 \epsilon^2 (H t)^3} + \left({\frac{86}3 \epsilon^2 (H t)^2 \over 1 - \frac{172}9 \epsilon^2 (H t)^3}\right)^2 \right\} \; .\end{aligned}$$ Note that we cannot neglect the single denominator term compared to the double one; in fact the former dominates the latter. The most useful form in which to give the pressure is added to the energy density: $$\begin{aligned} \rho(t) + p(t) & \approx & - {1 \over 8 \pi G} {d^2 \over dt^2} \ln(1 + A) \; , \\ & \approx & {1 \over 8 \pi G} \left({ \dot{A} \over 1 + A} \right)^2 \; , \\ & \approx & {H^2 \over 8 \pi G} \left({\frac{172}3 \epsilon^2 (H t)^2 \over 1 - \frac{172}9 \epsilon^2 (H t)^3}\right)^2 \; .\end{aligned}$$ Note that in passing to the middle expression we have neglected the term, $\ddot{A}/(1 + A)$, which is still insignificant when $\dot{A}/(1+A)$ is of order one. Note also that when $\dot{A}/(1 + A)$ is small, its square is even smaller. Therefore $\rho + p$ is quite near zero until screening becomes significant. Discussion ========== Our previous work [@tw1; @tw3; @tw4] has established that quantum gravitational corrections slow the expansion of an initially inflating universe by an amount that becomes non-pertutbatively large at late times. In this paper we have exploited exact relations between the objects which are actually computed in perturbative quantum gravity and the invariant quantities of physical interest. We conclude that the mechanism by which perturbation theory breaks down is the approach to $-1$ of the spatial trace coefficient $A(u)$. Furthermore, this approach is effected by two loop corrections at a time well before the higher loop corrections have become significant. This insight has a number of consequences, starting with a revised estimate for the number of inflationary e-foldings: $$N \sim \left({9 \over 172}\right)^{\frac13} \left({3 \pi \over G \Lambda} \right)^{\frac23} = \left({81 \over 11008}\right)^{\frac13} \left({M_P \over M}\right)^{\frac83} \; ,$$ where $M$ is the mass scale of inflation and $M_P$ is the Planck mass. For inflation on the GUT scale this gives $N \sim 10^7$ e-foldings. Electroweak inflation should last about $N \sim 10^{45}$ e-foldings. These numbers are smaller than our previous estimates, but still much longer than in typical models. We stress that this long period of inflation is a natural consequence of the fact that gravity is a weak interaction. One can also estimate the rapidity with which inflation ends once the effect becomes noticeable. Suppose we expand around the critical time: $$H t = N - H {\Delta t} \; .$$ When $H {\Delta t} \ll N$ our expression (\[eq:Hexplicit\]) for the effective Hubble constant becomes: $$H_{\rm eff}(t) \approx H \left\{1 - {1 \over 2 H {\Delta t}}\right\} \; .$$ It follows that inflation must end rapidly. To be precise, let us define the end of inflation as the period from when the effective Hubble constant falls from $\frac9{10}$ to $\frac1{10}$ of its initial value. From the previous formula, $H_{\rm eff}$ reaches $\frac9{10} H$ at $H {\Delta t} = 5$, and it falls to $\frac1{10} H$ at $H {\Delta t} = \frac59$, making for a transition time of $4 \frac49$ e-foldings. Of course one cannot trust perturbation theory during this period but it is reasonable to conclude that the end of inflation is likely to be sufficiently violent to give a substantial amount of re-heating. The end of inflation is also likely to be sudden enough to justify assuming that the observationally relevant density perturbations crossed the causal horizon during the period when our perturbative expressions are still valid. This means that we do not need to solve the non-perturbative problem in order to make predictions. ACKNOWLEDGEMENTS We thank the University of Crete for its hospitality during the execution of this project. This work was partially supported by DOE contract DE-FG02-97ER41029, by NSF grant 94092715, by EEC grant 961206, by NATO Collaborative Research Grant 971166 and by the Institute for Fundamental Theory. [99]{} J. Schwinger, [*J. Math. Phys.*]{} [**2**]{} 407; [*Particles, Sources and Fields*]{} (Addison-Wesley, Reading, MA, 1970). R. D. Jordan, [*Phys. Rev.*]{} [**D33**]{} (1986) 444. N. C. Tsamis and R. P. Woodard, [*Ann. Phys.*]{} [**253**]{} (1997) 1. N. C. Tsamis and R. P. Woodard, [*Phys. Lett.*]{} [**B301**]{} (1993) 351. N. C. Tsamis and R. P. Woodard, [*Class. Quant. Grav.*]{} [**11**]{} (1994) 2969. N. C. Tsamis and R. P. Woodard, [*Commun. Math. Phys.*]{} [**162**]{} (1994) 217. N. C. Tsamis and R. P. Woodard, [*Ann. Phys.*]{} [**238**]{} (1995) 1. [^1]: One traditionally defines the “scale of inflation” $M$ so that $M^4$ equals the energy density of the cosmological constant, $\Lambda/(8\pi G)$. Since the Planck mass is $M_P = G^{-1/2}$, the dimensionless coupling constant that characterizes quantum gravitational effects on inflation can be expressed as: $$G \Lambda = 8 \pi \left({M \over M_P}\right)^4 \; .$$ For GUT scale inflation this works out to about $G \Lambda \sim 10^{-11}$. The comparable figure for inflation on the electroweak scale would be about $G \Lambda \sim 10^{-67}$. [^2]: Note that the effective Hubble constant is an invariant by virtue of its relation to the Einstein tensor, $G_{00} = 3 \dot{b}^2$, and by the fact that co-moving coordinates are unique up to constant rescalings of space. It can also be shown to be gauge independent [@tw1]. [^3]: These terms are 1PI except for the external propagator.

ArXiv

--- abstract: 'Using a general theorem on conservation laws for arbitrary differential equations proved by Ibragimov, we have derived conservation laws for Dirac’s symmetrized Maxwell-Lorentz equations under the assumption that both the electric and magnetic charges obey linear conductivity laws (dual Ohm’s law). We find that this linear system allows for conservation laws which are non-local in time.' author: - 'Nail H. Ibragimov' - Raisa Khamitova - Bo Thidé bibliography: - 'nail.bib' title: 'Conservation laws for the Maxwell-Dirac equations with a dual Ohm’s law' --- Introduction ============ In all areas of physics, conservation laws are essential since they allow us to draw conclusions of a physical system under study in an efficient way. Electrodynamics, in terms of the standard Maxwell electromagnetic equations for fields in vacuum, exhibit a rich set of symmetries to which conserved quantities are associated. Recently, there has been a renewed interest in the utilisation of such quantities. Here we use a theorem of @Ibragimov:JMAA:2006b to derive conservation laws for Dirac’s symmetric version of the Maxwell-Lorentz microscopic equations, allowing for magnetic charges and magnetic currents, where the latter, just as electric currents, are assumed to be described by a linear relationship between the field and the current, i.e., an Ohm’s law. The method of @Ibragimov:JMAA:2006b produces two new adjoint vector fields which fulfil Maxwell-like equations. In particular, we obtain conservation laws for the symmetrized electromagnetic field which are non-local in time. Preliminaries ============= Notation -------- We will use the following notation (see, e.g., Ref.). Let $x=(x^1,\ldots,x^n)$ be independent variables and $u=(u^1,\ldots,u^m)$ be dependent variables. The set of the first-order partial derivatives $u^{\alpha}_i=\partial u^\alpha/\partial x^i$ will be denoted by $u_{(1)}=\{u^{\alpha}_i\}$, where $\alpha=1,\ldots.,m$ and $i,j,\ldots=1,\ldots,n$. The symbol $D_i$ denotes the total differentiation with respect to the variable $x^i$: $$D_i = \frac{\partial}{\partial x^i} + u^\alpha_{i}\frac{\partial}{\partial u^\alpha} + u^\alpha_{ij}\frac{\partial}{\partial u^\alpha_j} + u^\alpha_{ijk}\frac{\partial}{\partial u^\alpha_{jk}} + \cdots$$ We employ the usual convention of summation in repeated indices. Recall that a necessary condition for extrema of a variational integral $$\label{F1} \int_V\mathcal{L}(x,u,u_{(1)})\,{\mathrm{d}x}$$ with a Lagrangian $\mathcal{L}(x,u,u_{(1)})$, depending on first-order derivatives, is given by the Euler-Lagrange equations $$\frac{\delta\mathcal{L}}{\delta u^\alpha} \equiv \frac{\partial{\mathcal{L}}}{\partial u^\alpha} - D_i\bigg(\frac{\partial{\mathcal{L}}}{\partial u^\alpha_i}\bigg) = 0, \quad \alpha = 1,\ldots, m. \label{F2}$$ We will understand by a *symmetry* of a certain system of differential equations a generator $$\label{F3} X=\xi^i(x,u)\frac{\partial}{\partial x^i} + \eta^\alpha (x,u) \frac{\partial}{\partial u^\alpha}$$ of a continuous transformation group admitted by differential equations under consideration. A vector field $C=(C^1,\ldots,C^n)$ is said to be a *conserved vector* for the differential equations (\[F2\]) if the equation $$\label{C1} D_i\big(C^i\big) = 0$$ holds for any solution of Eq. (\[F2\]). If one of the independent variables is time, e.g., $x^n=t$, then the conservation law is often written in the form $${\frac{{\mathrm{d}}{E}}{{\mathrm{d}t}}} = 0,$$ where $$\label{F9} E = \int\limits_{{\ensuremath{\mathbb{R}}\xspace}^{n-1}} C^n (x, u(x), u_{(1)}(x))\,{\mathrm{d}x}^1\cdots{\mathrm{d}x}^{n-1}.$$ Accordingly, $C^n$ is termed the *density* of the conservation law. Basic conservation theorem -------------------------- We will employ the recent general theorem [@Ibragimov:JMAA:2006b] on a connection between symmetries and conservation laws for arbitrary systems of $s$th-order partial differential equations $$F_\alpha \big(x, u, u_{(1)}, \ldots, u_{(s)}\big) = 0, \quad \alpha = 1, \ldots, m,$$ where $F_\alpha (x, u, u_{(1)},\ldots, u_{(s)})$ involves $n$ independent variables $x = (x^1, \ldots, x^n)$ and $m$ dependent variables ${u=(u^1,\ldots,u^m)}$, $u = u(x)$ together with their derivatives up to an arbitrary order $s$. For our purposes, we formulate the theorem in the case of systems of first-order differential equations. \[bct.th\] (See Ref. , Theorem 3.5). Let an operator (\[F3\]) be a symmetry of a system of first-order partial differential equations $$\label{adeq1} F_\alpha \big(x, u, u_{(1)}\big) = 0, \quad \alpha = 1, \ldots, m.$$ where $v=(v^1,\ldots,v^m)$. Then the quantities $$\label{F4} C^i = v^\beta \bigg[\xi^i F_\beta+ \big(\eta^\alpha - \xi^j u_j^\alpha\big) \frac{\partial F_\beta}{\partial u_i^\alpha}\bigg], \quad i = 1, \ldots, n,$$ furnish a conserved vector $C=(C^1,\ldots,C^n)$ for the equations (\[adeq1\]) considered together with the adjoint system $$\label{adeq2} F^*_\alpha \big(x, u, v,u_{(1)},v_{(1)}\big) \equiv \frac{\delta{\mathcal{L}}}{\delta u^\alpha} = 0, \quad \alpha = 1, \ldots, m,$$ where $$\frac{\delta}{\delta u^\alpha} = \frac{\partial}{\partial u^\alpha} - D_i\,\frac{\partial}{\partial u^\alpha_i} \,,\quad \alpha=1, \ldots, m,$$ and $v = (v^1, \ldots, v^m)$ are are new dependent variables, i.e., ${v=v(x)}$. \[bct.re1\] The simultaneous system of equations (\[adeq1\]) and (\[adeq2\]) with $2m$ dependent variables ${u=(u^1,\ldots,u^m)}$, ${v=(v^1,\ldots,v^m)}$ can be obtained as the Euler-Lagrange equations (\[F2\]) with the Lagrangian $$\label{lagr} \mathcal{L} = v^\beta F_\beta \big(x, u, u_{(1)},\ldots, u_{(s)}\big)$$ Indeed, $$\begin{aligned} &\frac{\delta\mathcal{L}}{\delta v^\alpha} = F_\alpha\big(x,u,u_{(1)}\big), \label{lagr1}\\ &\frac{\delta\mathcal{L}}{\delta u^\alpha} = F^*_\alpha\big(x,u,v,u_{(1)},v_{(1)}\big). \label{lagr2}\end{aligned}$$ \[bct.re2\] The conserved quantities (\[F4\]) can be written in terms of the Lagrangian (\[lagr\]) as follows: $$\label{F4L} C^i = \mathcal{L} \xi^i + \big(\eta^\alpha - \xi^j u_j^\alpha\big) \frac{\partial \mathcal{L}}{\partial u_i^\alpha}\,.$$ \[bct.re3\] If Eqs. (\[adeq1\]) have $r$ symmetries $X_1,\ldots,X_r$ of the form (\[F3\]), $$\begin{aligned} X_\mu = \xi_\mu^i(x,u)\frac{\partial}{\partial x^i} + \eta_\mu^\alpha (x,u) \frac{\partial}{\partial u^\alpha}\,, \quad \mu = 1, \ldots, r,\end{aligned}$$ then Eqs. (\[F4\]) provide $r$ conserved vectors $C_1,\ldots,C_r$ with the components $$\begin{aligned} C_\mu^i = \mathcal{L}\xi_\mu^i + \big(\eta_\mu^\alpha - \xi_\mu^j u_j^\alpha\big) \frac{\partial\mathcal{L}}{\partial u_i^\alpha}\,, \quad \mu= 1, \ldots, r; \ \ i = 1, \ldots, n.\end{aligned}$$ Electromagnetic equations ========================= Basic equations and the Lagrangian ---------------------------------- Adopting Dirac’s ideas on the existence of magnetic monopoles [@Dirac:PRSL:1931], one can formulate a symmetrized version of Maxwell’s electromagnetic equations [@Schwinger:Science:1969]. In SI units and in microscopic (Lorentz) form, these equations are \[cf.Ref. \]: \[eq:Maxwell-Dirac\] $$\begin{aligned} &{{\bm{\nabla}}{\bm{\times}}}{\mathbf{E}} + {\frac{\partial{{\mathbf{B}}}}{\partial t}} + \mu_0{\mathbf{j}}_m={\mathbf{0}}, \label{IKT2.1}\\ &{{\bm{\nabla}}{\bm{\times}}}{\mathbf{B}} - \frac{1}{c^2}{\frac{\partial{{\mathbf{E}}}}{\partial t}} - \mu_0{\mathbf{j}}_e={\mathbf{0}}, \label{IKT2.2}\\ &{{\bm{\nabla}}\cdot}{\mathbf{E}} - \mu_0c^2\rho_e =0,\label{IKT2.3}\\ &{{\bm{\nabla}}\cdot}{\mathbf{B}} - \mu_0\rho_m =0, \label{IKT2.4}\end{aligned}$$ together with the dual Ohm’s law $$\label{gohm} {\mathbf{j}}_e = \sigma_e{\mathbf{E}}, \quad{\mathbf{j}}\,_m = \sigma_m{\mathbf{B}}.$$ where $\sigma_m$ and $\sigma_e$ are constant scalar (rank zero) quantities. The first equation in (\[gohm\]) is Ohm’s law for electric currents. The second equation is a dual Ohm’s law for magnetic currents, that, for symmetry reasons, was introduced in Ref. ; see also Eq. (5) in Ref. , Eq. (38) in Ref. , and its generalization Eq. (8) in Ref. . Now we substitute Eqs. (\[gohm\]) into Eqs. (\[IKT2.1\])–(\[IKT2.2\]). The ensuing equations involve, along with the light velocity $c$, three other constants, $\sigma_e, \sigma_m$ and $\mu_0$. We eliminate two constants by setting $$\begin{aligned} \label{dimless} &\tilde t = ct, \quad \widetilde{{\mathbf{B}}} = c {\mathbf{B}}, \quad \tilde{\sigma}_e = c \mu_0 \sigma_e, \quad \tilde{\sigma}_m = \frac{\mu_0}{c}\, \sigma_m, \notag\\[1ex] &\tilde{\rho}_e = c^2 \mu_0 \rho_e, \quad \tilde{\rho}_m = c \mu_0 \rho_m\end{aligned}$$ and rewrite our basic Maxwell-Dirac equations (\[eq:Maxwell-Dirac\]), discarding tilde, as follows: $$\begin{aligned} \label{emeq} \begin{split} &{{\bm{\nabla}}{\bm{\times}}}{\mathbf{E}} + {\frac{\partial{{\mathbf{B}}}}{\partial t}} + \sigma_m{\mathbf{B}}={\mathbf{0}},\\ &{{\bm{\nabla}}{\bm{\times}}}{\mathbf{B}} - {\frac{\partial{{\mathbf{E}}}}{\partial t}} - \sigma_e{\mathbf{E}}={\mathbf{0}},\\ &{{\bm{\nabla}}\cdot}{\mathbf{E}} - \rho_e =0,\\ &{{\bm{\nabla}}\cdot}{\mathbf{B}} - \rho_m =0, \end{split}\end{aligned}$$ The system (\[emeq\]) has eight equations for eight dependent variables: six coordinates of the electric and magnetic vector fields ${\mathbf{E}}=(E^1,E^2,E^3)$ and ${\mathbf{B}}=(B^1,B^2,B^3)$, respectively, and two scalar quantities, viz., the electric and magnetic monopole charge densities $\rho_e$ and $\rho_m$. Using the method of @Ibragimov:JMAA:2006b we write the Lagrangian (\[lagr\]) for Eqs. (\[emeq\]) in the following form: $$\begin{aligned} \label{emlag.1} \begin{split} \mathcal{L} =&{\mathbf{V}}\cdot\Big({{\bm{\nabla}}{\bm{\times}}}{\mathbf{E}}+{\frac{\partial{{\mathbf{B}}}}{\partial t}}+\sigma_m{\mathbf{B}}\Big) + R_e \Big({{\bm{\nabla}}\cdot}{\mathbf{E}} - \rho_e\Big) \\ &\mbox{}+{\mathbf{W}}\cdot\Big({{\bm{\nabla}}{\bm{\times}}}{\mathbf{B}} - {\frac{\partial{{\mathbf{E}}}}{\partial t}} - \sigma_e{\mathbf{E}}\Big) + R_m \Big({{\bm{\nabla}}\cdot}{\mathbf{B}} - \rho_m\Big), \end{split}\end{aligned}$$ where ${\mathbf{V}},{\mathbf{W}},R_{e},R_{m}$ are adjoint variables (we note in passing that ${\mathbf{V}}$ is a pseudovector and $R_m$ a pseudoscalar). With this Lagrangian we have: $$\begin{aligned} \label{emlag.2} \begin{split} & \frac{\delta\mathcal{L}}{\delta{\mathbf{V}}} = {{\bm{\nabla}}{\bm{\times}}}{\mathbf{E}} + {\frac{\partial{{\mathbf{B}}}}{\partial t}} + \sigma_m{\mathbf{B}}, \quad \frac{\delta\mathcal{L}}{\delta R_e} = {{\bm{\nabla}}\cdot}{\mathbf{E}} - \rho_e,\\ & \frac{\delta\mathcal{L}}{\delta{\mathbf{W}}} = {{\bm{\nabla}}{\bm{\times}}}{\mathbf{B}} - {\frac{\partial{{\mathbf{E}}}}{\partial t}} - \sigma_e{\mathbf{E}}, \quad \frac{\delta\mathcal{L}}{\delta R_m} = {{\bm{\nabla}}\cdot}{\mathbf{B}} - \rho_m, \end{split}\end{aligned}$$ and $$\begin{aligned} \label{emlag.3} \begin{split} & \frac{\delta\mathcal{L}}{\delta {\mathbf{E}}} = {{\bm{\nabla}}{\bm{\times}}}{\mathbf{V}} + {\frac{\partial{{\mathbf{W}}}}{\partial t}} - \sigma_e{\mathbf{W}} - {\bm{\nabla}}R_e, \quad \frac{\delta\mathcal{L}}{\delta \rho_e} = - R_e, \\ & \frac{\delta\mathcal{L}}{\delta {\mathbf{B}}} = {{\bm{\nabla}}{\bm{\times}}}{\mathbf{W}} - {\frac{\partial{{\mathbf{V}}}}{\partial t}} + \sigma_m{\mathbf{V}} - {\bm{\nabla}}R_m, \quad \frac{\delta\mathcal{L}}{\delta\rho_m} = - R_m. \end{split}\end{aligned}$$ It follows from Eqs. (\[emlag.2\])–(\[emlag.3\]) that the Euler-Lagrange equations (\[F2\]) for the Lagrangian (\[emlag.1\]) provide the electromagnetic equations (\[emeq\]) and the following adjoint equations for the new dependent variables ${\mathbf{V}},{\mathbf{W}},R_e,R_m$: $$\begin{aligned} \label{emeq.adj} \begin{split} &{{\bm{\nabla}}{\bm{\times}}}{\mathbf{V}} + {\frac{\partial{{\mathbf{W}}}}{\partial t}} - \sigma_e{\mathbf{W}}=0,\\ &{{\bm{\nabla}}{\bm{\times}}}{\mathbf{W}} - {\frac{\partial{{\mathbf{V}}}}{\partial t}} + \sigma_m{\mathbf{V}}=0,\\ &R_e = 0, \qquad R_m =0. \end{split}\end{aligned}$$ \[emlag.4\] Let the spatial coordinates $x^1,x^2,x^3$ be $x,y,z$. For computing the variational derivatives $\delta\mathcal{L}/\delta{\mathbf{E}}$ and $\delta \mathcal{L}/\delta{\mathbf{B}}$ in Eqs. (\[emlag.3\]), it is convenient to use the coordinate representation of the Lagrangian (\[emlag.1\]), namely: $$\begin{aligned} \label{emlag.coord} \mathcal{L} & = V^1\,(E^3_y - E^2_z + B^1_t + \sigma_m B^1) + V^2\,(E^1_z - E^3_x + B^2_t + \sigma_m B^2)\notag\\[1ex] &+ V^3\,(E^2_x - E^1_y + B^3_t + \sigma_m B^3) + R_e (E^1_x + E^2_y + E^3_z - \rho_e)\\[1ex] &+ W^1\,(B^3_y - B^2_z - E^1_t - \sigma_e E^1) + W^2\,(B^1_z - B^3_x - E^2_t - \sigma_e E^2)\notag\\[1ex] &+ W^3\,(B^2_x - B^1_y - E^3_t - \sigma_e E^3) + R_m (B^1_x + B^2_y + B^3_z - \rho_m).\notag\end{aligned}$$ Symmetries ---------- Eqs. (\[emeq\]) are invariant under the translations of time $t$ and the position vector ${\mathbf{x}} = (x, y, z)$ as well as the simultaneous rotations of the vectors ${\mathbf{x}},{\mathbf{E}}$ and ${\mathbf{B}}$ due to the vector formulation of Eqs. (\[emeq\]). These geometric transformations provide the following seven infinitesimal symmetries: $$\begin{aligned} \label{emeq.sym} \begin{split} &\qquad X_0 = {\frac{\partial{\,}}{\partial t}}, \quad X_1 = {\frac{\partial{\,}}{\partial x}}, \quad X_2 = {\frac{\partial{\,}}{\partial y}}, \quad X_3 = {\frac{\partial{\,}}{\partial z}},\\ & X_{12} = y\frac{\partial}{\partial x} - x\frac{\partial}{\partial y} + E^2\frac{\partial}{\partial E^1} - E^1\frac{\partial}{\partial E^2} + B^2\frac{\partial}{\partial B^1} - B^1\frac{\partial}{\partial B^2}\,,\\ & X_{13} = z\frac{\partial}{\partial x} - x\frac{\partial}{\partial z} + E^3\frac{\partial}{\partial E^1} - E^1\frac{\partial}{\partial E^3} + B^3\frac{\partial}{\partial B^1} - B^1\frac{\partial}{\partial B^3}\,,\\ & X_{23} = z\frac{\partial}{\partial y} - y\frac{\partial}{\partial z} + E^3\frac{\partial}{\partial E^2} - E^2\frac{\partial}{\partial E^3} + B^3\frac{\partial}{\partial B^2} - B^2\frac{\partial}{\partial B^3}\,. \end{split}\end{aligned}$$ The infinitesimal symmetries for the adjoint system (\[emeq.adj\]) are obtained from (\[emeq.sym\]) by replacing the vectors ${\mathbf{E}}$ and ${\mathbf{B}}$ by ${\mathbf{V}}$ and ${\mathbf{W}}$, respectively. Moreover, since Eqs. (\[emeq\]) are homogeneous, they admit simultaneous dilations of all dependent variables with the generator $$\label{hom} T = {\mathbf{E}} \cdot \frac{\partial}{\partial {\mathbf{E}}} +{\mathbf{B}} \cdot \frac{\partial}{\partial {\mathbf{B}}} + \rho_e \frac{\partial}{\partial \rho_e} + \rho_m \frac{\partial}{\partial \rho_m}\,,$$ where $${\mathbf{E}} \cdot \frac{\partial}{\partial {\mathbf{E}}} = \sum\limits_{i=1}^3 E^i \frac{\partial}{\partial E^i}\,, \quad {\mathbf{B}} \cdot \frac{\partial}{\partial {\mathbf{B}}} = \sum\limits_{i=1}^3 B^i \frac{\partial}{\partial B^i}\,.$$ Recall that the Maxwell equations in vacuum admit also the one-parameter group of *Heaviside-Larmor-Rainich duality transformations* $$\label{eq:HLR} \overline{{\mathbf{E}}}= {\mathbf{E}} \cos\alpha-{\mathbf{B}}\sin\alpha, \quad \overline{{\mathbf{B}}}={\mathbf{E}}\sin\alpha+{\mathbf{B}}\cos\alpha$$ with the generator $$X = {\mathbf{E}} \cdot \frac{\partial}{\partial {\mathbf{B}}} -{\mathbf{B}} \cdot \frac{\partial}{\partial {\mathbf{E}}} \equiv \sum\limits_{i=1}^3 \bigg(E^i \frac{\partial}{\partial B^i} - B^i \frac{\partial}{\partial E^i}\bigg)\,.$$ Also recall that the “mixing angle” $\alpha$ in (\[eq:HLR\]) is a pseudoscalar. It was shown in [@Ibragimov:JMAA:2006b] that the group (\[eq:HLR\]) provides the conservation of energy for the Maxwell equations. Let us clarify whether Eqs. (\[emeq\]) admit a similar group. Let therefore $$\label{IKT2.7} X = {\mathbf{E}} \cdot \frac{\partial}{\partial {\mathbf{B}}} -{\mathbf{B}} \cdot \frac{\partial}{\partial {\mathbf{E}}} + \rho_e \frac{\partial}{\partial \rho_m} - \rho_m \frac{\partial}{\partial \rho_e}\,.$$ The prolongation of the operator (\[IKT2.7\]) is written $$\begin{gathered} \label{IKT2.8} X = {\mathbf{E}}\cdot\frac{\partial}{\partial{\mathbf{B}}} -{\mathbf{B}}\cdot\frac{\partial}{\partial {\mathbf{E}}} +\rho_e\frac{\partial}{\partial\rho_m} -\rho_m\frac{\partial}{\partial\rho_e} \\ +{\mathbf{E}}_t\cdot\frac{\partial}{\partial{\mathbf{B}}_t} -{\mathbf{B}}_t\cdot\frac{\partial}{\partial{\mathbf{E}}_t} +{\mathbf{E}}_x\cdot\frac{\partial}{\partial{\mathbf{B}}_x} -{\mathbf{B}}_x\cdot\frac{\partial}{\partial{\mathbf{E}}_x} \\ +{\mathbf{E}}_y\cdot\frac{\partial}{\partial{\mathbf{B}}_y} -{\mathbf{B}}_y\cdot\frac{\partial}{\partial{\mathbf{E}}_y} +{\mathbf{E}}_z\cdot\frac{\partial}{\partial{\mathbf{B}}_z} -{\mathbf{B}}_z\cdot\frac{\partial}{\partial{\mathbf{E}}_z}\,.\end{gathered}$$ Reckoning shows that the operator (\[IKT2.8\]) acts on the left-hand sides of Eqs. (\[emeq\]) as follows: $$\begin{gathered} X\big({{\bm{\nabla}}{\bm{\times}}}{\mathbf{E}} + {\mathbf{B}}_t + \sigma_m {\mathbf{B}}\big) = - \big({{\bm{\nabla}}{\bm{\times}}}{\mathbf{B}} - {\mathbf{E}}_t - \sigma_m {\mathbf{E}}\big),\\ X\big({{\bm{\nabla}}{\bm{\times}}}{\mathbf{B}} - {\mathbf{E}}_t - \sigma_e {\mathbf{E}}\big) = {{\bm{\nabla}}{\bm{\times}}}{\mathbf{E}} + {\mathbf{B}}_t + \sigma_e {\mathbf{B}},\\ X\big({{\bm{\nabla}}\cdot}{\mathbf{E}} - \rho_e\big) = - \big({{\bm{\nabla}}\cdot}{\mathbf{B}} - \rho_m\big),\\ X\big({{\bm{\nabla}}\cdot}{\mathbf{B}} - \rho_m\big) = {{\bm{\nabla}}\cdot}{\mathbf{E}} - \rho_e.\end{gathered}$$ It follows that the operator (\[IKT2.7\]) is admitted by Eqs. (\[emeq\]) only in the case $$\label{emlag.5} \sigma_m = \sigma_e.$$ Conservation laws ================= Derivation of conservation laws {#dcl} ------------------------------- We will write the conservation law (\[C1\]) in the form $$\label{IKT2.9} D_t(\tau) + \mathrm{div}\,\bm{\chi} = 0,$$ where the pseudoscalar $\tau$ is the density of the conservation law (\[IKT2.9\]), the pseudovector current ${\bm{\chi}=(\chi^1,\chi^2,\chi^3)}$, and $$\mathrm{div}\,\bm{\chi} \equiv {{\bm{\nabla}}\cdot}\bm{\chi} = D_x(\chi^1) + D_y(\chi^2) + D_z(\chi^3).$$ Let us find the conservation law furnished by the symmetry (\[IKT2.7\]) when the condition (\[emlag.5\]) is satisfied, $\sigma_m=\sigma_e$. Applying the formula (\[F4\]) to the symmetry (\[IKT2.7\]) and to the Lagrangian (\[emlag.1\]), we obtain the following density of the conservation law (\[IKT2.9\]): $$\tau = {\mathbf{E}}\cdot\frac{\partial\mathcal{L}}{\partial {\mathbf{B}}_t} -{\mathbf{B}}\cdot\frac{\partial\mathcal{L}}{\partial {\mathbf{E}}_t} ={\mathbf{E}}\cdot{\mathbf{V}} + {\mathbf{B}}\cdot{\mathbf{W}}.$$ Thus, $$\label{IKT2.10} \tau = {\mathbf{E}}\cdot{\mathbf{V}} + {\mathbf{B}}\cdot{\mathbf{W}}.$$ The pseudovector $\bm{\chi}$ is obtained likewise. For example, using the Lagrangian in the form (\[emlag.coord\]), we have: $$\chi^1 = {\mathbf{E}}\cdot\frac{\partial\mathcal{L}}{\partial {\mathbf{B}}_x} -{\mathbf{B}}\cdot\frac{\partial\mathcal{L}}{\partial{\mathbf{E}}_x} = E^2 W^3 - E^3 W^2 - B^2 V^3 + B^3 V^2.$$ The other coordinates of $\bm{\chi}$ are computed likewise, and the final result is: $$\label{IKT2.11} \bm{\chi} = ({\mathbf{E}}{\bm{\times}}{\mathbf{W}}) - ({\mathbf{B}}{\bm{\times}}{\mathbf{V}}).$$ One can readily verify that (\[IKT2.10\]) and (\[IKT2.11\]) provide a conservation law for Eqs. (\[emeq\]) considered together with the adjoint equations (\[emeq.adj\]). Indeed, using the well-known formula ${{\bm{\nabla}}\cdot}({\mathbf{a}}{\bm{\times}}{\mathbf{b}})={\mathbf{b}}\cdot({{\bm{\nabla}}{\bm{\times}}}{\mathbf{a}}) -{\mathbf{a}}\cdot({{\bm{\nabla}}{\bm{\times}}}{\mathbf{b}})$ and Eqs. (\[emeq\]) and (\[emeq.adj\]), we obtain: $$\begin{aligned} D_t (\tau) &= {\mathbf{E}}_t\cdot{\mathbf{V}} + {\mathbf{E}}\cdot{\mathbf{V}}_t + {\mathbf{B}}_t\cdot{\mathbf{W}}+ {\mathbf{B}}\cdot{\mathbf{W}}_t\\ &= {\mathbf{V}}\cdot({{\bm{\nabla}}{\bm{\times}}}{\mathbf{B}} - \sigma_e{\mathbf{E}}) + {\mathbf{E}}\cdot({{\bm{\nabla}}{\bm{\times}}}{\mathbf{W}} + \sigma_m{\mathbf{V}})\\ &\quad\mbox{}- {\mathbf{W}}\cdot({{\bm{\nabla}}{\bm{\times}}}{\mathbf{E}} + \sigma_m{\mathbf{B}}) - {\mathbf{B}}\cdot({{\bm{\nabla}}{\bm{\times}}}{\mathbf{V}} - \sigma_e {\mathbf{W}}),\notag\\[1.5ex] {{\bm{\nabla}}\cdot}\bm{\chi} &= {{\bm{\nabla}}\cdot}({\mathbf{E}}{\bm{\times}}{\mathbf{W}}) - {{\bm{\nabla}}\cdot}{\mathbf{B}}{\bm{\times}}{\mathbf{V}})\\ &= {\mathbf{W}}\cdot({{\bm{\nabla}}{\bm{\times}}}{\mathbf{E}}) - {\mathbf{E}}\cdot({{\bm{\nabla}}{\bm{\times}}}{\mathbf{W}}) \\ &\quad\mbox{}- {\mathbf{V}}\cdot({{\bm{\nabla}}{\bm{\times}}}{\mathbf{B}}) + {\mathbf{B}}\cdot({{\bm{\nabla}}{\bm{\times}}}{\mathbf{V}}).\end{aligned}$$ Whence, $$D_t (\tau) + \mathrm{div}\, \bm{\chi} = (\sigma_m - \sigma_e) ({\mathbf{E}}\cdot{\mathbf{V}} - {\mathbf{B}}\cdot{\mathbf{W}}).$$ It follows again that the conservation law is valid only if ${\sigma_m=\sigma_e}$. \[emr.1\] The conservation law given by (\[IKT2.10\])–(\[IKT2.11\]) depends on solutions $({\mathbf{V}},{\mathbf{W}})$ of the adjoint system (\[emeq.adj\]). However, substituting into Eqs. (\[IKT2.10\]) and (\[IKT2.11\]) any particular solution $({\mathbf{V}},{\mathbf{W}})$ of the adjoint system (\[emeq.adj\]) with $\sigma_m=\sigma_e$, one obtains the conservation law for Eqs. (\[emeq\]) not involving ${\mathbf{V}}$ and ${\mathbf{W}}$. Let us denote $\sigma_m=\sigma_e=\sigma$ and take, e.g., the following simple solution of the adjoint system (\[emeq.adj\]): $$V^1 = e^{\sigma t}, \ V^2 = V^3 = 0; \quad W^1 = e^{\sigma t}, \ W^2 = W^3 = 0.$$ Then Eqs. (\[IKT2.10\])–(\[IKT2.11\]) yield: $$\begin{gathered} \tau = (E^1 + B^1)e^{\sigma t} \\ \chi^1 = 0 \quad \chi^2 = (E^3 - B^3)e^{\sigma t} \quad \chi^3 = (B^2 - E^2)e^{\sigma t}\,.\end{gathered}$$ The operator (\[IKT2.7\]) generates the one-parameter group $$\begin{gathered} \overline{{\mathbf{E}}}= {\mathbf{E}} \cos\alpha-{\mathbf{B}}\sin\alpha, \quad \overline{{\mathbf{B}}}={\mathbf{E}}\sin\alpha+{\mathbf{B}}\cos\alpha,\\ \overline{\rho}\,^e= \rho_e \cos\alpha-\rho_m\sin\alpha, \quad \overline{\rho}\,^m= \rho_e\sin\alpha+\rho_m\cos\alpha\,.\end{gathered}$$ where, again, the “mixing angle” $\alpha$ is a pseudoscalar. In the original variables used in Eqs. (\[IKT2.1\])–(\[IKT2.2\]) and (\[gohm\]), the operator (\[IKT2.7\]) is written: $$X = \frac{1}{c}\,{\mathbf{E}}\cdot\frac{\partial}{\partial{\mathbf{B}}} -c \,{\mathbf{B}}\cdot\frac{\partial}{\partial{\mathbf{E}}} + c\,\rho_e \frac{\partial}{\partial\rho_m} - \frac{1}{c}\,\rho_m \frac{\partial}{\partial\rho_e}\,.$$ Applying similar calculations to the generator (\[hom\]) of the dilation group provides the conservation law with $$\label{hom.1} \tau = {\mathbf{B}}\cdot{\mathbf{V}} - {\mathbf{E}}\cdot{\mathbf{W}}, \quad \bm{\chi} = ({\mathbf{E}}{\bm{\times}}{\mathbf{V}}) + ({\mathbf{B}}{\bm{\times}}{\mathbf{W}}).$$ This conservation law is valid for arbitrary $\sigma_m$ and $\sigma_e$. Indeed, $$\begin{aligned} D_t (\tau) &= {\mathbf{B}}_t\cdot {\mathbf{V}} + {\mathbf{B}}\cdot{\mathbf{V}}_t - {\mathbf{E}}_t\cdot {\mathbf{W}} - {\mathbf{E}}\cdot{\mathbf{W}}_t\\ &= -{\mathbf{V}}\cdot({{\bm{\nabla}}{\bm{\times}}}{\mathbf{E}} + \sigma_m {\mathbf{B}}) + {\mathbf{B}}\cdot({{\bm{\nabla}}{\bm{\times}}}{\mathbf{W}} + \sigma_m{\mathbf{V}})\\ &\quad\mbox{}-{\mathbf{W}}\cdot({{\bm{\nabla}}{\bm{\times}}}{\mathbf{B}} - \sigma_e{\mathbf{E}}) + {\mathbf{E}}\cdot({{\bm{\nabla}}{\bm{\times}}}{\mathbf{V}} - \sigma_e{\mathbf{W}})\\ &= -{\mathbf{V}}\cdot({{\bm{\nabla}}{\bm{\times}}}{\mathbf{E}}) + {\mathbf{B}}\cdot({{\bm{\nabla}}{\bm{\times}}}{\mathbf{W}})\\ &\quad\mbox{}-{\mathbf{W}}\cdot({{\bm{\nabla}}{\bm{\times}}}{\mathbf{B}}) + {\mathbf{E}}\cdot({{\bm{\nabla}}{\bm{\times}}}{\mathbf{V}})\,,\\[1.5ex] {{\bm{\nabla}}\cdot}\bm{\chi} &= {{\bm{\nabla}}\cdot}({\mathbf{E}}{\bm{\times}}{\mathbf{V}}) + {{\bm{\nabla}}\cdot}({\mathbf{B}}{\bm{\times}}{\mathbf{W}})\\ &= {\mathbf{V}}\cdot({{\bm{\nabla}}{\bm{\times}}}{\mathbf{E}}) -{\mathbf{E}}\cdot({{\bm{\nabla}}{\bm{\times}}}{\mathbf{V}})\\ &\quad\mbox{}+{\mathbf{W}}\cdot({{\bm{\nabla}}{\bm{\times}}}{\mathbf{B}}) - {\mathbf{B}}\cdot({{\bm{\nabla}}{\bm{\times}}}{\mathbf{W}})\,.\end{aligned}$$ Hence, $D_t(\tau)+\mathrm{div}\,\bm{\chi}=0$. Let us find the conservation law provided by the symmetry $X_0={\partial}/{\partial t}$ from (\[emeq.sym\]). Formula (\[F4\]) yields: $$\begin{aligned} \tau = \mathcal{L} - {\mathbf{E}}_t\cdot\frac{\partial\mathcal{L}}{\partial{\mathbf{E}}_t} -{\mathbf{B}}_t\cdot \frac{\partial\mathcal{L}}{\partial {\mathbf{B}}_t} = \mathcal{L} + {\mathbf{E}}_t \cdot {\mathbf{W}} - {\mathbf{B}}_t \cdot {\mathbf{V}}.\end{aligned}$$ Since the Lagrangian $\mathcal{L}$ given by (\[emlag.1\]) vanishes on the solutions of Eqs. (\[emeq\]), we can take $$\label{IKT2.12A} \tau = {\mathbf{E}}_t\cdot{\mathbf{W}} - {\mathbf{B}}_t\cdot{\mathbf{V}}$$ or $$\label{IKT2.12} \tau = {\mathbf{W}}\cdot[({{\bm{\nabla}}{\bm{\times}}}\bm{B}) - \sigma_e{\mathbf{E}}] + {\mathbf{V}}\cdot[({{\bm{\nabla}}{\bm{\times}}}{\mathbf{E}}) + \sigma_m {\mathbf{B}}].$$ Let us calculate the pseudovector $\bm{\chi}$. Formula (\[F4\]) yields: $$\chi^1 = - {\mathbf{E}}_t\cdot\frac{\partial\mathcal{L}}{\partial {\mathbf{E}}_x} -{\mathbf{B}}_t\cdot \frac{\partial\mathcal{L}}{\partial {\mathbf{B}}_x}\,.$$ Using the Lagrangian in the form (\[emlag.coord\]), we have: $$\chi^1 = - E^2_t V^3 + E^3_t V^2 - B^2_t W^3 + B^3_t W^2.$$ The other coordinates of $\bm{\chi}$ are computed similarly, and the final result is $$\label{IKT2.13} \bm{\chi} = ({\mathbf{V}}{\bm{\times}}{\mathbf{E}}_t) + ({\mathbf{W}}{\bm{\times}}{\mathbf{B}}_t).$$ Thus, the time translational invariance of Eqs. (\[emeq\]) leads to the conservation law (\[IKT2.9\]) with $\tau$ and $\bm{\chi}$ given by (\[IKT2.12\]) and (\[IKT2.13\]), respectively. Let us substitute in Eqs. (\[IKT2.12\]) and (\[IKT2.13\]) the following simple solution of the adjoint system (cf. Remark \[emr.1\]): $$V^1 = e^{\sigma_m t}, \ V^2 = V^3 = 0; \quad W^1 = e^{\sigma_e t}, \ W^2 = W^3 = 0.$$ Then Eqs. (\[IKT2.12\])–(\[IKT2.13\]) yield: $$\begin{gathered} \tau = (B^3_y - B^2_z - \sigma_e E^1) e^{\sigma_e t} + (E^3_y - E^2_z + \sigma_m B^1) e^{\sigma_m t},\\ \chi^1 = 0, \quad \chi^2 = - E^3_t e^{\sigma_m t} - B^3_t e^{\sigma_e t}, \quad \chi^3 = E^2_t e^{\sigma_m t} + B^2_t e^{\sigma_e t}.\end{gathered}$$ The conservation law provided by the symmetry $X_1=\partial/\partial x$ from (\[emeq.sym\]) has the following density: $$\tau = - {\mathbf{E}}_x\cdot \frac{\partial\mathcal{L}}{\partial {\mathbf{E}}_t} -{\mathbf{B}}_x\cdot \frac{\partial\mathcal{L}}{\partial {\mathbf{B}}_t} = {\mathbf{E}}_x\cdot {\mathbf{W}} - {\mathbf{B}}_x \cdot {\mathbf{V}}.$$ For the pseudovector $\bm{\chi}$ the formula (\[F4\]) yields: $$\chi^1 = \mathcal{L}- {\mathbf{E}}_x\cdot \frac{\partial\mathcal{L}}{\partial {\mathbf{E}}_x} -{\mathbf{B}}_x\cdot \frac{\partial\mathcal{L}}{\partial{\mathbf{B}}_x}\,.$$ Using the Lagrangian in the form (\[emlag.coord\]), we have: $$\chi^1 = \mathcal{L}- E^2_x V^3 + E^3_x V^2 - B^2_x W^3 + B^3_x W^2.$$ The other coordinates of $\bm{\chi}$ are calculated similarly: $$\begin{gathered} \chi^2 = E^1_x V^3 - E^3_x V^1 + B^1_x W^3 - B^3_x W^1,\\ \chi^3 = -E^1_x V^2 + E^2_x V^1 - B^1_x W^2 + B^2_x W^1.\end{gathered}$$ We can ignore $\mathcal{L}$ in $\chi^1$ since $D_x\mathcal{L}=0$ on solutions of Eqs. (\[emeq\]) and (\[emeq.adj\]), and the final result is $$\label{IKT2.131} \bm{\chi} = ({\mathbf{V}}{\bm{\times}}{\mathbf{E}}_x) + ({\mathbf{W}}{\bm{\times}}{\mathbf{B}}_x).$$ Replacing $x$ by $y$ and $z$ we obtain the following conservation laws corresponding to $X_2=\partial/\partial y$ and $X_3=\partial/\partial z$, respectively: $$\tau ={\mathbf{E}}_y\cdot{\mathbf{W}} - {\mathbf{B}}_y\cdot{\mathbf{V}},\quad \bm{\chi} = ({\mathbf{V}}{\bm{\times}}{\mathbf{E}}_y) + ({\mathbf{W}}{\bm{\times}}{\mathbf{B}}_y)$$ and $$\tau ={\mathbf{E}}_z\cdot{\mathbf{W}} - {\mathbf{B}}_z\cdot{\mathbf{V}},\quad \bm{\chi} = ({\mathbf{V}}{\bm{\times}}{\mathbf{E}}_z) + ({\mathbf{W}}{\bm{\times}}{\mathbf{B}}_z).$$ Applying Formula (\[F4\]) to the symmetry $X_{12}$ and to the Lagrangian (\[emlag.coord\]), we obtain the following density of the conservation law: $$\begin{aligned} \tau =& E^{2}\frac{\partial\mathcal{L}}{\partial E^{1}_t}- E^{1}\frac{\partial\mathcal{L}}{\partial E^{2}_t} + (x{\mathbf{E}}_y -y {\mathbf{E}}_x)\cdot \frac{\partial\mathcal{L}}{\partial{\mathbf{E}}_t} \\ &\mbox{}+ B^{2}\frac{\partial\mathcal{L}}{\partial B^{1}_t}- B^{1}\frac{\partial\mathcal{L}}{\partial {\mathbf{B}}^{2}_t} + (x{\mathbf{B}}_y -y {\mathbf{B}}_x)\cdot \frac{\partial\mathcal{L}}{\partial{\mathbf{B}}_t}\\ =& W^{2}E^{1}-W^{1}E^{2} + (y{\mathbf{E}}_x -x{\mathbf{E}}_y)\cdot{\mathbf{W}} \\ &\mbox{} - (V^{2}B^{1}-V^{1}B^{2})- (y{\mathbf{B}}_x -x{\mathbf{B}}_y) \cdot {\mathbf{V}}.\end{aligned}$$ The densities of the conservation laws for $X_{13}$ and $X_{23}$ are $$\begin{gathered} \tau = W^{3}E^{1}-W^{1}E^{3} + (z{\mathbf{E}}_x -x{\mathbf{E}}_z)\cdot {\mathbf{W}} \\ - (V^{3}B^{1}-V^{1}B^{3})- (z{\mathbf{B}}_x -x{\mathbf{B}}_z) \cdot {\mathbf{V}}\end{gathered}$$ and $$\begin{gathered} \tau = W^{3}E^{2}-W^{2}E^{3} + (z{\mathbf{E}}_y -y{\mathbf{E}}_z)\cdot {\mathbf{W}} \\ \mbox{} - (V^{3}B^{2}-V^{2}B^{3})- (z{\mathbf{B}}_y -y{\mathbf{B}}_z) \cdot {\mathbf{V}},\end{gathered}$$ respectively. Finally, the densities of conservation laws corresponding to the rotation generators $X_{ij}$ can be written as one vector: $$\label{IKT2.132} {\mathbf{\tau}} ={\mathbf{W}}{\bm{\times}}{\mathbf{E}} + {\mathbf{W}}\cdot({\mathbf{x}}{\bm{\times}}{\bm{\nabla}}){\mathbf{E}} - {\mathbf{V}}{\bm{\times}}{\mathbf{B}}- {\mathbf{V}}\cdot({\mathbf{x}}{\bm{\times}}{\bm{\nabla}}){\mathbf{B}},$$ where ${\mathbf{x}}=(x,y,z)$. The operator $X_{12}$ provides the following pseudovector $\bm{\chi}$: $$\begin{aligned} \chi^1 =&-V^{3}E^{1}- y(E^2_x V^3 - E^3_x V^2)+x(E^2_y V^3 - E^3_y V^2) \\ &\mbox{}-W^{3}B^{1} - y(B^2_x W^3 - B^3_x W^2)+x(B^2_y W^3 - B^3_yW^2), \\ \chi^2 =&-V^{3}E^{2}+ y(E^1_x V^3 - E^3_x V^1)-x(E^1_y V^3 - E^3_y V^1) \\ &\mbox{}-W^{3}B^{2} + y(B^1_x W^3 - B^3_x W^1)-x(B^1_y W^3 - B^3_yW^1),\\ \chi^3 =& V^{1}E^{1}+ V^{2}E^{2}- y(E^1_x V^2 - E^2_x V^1)+x(E^1_y V^2 - E^2_y V^1) \\ &\mbox{}+W^{1}B^{1}+ W^{2}B^{2}- y(B^1_x W^2 - B^2_x W^1)+x(B^1_y W^2 - B^2_y W^1)\,.\end{aligned}$$ The pseudovector $\bm{\chi}$ for the operator $X_{13}$ has the following form: $$\begin{aligned} \chi^1 =& V^{2}E^{1}- z(E^2_x V^3 - E^3_x V^2)+x(E^2_z V^3 - E^3_z V^2) \\ &\mbox{}+W^{2}B^{1} - z(B^2_x W^3 - B^3_x W^2)+x(B^2_z W^3 - B^3_zW^2),\\ \chi^2 =&-V^{1}E^{1}-V^{3}E^{3}+ z(E^1_x V^3 - E^3_x V^1)-x(E^1_z V^3 - E^3_z V^1) \\ &\mbox{}-W^{1}B^{1}-W^{3}B^{3} + z(B^1_x W^3 - B^3_x W^1)-x(B^1_z W^3 - B^3_zW^1),\\ \chi^3 =&V^{2}E^{3}- z(E^1_x V^2 - E^2_x V^1)+x(E^1_z V^2 - E^2_z V^1) \\ &\mbox{} + W^{2}B^{3}- z(B^1_x W^2 - B^2_x W^1)+x(B^1_z W^2 - B^2_z W^1)\,.\end{aligned}$$ The operator $X_{23}$ provides the following pseudovector $\bm{\chi}$: $$\begin{aligned} \chi^1 =& V^{2}E^{2}+V^{3}E^{3}- z(E^2_y V^3 - E^3_y V^2)+y(E^2_z V^3 - E^3_z V^2) \\ &\mbox{}+W^{2}B^{2}+W^{3}B^{3}- z(B^2_y W^3 - B^3_y W^2)+y(B^2_z W^3 - B^3_zW^2),\\ \chi^2 =&-V^{1}E^{2}+ z(E^1_y V^3 - E^3_y V^1)-y(E^1_z V^3 - E^3_z V^1) \\ &\mbox{} -W^{1}B^{2} + z(B^1_y W^3 - B^3_y W^1)-y(B^1_z W^3 - B^3_zW^1)\,,\\ \chi^3 =&-V^{1}E^{3}- z(E^1_y V^2 - E^2_y V^1)+y(E^1_z V^2 - E^2_z V^1) \\ &\mbox{} -W^{1}B^{3}- z(B^1_yW^2 - B^2_y W^1)+y(B^1_z W^2 - B^2_z W^1)\,.\end{aligned}$$ Two-solution representation of conservation laws {#tsr} ------------------------------------------------ The conserved quantities obtained in Subsection \[dcl\] involve solutions ${\mathbf{V}}, {\mathbf{W}}$ of the adjoint equations (\[emeq.adj\]). It may be useful for applications to give an alternative representation of the conserved quantities in terms of the electric and magnetic vector fields ${\mathbf{E}}, {\mathbf{B}}$ only. We suggest here one possibility based on the observation that one can satisfy the adjoint system (\[emeq.adj\]) by letting $$\begin{aligned} \label{tsr1} \begin{split} & {\mathbf{V}}({\mathbf{x}},t) = {\mathbf{B}}({\mathbf{x}},-t), \\ & {\mathbf{W}}({\mathbf{x}},t) = {\mathbf{E}}({\mathbf{x}},-t), \\ & R_e(\bm{x},t) = {{\bm{\nabla}}\cdot}{\mathbf{E}}({\mathbf{x}},-t) -\rho_e({\mathbf{x}},-t), \\ & R_m(\bm{x},t) = {{\bm{\nabla}}\cdot}{\mathbf{B}}({\mathbf{x}},-t) -\rho_m({\mathbf{x}},-t), \end{split}\end{aligned}$$ where ${\mathbf{E}}({\mathbf{x}},s),{\mathbf{B}}({\mathbf{x}},s)$ solve Eqs. (\[emeq\]) with $s=-t$. Indeed, employing the substitution (\[tsr1\]) and the notation $s=-t$ we have $$\begin{aligned} \begin{split} &{{\bm{\nabla}}{\bm{\times}}}{\mathbf{V}} + {\frac{\partial{{\mathbf{W}}}}{\partial t}}- \sigma_e{\mathbf{W}} \\ &\qquad= {{\bm{\nabla}}{\bm{\times}}}{\mathbf{B}}({\mathbf{x}},s) + {\frac{\partial{{\mathbf{E}}({\mathbf{x}},s)}}{\partial s}}{\frac{\partial{s}}{\partial t}} \mbox{}- \sigma_e\,{\mathbf{E}}({\mathbf{x}},s)\,, \\ &{{\bm{\nabla}}{\bm{\times}}}{\mathbf{W}} - {\frac{\partial{{\mathbf{V}}}}{\partial t}} + \sigma_m {\mathbf{V}} \\ &\qquad = {{\bm{\nabla}}{\bm{\times}}}{\mathbf{E}}({\mathbf{x}},s) - {\frac{\partial{{\mathbf{B}}({\mathbf{x}},s)}}{\partial s}} {\frac{\partial{s}}{\partial t}} + \sigma_m{\mathbf{B}}({\mathbf{x}},s)\,, \\ & R_e = {{\bm{\nabla}}\cdot}{\mathbf{E}}({\mathbf{x}},s) - \rho_e({\mathbf{x}},s)=0\,, \\ & R_m ={{\bm{\nabla}}\cdot}{\mathbf{B}}({\mathbf{x}},s) - \rho_m({\mathbf{x}},s)\,. \end{split}\end{aligned}$$ Hence, the adjoint equations (\[emeq.adj\]) reduce to (\[emeq\]): $$\begin{aligned} \label{tsr2} \begin{split} & {{\bm{\nabla}}{\bm{\times}}}{\mathbf{E}}({\mathbf{x}},s) + {\frac{\partial{{\mathbf{B}}({\mathbf{x}},s)}}{\partial s}} + \sigma_m{\mathbf{B}}({\mathbf{x}},s)=0, \\ & {{\bm{\nabla}}\cdot}{\mathbf{B}}({\mathbf{x}},s) - {\frac{\partial{{\mathbf{E}}({\mathbf{x}},s)}}{\partial s}} - \sigma_e{\mathbf{E}}({\mathbf{x}},s)=0, \\ & {{\bm{\nabla}}\cdot}{\mathbf{E}}({\mathbf{x}},s) - \rho_e({\mathbf{x}},s)=0, \\ & {{\bm{\nabla}}\cdot}{\mathbf{B}}({\mathbf{x}},s) - \rho_m({\mathbf{x}},s)=0\,. \end{split}\end{aligned}$$ Let $\big({\mathbf{E}}({\mathbf{x}},t), {\mathbf{B}}({\mathbf{x}},t)\big)$ and $\big({\mathbf{E}}'({\mathbf{x}},t),{\mathbf{B}}'({\mathbf{x}},t)\big)$ be any two solutions of the electromagnetic equations (\[emeq\]). Substituting in (\[tsr1\]) the solution $\big({\mathbf{E}}',{\mathbf{B}}'\big),$ we obtain the *two-solution representations* of the conservation laws. For example, the conservation law given by (\[IKT2.10\])–(\[IKT2.11\]) has in this representation the following coordinates: $$\begin{aligned} \label{IKT2.10T} \begin{split} & \tau = {\mathbf{E}}({\mathbf{x}},t)\cdot{\mathbf{B}}'({\mathbf{x}},-t) + {\mathbf{B}}({\mathbf{x}},t)\cdot{\mathbf{E}}'({\mathbf{x}},-t),\\ & \bm{\chi} = [{\mathbf{E}}({\mathbf{x}},t){\bm{\times}}{\mathbf{E}}'({\mathbf{x}},-t)] - [{\mathbf{B}}({\mathbf{x}},t){\bm{\times}}{\mathbf{B}}'({\mathbf{x}},-t)]\,. \end{split}\end{aligned}$$ In particular, if the solutions $\big({\mathbf{E}}({\mathbf{x}},t),{\mathbf{B}},t)\big)$ are identical, (\[IKT2.10T\]) provides the *one-solution representation*: $$\begin{aligned} \label{IKT2.10Ta} \begin{split} &\tau = {\mathbf{E}}({\mathbf{x}},t)\cdot{\mathbf{B}}({\mathbf{x}},-t) + {\mathbf{B}}({\mathbf{x}}, t)\cdot{\mathbf{E}}({\mathbf{x}},-t),\\ & \bm{\chi} = [{\mathbf{E}}({\mathbf{x}},t){\bm{\times}}{\mathbf{E}}({\mathbf{x}},-t)] - [{\mathbf{B}}({\mathbf{x}},t){\bm{\times}}{\mathbf{B}}({\mathbf{x}},-t)]\,. \end{split}\end{aligned}$$ All other conservation laws can be treated likewise, e.g., the conservation law given by (\[IKT2.12A\]) and (\[IKT2.13\]) has the following two-solution representation: $$\begin{aligned} \label{IKT2.12AT} \begin{split} & \tau = {\mathbf{E}}_t({\mathbf{x}},t)\cdot{\mathbf{E}}'({\mathbf{x}},-t) -{\mathbf{B}}_t({\mathbf{x}},t)\cdot{\mathbf{B}}'({\mathbf{x}},-t),\\ & \bm{\chi} = [{\mathbf{B}}'({\mathbf{x}},-t){\bm{\times}}{\mathbf{E}}_t({\mathbf{x}},t)] + [{\mathbf{E}}'({\mathbf{x}},-t){\bm{\times}}{\mathbf{B}}_t({\mathbf{x}},t)]\,. \end{split}\end{aligned}$$ One of the authors (B.T.) gratefully acknowledges the financial support from the Swedish Governmental Agency for Innovation Systems (VINNOVA).

ArXiv

--- author: - 'J.P.Pridham[^1]' bibliography: - 'references.bib' title: The deformation theory of representations of the fundamental group of a smooth variety --- Introduction {#introduction .unnumbered} ============ In [@sdc], I developed the theory of Simplicial Deformation Complexes (SDCs), proposed as an alternative to differential graded Lie algebras (DGLAs). They can be constructed for a whole range of deformation problems and capture more information than just the deformation functor. On a more practical level, DGLAs can be extremely useful; they enable Goldman and Millson in [@GM] to show, with certain restrictions, that the hull of the functor describing deformations of a real representation of the fundamental group of a compact Kähler manifold is defined by homogeneous quadratic equations. Even for the problem considered in [@GM], where a perfectly satisfactory governing DGLA was constructed, SDCs can be helpful. Goldman and Millson proceed via a chain of groupoid equivalences from the deformation groupoid to the groupoid associated to the DGLA. However, it is possible to write down an SDC immediately, passing via a natural chain of quasi-isomorphisms to the DGLA in [@GM]. This substantially shortens the reasoning, since it is no longer necessary to find topological interpretations of each intermediate object and functor, and quasi-isomorphisms are quicker to establish than groupoid equivalences. [@GM] was motivated by [@DGMS], which shows that the homotopy type of a compact Kähler manifold is a formal consequence of its cohomology, the idea being to replace statements about DGAs with those about DGLAs. In [@Morgan], Morgan proved an analogue of the results in [@DGMS] for smooth complex varieties. This indicates that the fundamental group of such a variety should have similar properties to those established in [@GM] for compact Kähler manifolds. In Section \[toprep\], I use the theory of SDCs to give a shorter proof of Goldman and Millson’s result. I also prove that the hull of the deformation functor of representations of the (topological) fundamental group of a smooth (non-proper) complex variety has a mixed Hodge structure. Using the weight restrictions established in Hodge II ([@Hodge2]), it follows that the hull is defined by equations of degree at most four. Neither of these results employs the theory of SDCs in an essential way. However, simplicial methods are vital for the final result of the section — that there is a mixed Hodge structure on the hull, even when the variety is not smooth. This has few consequences, since the weight restrictions of Hodge III ([@Hodge3]) give no bound on the degree of the defining equations. The results in [@DGMS] were inspired by the yoga of weights, motivated by the Weil Conjectures. With this in mind, it is reasonable to expect the characteristic zero results to have finite characteristic analogues. Section \[algrep\] is concerned with proving results on the structure of the hull of the functor of deformations of a continuous $l$-adic representation of the algebraic fundamental group of a smooth variety in finite characteristic. This is done by studying the behaviour of the Frobenius action on the hull. As the obstruction maps are Frobenius equivariant, the SDC allows us to describe the hull in terms of cohomology groups. The weight restrictions in Weil II ([@Weil2]) then imply that the hull is quadratic for smooth proper varieties, and defined by equations of degree at most four for smooth, non-proper varieties. Although the two sections of this paper prove similar results by using the same underlying philosphy, namely that weights determine the structure of the hull, they are logically independent. Representations of the Topological Fundamental Group {#toprep} ==================================================== Fix a connected topological space $X$ (sufficiently nice to have a universal covering space) and a point $x \in X$. Denote $\pi_1(X,x)$ by $\Gamma$. Throughout, $A$ will denote a ring in $\C_{\R}$. Let $G$ be a real algebraic Lie group, and fix a representation $\rho_0: \Gamma \to G(\R)$. Given a representation $\rho: \Gamma \to G(A)$, let $$\Bu_{\rho}:=\F_{\tilde{X}}\by_{\Gamma, \rho}G(A)=(\pi_*G(A))^{\Gamma,\rho},$$ where $\F_{\tilde{X}}$ is the sheaf on $X$ whose espace étalé is $\tilde{X}$, and $ \tilde{X}\xra{\pi} X$ is the universal covering space of $X$. The map $$\begin{aligned} \r_A(\rho_0) &\to& \mathfrak{B}_A(\Bu_0)\\ \rho &\mapsto& \Bu_{\rho}\end{aligned}$$ is an equivalence of categories, where $\r_A(\rho_0)$ has as objects representations $$\rho \equiv \rho_0 \mod \m_A,$$ while $\mathfrak{B}_A(\Bu_0)$ has as objects principal $G(A)$-sheaves $$\Bu \equiv \Bu_0 \mod \m_A$$ on $X$. In both cases morphisms must be the identity mod $ \m_A$. This is essentially the same as the opening pages of [@De]. It suffices to construct a quasi-inverse. Given a principal $G(A)$-sheaf $\Bu$ on $X$, $\pi^{-1}\Bu$ will be a principal $G(A)$-sheaf on $\tilde{X}$, so will be a constant sheaf, since $\tilde{X}$ is simply connected. Fix a point $y \in \tilde{X}$ above $x$. We have, for each $\gamma \in \Gamma$, isomorphisms $$\Bu_x \cong (\pi^{-1}\Bu)_y \cong (\pi^{-1}\Bu)_{\gamma y} \cong \Bu_x$$ of principal $G(A)$-spaces, the second isomorphism arising from the constancy of the sheaf. The composite isomorphism can be taken to be right multiplication by $\rho(\gamma)$, thus defining a representation $\rho$. Thus, as in [@sdc] Section \[sdc-smoothgroup\], this deformation problem is governed by the SDC $$E^n=\Hom(u^{-1}\Bu_0,(u^{-1}u_*)^n u^{-1}\Bu_0)^{G(\R)}_{u^{-1}\alpha^n}.$$ The cohomology groups are isomorphic to $$\H^*(X,\bu_0),$$ where $\bu_0$ is the tangent space of $\aut(\Bu_0)$ at the identity. Equivalently, $$\bu_0=\g \by^{G(\R)}\Bu_0= (\pi_* \g)^{\Gamma,\rho_0},$$ for $\g$ the tangent space of $G(\R)$ at the identity, and $G(\R)$ acting via the adjoint action, so the cohomology groups are $$\H^*(X,(\pi_*\g)^{\Gamma,\rho_0}).$$ Recall from [@sdc] Section \[sdc-sdcdgla\] that there is a functor $\cE$ from DGLAs to SDCs. In particular, if $L=\g$, a Lie algebra concentrated in degree $0$, then $$\cE^i(\g)=\ker(G(A) \to G(\R))=\exp(\g \ten \m_A),$$ where $G$ is the corresponding Lie group, with all $\sigma, \pd$ the identity, and $*$ given by the Alexander-Whitney cup product $$e*f = (\pd^{m+n}\ldots \pd^{m+2}\pd^{m+1}e)\cdot (\pd^0)^m f,$$ for $e \in \cE^m,\,f \in \cE^n$. Provided that $\H^1(X,\bu_0)$ is finite dimensional, the deformation functor will have a hull. Observe the quasi-isomorphism of SDCs $$\begin{aligned} \label{bungp} \Gamma(X', (u^{-1}u_*)^n u^{-1} \exp(\bu_0)) &\to& \Hom(u^{-1}(\Bu_0),(u^{-1}u_*)^n u^{-1}(\Bu_0))^{G(\R)}_{u^{-1}\alpha^n},\\ g &\mapsto& (b \mapsto g\cdot b),\end{aligned}$$ where the SDC structure of the term on the left is given by the Alexander-Whitney cup product. Now, write $$\CC^n(X,\sF):= \Gamma(X', (u^{-1}u_*)^n u^{-1}\sF),$$ for sheaves $\sF$ on $X$, and $$\sC^n(\sF):=(u_* u^{-1})^{n+1}\sF,$$ so that $\CC^n(X,\sF)=\Gamma(X, \sC^n(\sF))$. Now assume that $X$ is a differentiable manifold, and let $\sA^n$ be the sheaf of real-valued $\mathrm{C}^{\infty}$ $n$-forms on $X$. We have the following quasi-isomorphisms of SDCs: $$\begin{matrix} \CC^n(X,\exp(\bu_0)) \\ \| \\ \CC^n(X,\cE(\bu_0)^n) \\ \downarrow \\ \CC^n(X,\cE(\bu_0 \ten \sA^{\bullet})^n)\\ \uparrow \\ \cE(\Gamma(X,\bu_0 \ten \sA^{\bullet}))^n, \end{matrix}$$ since on cohomology we have: $$\begin{matrix} \H^n(X,\bu_0) \\ \downarrow \\ \bH^n(X,\bu_0 \ten \sA^{\bullet})\\ \uparrow \\ \H^n(\Gamma(X,\bu_0 \ten \sA^{\bullet})), \end{matrix}$$ the first quasi-isomorphism arising because $\R \to \sA^{\bullet}$ is a resolution, and the second from the flabbiness of $\sA^{\bullet}$. Hence $\r_A(\rho_0)$ is governed by the DGLA $$\Gamma(X,\bu_0 \ten \sA^{\bullet}).$$ Compact Kählerian manifolds {#compactkahler} --------------------------- This section is a reworking of [@GM]. Let $X$ be a compact complex Kählerian manifold. If $\ad(\rho_0)(\Gamma) \subset K$, for some compact subgroup $K \le \GL(\g)$, then we can define $K$-invariant Haar measure on $\g$, and thus a global positive definite inner product on $\bu_0$, so it becomes an orthogonal local system. On $\sA^*$, we have not only the operator $d$, but also $$\dc:=J^{-1}dJ=\sqrt{-1}(\bar{\pd}-\pd),$$ where $J$ is the integrable almost-complex structure on $X$. Define $$\sZ^{\bullet}_{\dc}:= \ker (\dc: \sA^{\bullet} \to \sA^{\bullet}).$$ We now apply Hodge theory for orthogonal local systems, in the form of the $d \dc$ Lemma (identical to [@DGMS] 5.11): On $\Gamma(X, \sA^n \ten \bu_0)$, $$\begin{aligned} \ker(\dc)\cap \im(d)=\im(d \dc),\\ \ker(\d)\cap \im(\dc)=\im(d \dc).\end{aligned}$$ It is worth noting that the $d\dc$ lemma holds on any manifold which can be blown up to a compact Kähler manifold, hence on any compact Moishezon manifold, so we may take $X$ to be of this form. Of course, this gives no more information about fundamental groups, since blowing-up does not change the fundamental group. We thus obtain quasi-isomorphisms of DGLAs: $$\begin{matrix} \Gamma(X,\bu_0 \ten \sA^{\bullet})\\ \uparrow \\ \Gamma(X,\bu_0 \ten \sZ_{\dc}^{\bullet})\\ \downarrow \\ (\H_{\dc}^*(\bu_0\ten \sA^{\bullet}),d=0)\\ \| \\ (\H^*(X,\bu_0),0). \end{matrix}$$ To see that these are quasi-isomorphisms, we look at the cohomology groups: $$\begin{matrix} \H^n(\Gamma(X,\bu_0 \ten \sA^{\bullet}))\\ \uparrow \\ \H^n(\Gamma(X,\bu_0 \ten \sZ_{\dc}^{\bullet},d))\\ \downarrow \\ (\H^n_{\dc}(\Gamma(X,\bu_0\ten \sA^{\bullet})). \end{matrix}$$ The last two quasi-isomorphisms are a consequence of the $d\dc$ Lemma. However, the DGLA $(H^*(\bu_0),d=0)$ is formal, in the sense of [@Man], so the hull of the deformation functor is given by homogeneous quadratic equations. Note that a hull $S$ must exist, since $\dim \H^1(\bu_0) < \infty$. Explicitly, $$S \cong \R[[\H^1(\bu_0)^{\vee}]]/(\check{\cup}\,\H^2(\bu_0)^{\vee}),$$ where $$\check{\cup}\,:\H^2(\bu_0)^{\vee} \to \SS^2(\H^1(\bu_0)^{\vee})$$ is dual to the cup product. Note further that we have a Hodge decomposition on $S \ten \Cx$, where $S$ is the hull. Since $\m_S/\m_S^2$ is pure of weight $-1$, $\m_S^n/\m_S^{n+1}$ is pure of weight $-n$, so must have even dimension for $n$ odd. Smooth complex varieties {#smoothcx} ------------------------ In this case we have to use the mixed Hodge theory of [@Hodge2], rather than pure Hodge theory. Let $X$ be a smooth complex variety. Using [@Nagata] and [@Hironaka], we may write $X=\bar{X} - D$, for $\bar{X}$ a complete smooth variety, and $D$ a divisor with normal crossings. Again, the governing DGLA is $$\Gamma(X,\bu_0 \ten \sA^{\bullet}).$$ If $\ad(\rho_0)(\Gamma) \subset K$, for some compact subgroup $K \le \GL(\g)$, then we can define $K$-invariant Haar measure on $\g$, and thus a global positive definite inner product on $\bu_0$, so that it becomes an orthogonal local system. Where [@GM] mimicked [@DGMS], we must now mimic [@Morgan], and will adopt the notation of that paper. Observe that [@Morgan] §2 works for twisted coefficients (in our case $\bu_0$), since $\bu_0$ is trivial on each simplex, giving us a quasi-isomorphism of DGLAs $$\Einf(X, \bu_0) \to \Gamma(X,\bu_0 \ten \sA^{\bullet}),$$ There is then a quasi-isomorphism $$\Einf(X, \bu_0)\ten \Cx \to (\Gamma(\bar{X},\widetilde{\bu_0}\ten \Omega_{\bar{X}}^{*}(\log D)\ten \sA_{\Cx}^{0,*}), \nabla),$$ where $(\widetilde{\bu_0}, \nabla)$ is the holomorphic Deligne extension of $\bu_0$, as defined in [@De]. The mixed Hodge theory of [@Timm] now proves that the latter quasi-isomorphism is a mixed Hodge diagram, in the sense of [@Morgan] 3.5 (except that we work with DGLAs rather than DGAs). In [@Sullivan], Sullivan defines the minimal model of a connected DGA. Replacing free graded algebras by free graded Lie algebras yields the same construction for a DGLA $L$ in non-negative degrees with $\H^0(L)=0$. The minimal model $M\to L$ will then in some sense be universal in the quasi-isomorphism class of $L$. If we only require quasi-isomorphisms in degrees $\ge 1$, then we can analogously construct a minimal model $M \to L$, for which automatically $\H^0(M)=0$. Let $M$ denote the Sullivan minimal model of $ \Gamma(X,\bu_0 \ten \sA^{\bullet})$. By [@Morgan] Theorems 6.10 and 8.6, there is a mixed Hodge structure on $M$. By [@Morgan] Theorem 10.1, this induces a weight decomposition on $M$. \[morthm\] The hull $S$ of the functor $\ddef_{\rho_0}$ is defined by equations of degree at most $4$. Moreover, there is a mixed Hodge structure on $S$, with $\m_S/\m_S^2$ of weights $-1$ and $-2$. Choose a decomposition $$M^1=\H^1(M)\oplus C^1, \quad \z^2(M)= \cH^2 \oplus \b^1(M), \quad M^2= \z^2(M)\oplus C^2,$$ respecting the filtrations $(F,W)$ on $M$ (this is straightforward, since $F,\bar{F},W$ give a bigraded decomposition of $M\ten \Cx$). The hull of $\ddef_M$ is then given by the Kuranishi functor as defined in [@Man] §4, which will now be pro-represented by $$\R[[\H^1(M)^{\vee}]]/(f\cH^2(M)^{\vee}),$$ where $$f:\cH^2(M)^{\vee} \to (\H^1(M)^{\vee})^2 \triangleleft \R[[\H^1(M)^{\vee}]]$$ respects the filtrations. By [@Timm] §6, $\H^1(X,\bu_0)$ is of weights $1$ and $2$, while $\H^2(X,\bu_0)$ is of weights $2, 3$ and $4$. Thus $f$ corresponds to functions: $$\begin{aligned} (W_{-1})^2 &\text{weight }-2 \\ (W_{-1})^3 +(W_{-1})(W_{-2}) &\text{weight }-3\\ (W_{-1})^4 +(W_{-1})^2(W_{-2}) +(W_{-2})^2 &\text{weight }-4,\end{aligned}$$ hence equations of degree at most $4$. Moreover, $f$ respects the Hodge structure, so we have a mixed Hodge structure on $S$. Note that the quotient map $$f:\H^2(M)^{\vee} \to (\H^1(M)^{\vee})^2/(\H^1(M)^{\vee})^3 \cong S^2(\H^1(M)^{\vee})$$ is dual to half the cup product $$\begin{aligned} \H^1(X,\bu_0) &\to& \H^2(X,\bu_0),\\ a &\mapsto& \half a\cup a.\end{aligned}$$ \[hodgedef\] It ought to be possible to prove this result without recourse to minimal models and the methods of [@Morgan]. The following approach should work: 1. Define a category $\mathrm{Hdg}(\C_{\R})$ of real Artinian algebras, with mixed Hodge structures on their maximal ideals. Show that Schlessinger’s theorems carry over into this context (in particular, tangent and obstruction spaces will have mixed Hodge structures). 2. Given a filtration $F$ on a complex DGLA $L$, and an algebra $A \in \mathrm{Hdg}(\C_{\R})$, define $$\begin{aligned} F \mcl(A)&:=& \mcl(A\ten \Cx)\cap F^0(L^1\ten \m_A)\\ F \gl(A)&:=& \exp (d^{-1} F^0(L^1\ten \m_A))\\ F \defl(A)&:=& F \mcl(A)/F \gl(A).\end{aligned}$$ We have similar definitions for a filtration $\bar{F}$ on a complex DGLA, and for a filtration $W$ on a real DGLA. 3. Consider the DGLA $$N=(\Gamma(\bar{X},\widetilde{\bu_0}\ten \Omega_{\bar{X}}^{*}(\log D)\ten \sA_{\Cx}^{0,*}), \nabla),$$ defined as above. This has filtrations $W$ and $F$. We have quasi-isomorphisms (of $W$-filtered DGLAs) $$N \to \Gamma(X,\bu_0 \ten \sA^{\bullet})\ten \Cx \leftarrow \bar{N}.$$ Writing $L=\Gamma(X,\bu_0 \ten \sA^{\bullet})$, we can use these quasi-isomorphisms to identify the functors $\ddef_N, \ddef_{\bar{N}}$ and $\ddef_{L\ten \Cx}$. Now, consider the functor on $\mathrm{Hdg}(\C_{\R})$ given by $$A \mapsto W \defl(A) \cap F\defn(A\ten \Cx) \cap \bar{F}\ddef_{\bar{N}}(A\ten \Cx),$$ the intersection being as subsets of $\defl(A\ten \Cx)$. This functor should have a hull $S$, with tangent space $\H^1(X,\bu_0)$ and obstruction space $\H^2(X,\bu_0)$ (with the usual Hodge structures). It would remain only to show that $S$ (without its Hodge structures) is a hull for $\defl$ over $\C_{\R}$. Arbitrary complex varieties {#arbcx} --------------------------- In this section, the methods of [@Hodge3] will be used. Let $X$ be a complex variety. As in [@Hodge3] 8.1.12, we take a simplicial resolution $Y_{\bullet} \to X$ of $X$ by smooth varieties, and compactify to obtain a smooth proper simplicial variety $\overline{Y}_{\!\!\bullet}$, and a divisor $D_{\bullet}$ of normal crossings in $\overline{Y}_{\!\!\bullet}$, such that $Y_{\bullet}=\overline{Y}_{\!\!\bullet} - D_{\bullet}$. Now, the SDC governing deformations of $\rho_0$ is $$\CC^n(X, \exp(\bu_0)).$$ Cohomological descent implies that we have a quasi-isomorphism of SDCs: $$\CC^n(X, \exp(\bu_0)) \to \CC^n(Y_n, \exp(\bu_0)).$$ We now proceed as in the previous section, obtaining quasi-isomorphisms $$\begin{matrix} \CC^n(Y_n, \exp(\bu_0)) \\ \downarrow\\ \CC^n(Y_n,\cE(\bu_0 \ten \sA_{Y_n}^{\bullet})^n) \\ \uparrow\\ \cE(\Gamma(Y_n,\bu_0 \ten \sA_{Y_n}^{\bullet}))^n\\ \uparrow\\ \cE(\Einf(Y_n, \bu_0))^n . \end{matrix}$$ As in [@Morgan] §3, there is a quasi-isomorphism $$\Einf(Y_n, \bu_0)\ten \Cx \to (\Gamma(\overline{Y}_{\!\!n},\widetilde{\bu_0}\ten \Omega_{\overline{Y}_{\!\!n}}^{*}(\log D_n)\ten \sA_{\Cx}^{0,*}), \nabla),$$ with $\nabla$ as in [@Timm]. We have a weight filtration $W$ on the former, and a compatible pair $(W,F)$ of filtrations on the latter. In [@Hodge3] §7, Deligne defines filtrations $F,\delta(W,L)$ on the total complex of a bigraded cochain complex. Since we are working simplicially, we must, via the Eilenberg-Zilber Theorem, define the corresponding filtration on the diagonal complex of a bi-cosimplicial complex. As in [@sdc] Section \[sdc-explicit\], we form the cosimplicial Lie algebras $$\mathpzc{e}(\Einf(Y_n, \bu_0))\ten \Cx \to \mathpzc{e}(\Gamma(\overline{Y}_{\!\!n},\widetilde{\bu_0}\ten \Omega_{\overline{Y}_{\!\!n}}^{*}(\log D_n)\ten \sA_{\Cx}^{0,*}), \nabla),$$ Now, we have $$\mathpzc{e}(\Einf(Y_n, \bu_0))^n = \diag\mathpzc{e}(\Einf(Y_m, \bu_0))^n,$$ and similarly for the other complex, on which we wish to define filtrations. Given a cosimplicial complex $K$ of vector spaces, the Eilenberg-Zilber map $$(\diag K)^n= K^{nn} \xra{\nabla} \bigoplus_{p+q=n} K^{pq} =(\mathrm{Tot} K)^n$$ induces an isomorphism on cohomology. Moreover, if $$\alpha \in N^n(\diag K)= \bigcap_{i=0,\ldots,n} \ker(\sigmah^i\sigmav^i: (\diag K)^n \to (\diag K)^{n-1}),$$ then $\nabla(\alpha)_{pq} \in N^p_{\mathrm{h}}\cap N^q_{\mathrm{v}}$. For a DGLA $L$, observe that $N(\mathpzc{e}(L))=L$. We therefore define $$\delta(W,L)_n(N(\diag \mathpzc{e}(\Einf(Y_m, \bu_0))^n)) = \nabla^{-1} \delta(W,L)_n (\mathrm{Tot} \Einf(Y_{\bullet}, \bu_0)^{\bullet}),$$ the latter being defined as in [@Hodge3] §7. Similarly, we obtain filtrations $\delta(W,L)$ and $F$ on $$N(\mathpzc{e}(\Gamma(\overline{Y}_{\!\!n},\widetilde{\bu_0}\ten \Omega_{\overline{Y}_{\!\!n}}^{*}(\log D_n)\ten \sA_{\Cx}^{0,*}))).$$ Now, given a cosimplicial complex $K^*$, and a filtration $\mathrm{Fil}$ on $N(K)$, define $$\mathrm{Fil}_m(K^n)= \sum_{0 \le j_1 < \ldots < j_t< n} (\sigma^{j_1}\ldots \sigma^{j_t})^{-1} \mathrm{Fil}_m(N^{n-t}(K)).$$ The reasoning behind these constructions is that they preserve the filtrations induced on cohomology. We have therefore defined filtrations $\delta(W,L),F$ (as applicable) on $$\mathpzc{e}(\Einf(Y_n, \bu_0))\ten \Cx \to \mathpzc{e}(\Gamma(\overline{Y}_{\!\!n},\widetilde{\bu_0}\ten \Omega_{\overline{Y}_{\!\!n}}^{*}(\log D_n)\ten \sA_{\Cx}^{0,*}), \nabla),$$ Exponentiation induces corresponding filtrations on $$\cE(\Einf(Y_n, \bu_0)\ten \Cx) \to \cE(\Gamma(\overline{Y}_{\!\!n},\widetilde{\bu_0}\ten \Omega_{\overline{Y}_{\!\!n}}^{*}(\log D_n)\ten \sA_{\Cx}^{0,*}), \nabla),$$ and hence (by functoriality) on $$\cL(\cE(\Einf(Y_n, \bu_0)))\ten \Cx \to \cL(\cE(\Gamma(\overline{Y}_{\!\!n},\widetilde{\bu_0}\ten \Omega_{\overline{Y}_{\!\!n}}^{*}(\log D_n)\ten \sA_{\Cx}^{0,*}))).$$ Explicitly, we obtain these filtrations by considering the cofiltrations induced on the rings pro-representing these SDCs and DGLAs ($\cQ_*$ and $\cT_{\bullet}$ in the notation of [@sdc] Section \[sdc-sdcdgla\]). The only technicality to check is that the resulting filtrations on the Lie algebra do indeed respect the Lie bracket (i.e. $[F^i,F^j] \subset F^{i+j}$). This is done by observing that we can recover the Lie bracket from the product (using a formula in some sense inverse to the Campbell-Baker-Hausdorff formula), giving a comultiplicativity condition on the cofiltrations satisfied by $\rho=*^{\sharp}$, and preserved by the functor $\cE$. It then follows from [@Hodge3] that this is a mixed Hodge diagram in the sense of [@Morgan], so we proceed as before. \[arbcxthm\] Let $S$ be the hull of the functor $\ddef_{\rho_0}$. Then $$S \cong \R[[\H^1(X,\bu_0)^{\vee}]]/(f(\H^2(X,\bu_0)^{\vee})),$$ where $$f:\H^2(X,\bu_0)^{\vee} \to (\H^1(X,\bu_0)^{\vee})^2 \triangleleft \R[[\H^1(X,\bu_0)^{\vee}]]$$ preserves the mixed Hodge structures. In fact, the quotient map $$f:\H^2(X,\bu_0)^{\vee} \to (\H^1(X,\bu_0)^{\vee})^2/(\H^1(X,\bu_0)^{\vee})^3 \cong S^2(\H^1(X,\bu_0)^{\vee})$$ is dual to half the cup product $$\begin{aligned} \H^1(X,\bu_0) &\to& \H^2(X,\bu_0),\\ a &\mapsto& \half a\cup a.\end{aligned}$$ As shown above, $\ddef_{\rho_0}$ is governed by the SDC $\cE(\Einf(Y_n, \bu_0))^n$, hence by the DGLA $\cL(\cE(\Einf(Y_n, \bu_0)))$. The proof is now identical to Theorem \[morthm\]. Note that since weight $0$ is permitted in both $\H^1(X,\bu)$ and $\H^2(X,\bu)$, we can draw no conclusions concerning the defining equations. \[subfields\] As in [@Morgan], we may replace $\R$ by any subfield $k$, taking $G/k$ algebraic. In fact, using Hodge theory for unitary local systems (as in [@Timm]), we may replace $\R$ by $\Cx$. As before, we obtain a bigraded decomposition on the complex cohomology, the only difference being that we must define the filtrations $F,\bar{F}$ separately, as complex conjugation no longer makes sense. In the general case, a sufficient hypothesis for $\bu_0$ to be a unitary local system becomes that $$\ad(\rho_0)(\Gamma) \subset K,$$ for some $K \le \GL(\g)$ a compact Lie group. Indeed, we may take any subfield $k\subset \Cx$ instead of $\Cx$, giving a bigraded decomposition on complex cohomology, for which the correponding weight decomposition will descend to $k$. A much cleaner proof would be possible if the approach suggested in Remark \[hodgedef\] were developed, since the filtrations on the SDCs $$\cE(N_n)^n \to \cE(\Gamma(Y_n,\bu_0 \ten \sA^{\bullet}_{Y_n})\ten \Cx)^n \leftarrow \cE(\bar{N_n})^n,$$ where $$N_n=(\Gamma(\bar{Y_n},\widetilde{\bu_0}\ten \Omega_{\bar{Y_n}}^{*}(\log D_n)\ten \sA_{\Cx}^{0,*}), \nabla),$$ allow us to define filtered deformation functors, with the same definitions as for DGLAs. This method would avoid having to make use of the noisome functor $\cL$. Representations of the Algebraic Fundamental Group {#algrep} ================================================== Let $k=\bF_q$, fix a connected variety $X_k/k$, and let $X= X_k \ten_k \bar{k}$. Throughout, we will assume that $l$ is a prime not dividing $q$. Fix a closed point $x$ of $X$, and denote the associated geometric point $x \to X$ by $\bar{x}$. The Weil group $W(X_k, \bar{x})$ is defined by: $$\xymatrix{ 1 \ar[r] &\pi_1(X_k,\bar{x}) \ar[r] & \pi_1(X,\bar{x}) \ar[r] &\hat{\Z} \ar[r] & 0 \\ 1 \ar[r] & W(X_k, \bar{x}) \ar[r] \ar@{^{(}->}[u] & \pi_1(X,\bar{x}) \ar@{=}[u]\ar[r] &\Z \ar@<-2pt>@{^{(}->}[u] \ar[r] & 0, }$$ with both rows exact. Each closed point $y \in |X_0|$ gives rise to a Frobenius element $\phi_y \in W(X,x)$, defined up to conjugation, and a representation of $W(X_k, \bar{x})$ is said to be pure of weight $n$ if the eigenvalues of all the Frobenius elements are algebraic, with all their complex conjugates of norm $q^{n/2}$. $\pi_1(X,\bar{x})$ and $W(X_k, \bar{x})$ are both topological groups, the former with the profinite topology, and the latter as a subgroup of $\pi_1(X_k,\bar{x})$, which has the profinite topology. Representations to $\GL_n(\Q_l)$ -------------------------------- Fix a continuous representation $$\rho_0:\pi_1(X,\bar{x}) \to \GL_n(\Q_l).$$ We define the functor $$\ddef_{\rho_0}:\C_{\Q_l} \to \Set$$ by setting $\ddef_{\rho_0}(A)$ to be the set of isomorphism classes of continuous representations deforming $\rho_0$. We will first need to generalise the notion of constructible $\Q_l$-sheaf, as in [@Weil2] 1.1.1. For $A \in \C_{\Q_l}$, the category of constructible locally free $A$-sheaves will have as objects locally free $A$-sheaves $\sF$ which are constructible as $\Q_l$-sheaves in the sense of [@Weil2] 1.1.1. It is well known that there is an equivalence of categories between continuous $\Q_l$-representations of the fundamental group, and constructible locally free $\Q_l$-sheaves. Let $\vv_0$ be the sheaf corresponding to $\rho_0$, with $\ww_0$ the underlying $\Z_l$-sheaf. \[vv\] For $A \in \C_{\Q_l}$, there is a functorial equivalence of groupoids between $\mathfrak{R}(A)$, the groupoid of continuous representations $\rho:\pi_1(X,\bar{x}) \to \GL_n(A)$ deforming $\rho_0$, and $\mathfrak{V}(A)$, the groupoid of rank $n$ constructible locally free $A$-sheaves deforming $\vv_0$. The representation $\rho: \pi_1(X,\bar{x}) \to \GL_n(A)$ is continuous, so has compact image. Since $A$ is a finite dimensional vector space over $\Q_l$, [@Se] LG 4 Appendix 1 provides the existence of a $\Z_l$-lattice $W$ generating $A^n$ such that $\rho$ factorises through $\GL(W)$. As in [@sdc] Section \[sdc-gln\], this gives a constructible $\Z_l$-sheaf $\ww$ on $X$. The $A$-module structure of $W\ten_{\Z_l} \Q_l$ now provides $\vv:=\ww\ten_{\Z_l} \Q_l$ with the structure of a constructible $A$-sheaf. To construct a quasi-inverse, start with a constructible $A$-sheaf $\vv=\ww\ten_{\Z_l} \Q_l$. Let $W:=\ww_{\bar{x}}$ and $V=W \ten_{\Z_l} \Q_l$. Then, as in [@sdc] Section \[sdc-gln\], we obtain a representation $$\rho:\pi_1(X,\bar{x}) \to \GL(W) \into \GL(V).$$ We now choose an isomorphism $V \cong A^n$ compatible with the canonical isomorphism $(\vv_0)_{\bar{x}}\cong \Q_l^n$. For a constructible locally free $\Z_l$-sheaf $\sF$, define $$\CC^n(X,\sF):=\lim_{\substack{\longleftarrow \\m}} \CC^n(X,\sF/l^m\sF),$$ where $$\CC^n(X,\sF):= \Gamma(X', (u^{*}u_*)^n u^{*}\sF),$$ for sheaves $\sF$ on $X$, and $$\sC^n(\sF):=(u_* u^{*})^{n+1}\sF,$$ so that $\CC^n(X,\sF)=\Gamma(X, \sC^n(\sF))$. For a constructible locally free $\Q_l$-sheaf $\sF \ten \Q_l$ of finite rank, define $$\begin{aligned} \sC^n(\sF\ten_{\Z_l} \Q_l)&:=&\sC^n(\sF)\ten_{\Z_l} \Q_l,\\ \CC^n(X,\sF\ten_{\Z_l} \Q_l)&:=&\CC^n(X,\sF)\ten_{\Z_l} \Q_l.\end{aligned}$$ Note that this construction is independent of the choice of $\sF$, since $\sF$ is of finite rank. \[defglnql\] Deformations of $\rho_0$ are described by the SDC $$E^n(A)=\exp(\CC^n(X,\ENd(\vv_0))\ten_{\Q_l}\m_A),$$ with product given by the Alexander-Whitney cup product. Given $\omega \in \mc_G(A)$, define $$\vv_{\omega}:= \{ v \in \sC^0(\vv_0 \ten_{\Q_l} A): \alpha_{\sC^0(\vv_0 \ten A)}(v)=\omega \cup v\},$$ where, for any sheaf $\sF$ on $X$, $\alpha_{\sF}$ is the canonical map $\alpha_{\sF}: \sF \to \sC^0(\sF)$. Since $\m_A$ is a finite dimensional $\Q_l$ vector space, with nilpotent product, we may find a multiplicatively closed lattice $I \subset \m_A$ with $$\omega \in \exp(\CC^n(X,\ENd(\ww_0))\ten_{\Z_l} I).$$ As in [@sdc] Section \[sdc-etshf\], this gives a constructible $(\Z_l\oplus I)$-sheaf $\ww_{\omega}$, and so is a constructible $A$-sheaf. Conversely, given a constructible $A$-sheaf $\vv$ deforming $\vv_0$, let $\ww$ be an underlying $\Z_l$-sheaf. We choose an isomorphism $\sC^0(\vv) \cong \sC^0(\vv_0)\ten_{\Q_l} \m_A$. The transition functions of $\ww$ then provide a suitable $\omega \in \mc_G(A)$. Equivalently, we may note that $L^n:=\CC^n(X,\ENd(\vv_0))$ has the structure of a DGAA (cup product being associative), with differential $d=\sum(-1)^i\pd^i$, so *a fortiori* a DGLA. Given $\omega \in \mcl(A)$, let $$\vv_{\omega}:=\ker(d + \omega \cup:\sC^0(\vv_0)\ten_{\Q_l} A \to \sC^1(\vv_0)\ten_{\Q_l} A).$$ That this gives the required deformation functor follows by observing that we have isomorphisms on tangent and obstruction spaces. \[glnqlcoho\] If $\H^i(X,\ENd(\ww_0))$ and $\H^{i-1}(X,\ENd(\ww_0))$ are finitely generated, then $$\H^i(E)=\H^i(X,\ENd(\vv_0)).$$ $$\begin{aligned} \H^i(E)&=& \H^i(\CC^{\bullet}(X,\ENd(\vv_0)))\\ &=& \H^i(\CC^{\bullet}(X,\ENd(\ww_0)))\ten_{\Z_l}\Q_l\\ &=& \H^i( \lim_{\substack{\longleftarrow \\n}} \CC^{\bullet}(X,\ENd(\ww_0)\ten \Z/l^n))\ten_{\Z_l}\Q_l\\ &=& \H^i( \lim_{\substack{\longleftarrow \\_n}} \CC^{\bullet}(X,\ENd(\ww_0))\ten \Z/l^n)\ten_{\Z_l}\Q_l.\end{aligned}$$ Now, the tower $$\cdots \to \CC^{\bullet}(X,\ENd(\ww_0))\ten \Z/l^{n+1} \to \CC^{\bullet}(X,\ENd(\ww_0))\ten \Z/l^n \to \cdots$$ clearly satisfies the Mittag-Leffler condition, so, by [@W] Theorem 3.5.8, we have the exact sequence: $$\xymatrix@=3ex{ 0 \ar[r] & \Lim^1 \H^{i-1}(\CC^{\bullet}(X,\ENd(\ww_0))_n) \ar[r] \ar@{=}[d] & \H^i(\Lim_n \CC^{\bullet}(X,\ENd(\ww_0))_n) \ar[r] \ar@{=}[d]& \Lim_n \H^i(\CC^{\bullet}(X,\ENd(\ww_0))_n) \ar[r] \ar@{=}[d]& 0\\ 0 \ar[r]& \Lim^1 \H^{i-1}(X, \ENd(\ww_0)_n) \ar[r]& \H^i( \Lim_n \CC^{\bullet}(X,\ENd(\ww_0))_n) \ar[r]& \H^i(X,\ENd(\ww_0)) \ar[r]& 0. }$$ From [@Mi] Lemma V 1.11, it follows that the $\H^{i-1}(X, \ENd(\ww_0)\ten \Z/l^n)$ are finitely generated $\Z/l^n$-modules, so satisfy DCC on submodules, so this inverse system satisfies the Mittag-Leffler condition, making the $\Lim^1$ on the left vanish. Thus $$\H^i(E)=\H^i(X,\ENd(\ww_0))\ten_{\Z_l}\Q_l= \H^i(X,\ENd(\vv_0)).$$ Representations to an arbitrary group variety over $\Q_l$ {#arbql} --------------------------------------------------------- Let $G$ be a group variety over $\Q_l$, with associated Lie algebra $\g$. We will consider deformations over $\C_{\Q_l}$. Given $A \in \C_{\Q_l}$, $G(A)$ has the structure of an $l$-adic Lie group. Fix a continuous representation $$\rho_0:\pi_1(X,\bar{x}) \to G(\Q_l).$$ We define the functor $$\ddef_{\rho_0}:\C_{\Q_l} \to \Set$$ by setting $\ddef_{\rho_0}(A)$ to be the set of isomorphism classes of continuous representations deforming $\rho_0$. Given a pro-$l$ group $K$, define a constructible principal $K$-sheaf to be a principal $K$-sheaf $\bD$, such that $$\bD=\lim_{\substack{\longleftarrow \\{K \to F \text{ finite}}}} F \by^K \bD.$$ Given an $l$-adic Lie group $G$, a constructible principal $G$-sheaf is a $G$-sheaf $\bB$ for which there exists a constructible principal $K$-sheaf $\bD$, for some $K \le G$ compact, with (observe that compact and totally disconnected is equivalent to pro-finite). For $A \in \C_{\Q_l}$, there is a functorial equivalence of groupoids between $\mathfrak{R}(A)$, the groupoid of continuous representations $\rho:\pi_1(X,\bar{x}) \to G(A)$ deforming $\rho_0$, and $\mathfrak{B}(A)$, the groupoid of constructible principal $G(A)$-sheaves deforming $\Bu_0$. Similar to Lemma \[vv\]. Since $\ad \rho_0(\pi_1(X,\bar{x})) \le \GL(\g)$ is compact, the corresponding sheaf $\bu_0$ is a constructible $\Q_l$-sheaf of Lie algebras. Hence the sheaves $\sC^n(\bu_0)$ are sheaves of Lie algebras. Deformations of $\rho_0$ are described by the SDC $$E^n(A)=\exp(\CC^n(X,\bu_0)\ten_{\Q_l}\m_A),$$ with product given by the Alexander-Whitney cup product. As in Theorem \[defglnql\] and [@sdc] Section \[sdc-smoothgroup\]. If we write $\bB=G\by^K\bD$, for some compact $K \le G$, and $\H^i(X,\ad \bD_0)$ and $\H^{i-1}(X,\ad \bD_0)$ are finitely generated, then $$\H^i(E)=\H^i(X,\ad \bB_0).$$ As for Lemma \[glnqlcoho\]. Structure of the hull {#weilhull} --------------------- Given a mixed Weil sheaf $\sF$ over $\Q_l$ on $X$, for $X$ either smooth or proper, [@Weil2] shows that all the eigenvalues of Frobenius acting on the cohomology group $\H^i(X,\sF)$ are algebraic numbers $\alpha$, and for each $\alpha$, there exists a weight $n$, such that all complex conjugates of $\alpha$ have norm $q^{n/2}$. This provides us with a weight decomposition $$\H^i(X,\sF)=\bigoplus_n \cW_n \H^i(X,\sF).$$ \[frobhull\] Let $\rho_0: \pi_1(X,\bar{x}) \to G(\Q_l)$ be a continuous representation. Assume that $$\ad(\rho_0): \pi_1(X,\bar{x}) \to \Aut(\g),$$ where $\Aut$ here denotes Lie algebra automorphisms, extends to a continuous mixed representation $$\widetilde{\ad(\rho_0)}: W(X_k,x) \to \Aut(\g).$$ Then the deformation functor $$\ddef_{\rho_0}:\C_{\Q_l} \to \Set$$ has a hull of the form: $$\Q_l[[\H^1(X,\bu_0)^{\vee}]]/(f(\H^2(X,\bu_0)^{\vee})),$$ where $$f:\H^2(X,\bu_0)^{\vee} \to (\H^1(X,\bu_0)^{\vee})^2 \triangleleft \Q_l[[\H^1(X,\bu_0)^{\vee}]]$$ preserves the Frobenius decompositions. In fact, the quotient map $$f:\H^2(X,\bu_0)^{\vee} \to (\H^1(X,\bu_0)^{\vee})^2/(\H^1(X,\bu_0)^{\vee})^3 \cong S^2(\H^1(X,\bu_0)^{\vee})$$ is dual to half the cup product $$\begin{aligned} \H^1(X,\bu_0) &\to& \H^2(X,\bu_0),\\ a &\mapsto& \half a\cup a.\end{aligned}$$ As in [@Weil2], the representation of the Weil group correponds to an isomorphism $\Phi:F^*\bu_0 \to \bu_0$, where $F$ is the Frobenius endomorphism of $X$ over $\bF_q$. Since $\ad \rho_0$ is a representation to Lie algebra automorphisms, $\Phi$ is an isomorphism of constructible sheaves of Lie algebras. Hence we have the following morphisms of SDCs: $$\exp(\CC^n(X,\bu_0)\ten_{\Q_l}\m_A) \xra{F^*} \exp(\CC^n(X,F^*\bu_0)\ten_{\Q_l}\m_A) \xra{\CC^n(X,\Phi)} \exp(\CC^n(X,\bu_0)\ten_{\Q_l}\m_A).$$ We therefore have a Frobenius endomorphism on our SDC. We could now proceed as in [@Weil2] §5, by forming the associated DGLA (as in [@sdc] Section \[sdc-sdctodgla\]), which will, by functoriality, also have a Frobenius endomorphism. We would then form the minimal model, as in Section \[smoothcx\], and construct a weight decomposition and quasi-isomorphism to a formal DGLA, as in [@Weil2] Corollary 5.3.7. Instead, we will take a more direct approach, involving neither DGLAs nor minimal models. By abuse of notation, write $F^*$ for the Frobenius endomorphism on $E^{\bullet}$ given by the composition above. Since $F^*$ respects the structures, this gives us an endomorphism $F^*: \ddef_E \to \ddef_E$. Let $R$ be the hull of $\ddef_E$. We have: $$\xymatrix{ h_R \ar[r]^-{\etale} \ar@{-->}[d]_{\widetilde{F^*}} & \ddef_E \ar[d]^{F^*} \\ h_R \ar[r]^-{\etale}& \ddef_E, }$$ for some (non-unique) lift $\widetilde{F^*}$. Now let $\tilde{F}$ be the dual morphism $\tilde{F}:R \to R$ of complete local $\Q_l$-algebras. We have canonical isomorphisms $$(\m_R/\m_R^2)^{\vee} \cong t_{\ddef_E} \cong \H^1(X,\bu_0),$$ compatible with Frobenius. Since $\m_R/\m_R^n$ is a finite dimensional vector space over $\Q_l$ for all $n$, we may use the Jordan decomposition (over $\bar{\Q_l}$) successively to lift the cotangent space $\m_R/\m_R^2$ to $\m_R/\m_R^n$, obtaining a space of generators $V \subset \m_R$, with the map $V \to \m_R/\m_R^2$ an isomorphism preserving the Frobenius decompositions. Now look at the map $$S=\Q_l[[V]] \onto R;$$ let its kernel be $J \subset (V)^2$. $(J/VJ)^{\vee}$ is then a universal obstruction space for $h_R$. Since is smooth, and $\H^2(X, \bu_0)$ is an obstruction space for $\ddef_{\rho_0}$, there is a unique injective map of obstruction spaces ([@Man] Proposition 2.18) $$(J/VJ)^{\vee} \into \H^2(X, \bu_0).$$ Observe that both of these obstruction theories are Frobenius equivariant: given a functor $H$ on which $F$ (Frobenius) acts, we say an obstruction theory $(W,w_e)$ is Frobenius equivariant if there is a Frobenius action on the space $W$ such that for every small extension $$e:\quad 0 \to I \to B \to A \to 0,$$ and every $h \in H(A)$, we have $w_e(Fh)=Fw_e(h)$. Since $h_R \to \ddef_E$ commutes with Frobenius, the corresponding map of obstruction spaces does (by uniqueness). As before, we may use the Jordan decomposition on the successive quotients to obtain a space of generators $W \subset J$, such that $W \to J/VJ$ is an isomorphism preserving the Frobenius decompositions. Hence we have maps $$\H^2(X, \bu_0)^{\vee} \onto J/VJ \cong W \into \Q_l[[V]],$$ and $$V \cong \m_R/\m_R^2 \cong \H^1(X, \bu_0)^{\vee},$$ preserving the Frobenius decompositions. Let $f:\H^2(X, \bu_0)^{\vee} \to \Q_l[[\H^1(X, \bu_0)^{\vee}]]$ be the composition. Then $$R=\Q_l[[V]]/J \cong \Q_l[[\H^1(X, \bu_0)^{\vee}]]/(f(\H^2(X, \bu_0)^{\vee})),$$ as required. It only remains to prove the statement about the cup product, which follows since half the cup product is the primary obstruction map. Let $X$ be smooth and proper, $\rho_0$ as above, and assume that $ \widetilde{\ad(\rho_0)} $ is pure of weight $0$. Then the deformation functor $$\ddef_{\rho_0}:\C_{\Q_l} \to \Set$$ has a hull $R$ defined by homogeneous quadratic equations (given by the cup product). This follows since, under these hypotheses, [@Weil2] Corollaries 3.3.4–3.3.6 imply that $\H^1(X,\bu_0)$ is pure of weight $1$, and $\H^2(X,\bu_0)$ is pure of weight $2$. Let $X$ be smooth, $\rho_0$ as above, and assume that $ \widetilde{\ad(\rho_0)} $ is pure of weight $0$. Then the deformation functor $$\ddef_{\rho_0}:\C_{\Q_l} \to \Set$$ has a hull defined by equations of degree at most four. This follows since, under these hypotheses, [@Weil2] Corollaries 3.3.4–3.3.6 imply that $\H^1(X,\bu_0)$ is of weights $1$ and $2$, while $\H^2(X,\bu_0)$ is of weights $2, 3$ and $4$. Hence the image of $f:\H^2(X,\bu_0)^{\vee} \to \Q_l[[\H^1(X,\bu_0)^{\vee}$ gives equations: $$\begin{aligned} (\cW_{-1})^2 &\text{weight }-2 \\ (\cW_{-1})^3 +(\cW_{-1})(\cW_{-2}) &\text{weight }-3\\ (\cW_{-1})^4 +(\cW_{-1})^2(\cW_{-2}) +(\cW_{-2})^2 &\text{weight }-4,\end{aligned}$$ hence equations of degree at most $4$. \[qlbar\] 1. In the above working, we may replace $\Q_l$ by any finite field extension $E/\Q_l$. Indeed, since $\pi_1(X,\bar{x})$ is pro-finitely generated, any representation to $G(\bar{\Q_l})$ will factor through $G(E)$ for some finite extension $E$ of $\Q_l$, so we may replace $\Q_l$ by $\bar{\Q_l}$, with the usual conventions for constructible $\bar{\Q_l}$-sheaves. 2. If $G=\GL_n$, a large class of examples can be produced to satisfy the hypotheses. For each irreducible continuous representation $\tilde{\rho_0}: W(X_k,\bar{x}) \to \GL_n(E)$, the induced representation $\rho_0: \pi_1(X,\bar{x}) \to \GL_n(E)$ satisfies the hypotheses. This follows from [@Weil2] Conjecture 1.2.10 (and remarks preceding) which imply that every irreducible Weil sheaf over $\bar{\Q_l}$ is of the form $$\sF = \sP \ten \bar{\Q_l}^{(b)},$$ for a pure sheaf $\sP$. Here, $\bar{\Q_l}^{(b)}$ is the constant sheaf $\Q_l$, on which the Frobenius action is multiplication by $b$, for $b \in \bar{\Q_l}^*$. Hence $$\ENd(\sF)\cong \sF \ten \sF^{\vee} \cong \sP \ten \sP^{\vee},$$ which is pure of weight $0$. Setting $\widetilde{\ad \rho_0}:= \ad \tilde{\rho_0}$ gives the desired result. Lafforgue has proved [@Weil2] Conjecture 1.2.10 (known as the Deligne Conjecture) as [@La] Theorem VII.6. Representations over $\Z_l$ --------------------------- We will consider deformations as in [@sdc] Section \[sdc-smoothgroup\], with $\L =\Z_l$. Given a mixed Weil sheaf $\sF$ over $\Z_l$, consider the Frobenius decompositions on $\H^i(X,\sF)\ten_{\Z_l}\Q_l$. Since the characteristic polynomial of $F^*$ is, in this case, defined over $\Z_l$, Gauss’s lemma and basic linear algebra provide a weight decomposition of the torsion-free quotient of $\H^i(X,\sF)$. Assume that we have a representation $$\rho_0: \pi_1(X,\bar{x}) \to G(\bF_l),$$ for which there exists a continuous representation $$\tilde{\rho_0}:W(X_k,\bar{x}) \to G(\Z_l),$$ making the following diagram commute: $$\begin{CD} \pi_1(X,\bar{x}) @>{\rho_0}>> G(\bF_l)\\ @VVV @AAA \\ W(X_k,\bar{x}) @>{\tilde{\rho_0}}>> G(\Z_l). \end{CD}$$ Write $\Bu_0$ for the principal $G(\bF_l)$-space corresponding to $\rho_0$, and $\widetilde{\Bu_0}$ the Weil sheaf of principal $G(\Z_l)$-spaces corresponding to $\tilde{\rho_0}$. Assume, moreover, that the representation $\ad \tilde{\rho_0}$ is mixed, and that the cohomology groups $$\H^i(X,\ad \widetilde{\Bu_0})\quad \quad i=1,2,3$$ are torsion free. Then the deformation functor $$\ddef_{\rho_0}:\C_{\Z_l} \to \Set$$ has a hull of the form: $$\Z_l[[\H^1(X,\ad \widetilde{\Bu_0})^{\vee}]]/(f(\H^2(X,\ad \widetilde{\Bu_0}))^{\vee})),$$ where $f$ preserves the Frobenius decompositions. As in Theorem \[frobhull\], we have a Frobenius endomorphism on the governing SDC $$E^n(A)= \CC^n(X, \ker(\aut(G(A)\by^{G(\Z_l)}\widetilde{\Bu_0})\to \aut(\Bu_0))).$$ This gives us an endomorphism $F^*: \ddef_E \to \ddef_E$. The representation $\tilde{\rho_0}$ gives a Frobenius-invariant point $\omega_0 \in \ddef_E(\Z_l)$. Let $R$ be the hull of $\ddef_E$; $\omega_0$ gives an augmentation $R \to \Z_l$. As in Theorem \[frobhull\], we obtain a map $\tilde{F}:R \to R$ of complete local *augmented* $\Z_l$-algebras. Let $I_R$ be the augmentation ideal of R. We have canonical isomorphisms $$(I_R/I_R^2)^{\vee} \cong \ker (\ddef_E(\Z_l[\eps]) \to \ddef_E(\Z_l)) \cong \H^i(X,\ad \widetilde{\Bu_0}),$$ where $\ker$ denotes the inverse image of $\omega_0$, compatible with Frobenius. Since $I_R/I_R^n$ is a finite rank $\Z_l$-module for all $n$, we may use the Jordan decomposition (over $\bar{\Q_l}$) successively to lift the cotangent space $I_R/I_R^2$ to $I_R/I_R^n$, obtaining a space of generators $V \subset I_R$, with the map $V \to I_R/I_R^2$ an isomorphism preserving the Frobenius decompositions. Now look at the map of augmented algebras $$S=\Z_l[[V]] \onto R;$$ let its kernel be $J$. Since the $\H^i(X,\ad \widetilde{\Bu_0})$ are torsion free for $i=1,2,3$, it follows that $$\H^i(X,\ad \widetilde{\Bu_0})\ten\bF_l \cong \H^i(X, \bu_0) \quad \quad i=1,2.$$ $((J/VJ)\ten \bF_l)^{\vee}$ is then a universal obstruction space for $h_R$. Since $h_R \to \ddef_E$ is smooth, and $\H^2(X, \bu_0)$ is an obstruction space for $\ddef_{\rho_0}$, there is a unique injective map of obstruction spaces ([@Man] Proposition 2.18) $$\label{zlobstrn} ((J/VJ)\ten \bF_l)^{\vee} \into \H^2(X, \bu_0)\cong \H^2(X,\ad \widetilde{\Bu_0})\ten\bF_l.$$ Now, consider the extension $$0 \to J/VJ \to S/VJ \to S/J =R \to 0.$$ Although this is not a small extension, it is close enough for our purposes, being, in some sense, a small extension of augmented algebras. We seek to imitate [@sdc] Lemma \[sdc-obstrsdc\]. Take the canonical element $\xi \in \ddef_E(R)$ arising from the map $h_R \to \ddef_E$. Lift $\xi$ to $\zeta \in \mc_E(R)$, and use the smoothness of $E^1$ to lift $\zeta$ to $\tilde{\zeta} \in E^1(S/VJ)$. We have the advantage over [@sdc] Lemma \[sdc-obstrsdc\] in that our SDC has the (stronger) structure of a cosimplicial group. Let $$c= \pd^0(\tilde{\zeta})\cdot \pd^2(\tilde{\zeta})\cdot \pd^1(\tilde{\zeta})=\tilde{\zeta}*\tilde{\zeta}\cdot \pd^1(\tilde{\zeta}) \in \ker(E^2(S/VJ) \to E^2(R)) = \mathfrak{e}^2 \ten J/VJ,$$ where $\mathfrak{e}^2$ is the $\Z_l$-Lie algebra $\ker(E^2(\Z_l[\eps]) \to E^2(\Z_l))$. The final equality follows since the augmentation ideal of $S/VJ$ is $(V)$, and $J/VJ\cdot (V)=0$. In fact, $c \in \z^2(\mathfrak{e})$ and we obtain $$[c] \in \H^2(\mathfrak{e})\ten J/VJ= \H^2(X,\ad \widetilde{\Bu_0})\ten J/VJ$$ Hence we get a map (Frobenius equivariant by the construction of $[c]$): $$\H^2(X,\ad \widetilde{\Bu_0})^{\vee} \to J/VJ.$$ Comparing our construction with [@sdc] Lemma \[sdc-obstrsdc\], we see that, mod $ l$, this map is the dual of the map in Equation \[zlobstrn\], so is surjective. As before, we may obtain a space of generators $W \subset J$, such that $W \to J/VJ$ is an isomorphism preserving the Frobenius decompositions. Thus $$R \cong \Z_l[[V]]/(W) \cong \Z_l[[\H^1(X, \bu_0)^{\vee}]]/(f(\H^2(X, \bu_0)^{\vee})),$$ where $f$ preserves the Frobenius decompositions. In fact, the quotient map $$f:(\H^2(X, \bu_0)^{\vee}) \to (\H^1(X, \bu_0)^{\vee})^2/(\H^1(X, \bu_0)^{\vee})^3$$ will be dual to half the cup product. 1. Let $X$ be smooth and proper, $\rho_0$ as above, and assume that $ \widetilde{\ad(\rho_0)} $ is pure of weight $0$. Then the deformation functor $$\ddef_{\rho_0}:\C_{\Z_l} \to \Set$$ has a hull $R$ defined by homogeneous quadratic equations. 2. Let $X$ be smooth, $\rho_0$ as above, and assume that $ \widetilde{\ad(\rho_0)} $ is pure of weight $0$. Then the deformation functor $$\ddef_{\rho_0}:\C_{\Z_l} \to \Set$$ has a hull defined by equations of degree at most four. [^1]: The author is supported by Trinity College, Cambridge and the Isle of Man Department of Education.

ArXiv

--- abstract: | In this work we include, for the Carnot cycle, irreversibilities of linear finite rate of heat transferences between the heat engine and its reservoirs, heat leak between the reservoirs and internal dissipations of the working fluid. A first optimization of the power output, the efficiency and ecological function of an irreversible Carnot cycle, with respect to: internal temperature ratio, time ratio for the heat exchange and the allocation ratio of the heat exchangers; is performed. For the second and third optimizations, the optimum values for the time ratio and internal temperature ratio are substituted into the equation of power and, then, the optimizations with respect to the cost and effectiveness ratio of the heat exchangers are performed. Finally, a criterion of partial optimization for the class of irreversible Carnot engines is herein presented. **Keywords:** Internal and external irreversibilities,heat engines, finite time and size thermodynamics, cost and effectiveness optimization. author: - 'G Aragón-González, A. Canales-Palma,' - | A. León-Galicia, J. R. Morales-Gómez\ PDPA. UAM- Azcapotzalco. Av. San Pablo \# 180. Col. Reynosa.\ Azcapotzalco, 02800, D.F. Teléfono y FAX: (55) 5318-9057.\ e-mail: gag@correo.azc.uam.mx. title: 'Maximum power, ecological function and efficiency of an irreversible Carnot cycle. A cost and effectiveness optimization.' --- **Nomenclature** $I$: internal irreversible factor. $Q$: heat transfer. $q$: dimensionless heat transfer. $W$: work. $P$: power $p$: dimensionless power. $S$: entropy-generation rate. $s$: dimensionless entropy-generation rate. $K$: thermal conductance for heat loss. $t$: time $x$: internal temperatures ratio. $y$: time ratio. $z$: allocation ratio. $U$: global heat transfer coefficient. $A$: total heat transfer area. $L$: thermal conductances ratio. $C$: total cost. Greek Symbols $\alpha $: thermal conductance of hot side. $\beta $: thermal conductance of cold side. $\eta $ : efficiency. $\sigma $: dimensionless dissipation. $\epsilon $ : dimensionless ecological function. $\mu $ : temperatures ratio of hot and cold sides. Subscripts $C$: Carnot. $CI$: Carnot-like. $CA$: endoreversible or Curzon-Ahlborn. $H$: hot-side. $L$: cold-side. $max$: maximum. $mp$: maximum power. $me$: maximum efficiency. $mec$: maximum ecological function. Superscripts $\ast $: second optimization. $\ast \ast $: third optimization. Introduction ============ The thermal efficiency of a reversible Carnot cycle is an upper limit of efficiency for heat engines. In according to classical thermodynamics, the Carnot efficiency is:$$\eta _{C}=1-\frac{T_{L}}{T_{H}} \label{effC}$$where $T_{L}$ and $T_{H}$ are the temperatures of the hot and cold reservoirs between which the heat engine operates. The thermal efficiency $% \eta _{C}$ can only be achieved through the infinitely slow process required by thermodynamic equilibrium. Therefore, it is not possible to obtain a certain amount of power output by using heat exchangers with finite heat transfer areas. Thus, the thermal efficiency given in equation (\[effC\]) does not have great significance and is a poor guide for the performances of real heat engines. A more realistic upper bound could be placed on the efficiency of a heat engine operating at its maximum power point; the so-called CA efficiency (Curzon-Alhborn [@curzon]):$$\eta _{CA}=1-\sqrt{\frac{T_{L}}{T_{H}}}$$where the only source of irreversibility in the engine is a linear finite rate heat transfer between the working fluid and its two heat reservoirs. Real heat engines are complex devices. Besides the irreversibility of finite-rate heat transfer in finite time taken into account in the Curzon-Ahlborn engine (CA-engine), there are also other sources of irreversibility, such as heat leaks, dissipative processes inside the working fluid and so on. Thus, it is necessary to investigate more comprehensively the influence of finite-rate heat transfer together with other major irreversibilities on the performance of heat engines. For this aim, we must consider general irreversible Carnot engines including three major irreversibilities, which often exist in heat engines, and use it to optimize the performance of an irreversible Carnot engine for several objective functions. In the past decade some new models of irreversible Carnot engines which include other irreversibilities, besides thermal resistance, have been established: heat leak and internal dissipations of the working fluid (see [@hoffman], [@gordon], [@chen], [@yan] , [@ahmet], [arias]{}, [@calvo], [@lchen], [@aragon] and included references there). Nevertheless, there are another parameters involved in the performance and optimization of an irreversible Carnot cycle; for instance, the allocation ratio of the heat exchangers, cost and effectiveness ratio of the heat exchangers and so on (see [@calvo], [@lchen] and [lewins]{}). In the optimization of Carnot cycles, including those irreversibilities, has appeared four objective functions: power, efficiency, ecological and entropy generation. The maximum power and efficiency have been obtained in [chen]{}, [@yan] and [@aragon]. The maximum ecological function was obtained in [@angulo] for the CA-engine and in form more general in [arias]{}. Bejan [@bejan] has considered the minimization of the entropy generation. In general, these optimizations were performed with respect to only one characteristic parameter: internal temperature ratio. In the first analysis of the CA-engine the time ratio of heat transfer from hot to cold side was considered, but in further works this ratio was not taken into account (see [@hoffman] for details). In this work this relation is considered as a parameter. On the other hand, [@bejan] has performed the optimization, also, with respect to other parameter: the allocation ratio of the heat exchangers; and [@lewins] has considered as parameters the cost and effectiveness ratio of the heat exchangers for the CA-engine. In this work we include, for the Carnot cycle, irreversibilities of linear finite rate of heat transferences between the heat engine and its reservoirs, heat leak between the reservoirs and internal dissipations of the working fluid. A first optimization of the power output, the efficiency and ecological function of an irreversible Carnot cycle, with respect to: internal temperature ratio, time ratio for the heat exchange and the allocation ratio of the heat exchangers; is performed. For the second and third optimizations, the optimum values for the time ratio and internal temperature ratio are substituted into the equation of power and, then, the optimizations with respect to the cost and effectiveness ratio of the heat exchangers are performed. Finally, a criterion of partial optimization for the class of irreversible Carnot engines is herein presented. This paper is organized as follows. In the section $2$ the relations for the dimensionless power, efficiency, entropy generation and ecological function of a class of irreversible Carnot engines are presented. In the section $3$, the optimal analytical expressions for the efficiencies corresponding to power and ecological function; and maximum efficiency are shown. In section $% 4$, the optimum values for the time ratio and internal temperature ratio are substituted in the expression for dimensionless power. Then a second and third optimizations of dimensionless power, are performed with respect to the cost and effectiveness ratio of the heat exchangers. In the section of Conclusions, a criterion of partial optimization for power, ecological function, efficiency and entropy generation is presented. Irreversible Carnot engine. =========================== In considering the class of irreversible Carnot engines (see [@hoffman]) shown in Figure $1$, which satisfy the following five conditions: {width="10.0cm"} \(i) The cycle of the engine consists of two isothermal and two adiabatic processes. The temperatures of the working fluid in the hot and cold isothermal processes are, respectively, $T_{1}$ and $T_{2}$, and the times of the two isothermal processes are, respectively, $t_{H}$ and $t_{L}$. The temperatures of the hot and cold heat reservoirs are, respectively, $T_{H}$ and $T_{L}$. \(i) There is thermal resistance between the working fluid and the heat reservoirs. \(ii) There is a heat lost $Q_{leak\textrm{ }}$from the hot reservoir to the cold reservoir [@bejan]. In real engines heat leaks are unavoidable, there are many features of an actual power plant which fall under that kind of irreversibility, such as the heat lost through the walls of a boiler, a combustion chamber, or a heat exchanger, and heat flow through the cylinder walls of an internal combustion engine, and so on. \(iii) All heat transfer is assumed to be linear in temperature differences, that is, Newtonian. \(iv) Besides thermal resistance and heat loss, there are other irreversibilities in the cycle, the internal irreversibilities. For many devices such as gas turbines, automotive engines, and thermoelectric generator, there are other loss mechanisms, like friction or generators losses, etc. that play an important role, but are hard to model in detail. Some authors use the compressor (pump) and turbine isentropic efficiencies to model the internal loss in the gas turbines or steam plants. Others, in Carnot cycles, use simply one parameter to describe the internal losses. Such a parameter is associated with the entropy produced inside the engine during a cycle. Specifically, this parameter makes the Claussius inequality becomes an equality (for details see [@hoffman]):$$\frac{Q_{2}}{T_{2}}-I\frac{Q_{1}}{T_{1}}=0 \label{claus}$$where $I=\frac{\Delta S_{2}}{\Delta S_{1}}\geq 1$ ([@chen]). Thus, the irreversible Carnot engine operates with fixed time $t$ allowed for each cycle. The heat leakage $Q_{leak}$ is ([@bejan]): $$Q_{leak}=K(T_{H}-T_{L})t$$ The heats $Q_{H},$ $Q_{L\textrm{ }}$transferred from the hot-cold reservoirs are given by: $$\begin{aligned} Q_{H} &=&Q_{1}+Q_{leak}=\alpha (T_{H}-T_{1})t_{H}+K(T_{H}-T_{L})t \label{QH} \\ Q_{L} &=&Q_{2}+Q_{leak}=\beta (T_{2}-T_{L})t_{L}+K(T_{H}-T_{L})t\end{aligned}$$where $\alpha ,$ $\beta $ and $K$ are the thermal conductances and $% t_{H},t_{L}$ are the time for the heat transfer in the isothermal branches, respectively. The connecting adiabatic branches are often assumed to proceed in negligible time ([@gordon]), such that the cycle contact total time $% t $ is [@rubin]: $$t=t_{H}+t_{L} \label{time}$$ By first law and combining equations (\[QH\]) and (\[claus\]) we obtain : $$W=Q_{1}(1-Ix)=\frac{T_{H}(1-Ix)\left( 1-\frac{\mu }{x}\right) }{\frac{1}{% \alpha t_{H}}+\frac{I}{\beta t_{L}}} \label{W}$$ $$Q_{H}=Q_{1}+Q_{leak}=\frac{W}{1-Ix}+K(T_{H}-T_{L})t \label{QHf}$$ where $\mu =\frac{T_{L}}{T_{H}}$. And $x=\frac{T_{2}}{T_{1}}$ is a characteristic parameter of the engine. Now, the equation (\[time\]) gives us the total time of the cycle, so it can be parametrized as:$$t_{H}=yt;\textrm{ }t_{L}=(1-y)t$$where $y=\frac{t_{H}}{t}=\frac{t_{H}}{t_{H}+t_{L}}$ is other characteristic parameter of the engine. Another parameter is the allocation of the exchangers heat [@bejan]. The thermal conductances can be written as: $$\alpha =UA_{H};\textrm{ }\beta =UA_{L}$$where $U$ is overall heat transfer coefficient and $A_{H}$ and $A_{L}$ are the available areas for heat transfer. Then, an approach might be to suppose that $U$ is fixed, the same for the hot side and the cold side heat exchangers, and that the area $A$ can be allocated between both. The optimization problem is then selected, besides of the optimum temperature ratio and the time ratio, as the best allocation ratio. To take $UA$ as a fixed value can be justified in terms of the area purchased, and the fixed running costs and capital costs that altogether determine the overall heat transfer coefficient (see ([@lewins])). Thus, for the optimization we can take: $$\frac{\alpha }{U}+\frac{\beta }{U}=A \label{rule1}$$and parametrize it as:$$\alpha =zUA;\textrm{ }\beta =(1-z)UA$$$$\frac{\alpha }{\beta }=\frac{z}{(1-z)} \label{z}$$ Therefore, the dimensionless power output, $p=\frac{W}{AUtT_{H}}$, and the dimensionless heat transfer rate $q_{H}=$ $\frac{Q_{H}}{AUtT_{H}}$ are (by equations(\[W\]) and (\[QHf\])):$$p=\frac{z\left( 1-z\right) y\left( 1-y\right) \left( 1-Ix\right) \left( 1-% \frac{\mu }{x}\right) }{\left( 1-z\right) \left( 1-y\right) +zyI} \label{powerIL}$$$$q_{H}=\frac{z\left( 1-z\right) y\left( 1-y\right) \left( 1-\frac{\mu }{x}% \right) }{\left( 1-z\right) \left( 1-y\right) +zyI}+L(1-\mu ) \label{qH}$$where$\ L=\frac{K}{AU}$. And $z=\frac{\alpha }{UA}$ is the third characteristic parameter of the engine. The thermal efficiency is given by:$$\eta =\frac{z\left( 1-z\right) y\left( 1-y\right) \left( 1-Ix\right) \left( 1-\frac{\mu }{x}\right) }{z\left( 1-z\right) y\left( 1-y\right) \left( 1-% \frac{\mu }{x}\right) +L(1-\mu )\left( \left( 1-z\right) \left( 1-y\right) +zyI\right) } \label{eff_IL}$$ The entropy-generation rate, $s_{gen}=\frac{S_{gen}}{AUtT_{H}}$, multiplied by the temperature of the cold side, give us a dimensionless function $% \sigma ,$ which is (equations (\[powerIL\],\[qH\])):$$\sigma =T_{L}s_{gen}=T_{L}\left( \frac{q_{H}-p}{T_{L}}-\frac{q_{H}}{T_{H}}% \right) =q_{H}(1-\mu )^{2}-p$$so, $$\sigma =\frac{z\left( 1-z\right) y\left( 1-y\right) \left( 1-\frac{\mu }{x}% \right) \left( Ix-\mu \right) }{\left( 1-z\right) \left( 1-y\right) +zyI}% \allowbreak +L(1-\mu )^{2} \label{entroIL}$$ Finally, the ecological function [@angulo], when $T_{L}$ is the environmental temperature, is: $$\epsilon =p-\sigma =p\frac{2Ix-1-\mu }{Ix-1}+L(1-\mu )$$$\allowbreak $then,$$\epsilon =\frac{z\left( 1-z\right) y\left( 1-y\right) \left( 1-\frac{\mu }{x}% \right) \left( 1-2Ix\right) }{\left( 1-z\right) \left( 1-y\right) +zyI}% +L(1-\mu ) \label{ecolIL}$$when $I=1$ and $L=0$ the expressions for the CA-engine are obtained. Maximum power, ecological function and efficiency. ================================================== In using the equation (\[powerIL\]) and the extremes conditions: $$\frac{\partial p}{\partial x}{\Huge |}_{(x_{mp},y_{mp},z_{mp})}=0;\textrm{ \ \ }\frac{\partial p}{\partial y}{\Huge |}_{(x_{mp},y_{mp},z_{mp})}=0;\textrm{\ }% \frac{\partial p}{\partial z}{\Huge |}_{(x_{mp},y_{mp},z_{mp})}=0$$when the power reaches its maximum, $x_{mp},$ $y_{mp}$ and $z_{mp}$ are given by: $$x_{mp}=\sqrt{\frac{\mu }{I}} \label{xmp}$$ $$y_{mp}=z_{mp}=\frac{1}{\sqrt[3]{I}+1} \label{yzmp}$$ Clearly $p$ reaches its maximum in $(x_{mp},y_{mp},z_{mp})$. Indeed, all the critical points are (necessary condition): $$\begin{tabular}{c} $\left\{ z=0,y=y,x=\mu \right\} ,\left\{ x=\frac{1}{I},y=1,z=z\right\} ,\left\{ y=1,z=z,x=\mu \right\} $ \\ $\left\{ y=0,z=z,x=\mu \right\} ,\left\{ x=\frac{1}{I},z=0,y=y\right\} ,\left\{ y=0,z=0,x=x\right\} $ \\ $\left\{ z=1,y=1,x=x\right\} ,\left\{ z=1,y=y,x=\mu \right\} ,\left\{ x=% \frac{1}{I},z=1,y=y\right\} $ \\ $\left\{ x=\frac{1}{I},y=0,z=z\right\} ,\left\{ x=\pm \sqrt{\frac{\mu }{I}}% ,y=\frac{1}{\sqrt[3]{I}+1},z=\frac{1}{\sqrt[3]{I}+1}\right\} $% \end{tabular}%$$ In eliminating the solutions without physical meaning, we see that there is only one global critical point given by the equations (\[xmp\], \[yzmp\]). Moreover, at this critical point maximum power developed. Indeed, a sufficient condition for maximum power is, the eingenvalues of the Hessian ($% \left[ \frac{\partial ^{2}p}{\partial w\partial u}|_{\left( x_{mp},y_{mp},z_{mp}\right) }\right] _{w,u=x,y,z}$) must be negatives ([panos]{}). It is clearly fulfilled that: $$\left[ \begin{array}{ccc} \allowbreak -\frac{2\allowbreak I^{\frac{3}{2}}}{\sqrt{\mu }\left( 1+\sqrt[3]% {I}\right) ^{3}} & 0 & 0 \\ 0 & \allowbreak -\frac{2\left( 1-\sqrt{I\mu }\right) ^{2}}{\sqrt[3]{I}\left( 1+\sqrt[3]{I}\right) } & 0 \\ 0 & 0 & -\frac{2\left( 1-\sqrt{I\mu }\right) ^{2}}{\sqrt[3]{I}\left( 1+\sqrt[% 3]{I}\right) }% \end{array}% \right]$$ The efficiency that maximizes the power $\eta _{mp}$ is given by (see equation (\[eff\_IL\]) and Figure 2), $$\eta _{mp}=\frac{\left( 1-\sqrt{I\mu }\right) }{1+\frac{L\left( 1-\mu \right) \left( \sqrt[3]{I}+1\right) ^{3}}{\left( 1-\sqrt{I\mu }\right) }} \label{efmp}$$ The generation of entropy is minimum when $y$ and $z$ are given by equation (\[yzmp\]) and $x=\frac{\mu }{\sqrt{I}}.$ Nevertheless, for these values it is seen that the corresponding power does not have physical meaning. For $x=% \frac{\mu }{I}$ ($y=0,1$ or $z=0,1$)$,$ makes the first term of the equation(\[entroIL\]) zero. The corresponding values of $y,z$ are also without physical meaning. For $x=\mu $ ($y=0,1$ or $z=0$, $1)$ do not have physical meaning either. Therefore, for this kind of Carnot engine, the entropy generation does not have a global minimum within the valid interval. In [lchen1]{} an engine that corresponds with the kind of irreversible Carnot cycles herein presented is analyzed but the calculations leading to the minimization of entropy generation are at fault, since they do not have physical meaning. It results that the obtained power is negative! Thus, it is only possible to minimize the entropy generation partially for the variables $y,z$ and those values are given by: $$y_{m\sigma }=z_{m\sigma }=\frac{1}{\sqrt[3]{I}+1} \label{yzment}$$ In doing a analogous analysis for the ecological function, we have by the equation (\[ecolIL\]) that the unique critical point of ecological function solutions with physical meaning is: $$x_{mec}=\sqrt{\frac{\mu \left( 1+\mu \right) }{2I}}, \label{xmec}$$$$y_{mec}=z_{mec}=\frac{1}{\sqrt[3]{I}+1} \label{yzmec}$$and newly can see that its Hessian has all its negative eingenvalues. The efficiency that maximizes the ecological function $\eta _{mec}$ is given by (equation (\[eff\_IL\])):$$\eta _{mec}=\frac{\left( 1-\sqrt{\frac{\mu \left( 1+\mu \right) I}{2}}% \right) }{1+\frac{L\left( 1-\mu \right) \left( \sqrt[3]{I}+1\right) ^{3}}{% \left( 1-\sqrt{\frac{2I\mu }{\mu +1}}\right) }} \label{efimec}$$ Similarly, it’s easily seen that there is only one critical point, with physical meaning, for the efficiency, and it is given by: $$x_{me}=\frac{\mu +\sqrt{L\mu \left( 1-\mu \right) \left( 1+\sqrt[3]{I}% \right) ^{3}\left( L\left( \sqrt[3]{I}+1\right) ^{3}\left( 1-\mu \right) +1-I\mu \right) }}{I^{2}\left( L\left( \sqrt[3]{I}+1\right) ^{3}\left( 1-\mu \right) +1\right) } \label{xme}$$ $$y_{me}=z_{me}=\frac{1}{\sqrt[3]{I}+1} \label{yzme}$$ To see, as above, that the efficiency reaches a maximum, becomes to cumbersome a task if the solution of systems of equations are undertaken. Therefore, an alternative way is presented in that follows, to obtain equation (\[xme\]). And when the efficiency reaches its maximum $\left( x_{me},y_{me},z_{me}\right) $ is given by the equations(\[xme\],\[yzme\]). Indeed, clearly the values of $y_{me},z_{me}$ given by the equation ([yzme]{}) fulfill the following two extreme conditions:$$\frac{\partial \eta }{\partial y}=0;\frac{\partial \eta }{\partial z}=0$$ Furthermore, as it was seen above, the optimal time ratio and the allocation ratio are the same for both maximum power and ecological function (equations (\[xmec\]), (\[yzmec\])). Therefore,$$y_{mp}=y_{mec}=y_{me}=z_{mp}=z_{mec}=z_{me}=\frac{1}{\sqrt[3]{I}+1}$$ Thus, this values could be included in the equations of power and heat transfer (equations (\[powerIL\],\[qH\])) and proceed to optimizes the efficiency (equation (\[eff\_IL\]) by the following criterion valid when there is only one parameter([@aragon]): Criterion (Maximum efficiency) : *Let* $\eta =\frac{p}{q_{H}}$* *$.$ *Suppose  *$\frac{\partial ^{2}p}{\partial x^{2}}% |_{x}=\frac{\partial ^{2}q_{H}}{\partial x^{2}}|_{x},$* for some* $% x. $* Then the maximum efficiency *$\eta _{\max }$* is given by* $$\eta _{\max }=\frac{\frac{\partial p}{\partial x}|_{x_{me}}}{\frac{\partial \mathit{\ }q_{H}}{\partial x}|_{x_{me}}} \label{maxeff}$$*where* $x_{me}$* is the point in which* $\eta $ *achieves a maximum value.* Then, by the equations (\[powerIL\]) and (\[QH\]) we obtain the relationships of $p$ and $q_{H}$ with respect to $x.$$$p=\frac{\left( 1-Ix\right) (1-\frac{\mu }{x})}{\left( \sqrt[3]{I}+1\right) ^{3}}$$$$q_{H}=\frac{(1-\frac{\mu }{x})}{\left( \sqrt[3]{I}+1\right) ^{3}}+L(1-\mu )$$ The conditions of the criterion are clearly satisfied. Indeed,$$\frac{\partial ^{2}p}{\partial x^{2}}=\frac{\partial ^{2}q_{H}}{\partial x^{2}}=\allowbreak -\frac{2\mu }{x^{3}\left( \sqrt[3]{I}+1\right) ^{3}}<0$$since $x>0.$Therefore (equation(\[maxeff\])), $$\eta _{\max }=1-\frac{x_{me}^{2}I}{\mu } \label{efirrmax}$$where $x_{me}$ must, by the second law, satisfies the inequality ([aragon]{}): $$\textrm{\ \ }\frac{\mu }{I}\leq x_{me}\leq \sqrt{\frac{\mu }{I}} \label{efimec1}$$ if we apply the preceding statement and the equation (\[efirrmax\]), the following inequality is obtained $$\eta _{mp}=1-\sqrt{I\mu }\leq \eta _{\max }\leq 1-I\mu =\eta _{CI} \label{desefmaxcl}$$where $\eta _{mp}=1-\sqrt{I\mu }$ and $\eta _{CI}=1-I\mu $ corresponding to (Curzon-Ahlborn)-like and Carnot-like efficiencies; which includes the internal irreversibilities in the $I$ factor. Nevertheless, we can calculate easily $x_{me}$ from the following cubic equation: $$1-\frac{x_{me}^{2}I_{S}}{\mu }=\frac{p|_{x_{me}}}{q_{H}|_{x_{me}}}=\frac{% (1-x_{me}I)(1-\frac{\mu }{x_{me}})}{(1-\frac{\mu }{x_{me}})+L(1-\mu )\left( \sqrt[3]{I}+1\right) ^{3}}$$In solving this equation and taking into account the inequality ([efimec1]{}), we obtain the equation (\[xme\]). Finally, the maximum efficiency $\eta _{\max }$ is given by (equation ([efirrmax]{}) ):$$\eta _{\max }=1-\left( \frac{\sqrt{I\mu }+\sqrt{L\left( 1-\mu \right) \left( 1+\sqrt[3]{I}\right) ^{3}\left( L\left( 1-\mu \right) \left( 1+\sqrt[3]{I}% \right) ^{3}+1-I\mu \right) }}{1+L\left( 1-\mu \right) \left( \sqrt[3]{I}% +1\right) ^{3}}\right) ^{2} \label{effmay}$$ The behavior of the efficiencies $\eta _{mp},\eta _{mec}$ and $\eta _{\max }$ is shown in the Figure 2. In general it has been supposed that $I\geq 1;$ but sometimes can be considered that $I=1$. In this case the internal irreversibilities can be physically interpreted as part of the engine’s heat leak that brings us to the engine modeled in [@hoffman] and [@bejan]. So, substitution of $% I=1$ into equations (\[xmp\]; \[xmec\], \[xme\]) and (\[yzme\]) gives:$$\begin{aligned} x_{mp} &=&\sqrt{\mu };\textrm{ }x_{mec}=\frac{\left( 1-\sqrt{\frac{2\mu }{\mu +1}}\right) \left( 1-\sqrt{\frac{\mu \left( 1+\mu \right) }{2}}\right) ^{2}}{% 8L\left( 1-\mu \right) +\left( 1-\sqrt{\frac{2\mu }{\mu +1}}\right) }\textrm{;} \\ \textrm{ }x_{me} &=&\frac{\mu +\sqrt{8L\mu \left( 1-\mu \right) \left( 8L\left( 1-\mu \right) +1-I\mu \right) }}{8\left( 8L\left( 1-\mu \right) +1\right) }\end{aligned}$$$$y_{mp}=y_{mec}=y_{me}=z_{mp}=z_{mec}=z_{me}=\frac{1}{2}$$ {width="10.0cm"} The equations:$$x_{mp}=\sqrt{\mu }\textrm{ and }z_{mp}=\frac{1}{2}$$are the same as the presented in [@bejan] and $$x_{mp}=\sqrt{\mu }\textrm{ and }y_{mp}=\frac{1}{2}$$corresponding to the CA-engine. Further, the following results are obtained (see equations (\[efmp\]; \[efimec\], \[effmay\])):$$\eta _{mp}=\frac{1-\sqrt{\mu }}{2+\frac{8L\left( 1-\mu \right) }{1-\sqrt{\mu }}}$$$$\eta _{mec}=\frac{1-\sqrt{\frac{\mu \left( 1+\mu \right) }{2}}}{1+\frac{% 8L\left( 1-\mu \right) }{\left( 1-\sqrt{\frac{2\mu }{\mu +1}}\right) }}$$$$\eta _{\max }=1-\left( \frac{\sqrt{\mu }+\sqrt{8L\left( 1-\mu \right) \left( 8L\left( 1-\mu \right) +1-\mu \right) }}{1+8L\left( 1-\mu \right) }\right) ^{2}$$some of these results have been reported in the literature [@ahmet]. On the other hand, the optimization performed in this work gives results that could be applied to the design of power plants. For instance, in the third section it is found that for $$y_{mp}=y_{m\sigma }=y_{mec}=y_{me}=z_{mp}=z_{m\sigma }=z_{mec}=z_{me}=\frac{1% }{\sqrt[3]{I}+1} \label{yzmax}$$the engine operates at maximum power, efficiency and ecological function and entropy generation local minimum. Therefore, the time rate in the isothermal processes satisfies:$$\frac{t_{L}}{t_{H}}=\sqrt[3]{I}\geq 1 \label{ratet}$$This result generalizes to one presented in [@hoffman]. And when $I=1,$$$t_{L}=t_{H}$$ Similarly, it follows from the equation (\[yzmax\]) that when the engine operates at maximum power, efficiency and ecological function, the relation for the heat transfer areas for the cold side to the hot side, is:$$\frac{A_{L}}{A_{H}}=\sqrt[3]{I}\geq 1;\frac{\beta }{\alpha }=\frac{UA_{L}}{% UA_{H}}=\sqrt[3]{I} \label{rateA}$$This result shows that the size of the heat exchanger in the cold side must be larger than the size of heat exchanger in the hot side. Thus, in accordance with the definitions adopted for the thermal conductance, if $I>1$ the one for the cold side results greater than the hot side$.$ Furthermore, if $I=1$$$A_{L}=A_{H}$$which implies that the allocation of the heat exchangers is balanced ([bejan]{}) . By the equations (\[ratet\]; \[rateA\]) we have:$$\frac{t_{L}}{t_{H}}=\frac{A_{L}}{A_{H}}$$which is satisfied when the heat engine operates to maximum power, ecological function and efficiency, and minimum entropy generation. In [aragon1]{} the above relationship was obtained, by a double optimization of power and efficiency. For $I>1,$ the irreversibility produces an inverse relationship between the total area and the total contact time; that is, a less time is needed to transfer the heat that the engine processes. This is due to the fact that less heat goes through the engine. Part of the heat is lost because of internal irreversibility. For $I=1,$ the relationship between area and contact time is inversely proportional; that is, if the area is augmented the time is reduced. This result does not depend explicitly of $I$ and differs of one presented in [@chen] and [hoffman]{}. A cost and effectiveness optimization. ====================================== A more detailed model would involve acknowledgement that the cost of providing the same heat transfer capability differs between the cold and hot sides. Let this represented as having a cost per unit heat transfer to be $% c_{L}$ on the cold side but $c_{H}$ on the hot side ([@lewins]). Then$$c_{H}\alpha +c_{L}\beta =C$$where $C$ is fixed total cost. Thus we have that the third characteristic parameter $z_{1}$ changes to:$$z^{\ast }=\frac{c_{H}\alpha }{C};1-z^{\ast }=\frac{c_{H}\beta }{C}$$In including the optimal values $x_{mp},y_{mp}$ (equations (\[xmp\], [yzmp]{})) in the equation (\[W\]) the dimensionless power $p^{\ast }=\frac{W% }{C^{\ast }T_{H}t_{H}}$is given by: $$p^{\ast }=\frac{\left( 1-\sqrt{I\mu }\right) ^{2}}{\frac{1}{z^{\ast }}+\frac{% cI^{\frac{2}{3}}}{1-z^{\ast }}}$$where $\allowbreak c=\frac{c_{L}}{c_{H}}>1$ (equation(\[rateA\])) and $% C^{\ast }=\frac{C}{c_{H}}$. The efficiency is given by:$$\eta ^{\ast }=\frac{\left( 1-\sqrt{I\mu }\right) ^{2}}{1-\sqrt{I\mu }% +L\left( 1-\mu \right) \left( \sqrt[3]{I}+1\right) \left( \frac{1}{z^{\ast }}% +\frac{cI^{\frac{2}{3}}}{\left( 1-z^{\ast }\right) }\right) }$$ In optimizing the power with respect to $z^{\ast },$ we have:$$z_{mp}^{\ast }=\frac{1}{1+\sqrt{c}\sqrt[3]{I}}$$or equivalently$$\frac{\beta }{\alpha }=\frac{\sqrt[3]{I}}{\sqrt{c}}$$Of course this reverts to the earlier form (equation(\[rateA\])) if $c=1.$ The efficiency that develops maximum power is:$$\eta _{mp}^{\ast }=\frac{1-\sqrt{I\mu }}{1+\frac{L\left( 1-\mu \right) \left( 1+\sqrt[3]{I}\right) \left( 1+\sqrt[3]{I}\sqrt{c}\right) ^{2}}{1-% \sqrt{I\mu }}}$$The Figure $3$ shows the behavior of $\eta _{mp}^{\ast }$versus $c$. {width="10.0cm"} In general, there are two design rules for heat exchange at the two ends of the heat engine ([@salah]). The first rule is that the thermal conductance is constrained: $$\alpha +\beta =\gamma$$where $\gamma $ is a constant; which was applied herein for the allocation of the heat exchangers (see equation (\[z\]) with $\gamma =UA$). The second rule is that the total is constrained by $$A=A_{H}+A_{L}$$where $A_{H},A_{L}$ are heat transfer areas on hot and cold side. To apply the second rule, we may be faced with an existing heat exchange apparatus which is to be redistributed between hot and cold sides to achieve maximum power. Now, the total area $A$ is fixed but when distributed it has different overall heat transfer coefficients and hence different effectiveness on hot and cold sides. Thus, $$A=A_{H}+A_{L}=\frac{\alpha }{U_{H}}+\frac{\beta }{U_{L}}$$where $U_{H},U_{L}$ are overall heat transfer coefficients on hot and cold side. In parametrizing again. $$z^{\ast \ast }=\frac{\alpha }{U_{H}A};1-z^{\ast \ast }=\frac{\beta }{U_{L}A}$$Again, including the optimal values $x_{mp},y_{mp}$ (equations (\[xmp\], \[yzmp\])) in the equation (\[W\]) the dimensionless power $p^{\ast \ast }=\frac{W}{AU_{H}t_{H}T_{H}}$is given by:$$p^{\ast \ast }=\frac{\left( 1-\sqrt{I\mu }\right) ^{2}}{\frac{1}{z}+\frac{I^{% \frac{2}{3}}}{\left( 1-z\right) U}}$$where $U=\frac{U_{L}}{U_{H}}$. And the efficiency is now given by:$$\eta ^{\ast \ast }=\frac{\left( 1-\sqrt{I\mu }\right) ^{2}}{1-\sqrt{I\mu }% +L\left( 1-\mu \right) \left( \sqrt[3]{I}+1\right) \left( \frac{1}{z}+\frac{% I^{\frac{2}{3}}}{\left( 1-z\right) U}\right) }$$In optimizing the power with respect to $z^{\ast \ast },$ we have:$$z_{mp}^{\ast \ast }=\frac{\sqrt{U}}{\sqrt[3]{I}+\sqrt{U}}$$or equivalently$$\frac{\beta }{\alpha }=\sqrt[3]{I}\sqrt{U}\,;\textrm{ \ \ }\frac{A_{L}}{A_{H}}=% \frac{\sqrt[3]{I}}{\sqrt{U}}=\sqrt[3]{I}\sqrt{\frac{U_{H}}{U_{L}}}$$Then, the optimal distribution of the heat exchangers areas is:$$\begin{aligned} A_{H} &=&\frac{A}{1+\sqrt[3]{I}\sqrt{\frac{U_{H}}{U_{L}}}}; \\ A_{L} &=&\frac{A}{1+\frac{1}{\sqrt[3]{I}}\sqrt{\frac{U_{L}}{U_{H}}}}\end{aligned}$$This result has been reported by [@kodal] (when $I=1)$ using another thermoeconomic criterion. However, it differs when $I\neq 1$ [@kodal1]. Then, the efficiency that develops maximum power is:$$\eta _{mp}^{\ast \ast }=\frac{1-\sqrt{I\mu }}{1+\frac{L\left( 1-\mu \right) \left( \sqrt[3]{I}+1\right) \left( 1+\allowbreak \frac{\sqrt[3]{I}}{\sqrt{U}}% \right) ^{2}}{1-\sqrt{I\mu }}}$$ The Figure 4 shows the behavior of $\eta _{mp}^{\ast \ast }$ versus $U$. {width="10.0cm"} Conclusion. =========== In the above section, the optimization has been carried for one objective function, that is, the power developed, respect to two additional parameters. These parameters could be treated as economic ones. Moreover, the obtained values, also optimizes the other considered objective functions. Such as, entropy generation, ecological function and efficiency. This is due to the fact that the analyzed Carnot engine satisfy the following partial criterion: Criterion : If $f_{i}=f_{i}(x,z,z^{\ast },z^{\ast \ast }...)$ represents one of the four objective functions, that is, power, efficiency, ecological function or entropy generation; with (power, $i=1,...4;$ $x$ as the internal temperature ratio; $z,z^{\ast },z^{\ast \ast }....$ are the characteristic-economic parameters of Carnot cycles belonging to the class of Carnot irreversible cycles analyzed. Moreover, if $z_{mj},z_{mj}^{\ast },z_{mj}^{\ast \ast }...$are the optimum values for functions $f_{j}$ for some $j$, then $\ z_{mj},z_{mj}^{\ast },z_{mj}^{\ast \ast }...$are also the optimum values for the functions $f_{i},$ for $i\neq j.$ It is a fact that it suffices to develop the optimization for a couple of objective functions, say the power and the efficiency to obtain the parameters that optimizes the remaining objective functions. For instance, the power and the efficiency satisfy the following functional relationship (equation (\[eff\_IL\])): $$\eta =\frac{p}{A+g(x)p} \label{rel_func}$$ with $A=L(1-\mu )$ and $g(x)=\frac{1}{1-Ix}.$ Let $\phi =z,z^{\ast },z^{\ast \ast }...$Then,$$\frac{\partial \eta }{\partial \phi }=\frac{\frac{\partial p}{\partial \phi }% \left( 1-\eta \right) }{A+g(x)p}$$Therefore,$$\frac{\partial \eta }{\partial \phi }{\Huge |}_{\phi _{mp}=\phi _{me}}=0\Longleftrightarrow \frac{\partial p}{\partial \phi }{\Huge |}_{\phi _{mp}=\phi _{me}}=0$$ This implies that their roots are the same (necessary condition). It is easily seen that for $\phi _{mp}=\phi _{me}$, the power and the efficiency reach a maximum (sufficiency condition) A remarkable conclusion of this work is that it is sufficient to find the extreme of some of the functions $f_{i}$, say the power so that $$\frac{\partial p}{\partial x}=0;\textrm{ }\frac{\partial p}{\partial \phi }=0$$where $\phi =z,z^{\ast },z^{\ast \ast }...$and then substitute in the appropriate functional relationship (for the efficiency is equation([rel\_func]{})) the values of $\phi _{mp}=\phi _{me}.$ The obtained $% f_{i}=f_{i}\left( x\right) $ ($\eta =\eta (x)$) are then optimized respect to the $x$ parameter only. It is found that the result optimizes the other objective functions. In other words, for the class of irreversible Carnot engines considered in this work, the $x$ parameter could be considered as the fundamental characteristic parameter of the engine. This is the only parameter that changes its optimal value according to the engine operation conditions. The remaining parameters maintain their optimal value independently of the operation condition of the engine. Finally, the above mentioned criteria could be applied and extended to other models of irreversible engines [@ahmet]. Further work is underway. **Acknowledgement:** This work was supported by the Program for the Professional Development in Automation, through the grant from the Universidad Autónoma Metropolitana and Parker Haniffin - México. [99]{} F. L. Curzon and B. Ahlborn . *Efficiency of a Carnot engine at maximum power output*. Am. J. Phys. 1975; 43: 22-24. K. H. Hoffman, J. M. Burzler and S. Shuberth . *Endoreversible Thermodynamics.* J. Non-Equilib. Thermodyn. 1997: 22 (4): 311-55. J. M. Gordon . *On optimizing maximum power heat engines*. J. Appl. Phys. 1991; 69: 1-7. J. Chen . The maximum power output and maximum efficiency of an irreversible Carnot heat engine. J. Phys. D: Appl. Phys. 1994; 27: 1144-49. Z. Yan and L. Chen . *The fundamental optimal relation and the bounds of power output and efficiency for and irreversible Carnot engine*. J. Phys. A: Math. Gen. 1995; 28: 6167-75. A. Durmayaz, O. S. Sogut, B. Sahin and H. Yavuz . *Optimization of thermal systems based on finite time thermodynamics and thermoleconomics.* Progr. Energ. and Combus. Sci. 2004; 30: 175-217. L. A. Arias-Hernández, G. Ares de Parga and F. Angulo-Brown. *On Some Nonendoreversible Engine Models with Nonlinear Heat Transfer Laws.* Open Sys. & Information Dyn. 2003; 10: 351-75. A. Calvo, A. Medina, J. M. Roco, J. A. White and S. Velasco . *Unified optimization criterion for energy converters.* Phys. Rev. E 2001; 63(4): 0371021–5. Lingen Chen, Xiaoqin Zhi, Fengrui Sun, Chih Wu. A . *Optimal configuration and performance for a generalized Carnot cycle assuming the heat-transfer generalized law*. Open Sys. & Information Dyn. 2004; 10: 351-75. G. Aragón-González, A. Canales-Palma, A. León-Galicia and M. Musharrafie-Martínez . *A criterion to maximize the irreversible efficiency in heat engines*. J. Phys. D: Appl. Phys. 2003; 36: 280-87. M. H. Rubin **.** *Optimal configurations of a class of irreversible heat engines.* Phys. Rev. **A** 1979; 15: 2094-102. F. Angulo-Brown . *An ecological optimization-criterion for finite-time heat engines.* J. Appl. Phys. 1991; 69 (11): 7465–9. A. Bejan .* Entropy Generation Minimization.* CRC Press, Boca Raton, FL; 1996. J. D. Lewins . *The endo-reversible thermal engine: cost and effectiveness optimization.* Int. J. Mech. Eng. Edu. 2000; 28 (1): 41-6. P. Y. Papalambros and D. J. Wilde . *Principles of optimal design.* Cambridge: University Press, p. 142; 2000. L. Chen, J. Zhou, F. Sun, C. Wu. A . *Ecological optimization for generalized irreversible Carnot engines*. Appl. Energ. 2004; 77: 327-38. G. Aragón-González, A. Canales-Palma, A. León-Galicia and J. R. Morales-Gómez. *Optimization of an irreversible Carnot engine in finite time and finite size.* To appear in Rev. Mex. Fis. M. M. Salah EL-Din . *Second law analysis heat engines with variable temperature heat reservoirs.* Energ. Convers. Manage. 2001; 42: 189-200. B. Sahin and A. Kodal . *Performance of an endoreversible heat engine based on a new thermoeconomic optimization criterion.* Energ. Convers. Manage. 2001; 42: 1085-93. B. Sahin and A. Kodal . *Finite size thermoeconomic optimization for irreversible heat engines.* Int. J. Ther. Sci. 2003; 42: 777–82.

ArXiv

--- abstract: 'The Large Area Telescope (LAT) on–board the [*Fermi*]{} Gamma–ray Space Telescope began its on–orbit operations on June 23, 2008. Calibrations, defined in a generic sense, correspond to synchronization of trigger signals, optimization of delays for latching data, determination of detector thresholds, gains and responses, evaluation of the perimeter of the South Atlantic Anomaly (SAA), measurements of live time, of absolute time, and internal and spacecraft boresight alignments. Here we describe on–orbit calibration results obtained using known astrophysical sources, galactic cosmic rays, and charge injection into the front-end electronics of each detector. Instrument response functions will be described in a separate publication. This paper demonstrates the stability of calibrations and describes minor changes observed since launch. These results have been used to calibrate the LAT datasets to be publicly released in August 2009.' --- [**The On–orbit calibration of the [*Fermi*]{} Large Area Telescope**]{} *The Fermi LAT Collaboration* A. A. Abdo$^{1,2}$, M. Ackermann$^{3}$, M. Ajello$^{3}$, J. Ampe$^{2}$, B. Anderson$^{4}$, W. B. Atwood$^{4}$, M. Axelsson$^{5,6}$, R. Bagagli$^{7}$, L. Baldini$^{7}$, J. Ballet$^{8}$, G. Barbiellini$^{9,10}$, J. Bartelt$^{3}$, D. Bastieri$^{11,12}$, B. M. Baughman$^{13}$, K. Bechtol$^{3}$, D. Bédérède$^{14}$, F. Bellardi$^{7}$, R. Bellazzini$^{7}$, F. Belli$^{15,16}$, B. Berenji$^{3}$, D. Bisello$^{11,12}$, E. Bissaldi$^{17}$, E. D. Bloom$^{3}$, G. Bogaert$^{18}$, J. R. Bogart$^{3}$, E. Bonamente$^{19,20}$, A. W. Borgland$^{3}$, P. Bourgeois$^{14}$, A. Bouvier$^{3}$, J. Bregeon$^{7}$, A. Brez$^{7}$, M. Brigida$^{21,22}$, P. Bruel$^{18}$, T. H. Burnett$^{23}$, G. Busetto$^{11,12}$, G. A. Caliandro$^{21,22}$, R. A. Cameron$^{3}$, M. Campell$^{3}$, P. A. Caraveo$^{24}$, S. Carius$^{25}$, P. Carlson$^{5,26}$, J. M. Casandjian$^{8}$, E. Cavazzuti$^{27}$, M. Ceccanti$^{7}$, C. Cecchi$^{19,20}$, E. Charles$^{3}$, A. Chekhtman$^{28,2}$, C. C. Cheung$^{29}$, J. Chiang$^{3}$, R. Chipaux$^{30}$, A. N. Cillis$^{29}$, S. Ciprini$^{19,20}$, R. Claus$^{3}$, J. Cohen-Tanugi$^{31}$, S. Condamoor$^{3}$, J. Conrad$^{5,26,32}$, R. Corbet$^{29}$, S. Cutini$^{27}$, D. S. Davis$^{29,33}$, M. DeKlotz$^{34}$, C. D. Dermer$^{2}$, A. de Angelis$^{35}$, F. de Palma$^{21,22}$, S. W. Digel$^{3}$, P. Dizon$^{36}$, M. Dormody$^{4}$, E. do Couto e Silva$^{3}$[^1], P. S. Drell$^{3}$, R. Dubois$^{3}$, D. Dumora$^{37,38}$, Y. Edmonds$^{3}$, D. Fabiani$^{7}$, C. Farnier$^{31}$, C. Favuzzi$^{21,22}$, E. C. Ferrara$^{29}$, O. Ferreira$^{18}$, Z. Fewtrell$^{2}$, D. L. Flath$^{3}$, P. Fleury$^{18}$, W. B. Focke$^{3}$, K. Fouts$^{3}$, M. Frailis$^{35}$, D. Freytag$^{3}$, Y. Fukazawa$^{39}$, S. Funk$^{3}$, P. Fusco$^{21,22}$, F. Gargano$^{22}$, D. Gasparrini$^{27}$, N. Gehrels$^{29,40}$, S. Germani$^{19,20}$, B. Giebels$^{18}$, N. Giglietto$^{21,22}$, F. Giordano$^{21,22}$, T. Glanzman$^{3}$, G. Godfrey$^{3}$, J. Goodman$^{3}$, I. A. Grenier$^{8}$, M.-H. Grondin$^{37,38}$, J. E. Grove$^{2}$, L. Guillemot$^{37,38}$, S. Guiriec$^{31}$, M. Hakimi$^{3}$, G. Haller$^{3}$, Y. Hanabata$^{39}$, P. A. Hart$^{3}$, P. Hascall$^{41}$, E. Hays$^{29}$, M. Huffer$^{3}$, R. E. Hughes$^{13}$, G. Jóhannesson$^{3}$, A. S. Johnson$^{3}$, R. P. Johnson$^{4}$, T. J. Johnson$^{29,40}$, W. N. Johnson$^{2}$, T. Kamae$^{3}$, H. Katagiri$^{39}$, J. Kataoka$^{42}$, A. Kavelaars$^{3}$, H. Kelly$^{3}$, M. Kerr$^{23}$, W. Klamra$^{5,26}$, J. Knödlseder$^{43}$, M. L. Kocian$^{3}$, F. Kuehn$^{13}$, M. Kuss$^{7}$, L. Latronico$^{7}$, C. Lavalley$^{31}$, B. Leas$^{2}$, B. Lee$^{41}$, S.-H. Lee$^{3}$, M. Lemoine-Goumard$^{37,38}$, F. Longo$^{9,10}$, F. Loparco$^{21,22}$, B. Lott$^{37,38}$, M. N. Lovellette$^{2}$, P. Lubrano$^{19,20}$, D. K. Lung$^{41}$, G. M. Madejski$^{3}$, A. Makeev$^{28,2}$, B. Marangelli$^{21,22}$, M. Marchetti$^{15,16}$, M. M. Massai$^{7}$, D. May$^{2}$, G. Mazzenga$^{15,16}$, M. N. Mazziotta$^{22}$, J. E. McEnery$^{29}$, S. McGlynn$^{5,26}$, C. Meurer$^{5,32}$, P. F. Michelson$^{3}$, M. Minuti$^{7}$, N. Mirizzi$^{21,22}$, P. Mitra$^{3}$, W. Mitthumsiri$^{3}$, T. Mizuno$^{39}$, A. A. Moiseev$^{44}$, M. Mongelli$^{22}$, C. Monte$^{21,22}$, M. E. Monzani$^{3}$, E. Moretti$^{9,10}$, A. Morselli$^{15}$, I. V. Moskalenko$^{3}$, S. Murgia$^{3}$, D. Nelson$^{3}$, L. Nilsson$^{25,45}$, S. Nishino$^{39}$, P. L. Nolan$^{3}$, E. Nuss$^{31}$, M. Ohno$^{46}$, T. Ohsugi$^{39}$, N. Omodei$^{7}$, E. Orlando$^{17}$, J. F. Ormes$^{47}$, M. Ozaki$^{46}$, A. Paccagnella$^{11,48}$, D. Paneque$^{3}$, J. H. Panetta$^{3}$, D. Parent$^{37,38}$, V. Pelassa$^{31}$, M. Pepe$^{19,20}$, M. Pesce-Rollins$^{7}$, P. Picozza$^{15,16}$, M. Pinchera$^{7}$, F. Piron$^{31}$, T. A. Porter$^{4}$, S. Rainò$^{21,22}$, R. Rando$^{11,12}$, E. Rapposelli$^{7}$, W. Raynor$^{2}$, M. Razzano$^{7}$, A. Reimer$^{3}$, O. Reimer$^{3}$, T. Reposeur$^{37,38}$, L. C. Reyes$^{49}$, S. Ritz$^{29,40}$, S. Robinson$^{50,23}$, L. S. Rochester$^{3}$, A. Y. Rodriguez$^{51}$, R. W. Romani$^{3}$, M. Roth$^{23}$, F. Ryde$^{5,26}$, A. Sacchetti$^{22}$, H. F.-W. Sadrozinski$^{4}$, N. Saggini$^{7}$, D. Sanchez$^{18}$, L. Sapozhnikov$^{3}$, O. H. Saxton$^{3}$, P. M. Saz Parkinson$^{4}$, A. Sellerholm$^{5,32}$, C. Sgrò$^{7}$, E. J. Siskind$^{52}$, D. A. Smith$^{37,38}$, P. D. Smith$^{13}$, G. Spandre$^{7}$, P. Spinelli$^{21,22}$, J.-L. Starck$^{8}$, T. E. Stephens$^{29}$, M. S. Strickman$^{2}$, A. W. Strong$^{17}$, M. Sugizaki$^{3}$, D. J. Suson$^{53}$, H. Tajima$^{3}$, H. Takahashi$^{39}$, T. Takahashi$^{46}$, T. Tanaka$^{3}$, A. Tenze$^{7}$, J. B. Thayer$^{3}$, J. G. Thayer$^{3}$, D. J. Thompson$^{29}$, L. Tibaldo$^{11,12}$, O. Tibolla$^{54}$, D. F. Torres$^{55,51}$, G. Tosti$^{19,20}$, A. Tramacere$^{56,3}$, M. Turri$^{3}$, T. L. Usher$^{3}$, N. Vilchez$^{43}$, N. Virmani$^{36}$, V. Vitale$^{15,16}$, L. L. Wai$^{3,57}$, A. P. Waite$^{3}$, P. Wang$^{3}$, B. L. Winer$^{13}$, D. L. Wood$^{2}$, K. S. Wood$^{2}$, H. Yasuda$^{39}$, T. Ylinen$^{25,5,26}$, M. Ziegler$^{4}$ 1. National Research Council Research Associate 2. Space Science Division, Naval Research Laboratory, Washington, DC 20375 3. W. W. Hansen Experimental Physics Laboratory, Kavli Institute for Particle Astrophysics and Cosmology, Department of Physics and SLAC National Laboratory, Stanford University, Stanford, CA 94305 4. Santa Cruz Institute for Particle Physics, Department of Physics and Department of Astronomy and Astrophysics, University of California at Santa Cruz, Santa Cruz, CA 95064 5. The Oskar Klein Centre for Cosmo Particle Physics, AlbaNova, SE-106 91 Stockholm, Sweden 6. Department of Astronomy, Stockholm University, SE-106 91 Stockholm, Sweden 7. Istituto Nazionale di Fisica Nucleare, Sezione di Pisa, I-56127 Pisa, Italy 8. Laboratoire AIM, CEA-IRFU/CNRS/Université Paris Diderot, Service d’Astrophysique, CEA Saclay, 91191 Gif sur Yvette, France 9. Istituto Nazionale di Fisica Nucleare, Sezione di Trieste, I-34127 Trieste, Italy 10. Dipartimento di Fisica, Università di Trieste, I-34127 Trieste, Italy 11. Istituto Nazionale di Fisica Nucleare, Sezione di Padova, I-35131 Padova, Italy 12. Dipartimento di Fisica “G. Galilei", Università di Padova, I-35131 Padova, Italy 13. Department of Physics, Center for Cosmology and Astro-Particle Physics, The Ohio State University, Columbus, OH 43210 14. IRFU/Dir, CEA Saclay, 91191 Gif sur Yvette, France 15. Istituto Nazionale di Fisica Nucleare, Sezione di Roma “Tor Vergata", I-00133 Roma, Italy 16. Dipartimento di Fisica, Università di Roma “Tor Vergata", I-00133 Roma, Italy 17. Max-Planck Institut für extraterrestrische Physik, 85748 Garching, Germany 18. Laboratoire Leprince-Ringuet, École polytechnique, CNRS/IN2P3, Palaiseau, France 19. Istituto Nazionale di Fisica Nucleare, Sezione di Perugia, I-06123 Perugia, Italy 20. Dipartimento di Fisica, Università degli Studi di Perugia, I-06123 Perugia, Italy 21. Dipartimento di Fisica “M. Merlin" dell’Università e del Politecnico di Bari, I-70126 Bari, Italy 22. Istituto Nazionale di Fisica Nucleare, Sezione di Bari, 70126 Bari, Italy 23. Department of Physics, University of Washington, Seattle, WA 98195-1560 24. INAF-Istituto di Astrofisica Spaziale e Fisica Cosmica, I-20133 Milano, Italy 25. School of Pure and Applied Natural Sciences, University of Kalmar, SE-391 82 Kalmar, Sweden 26. Department of Physics, Royal Institute of Technology (KTH), AlbaNova, SE-106 91 Stockholm, Sweden 27. Agenzia Spaziale Italiana (ASI) Science Data Center, I-00044 Frascati (Roma), Italy 28. George Mason University, Fairfax, VA 22030 29. NASA Goddard Space Flight Center, Greenbelt, MD 20771 30. IRFU/SEDI, CEA Saclay, 91191 Gif sur Yvette, France 31. Laboratoire de Physique Théorique et Astroparticules, Université Montpellier 2, CNRS/IN2P3, Montpellier, France 32. Department of Physics, Stockholm University, AlbaNova, SE-106 91 Stockholm, Sweden 33. University of Maryland, Baltimore County, Baltimore, MD 21250 34. Stellar Solutions Inc., 250 Cambridge Avenue, Suite 204, Palo Alto, CA 94306 35. Dipartimento di Fisica, Università di Udine and Istituto Nazionale di Fisica Nucleare, Sezione di Trieste, Gruppo Collegato di Udine, I-33100 Udine, Italy 36. ATK Space Products, Beltsville, MD 20705 37. CNRS/IN2P3, Centre d’Études Nucléaires Bordeaux Gradignan, UMR 5797, Gradignan, 33175, France 38. Université de Bordeaux, Centre d’Études Nucléaires Bordeaux Gradignan, UMR 5797, Gradignan, 33175, France 39. Department of Physical Science and Hiroshima Astrophysical Science Center, Hiroshima University, Higashi-Hiroshima 739-8526, Japan 40. University of Maryland, College Park, MD 20742 41. Orbital Network Engineering, 10670 North Tantau Avenue, Cupertino, CA 95014 42. Department of Physics, Tokyo Institute of Technology, Meguro City, Tokyo 152-8551, Japan 43. Centre d’Étude Spatiale des Rayonnements, CNRS/UPS, BP 44346, F-30128 Toulouse Cedex 4, France 44. Center for Research and Exploration in Space Science and Technology (CRESST), NASA Goddard Space Flight Center, Greenbelt, MD 20771 45. Mätfakta i Kalmar AB, 30477 Kalmar, Sweden 46. Institute of Space and Astronautical Science, JAXA, 3-1-1 Yoshinodai, Sagamihara, Kanagawa 229-8510, Japan 47. Department of Physics and Astronomy, University of Denver, Denver, CO 80208 48. Dipartimento di Ingegneria dell’Informazione, Università di Padova, I-35131 Padova, Italy 49. Kavli Institute for Cosmological Physics, University of Chicago, Chicago, IL 60637 50. Current address: Pacific Northwest National Laboratory, Richland, WA 99352 51. Institut de Ciencies de l’Espai (IEEC-CSIC), Campus UAB, 08193 Barcelona, Spain 52. NYCB Real-Time Computing Inc., Lattingtown, NY 11560-1025 53. Department of Chemistry and Physics, Purdue University Calumet, Hammond, IN 46323-2094 54. Max-Planck-Institut für Kernphysik, D-69029 Heidelberg, Germany 55. Institució Catalana de Recerca i Estudis Avançats (ICREA), Barcelona, Spain 56. Consorzio Interuniversitario per la Fisica Spaziale (CIFS), I-10133 Torino, Italy 57. Current address: Yahoo! Inc., Sunnyvale, CA 94089 [**Keywords:**]{} GLAST, Fermi, LAT, gamma-ray, calibrations [**PACS classification codes:**]{} 07.87.+v; 95.55.Ka Introduction {#sec:intro} ============ The Fermi Gamma–ray Space Telescope, hereafter [*Fermi*]{}, represents the next generation of satellite–based high-energy gamma-ray observatory. The [*Fermi*]{} satellite hosts two instruments: the Large Area Telescope (LAT) [@latpaper] and the Gamma-ray Burst Monitor (GBM) [@gbmpaper]. The former employs a pair-conversion technique to measure photons from 20 MeV to energies greater than 300 GeV, while the latter uses NaI and BGO scintillation counters to record transient phenomena in the sky in the energy range from 8 keV to 40 MeV. The LAT has no consumables, and a very stable response unlike its predecessor, the Energetic Gamma Ray Emission Telescope (EGRET) [@egret]. For the energy range above 10 GeV the sensitivity of the LAT is at least one order of magnitude greater than that of EGRET, allowing the sky to be explored at these energies essentially for the first time [@latpaper]. The LAT consists of a tracker/converter (TKR) for direction measurements [@Atwood:2007ra; @ssd; @Johnson98; @Baldini:2006pv], followed by a calorimeter (CAL) for energy measurements [@cal]. Sixteen TKR and CAL modules are combined to form sixteen towers, which are assembled in a 4$\times$4 mechanical support structure. An anticoincidence detector (ACD), enclosed by a micrometeoroid shield, surrounds the TKRs and rejects charged cosmic-ray background [@acdover; @acddet]. The LAT has about one million detector readout channels. On-orbit calibrations relate to all aspects of LAT measurements and data analysis results, from absolute timing to energy and direction measurements for individual events, to fluxes and positions of gamma-ray sources. The accurate timestamps of the LAT are obtained using the Global Positioning System (GPS) of the [*Fermi*]{} spacecraft, which provides timing and position information. Those are needed for phase folding pulsars and correlating gamma-ray observations with those at other wavelengths. As the number of photons in an observation increases, the centroid of their spatial distribution becomes better measured, and eventually the error is dominated by uncertainties in the alignment of the LAT, both internal and with respect to the [*Fermi*]{} spacecraft. Source localization at GeV energies enables the LAT to resolve bright, adjacent sources previously labeled as unidentified [@egret] and will help elucidate the origin of gamma-ray emissions from galactic cosmic rays accelerated in supernova remnants. The energy calibrations at higher energies are of utmost importance for detection of dark matter particle signals. Some extensions to the Standard Model predict narrow spectral lines due to the annihilations of as-yet unknown massive particles. For detecting and characterizing these features, accurate energy determination is vital. Even though the LAT was designed to measure gamma-rays it can also study, though not separately, the cosmic-ray electron and positron spectra. Cosmic-ray electron and positron spectra and intensities may also contain signatures for new physics. In this case, energy calibrations and position determination using extrapolated tracks into the CAL play an important role. At lower energies, broad features in the photon energy spectra of active galactic nuclei, or supernovae remnants originating from pion decays and bremsstrahlung may help unravel outstanding questions concerning particle acceleration in these sources. The purpose of this paper is to document the on-orbit calibration procedures used by the LAT; it begins with an overview of calibrations in Section \[sec:overview\]. Details on trigger, ACD, CAL and TKR calibrations are described in Sections \[sec:daq\], \[sec:acd\], \[sec:cal\] and \[sec:tkr\], respectively. The evaluation and updates to the perimeter of the South Atlantic Anomaly (SAA) are presented in Section \[sec:saa\] and measurements of live time are discussed in Section \[sec:livetime\]. The results from absolute timing follow in Section \[sec:absolute\]. Finally, internal and spacecraft boresight alignments are explained in Section \[sec:ta\]. We conclude with a table that summarizes the calibration results in Section \[sec:conclusion\]. Assessment of the current LAT performance is described in a separate publication [@perfpaper], while tests performed at particle accelerators are presented elsewhere [@bt97; @bt99; @bt06]. The section describing each calibration in shown in the last column of Table \[table:calib\]. Overview of LAT calibrations {#sec:overview} ============================ For this paper, the word calibration represents synchronization of trigger signals, optimization of delays for latching data, determination of detector thresholds, gains and responses, evaluation of the perimeter of the South Atlantic Anomaly (SAA), measurements of live time and of absolute time and internal and spacecraft boresight alignments. Table \[table:calib\] summarizes all calibration types, classified by category. Configuration refers to operational settings that define the state of the hardware, which are fixed before data are acquired. Calibrations are related to quantities that can change after data are processed and analyzed. As shown in Table \[table:calib\], for some types of calibrations, updates are not as frequent as the acquisition of the relevant data. ----------- ---------------------------- -------- -------------- -------------- ------------------- -- Title Type Frequency to Frequency of Sec. acquire data updates Trigger Time coincidence window config 1 year 1 year \[sec:tsync\] Trigger Fast trigger delays config 1 year 1 year \[sec:tsync\] Trigger Delays for latching data config 1 year 1 year \[sec:tsync\] ACD Pedestal both continuous 3 months \[sec:acdpeds\] ACD Coherent noise calib 3 months 3 months \[sec:acdpeds\] ACD MIP peak calib continuous 3 months \[sec:acdmip\] ACD High range (CNO) calib continuous 3 months \[sec:acdmip\] ACD Veto threshold config 3 months 3 months \[sec:veto\] ACD High level discriminator config 3 months 3 months \[sec:veto\] CAL Pedestal both continuous 3 months \[sec:calpeds\] CAL Electronics linearity calib 3 months 3 months \[sec:callin\] CAL Energy scales calib continuous 6 months \[sec:callin\] CAL Light asymmetry calib continuous 6 months \[sec:calpos\] CAL Zero-suppression threshold config continuous 3 months \[sec:calthresh\] CAL Low-energy threshold config continuous 6 months \[sec:calthresh\] CAL High-energy threshold config continuous 6 months \[sec:calthresh\] CAL Upper level discriminators config continuous 1 year \[sec:calthresh\] TKR Noisy channels config continuous 3 months \[sec:tkrch\] TKR Trigger threshold config 3 months 3 months \[sec:tkrdac\] TKR Data latching threshold calib 3 months 3 months \[sec:tkrdac\] TKR ToT conversion parameters calib 1 year 1 year \[sec:tkrtot\] TKR MIP scale calib continuous 3 months \[sec:tkrmip\] SAA SAA polygon config continuous 1 year \[sec:saa\] Timing LAT timestamps calib continuous continuous \[sec:absolute\] Alignment Intra tower calib continuous 1 year \[sec:intra\] Alignment Inter tower calib continuous 1 year \[sec:inter\] Alignment LAT boresight calib continuous 1 year \[sec:sc\] ----------- ---------------------------- -------- -------------- -------------- ------------------- -- : List of LAT calibrations (calib) and configurations (config) that can impact LAT scientific results. Note that pedestals are used both as a configuration on-board and as a calibration on the ground. Detailed descriptions are given in the sections listed in the last column. Frequency of updates correspond to current best estimates for the period beyond the first year of operations.[]{data-label="table:calib"} Operationally, calibration data are acquired in two distinct modes (see Section \[sec:tmodes\]): 1. Dedicated, meaning that the trigger, detector and software filter settings are incompatible with nominal science data taking. 2. Continuous, meaning the trigger and software can, with only a small penalty in live time, acquire specialized data that is used to calibrate, or, more generally, monitor the performance of the LAT during nominal science data-taking. Within a run the LAT acquires data with fixed instrument configurations. For the most part, the dedicated calibration runs are concerned with characterizing the electronics’ response to known stimuli while the continuous calibrations are aimed at calibrating the electronics with a known physics input. The stability of calibrations has been such that operations in dedicated-mode amount to approximately 2.5 hours every three months. Sea-level cosmic ray muons were used to calibrate the low-energy scales and trigger thresholds, but muons do not deposit enough energy in the detector elements to calibrate the high-energy scales and high-energy trigger thresholds. Instead, we used charge injection into the front-end electronics to calibrate the high-energy scales. Because of rise-time slewing effects, the optimal synchronization of trigger signals and optimal delays for data latching are energy dependent. We used pre-launch tests to provide a best approximation of the optimal trigger timing, and verified and corrected the synchronization and delays with on-orbit data. All these calibrations are revisited with on-orbit data selected from galactic cosmic rays and, as it will be shown later, there are only minor deviations when comparing results prior and after launch. Overview of trigger and readout {#sec:daq} =============================== The trigger and data readout system controls the composition and flow of data from the source in the detector elements to the Solid State Recorder (SSR) of the [*Fermi*]{} spacecraft. There are two distinct stages: the latching and movement of data from the front-end detector elements to a LAT Event Processing Unit (EPU), and the processing of the data on the EPU and its subsequent transfer to the SSR. The first stage is controlled by hardware, namely the trigger, and the second stage by software. A full description of the LAT multilevel trigger and data readout can be found elsewhere [@latpaper; @tdaqpaper]. Acquisition modes: dedicated or continuous {#sec:tmodes} ------------------------------------------ During dedicated calibration runs the trigger system and detector electronics are configured to acquire data useful for calibrating the thresholds and responses of detectors and synchronizing the arrival times of signals from various parts of the detectors. We call these runs dedicated because the trigger and detector configurations necessary to acquire these specialized data are incompatible with acquiring our primary science data (i.e. high-energy gamma rays from celestial sources). During continuous calibration acquisitions, the trigger system and detector electronics are configured to detect and latch not only those events thought to be gamma rays, but also those useful for calibration. As used on orbit, the LAT trigger system takes inputs from the ACD, TKR, and CAL front-end electronics and from a programmable, internal periodic trigger. Under control of LAT flight software, the periodic trigger can be used with a programmable charge-injection system to calibrate the detector electronics, or it can be used simply to read out the detector front-ends at a specified cadence. The programmed control under the charge injection system is used only in dedicated calibration runs. There the trigger is configured to collect a specified number of events at a particular rate, regardless of input from the detectors. In most cases, the trigger is configured to instruct the detector electronics to inject a known, programmable amount of charge at a specified time relative to each trigger. In other cases, no charge is injected. For each sequence of these events, flight software sets configuration registers throughout the instrument that control, for example, the amount of charge injected. By collecting a number of events with the same injected charge, and then varying the amount of injected charge, the electronic response can be accurately calibrated. Other dedicated calibration runs and all continuous calibration runs have the instrument triggered by signals from the detectors, not a programmed sequence, and charge injection is not used. For example, dedicated calibration runs in this mode are those used to synchronize trigger signals from the detectors. Data are collected using a sequence of configurations that sweep values through the various registers that control the delay times of the trigger signals. The data captured are then analyzed on the ground to identify the delay settings that best synchronize the arrival times of the trigger signals from each detector system. During continuous calibration runs, the trigger is configured to detect and latch both events thought to be gamma rays and other events useful for calibration. Based on the signals recieved from the detectors, the trigger specifies the type of ACD and CAL readout. There are four types (see Table \[table:cont\]) formed by enabling or disabling zero-suppression, in which values are committed to the data stream only if they exceed a programmable threshold, and selecting CAL single-range or four-range mode, in which the electronics takes respectively only the “best” range or all four values (see Section \[sec:calthresh\] for details). The ACD does not support a readout mode to select both readout ranges. It does, however, respect the zero-suppress mode. Purpose Readout Mode --------- --------------------- -- Zero-suppressed and best range Zero-suppressed and four-range Zero-suppressed and best range Non-zero-suppressed and four-range : Readout modes used for continuous calibrations (nominal science operations).[]{data-label="table:cont"} After an event passes the hardware trigger it is inspected by on-board software filters, each configured to identify events likely to be useful for one or more scientific or calibration purposes. If any filter accepts an event, it is included in the LAT data stream and forwarded to the SSR for transmission to the ground. In addition, the filters can be configured to allow a fraction of events that would otherwise have been rejected to be included in the data stream. Such events are used primarily to study filter performance. The input trigger rate of about 2.2 kHz, averaged over many orbits, is reduced to about 450 Hz by the software filters. Table \[table:obf\] describes filter types, purpose and average output rate. Filter Type Purpose Average Rate (Hz) ------------- ------------------------------------ --------------------- -- select gamma-ray candidates 410 and events $>$ 20 GeV Heavy ion calibration of high-energy scales, 2.5 select non-interacting protons 0 (nominal) 10 (dedicated-mode) Diagnostic filter performance, background 22 : On-board filters used to select events for calibration acquisitions. Rates depend on geomagnetic and other orbital variations. Here we list average rates.[]{data-label="table:obf"} Synchronization of trigger signals and delays for latching data {#sec:tsync} --------------------------------------------------------------- Fast trigger signals ($\sim$ few hundred ns) from the detectors must be synchronized with respect to each other for the trigger to operate efficiently. There are five such signals: veto and high level discriminator in the ACD (see Section \[sec:acd\]), low and high-energy in the CAL (see Section \[sec:cal\]) and TKR (see Sec \[sec:tkr\]). The earliest arriving trigger signal initiates a readout. The arrival times of the other trigger signals relative to this signal are captured in the event data with 50 ns precision (i.e. the period of the LAT system clock), allowing a direct comparison of trigger signals on an event-by-event basis. Once the relative timing is determined, the settings of various delay registers (i.e. the instrument configuration) are modified to synchronize these trigger signals in subsequent data acquisitions. There are two types of delay associated with the trigger: the fast trigger signal delay and the delay for latching the data. To synchronize the detectors we use the time of arrival of trigger signals, referred to as condition arrival times. This is event information containing the number of 50 ns clock ticks that have passed between the opening of the coincidence time window and the arrival of a given trigger signal. Since the trigger signals for TKR and CAL have significant time walk depending on the ratio of signal size to threshold value, it is important to choose an appropriate dataset to optimize the timing. We select MIP data for the relative timing of ACD and TKR, since the main purpose of the ACD is to reject charged particle background. The relative timing between TKR and CAL is optimized using photon candidates selected by the gamma filter. The CAL low and high-energy signals are controlled by a single delay and cannot be tuned independently. Figure \[fig:treq\] shows the arrival times for CAL with respect to the TKR. Negative values indicate that the TKR trigger arrived earlier than the CAL trigger signal. Distributions are fully contained inside the time coincidence window[^2]of 700 ns (14 ticks). The arrival times are affected by the ratio of the crystal energy over the threshold, thus influencing the shape of the trigger signal curves. The spectrum is slightly altered from its original form due to a trigger event selection. There is no difference in arrival times between CAL high- and CAL low-energy triggers at energies very far above the two thresholds. ![On-orbit arrival times of the CAL low-energy (FLE) and high-energy (FHE) trigger signals with respect to the TKR trigger. This special dataset was acquired with the time coincidence window of 1550 ns (31 ticks). Most of the entries fall within the window used for nominal science operations (700 ns or $\pm$13 ticks).[]{data-label="fig:treq"}](condarrcaltkr1.eps){width="14.cm"} Measurement ACD Delay TKR Delay CAL Delay ------------- ------------------- ------------------ ----------- Ground 800 ns (16 ticks) 250 ns (5 ticks) 0 Orbit 750 ns (15 ticks) 200 ns (4 ticks) 0 : Trigger signal delays for measurements on the ground and on orbit.[]{data-label="table:treq"} On the ground, only muons were available for timing calibrations so a small change in parameters was observed on orbit. The results from the synchronization of the trigger signals are shown in Table \[table:treq\]. The optimal delay is obtained after analyzing data acquired with fixed delay values. For the CAL we fit the MIP peak for each dataset with a fixed delay. Figure \[fig:tack\]a shows an example for a delay setting of 750 ns (25 ticks). Figure \[fig:tack\]b shows CAL MIP peak positions for six delay values for latching data. The optimal setting is the peak position obtained from the fit to the data. There is a data point slightly off the curve due to changes in geomagnetic conditions along the orbit. Since data were not recorded at the same location in orbit, the MIP selection cuts are designed to keep variations to $<$ 0.1% of the MIP peak value. A similar procedure is applied to the ACD. For the TKR, the quantity of interest is not the MIP peak position but instead the detector efficiency, which is defined as the number of layer hits between the first and the last hit of the track divided by the expected number of layers crossed by the track. The optimal[^3]settings for the on-orbit delays for latching data are 200 ns (4 ticks), 0 and 2450 ns (49 ticks) for the ACD, CAL and TKR delays, respectively. ![a) Energy deposited in a single CAL crystal (averaged over all towers) for a fixed value of the delay for latching data. The fit (curve) is used to determine the MIP peak position in the CAL, b) MIP peak positions for different CAL delay values (all towers). The curve is a parabola fit to the data.[]{data-label="fig:tack"}](tkrcaldelay.eps){width="14.cm"} ACD calibrations {#sec:acd} ================ The ACD is the LAT first-level defense against the charged particle cosmic-ray background that outnumbers the gamma-ray signals by 3-5 orders of magnitude. It consists of 97 separate plastic scintillating detectors - 89 scintillator tiles and 8 scintillator ribbons, each viewed by two photomultiplier tubes (PMTs) for redundancy. The overall ACD detection efficiency is $>$ 0.9997, which is provided by ensuring high uniformity of the detectors’ response and a large number of photoelectrons. The segmentation is needed in order to minimize pulse height variations over the ACD area and to minimize unwanted self-veto due to backsplash of soft photons from the developing electromagnetic shower in the CAL. Such self-veto caused a significant reduction in effective area in EGRET at high energies [@egret; @backsplash]. Detailed information about the ACD can be obtained elsewhere [@acdover; @acddet]. ACD calibrations include the determination of the mean values of pedestals, of the signal pulse heights produced by single MIP particles in each ACD scintillator, and the veto threshold settings; and the high-energy and coherent noise calibrations. All those parameters are determined for each PMT, so that each ACD tile or ribbon has two calibrated values. Every ACD channel has two ranges, low and high, in order to expand the dynamic range of processed signals. The low range covers the signals below 4-8 MIPs (depending on the channel), and the high range extends well above 1000 MIPs. The switching between ranges occurs automatically depending on the amplitude of the signal. In the future there may be changes to the MIP peak positions due to degradation of PMT photocathodes, scintillators or in an optical path between them. A possible way to mitigate these effects is to raise the high voltage for the PMTs. Since each high voltage is common to groups of 16 or 17 PMTs, this requires re-calibration of all channels belonging to that particular PMT group. Pedestals and coherent noise {#sec:acdpeds} ---------------------------- Pedestals are offset voltages present at the Analog-to-Digital Converter (ADC) inputs in the low and high range readouts. We extract the ACD pedestals for the low range readout from the 2 Hz periodic triggers, which currently provide approximately 10,000 random samples per orbit. Since a small fraction of these events contains particle signals or electronics noise or tails from previous signals in the ACD, we extract the pedestal values by performing a Gaussian fit to the central 80% of the pulse height distribution (truncated mean). Figure. \[fig:acdpeds\]a shows a typical pedestal distribution for a single PMT, where the width of the Gaussian (truncated) is about 2-4 pulse height bins ($\sim$ 0.01 MIPs). The narrow core of the distribution shows the intrinsic electronic noise; tails are residual signals from particles near in time to the periodic trigger. The small peak at about 250 is dominated by coherent noise contributions. ![a) Pedestal distribution from periodic triggers for a single PMT, b) Long term pedestal trending for a single PMT.[]{data-label="fig:acdpeds"}](calib_figure_ped_a.eps "fig:"){width="6.8cm"} ![a) Pedestal distribution from periodic triggers for a single PMT, b) Long term pedestal trending for a single PMT.[]{data-label="fig:acdpeds"}](calib_figure_ped_b.eps "fig:"){width="6.8cm"} Extracting the pedestal for the high range readout requires a special data-taking configuration, which forces a series of randomly triggered events to be read out in the high range. Since this configuration is incompatible with regular data taking and the pedestals are reasonably stable, these data are only acquired during the quarterly calibration periods where the LAT is in dedicated-mode. The data analysis is similar to that of the low range pedestals. Figure \[fig:acdpeds\]b shows the long term pedestal trending data for a single PMT. All values are plotted relative to the calibration being used, at the time of writing, in the offline reconstruction software (Mission week 25). To reduce data volume, we use the low range pedestal values on-board. We reject all signals less than 25 counts above pedestal, which corresponds to approximately 8 times the electronics noise, or 0.05 times the MIP signal in a typical tile. We also use pedestal values in the flight software data compression algorithm (for both ACD and CAL). In this case we reduce the data size by referencing signals against pedestal values rather than against zero. During ground testing of the ACD we discovered that the readout process causes the electronics pedestals to ring. These oscillations occur when elements of its internal circuitry resonate at their characteristic frequency. This reproducible effect can be quantified. Figure \[fig:acdcn\] shows the difference ($\Delta$ Pedestal) between the coherent noise and the regular pedestal values for a single PMT. This truncated mean (see Section \[sec:acdpeds\]) is displayed versus the time difference between consecutive readout cycles in 50 ns clock ticks. Data were obtained using periodic triggers and the error bars correspond to the RMS ADC values. The curve represents a fit to a sinusoidal oscillation inside a decaying exponential envelope, which falls to the electronics noise after about 200 $\mu$s. ![Readout related pedestal ringing in a single PMT. The truncated mean and RMS ADC values obtained using periodic triggers versus the time difference between consecutive readout cycles in 50 ns clock ticks. The curve corresponds to a fit of a sinusoidal oscillation inside a decaying exponential envelope.[]{data-label="fig:acdcn"}](calib_figure_cNoise.eps){width="13.cm"} As this effect is present in every channel of the ACD, events read out when the pedestal peaks at about 50 $\mu$s (1000 ticks) after the previous readout can lead to small signals in many PMTs. This correction is applied to the offline data to avoid overestimating the total energy in the ACD and compromising background rejection and photon selection. For events taken at the peak of the coherent oscillation ($\sim$ 500 ticks), this calibration reduces the coherent noise contribution from 0.05 MIPs per PMT to less that 0.005 MIPs per PMT. The effect is so small that we do not need to apply corrections to the on-board processing. Temperature dependences in ACD pedestals are negligible. MIP peak and high range calibrations {#sec:acdmip} ------------------------------------ The MIP peak values are determined using reconstructed tracks pointing at an ACD tile or ribbon that recorded a signal. The peak value of the pathlength corrected pulse-height distribution for each PMT is the MIP peak for that channel. This calibration is a heuristic attempt to quantify the average MIP signal seen in the ACD and not a precise determination of all details of the energy deposition in the ACD sensors. Figure \[fig:acdmip\]a shows a distribution of pedestal subtracted and pathlength corrected signals for a single PMT in the ACD. The MIP peak at about 600 is clearly seen and the peak below 100 corresponds to soft X-ray background. Figure \[fig:acdmip\]b displays the long term trending of the MIP peak for a single PMT, where pedestals from mission week 3 are used as a reference. The stability of the measurement is about 10% of the MIP peak for a period of 20 weeks. ![a) Distribution of pedestal subtracted and pathlength corrected signals for a single PMT in the ACD; the MIP peak is clearly visible and the curve is a fit to the data, b) Long term trending of the MIP peak versus mission week.[]{data-label="fig:acdmip"}](calib_figure_mip_a.eps "fig:"){width="6.8cm"} ![a) Distribution of pedestal subtracted and pathlength corrected signals for a single PMT in the ACD; the MIP peak is clearly visible and the curve is a fit to the data, b) Long term trending of the MIP peak versus mission week.[]{data-label="fig:acdmip"}](calib_figure_mip_b.eps "fig:"){width="6.8cm"} During the offline reconstruction we use the MIP peak calibration values to express raw signal pulse height measurements in MIP equivalent values. These are converted into energy using the energy deposition in the scintillator ($\sim$2 MeV/cm). MIP peaks in the ACD tiles occur between 400 and 1000 pulse height bins above pedestal, and are determined with an accuracy better than 5%. The high and low range readout are calibrated in an analogous way. The main difference is that for the high-range readout we use tracks identified as carbon nuclei by the CAL, since the proton MIP-like signals are too small. The algorithm requires the deposited energy in the CAL, which is pathlength corrected, to be consistent with that of carbon and the first few layers to have similar energy to avoid carbon interacting events. Also, in converting from raw pulse heights to MIP equivalent values we allow for non-linearities caused by signal saturation in the electronics and scintillators. Figure \[fig:acdrange\] shows the deposited energy for the full dynamic range of a single PMT. ![Deposited energy for the full dynamic range of a single PMT. The solid line shows events read out in the low range, the dotted line shows events read out in the high range. The MIP peak occurs $\sim$ 1.3 MIPs because pathlength corrections are not applied.[]{data-label="fig:acdrange"}](calib_figure_range_log.eps){width="12cm"} The solid line shows events read out in the low range and the dotted line shows events read out in the high range. Figure \[fig:acdcarbon\] shows a distribution of pedestal subtracted and pathlength corrected signals for carbon in a single PMT. ![Distribution of pedestal subtracted and pathlength corrected carbon signals for a single PMT in the ACD in the high range. The carbon peak is clearly visible at $\sim$200 counts and the curve shows the fit to the data.[]{data-label="fig:acdcarbon"}](calib_figure_carbon.eps){width="12cm"} Veto threshold and high-level discriminator {#sec:veto} ------------------------------------------- The on-board thresholds are controlled by DAC settings in the front-end electronics. These are calibrated by scanning three settings and measuring the resulting veto threshold in pulse height bins. We combine that information with the MIP peak to pulse height scale, or carbon peak for the high range, and set the veto and high-level discriminator thresholds as a fraction of the MIP and carbon signals, respectively. The veto thresholds for each PMT are set separately to an accuracy of about 0.01 MIPs ($\sim$ 20 keV) relative to the calibrated MIP peak. As shown in Fig \[fig:acdveto\], the veto turns on between 0.4-0.5 MIPs and the 50% efficiency point is close to 0.45 MIP. For events below $<$ 0.4 MIPs the ADC values can be artificially small, because the readout electronics are relatively slow when compared to the veto electronics. Events arriving when the electronics are overshooting the return to baseline tend to fire the veto discriminator, even though they are not expected to have any ADC counts. ![Veto turn-on curve for a single PMT. The small number of events (0.1%) $<$ 0.4 MIPs correspond to overshoots in the ADC value from a previous event.[]{data-label="fig:acdveto"}](calib_figure_veto_log.eps){width="14cm"} The turn-on for the high level discriminator occurs between 24-26 MIPs and the RMS width of the carbon peak is 20% of the mean value. We have not yet monitored the stability of the carbon peak since it needs large statistical samples. At least four months of data are required to obtain a carbon peak value with reasonable statistics. CAL calibrations {#sec:cal} ================ The CAL is designed to measure the energy of incident photons and charged particles, and to determine the direction and energy of photons and charged particles for which the TKR did not provide direction information, either because their trajectories did not cross the TKR or because they did not pair-produce in the TKR. Its imaging properties are also a key ingredient in seeding the track reconstruction process in TKR data analysis and in the rejection of charged-particle background [@latpaper]. The CAL consists of 16 identical modules. Each module is composed of 96 CsI(Tl) scintillation crystals arranged in a hodoscopic configuration with eight layers each containing 12 crystals. Each layer is rotated 90$^\circ$ with respect to its neighbors, forming an $x$-$y$ array. Crystals are read out by two dual-PIN-photodiode assemblies, one at each end, that measure the scintillation light produced in the crystal. Each photodiode assembly contains a large-area photodiode to measure small energy depositions and a small-area photodiode to measure large energy depositions. The active areas of the large and small diodes have a ratio of 6:1 with a spectral response well matched to the scintillation spectrum of CsI(Tl). Each of the 3072 photodiode assemblies is read out by an amplifier-discriminator ASIC, the GLAST Calorimeter Front-End Electronics (GCFE). To cover a large dynamic range of 5 $\times$ 10$^{5}$ in each GCFE with commercially available 12-bit ADCs for digitization, the low and high energy photodiodes each have their own independent signal chains, the low-energy and the high-energy, and each chain operates with two track and hold gains (low and high). This arrangement results in four overlapping energy ranges from 2 MeV to 70 GeV overall, as shown in Table \[table:calranges\]. Range overlap allows cross-calibration of the electronics. Table \[table:calranges\] also shows the approximate factors used to convert ADC readout to energy units (MeV). A detailed description of the CAL is found elsewhere [@cal]. Name Energy Gain Energy range MeV/ADC ------ -------- ------ ------------------ --------- LEX8 low high 2 MeV to 100 MeV 0.033 LEX1 low low 2 MeV to 1 GeV 0.30 HEX8 high high 30 MeV to 7 GeV 2.3 HEX1 high low 30 MeV to 70 GeV 20 : Four overlapping readout ranges for the CAL and their conversion factors.[]{data-label="table:calranges"} Here we describe the on-orbit measurements of the following: CAL pedestals; crystal energy scale, derived from the electronics linearity and crystal light output; crystal light asymmetry, which calibrates position measurements along the crystal; and threshold settings. These calibrations allow determination of the location and amount of energy deposited in each crystal. Processes for estimation of incident photon energy and the resulting overall energy resolution are discussed elsewhere [@perfpaper; @bt06]. Pedestals {#sec:calpeds} --------- As discussed in Section \[sec:acdpeds\], pedestals are offset voltages for each of the four CAL energy ranges that set the “zero point” for the energy scale. We measure pedestals on orbit from the periodic triggers issued at 2 Hz by the LAT trigger system during all nominal science data acquisitions. Chance coincidence energy deposits result in a small tail to the pedestal distribution, but this is suppressed both by the pedestal distribution fitting techniques used and by comparison of the various energy ranges. Figure \[fig:calpedwidth\] shows that typical pedestal widths (RMS) are 0.2 MeV for the LEX1 and LEX8 ranges and 7-10 MeV for the HEX1 and HEX8 ranges. The pedestal values are regularly monitored on orbit and have been extremely stable since an initial settling period. ![On-orbit pedestal widths (for all channels) for: a) LEX8, b) HEX8, c) LEX1 and d) HEX1 ranges.[]{data-label="fig:calpedwidth"}](calpedwidthall.eps){width="14cm"} Pedestal measurements made during thermal-vacuum tests in January 2008 indicated a channel-dependent linear dependence of pedestal position with temperature, where the drift magnitude varies from -3 to +3 ADC units per degree for the LEX8 and HEX8 ranges and $\sim$10 times smaller for LEX1 and HEX1, reflecting their smaller gains. When [*Fermi*]{} is in Pointed observing mode, the exposure of the LAT thermal control system to the warm Earth changes relative to the Sky Survey observing mode, and the LAT detector temperatures change modestly. In the CAL, the temperature changes 1–2 degrees within a few hours, leading to a temperature drift of the pedestals of a few ADC units at most. Since Log Accept (LAC) thresholds, which determine whether a given crystal readout is included in the data stream, are set in absolute ADC unit values (i.e. not relative to the pedestal), changes in the pedestal values change the corresponding threshold energies. A large enough pedestal change could thus influence data volume. However, only very large temperature changes ($\sim$ 20 degrees Celsius), which are not anticipated during the mission, could lead to significant increases in data volume. Individual crystal energy scales {#sec:callin} -------------------------------- Calibration of the individual crystal energy scales involves the determination of the parameters of a transfer function between the energy deposited in the crystal and the signal output in ADC units. The transfer function consists of the response of the front-end electronics and the light emission and collection properties of the scintillation crystals. We use charge injection calibrations to describe the electronics and ionization energy depositions from cosmic-ray muons or protons and various nuclei to calibrate crystal response. Ionization energy losses over a known path length are very predictable and hence make a good calibration tool. Charge injection calibrations are used to characterize the non-linear behavior of the electronics chain. A pulsed signal of known amplitude (controlled by a charge injection calibration DAC) is sent to the preamplifier input of each CAL front-end electronics channel. The nonlinearity of this DAC is specified by the manufacturer to be $<$0.1% and hence negligible when compared to the nonlinearity of the front-end electronics. For each fixed DAC setting we inject 100 pulses onto each of the electronics channels and average the resulting ADC output values. A spline function is used to fit the resulting DAC versus ADC curve. The functions (one for each channel), describe both electronics gain and non-linearity. Using these functions, signals from each channel can be converted to a linear scale. Figure \[fig:calnonlin\] shows the normalized ADC/DAC versus energy, where deviations from one indicate non-linear behavior for measurements made at fixed energy values. The largest deviations ($\sim$12%) are seen in the LEX1 range displayed in Figure \[fig:calnonlin\]a. Figure \[fig:calnonlin\]b illustrates the case for the HEX1 range with 4% non-linearities. The feature around 2 GeV in the HEX1 curve corresponds to cross-talk between LEX1, which saturates around 1 GeV, and HEX1. For the other ranges (LEX8 and HEX8) deviations are $<$1%. After applying the measured nonlinearity calibration, residual nonlinearity is $\le$ 1% of the measured energy, resulting in a negligible systematic effect in spectrum determination. ![Characterization of electronics non-linearities. Normalized ADC/DAC versus energy for: a) LEX1 and b) HEX1 ranges.[]{data-label="fig:calnonlin"}](nonlin_lex1.eps "fig:"){width="6.7cm"} ![Characterization of electronics non-linearities. Normalized ADC/DAC versus energy for: a) LEX1 and b) HEX1 ranges.[]{data-label="fig:calnonlin"}](nonlin_hex1.eps "fig:"){width="6.7cm"} The crystal response calibration, i.e. the function that relates deposited energy to the linearized signal described above, has been performed using different ionizing particles on the ground and in orbit. In both cases, ion incident energies are, for the most part, in the slow relativistic rise region of the Bethe-Bloch curve, so the predicted energy loss per unit path length ($\frac{dE}{dx}$) is only weakly dependent on incident energy. Using simulated incident spectra for ground and orbit environments, we determine an expected $\frac{dE}{dx}$ for each incident particle species. We then collect spectra of the these species, correcting each event for path length, and compare the peak position in pulse height units to the predicted position in energy to yield a calibration. Variations in incident spectrum with orbital position result in slight broadening of the energy deposit peaks, but, given integration over multiple orbits, the peak most-probable-value is well-determined and usable for calibration. On the ground, the low-energy scales were calibrated using sea-level cosmic ray muons while high energy ranges were calibrated using muons with the HEX1 and HEX8 channels set to a special “muon mode" gain setting that increased the gain by a factor of $\sim$10. On-orbit, we used a technique we refer to as “proton inter-range calibration”. The low-energy scales are calibrated using protons. Higher energies are calibrated by using energy deposits that meet two criteria: first they must be in the overlap range between LEX1 and HEX8 and second they must result in a “heavy ion” trigger, the only common trigger that produces the required 4 energy range readout rather than the normal single range readout. These events are a combination of galactic cosmic ray (GCR) primary carbon nuclei, interacting protons and other interacting or ionizing GCRs. From events that meet these criteria, we can construct a cross calibration of the low and high energy ranges. In both ground and on-orbit cases, the ionization calibration, together with the charge injection results yield a usable energy scale that converts ADC units to deposited energy. A week of nominal science operations data is sufficient to calibrate the energy scales using relativistic protons and heavy ion trigger events in the overlap energy range. Protons that are accepted by either the MIP filter (when active) or the diagnostic filter (see Table \[table:obf\]) are required to pass a number of cuts. In order to reject all events that are not contained in a single CAL module and to eliminate “corner clipping” events, for which path length determination is not sufficiently accurate, we require extrapolated TKR tracks to cross the top and the bottom surfaces of a single CAL crystal, and be at least 5 mm away from the crystal edges. In addition, we require single TKR track events with $>$ 20 TKR hits and a chi-square for the track $<$ 3. To reject low-energy re-entrant albedo protons that could broaden and bias the energy distribution, we require the multiple scattering angle, calculated by the Kalman filter used for TKR track reconstruction, to be $<$ 0.01. Finally, we reject nuclear-interacting protons by selecting events with two or fewer hits in each CAL layer and no additional crystals hit in the layer containing the crystal being calibrated. In order to determine a calibration peak shape that represents the data well but minimizes the number of free parameters, we use a two-step process. First, we produce a spectrum of path length corrected signals for all the crystals together. Each peak is fit with a Landau distribution convolved with a Gaussian, for which all parameters are left free. From this fit, we determine both the Landau and Gaussian widths, which are then fixed. In the second step, we fit the Gaussian-convolved Landau function to spectra from each crystal separately, allowing peak position and amplitude to vary but using the fixed widths determined above. The peak most probable value (MPV) in energy units from the simulations (10.6 MeV for protons) is divided by the peak MPV in pulse height units determined by the fitting process just described to yield the desired calibration quantity for each crystal. Figure \[fig:calproton\] shows the results, where the most probable value is the peak position. ![Energy deposited in a crystal (pathlength corrected). The position of the proton peak is given by the fit. The signal is corrected for electronics non-linearities.[]{data-label="fig:calproton"}](proton_peak_1chan.eps){width="15.cm"} Originally, we intended to use “heavy” GCR primary nuclei to calibrate the higher energy scales. This would be desirable since $\frac{dE}{dx}$ varies as $Z^2$ where $Z$ is the atomic number of the incident ion. The major difference between use of GCR heavy nuclei and protons or muons lies in a phenomenon known as “quenching”, in which the crystal light output per unit deposited energy ($\frac{dL}{dE}$) is thought to be less for higher $Z$ nuclei than for protons or muons. In addition to being $Z$ dependent, this phenomenon also depends on the ion incident energy. Since quenching is not well studied for the combination of $Z$ and incident energies relevant to on-orbit CAL calibration, we measured the response of the calorimeter CsI(Tl) crystals to relativistic nuclei (from carbon to iron) at the GSI facility in 2003 and 2006. The results of these studies indicated that for the ion energies examined at GSI, which were rather higher than those measured previously for CsI(Tl), $\frac{dL}{dE}$ was actually higher for measured nuclei with $6 \leq Z \leq 14$ than for protons [@lott2006]. Due to the lack of a physical model or understanding for this “anti-quenching” behavior, we felt that the systematic uncertainties introduced in using the heavy ion GCR calibration were unacceptable at this time and have used the proton inter-range technique instead. In order to study any possible changes in the energy scale calibration between pre and post launch measurements and during early on-orbit operations, we calculate, for each crystal, the ratio of the energy scale measured after launch to that measured prior to launch. A Gaussian fit to this distribution leads to a mean bias, for low-energy diodes, of $\sim$1% and a standard deviation which indicates crystal-to-crystal variations of 0.8%. The latter characterizes the statistical precision of this calibration procedure. For the high-energy diodes, the bias with respect to ground calibration is 5%, which is explained by the lack of any high-energy signal to reliably calibrate the ratio between the two diodes on the ground, as the muon signal is too small to be visible in the high-energy diode with flight gain settings. The comparison of on-orbit results from July and October 2008 shows a shift of only 1-2% and a spread, crystal-to-crystal, of $<$ 1%. The stability of our calibrations is demonstrated in Figure \[fig:calbias\]a for the low-energy diode and Figure \[fig:calbias\]b for the high-energy diode. ![Ratio of the energy scales for the on-orbit calibrations performed in October and July 2008: a) low-energy diodes (mean 1.014, sigma 0.009) and b) high-energy diodes (mean 1.019, sigma 0.009). []{data-label="fig:calbias"}](fig12a_newer_mpd_ratio_le.eps "fig:"){width="6.7cm"} ![Ratio of the energy scales for the on-orbit calibrations performed in October and July 2008: a) low-energy diodes (mean 1.014, sigma 0.009) and b) high-energy diodes (mean 1.019, sigma 0.009). []{data-label="fig:calbias"}](fig12b_newer_mpd_ratio_he.eps "fig:"){width="6.7cm"} Despite the fact that they are not suitably well understood for absolute calibrations, we do use energy deposits from from GCR heavy nuclei (500 MeV for carbon nuclei and 8 GeV for iron) for independent monitoring of the energy scale at high energies. The pathlength-corrected spectrum shown in Figure \[fig:heavyion\] is obtained by selecting crystal hits in low-multiplicity layers, thus rejecting nuclear interactions. Narrow peaks (the carbon peak has a 5% width) are easily identified. It is worth noting that the measured energies deposited by cosmic rays are consistent with the anti-quenching effect observed in the beam test data acquired at GSI. For example, the $\frac{dE}{dx}$ for carbon is observed to be about 20% higher than expected from a $Z^2$ scaling of the $\frac{dE}{dx}$ for protons. The count rate in the charge peaks is similar to that of the primary galactic cosmic-ray abundance, modified by loss of particles through charge-changing interactions above the CAL and by the decreasing efficiency of the on-board heavy-ion filter for higher $Z$ nuclei. ![Energy deposited in all crystals from heavy nuclei collected during 4 days of on-orbit operations. Pathlength corrections are applied.[]{data-label="fig:heavyion"}](heavyions.eps){width="11.5cm"} For example, Figure \[fig:carbonpeak\] shows that, although there is a slow systematic shift, the carbon peak position is stable to within 0.5% (for the whole CAL) after 2 months of operations. ![Position of carbon peak for 2 months of on-orbit data.[]{data-label="fig:carbonpeak"}](Cstability2.eps){width="11.5cm"} Light asymmetry {#sec:calpos} --------------- The design of the CAL crystals deliberately “tapers” the light propagation properties of the crystal so that we can determine the longitudinal position of an energy deposit by comparison of the signal at each crystal end. We refer to the measured quantity as the signal asymmetry, defined as the logarithm of the ratio between the signals read out at the two ends of the same crystal. To calibrate light asymmetry, we select non-interacting heavy nuclei in a similar way to protons (see Section \[sec:callin\]), but with looser TKR track quality. The TKR-determined position in a CAL crystal is obtained by extrapolating the TKR track to the center of the crystal (half-way through its thickness). Each track enters the crystal through one of the twelve evenly spaced bins defined along its length. The distribution of asymmetry signals for each bin is collected by computing the asymmetry for each event for the appropriate bin. The first and the last bins, near the crystal ends, are not used because of known non-uniformities in light collection near the diodes. The average asymmetry is calculated for each of the ten central bins along each crystal and is fit with a [*spline*]{} function for purposes of interpolation. We use the signals from the two ends of each crystal to obtain the weighted centroid of the energy depositions along the crystal. The position resolution along the crystal is defined as the RMS of the difference of the position from light asymmetry to that from track extrapolation. We calibrated the light asymmetry of each crystal on the ground with sea-level cosmic muons, and we recalibrated on orbit with GCRs. The calibration constants derived on orbit are more precise than those derived on the ground because the GCRs suffer less multiple scattering than muons and create larger scintillation signals. The position resolution measurement for energy depositions from 200 to 900 MeV, for both LEX1, and HEX8 ranges, improves from 4 mm to 2 mm and from 13 mm to 9 mm, respectively. Figure \[fig:posres\] shows position resolutions with pre- and post-launch constants in the energy range from 200 to 900 MeV. Results are dominated by the carbon events which peak $\sim$500 MeV. The improvement by using flight calibration constants is clearly seen. ![CAL position resolution from 200 to 900 MeV using ground calibration constants a) LEX1 and b) HEX8, and using flight calibrations for: c) LEX1 and d) HEX8.[]{data-label="fig:posres"}](posdev.eps){width="15cm"} Trigger thresholds and upper level discriminators {#sec:calthresh} ------------------------------------------------- To reduce the data volume generated by the CAL and the additional dead time that would be created by moving large events through the LAT data acquisition system, each GCFE has a zero-suppression discriminator with a programmable threshold DAC. This threshold, known as the log-accept (LAC) threshold, is nominally set to 2 MeV, which is approximately 10 times higher than the average electronic noise. The zero-suppression for an entire crystal is performed on the logical OR of the LAC discriminator states at the two ends; thus data from [*both*]{} ends of a crystal are included in the CAL data stream if the LAC discriminator on either end fires. Extensive ground testing with the LAT charge injection system and sea-level cosmic ray muons established the functionality, linearity, and energy scale for each LAC discriminator. We launched with LAC settings derived from the ground calibrations, but the LAC settings are temperature dependent, primarily because the pedestal values are temperature dependent. Thus we revised the LAC calibration constants on orbit once the LAT had achieved its stable operating temperature. For each discriminator, we characterize the relationship between DAC setting and LAC value (in MeV) with a linear model that is derived from calibration data acquired with the LAC set at two values near the nominal setting. We calibrate one end of all crystals at a time using a set of four configurations such that the LAC threshold at the crystal end [*not*]{} being calibrated is set to its maximum possible value, preventing it from initiating the readout, and the LAC threshold at the end being calibrated is set to the test value. This process give a LAC measurement for each channel with a statistical precision of 5% ($\sim$ 0.1 MeV). Figure \[fig:lac\] shows an example of the signals measured at one end of a crystal (LEX8 range) and the fit that determines the LAC threshold. The values can be easily converted into energy by using results from Table \[table:calranges\]. ![Signals from one end of a crystal (LEX8 range) and a fit used to determine the LAC threshold.[]{data-label="fig:lac"}](lac.eps){width="14cm"} We are monitoring four GCFEs with out-of-family electronic noise (out of 3072 in the CAL). In February 2009, we inhibited one of the channels from participating in zero-suppression decisions because its noise level reached 1.5 MeV. This has no impact on the scientific performance of the CAL. We monitor the stability of each LAC threshold in all nominal science operations data acquisitions. To measure the threshold value, we select events for which we are certain which discriminator qualified the crystal for inclusion in the data stream. Because the LAC value at each crystal end is within 10% of the setting at the opposite end, we achieve this certainty by selecting events where the signal from the two ends differs by more than 10%. The CAL provides two fast signals that participate in the formation of LAT trigger, the low energy CAL\_LO and high energy CAL\_HI triggers. The CAL\_LO and CAL\_HI trigger requests are formed as the logical OR of the outputs of the programmable fast low-energy and fast high-energy trigger discriminators, respectively FLE and FHE, at each end of a crystal. The nominal values for the FLE and FHE thresholds are 100 MeV and 1000 MeV per crystal, respectively, as measured at the center of the crystal by each GCFE. Extensive testing on the ground with the LAT charge injection system clearly demonstrated the functionality and linearity of each FLE and FHE discriminator; however it gave only an approximate absolute calibration (i.e. in MeV deposited) of the threshold DACs. The FLE and FHE thresholds are calibrated on orbit using background events recorded in dedicated-mode with additional information provided by the tower electronics module. However, this additional trigger diagnostic information only provides the logical OR of the combination of all 12 FLE or all 12 FHE discriminators for each CAL layer-end; thus it does not clearly identify which crystal end produced the trigger signal. Distinct procedures for FLE and FHE thresholds are necessary to resolve this 12-fold ambiguity. To calibrate the FLE discriminators near the nominal 100 MeV setting, we separately enable the trigger for each of two groups of six crystals in a layer (the six odd-numbered and the six even-numbered crystals) and require that five of the six enabled crystals have a signal below 50 MeV. Most showers share their energy between adjacent crystals in a layer, so the separation into two sets of six non-adjacent crystals readily resolves that ambiguity. The 50 MeV energy cut ensures that only one of the six is the source of the trigger signal. To calibrate the FHE discriminators, we allow only the CAL\_LO signal to initiate a LAT trigger, and we enable the FHE discriminators in two groups of six crystals per layer while we read the state of the FHE trigger diagnostic information. The efficiency of a discriminator as a function of signal (in MeV) is determined by calculating the ratio of the spectrum from events for which the discriminator fires to the spectrum for all events. As shown in Figure \[fig:caleff\], the value of the threshold is obtained by fitting a step function to this ratio. Having established that the FLE and FHE discriminators are linear in threshold DAC setting, we calibrate each discriminator at two settings near nominal value and fit the measurements with a linear model. We calibrate FLE at 100 MeV and 150 MeV, and we calibrate FHE at 1000 MeV and 1500 MeV. The statistical error in determining the threshold values is $<$ 1% for FLE and $<$ 2% for FHE thresholds. Figure \[fig:caleff\] illustrates how these efficiencies are obtained for the FLE and FHE thresholds for one GCFE. ![Efficiency versus energy for the FLE and FHE thresholds.[]{data-label="fig:caleff"}](fig17a_fle_new_noerrors.eps "fig:"){width="6.7cm"} ![Efficiency versus energy for the FLE and FHE thresholds.[]{data-label="fig:caleff"}](fig17b_fhe_new_noerrors.eps "fig:"){width="6.7cm"} These data are from dedicated calibration runs taken during early operations (July 2008), when thresholds were set using calibration coefficients derived from ground tests. As it happens, the ground calibration gave threshold values for FLE and FHE somewhat higher than intended, viz. $\sim$140 MeV and $\sim$1200 MeV, respectively. The ground FLE and FHE calibration relied on the charge-injection system, which gives pulse shapes that differ from those produced by CsI(Tl) scintillation signals. Since the trigger signal is fast ($\sim$ 250 ns) and the energy measurement depends on the slow shaper ($\sim$ 4000 ns), the difference in the shape of the pulses is important and creates the 40% and 20% bias we observed. We then adjusted the FLE and FHE settings using the calibration constants derived from the on-orbit calibration. We continue to monitor the FLE and FHE threshold values with data from the nominal science acquisitions. The programmable Upper Level Discriminator (ULD) in each GCFE is responsible for switching between CAL energy ranges. To select the best range for digitization, three ULDs in each GCFE compare the outputs of three ranges (LEX8, LEX1 and HEX8) with corresponding ULD threshold (one setting per GCFE). The output of these discriminators is analyzed by the range selection logic which selects the best range as the highest range without an ULD signal [@cal]. All three ULD thresholds of each crystal end are defined by one DAC and are set to $\sim$5% below the saturation level of the ADC. We calibrated the ULD threshold DACs on the ground with charge injection and verified those settings and the linear calibration model on orbit during nominal science continuous acquisitions. We measured the ULDs on orbit by finding, for each energy range, the highest observed signal for each individual crystal end. Most on-orbit data-taking configurations read out only one of the four available ranges, namely the one providing largest signal below the ADC saturation, but the nominal science data explore all four ranges fully. We updated the LAC, FLE, FHE, and ULD threshold settings after their initial on-orbit calibrations and later adjusted them to accommodate the small settling drift in pedestal values (see Section \[sec:calpeds\] for details). Figure \[fig:calthresholds\] shows the distribution of all thresholds for all channels. The LAC threshold data from August (dashed) and November 2008 (solid) are used in Figure \[fig:calthresholds\], where the effect of pedestal evolution since launch is seen as a slight broadening of the distribution. This effect is negligible for the other thresholds. The slight asymmetry in the distribution of ULD values (Figure \[fig:calthresholds\]d) has absolutely no effect on CAL performance; it means only that for a small fraction of channels the range switching will happen at a slightly lower energy than for the majority of channels. ![On-orbit measurements of threshold values for November 2008: a) LAC, b) FLE, c) FHE and d) ULD. The dashed line in the LAC histogram shows the effects of pedestal drifts seen in August 2008, prior to stabilization. The ULD thresholds are expressed in non-pedestal-subtracted ADC units. This makes it easier to judge how thresholds are set. The saturation limit corresponds to 4095 ACD units.[]{data-label="fig:calthresholds"}](lac_dist.eps "fig:"){width="6.7cm"} ![On-orbit measurements of threshold values for November 2008: a) LAC, b) FLE, c) FHE and d) ULD. The dashed line in the LAC histogram shows the effects of pedestal drifts seen in August 2008, prior to stabilization. The ULD thresholds are expressed in non-pedestal-subtracted ADC units. This makes it easier to judge how thresholds are set. The saturation limit corresponds to 4095 ACD units.[]{data-label="fig:calthresholds"}](fle_dist.eps "fig:"){width="6.7cm"} ![On-orbit measurements of threshold values for November 2008: a) LAC, b) FLE, c) FHE and d) ULD. The dashed line in the LAC histogram shows the effects of pedestal drifts seen in August 2008, prior to stabilization. The ULD thresholds are expressed in non-pedestal-subtracted ADC units. This makes it easier to judge how thresholds are set. The saturation limit corresponds to 4095 ACD units.[]{data-label="fig:calthresholds"}](fhe_dist.eps "fig:"){width="6.7cm"} ![On-orbit measurements of threshold values for November 2008: a) LAC, b) FLE, c) FHE and d) ULD. The dashed line in the LAC histogram shows the effects of pedestal drifts seen in August 2008, prior to stabilization. The ULD thresholds are expressed in non-pedestal-subtracted ADC units. This makes it easier to judge how thresholds are set. The saturation limit corresponds to 4095 ACD units.[]{data-label="fig:calthresholds"}](uld_dist.eps "fig:"){width="6.7cm"}  calibrations {#sec:tkr} ============= The TKR is used to convert the photon to an e$^+$/e$^-$ pair and to determine the incoming photon direction. It is also the main contributor to the LAT trigger. It consists of sixteen modules each composed of a stack of 19 trays. A tray is a stiff, lightweight carbon-composite panel with silicon-strip detectors (SSDs) mounted on both sides with strips oriented along the same direction. All but the three bottommost trays in each TKR module, contain an array of tungsten foils, which matches the active area of each SSD. These foils act as photon converters. Depending on their location within the tower (front or back), foils are 3% and 18% of a radiation length. Each tray is rotated 90 with respect to the one above and the one below. Therefore, two consecutive trays are needed to provide orthogonal measurements of the $x$,$y$ coordinates. Details of the  design are described elsewhere [@Atwood:2007ra]. Each side (top or bottom) of the tray consists of 1536 silicon strips read out by twenty four 64-channel amplifier-discriminator ASICs, GLAST Tracker Front-end Electronics (GTFE), which are controlled by two digital readout-controller ASICs, GLAST Tracker Readout Controller (GTRC). Each channel in the GTFE has a preamplifier, shaping amplifier, and discriminator similar, although not identical, to the prototype circuits described in [@Johnson98]. The amplified detector signals are discriminated by a single threshold per GTFE chip; no other measurement of the signal size is made within the GTFE. The  electronics is discussed in detail elsewhere [@Baldini:2006pv]. The trigger information is formed within each GTFE chip from a logical OR of the 64 channels. Any latched, noisy or inoperable channel can be masked. The OR signal is passed to the left or to the right, depending on how the chip is configured, and combined with the OR of the neighbor. This procedure continues down the line, until the GTRC receives a logical OR of all non-masked channels it controls. This “layer-OR" initiates a one-shot pulse of adjustable length in the GTRC, which is sent as a fast trigger signal input for the trigger decision. In addition, a counter in the GTRC measures the length of the layer-OR signal, i.e. the time-over-threshold (ToT), and buffers the result for inclusion in the event data stream. Upon receipt of a signal that acknowledges the trigger decision, each GTFE chip latches the status of all 64 channels into one of the four internal event buffers. Another 64-bit mask, which is separate from the trigger mask mentioned above, can be used to mask any subset of channels from contributing data, as may be necessary in case of noisy channels. TKR calibrations include the determination of the noisy channels that form the trigger and data masks, of the trigger threshold settings, and ToT calibrations. Noisy channels {#sec:tkrch} -------------- Since noisy channels can increase the false trigger rate, and can affect instrumental dead time,  measurements and data volume, they are disabled in trigger and/or data masks. The trigger mask determines the active channels that can participate in the formation of the layer-OR trigger signals. The noise occupancy for each strip is measured using not only periodic triggers but all available events, thus increasing statistics by almost 2 orders of magnitude. The computation of noise occupancy for non-periodic triggered events excludes consecutive (2 or more) layers with at least one hit in each. The occupancy measured by this method is consistent with that obtained from periodic triggers. Noisy channels produce off-timing trigger signals resulting in a dead time of $\sim 1~\mu$s for the layer involved. Noisy channels also lead to incorrect  measurements if the noise hit occurs at the tail of the main pulse. To minimize this effect, we mask any channel with occupancy greater than 0.7%[^4]. Furthermore, if the occupancy of layer-OR is greater than 8%, we mask the highest-occupancy channels until we reduce it to this fraction. The data mask determines the active channels whose data can be transmitted to the ground. Since the offline track reconstruction software is tolerant of high-occupancy channels, data mask is driven by the constraint on the data rate given by noisy channels. We mask any channel with occupancy greater than 50% since it does not carry any useful information. We limit the data size due to noisy channels to be less than 10% of TKR total data size, by requiring the average strip occupancy to be less than $5\times10^{-5}$, which corresponds to 44 strip hits per event. The typical strip occupancy of $\sim2-3\times10^{-6}$ is dominated by accidental hits due to off-timing cosmic-ray tracks. The strip occupancy due to electronics noise is $10^{-7}$ or less. We mask highest-occupancy channels until the TKR average occupancy is reduced to $<$ $5\times10^{-5}$. The number of masked channels for trigger and data purposes was 203 before launch and changed to 206 in July 2008, to 220 in August 2008, to 284 in Octobter 2008 and finally to 316 in January 2009. These additional 113 channels are distributed across seven SSDs, while 60 of these channels are concentrated in a region of one SSD. The total number of disabled channels corresponds to only 0.04% of the total number of TKR channels. Trigger and data latching thresholds {#sec:tkrdac} ------------------------------------ In order to minimize the noise occupancy while maximizing the hit efficiency, the nominal threshold level is set to 1.4 fC ($\sim$0.28 MIP). The threshold DAC value for each GTFE is calibrated using charge injection. The charge injection DAC is set to the value corresponding to 1.4 fC and the threshold DAC is scanned (see Section\[sec:tkrmip\]). The best threshold for each channel is determined by a fit to the occupancy versus threshold using the error function (integral of a Gaussian). The average threshold for each GTFE is obtained by calculating the mean value of the threshold DAC values after removing all dead and masked channels, and 5% of the channels with the largest and the smallest values. The calibration of the threshold DAC performed after launch yields identical DAC values to those before launch for 86% of GTFEs. Only 0.04% of the total number of GFTEs (6 out of 14 000) exhibits a difference of more than one DAC value. This was already known from pre-launch measurements and it corresponded to additional noise in the system. The GTFE data is latched following a trigger acknowledge signal, which is 0.8 $\mu$s later than the typical peak of the  pulse shape. This value was determined using an external trigger during pre-launch tests[^5]. Due to the delayed data latch timing, the effective threshold for the data is different from the trigger threshold. Once the threshold DAC value is determined for all GTFEs, we measure the effective thresholds at the time of the data capture by scanning charge injection calibration DAC values. The best threshold for each channel is determined by a fit to the occupancy versus threshold data using the error function. The ratio of data latching threshold measurements before and after launch yields a 2% shift in RMS and 0.5% shift in the mean value, which implies no significant changes from values measured prior to launch. Therefore, trigger thresholds were assumed not to have changed after launch and were not recalibrated. Figure \[fig:tkrthresh\] shows the trigger thresholds obtained prior to launch and the data latching thresholds measured on-orbit. Because of the delay in latching the data, the threshold for the data capture is slightly higher and has a broader peak than that of the trigger threshold. The RMS dispersion is $\sim$5% and $\sim$12% for the trigger and data latching thresholds, respectively. ![Effective threshold for data capture (solid) and trigger threshold (dashed).[]{data-label="fig:tkrthresh"}](tkr_triggerdata.eps){width="14.0cm"} Time-over-threshold conversion parameters {#sec:tkrtot} ----------------------------------------- To determine the conversion parameters from  (ns) to charge deposit (fC), we measure for each channel, the  value for several settings of the charge injection calibration DAC. Since by definition the  values cannot be negative, the  measurements near threshold are biased toward positive values and result in slightly biased conversion parameters. Figure \[fig:totconvfit\] shows the amplitude of charge injected versus ToT values, where the values for charge injection have not been corrected by the MIP scale calibration described in Section \[sec:tkrmip\]. In the fit shown in Figure \[fig:totconvfit\], the  is described as a second order polynomial of injected charge, whose offset corresponds to the threshold value. To avoid biases near the threshold, the fit uses a fixed value for the intercept that corresponds to the calibrated threshold. The statistical error in the fit of order 8% and is estimated by comparing two measurements of the same curve done prior to launch. This calibration is important since the  gain may vary by as much as a factor of three within a GTFE. ![Amplitude of charge injected versus ToT values. The curve corresponds to a second order polynomial fit to the data.[]{data-label="fig:totconvfit"}](tkr_totfit.eps){width="11.0cm"} We have not performed this calibration after launch, since other  results (see Section \[sec:tkrdac\]) indicated little change in the response of  pulses and this calibration requires more than 10 hours of data-taking in dedicated-mode. We expect to recalibrate these values annually. MIP scale calibration {#sec:tkrmip} --------------------- The absolute calibration of the charge injection DAC was performed prior to launch using the charge deposited by surface cosmic rays. A correction factor was defined by the ratio of MIP peaks between data and Monte Carlo simulations. The distribution of this ratio shown in Figure \[fig:totratio\] exhibits an RMS dispersion of $\sim$9%. ![Measured MIP peak divided by simulated MIP peak obtained prior to launch. This charge scale correction factor is obtained for all GTFEs. []{data-label="fig:totratio"}](tkr_scale.eps){width="13cm"} The MIP deposited in each channel was calibrated using these correction factors. On-orbit, we select single-track events, close to normal incidence ($\cos\theta<-0.85$) and with CAL energy consistent with that of a MIP. We require the RMS values of TKR hit positions with respect to the reconstructed track position to be consistent with the resolution needed to reject low-energy tracks. To avoid confusion with charge sharing between adjacent strips, we only consider layers with single hit strips. The charge deposited by a particle is corrected by taking into account its path length in the silicon. The measured  of the hit associated with the track is converted to charge using the conversion parameters. Following that, we fit the data for each GTFE with a Landau distribution convolved with a Gaussian. More than 10 million MIP tracks are required to accumulate sufficient entries for all GTFEs, which takes 5 days during nominal science operations. Figure \[fig:tot\] shows the MIP charge deposit distribution for all channels before (dotted histogram) and after (solid histogram) each GTFE is calibrated. The dispersion correction factor of 9% is included in the calibrated results. ![MIP signal (pathlength corrected) before (dotted) and after (solid) calibrations.[]{data-label="fig:tot"}](tkr_mip.eps){width="13cm"} Determination of SAA polygon {#sec:saa} ============================ The orbit of [*Fermi*]{} intersects the Earth’s inner radiation belt in a region which is known as the South Atlantic Anomaly (SAA). This region features geomagnetically trapped protons with energies up to hundreds of MeV and electrons with energies up to tens of MeV. The flux of protons and electrons in the LAT energy range reach levels which are several orders of magnitudes above those of primary cosmic rays. This extreme particle flux imposes constraints on LAT operations. The TKR electronics saturate due to the increase in the charge deposited per live time, leading to large dead time fractions, thereby hampering scientific observations. The continuous influx of particles generates high currents in the ACD photomultiplier tubes (PMT), thus exceeding safe operating limits, which leads to slow deterioration. Therefore, during SAA passages, triggering, recording and transmission of science data are stopped and the bias voltages of the PMTs are lowered from 900V to $\sim$400V. Only LAT housekeeping data are recorded and transmitted to the ground. The position along the orbit defined by the GPS receiver aboard the [*Fermi*]{} spacecraft determines the transition between nominal science operations and the SAA transit mode. The latitude and longitude of the [*Fermi*]{} position are compared to the bounds of a polygon defined by 12 latitude-longitude vertices stored in the spacecraft memory. As the spacecraft position crosses this polygonal boundary it triggers the SAA transit mode. To avoid multiple entries and exits during a single orbit, a convex polygon is used to define the SAA region. We chose a conservative definition for this initial SAA boundary, with the expectation that we would update the boundary based on particle rate measurements made with the LAT once it was on orbit. The first version of the polygon, or SAA boundary, was determined before launch based on models of the Earth radiation belts and data from other spacecraft. The inner radiation belt was modeled using trapped radiation models: AP-8 [@ap8] and PSB97 [@psb97] in conjunction with the current version of the International Geomagnetic Reference Field (IGRF-10) [@igrf]. The 12-edge SAA boundary polygon was calculated from these models based on the contour in latitude and longitude, where the $E>20$ MeV trapped proton flux reached 1 cm$^{-2}$s$^{-1}$. For regions where two models predictions disagreed, we chose the larger flux. The smallest convex polygon circumscribing this contour was selected and padded by a 4$^\circ$ margin. Figure \[fig:saa1\]a shows the trapped proton flux profiles above 20 MeV predicted by trapped radiation models versus geographic latitude and longitude. Figure. \[fig:saa1\]b shows top scintillator count rates in the altitude range between 532 km and 575 km, reported by the PAMELA experiment [@pamela_icrc], which is similar to that of the [*Fermi*]{} orbit. The red polygon shows the locations of the SAA boundary edges determined before launch and used during the initial phase of the [*Fermi*]{} LAT mission. This definition of the SAA boundary resulted in a loss of observation time of about 17%. ![Trapped radiation models versus geographic latitude and longitude: a) trapped proton flux above 20 MeV predicted by the AP-8 [@ap8] (black contour lines) and PSB97 [@psb97] (color contours), b) top scintillator count rates reported by the PAMELA experiment [@pamela_icrc]. In both of the plots, the red polygon shows the locations of the SAA boundary edges determined before launch and used during the initial phase of the LAT operations.[]{data-label="fig:saa1"}](SAA_boundary_prelaunch.eps){width="15cm"} After launch, diagnostic data of the LAT were used to refine the size of the polygon. Even though science triggers are disabled during SAA passages, fast trigger signals remain operational. Special TKR and ACD counters can sample the rate of fast trigger signals to determine position-dependent rates of the LAT along the orbit. Figure \[fig:saa2\] shows the rates recorded in the TKR counters versus spacecraft position. A rate increase is visible at the edges of the SAA before the TKR electronics saturates and suppress fast trigger signals, thus bringing the count rates to zero. To define the SAA boundary using these data one has to account for particle rates and rate variations from primary and secondary cosmic rays, where both depend on the local geomagnetic cutoff rigidity at the [*Fermi*]{} location. Therefore, we used data from the region outside the pre-launch SAA boundary to determine these rates and set an upper limit on the expected number of cosmic-ray counts per second as a function of the local geomagnetic rigidity cutoff. The optimized SAA boundary polygon is calculated including the points that exceeded this limit. A padding of 1$^\circ$ is applied to account for the limited resolution and sensitivity of the measurement. Figure \[fig:saa2\] shows the average rate of TKR counters obtained during 26.6 days of LAT nominal science operations versus geographic latitude and longitude. Superimposed are the pre-launch SAA boundary (red) used during the initial phase of the mission, and the refined polygon (yellow) uploaded to spacecraft memory. The updated polygon reduced the loss in observation time to approximately 13% of the total on-orbit time. This polygon has been the default for the LAT operations since July 28, 2008. A cross-check during nominal science operations is performed with the ACD trigger signal counters. These are more sensitive to the low-energy component (E $\approx$ 10-60 MeV) of the trapped particle flux than those from the TKR. There is no significant increase in the rate of ACD fast trigger signals as [*Fermi*]{} approaches the SAA boundary, thus validating the optimized polygon. ![Average rate of TKR counters obtained during 26.6 days of LAT nominal science operations versus geographic latitude and longitude. Superimposed are the pre-launch SAA boundary (red) used during the initial phase of the mission, and the updated SAA boundary (yellow) derived from measurements of the TKR counter data (see text)[]{data-label="fig:saa2"}](SAA_optimized_tkr_cnt_rates.eps){width="15cm"} Since the SAA moves at a rate of a few tenths of a degree per year and its size and particle fluxes vary with the solar cycle, we expect annual updates to the SAA boundary. Live time {#sec:livetime} ========= The live time is accumulated taking into account the variety of dead time effects. Science data taking is disabled during SAA passages (see Section \[sec:saa\]). Instrumental dead time, during event latching and readout, is about 8% on average outside the SAA, although this fraction depends on the trigger rate (primarily background). The dependence on the geomagnetic latitude of [*Fermi*]{} is secondary. Other losses are caused by failures in transmission and ground processing and dedicated-mode calibrations. Accurate accounting for live time is essential for obtaining calibrated fluxes and spectra for astrophysical sources of gamma rays. The live time relates the effective collecting area of the LAT to the overall exposure. Owing to the very large field of view of the LAT ($>$2 sr) and the relatively slow scanning rate ($\sim$4 deg min$^{-1}$) the accumulated live time typically is needed only coarsely ($\sim$30 s intervals) for accurate exposures to be calculated. For very bright transient sources, finer accounting for live time is used, owing to the high and variable rates of triggers. For example, a bright GRB in the field of view of the LAT, such as GRB08019C [@grb0801916C], can induce a dead time fraction of about 16% during the impulsive phase of the burst. The dead time for event latching and readout is tracked on-board the LAT every 50 ns using the 20 MHz LAT system clock. The stability of the 20 MHz clock is closely tracked with a 1 pulse-per-second (PPS) signal from the GPS system of the spacecraft. During nominal science operations, the instrumental dead time is dominated by the fixed time to read out the event timing and trigger information, which imposes a minimum dead time of 26.5 $\mu$s per event. Every time the LAT triggers, further data taking is disabled until the data from the event are read out. After the end of the time coincidence window the latency of the trigger system is 100 ns before the new time coincidence window is available, even if the previous event was not read out. Single front-end electronic channels can be dead for several microseconds while the pulse is above threshold. In special cases, such as when the periodic trigger is enabled (2 Hz), the entire CAL is read out with no zero suppression and in this case, the dead time can be as large as 620 $\mu$s. During nominal science operations one of the trigger combinations is configured to read the CAL with all four energy ranges and zero suppression for which the dead time is about 65 $\mu$s. Losses of data in transmission can effectively remove events from the data stream both directly and indirectly. The indirect effects relate to how the events are assembled in the telemetry stream. In each packet, the event times are encoded as times relative to the start of the packet. In addition, the 20 MHz counters mentioned above roll over every 1.6 s and the times of the roll overs are also encoded in the packets. Loss of certain parts of a packet can cause times and live time accumulations to be lost for up to $\sim$200 events. At the 450 Hz nominal event rate in telemetry, these losses can be as large as 0.4 s. The mission specification for data loss is less than 2%, and in practice loss due to transmission errors has been much less than this, in part because on-board the LAT data are retained for approximately 24 hours and retransmissions can be requested. At the ground station, [*Fermi*]{} telemetry is buffered for 1 week and losses in ground transmission generally can be recovered. The overall average loss has been much less than 1% to date. Losses in ground processing of event data are also rare and are significantly less than 1%. Overall timing accuracy {#sec:absolute} ======================= Recording accurate arrival times of LAT photons is essential for studies of gamma-ray bursts and pulsar timing. Absolute timing tests were performed during pre- and post-launch activities. A discussion of LAT pulsar timing can be found elsewhere [@pulsartiming]. During pre-launch tests we recorded cosmic rays to measure the time difference between two GPS systems. As shown in Figure \[fig:MuTimeTestSketch\] a pair of scintillator tiles provided a reference for the LAT timestamps. The coincidence signal from these tiles triggered a VME-based GPS time system previously used by the ground gamma-ray telescope CELESTE. Its absolute time accuracy was previously demonstrated by measuring the Crab optical pulsar [@celeste]. ![Diagram of the muon scintillator telescope placed next to the [*Fermi*]{} satellite during pre-launch tests. \[fig:MuTimeTestSketch\]](latmuontelescope.eps){width="9cm"} Reconstructed muon tracks traversing the LAT detector were extrapolated to their impact point on the laboratory floor and their timestamps were measured with respect to the GPS of the [*Fermi*]{} satellite. If a muon passed through the pair of scintillators placed next to [*Fermi*]{}, a GPS timestamp from a standalone VME data acquisition system was also recorded. Figure \[fig:MuTimeTestResults\] shows that the LAT timestamps agreed with the reference GPS to within 0.3 $\mu$s. ![Results from pre-launch tests: a) differences in the times recorded with the two acquisition systems, versus elapsed time; b) histogram of time differences indicating mean and RMS values around 0.3 $\mu$s. \[fig:MuTimeTestResults\]](absolutetime.eps){width="15cm"} GPS receivers use the arrival times of reference signals from other GPS satellites to calculate their time and position and transmit that information to processors on [*Fermi*]{}. This is accompanied by an electronic “Pulse Per Second” (PPS) at the moment of validity of the timestamp word. The processors, using precision oscillators on-board the spacecraft, maintain the PPS accuracy in the case of occasional short losses of GPS signal reception. Bright gamma-ray pulsars were used to verify that the integer seconds of absolute time from the GPS receiver conform to Coordinated Universal Time (UTC), since an integer offset in the [*Fermi*]{} clocks would make a large shift in observed gamma-ray phase, different for each pulsar. The rotational phase of the gamma-rays peaks of the Vela pulsar relative to the radio peak agree with that measured by previous experiments [@vela]. The first gamma-ray peak of the Crab pulsar is aligned with the radio reference, since Crab radio and gamma-ray beams appear to come from the same part of the neutron star magnetosphere [@celeste]. The fractional part of LAT event timestamps is obtained from the counts of a 20 MHz oscillator recorded by scalers latched at the reception of a GPS PPS signal and at the reception of an event trigger. The behavior of the oscillator was extensively characterized during the ground tests, and its frequency is recalibrated each second using the PPS-latched scaler values. On-orbit telemetry monitoring shows that the internal spacecraft timing signals behave as before launch, from which we conclude that LAT timestamps are still well within 1 $\mu$s of the GPS times used by the spacecraft. GPS times are maintained within 20 ns (1 sigma) of UTC [@gpsStandard]. At present the best observational validation of the on-orbit clock performance comes from the pulsars PSR J0030+0451 and PSR J1939+2134. The peak width of $<1$20 $\mu$s reported for PSR J0030+0451 in [@0451] demonstrates the stability of the LAT event times over six months of data-taking, but not their absolute accuracy. However, the 1.56 ms pulsar PSR J1939+2134 appears to have a gamma peak aligned with the radio peak to better than 1/20 of a rotation of the neutron star, that is, 80 $\mu$s [@2134]. This translates to an absolute time accuracy if one assumes that both the gamma and radio emission come from the same region in the pulsar’s magnetosphere. Internal and spacecraft boresight alignments {#sec:ta} ============================================ The accuracy of the directions of reconstructed events dependes on the knowledge of the exact position of each element of the TKR, i.e individual SSD or planes. Misalignments of any element can broaden the instrument response function, thus lowering the LAT sensitivity. We perform the following alignment procedures: 1. intra-tower alignment to determine the position and orientation of [*each element*]{} of a single tower with respect to an ideal coordinate system; 2. inter-tower alignment to determine the position and orientation of [*each tower*]{} with respect to an ideal coordinate system; 3. spacecraft alignment to determine the rotation of the LAT with respect to the [*Fermi*]{} on-board guidance, navigation and control system. Intra-tower alignment {#sec:intra} --------------------- As described in Section \[sec:tkr\], each TKR tower consists of 36 silicon planes each instrumented with sixteen silicon microstrip detectors. All strips in a plane are oriented along the same direction. The intra-tower procedure aligns the planes horizontally (along the measured coordinate) and vertically (perpendicular to the silicon plane), and determines all rotations of the planes within the TKR tower. Due to the procedures for construction and assembly of towers, we expect only minor deviations from the original positions. This simplifies the formalism used for the alignment, and we assume first-order approximations to deviations. The trajectory of a particle traversing the TKR is characterized by a reconstructed track, which consists of a list of associated silicon strip hits joined by a straight line. As the particle encounters material in its path its direction is affected by multiple scattering. The effect is more pronounced in the presence of dense materials such as the tungsten converter foils. Thus, a fit to an otherwise straight track results in deviations of the real hit positions from an ideal straight track referred to as residuals. The diagram in Figure \[fig:align\] shows two tracks in the $xz$ plane and illustrates how residuals relate to measured and ideal positions. ![The diagram shows five silicon layers (x0 to x4) and the ideal x2 layer, crossed by two tracks. Measured hits are denoted as filled diamonds, extrapolated hits (assuming the ideal position of the x2 plane) as open diamonds. The residual is composed of two contributions: $\Delta x$ due to horizontal misalignment, and $\Delta z\cdot\tan(\theta)$ due to vertical misalignment.[]{data-label="fig:align"}](intra4.eps){height="6cm"} A distribution of residuals should be centered at 0, and any deviation from zero is an indication that the element studied is not at its assumed position. Figure \[fig:hz\] illustrates how straight tracks are used to determine the horizontal and vertical misalignments. It shows the hit residuals versus $\tan(\theta)$ (i.e. inverse of the slope) of the track, for one silicon plane (arbitrarily chosen). A straight line fit through these points produces an offset and a slope, which correspond to deviations of the real position from the assumed one. Results from the fit determine horizontal ($\Delta x$) and vertical ($\Delta z$) deviations, corresponding to shifts along and perpendicular to the strip coordinates, respectively. ![Residual versus $\tan(\theta)$ of the track, for a silicon plane arbitrarily chosen. A fit with a straight line provides an offset and a slope, which correspond to deviations of the real position to the assumed one.[]{data-label="fig:hz"}](hz4.eps){height="9cm"} Intra- and inter-tower TKR alignments are iterative processes that use reconstructed events recorded during nominal science operations with no requirement on any on-board filter. As a result, each TKR element is subject to a wide range of event types, thus minimizing the possibility of selection bias. Here we briefly describe a method common to both intra- and inter-tower alignments. The event reconstruction associates hits to tracks and classifies them according to their track length and straightness. The internal TKR alignment procedures benefits from track reconstruction by using only the first (best) track in the event, i.e. all other tracks are discarded. The list of the hits is then input to the alignment algorithm. The alignment algorithms rely on two important pieces: the raw position information, i.e. the plane and strip containing a hit, and the [*smoothed*]{} position, which results from track finding and optimization through the Kalman filter algorithms [@latpaper]. Since the raw position cannot be used to determine the location of the hit along the strip, the algorithm uses the corresponding [*smoothed*]{} position, instead. Finally the algorithm fits the hit positions with a straight line, computes the $\chi^2$ of the fit, and evaluates the residuals, and the angle $\theta$ of the track with respect to the tower axis. Correlations similar to those depicted in Figure \[fig:hz\] are used to determine all alignment parameters listed in Table \[table:corr\]. Note that the angle $\theta$ is always measured in the coordinate of the silicon plane (e.g. $\theta_x$ for $x$-planes). The sign of the correlation for rot$z$ is different for the $x$ and the $y$-planes. There are six parameters to determine, but only four correlations. Each correlation gives two parameters: the horizontal position, either $\Delta x$ (for $x$ planes) or $\Delta y$ (for $y$ planes), and $\Delta z$ or any of the three rotations. Correlation of track parameters Alignment Parameters ------------------------------------------------ ------------------------- -- -- residual versus $\tan(\theta)$ $\Delta h$ + $\Delta z$ residual versus $x\cdot\tan(\theta)$ $\Delta h$ + rot$x$ residual versus $-y\cdot\tan(\theta)$ $\Delta h$ + rot$y$ residual versus $\mp$ position along the strip $\Delta h$ + rot$z$ : Correlations of track parameters, and resulting alignment parameters. $\Delta h$ is the measured horizontal coordinate of a plane, i.e. $\Delta x$ for $x$-planes and $\Delta y$ for $y$-planes. See text for more details.[]{data-label="table:corr"} For each TKR element the iterative algorithm determines five parameters that are measured relative to other elements whose positions are also unknown. The knowledge of the precise positions and rotations increases with successive iterations, and each parameter converges to its correct value. The decision for convergence comes from comparing the largest deviation of any positional (rotational) parameter between two successive iterations and by requiring the maximum deviations to be 0.01$\mu$m (1$\mu$rad). Convergence typically occurs after 100 iterations. Currently, the smallest elements considered in intra-tower alignment are single silicon planes (top and bottom of trays as defined in Section \[sec:tkr\]). In the future, we will extend the method to align SSD’s. Table \[table:intrafinal\] shows the results from the on-orbit intra-tower alignment of 576 TKR planes. Offsets and slopes (as in Figure \[fig:hz\]) are entered in a histogram and the standard deviation ($\sigma$) of each distribution is shown in Table \[table:intrafinal\]. Since planes are aligned with respect to an ideal frame, by construction, the sum of all shifts (for all parameters) average to the mean value of 0. Results clearly demonstrate the quality of the assembly of individual towers since standard deviations are within $\pm$ 61 $\mu$m and $\pm$ 220 $\mu$rad. Parameter Standard deviation ($\sigma$) ----------- ------------------------------- $\Delta$x $\pm$ 43 $\mu$m $\Delta$y $\pm$ 59 $\mu$m $\Delta$z $\pm$ 61 $\mu$m rot$_x$ $\pm$ 220 $\mu$rad rot$_y$ $\pm$ 220 $\mu$rad rot$_z$ $\pm$ 210 $\mu$rad : On-orbit intra-tower alignment constants, for 576 planes (averaged for all towers).[]{data-label="table:intrafinal"} Inter-tower alignment {#sec:inter} --------------------- The inter-tower alignment procedure aligns towers spatially and determines their rotations with respect to the LAT reference frame, and is performed only after the intra-tower alignment constants have been determined. While the latter uses the residuals of each hit to determine the alignment constants, the former relies on track segments of events that cross tower boundaries. We determine the orientation of each tower by evaluating the angles between pairs of track segments. The rotation in one axis is given by the scalar product of the track segments, projected on the plane spanned by the two other axes, and averaged over all selected events: $$\label{TA:scalar} cos(\mbox{rot}_i) = \left< \sum \frac{\vec{t} \,\vec{t}'}{|t| |t'|} \right>,$$ where $\vec{t}$ and $\vec{t}'$ are the track segments in vectorial form and $i$ corresponds to $x$, $y$, or $z$. For convenience, Eq.(\[TA:scalar\]) can be rewritten in terms of track slopes $cot(\theta)$, yielding $$\label{TA:rot} \mbox{rot}_i = \left< \sum \frac{\Delta\cot(\theta_i)}{1 + \cot(\theta_i)^2} \right>.$$ After fixing rotations, each pair of track segments is projected onto a reference plane between both towers, as depicted in Figure \[fig:inter\]. ![ Horizontal and vertical alignment for a pair of towers. the real positions are drawn with solid lines, the assumed position of tower B with dashed ones. The black line between the towers marks the reference plane. The real particle track is shown with solid line, the extrapolated track based on the assumed position dashed. []{data-label="fig:inter"}](inter4.eps){height="8cm"} Each intersection defines a position on this plane. For perfect alignment, these positions are identical. The differences between both positions are analyzed as in Section \[sec:intra\], i.e. by searching for correlations between the position differences and $\cot(\theta)$ of the track. In the example of Figure \[fig:inter\], the residual has a $\Delta$z and a $\Delta$x$\cdot\cot(\theta)$ contribution due to vertical and horizontal misalignments. Table \[table:inter\] summarizes the results from the on-orbit inter-tower alignment of 16 TKR towers. Contrary to the intra-tower alignment, only a small number of values (16 towers instead of 576 planes) are available to determine the $\sigma$ of the distribution for each of the alignment parameters. Because of that, Table \[table:inter\] has an additional column that shows the largest and smallest value for each parameter. Results clearly demonstrate the quality of the assembly of towers in the LAT since standard deviations are within $\pm$ 120 $\mu$m and $\pm$ 260 $\mu$rad. Parameter Values (mix,max) Standard deviation ($\sigma$) ----------- ---------------------- ------------------------------- $\Delta$x (-250,+190) $\mu$m $\pm$ 119 $\mu$m $\Delta$y (-90,+130) $\mu$m $\pm$ 68 $\mu$m $\Delta$z (-150,+150) $\mu$m $\pm$ 87 $\mu$m rot$x$ (-450,+400) $\mu$rad $\pm$ 260 $\mu$rad rot$y$ (-360,+480) $\mu$rad $\pm$ 250 $\mu$rad rot$z$ (-360,+360) $\mu$rad $\pm$ 230 $\mu$rad : On-orbit inter-tower alignment constants for 16 towers.[]{data-label="table:inter"} Alignment procedures were validated using a simulated dataset of cosmic protons, generated with randomly misaligned geometry for the LAT. Figure \[fig:data\] shows the results when the alignment procedure is applied to data acquired before and after launch. Here the standard deviations ($\sigma_{\Delta}$) are calculated from a histogram of differences between ground and on-orbit measurements. Figure \[fig:data\]a shows the correlation for intra-tower positional alignment constants, while Figure \[fig:data\]b displays the histogram of these residual differences. ![On-orbit data: a) correlation between ground and on-orbit inter-tower positional alignment constants, b) the histogram of the differences ($\Delta\sigma$), where $x$, $y$ and $z$ correspond to squares, diamonds and triangles, respectively.[]{data-label="fig:data"}](data4.eps){height="7cm"} Table \[table:data\] compares unbiased datasets obtained on ground ($\sim$8 million surface muons) with orbit data ($\sim$20 million events, mostly cosmic protons) and Parameter $\sigma_{\Delta}$ (intra-tower) $\sigma_{\Delta}$ (inter-tower) ----------- --------------------------------- --------------------------------- $\Delta$x 2.4 $\mu$m 30 $\mu$m $\Delta$y 2.5 $\mu$m 32 $\mu$m $\Delta$z 8.8 $\mu$m 17 $\mu$m rot$x$ 51 $\mu$rad 53 $\mu$rad rot$y$ 53 $\mu$rad 42 $\mu$rad rot$z$ 36 $\mu$rad 24 $\mu$rad : Standard deviations ($\sigma_{\Delta}$) of the differences between ground and on-orbit alignments. Inter-tower values are averaged over all three positions and rotations.[]{data-label="table:data"} lists the average differences for both intra- and inter-alignment, for all parameters. Results from Figure \[fig:data\] and Table \[table:data\] demonstrate that misalignments due to launch or temperature variations on-orbit are small and limited to within $\pm$ 35 $\mu$m and $\pm$ 55 $\mu$rad. Finally, Figure \[fig:pointing\] ![Correlation between the different rotation angles of each tower (the number of each tower is denoted in the diagram). Squares denote the correlation between rot$x$ ($x$-axis) and roty (left $y$-axis label), and diamonds between rot$x$ and rot$z$ (right $y$-axis label).[]{data-label="fig:pointing"}](pointing4.eps){height="14cm"} shows all correlations between the rotation angles obtained for each tower using on-orbit data acquired by the end of October 2008. Tower numbers are labeled from 0 to 15. Squares denote the correlation between rot$x$ ($x$-axis) and rot$y$ (left $y$-axis label), and diamonds between rot$x$ and rot$z$ (right $y$-axis label). The spread in the correlation between rot$x$ and rot$y$ translates directly into a pointing inaccuracy on the celestial sphere, e.g. a misalignment of 0.3 mrad in rot$x$ or rot$y$ for one tower would cause a source position error of the same size. The absence of clustering in this distribution indicates that there are no obvious systematic effects. The alignment of the TKR has been checked twice before launch (since October 2006) and four times after launch (since June 2008). No significant changes of the positions and rotations were observed. We expect updates to be made annually. LAT alignment with respect to the spacecraft {#sec:sc} -------------------------------------------- The LAT reconstructs the direction of each photon in the LAT reference system. To convert this to celestial coordinates, we require the celestial orientation of the LAT. The spacecraft orientation is provided by the Guidance, Navigation and Control system, or star-tracker. This system involves optical telescopes on an optical bench, and a star field pattern recognition system, which was calibrated before launch. The LAT system is nominally the same as the spacecraft system, but small deviations are expected due to uncertainty in the ground alignment process, thermal variations, launch vibrations, relaxation in 0g, or the inter-tower alignment process. Measuring these deviations, by comparing the LAT view of the sky with that seen by the star-tracker is called “boresight alignment". The boresight alignment is determined via likelihood maximization for gamma rays near bright, identified celestial point sources of gamma-ray emission. The likelihood analysis is “binned” in the sense that the gamma rays are not considered individually but binned into maps in celestial coordinates. Separate maps are used for different energy ranges because the point-spread function (PSF) depends strongly on energy. The likelihood is defined in terms of the Point Spread Function (PSF), the background (sum of celestial sources and residual charged particle bacgrounds, estimated independently for each source), and the boresight alignment angles. We measure the boresight alignment by maximizing the total binned likelihood of the point source model (background is included) with respect to three angles characterizing an arbitrary rotation. We chose the convention of performing consecutive rotations about the $x$, $y$, and $z$ axes. These are defined by the spacecraft: $z$ is along the symmetry axis (pointing direction), and $y$ is along the solar panels. Since the angles are small, less than a degree, the order of the rotations is irrelevant. Rotations about the $x$ and $y$ axes thus correspond to angular deviations of the same size, while the rotation about the $z$ axis affects only tracks that traverse the detector at off-axis angles. The procedure for calculating the likelihood transforms both the source position (i.e. celestial coordinates) and the photon position into spacecraft coordinates, rotates the photon’s position, and determines the likelihood that the photon is consistent with the source location. Details of the likelihood method are discussed elsewhere [@like]. We first define a set of reference bright point sources using the likelihood itself to select the most significant. We select photons within the 99% PSF containment of the sources. Since the angular containment is energy dependent, photons are divided into energy bands, using an average PSF for each band to define the expected deviation distribution. The PSF is non-Gaussian, and has been found to be well modeled by the following power-law expression [@perfpaper], $$PSF(\delta,E_i)=\left( 1- \frac{1}{\gamma(E_i)} \right)\left( 1+\frac{\delta^{2}}{2\sigma(E_i)^{2}\gamma(E_i)}\right)^{-\gamma(E_i)},$$ where $E_i$ is the energy bin of the photon, $\delta$ is the angular separation between reconstructed and true direction, and $\gamma(E_i)$ and $\sigma(E_i)$ are energy dependent parameters. The parameter $\gamma(E_i)$ determines the tails of the distribution and at low energues has a value of $\sim$2.25, which decreases at high energies, creating longer PSF tails. The parameter $\sigma(E_i)$ has a power-law dependence at lower energies and reaches an asymptotic value at high energies determined by the silicon detector readout pitch. We compute $\delta$ for all spatial bins and weight the likelihood by the number of counts per bin. The parameters of the PSF were estimated by extensive Monte Carlo simulations, but have been since refined with on-orbit data. We select photons from energy bands above 500 MeV, which contain little or no background. We calculate a likelihood for each source by modeling it as point source in a uniform background. Treating the remaining background on the angular scale of the PSF as uniform is a good approximation, and would be significant only in the tails where the contribution to the likelihood is small. The bright point sources for daily alignment check are selected based on the Test Statistic (TS), which is twice the difference of the log likelihood of a point source in a uniform background and the log likelihood of a purely uniform background for a particular energy band. The total TS is then the sum of the TS’s for the energy bands above 500 MeV. Since the background fraction is already small, the assumption of uniformity is a good approximation to the distribution of the background events, and we select all sources with a total TS greater than 25. For the example shown in Figure \[fig:ellipse\], 6 sources match this criterion, and these are used as seeds for the alignment procedure. The largest contribution comes from the Vela pulsar, which is dominated by photons between 1 GeV and its 10 GeV cutoff. The likelihood is maximized with respect to the background for each energy band, since the shape of the PSF depends on energy. The total likelihood for a point source is then the product of the likelihood in each energy band. Figure \[fig:ellipse\] shows typical projected error ellipses resulting from the likelihood fit. The crosses correspond to the ideal location and the dashed and solid contour lines represent 1 and 2 sigma contours derived from the log likelihood, respectively. A typical day in orbit yields a precision of 1.5 $arcmin$ in $x$ and $y$, and 2.4 $arcmin$ in $z$, when using 174 photons collected from six astronomical sources. We expect, after the analysis of the first year data to reach the required value of 4 arcsec for the boresight alignment residuals. ![Projection of the likelihood surfaces into each of the planes a) $yx$, b) $zx$ and c) $zy$ for one day of data (August 3, 2008). Crosses correspond to the ideal location (perfect alignment), and contours are 68% (dashed) and 95% (solid) containment.[]{data-label="fig:ellipse"}](1day_unc_1.eps "fig:"){width="17cm"} ![Projection of the likelihood surfaces into each of the planes a) $yx$, b) $zx$ and c) $zy$ for one day of data (August 3, 2008). Crosses correspond to the ideal location (perfect alignment), and contours are 68% (dashed) and 95% (solid) containment.[]{data-label="fig:ellipse"}](1day_unc_2.eps "fig:"){width="8.5cm"} As shown in Figure \[fig:daily\] we make independent measurements of the boresight alignment for each week, to monitor the stability. The figure shows the cumulative mean (dash-dotted line) for the rotation angles about the $x$, $y$ and $z$ axes for a period of 3 months. The cumulative mean is just the average of the parameters weighted by the errors. We use the same sources and data to optimize the combined likelihood with respect to the PSF parameters themselves, and feed this back into the analysis. Weekly measurements indicate a stability of 0.3 $arcmin$, constant with the statistical errors. ![Results of the likelihood fit for weekly measurements over a period of 3 months for the rotation angles about the a) $x$, b) $y$ and c) $z$ axes. The cumulative mean is displayed as dash-dotted line, while the dashed line shows the reference at zero ($y$ axis scales are different). The start date for these measurements on Mission week 10 corresponds to August 10, 2008.[]{data-label="fig:daily"}](weekly_1.eps "fig:"){width="16cm"} ![Results of the likelihood fit for weekly measurements over a period of 3 months for the rotation angles about the a) $x$, b) $y$ and c) $z$ axes. The cumulative mean is displayed as dash-dotted line, while the dashed line shows the reference at zero ($y$ axis scales are different). The start date for these measurements on Mission week 10 corresponds to August 10, 2008.[]{data-label="fig:daily"}](weekly_2.eps "fig:"){width="7.6cm"} We repeated the likelihood fits for a period of about 5 months. The results for the rotation angles are shown in Figure \[fig:months\]. We clearly see the improvement with more statistics. As expected, rotations about $x$ and $y$ axis lead to similar results (6 $arcsec$), while for rotations around $z$ we obtain values approaching 8 $arcsec$. These values should be added in quadrature to other contributions to obtain the systematic uncertainties for localizing point sources. ![Data correspond to results of the likelihood fit for measurements over a period of 5 months for the rotation angles about the a) $x$, b) $y$ and c) $z$ axes.[]{data-label="fig:months"}](norm_res500.eps){width="14cm"} Conclusions {#sec:conclusion} =========== We have discussed the on-orbit calibrations for the [*Fermi*]{} LAT, which include synchronization of trigger signals, optimization of delays for latching data, determination of detector thresholds, gains and responses, evaluation of the perimeter of the South Atlantic Anomaly (SAA), measurements of live time and of absolute time and internal and spacecraft boresight alignments. The results summarized in Table \[table:summary\] were obtained using known astrophysical sources, galactic cosmic rays, and charge injection. There were only minor changes to calibration constants since launch and these quantities have been stable during the first eight months of operations. The LAT with almost a million channels is a remarkably stable instrument and is expected to operate during the next few years with dedicated calibration runs of about three hours every three months. The frequency with which calibration updates are expected to occur varies from three to twelve months, but these do not necessarily impact the LAT performance in any significant way. As a consequence, changes to high level datasets due to these calibrations will be infrequent. The results reported here have been used to calibrate the LAT datasets to be publicly released in August 2009. Category Title Summary Sec. ----------- ---------------------------- --------------------------------------------------------------------------------- ------------------- Trigger Time coincidence window 700 ns, settings within 50 ns \[sec:tsync\] Trigger Fast trigger delays synchronization within 50 ns \[sec:tsync\] Trigger Delays for latching data optimized within 50 ns \[sec:tsync\] ACD Pedestal width $<$ 4 ADC bins or $<$ 0.01 MIPs \[sec:acdpeds\] ACD Coherent noise removed down to 0.005 MIP \[sec:acdpeds\] ACD MIP peak stability $<$10% \[sec:acdmip\] ACD High range (CNO) width of carbon peak$\sim$20% of peak \[sec:acdmip\] ACD Veto threshold turn-on at 0.4-0.5 MIP, set within $\pm$ 0.01 MIP \[sec:veto\] ACD High level discriminator turn-on at 24-26 MIP, set within $\pm$ 1 MIP \[sec:veto\] CAL Pedestal RMS within 0.1-0.2 MeV \[sec:calpeds\] CAL Electronics linearity corrected to $\le$1% of the measured energy \[sec:callin\] Energy scales spread crystal-to-crystal $\le$1% \[sec:callin\] stability of peaks: MIP($\le$2%), carbon($\le$1%) CAL Light asymmetry 2 mm (LEX1), 9 mm (HEX8) from 200-900 MeV \[sec:calpos\] CAL Zero-suppression threshold set at 2 MeV, $\sim$ 10 x electronics noise \[sec:calthresh\] CAL Low-energy threshold set at 100 MeV ($\pm$ 1%) \[sec:calthresh\] CAL High-energy threshold set at 1 GeV ($\pm$ 2%) \[sec:calthresh\] CAL Upper level discriminators set at 5% below saturation level \[sec:calthresh\] Noisy channels avg strip occ. (10$^{-5}$), add $\le$10% to the TKR data \[sec:tkrch\] electronic noise occupancy (10$^{-7}$), 0.04% disabled Trigger threshold set at $\sim$0.28 MIP, spread channel-to-channel$\sim$5% \[sec:tkrdac\] $\le$ 1 ADC shift since launch (99.96% of channels) TKR Data latching threshold spread channel-to-channel$\sim$12% \[sec:tkrdac\] TKR ToT conversion parameters fitted to $\sim$ 8% (statistical error) \[sec:tkrtot\] TKR MIP scale RMS of correction factor (9%) \[sec:tkrmip\] SAA SAA polygon in SAA for$\sim$13% of the orbit time \[sec:saa\] Timing LAT timestamps $<$ 0.3 $\mu$s with respect to a reference GPS \[sec:absolute\] Alignment Intra tower for details see Tables \[table:intrafinal\] and \[table:data\] \[sec:intra\] Alignment Inter tower for details see Tables \[table:inter\] and \[table:data\] \[sec:inter\] Alignment LAT boresight $\theta_x$ , $\theta_y$ $\le$ 6$^{''}$ and $\theta_z$ $\le$ 8$^{''}$ (5 months) \[sec:sc\] : Summary of the on-orbit Fermi LAT calibrations.[]{data-label="table:summary"} Acknowledgments =============== The [*Fermi*]{} LAT Collaboration acknowledges the generous ongoing support of a number of agencies and institutes that have supported both the development and the operation of the LAT as well as scientific data analysis. These include the National Aeronautics and Space Administration and the Department of Energy in the United States, the Commissariat à l’Energie Atomique and the Centre National de la Recherche Scientifique / Institut National de Physique Nucléaire et de Physique des Particules in France, the Agenzia Spaziale Italiana and the Istituto Nazionale di Fisica Nucleare in Italy and the Istituto Nazionale di Astrofisica, the Ministry of Education, Culture, Sports, Science and Technology (MEXT), High Energy Accelerator Research Organization (KEK) and Japan Aerospace Exploration Agency (JAXA) in Japan, and the K. A. Wallenberg Foundation and the Swedish National Space Board in Sweden. Additional support for science analysis during the operations phase from the following agencies is also gratefully acknowledged: the Istituto Nazionale di Astrofisica in Italy and the K. A. Wallenberg Foundation in Sweden for providing a grant in support of a Royal Swedish Academy of Sciences Research fellowship for JC. [99]{} , in press [*Astrophysical Journal*]{}. C. Meegan [*et al*]{}, in preparation. D.J. Thompson [*et al.*]{}, ApJ. Suppl. Vol [**86**]{} (1993) 629. W. B. Atwood [*et al.*]{}, Astropart. Phys.  [**28**]{}, (2007) 422. T. Ohsugi [*et al*]{}, Nucl. Inst. and Meth. [**A 541**]{} (2005) 29. R. P. Johnson, P. Poplevin, H. F.-W. Sadrozinski, and E. N. Spencer, IEEE Trans. Nucl. Sci. [**45**]{}, (1998) 927. L. Baldini [*et al.*]{}, IEEE Trans. Nucl. Sci.  [**53**]{}, (2006) 466. J. E. Grove [*et al.*]{}, in preparation. A. Moiseev [*et al.*]{}, Astropart. Phys. **27**, (2007) 339. A. Moiseev [*et al.*]{}, Nucl. Inst. and Meth. [**A 583**]{}, (2007) 372. A. Abdo [*et al.*]{}, in preparation. W. Atwood [*et al.*]{}, Nucl. Inst. and Meth. [**A 446**]{} (2000) 444. E. do Couto e Silva [*et al.*]{}, Nucl. Inst. and Meth. [**A 474**]{}, (2001) 19. A. Abdo [*et al.*]{}, in preparation. A. Abdo [*et al.*]{}, in preparation. A. Moiseev [*et al.*]{}, Astropart. Phys. **22**, (2004) 275. B. Lott [*et al.*]{}, Nucl. Inst. and Meth. **560**, (2006) 395. D. M. Sawyer and J. I. Vette, [*AP-8 Trapped Proton Environment for Solar Maximum and Solar Minimum*]{}, NSSDC/WDC-A-R&S 76-06, 1976. D. Heynderickx, M. Kruglanski, V. Pierrard, J. Lemaire, M. D. Looper, and J. B. Blake, IEEE Trans. Nucl. Sci., [**46**]{}, (1999) 1475. IGRF-10, IAGA Division V-MOD Geomagnetic Field Modeling, http://www.ngdc.noaa.gov/IAGA/vmod/igrf.html M. Casolino, et al. [*Observations of primary, trapped and quasi-trapped particles with the PAMELA experiment*]{}. to appear in the proceedings of the 30$^{th}$ International Cosmic Ray Conference (ICRC 2007) 3-11 Jul 2007, Merida, Yucatan, Mexico. A. Abdo [*et al*]{} submitted to [*Science*]{} (2008). D. A. Smith [*et al.*]{} A&A [**492**]{} (2008) 923. M. de Naurois, J. Holder [*et al.*]{} ApJ [**566**]{}, (2002) 343. A. Abdo [*et al.*]{}, in press [*Astrophysical Journal*]{} (2008). Global Postioning System Standard Positioning Service Performance Standard’, 4th edition (September 2008), http://pnt.gov/public/docs/2008-SPSPS.pdf, p. A-16, section A.4.8 A. Abdo [*et al.*]{}, submitted to ApJ. A. Abdo [*et al.*]{}, in preparation. A. Abdo [*et al.*]{}, in preparation. [^1]: Corresponding author. Tel. +1 650 926 2698, email: eduardo@slac.stanford.edu [^2]: Prior to launch the time coincidence window was set to 600 ns (12 ticks). [^3]: The efficiency for latching TKR data is unchanged up to $\ge$ 750 ns (15 ticks). [^4]: 0.7% corresponds to the value in which the loss in efficiency due to dead time is comparable to that from masking noisy channels. [^5]: Although this implies that the TKR delay should be negative, the smallest allowed value is zero. This has no effect on the TKR performance.

ArXiv

= 10000 -1.0in 0.025in 0.025in 9.35in 6.45in [**X-RAY ABSORPTION STUDY OF PULSED LASER DEPOSITED BORON NITRIDE FILMS**]{}\ A. CHAIKEN$^*$, L.J. TERMINELLO$^*$, J. WONG$^*$, G.L. DOLL$^{\dag}$, AND T. SATO$^{\ddag}$\ $^*$Lawrence Livermore National Laboratory, Livermore, CA 94550\ $^{\dag}$GM Research and Development Laboratory, Warren, MI 48090\ $^{\ddag}$NIRIM, Tsukuba City, Japan [**ABSTRACT**]{} B and N K-edge x-ray absorption spectroscopy measurements have been performed on three BN thin films grown on Si substrates using ion-assisted pulsed laser deposition. Comparison of the films’ spectra to those of several single-phase BN powder standards shows that the films consist primarily of $sp^2$ bonds. Other features in the films’ spectra suggest the presence of secondary phases, possibly cubic or rhombohedral BN. Films grown at higher deposition rates and higher ion-beam voltages are found to be more disordered, in agreement with previous work. [**INTRODUCTION**]{} Near-edge x-ray absorption spectroscopy is well-known as a method for characterizing the bonding and orientation of organic molecules adsorbed on thin-film surfaces.[@stohrbook] Subsequent work has shown that core-level x-ray absorption is also a useful technique for characterizing the unoccupied electronic states of low-atomic-number solids. A number of x-ray absorption studies have focussed on graphite and diamond,[@morar85] the two common crystalline phases of carbon. Other researchers have compared the spectra of amorphous carbon and hydrogenated diamond-like carbon thin films to those of the bulk crystalline phases.[@denley80] Boron nitride is isoelectronic to carbon and has both hexagonal and cubic phases analogous to graphite and diamond. In hexagonal BN (hBN), B and N atoms in alternate layer planes lie directly on top of one another in an AA$^{\prime}$A stacking arrangement, as opposed to the staggered ABA stacking in graphite, which causes C atoms in neighboring layers to be offset. Rhombohedral BN (rBN) differs from hBN only in its staggered ABCA stacking. Previous NEXAFS studies have compared the B and N K-edge spectra of well-ordered hexagonal, cubic, wurtzite and rhombohedral BN powders.[@terminello93] The insight obtained through analysis of the powder spectra can be used to interpret similar data taken on thin BN films, which are likely to find use as both tribological coatings and as wide-gap semiconducting materials. [**EXPERIMENTAL DETAILS**]{} The three BN films used in this study were produced by ion-assisted pulsed laser deposition onto heated silicon substrates, as detailed previously.[@ballal92] Deposition conditions are summarized in Table 1. Infrared transmission measurements (not shown) on these films revealed absorption at wavelengths characteristic of both the cubic and hexagonal phases, allowing an estimate of their volume fraction. The purity of the three types of BN powder was determined using x-ray diffraction (not shown) to be greater than 90%. B and N K-edge x-ray absorption spectra were obtained at both the 8-1 and 8-2 beamlines at SSRL, as well as at the IBM/U8 beamline at NSLS. Details of the data acquisition have been reported previously.[@terminello93; @BNapl] A simple polynomial was adequate to fit the background of the powder spectra but it was necessary to use a gaussian-broadened step function[@stohrbook] to fit the background intensity for all the films’ B K-edge spectra. The position of the step was fixed at 2.125 eV above the B 1s $\pi^*$ peak, the energy where $\sigma ^*$ absorption begins in cBN. The absolute energies of spectral features are reproducible to within 0.3 eV from run to run, but the relative energies of the features are reproducible to within 0.05 eV. [**RESULTS**]{} Figure \[b1sspectra\]a shows the B K-edge spectra of the three BN/Si films, while Figure \[b1sspectra\]b shows the spectra of hBN, rBN and cBN powders, which have been described in detail previously.[@terminello93] The most notable feature of the hBN and rBN powder data is the presence of the $\pi^*$ feature at 192.0 eV, which is characteristic of $sp^2$ bonding.[@fomichev68; @davies81] This $\pi^*$ feature is absent in cBN powder due to the $sp^3$ nature of the bonding. In the $sp^2$-bonded layered materials, the $\pi^*$ feature has been described as a core exciton whose position below the conduction-band edge is a result of a 1.3 eV excitonic binding energy.[@carson87] Comparison of the data Figures \[b1sspectra\]a and \[b1sspectra\]b shows that the bonding in the BN/Si films is predominantly $sp^2$, similar to the hBN powder. However, there are three sharp peaks that appear in the B spectra of all three BN films just above the $\pi^*$ peak but which don’t appear in the hBN data. Parameters obtained from fits to these smaller peaks are collected in Table 1 along with fit parameters from the hBN and rBN powders. The BN/Si films’ spectra also have a peak at 199 eV in the middle of the empty $\sigma$ band which is not present in any of the powder spectra. N K-edge spectra for the BN/Si films and the three BN powders are shown in Figure \[n1sspectra\]. The first peak at 402 eV in the $sp^2$-bonded materials is again a $\pi^*$ feature, although it is not as intense or as far below the conduction-band edge as in the B K-edge spectra. In the Wannier model of excitonic behavior, the intensity and binding energy of an exciton are reduced when the core hole is created on the anionic site.[@carson87] Overall the film data is similar to that for hBN and rBN, although the intensity of the $\pi^*$ peak at 402 eV is reduced with respect to that of the the $\sigma^*$ peak at 408.5 eV in the films compared to the hBN or rBN standards. [**DISCUSSION**]{} Because infrared spectroscopy on these BN films has indicated the presence of cBN, it is logical to ask whether the unexplained features in the films’ spectra may result from a combination of hBN and cBN states. In fact, the occurrence of a step near 194 ev in the background intensity of the films’ B K-edge spectra is probably an indication of the cBN phase. The peak at 199 eV in the BN/Si B K-edge data may also be due to a contribution from cBN. Also attributable to cBN is the increased intensity of the $\sigma^*$ peak with respect to the $\pi^*$ peak in the films N K-edge spectra. The size of the above-mentioned deviations from the hBN spectra appears to be correlated with the amount of cBN indicated in each film by the IR measurements. The three fairly narrow peaks near 192.6, 193.2, and 194.0 eV in the films’ B K-edge data (see Table 1) do not have analogs in the cBN or hBN data, nor are there similar features in the K-edge spectra of disordered C films.[@denley80] With much lower energy resolution, Fomichev and Rumsh previously reported a single broad peak near 194.0 eV in their study of hBN powder.[@fomichev68] The nearly constant energy spacing shows that these peaks are not a Rydberg series[@stohrbook] and the approximately 0.6 eV magnitude of the spacing rules out an origin involving lattice vibrations. The proximity of these sharp peaks to the $\pi^*$ excitonic peak suggests a related origin. One possibility is that stacking in the films is disordered and that the individual peaks represent different stacking configurations. Given that BN films are often non-stoichiometric,[@wada92] another possibility is that these peaks represent a quasi-bound final state formed by a photoelectron and a N vacancy. The most obvious possibility is that these peaks are due to the presence of rBN, whose B K-edge spectrum shows similar small peaks, as documented in Table 1. The unidentified peaks would then be due to modification of the $\pi$ bonding by the ABCA stacking that is characteristic of rBN. Positive identification rBN as a minor constituent of these films would require further study. A closer look at Table 1 and Figures \[b1sspectra\]a and \[n1sspectra\]a shows that the peaks for the film synthesized with the higher ion beam voltage and higher deposition rate are considerably broader than those for the other two films. Energetic ions have likely caused damage to this film, corroborating the findings of a previous IR and photoemission spectroscopy study by Wada and Yamashita on ion-assisted evaporation of BN films.[@wada92] Wada and Yamashita also observed metallic B and a smaller amount of cBN in material grown at deposition rates above 1 Å/s.[@wada92] These results are consistent with the observation (see Table 1) that the film synthesized at the higher deposition rate has a lower fraction of $sp ^3$-bonded material. Given the evident degree of disorder in this film, the recent observation of a preferred crystallographic orientation came as a surprise. NEXAFS and transmission electron microscopy experiments showed the hexagonal layer planes were oriented close to orthogonal to the substrate.[@BNapl] This orientation is contrary to that usually observed in thin graphite films, where the c-axis is typically normal to the film-substrate interface.[@denley80] [**CONCLUSIONS**]{} B and N K-edge spectra have been acquired on both single-phase BN powders and on BN/Si films fabricated using ion-assisted pulsed-laser deposition. Comparison with the powder data shows the films to be primarily $sp^2$-bonded. Systematic differences between the films’ spectra and that of pure hBN suggest that another phase is present, probably the cBN that has been observed with infrared transmission measurements. Three sharp peaks appear in the films’ B K-edge spectra between the usual $\pi^*$ and $\sigma^*$ bands. These peaks may be related to disorder or may indicate the presence of the rBN phase in the films. Angle-dependent x-ray absorption studies have shown that the BN/Si films are preferentially oriented with the hexagonal axis nearly in the film plane. Transmission electron microscopy and photoemission studies are underway to help clarify issues of film morphology and stoichiometry. The near-edge x-ray absorption spectra of thin BN films are proving to be more complex to interpret than those of vapor-deposited C films. We would like to thank A.K. Ballal and L. Salamanca Riba for the TEM pictures, D.K. Shuh for assistance with the data collection and E.A. Hudson for helpful discussions. Part of this work was performed under the auspices of the U.S. Department of Energy by LLNL under contract No. W-7405-ENG-48. [99]{} J. Stöhr, , (Springer Verlag, New York, 1992). R.A. Rosenberg, P.J. Love, and V. Rehn, Phys. Rev. [**B 33**]{}, 4034 (1986). J.F. Morar, F.H. Himpsel, G. Hollinger, G. Hughes, and J.L. Jordan, Phys. Rev. Lett. [**54**]{}, 1960 (1985). D. Denley, P. Perfetti, R.S. Williams, D.A. Shirley, and J. Stöhr, Phys. Rev. [**B21**]{}, 2267 (1980). L.J. Terminello, A. Chaiken, D.A. Lapiano-Smith, G.L. Doll and T. Sato, submitted to J. Vac. Sci. Techn. A.K. Ballal, L. Salamanca Riba, G.L. Doll, C.A. Taylor, and R. Clarke, J. Mater. Res. [**7**]{}, 1618 (1992). A. Chaiken, L.J. Terminello, J. Wong, G.L. Doll and C.A. Taylor II, Appl. Phys. Lett. [**63**]{}, 2112 (1993). V.A. Fomichev and M.A. Rumsh, J. Phys. Chem. Solids [**29**]{}, 1015 (1968). B.M. Davies, F. Bassani, F.C. Brown, and C.G. Olson, Phys. Rev. [**B24**]{}, 3537 (1981). R.D. Carson and S.E. Schnatterly, Phys. Rev. Lett. [**59**]{}, 319 (1987). T. Wada and N. Yamashita, J. Vac. Sci. Techn. [**A10**]{}, 515 (1992).

ArXiv

--- abstract: 'We discuss theoretically the behavior of the velocity autocorrelation function in the Dissipative Particle Dynamics model. Two dynamical regimes are identified depending on the dimensionless model parameters. For low frictions a mean field behavior is observed in which the kinetic theory for DPD provides good predictions. For high frictions, collective hydrodynamic effects are dominant. We have performed numerical simulations that validate the theory presented.' address: | Departamento de Física Fundamental, Universidad Nacional\ de Educación a Distancia, C/ Senda del Rey s/n, E-28040, Madrid,\ Spain author: - Pep Español and Mar Serrano title: Dynamical regimes in DPD --- Introduction ============ The Dissipative Particle Dynamics model allows to simulate hydrodynamics at mesoscopic scales in which thermal fluctuations are important [@hoo92; @esp95]. For this reason, it appears as a good simulation technique for the study of complex fluids like polymer or colloidal suspensions where both hydrodynamic interactions and Brownian motion are important [@sch95; @boe97; @cov97; @gro97; @ijmp97]. Being an off lattice technique, it does not suffer from the restrictions imposed by the lattice as in Lattice Gas Cellular Automata or the Lattice Boltzmann approach and it is much more flexible for modeling. Even though the technique has a well sounded theoretical support and many applications have been undertaken, there is at present no systematic study of the region of parameters much suitable for simulation of particular hydrodynamic problems. In addition, recent simulations [@pag98] have shown deviations from the transport coefficients predicted by the kinetic theory developed by Marsh [*et al.*]{} [@mar97]. The two approximations involved in this kinetic theory are the small gradient expansion around local equilibrium and the molecular chaos hypothesis. However, it is difficult to investigate the origin of the discrepancies between theory and simulations within the kinetic theory context. The theory just produces the explicit expressions for the transport coefficients with no hint about its range of validity. It has been suggested that it is precisely in the region of parameters where kinetic theory fails where it is more sensible to conduct simulations that reproduce hydrodynamic behavior [@pag98]. We shed some light into the problem by presenting a theory that allows to compute the velocity autocorrelation function (vaf) of the dissipative particles. The theory is based on the physical picture in which the DPD particles are regarded as Brownian-like particles moving in an environment created by the rest of DPD particles. Strictly speaking, though, they are not Brownian particles because the total moment of the system is conserved. This approach was introduced by Groot and Warren as a way of computing the diffusion coefficient [@gro97]. We identify the basic dimensionless parameters which allow to classify and discuss the dynamical regimes displayed by the model. By assuming that the environment of DPD particle behaves hydrodynamically, it is possible to obtain an explicit analytical expression for the velocity autocorrelation function. Finally, we present numerical simulations that allow to validate the hypothesis made in the theory. The DPD model ============= The stochastic differential equations (SDE) that govern the position ${\bf r}_{i}$ and velocity ${\bf v}_{i}$ of the $i$-th particle of mass $m$ in DPD are given by [@esp95] $$\begin{aligned} d{\bf r}_i&=&{\bf v}_idt, \nonumber \\ md{\bf v}_i &=&-\gamma m\sum_{j}\omega(r_{ij}) ({\bf e}_{ij}\!\cdot\!{\bf v}_{ij}) {\bf e}_{ij}dt \nonumber\\ &+&\sigma\sum_j\omega^{1/2}(r_{ij}){\bf e}_{ij}dW_{ij}, \label{sde}\end{aligned}$$ where the following quantities are defined $$\begin{aligned} {\bf e}_{ij} &\equiv&\frac{{\bf r}_{ij}}{r_{ij}} \nonumber \\ {\bf r}_{ij}&\equiv&{\bf r}_i-{\bf r}_j, \nonumber \\ r_{ij}&\equiv&|{\bf r}_i-{\bf r}_j|, \nonumber \\ {\bf v}_{ij} &\equiv& {\bf v}_i-{\bf v}_j. \label{def1}\end{aligned}$$ In order to compare with the kinetic theory in Ref. [@mar97], it is assumed that the usual conservative force is not present. The noise amplitude $\sigma$ is given by the detailed balance condition $$\sigma=(2k_B T\gamma m)^{1/2}, \label{fd}$$ where $T$ is the temperature of the equilibrium state towards which the system relax (if the boundary conditions allow for it) and $k_B$ is Boltzmann’s constant. Finally, $dW_{ij}=dW_{ji}$ are independent increments of the Wiener process that obey the Itô calculus rule $$dW_{ij}dW_{{i'}{j'}} =(\delta_{i{i'}}\delta_{j{j'}} +\delta_{i{j'}}\delta_{j{i'}})dt, \label{ito}$$ i.e., $dW_{ij}$ is an infinitesimal of order $1/2$ [@gar83]. The dimensionless weight function $\omega(r)$ is normalized according to [@hoo92] $$\int d{\bf r} \omega(r) = \frac{1}{n}, \label{norm}$$ where $n$ is the number density of the system. In this paper we will work in two spatial dimensions (2D) and select the following weight function with range $r_c$, $$\omega(r) =\frac{3}{\pi r_c^2 n}\left(1-\frac{r}{r_c}\right), \label{ome}$$ if $r<r_c$ and zero if $r>r_c$. We discuss now which are the fundamental parameters for the DPD model. By appropriate choice of units of mass, time and space it is always possible to reduce the number of relevant parameters of the model. It is obvious that the dynamical regimes are independent of the units used, and will depend on [*dimensionless*]{} parameters only. There are six parameters in the model: $m, \gamma, r_c, k_BT, \lambda, L$, where $\lambda$ is the average distance between particles, related to the number density $n$ of particles by $\lambda = n^{-1/d}$, $d$ is the space dimension, and $L$ is the box size (or any other boundary length scale). From these six parameters we can form three dimensionless parameters. By defining the thermal velocity $v_T=(k_BT/m)^{1/2}$ we select $$\begin{aligned} \Omega &\equiv &\frac{\gamma r_c}{d v_T}=\frac{\tau_T}{d \tau_\gamma}, \nonumber \\ s &\equiv& \frac{r_c}{\lambda} , \nonumber\\ \mu &\equiv&\frac{L}{r_c}. \label{dim1} \end{aligned}$$ The physical meaning of these parameters is as follows: $\tau_T$ is the time taken by a particle moving at the thermal velocity to move a distance $r_c$, whereas $\tau_\gamma=\gamma^{-1}$ is the time associated to the friction. Therefore, the [*dimensionless friction*]{} $\Omega$ is the ratio of these two time scales. On the other hand, $s$ is the [*overlapping*]{} between particles which is related to the number of particles that are within the range of interaction (the [*action sphere*]{}) of a given one. Finally, $\mu$ is the dimensionless box length. These dimensionless parameters $\Omega, s,\mu$ fix the dynamical regimes of the model. Velocity autocorrelation function ================================= The velocity equation in (\[sde\]) can be written in the form $$\dot{{\bf v}}_i = -\gamma \left[\sum_{j\neq i}\omega(r_{ij}) {\bf e}_{ij}{\bf e}_{ij}\right]\!\cdot\!{\bf v}_{i} +\frac{\gamma}{d}{\bf V}_i(t) +\frac{\tilde{{\bf F}}_i}{m}, \label{rew}$$ where the random force is $\tilde{{\bf F}}_i dt=\sigma\sum_j\omega^{1/2}(r_{ij}){\bf e}_{ij}dW_{ij}$. We have introduced in (\[rew\]) the [*environment*]{} velocity by $${\bf V}_i(t)= d\sum_{j\neq i}\omega(r_{ij}){\bf e}_{ij}{\bf e}_{ij}\!\cdot\!{\bf v}_j. \label{vhidro}$$ This velocity is a weighted average of the velocities of the neighboring particles of particle $i$. Next, we observe that the factor of ${\bf v}_i$ in the right hand side of (\[rew\]) can be written as $$\sum_{j\ne i}\omega(r_{ij}){\bf e}_{ij}{\bf e}_{ij} = \int d{\bf r} \omega({\bf r}_i-{\bf r}) \frac{{\bf r}_i-{\bf r}}{|{\bf r}_i-{\bf r}|} \frac{{\bf r}_i-{\bf r}}{|{\bf r}_i-{\bf r}|} n({\bf r},t), \label{mic1}$$ where we have introduced the microscopic density field $n({\bf r},t)= \sum_{j\neq i}\delta({\bf r}_j-{\bf r})$. If we assume that the density field is constant with value $n$ (which will be confirmed by the results obtained later) then we may approximate $$\begin{aligned} && \int d{\bf r} \omega({\bf r}_i-{\bf r}) \frac{{\bf r}_i-{\bf r}}{|{\bf r}_i-{\bf r}|} \frac{{\bf r}_i-{\bf r}}{|{\bf r}_i-{\bf r}|} n({\bf r},t) \nonumber\\ &&\approx n \int d{\bf r} \omega({\bf r}) \frac{{\bf r}}{|{\bf r}|}\frac{{\bf r}}{|{\bf r}|} =\frac{{\bf 1}}{d} \label{mic2}\end{aligned}$$ The last equality is obtained by noting that the integral is an isotropic second order tensor, which must be proportional to the identity (the constant of proportionality can be obtained by taking the trace of the integral and using the normalization (\[norm\])). After using Eqns. (\[mic1\]) and (\[mic2\]) in Eqn. (\[rew\]) one obtains $$\begin{aligned} d{\bf r}_i&=&{\bf v}_idt, \nonumber \\ d{\bf v}_i &=&-\frac{\gamma}{d}\left[{\bf v}_{i}-{\bf V}_i\right]dt + \frac{\tilde{{\bf F}}_i}{m} dt. \label{sde-mf}\end{aligned}$$ We observe that the DPD particles behave similarly to Brownian particles but in a systematic velocity field determined by the rest of its neigbouring particles. The stochastic properties of the random force are not exactly those of a Brownian particle because the total momentum of the system is conserved but for the rest of the development they are irrelevant. The formal solution of Eqn. (\[sde-mf\]) is $$\begin{aligned} {\bf v}_i(t)&=&\exp\{-\gamma t/d\}{\bf v}_i(0) \nonumber\\ &+& \int_0^t dt'\exp\{-\gamma (t-t')/d\} \left[\frac{\gamma}{d}{\bf V}_i(t')+\frac{\tilde{{\bf F}}_i(t')}{m}\right]. \nonumber\\ \label{mfsol}\end{aligned}$$ By multiplying this equation by ${\bf v}_i(0)$ and ${\bf V}_i(0)$ and averaging one obtains a set of equations for the velocity autocorrelation function at equilibrium $$\begin{aligned} &&\frac{1}{d}\langle {\bf v}_i(t)\!\cdot\!{\bf v}_i(0)\rangle =\exp\{-\gamma t/d\}\frac{k_BT}{m} \nonumber\\ &+&\frac{\gamma}{d}\int_0^t dt'\exp\{-\gamma (t-t')/d\} \frac{1}{d}\langle{\bf V}_i(t')\!\cdot\!{\bf v}_i(0)\rangle, \nonumber\\ &&\frac{1}{d}\langle {\bf v}_i(t)\!\cdot\!{\bf V}_i(0)\rangle = \nonumber\\ &&\frac{\gamma}{d}\int_0^t dt'\exp\{-\gamma (t-t')/d\} \frac{1}{d}\langle{\bf V}_i(t')\!\cdot\!{\bf V}_i(0)\rangle, \nonumber\\ \label{acvf}\end{aligned}$$ where use has been made of the fact that the random force is not correlated with the velocity at present and past times and the property $\langle{\bf V}_i(0)\!\cdot\!{\bf v}_i(0)\rangle=0$ (which can be checked from the definition (\[vhidro\])). Substitution of the second equation in (\[acvf\]) into the first one leads to an expression that relates the particle vaf with the environment vaf, this is $$\begin{aligned} &&\frac{1}{d}\langle {\bf v}_i(t)\!\cdot\!{\bf v}_i(0)\rangle =\exp\{-\gamma t/d\}\frac{k_BT}{m} \nonumber\\ &&+\left(\frac{\gamma}{d}\right)^2 \int_0^t dt'(t-t')\exp\{-\gamma (t-t')/d\} \frac{1}{d}\langle{\bf V}_i(t')\!\cdot\!{\bf V}_i(0)\rangle. \nonumber\\ \label{acvf01}\end{aligned}$$ The second term in the right hand side represents [*collective effects*]{}. When this term is negligible we say that a [*mean field*]{} approximation is valid, in which the velocity autocorrelation function decays exponentially. The reason for the name “mean field” comes form the observation that Eqn. (\[sde-mf\]), in which the average value $\langle {\bf V}_i\rangle=0$ is used instead of the instantaneous value ${\bf V}_i$, produces an exponential decay of the velocity autocorrelation function. In the Appendix of Ref. [@gro97] it was computed the velocity autocorrelation function and the diffusion coefficient of the DPD particles by using this mean field approximation. The solution (\[acvf01\]) is still formal because we do not know explicitly the form of the correlation of the environment velocity (which will be given in the next section). Nevertheless, it is possible to extract useful information from this expression. This is most conveniently done by taking dimensionless variables. Let ${\overline t}$ be the dimensionless time $t v_T/r_c$, this is, the time expressed in units in which $r_c=1$ and $v_T=1$, and ${\overline {\bf v}}={\bf v}/v_T$ a dimensionless velocity. In these units, Eqns. (\[acvf\]) take the form $$\begin{aligned} \frac{1}{d}\langle {\overline {\bf v}}_i({\overline t})\!\cdot\!{\overline {\bf v}}_i(0)\rangle &=&\exp\{-\Omega {\overline t}\} \nonumber\\ &+&\Omega\int_0^{\overline t} d{\overline t}'\exp\{-\Omega ({\overline t}-{\overline t}')\} \frac{1}{d}\langle{\overline {\bf V}}_i({\overline t}')\!\cdot\!{\overline {\bf v}}_i(0)\rangle , \nonumber\\ \frac{1}{d}\langle {\overline {\bf v}}_i({\overline t})\!\cdot\!{\overline {\bf V}}_i(0)\rangle &=& \Omega\int_0^{\overline t} d{\overline t}'\exp\{-\Omega ({\overline t}-{\overline t}')\} \frac{1}{d}\langle{\overline {\bf V}}_i({\overline t}')\!\cdot\!{\overline {\bf V}}_i(0)\rangle, \nonumber\\ \label{acvfbis}\end{aligned}$$ and Eqn. (\[acvf01\]) takes the form $$\begin{aligned} &&\frac{1}{d} \langle {\overline {\bf v}}_i({\overline t})\!\cdot\!{\overline {\bf v}}_i(0)\rangle =\exp\{-\Omega {\overline t}\} \nonumber\\ &&+\Omega^2 \int_0^{\overline t} d{\overline t}'({\overline t}-{\overline t}') \exp\{-\Omega ({\overline t}-{\overline t}')\} \frac{1}{d} \langle{\overline {\bf V}}_i({\overline t}')\!\cdot\!{\overline {\bf V}}_i(0)\rangle \label{acvf02}\end{aligned}$$ For later notational convenience we introduce $$\begin{aligned} c({\overline t})&\equiv&\frac{1}{d} \langle {\overline {\bf v}}_i({\overline t}) \!\cdot\!{\overline {\bf v}}_i(0)\rangle, \nonumber\\ C({\overline t})&\equiv& \frac{1}{d}\langle{\overline {\bf V}}_i({\overline t}) \!\cdot\!{\overline {\bf V}}_i(0)\rangle. \label{dein}\end{aligned}$$ In order to complete the picture we need to evaluate the overall magnitude of the correlation of the environment velocity. This is determined by the value at the origin which is computed in the Appendix I with the result $$\frac{1}{d}\langle{\overline {\bf V}}_i(0)\!\cdot\!{\overline {\bf V}}_i(0)\rangle =\frac{1}{s^2}\frac{3 d}{2\pi}. \label{evt0}$$ We observe that the magnitude of this correlation decreases with the overlapping coefficient. This is physically meaningful because the environment velocity is a weighted average of the velocities of the particles that are within an action sphere. These velocities are distributed at random and, therefore, if there are many particles within an action sphere, the average will be proportionally smaller. Now, several qualitative predictions concerning the different dynamical regimes can be made from expression (\[acvfbis\]) or (\[acvf02\]). For fixed $\Omega$, the large overlapping $s$ limit produces a small contribution from the collective part and the velocity correlation function decays in an exponential way. For fixed overlapping $s$, when $\Omega$ is small (in the limit of small friction or high temperature) the behavior of the vaf is again exponential. In the opposite regime of large $\Omega$, the exponential contribution decays in a very short time and the main contribution for times larger than $\Omega^{-1}$ is given by the collective term. Actually, in the limit $\Omega\rightarrow\infty$ the exponential memory function acts as a delta function and for times larger than $\Omega ^{-1}$ one obtains $$\langle{\overline {\bf v}}_i({\overline t})\!\cdot\!{\overline {\bf v}}_i(0)\rangle \approx \langle{\overline {\bf V}}_i({\overline t})\!\cdot\!{\overline {\bf v}}_i(0)\rangle \approx \langle{\overline {\bf V}}_i({\overline t})\!\cdot\!{\overline {\bf V}}_i(0)\rangle. \label{highomeg}$$ The physical meaning of these expressions (\[highomeg\]) is also clear. When the friction is high, in a very short time the velocity of a given particle is slaved by the average velocity of its environment. Hydrodynamic hypothesis ======================= The environment velocity ${\bf V}_i$ defined in Eqn. (\[vhidro\]) can be rewritten as an average over the action sphere of the microscopic velocity field, this is, $${\bf V}_i(t)=d\int d{\bf r} \omega({\bf r}_i-{\bf r}) \frac{{\bf r}_i-{\bf r}}{|{\bf r}_i-{\bf r}|} \frac{{\bf r}_i-{\bf r}}{|{\bf r}_i-{\bf r}|} n({\bf r},t){\bf v}({\bf r},t), \label{hh1}$$ where $$n({\bf r},t){\bf v}({\bf r},t) =\sum_{j\neq i}{\bf v}_j\delta({\bf r}_j-{\bf r}).$$ The velocity field ${\bf v}({\bf r},t)$ obeys the equations of hydrodynamics when its characteristic length scale is much larger than the interparticle distance. We expect that the average involved in ${\bf V}_i$ will be dominated by the hydrodynamic modes whenever the range of interaction $r_c$ is much larger than the interparticle distance $\lambda$ (i.e., large overlapping $s$). In this section we present a calculation of the correlation function of the environment velocity which is based on this hydrodynamic argument. From Eqn. (\[hh1\]), the environment velocity is conveniently expressed in terms of the Fourier components of the velocity field ${\bf v}({\bf k},t)$, this is $${\bf V}_i(t) = dn\int \frac{d{\bf k}}{(2 \pi)^2} {\bf \omega}({\bf k})\!\cdot\! {\bf v}({\bf k},t) \exp\{i {\bf k}{\bf r}_i\}, \label{ve1}$$ where we have introduced the second order tensor $${\bf \omega}({\bf k})= \int d{\bf r}\omega(r){\bf \hat{r}}{\bf \hat{r}} \exp\{-i {\bf k }{\bf r}\}.$$ The explicit form of this tensor when the weight function is given by Eqn. (\[ome\]) is given in the Appendix II. The environment velocity correlation function is given by $$\begin{aligned} \langle {\bf V}_i(0)\!\cdot\!{\bf V}_i(t)\rangle &=&(d n )^2 \int \frac{ d{\bf k}}{(2 \pi)^2}\frac{ d{\bf k'}}{(2 \pi)^2} {\bf \omega}({\bf k}){\bf \omega}({\bf k}') \nonumber\\ &\times & \langle \exp\{i ({\bf k }{\bf r}_i(0)+{\bf k' }{\bf r}_i(t))\} {\bf v}({\bf k},0){\bf v}({\bf k'},t)\rangle. \nonumber\\ \label{evacf}\end{aligned}$$ We will assume that the position of particle $i$ is weakly correlated with the velocity field ${\bf v}({\bf k},t)$ in such a way that we can approximate $$\begin{aligned} &&\langle \exp\{i ({\bf k }{\bf r}_i(0)+{\bf k' }{\bf r}_i(t))\} {\bf v}({\bf k},0){\bf v}({\bf k'},t)\rangle \nonumber\\ &&\approx \langle \exp\{i ({\bf k }{\bf r}_i(0)+{\bf k' }{\bf r}_i(t))\}\rangle \;\;\langle {\bf v}({\bf k},0){\bf v}({\bf k'},t)\rangle. \label{2vm}\end{aligned}$$ We further assume that the correlation of the velocity field is given by the linear hydrodynamics result [@lin] $$\begin{aligned} &&\langle {\bf v}({\bf k'},0){\bf v}^{T}({\bf k},t)\rangle = \frac{k_BT}{nm} (2 \pi)^{2} \delta ({\bf k}+{\bf k}') \nonumber\\ &&\times [\exp\{-\nu k^{2} t\} ({\bf 1} -{\bf\hat{k}} {\bf\hat{k}}) +\exp\{-\Gamma k^{2} t\} \cos{kct} {\bf \hat{k}} {\bf \hat{k}}]. \label{vfacf}\end{aligned}$$ Here, $\nu$ is the kinematic viscosity, $\Gamma$ is the sound absorption coefficient, and $c$ is the sound speed of the DPD fluid. The correlation function (\[vfacf\]) is different from zero only when ${\bf k}=-{\bf k}'$. The first average in the right hand side of (\[2vm\]) is, therefore, given by the incoherent intermediate scattering function $F_s({\bf k },t)= \langle \exp\{i {\bf k }\!\cdot\!({\bf r}_i(0)-{\bf r}_i(t))\}\rangle$ [@lin]. By further assuming a hydrodynamic behavior for this function we obtain [@lin] $$F_s({\bf k },t) = \exp\{- D k^2 t\}, \label{diff}$$ where $D$ is the self-diffusion coefficient of the DPD particles. The final hydrodynamic expression for the environment velocity correlation function is found by substitution of (\[2vm\]),(\[vfacf\]),(\[diff\]), into (\[evacf\]) $$\begin{aligned} \frac{1}{d}\langle {\bf V}_i(0)\!\cdot\!{\bf V}_i(t)\rangle &=& \frac{3d k_BT}{4\pi r_c^2 n m} \left[\Phi\left (\frac{(\nu+D)t}{r_c^2}\right )\right. \nonumber\\ &+&\left. \Psi\left(\frac{(\Gamma+D)t}{r_c^2},\frac{ct}{r_c}\right )\right], \label{hp1}\end{aligned}$$ where the following functions are defined $$\begin{aligned} \Phi \left (x \right ) &=&\frac{\int d{\bf k} a^2(k) \exp\{-x k^{2} \}}{\int d{\bf k} a^2(k)} \nonumber\\ \Psi \left (y,z\right )&=& \frac{\int d{\bf k} (a(k)+b(k))^2 \exp\{-y k^{2}\} \cos{kz}} {\int d{\bf k} (a(k)+b(k))^2} , \label{hp0}\end{aligned}$$ which satisfy $\Phi(0)=1,\Psi(0,0)=1$, with the definitions for $a(k), b(k)$ given in Appendix II. In this way, the environment velocity correlation function is explicitly given in terms of hydrodynamic fluid properties, i.e., the transport coefficients of the DPD fluid. The prediction is not complete until particular values for these transport coefficients are provided. One possibility is to measure these transport coefficients in a simulation. Another possibility, which is the one we follow here, is to use the values for $\nu, D, \Gamma, c$ provided by kinetic theory [@mar97]. This has the advantage that one knows the explicit dependence of the transport coefficients in terms of the dimensionless parameters $\Omega,s$. Even though for some dynamical regimes the kinetic theory predictions are not in exact quantitative agreement with the measured transport coefficients [@pag98; @rev98] the discrepancies between theoretical and simulation transport coefficients are small. The kinetic theory results are [@mar97],[@esp98] $$\begin{aligned} D&=&\frac{d}{\gamma}\frac{k_BT}{m}, \nonumber\\ \nu &=&\frac{1}{2}\left[ \gamma n \frac{1}{d(d+2)}\int r^{2} \omega(r)d{\bf r} +c^{2} \frac{d}{\gamma n \int \omega (r) d {\bf r} }\right], \nonumber \\ \nu_b &=& \gamma n \frac{1}{2d^{2}}\int r^{2} \omega(r)d{\bf r} +c^{2} \frac{1}{\gamma n \int \omega (r) d {\bf r} }, \nonumber \\ c &=&\sqrt{\frac{k_B T}{m}}. \label{trans1}\end{aligned}$$ The sound absorption coefficient is given by $\Gamma = 2 \nu + \frac{1}{2}\nu_b$ for the case that the equation of state is that of the ideal gas [@lin]. We are now in position to write the hydrodynamic prediction (\[hp1\]) in terms of dimensionless variables. Substitution of (\[trans1\]) with the expression (\[ome\]) into (\[hp1\]) leads to $$\begin{aligned} C({\overline t}) &=& \frac{3 d }{4 \pi s^2 }\left[\Phi\left(\left[\frac{3}{80}\Omega +\frac{3}{2\Omega} \right]{\overline t}\right )\right. \nonumber\\ &+&\left.\Psi\left( \left[\frac{9}{80}\Omega +\frac{11}{4\Omega}\right]{\overline t},{\overline t} \right)\right]. \label{hp3}\end{aligned}$$ We observe that the time scale of the environment velocity correlation function is determined by $\Omega$, while its amplitude is determined by $s$. In Fig. \[fig1\] we show the theoretical prediction (\[hp3\]) for a particular value of $s$ and three different values of $\Omega$. An algebraic dependence as $t^{-1}$ is observed for very long times which is the celebrated long time tail $t^{-d/2}$ arising from the diffusive character of the shear mode. (974,675)(0,0) =cmr10 at 10pt (220.0,113.0) ------------------------------------------------------------------------ (198,113)[(0,0)\[r\][$10^{-6}$]{}]{} (890.0,113.0) ------------------------------------------------------------------------ (220.0,144.0) ------------------------------------------------------------------------ (900.0,144.0) ------------------------------------------------------------------------ (220.0,184.0) ------------------------------------------------------------------------ (900.0,184.0) ------------------------------------------------------------------------ (220.0,205.0) ------------------------------------------------------------------------ (900.0,205.0) ------------------------------------------------------------------------ (220.0,215.0) ------------------------------------------------------------------------ (198,215)[(0,0)\[r\][$10^{-5}$]{}]{} (890.0,215.0) ------------------------------------------------------------------------ (220.0,245.0) ------------------------------------------------------------------------ (900.0,245.0) ------------------------------------------------------------------------ (220.0,286.0) ------------------------------------------------------------------------ (900.0,286.0) ------------------------------------------------------------------------ (220.0,307.0) ------------------------------------------------------------------------ (900.0,307.0) ------------------------------------------------------------------------ (220.0,316.0) ------------------------------------------------------------------------ (198,316)[(0,0)\[r\][$10^{-4}$]{}]{} (890.0,316.0) ------------------------------------------------------------------------ (220.0,347.0) ------------------------------------------------------------------------ (900.0,347.0) ------------------------------------------------------------------------ (220.0,387.0) ------------------------------------------------------------------------ (900.0,387.0) ------------------------------------------------------------------------ (220.0,408.0) ------------------------------------------------------------------------ (900.0,408.0) ------------------------------------------------------------------------ (220.0,418.0) ------------------------------------------------------------------------ (198,418)[(0,0)\[r\][$10^{-3}$]{}]{} (890.0,418.0) ------------------------------------------------------------------------ (220.0,449.0) ------------------------------------------------------------------------ (900.0,449.0) ------------------------------------------------------------------------ (220.0,489.0) ------------------------------------------------------------------------ (900.0,489.0) ------------------------------------------------------------------------ (220.0,510.0) ------------------------------------------------------------------------ (900.0,510.0) ------------------------------------------------------------------------ (220.0,520.0) ------------------------------------------------------------------------ (198,520)[(0,0)\[r\][$10^{-2}$]{}]{} (890.0,520.0) ------------------------------------------------------------------------ (220.0,550.0) ------------------------------------------------------------------------ (900.0,550.0) ------------------------------------------------------------------------ (220.0,591.0) ------------------------------------------------------------------------ (900.0,591.0) ------------------------------------------------------------------------ (220.0,612.0) ------------------------------------------------------------------------ (900.0,612.0) ------------------------------------------------------------------------ (220.0,621.0) ------------------------------------------------------------------------ (198,621)[(0,0)\[r\][$10^{-1}$]{}]{} (890.0,621.0) ------------------------------------------------------------------------ (220.0,652.0) ------------------------------------------------------------------------ (900.0,652.0) ------------------------------------------------------------------------ (220.0,113.0) ------------------------------------------------------------------------ (220,68)[(0,0)[0.01]{}]{} (220.0,632.0) ------------------------------------------------------------------------ (262.0,113.0) ------------------------------------------------------------------------ (262.0,642.0) ------------------------------------------------------------------------ (316.0,113.0) ------------------------------------------------------------------------ (316.0,642.0) ------------------------------------------------------------------------ (345.0,113.0) ------------------------------------------------------------------------ (345.0,642.0) ------------------------------------------------------------------------ (358.0,113.0) ------------------------------------------------------------------------ (358,68)[(0,0)[0.1]{}]{} (358.0,632.0) ------------------------------------------------------------------------ (400.0,113.0) ------------------------------------------------------------------------ (400.0,642.0) ------------------------------------------------------------------------ (454.0,113.0) ------------------------------------------------------------------------ (454.0,642.0) ------------------------------------------------------------------------ (483.0,113.0) ------------------------------------------------------------------------ (483.0,642.0) ------------------------------------------------------------------------ (496.0,113.0) ------------------------------------------------------------------------ (496,68)[(0,0)[1]{}]{} (496.0,632.0) ------------------------------------------------------------------------ (538.0,113.0) ------------------------------------------------------------------------ (538.0,642.0) ------------------------------------------------------------------------ (592.0,113.0) ------------------------------------------------------------------------ (592.0,642.0) ------------------------------------------------------------------------ (621.0,113.0) ------------------------------------------------------------------------ (621.0,642.0) ------------------------------------------------------------------------ (634.0,113.0) ------------------------------------------------------------------------ (634,68)[(0,0)[10]{}]{} (634.0,632.0) ------------------------------------------------------------------------ (676.0,113.0) ------------------------------------------------------------------------ (676.0,642.0) ------------------------------------------------------------------------ (730.0,113.0) ------------------------------------------------------------------------ (730.0,642.0) ------------------------------------------------------------------------ (759.0,113.0) ------------------------------------------------------------------------ (759.0,642.0) ------------------------------------------------------------------------ (772.0,113.0) ------------------------------------------------------------------------ (772,68)[(0,0)[100]{}]{} (772.0,632.0) ------------------------------------------------------------------------ (814.0,113.0) ------------------------------------------------------------------------ (814.0,642.0) ------------------------------------------------------------------------ (868.0,113.0) ------------------------------------------------------------------------ (868.0,642.0) ------------------------------------------------------------------------ (897.0,113.0) ------------------------------------------------------------------------ (897.0,642.0) ------------------------------------------------------------------------ (910.0,113.0) ------------------------------------------------------------------------ (910,68)[(0,0)[1000]{}]{} (910.0,632.0) ------------------------------------------------------------------------ (220.0,113.0) ------------------------------------------------------------------------ (910.0,113.0) ------------------------------------------------------------------------ (220.0,652.0) ------------------------------------------------------------------------ (45,382)[(0,0)[$C({\overline t})$]{}]{} (565,23)[(0,0)[${\overline t}$]{}]{} (220.0,113.0) ------------------------------------------------------------------------ (220,604) (220.00,602.92)(1.958,-0.492)[19]{} ------------------------------------------------------------------------ (220.00,603.17)(38.623,-11.000)[2]{} ------------------------------------------------------------------------ (262.00,591.93)(1.368,-0.489)[15]{} ------------------------------------------------------------------------ (262.00,592.17)(21.579,-9.000)[2]{} ------------------------------------------------------------------------ (286.00,582.93)(1.485,-0.482)[9]{} ------------------------------------------------------------------------ (286.00,583.17)(14.440,-6.000)[2]{} ------------------------------------------------------------------------ (303.00,576.93)(1.123,-0.482)[9]{} ------------------------------------------------------------------------ (303.00,577.17)(10.994,-6.000)[2]{} ------------------------------------------------------------------------ (316.00,570.93)(1.155,-0.477)[7]{} ------------------------------------------------------------------------ (316.00,571.17)(8.966,-5.000)[2]{} ------------------------------------------------------------------------ (327.00,565.93)(1.044,-0.477)[7]{} ------------------------------------------------------------------------ (327.00,566.17)(8.132,-5.000)[2]{} ------------------------------------------------------------------------ (337.00,560.94)(1.066,-0.468)[5]{} ------------------------------------------------------------------------ (337.00,561.17)(6.132,-4.000)[2]{} ------------------------------------------------------------------------ (345.00,556.95)(1.355,-0.447)[3]{} ------------------------------------------------------------------------ (345.00,557.17)(4.855,-3.000)[2]{} ------------------------------------------------------------------------ (352.00,553.94)(0.774,-0.468)[5]{} ------------------------------------------------------------------------ (352.00,554.17)(4.547,-4.000)[2]{} ------------------------------------------------------------------------ (358.00,549.95)(1.132,-0.447)[3]{} ------------------------------------------------------------------------ (358.00,550.17)(4.132,-3.000)[2]{} ------------------------------------------------------------------------ (364.00,546.95)(0.909,-0.447)[3]{} ------------------------------------------------------------------------ (364.00,547.17)(3.409,-3.000)[2]{} ------------------------------------------------------------------------ (369.00,543.95)(0.909,-0.447)[3]{} ------------------------------------------------------------------------ (369.00,544.17)(3.409,-3.000)[2]{} ------------------------------------------------------------------------ (374,540.17) ------------------------------------------------------------------------ (374.00,541.17)(2.132,-2.000)[2]{} ------------------------------------------------------------------------ (378.00,538.95)(0.685,-0.447)[3]{} ------------------------------------------------------------------------ (378.00,539.17)(2.685,-3.000)[2]{} ------------------------------------------------------------------------ (382,535.17) ------------------------------------------------------------------------ (382.00,536.17)(2.132,-2.000)[2]{} ------------------------------------------------------------------------ (386,533.17) ------------------------------------------------------------------------ (386.00,534.17)(2.132,-2.000)[2]{} ------------------------------------------------------------------------ (390,531.17) ------------------------------------------------------------------------ (390.00,532.17)(1.547,-2.000)[2]{} ------------------------------------------------------------------------ (393.00,529.95)(0.462,-0.447)[3]{} ------------------------------------------------------------------------ (393.00,530.17)(1.962,-3.000)[2]{} ------------------------------------------------------------------------ (396,526.67) ------------------------------------------------------------------------ (396.00,527.17)(2.000,-1.000)[2]{} ------------------------------------------------------------------------ (400,525.17) ------------------------------------------------------------------------ (400.00,526.17)(1.000,-2.000)[2]{} ------------------------------------------------------------------------ (402,523.17) ------------------------------------------------------------------------ (402.00,524.17)(1.547,-2.000)[2]{} ------------------------------------------------------------------------ (405,521.17) ------------------------------------------------------------------------ (405.00,522.17)(1.547,-2.000)[2]{} ------------------------------------------------------------------------ (408,519.67) ------------------------------------------------------------------------ (408.00,520.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (410,518.17) ------------------------------------------------------------------------ (410.00,519.17)(1.547,-2.000)[2]{} ------------------------------------------------------------------------ (413,516.17) ------------------------------------------------------------------------ (413.00,517.17)(1.000,-2.000)[2]{} ------------------------------------------------------------------------ (415,514.67) ------------------------------------------------------------------------ (415.00,515.17)(1.500,-1.000)[2]{} ------------------------------------------------------------------------ (418,513.17) ------------------------------------------------------------------------ (418.00,514.17)(1.000,-2.000)[2]{} ------------------------------------------------------------------------ (420,511.67) ------------------------------------------------------------------------ (420.00,512.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (422,510.67) ------------------------------------------------------------------------ (422.00,511.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (424,509.17) ------------------------------------------------------------------------ (424.00,510.17)(1.000,-2.000)[2]{} ------------------------------------------------------------------------ (426,507.67) ------------------------------------------------------------------------ (426.00,508.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (428,506.67) ------------------------------------------------------------------------ (428.00,507.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (430,505.67) ------------------------------------------------------------------------ (430.00,506.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (431,504.17) ------------------------------------------------------------------------ (431.00,505.17)(1.000,-2.000)[2]{} ------------------------------------------------------------------------ (433,502.67) ------------------------------------------------------------------------ (433.00,503.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (435,501.67) ------------------------------------------------------------------------ (435.00,502.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (436,500.67) ------------------------------------------------------------------------ (436.00,501.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (438,499.67) ------------------------------------------------------------------------ (438.00,500.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (440,498.67) ------------------------------------------------------------------------ (440.00,499.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (441,497.67) ------------------------------------------------------------------------ (441.00,498.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (443,496.67) ------------------------------------------------------------------------ (443.00,497.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (444,495.67) ------------------------------------------------------------------------ (444.00,496.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (445,494.67) ------------------------------------------------------------------------ (445.00,495.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (447,493.67) ------------------------------------------------------------------------ (447.00,494.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (448,492.67) ------------------------------------------------------------------------ (448.00,493.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (449,491.67) ------------------------------------------------------------------------ (449.00,492.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (451,490.67) ------------------------------------------------------------------------ (451.00,491.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (452,489.67) ------------------------------------------------------------------------ (452.00,490.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (454,488.67) ------------------------------------------------------------------------ (454.00,489.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (456,487.67) ------------------------------------------------------------------------ (456.00,488.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (457,486.67) ------------------------------------------------------------------------ (457.00,487.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (458,485.67) ------------------------------------------------------------------------ (458.00,486.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (453.0,490.0) (460,484.67) ------------------------------------------------------------------------ (460.00,485.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (461,483.67) ------------------------------------------------------------------------ (461.00,484.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (462,482.67) ------------------------------------------------------------------------ (462.00,483.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (459.0,486.0) (464,481.67) ------------------------------------------------------------------------ (464.00,482.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (465,480.67) ------------------------------------------------------------------------ (465.00,481.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (466,479.67) ------------------------------------------------------------------------ (466.00,480.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (463.0,483.0) (468,478.67) ------------------------------------------------------------------------ (468.00,479.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (469,477.67) ------------------------------------------------------------------------ (469.00,478.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (467.0,480.0) (471,476.67) ------------------------------------------------------------------------ (471.00,477.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (472,475.67) ------------------------------------------------------------------------ (472.00,476.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (470.0,478.0) (474,474.67) ------------------------------------------------------------------------ (474.00,475.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (473.0,476.0) (475,475) (475,473.67) ------------------------------------------------------------------------ (475.00,474.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (476,472.67) ------------------------------------------------------------------------ (476.00,473.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (478,471.67) ------------------------------------------------------------------------ (478.00,472.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (477.0,473.0) (479.0,472.0) (480.0,471.0) (481,469.67) ------------------------------------------------------------------------ (481.00,470.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (482,468.67) ------------------------------------------------------------------------ (482.00,469.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (480.0,471.0) (483,469) (483,467.67) ------------------------------------------------------------------------ (483.00,468.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (485,466.67) ------------------------------------------------------------------------ (485.00,467.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (484.0,468.0) (486,467) (486,465.67) ------------------------------------------------------------------------ (486.00,466.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (487.0,466.0) (488.0,465.0) (489,463.67) ------------------------------------------------------------------------ (489.00,464.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (488.0,465.0) (490,464) (490,462.67) ------------------------------------------------------------------------ (490.00,463.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (491.0,463.0) (492.0,462.0) (493,460.67) ------------------------------------------------------------------------ (493.00,461.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (492.0,462.0) (494,461) (494,459.67) ------------------------------------------------------------------------ (494.00,460.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (495,460) (496,458.67) ------------------------------------------------------------------------ (496.00,459.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (495.0,460.0) (497,459) (497,457.67) ------------------------------------------------------------------------ (497.00,458.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (498,458) (498,456.67) ------------------------------------------------------------------------ (498.00,457.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (499,457) (500,455.67) ------------------------------------------------------------------------ (500.00,456.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (499.0,457.0) (501,456) (501,454.67) ------------------------------------------------------------------------ (501.00,455.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (502,455) (502,453.67) ------------------------------------------------------------------------ (502.00,454.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (503,454) (503.0,454.0) (504.0,453.0) (505,451.67) ------------------------------------------------------------------------ (505.00,452.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (504.0,453.0) (506,452) (506,450.67) ------------------------------------------------------------------------ (506.00,451.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (507,451) (507.0,451.0) (508.0,450.0) (509,448.67) ------------------------------------------------------------------------ (509.00,449.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (508.0,450.0) (510,449) (510,447.67) ------------------------------------------------------------------------ (510.00,448.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (511,448) (511.0,448.0) (512.0,447.0) (513,445.67) ------------------------------------------------------------------------ (513.00,446.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (512.0,447.0) (514,446) (514,446) (514,444.67) ------------------------------------------------------------------------ (514.00,445.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (515,445) (515.0,445.0) (516.0,444.0) (516.0,444.0) (517.0,443.0) (518,441.67) ------------------------------------------------------------------------ (518.00,442.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (517.0,443.0) (519,442) (519,442) (519.0,442.0) (520.0,441.0) (520.0,441.0) (521.0,440.0) (522,438.67) ------------------------------------------------------------------------ (522.00,439.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (521.0,440.0) (523,439) (523,439) (523.0,439.0) (524.0,438.0) (524.0,438.0) (525.0,437.0) (526,435.67) ------------------------------------------------------------------------ (526.00,436.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (525.0,437.0) (527,436) (527,436) (527,434.67) ------------------------------------------------------------------------ (527.00,435.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (528,435) (528.0,435.0) (529.0,434.0) (530,432.67) ------------------------------------------------------------------------ (530.00,433.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (529.0,434.0) (531,433) (531,433) (531.0,433.0) (532.0,432.0) (532.0,432.0) (533.0,431.0) (533.0,431.0) (534.0,430.0) (535,428.67) ------------------------------------------------------------------------ (535.00,429.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (534.0,430.0) (536,429) (536,429) (536.0,429.0) (537.0,428.0) (537.0,428.0) (538.0,427.0) (538.0,427.0) (539.0,426.0) (540,424.67) ------------------------------------------------------------------------ (540.00,425.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (539.0,426.0) (541,425) (541,425) (541.0,425.0) (542.0,424.0) (542.0,424.0) (543.0,423.0) (543.0,423.0) (544.0,422.0) (545,420.67) ------------------------------------------------------------------------ (545.00,421.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (544.0,422.0) (546,421) (546,421) (546,421) (546.0,421.0) (547.0,420.0) (547.0,420.0) (548.0,419.0) (548.0,419.0) (549.0,418.0) (550,416.67) ------------------------------------------------------------------------ (550.00,417.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (549.0,418.0) (551,417) (551,417) (551,417) (551.0,417.0) (552.0,416.0) (552.0,416.0) (553.0,415.0) (553.0,415.0) (554.0,414.0) (554.0,414.0) ------------------------------------------------------------------------ (556.0,413.0) (556.0,413.0) (557.0,412.0) (557.0,412.0) (558.0,411.0) (558.0,411.0) (559.0,410.0) (560,408.67) ------------------------------------------------------------------------ (560.00,409.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (559.0,410.0) (561,409) (561,409) (561,409) (561,409) (561.0,409.0) (562.0,408.0) (562.0,408.0) (563.0,407.0) (563.0,407.0) (564.0,406.0) (565,404.67) ------------------------------------------------------------------------ (565.00,405.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (564.0,406.0) (566,405) (566,405) (566,405) (566,405) (566,405) (566,403.67) ------------------------------------------------------------------------ (566.00,404.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (567,404) (567,404) (567,404) (567,404) (567.0,404.0) (568.0,403.0) (568.0,403.0) (569.0,402.0) (569.0,402.0) (570.0,401.0) (571,399.67) ------------------------------------------------------------------------ (571.00,400.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (570.0,401.0) (572,400) (572,400) (572,400) (572,400) (572,400) (572.0,400.0) (573.0,399.0) (573.0,399.0) (574.0,398.0) (574.0,398.0) (575.0,397.0) (575.0,397.0) (576.0,396.0) (576.0,396.0) (577.0,395.0) (578,393.67) ------------------------------------------------------------------------ (578.00,394.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (577.0,395.0) (579,394) (579,394) (579,394) (579,394) (579,394) (579.0,394.0) (580.0,393.0) (580.0,393.0) (581.0,392.0) (581.0,392.0) (582.0,391.0) (582.0,391.0) (583.0,390.0) (584,388.67) ------------------------------------------------------------------------ (584.00,389.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (583.0,390.0) (585,389) (585,389) (585,389) (585,389) (585,389) (585,389) (585,389) (585.0,389.0) (586.0,388.0) (586.0,388.0) (587.0,387.0) (587.0,387.0) (588.0,386.0) (588.0,386.0) (589.0,385.0) (589.0,385.0) (590.0,384.0) (590.0,384.0) (591.0,383.0) (591.0,383.0) ------------------------------------------------------------------------ (593.0,382.0) (593.0,382.0) (594.0,381.0) (594.0,381.0) (595.0,380.0) (595.0,380.0) (596.0,379.0) (596.0,379.0) (597.0,378.0) (597.0,378.0) (598.0,377.0) (599,375.67) ------------------------------------------------------------------------ (599.00,376.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (598.0,377.0) (600,376) (600,376) (600,376) (600,376) (600,376) (600,376) (600,376) (600,376) (600.0,376.0) (601.0,375.0) (601.0,375.0) (602.0,374.0) (602.0,374.0) (603.0,373.0) (603.0,373.0) (604.0,372.0) (604.0,372.0) (605.0,371.0) (605.0,371.0) (606.0,370.0) (606.0,370.0) (607.0,369.0) (608,367.67) ------------------------------------------------------------------------ (608.00,368.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (607.0,369.0) (609,368) (609,368) (609,368) (609,368) (609,368) (609,368) (609,368) (609,368) (609,368) (609,368) (609.0,368.0) (610.0,367.0) (610.0,367.0) (611.0,366.0) (611.0,366.0) (612.0,365.0) (612.0,365.0) (613.0,364.0) (613.0,364.0) (614.0,363.0) (614.0,363.0) (615.0,362.0) (615.0,362.0) (616.0,361.0) (616.0,361.0) (617.0,360.0) (617.0,360.0) (618.0,359.0) (618.0,359.0) (619.0,358.0) (620,356.67) ------------------------------------------------------------------------ (620.00,357.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (619.0,358.0) (621,357) (621,357) (621,357) (621,357) (621,357) (621,357) (621,357) (621,357) (621,357) (621,357) (621,357) (621,357) (621.0,357.0) (622.0,356.0) (622.0,356.0) (623.0,355.0) (623.0,355.0) (624.0,354.0) (624.0,354.0) (625.0,353.0) (625.0,353.0) (626.0,352.0) (626.0,352.0) (627.0,351.0) (627.0,351.0) (628.0,350.0) (628.0,350.0) (629.0,349.0) (629.0,349.0) (630.0,348.0) (630.0,348.0) (631.0,347.0) (631.0,347.0) (632.0,346.0) (632.0,346.0) (633.0,345.0) (633.0,345.0) (634.0,344.0) (634.0,344.0) (635.0,343.0) (635.0,343.0) (636.0,342.0) (637,340.67) ------------------------------------------------------------------------ (637.00,341.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (636.0,342.0) (638,341) (638,341) (638,341) (638,341) (638,341) (638,341) (638,341) (638,341) (638,341) (638,341) (638,341) (638,341) (638,341) (638,341) (638,341) (638,341) (638.0,341.0) (639.0,340.0) (639.0,340.0) (640.0,339.0) (640.0,339.0) (641.0,338.0) (641.0,338.0) (642.0,337.0) (642.0,337.0) (643.0,336.0) (643.0,336.0) (644.0,335.0) (644.0,335.0) (645.0,334.0) (645.0,334.0) (646.0,333.0) (646.0,333.0) (647.0,332.0) (647.0,332.0) (648.0,331.0) (648.0,331.0) (649.0,330.0) (649.0,330.0) (650.0,329.0) (650.0,329.0) (651.0,328.0) (651.0,328.0) (652.0,327.0) (652.0,327.0) (653.0,326.0) (653.0,326.0) (654.0,325.0) (654.0,325.0) (655.0,324.0) (655.0,324.0) (656.0,323.0) (656.0,323.0) (657.0,322.0) (657.0,322.0) (658.0,321.0) (658.0,321.0) (659.0,320.0) (659.0,320.0) (660.0,319.0) (660.0,319.0) (661.0,318.0) (661.0,318.0) (662.0,317.0) (662.0,317.0) (663.0,316.0) (663.0,316.0) (664.0,315.0) (664.0,315.0) (665.0,314.0) (665.0,314.0) (666.0,313.0) (666.0,313.0) (667.0,312.0) (667.0,312.0) (668.0,311.0) (668.0,311.0) (669.0,310.0) (669.0,310.0) (670.0,309.0) (670.0,309.0) (671.0,308.0) (671.0,308.0) (672.0,307.0) (672.0,307.0) (673.0,306.0) (673.0,306.0) (674.0,305.0) (674.0,305.0) ------------------------------------------------------------------------ (676.00,302.92)(0.544,-0.496)[41]{} ------------------------------------------------------------------------ (676.00,303.17)(22.887,-22.000)[2]{} ------------------------------------------------------------------------ (700.00,280.92)(0.607,-0.494)[25]{} ------------------------------------------------------------------------ (700.00,281.17)(15.784,-14.000)[2]{} ------------------------------------------------------------------------ (717.00,266.93)(0.728,-0.489)[15]{} ------------------------------------------------------------------------ (717.00,267.17)(11.593,-9.000)[2]{} ------------------------------------------------------------------------ (730.00,257.93)(0.798,-0.485)[11]{} ------------------------------------------------------------------------ (730.00,258.17)(9.488,-7.000)[2]{} ------------------------------------------------------------------------ (741.00,250.93)(0.852,-0.482)[9]{} ------------------------------------------------------------------------ (741.00,251.17)(8.409,-6.000)[2]{} ------------------------------------------------------------------------ (751.00,244.93)(0.821,-0.477)[7]{} ------------------------------------------------------------------------ (751.00,245.17)(6.464,-5.000)[2]{} ------------------------------------------------------------------------ (759.00,239.94)(0.920,-0.468)[5]{} ------------------------------------------------------------------------ (759.00,240.17)(5.340,-4.000)[2]{} ------------------------------------------------------------------------ (766.00,235.94)(0.774,-0.468)[5]{} ------------------------------------------------------------------------ (766.00,236.17)(4.547,-4.000)[2]{} ------------------------------------------------------------------------ (772.00,231.94)(0.774,-0.468)[5]{} ------------------------------------------------------------------------ (772.00,232.17)(4.547,-4.000)[2]{} ------------------------------------------------------------------------ (778.00,227.95)(0.909,-0.447)[3]{} ------------------------------------------------------------------------ (778.00,228.17)(3.409,-3.000)[2]{} ------------------------------------------------------------------------ (783.00,224.95)(0.909,-0.447)[3]{} ------------------------------------------------------------------------ (783.00,225.17)(3.409,-3.000)[2]{} ------------------------------------------------------------------------ (788.00,221.95)(0.685,-0.447)[3]{} ------------------------------------------------------------------------ (788.00,222.17)(2.685,-3.000)[2]{} ------------------------------------------------------------------------ (792.00,218.95)(0.685,-0.447)[3]{} ------------------------------------------------------------------------ (792.00,219.17)(2.685,-3.000)[2]{} ------------------------------------------------------------------------ (796.00,215.95)(0.685,-0.447)[3]{} ------------------------------------------------------------------------ (796.00,216.17)(2.685,-3.000)[2]{} ------------------------------------------------------------------------ (800,212.17) ------------------------------------------------------------------------ (800.00,213.17)(2.132,-2.000)[2]{} ------------------------------------------------------------------------ (804.00,210.95)(0.462,-0.447)[3]{} ------------------------------------------------------------------------ (804.00,211.17)(1.962,-3.000)[2]{} ------------------------------------------------------------------------ (807,207.17) ------------------------------------------------------------------------ (807.00,208.17)(1.547,-2.000)[2]{} ------------------------------------------------------------------------ (810,205.17) ------------------------------------------------------------------------ (810.00,206.17)(2.132,-2.000)[2]{} ------------------------------------------------------------------------ (814,203.17) ------------------------------------------------------------------------ (814.00,204.17)(1.000,-2.000)[2]{} ------------------------------------------------------------------------ (816,201.17) ------------------------------------------------------------------------ (816.00,202.17)(1.547,-2.000)[2]{} ------------------------------------------------------------------------ (819,199.17) ------------------------------------------------------------------------ (819.00,200.17)(1.547,-2.000)[2]{} ------------------------------------------------------------------------ (822,197.17) ------------------------------------------------------------------------ (822.00,198.17)(1.000,-2.000)[2]{} ------------------------------------------------------------------------ (824,195.17) ------------------------------------------------------------------------ (824.00,196.17)(1.547,-2.000)[2]{} ------------------------------------------------------------------------ (827,193.67) ------------------------------------------------------------------------ (827.00,194.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (829,192.17) ------------------------------------------------------------------------ (829.00,193.17)(1.547,-2.000)[2]{} ------------------------------------------------------------------------ (832,190.17) ------------------------------------------------------------------------ (832.00,191.17)(1.000,-2.000)[2]{} ------------------------------------------------------------------------ (834,188.67) ------------------------------------------------------------------------ (834.00,189.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (836,187.17) ------------------------------------------------------------------------ (836.00,188.17)(1.000,-2.000)[2]{} ------------------------------------------------------------------------ (838,185.67) ------------------------------------------------------------------------ (838.00,186.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (840,184.67) ------------------------------------------------------------------------ (840.00,185.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (842,183.17) ------------------------------------------------------------------------ (842.00,184.17)(1.000,-2.000)[2]{} ------------------------------------------------------------------------ (844,181.67) ------------------------------------------------------------------------ (844.00,182.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (845,180.67) ------------------------------------------------------------------------ (845.00,181.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (847,179.67) ------------------------------------------------------------------------ (847.00,180.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (848.67,178) ------------------------------------------------------------------------ (848.17,179.00)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (850,176.67) ------------------------------------------------------------------------ (850.00,177.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (852,175.67) ------------------------------------------------------------------------ (852.00,176.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (854,174.67) ------------------------------------------------------------------------ (854.00,175.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (855,173.67) ------------------------------------------------------------------------ (855.00,174.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (857,172.67) ------------------------------------------------------------------------ (857.00,173.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (858,171.67) ------------------------------------------------------------------------ (858.00,172.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (859,170.67) ------------------------------------------------------------------------ (859.00,171.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (861,169.67) ------------------------------------------------------------------------ (861.00,170.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (862,168.67) ------------------------------------------------------------------------ (862.00,169.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (863,167.67) ------------------------------------------------------------------------ (863.00,168.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (865,166.67) ------------------------------------------------------------------------ (865.00,167.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (866,165.67) ------------------------------------------------------------------------ (866.00,166.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (867,164.67) ------------------------------------------------------------------------ (867.00,165.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (868,163.67) ------------------------------------------------------------------------ (868.00,164.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (676.0,304.0) (871,162.67) ------------------------------------------------------------------------ (871.00,163.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (872,161.67) ------------------------------------------------------------------------ (872.00,162.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (873,160.67) ------------------------------------------------------------------------ (873.00,161.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (874,159.67) ------------------------------------------------------------------------ (874.00,160.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (870.0,164.0) (876,158.67) ------------------------------------------------------------------------ (876.00,159.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (877,157.67) ------------------------------------------------------------------------ (877.00,158.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (878,156.67) ------------------------------------------------------------------------ (878.00,157.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (875.0,160.0) (880,155.67) ------------------------------------------------------------------------ (880.00,156.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (881,154.67) ------------------------------------------------------------------------ (881.00,155.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (879.0,157.0) (883,153.67) ------------------------------------------------------------------------ (883.00,154.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (884,152.67) ------------------------------------------------------------------------ (884.00,153.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (882.0,155.0) (886,151.67) ------------------------------------------------------------------------ (886.00,152.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (887,150.67) ------------------------------------------------------------------------ (887.00,151.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (885.0,153.0) (888.0,151.0) (889,148.67) ------------------------------------------------------------------------ (889.00,149.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (889.0,150.0) (891,147.67) ------------------------------------------------------------------------ (891.00,148.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (890.0,149.0) (893,146.67) ------------------------------------------------------------------------ (893.00,147.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (892.0,148.0) (894.0,146.0) (895,144.67) ------------------------------------------------------------------------ (895.00,145.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (894.0,146.0) (896.0,145.0) (897.0,144.0) (898,142.67) ------------------------------------------------------------------------ (898.00,143.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (897.0,144.0) (899.0,143.0) (900.0,142.0) (901,140.67) ------------------------------------------------------------------------ (901.00,141.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (900.0,142.0) (902,141) (902,139.67) ------------------------------------------------------------------------ (902.00,140.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (903.0,140.0) (904.0,139.0) (905,137.67) ------------------------------------------------------------------------ (905.00,138.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (904.0,139.0) (906,138) (906,136.67) ------------------------------------------------------------------------ (906.00,137.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (907.0,137.0) (908.0,136.0) (908.0,136.0) (909.0,135.0) (909.0,135.0) (220,621) (220.00,619.93)(4.606,-0.477)[7]{} ------------------------------------------------------------------------ (220.00,620.17)(34.819,-5.000)[2]{} ------------------------------------------------------------------------ (262.00,614.94)(3.406,-0.468)[5]{} ------------------------------------------------------------------------ (262.00,615.17)(18.811,-4.000)[2]{} ------------------------------------------------------------------------ (286.00,610.95)(3.588,-0.447)[3]{} ------------------------------------------------------------------------ (286.00,611.17)(12.088,-3.000)[2]{} ------------------------------------------------------------------------ (303.00,607.95)(2.695,-0.447)[3]{} ------------------------------------------------------------------------ (303.00,608.17)(9.195,-3.000)[2]{} ------------------------------------------------------------------------ (316.00,604.95)(2.248,-0.447)[3]{} ------------------------------------------------------------------------ (316.00,605.17)(7.748,-3.000)[2]{} ------------------------------------------------------------------------ (327,601.17) ------------------------------------------------------------------------ (327.00,602.17)(5.641,-2.000)[2]{} ------------------------------------------------------------------------ (337,599.17) ------------------------------------------------------------------------ (337.00,600.17)(4.472,-2.000)[2]{} ------------------------------------------------------------------------ (345,597.17) ------------------------------------------------------------------------ (345.00,598.17)(3.887,-2.000)[2]{} ------------------------------------------------------------------------ (352,595.17) ------------------------------------------------------------------------ (352.00,596.17)(3.302,-2.000)[2]{} ------------------------------------------------------------------------ (358,593.17) ------------------------------------------------------------------------ (358.00,594.17)(3.302,-2.000)[2]{} ------------------------------------------------------------------------ (364,591.67) ------------------------------------------------------------------------ (364.00,592.17)(2.500,-1.000)[2]{} ------------------------------------------------------------------------ (369,590.17) ------------------------------------------------------------------------ (369.00,591.17)(2.717,-2.000)[2]{} ------------------------------------------------------------------------ (374,588.67) ------------------------------------------------------------------------ (374.00,589.17)(2.000,-1.000)[2]{} ------------------------------------------------------------------------ (378,587.17) ------------------------------------------------------------------------ (378.00,588.17)(2.132,-2.000)[2]{} ------------------------------------------------------------------------ (382,585.67) ------------------------------------------------------------------------ (382.00,586.17)(2.000,-1.000)[2]{} ------------------------------------------------------------------------ (386,584.17) ------------------------------------------------------------------------ (386.00,585.17)(2.132,-2.000)[2]{} ------------------------------------------------------------------------ (390,582.67) ------------------------------------------------------------------------ (390.00,583.17)(1.500,-1.000)[2]{} ------------------------------------------------------------------------ (393,581.67) ------------------------------------------------------------------------ (393.00,582.17)(1.500,-1.000)[2]{} ------------------------------------------------------------------------ (396,580.17) ------------------------------------------------------------------------ (396.00,581.17)(2.132,-2.000)[2]{} ------------------------------------------------------------------------ (400,578.67) ------------------------------------------------------------------------ (400.00,579.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (402,577.67) ------------------------------------------------------------------------ (402.00,578.17)(1.500,-1.000)[2]{} ------------------------------------------------------------------------ (405,576.67) ------------------------------------------------------------------------ (405.00,577.17)(1.500,-1.000)[2]{} ------------------------------------------------------------------------ (408,575.67) ------------------------------------------------------------------------ (408.00,576.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (410,574.17) ------------------------------------------------------------------------ (410.00,575.17)(1.547,-2.000)[2]{} ------------------------------------------------------------------------ (413,572.67) ------------------------------------------------------------------------ (413.00,573.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (415,571.67) ------------------------------------------------------------------------ (415.00,572.17)(1.500,-1.000)[2]{} ------------------------------------------------------------------------ (418,570.67) ------------------------------------------------------------------------ (418.00,571.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (420,569.67) ------------------------------------------------------------------------ (420.00,570.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (422,568.67) ------------------------------------------------------------------------ (422.00,569.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (424,567.67) ------------------------------------------------------------------------ (424.00,568.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (426,566.67) ------------------------------------------------------------------------ (426.00,567.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (428,565.67) ------------------------------------------------------------------------ (428.00,566.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (430,564.67) ------------------------------------------------------------------------ (430.00,565.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (433,563.67) ------------------------------------------------------------------------ (433.00,564.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (435,562.67) ------------------------------------------------------------------------ (435.00,563.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (436,561.67) ------------------------------------------------------------------------ (436.00,562.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (438,560.67) ------------------------------------------------------------------------ (438.00,561.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (440,559.67) ------------------------------------------------------------------------ (440.00,560.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (441,558.67) ------------------------------------------------------------------------ (441.00,559.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (431.0,565.0) ------------------------------------------------------------------------ (444,557.67) ------------------------------------------------------------------------ (444.00,558.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (445,556.67) ------------------------------------------------------------------------ (445.00,557.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (447,555.67) ------------------------------------------------------------------------ (447.00,556.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (448,554.67) ------------------------------------------------------------------------ (448.00,555.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (443.0,559.0) (451,553.67) ------------------------------------------------------------------------ (451.00,554.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (452,552.67) ------------------------------------------------------------------------ (452.00,553.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (449.0,555.0) ------------------------------------------------------------------------ (454,551.67) ------------------------------------------------------------------------ (454.00,552.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (456,550.67) ------------------------------------------------------------------------ (456.00,551.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (457,549.67) ------------------------------------------------------------------------ (457.00,550.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (453.0,553.0) (459,548.67) ------------------------------------------------------------------------ (459.00,549.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (460,547.67) ------------------------------------------------------------------------ (460.00,548.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (458.0,550.0) (462,546.67) ------------------------------------------------------------------------ (462.00,547.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (461.0,548.0) (464,545.67) ------------------------------------------------------------------------ (464.00,546.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (465,544.67) ------------------------------------------------------------------------ (465.00,545.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (463.0,547.0) (467,543.67) ------------------------------------------------------------------------ (467.00,544.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (468,542.67) ------------------------------------------------------------------------ (468.00,543.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (466.0,545.0) (470,541.67) ------------------------------------------------------------------------ (470.00,542.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (469.0,543.0) (472,540.67) ------------------------------------------------------------------------ (472.00,541.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (471.0,542.0) (474,539.67) ------------------------------------------------------------------------ (474.00,540.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (473.0,541.0) (475.0,539.0) (476,537.67) ------------------------------------------------------------------------ (476.00,538.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (475.0,539.0) (478,536.67) ------------------------------------------------------------------------ (478.00,537.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (477.0,538.0) (479.0,537.0) (480.0,536.0) (481,534.67) ------------------------------------------------------------------------ (481.00,535.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (480.0,536.0) (482.0,535.0) (483.0,534.0) (484,532.67) ------------------------------------------------------------------------ (484.00,533.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (483.0,534.0) (485.0,533.0) (486.0,532.0) (487,530.67) ------------------------------------------------------------------------ (487.00,531.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (486.0,532.0) (488,531) (488,529.67) ------------------------------------------------------------------------ (488.00,530.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (489.0,530.0) (490.0,529.0) (490.0,529.0) ------------------------------------------------------------------------ (492.0,528.0) (493,526.67) ------------------------------------------------------------------------ (493.00,527.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (492.0,528.0) (494,527) (494,525.67) ------------------------------------------------------------------------ (494.00,526.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (495,526) (495,524.67) ------------------------------------------------------------------------ (495.00,525.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (497,523.67) ------------------------------------------------------------------------ (497.00,524.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (496.0,525.0) (498,524) (498,522.67) ------------------------------------------------------------------------ (498.00,523.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (499,523) (500,521.67) ------------------------------------------------------------------------ (500.00,522.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (499.0,523.0) (501,522) (501,520.67) ------------------------------------------------------------------------ (501.00,521.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (502,521) (502.0,521.0) (503.0,520.0) (503.0,520.0) (504.0,519.0) (505,517.67) ------------------------------------------------------------------------ (505.00,518.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (504.0,519.0) (506,518) (506.0,518.0) (507.0,517.0) (507.0,517.0) (508.0,516.0) (509,514.67) ------------------------------------------------------------------------ (509.00,515.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (508.0,516.0) (510,515) (510.0,515.0) (511.0,514.0) (512,512.67) ------------------------------------------------------------------------ (512.00,513.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (511.0,514.0) (513,513) (513.0,513.0) (514.0,512.0) (514.0,512.0) (515.0,511.0) (516,509.67) ------------------------------------------------------------------------ (516.00,510.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (515.0,511.0) (517,510) (517,510) (517,508.67) ------------------------------------------------------------------------ (517.00,509.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (518,509) (518.0,509.0) (519.0,508.0) (519.0,508.0) (520.0,507.0) (521,505.67) ------------------------------------------------------------------------ (521.00,506.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (520.0,507.0) (522,506) (522.0,506.0) (523.0,505.0) (524,503.67) ------------------------------------------------------------------------ (524.00,504.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (523.0,505.0) (525,504) (525,504) (525,502.67) ------------------------------------------------------------------------ (525.00,503.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (526,503) (526,503) (526,501.67) ------------------------------------------------------------------------ (526.00,502.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (527,502) (527,502) (527.0,502.0) (528.0,501.0) (529,499.67) ------------------------------------------------------------------------ (529.00,500.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (528.0,501.0) (530,500) (530,500) (530,498.67) ------------------------------------------------------------------------ (530.00,499.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (531,499) (531,499) (531.0,499.0) (532.0,498.0) (532.0,498.0) (533.0,497.0) (533.0,497.0) (534.0,496.0) (535,494.67) ------------------------------------------------------------------------ (535.00,495.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (534.0,496.0) (536,495) (536,495) (536.0,495.0) (537.0,494.0) (537.0,494.0) (538.0,493.0) (538.0,493.0) (539.0,492.0) (540,490.67) ------------------------------------------------------------------------ (540.00,491.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (539.0,492.0) (541,491) (541,491) (541.0,491.0) (542.0,490.0) (542.0,490.0) (543.0,489.0) (543.0,489.0) (544.0,488.0) (544.0,488.0) (545.0,487.0) (546,485.67) ------------------------------------------------------------------------ (546.00,486.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (545.0,487.0) (547,486) (547,486) (547,486) (547.0,486.0) (548.0,485.0) (548.0,485.0) (549.0,484.0) (549.0,484.0) (550.0,483.0) (551,481.67) ------------------------------------------------------------------------ (551.00,482.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (550.0,483.0) (552,482) (552,482) (552,482) (552.0,482.0) (553.0,481.0) (553.0,481.0) (554.0,480.0) (554.0,480.0) (555.0,479.0) (555.0,479.0) (556.0,478.0) (556.0,478.0) (557.0,477.0) (558,475.67) ------------------------------------------------------------------------ (558.00,476.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (557.0,477.0) (559,476) (559,476) (559,476) (559,476) (559.0,476.0) (560.0,475.0) (560.0,475.0) (561.0,474.0) (561.0,474.0) (562.0,473.0) (562.0,473.0) (563.0,472.0) (564,470.67) ------------------------------------------------------------------------ (564.00,471.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (563.0,472.0) (565,471) (565,471) (565,471) (565,471) (565.0,471.0) (566.0,470.0) (566.0,470.0) (567.0,469.0) (567.0,469.0) (568.0,468.0) (568.0,468.0) (569.0,467.0) (569.0,467.0) (570.0,466.0) (570.0,466.0) (571.0,465.0) (572,463.67) ------------------------------------------------------------------------ (572.00,464.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (571.0,465.0) (573,464) (573,464) (573,464) (573,464) (573,464) (573.0,464.0) (574.0,463.0) (574.0,463.0) (575.0,462.0) (575.0,462.0) (576.0,461.0) (576.0,461.0) (577.0,460.0) (577.0,460.0) (578.0,459.0) (579,457.67) ------------------------------------------------------------------------ (579.00,458.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (578.0,459.0) (580,458) (580,458) (580,458) (580,458) (580,458) (580,458) (580.0,458.0) (581.0,457.0) (581.0,457.0) (582.0,456.0) (582.0,456.0) (583.0,455.0) (583.0,455.0) (584.0,454.0) (584.0,454.0) (585.0,453.0) (585.0,453.0) (586.0,452.0) (587,450.67) ------------------------------------------------------------------------ (587.00,451.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (586.0,452.0) (588,451) (588,451) (588,451) (588,451) (588,451) (588,451) (588,451) (588.0,451.0) (589.0,450.0) (589.0,450.0) (590.0,449.0) (590.0,449.0) (591.0,448.0) (591.0,448.0) (592.0,447.0) (592.0,447.0) (593.0,446.0) (594,444.67) ------------------------------------------------------------------------ (594.00,445.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (593.0,446.0) (595,445) (595,445) (595,445) (595,445) (595,445) (595,445) (595,445) (595,445) (595.0,445.0) (596.0,444.0) (596.0,444.0) (597.0,443.0) (597.0,443.0) (598.0,442.0) (598.0,442.0) (599.0,441.0) (599.0,441.0) (600.0,440.0) (601,438.67) ------------------------------------------------------------------------ (601.00,439.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (600.0,440.0) (602,439) (602,439) (602,439) (602,439) (602,439) (602,439) (602,439) (602,439) (602,439) (602.0,439.0) (603.0,438.0) (603.0,438.0) (604.0,437.0) (604.0,437.0) (605.0,436.0) (605.0,436.0) (606.0,435.0) (606.0,435.0) (607.0,434.0) (607.0,434.0) ------------------------------------------------------------------------ (609.0,433.0) (609.0,433.0) (610.0,432.0) (610.0,432.0) (611.0,431.0) (611.0,431.0) (612.0,430.0) (612.0,430.0) (613.0,429.0) (613.0,429.0) ------------------------------------------------------------------------ (615.0,428.0) (615.0,428.0) (616.0,427.0) (616.0,427.0) (617.0,426.0) (617.0,426.0) (618.0,425.0) (619,423.67) ------------------------------------------------------------------------ (619.00,424.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (618.0,425.0) (620,424) (620,424) (620,424) (620,424) (620,424) (620,424) (620,424) (620,424) (620,424) (620,424) (620,424) (620,424) (620.0,424.0) (621.0,423.0) (621.0,423.0) (622.0,422.0) (622.0,422.0) (623.0,421.0) (623.0,421.0) ------------------------------------------------------------------------ (625.0,420.0) (625.0,420.0) (626.0,419.0) (626.0,419.0) (627.0,418.0) (627.0,418.0) (628.0,417.0) (628.0,417.0) ------------------------------------------------------------------------ (630.0,416.0) (630.0,416.0) (631.0,415.0) (631.0,415.0) (632.0,414.0) (632.0,414.0) ------------------------------------------------------------------------ (634.0,413.0) (634.0,413.0) (635.0,412.0) (635.0,412.0) (636.0,411.0) (636.0,411.0) ------------------------------------------------------------------------ (638.0,410.0) (638.0,410.0) (639.0,409.0) (639.0,409.0) (640.0,408.0) (640.0,408.0) ------------------------------------------------------------------------ (642.0,407.0) (642.0,407.0) (643.0,406.0) (643.0,406.0) ------------------------------------------------------------------------ (645.0,405.0) (645.0,405.0) (646.0,404.0) (646.0,404.0) (647.0,403.0) (647.0,403.0) ------------------------------------------------------------------------ (649.0,402.0) (649.0,402.0) (650.0,401.0) (650.0,401.0) ------------------------------------------------------------------------ (652.0,400.0) (652.0,400.0) (653.0,399.0) (653.0,399.0) ------------------------------------------------------------------------ (655.0,398.0) (655.0,398.0) (656.0,397.0) (656.0,397.0) ------------------------------------------------------------------------ (658.0,396.0) (658.0,396.0) (659.0,395.0) (659.0,395.0) ------------------------------------------------------------------------ (661.0,394.0) (661.0,394.0) (662.0,393.0) (662.0,393.0) ------------------------------------------------------------------------ (664.0,392.0) (664.0,392.0) (665.0,391.0) (665.0,391.0) ------------------------------------------------------------------------ (667.0,390.0) (667.0,390.0) (668.0,389.0) (668.0,389.0) ------------------------------------------------------------------------ (670.0,388.0) (670.0,388.0) (671.0,387.0) (671.0,387.0) ------------------------------------------------------------------------ (673.0,386.0) (673.0,386.0) (674.0,385.0) (676.00,383.92)(0.708,-0.495)[31]{} ------------------------------------------------------------------------ (676.00,384.17)(22.620,-17.000)[2]{} ------------------------------------------------------------------------ (700.00,366.92)(0.712,-0.492)[21]{} ------------------------------------------------------------------------ (700.00,367.17)(15.616,-12.000)[2]{} ------------------------------------------------------------------------ (717.00,354.93)(0.728,-0.489)[15]{} ------------------------------------------------------------------------ (717.00,355.17)(11.593,-9.000)[2]{} ------------------------------------------------------------------------ (730.00,345.93)(0.692,-0.488)[13]{} ------------------------------------------------------------------------ (730.00,346.17)(9.651,-8.000)[2]{} ------------------------------------------------------------------------ (741.00,337.93)(0.852,-0.482)[9]{} ------------------------------------------------------------------------ (741.00,338.17)(8.409,-6.000)[2]{} ------------------------------------------------------------------------ (751.00,331.93)(0.671,-0.482)[9]{} ------------------------------------------------------------------------ (751.00,332.17)(6.685,-6.000)[2]{} ------------------------------------------------------------------------ (759.00,325.93)(0.710,-0.477)[7]{} ------------------------------------------------------------------------ (759.00,326.17)(5.630,-5.000)[2]{} ------------------------------------------------------------------------ (766.00,320.93)(0.599,-0.477)[7]{} ------------------------------------------------------------------------ (766.00,321.17)(4.796,-5.000)[2]{} ------------------------------------------------------------------------ (772.00,315.94)(0.774,-0.468)[5]{} ------------------------------------------------------------------------ (772.00,316.17)(4.547,-4.000)[2]{} ------------------------------------------------------------------------ (778.00,311.94)(0.627,-0.468)[5]{} ------------------------------------------------------------------------ (778.00,312.17)(3.755,-4.000)[2]{} ------------------------------------------------------------------------ (783.00,307.95)(0.909,-0.447)[3]{} ------------------------------------------------------------------------ (783.00,308.17)(3.409,-3.000)[2]{} ------------------------------------------------------------------------ (788.00,304.94)(0.481,-0.468)[5]{} ------------------------------------------------------------------------ (788.00,305.17)(2.962,-4.000)[2]{} ------------------------------------------------------------------------ (792.00,300.95)(0.685,-0.447)[3]{} ------------------------------------------------------------------------ (792.00,301.17)(2.685,-3.000)[2]{} ------------------------------------------------------------------------ (796,297.17) ------------------------------------------------------------------------ (796.00,298.17)(2.132,-2.000)[2]{} ------------------------------------------------------------------------ (800.00,295.95)(0.685,-0.447)[3]{} ------------------------------------------------------------------------ (800.00,296.17)(2.685,-3.000)[2]{} ------------------------------------------------------------------------ (804.00,292.95)(0.462,-0.447)[3]{} ------------------------------------------------------------------------ (804.00,293.17)(1.962,-3.000)[2]{} ------------------------------------------------------------------------ (807,289.17) ------------------------------------------------------------------------ (807.00,290.17)(1.547,-2.000)[2]{} ------------------------------------------------------------------------ (810,287.17) ------------------------------------------------------------------------ (810.00,288.17)(2.132,-2.000)[2]{} ------------------------------------------------------------------------ (814,285.17) ------------------------------------------------------------------------ (814.00,286.17)(1.000,-2.000)[2]{} ------------------------------------------------------------------------ (816,283.17) ------------------------------------------------------------------------ (816.00,284.17)(1.547,-2.000)[2]{} ------------------------------------------------------------------------ (819,281.17) ------------------------------------------------------------------------ (819.00,282.17)(1.547,-2.000)[2]{} ------------------------------------------------------------------------ (822,279.17) ------------------------------------------------------------------------ (822.00,280.17)(1.000,-2.000)[2]{} ------------------------------------------------------------------------ (824,277.17) ------------------------------------------------------------------------ (824.00,278.17)(1.547,-2.000)[2]{} ------------------------------------------------------------------------ (827,275.17) ------------------------------------------------------------------------ (827.00,276.17)(1.000,-2.000)[2]{} ------------------------------------------------------------------------ (829,273.67) ------------------------------------------------------------------------ (829.00,274.17)(1.500,-1.000)[2]{} ------------------------------------------------------------------------ (832,272.17) ------------------------------------------------------------------------ (832.00,273.17)(1.000,-2.000)[2]{} ------------------------------------------------------------------------ (834,270.17) ------------------------------------------------------------------------ (834.00,271.17)(1.000,-2.000)[2]{} ------------------------------------------------------------------------ (836,268.67) ------------------------------------------------------------------------ (836.00,269.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (838,267.17) ------------------------------------------------------------------------ (838.00,268.17)(1.000,-2.000)[2]{} ------------------------------------------------------------------------ (840,265.67) ------------------------------------------------------------------------ (840.00,266.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (842,264.67) ------------------------------------------------------------------------ (842.00,265.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (843.67,263) ------------------------------------------------------------------------ (843.17,264.00)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (845,261.67) ------------------------------------------------------------------------ (845.00,262.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (847,260.67) ------------------------------------------------------------------------ (847.00,261.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (849,259.67) ------------------------------------------------------------------------ (849.00,260.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (850,258.67) ------------------------------------------------------------------------ (850.00,259.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (852,257.17) ------------------------------------------------------------------------ (852.00,258.17)(1.000,-2.000)[2]{} ------------------------------------------------------------------------ (854,255.67) ------------------------------------------------------------------------ (854.00,256.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (855,254.67) ------------------------------------------------------------------------ (855.00,255.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (857,253.67) ------------------------------------------------------------------------ (857.00,254.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (858,252.67) ------------------------------------------------------------------------ (858.00,253.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (859,251.67) ------------------------------------------------------------------------ (859.00,252.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (861,250.67) ------------------------------------------------------------------------ (861.00,251.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (862,249.67) ------------------------------------------------------------------------ (862.00,250.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (863,248.67) ------------------------------------------------------------------------ (863.00,249.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (865,247.67) ------------------------------------------------------------------------ (865.00,248.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (866,246.67) ------------------------------------------------------------------------ (866.00,247.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (867,245.67) ------------------------------------------------------------------------ (867.00,246.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (674.0,385.0) ------------------------------------------------------------------------ (870,244.67) ------------------------------------------------------------------------ (870.00,245.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (871,243.67) ------------------------------------------------------------------------ (871.00,244.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (872,242.67) ------------------------------------------------------------------------ (872.00,243.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (873,241.67) ------------------------------------------------------------------------ (873.00,242.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (874,240.67) ------------------------------------------------------------------------ (874.00,241.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (868.0,246.0) ------------------------------------------------------------------------ (876,239.67) ------------------------------------------------------------------------ (876.00,240.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (877,238.67) ------------------------------------------------------------------------ (877.00,239.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (878,237.67) ------------------------------------------------------------------------ (878.00,238.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (875.0,241.0) (880,236.67) ------------------------------------------------------------------------ (880.00,237.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (881,235.67) ------------------------------------------------------------------------ (881.00,236.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (879.0,238.0) (883,234.67) ------------------------------------------------------------------------ (883.00,235.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (884,233.67) ------------------------------------------------------------------------ (884.00,234.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (882.0,236.0) (886,232.67) ------------------------------------------------------------------------ (886.00,233.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (887,231.67) ------------------------------------------------------------------------ (887.00,232.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (885.0,234.0) (888.0,232.0) (889,229.67) ------------------------------------------------------------------------ (889.00,230.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (889.0,231.0) (891,228.67) ------------------------------------------------------------------------ (891.00,229.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (890.0,230.0) (893,227.67) ------------------------------------------------------------------------ (893.00,228.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (892.0,229.0) (894.0,227.0) (895,225.67) ------------------------------------------------------------------------ (895.00,226.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (894.0,227.0) (896.0,226.0) (897.0,225.0) (898,223.67) ------------------------------------------------------------------------ (898.00,224.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (897.0,225.0) (899.0,224.0) (900.0,223.0) (901,221.67) ------------------------------------------------------------------------ (901.00,222.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (900.0,223.0) (902,222) (902,220.67) ------------------------------------------------------------------------ (902.00,221.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (903.0,221.0) (904.0,220.0) (905,218.67) ------------------------------------------------------------------------ (905.00,219.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (904.0,220.0) (906,219) (906,217.67) ------------------------------------------------------------------------ (906.00,218.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (907.0,218.0) (908.0,217.0) (908.0,217.0) (909.0,216.0) (909.0,216.0) (220,617) (220.00,615.93)(3.162,-0.485)[11]{} ------------------------------------------------------------------------ (220.00,616.17)(36.811,-7.000)[2]{} ------------------------------------------------------------------------ (262.00,608.93)(2.118,-0.482)[9]{} ------------------------------------------------------------------------ (262.00,609.17)(20.472,-6.000)[2]{} ------------------------------------------------------------------------ (286.00,602.94)(2.382,-0.468)[5]{} ------------------------------------------------------------------------ (286.00,603.17)(13.264,-4.000)[2]{} ------------------------------------------------------------------------ (303.00,598.94)(1.797,-0.468)[5]{} ------------------------------------------------------------------------ (303.00,599.17)(10.094,-4.000)[2]{} ------------------------------------------------------------------------ (316.00,594.95)(2.248,-0.447)[3]{} ------------------------------------------------------------------------ (316.00,595.17)(7.748,-3.000)[2]{} ------------------------------------------------------------------------ (327.00,591.94)(1.358,-0.468)[5]{} ------------------------------------------------------------------------ (327.00,592.17)(7.717,-4.000)[2]{} ------------------------------------------------------------------------ (337.00,587.95)(1.579,-0.447)[3]{} ------------------------------------------------------------------------ (337.00,588.17)(5.579,-3.000)[2]{} ------------------------------------------------------------------------ (345,584.17) ------------------------------------------------------------------------ (345.00,585.17)(3.887,-2.000)[2]{} ------------------------------------------------------------------------ (352.00,582.95)(1.132,-0.447)[3]{} ------------------------------------------------------------------------ (352.00,583.17)(4.132,-3.000)[2]{} ------------------------------------------------------------------------ (358,579.17) ------------------------------------------------------------------------ (358.00,580.17)(3.302,-2.000)[2]{} ------------------------------------------------------------------------ (364.00,577.95)(0.909,-0.447)[3]{} ------------------------------------------------------------------------ (364.00,578.17)(3.409,-3.000)[2]{} ------------------------------------------------------------------------ (369,574.17) ------------------------------------------------------------------------ (369.00,575.17)(2.717,-2.000)[2]{} ------------------------------------------------------------------------ (374,572.17) ------------------------------------------------------------------------ (374.00,573.17)(2.132,-2.000)[2]{} ------------------------------------------------------------------------ (378,570.17) ------------------------------------------------------------------------ (378.00,571.17)(2.132,-2.000)[2]{} ------------------------------------------------------------------------ (382,568.17) ------------------------------------------------------------------------ (382.00,569.17)(2.132,-2.000)[2]{} ------------------------------------------------------------------------ (386,566.17) ------------------------------------------------------------------------ (386.00,567.17)(2.132,-2.000)[2]{} ------------------------------------------------------------------------ (390,564.17) ------------------------------------------------------------------------ (390.00,565.17)(1.547,-2.000)[2]{} ------------------------------------------------------------------------ (393,562.67) ------------------------------------------------------------------------ (393.00,563.17)(1.500,-1.000)[2]{} ------------------------------------------------------------------------ (396,561.17) ------------------------------------------------------------------------ (396.00,562.17)(2.132,-2.000)[2]{} ------------------------------------------------------------------------ (400,559.67) ------------------------------------------------------------------------ (400.00,560.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (402,558.17) ------------------------------------------------------------------------ (402.00,559.17)(1.547,-2.000)[2]{} ------------------------------------------------------------------------ (405,556.67) ------------------------------------------------------------------------ (405.00,557.17)(1.500,-1.000)[2]{} ------------------------------------------------------------------------ (408,555.17) ------------------------------------------------------------------------ (408.00,556.17)(1.000,-2.000)[2]{} ------------------------------------------------------------------------ (410,553.67) ------------------------------------------------------------------------ (410.00,554.17)(1.500,-1.000)[2]{} ------------------------------------------------------------------------ (413,552.17) ------------------------------------------------------------------------ (413.00,553.17)(1.000,-2.000)[2]{} ------------------------------------------------------------------------ (415,550.67) ------------------------------------------------------------------------ (415.00,551.17)(1.500,-1.000)[2]{} ------------------------------------------------------------------------ (418,549.67) ------------------------------------------------------------------------ (418.00,550.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (420,548.67) ------------------------------------------------------------------------ (420.00,549.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (422,547.17) ------------------------------------------------------------------------ (422.00,548.17)(1.000,-2.000)[2]{} ------------------------------------------------------------------------ (424,545.67) ------------------------------------------------------------------------ (424.00,546.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (426,544.67) ------------------------------------------------------------------------ (426.00,545.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (428,543.67) ------------------------------------------------------------------------ (428.00,544.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (430,542.67) ------------------------------------------------------------------------ (430.00,543.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (431,541.67) ------------------------------------------------------------------------ (431.00,542.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (433,540.67) ------------------------------------------------------------------------ (433.00,541.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (435,539.67) ------------------------------------------------------------------------ (435.00,540.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (436,538.67) ------------------------------------------------------------------------ (436.00,539.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (438,537.67) ------------------------------------------------------------------------ (438.00,538.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (440,536.67) ------------------------------------------------------------------------ (440.00,537.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (441,535.67) ------------------------------------------------------------------------ (441.00,536.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (443,534.67) ------------------------------------------------------------------------ (443.00,535.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (444,533.67) ------------------------------------------------------------------------ (444.00,534.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (445,532.67) ------------------------------------------------------------------------ (445.00,533.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (447,531.67) ------------------------------------------------------------------------ (447.00,532.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (448,530.67) ------------------------------------------------------------------------ (448.00,531.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (451,529.67) ------------------------------------------------------------------------ (451.00,530.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (452,528.67) ------------------------------------------------------------------------ (452.00,529.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (453,527.67) ------------------------------------------------------------------------ (453.00,528.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (454,526.67) ------------------------------------------------------------------------ (454.00,527.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (449.0,531.0) ------------------------------------------------------------------------ (457,525.67) ------------------------------------------------------------------------ (457.00,526.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (458,524.67) ------------------------------------------------------------------------ (458.00,525.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (459,523.67) ------------------------------------------------------------------------ (459.00,524.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (456.0,527.0) (461,522.67) ------------------------------------------------------------------------ (461.00,523.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (462,521.67) ------------------------------------------------------------------------ (462.00,522.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (460.0,524.0) (464,520.67) ------------------------------------------------------------------------ (464.00,521.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (465,519.67) ------------------------------------------------------------------------ (465.00,520.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (463.0,522.0) (467,518.67) ------------------------------------------------------------------------ (467.00,519.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (468,517.67) ------------------------------------------------------------------------ (468.00,518.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (466.0,520.0) (470,516.67) ------------------------------------------------------------------------ (470.00,517.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (471,515.67) ------------------------------------------------------------------------ (471.00,516.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (469.0,518.0) (473,514.67) ------------------------------------------------------------------------ (473.00,515.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (472.0,516.0) (474.0,515.0) (475,512.67) ------------------------------------------------------------------------ (475.00,513.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (475.0,514.0) (477,511.67) ------------------------------------------------------------------------ (477.00,512.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (476.0,513.0) (479,510.67) ------------------------------------------------------------------------ (479.00,511.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (478.0,512.0) (480,511) (480,509.67) ------------------------------------------------------------------------ (480.00,510.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (482,508.67) ------------------------------------------------------------------------ (482.00,509.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (481.0,510.0) (483,509) (483,507.67) ------------------------------------------------------------------------ (483.00,508.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (485,506.67) ------------------------------------------------------------------------ (485.00,507.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (484.0,508.0) (486,507) (486,505.67) ------------------------------------------------------------------------ (486.00,506.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (487.0,506.0) (488.0,505.0) (489,503.67) ------------------------------------------------------------------------ (489.00,504.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (488.0,505.0) (490,504) (490,502.67) ------------------------------------------------------------------------ (490.00,503.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (491.0,503.0) (492.0,502.0) (493,500.67) ------------------------------------------------------------------------ (493.00,501.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (492.0,502.0) (494,501) (494,499.67) ------------------------------------------------------------------------ (494.00,500.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (495,500) (496,498.67) ------------------------------------------------------------------------ (496.00,499.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (495.0,500.0) (497,499) (497,497.67) ------------------------------------------------------------------------ (497.00,498.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (498,498) (498.0,498.0) (499.0,497.0) (500,495.67) ------------------------------------------------------------------------ (500.00,496.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (499.0,497.0) (501,496) (501.0,496.0) (502.0,495.0) (502.0,495.0) (503.0,494.0) (504,492.67) ------------------------------------------------------------------------ (504.00,493.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (503.0,494.0) (505,493) (505,491.67) ------------------------------------------------------------------------ (505.00,492.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (506,492) (506.0,492.0) (507.0,491.0) (508,489.67) ------------------------------------------------------------------------ (508.00,490.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (507.0,491.0) (509,490) (509.0,490.0) (510.0,489.0) (511,487.67) ------------------------------------------------------------------------ (511.00,488.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (510.0,489.0) (512,488) (512.0,488.0) (513.0,487.0) (513.0,487.0) (514.0,486.0) (515,484.67) ------------------------------------------------------------------------ (515.00,485.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (514.0,486.0) (516,485) (516.0,485.0) (517.0,484.0) (517.0,484.0) (518.0,483.0) (518.0,483.0) (519.0,482.0) (519.0,482.0) ------------------------------------------------------------------------ (521.0,481.0) (521.0,481.0) (522.0,480.0) (523,478.67) ------------------------------------------------------------------------ (523.00,479.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (522.0,480.0) (524,479) (524.0,479.0) (525.0,478.0) (525.0,478.0) (526.0,477.0) (527,475.67) ------------------------------------------------------------------------ (527.00,476.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (526.0,477.0) (528,476) (528.0,476.0) (529.0,475.0) (529.0,475.0) (530.0,474.0) (531,472.67) ------------------------------------------------------------------------ (531.00,473.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (530.0,474.0) (532,473) (532,473) (532.0,473.0) (533.0,472.0) (533.0,472.0) (534.0,471.0) (534.0,471.0) (535.0,470.0) (535.0,470.0) ------------------------------------------------------------------------ (537.0,469.0) (537.0,469.0) (538.0,468.0) (538.0,468.0) (539.0,467.0) (540,465.67) ------------------------------------------------------------------------ (540.00,466.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (539.0,467.0) (541,466) (541,466) (541.0,466.0) (542.0,465.0) (542.0,465.0) (543.0,464.0) (544,462.67) ------------------------------------------------------------------------ (544.00,463.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (543.0,464.0) (545,463) (545,463) (545,463) (545.0,463.0) (546.0,462.0) (546.0,462.0) (547.0,461.0) (547.0,461.0) (548.0,460.0) (549,458.67) ------------------------------------------------------------------------ (549.00,459.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (548.0,460.0) (550,459) (550,459) (550,459) (550.0,459.0) (551.0,458.0) (551.0,458.0) (552.0,457.0) (552.0,457.0) (553.0,456.0) (553.0,456.0) ------------------------------------------------------------------------ (555.0,455.0) (555.0,455.0) (556.0,454.0) (556.0,454.0) (557.0,453.0) (558,451.67) ------------------------------------------------------------------------ (558.00,452.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (557.0,453.0) (559,452) (559,452) (559,452) (559,452) (559.0,452.0) (560.0,451.0) (560.0,451.0) (561.0,450.0) (561.0,450.0) (562.0,449.0) (563,447.67) ------------------------------------------------------------------------ (563.00,448.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (562.0,449.0) (564,448) (564,448) (564,448) (564,448) (564.0,448.0) (565.0,447.0) (565.0,447.0) (566.0,446.0) (566.0,446.0) (567.0,445.0) (568,443.67) ------------------------------------------------------------------------ (568.00,444.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (567.0,445.0) (569,444) (569,444) (569,444) (569,444) (569.0,444.0) (570.0,443.0) (570.0,443.0) (571.0,442.0) (571.0,442.0) (572.0,441.0) (573,439.67) ------------------------------------------------------------------------ (573.00,440.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (572.0,441.0) (574,440) (574,440) (574,440) (574,440) (574,440) (574.0,440.0) (575.0,439.0) (575.0,439.0) (576.0,438.0) (576.0,438.0) (577.0,437.0) (577.0,437.0) (578.0,436.0) (578.0,436.0) ------------------------------------------------------------------------ (580.0,435.0) (580.0,435.0) (581.0,434.0) (581.0,434.0) (582.0,433.0) (582.0,433.0) (583.0,432.0) (584,430.67) ------------------------------------------------------------------------ (584.00,431.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (583.0,432.0) (585,431) (585,431) (585,431) (585,431) (585,431) (585,431) (585,431) (585.0,431.0) (586.0,430.0) (586.0,430.0) (587.0,429.0) (587.0,429.0) (588.0,428.0) (588.0,428.0) (589.0,427.0) (589.0,427.0) ------------------------------------------------------------------------ (591.0,426.0) (591.0,426.0) (592.0,425.0) (592.0,425.0) (593.0,424.0) (593.0,424.0) (594.0,423.0) (594.0,423.0) (595.0,422.0) (596,420.67) ------------------------------------------------------------------------ (596.00,421.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (595.0,422.0) (597,421) (597,421) (597,421) (597,421) (597,421) (597,421) (597,421) (597,421) (597.0,421.0) (598.0,420.0) (598.0,420.0) (599.0,419.0) (599.0,419.0) (600.0,418.0) (600.0,418.0) (601.0,417.0) (601.0,417.0) (602.0,416.0) (602.0,416.0) ------------------------------------------------------------------------ (604.0,415.0) (604.0,415.0) (605.0,414.0) (605.0,414.0) (606.0,413.0) (606.0,413.0) (607.0,412.0) (607.0,412.0) (608.0,411.0) (609,409.67) ------------------------------------------------------------------------ (609.00,410.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (608.0,411.0) (610,410) (610,410) (610,410) (610,410) (610,410) (610,410) (610,410) (610,410) (610,410) (610,410) (610.0,410.0) (611.0,409.0) (611.0,409.0) (612.0,408.0) (612.0,408.0) (613.0,407.0) (613.0,407.0) (614.0,406.0) (614.0,406.0) (615.0,405.0) (615.0,405.0) ------------------------------------------------------------------------ (617.0,404.0) (617.0,404.0) (618.0,403.0) (618.0,403.0) (619.0,402.0) (619.0,402.0) (620.0,401.0) (620.0,401.0) (621.0,400.0) (621.0,400.0) (622.0,399.0) (623,397.67) ------------------------------------------------------------------------ (623.00,398.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (622.0,399.0) (624,398) (624,398) (624,398) (624,398) (624,398) (624,398) (624,398) (624,398) (624,398) (624,398) (624,398) (624,398) (624,398) (624.0,398.0) (625.0,397.0) (625.0,397.0) (626.0,396.0) (626.0,396.0) (627.0,395.0) (627.0,395.0) (628.0,394.0) (628.0,394.0) (629.0,393.0) (630,391.67) ------------------------------------------------------------------------ (630.00,392.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (629.0,393.0) (631,392) (631,392) (631,392) (631,392) (631,392) (631,392) (631,392) (631,392) (631,392) (631,392) (631,392) (631,392) (631,392) (631,392) (631,392) (631.0,392.0) (632.0,391.0) (632.0,391.0) (633.0,390.0) (634.00,388.92)(0.600,-0.498)[67]{} ------------------------------------------------------------------------ (634.00,389.17)(40.796,-35.000)[2]{} ------------------------------------------------------------------------ (676.00,353.92)(0.668,-0.495)[33]{} ------------------------------------------------------------------------ (676.00,354.17)(22.685,-18.000)[2]{} ------------------------------------------------------------------------ (700.00,335.92)(0.712,-0.492)[21]{} ------------------------------------------------------------------------ (700.00,336.17)(15.616,-12.000)[2]{} ------------------------------------------------------------------------ (717.00,323.93)(0.728,-0.489)[15]{} ------------------------------------------------------------------------ (717.00,324.17)(11.593,-9.000)[2]{} ------------------------------------------------------------------------ (730.00,314.93)(0.692,-0.488)[13]{} ------------------------------------------------------------------------ (730.00,315.17)(9.651,-8.000)[2]{} ------------------------------------------------------------------------ (741.00,306.93)(0.852,-0.482)[9]{} ------------------------------------------------------------------------ (741.00,307.17)(8.409,-6.000)[2]{} ------------------------------------------------------------------------ (751.00,300.93)(0.671,-0.482)[9]{} ------------------------------------------------------------------------ (751.00,301.17)(6.685,-6.000)[2]{} ------------------------------------------------------------------------ (759.00,294.93)(0.710,-0.477)[7]{} ------------------------------------------------------------------------ (759.00,295.17)(5.630,-5.000)[2]{} ------------------------------------------------------------------------ (766.00,289.94)(0.774,-0.468)[5]{} ------------------------------------------------------------------------ (766.00,290.17)(4.547,-4.000)[2]{} ------------------------------------------------------------------------ (772.00,285.94)(0.774,-0.468)[5]{} ------------------------------------------------------------------------ (772.00,286.17)(4.547,-4.000)[2]{} ------------------------------------------------------------------------ (778.00,281.94)(0.627,-0.468)[5]{} ------------------------------------------------------------------------ (778.00,282.17)(3.755,-4.000)[2]{} ------------------------------------------------------------------------ (783.00,277.95)(0.909,-0.447)[3]{} ------------------------------------------------------------------------ (783.00,278.17)(3.409,-3.000)[2]{} ------------------------------------------------------------------------ (788.00,274.94)(0.481,-0.468)[5]{} ------------------------------------------------------------------------ (788.00,275.17)(2.962,-4.000)[2]{} ------------------------------------------------------------------------ (792.00,270.95)(0.685,-0.447)[3]{} ------------------------------------------------------------------------ (792.00,271.17)(2.685,-3.000)[2]{} ------------------------------------------------------------------------ (796,267.17) ------------------------------------------------------------------------ (796.00,268.17)(2.132,-2.000)[2]{} ------------------------------------------------------------------------ (800.00,265.95)(0.685,-0.447)[3]{} ------------------------------------------------------------------------ (800.00,266.17)(2.685,-3.000)[2]{} ------------------------------------------------------------------------ (804,262.17) ------------------------------------------------------------------------ (804.00,263.17)(1.547,-2.000)[2]{} ------------------------------------------------------------------------ (807.00,260.95)(0.462,-0.447)[3]{} ------------------------------------------------------------------------ (807.00,261.17)(1.962,-3.000)[2]{} ------------------------------------------------------------------------ (810,257.17) ------------------------------------------------------------------------ (810.00,258.17)(2.132,-2.000)[2]{} ------------------------------------------------------------------------ (814,255.17) ------------------------------------------------------------------------ (814.00,256.17)(1.000,-2.000)[2]{} ------------------------------------------------------------------------ (816,253.17) ------------------------------------------------------------------------ (816.00,254.17)(1.547,-2.000)[2]{} ------------------------------------------------------------------------ (819,251.17) ------------------------------------------------------------------------ (819.00,252.17)(1.547,-2.000)[2]{} ------------------------------------------------------------------------ (822,249.17) ------------------------------------------------------------------------ (822.00,250.17)(1.000,-2.000)[2]{} ------------------------------------------------------------------------ (824,247.17) ------------------------------------------------------------------------ (824.00,248.17)(1.547,-2.000)[2]{} ------------------------------------------------------------------------ (827,245.67) ------------------------------------------------------------------------ (827.00,246.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (829,244.17) ------------------------------------------------------------------------ (829.00,245.17)(1.547,-2.000)[2]{} ------------------------------------------------------------------------ (832,242.17) ------------------------------------------------------------------------ (832.00,243.17)(1.000,-2.000)[2]{} ------------------------------------------------------------------------ (834,240.67) ------------------------------------------------------------------------ (834.00,241.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (836,239.17) ------------------------------------------------------------------------ (836.00,240.17)(1.000,-2.000)[2]{} ------------------------------------------------------------------------ (838,237.67) ------------------------------------------------------------------------ (838.00,238.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (840,236.17) ------------------------------------------------------------------------ (840.00,237.17)(1.000,-2.000)[2]{} ------------------------------------------------------------------------ (842,234.67) ------------------------------------------------------------------------ (842.00,235.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (844,233.67) ------------------------------------------------------------------------ (844.00,234.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (845,232.67) ------------------------------------------------------------------------ (845.00,233.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (847,231.17) ------------------------------------------------------------------------ (847.00,232.17)(1.000,-2.000)[2]{} ------------------------------------------------------------------------ (849,229.67) ------------------------------------------------------------------------ (849.00,230.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (850,228.67) ------------------------------------------------------------------------ (850.00,229.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (852,227.67) ------------------------------------------------------------------------ (852.00,228.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (854,226.67) ------------------------------------------------------------------------ (854.00,227.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (855,225.67) ------------------------------------------------------------------------ (855.00,226.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (857,224.67) ------------------------------------------------------------------------ (857.00,225.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (857.67,223) ------------------------------------------------------------------------ (857.17,224.00)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (859,221.67) ------------------------------------------------------------------------ (859.00,222.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (861,220.67) ------------------------------------------------------------------------ (861.00,221.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (633.0,390.0) (863,219.67) ------------------------------------------------------------------------ (863.00,220.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (865,218.67) ------------------------------------------------------------------------ (865.00,219.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (866,217.67) ------------------------------------------------------------------------ (866.00,218.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (867,216.67) ------------------------------------------------------------------------ (867.00,217.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (868,215.67) ------------------------------------------------------------------------ (868.00,216.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (870,214.67) ------------------------------------------------------------------------ (870.00,215.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (871,213.67) ------------------------------------------------------------------------ (871.00,214.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (872,212.67) ------------------------------------------------------------------------ (872.00,213.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (862.0,221.0) (874,211.67) ------------------------------------------------------------------------ (874.00,212.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (875,210.67) ------------------------------------------------------------------------ (875.00,211.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (876,209.67) ------------------------------------------------------------------------ (876.00,210.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (873.0,213.0) (878,208.67) ------------------------------------------------------------------------ (878.00,209.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (879,207.67) ------------------------------------------------------------------------ (879.00,208.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (880,206.67) ------------------------------------------------------------------------ (880.00,207.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (877.0,210.0) (882,205.67) ------------------------------------------------------------------------ (882.00,206.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (883,204.67) ------------------------------------------------------------------------ (883.00,205.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (881.0,207.0) (885,203.67) ------------------------------------------------------------------------ (885.00,204.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (886,202.67) ------------------------------------------------------------------------ (886.00,203.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (884.0,205.0) (888,201.67) ------------------------------------------------------------------------ (888.00,202.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (887.0,203.0) (889.0,201.0) (890,199.67) ------------------------------------------------------------------------ (890.00,200.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (889.0,201.0) (892,198.67) ------------------------------------------------------------------------ (892.00,199.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (893,197.67) ------------------------------------------------------------------------ (893.00,198.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (891.0,200.0) (894,198) (894,196.67) ------------------------------------------------------------------------ (894.00,197.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (896,195.67) ------------------------------------------------------------------------ (896.00,196.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (895.0,197.0) (897,196) (897,194.67) ------------------------------------------------------------------------ (897.00,195.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (899,193.67) ------------------------------------------------------------------------ (899.00,194.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (898.0,195.0) (900.0,193.0) (901,191.67) ------------------------------------------------------------------------ (901.00,192.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (900.0,193.0) (902,192) (902,190.67) ------------------------------------------------------------------------ (902.00,191.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (903.0,191.0) (904.0,190.0) (904.0,190.0) ------------------------------------------------------------------------ (906.0,189.0) (907,187.67) ------------------------------------------------------------------------ (907.00,188.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (906.0,189.0) (908,188) (908,186.67) ------------------------------------------------------------------------ (908.00,187.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (909,187) (909,185.67) ------------------------------------------------------------------------ (909.00,186.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (358.00,652.00) (359,651)(16.889,-12.064)[0]{} (366,646)(16.889,-12.064)[0]{} (374.74,639.76) (380,636)(15.759,-13.508)[0]{} (391.13,627.05) (394,625)(16.889,-12.064)[0]{} (401,620)(16.889,-12.064)[0]{} (408.02,614.99) (415,610)(16.889,-12.064)[0]{} (424.90,602.93) (429,600)(15.759,-13.508)[0]{} (441.29,590.22) (443,589)(16.889,-12.064)[0]{} (450,584)(16.889,-12.064)[0]{} (458.18,578.16) (464,574)(16.889,-12.064)[0]{} (475.07,566.09) (478,564)(16.889,-12.064)[0]{} (491.49,553.43) (492,553)(16.889,-12.064)[0]{} (499,548)(16.889,-12.064)[0]{} (508.35,541.32) (513,538)(16.889,-12.064)[0]{} (525.24,529.26) (527,528)(16.889,-12.064)[0]{} (534,523)(15.759,-13.508)[0]{} (541.62,516.55) (548,512)(16.889,-12.064)[0]{} (558.51,504.49) (562,502)(15.945,-13.287)[0]{} (568,497)(16.889,-12.064)[0]{} (575.05,491.97) (582,487)(16.889,-12.064)[0]{} (591.74,479.65) (596,476)(16.889,-12.064)[0]{} (608.32,467.20) (610,466)(16.889,-12.064)[0]{} (617,461)(16.889,-12.064)[0]{} (625.21,455.13) (631,451)(16.889,-12.064)[0]{} (641.83,442.72) (645,440)(16.889,-12.064)[0]{} (658.49,430.36) (659,430)(16.889,-12.064)[0]{} (666,425)(16.889,-12.064)[0]{} (675.38,418.30) (680,415)(16.889,-12.064)[0]{} (691.92,405.79) (694,404)(16.889,-12.064)[0]{} (701,399)(16.889,-12.064)[0]{} (708.66,393.53) (715,389)(16.889,-12.064)[0]{} (725.55,381.47) (729,379)(16.889,-12.064)[0]{} (742.44,369.40) (743,369)(15.759,-13.508)[0]{} (750,363)(16.889,-12.064)[0]{} (758.82,356.70) (764,353)(16.889,-12.064)[0]{} (775.71,344.63) (778,343)(16.889,-12.064)[0]{} (785,338)(16.889,-12.064)[0]{} (792.52,332.48) (798,327)(16.889,-12.064)[0]{} (808.59,319.44) (812,317)(16.889,-12.064)[0]{} (825.48,307.37) (826,307)(16.889,-12.064)[0]{} (833,302)(16.889,-12.064)[0]{} (842.36,295.31) (847,292)(15.759,-13.508)[0]{} (858.75,282.61) (861,281)(16.889,-12.064)[0]{} (868,276)(16.889,-12.064)[0]{} (875.64,270.54) (882,266)(16.889,-12.064)[0]{} (892.53,258.48) (896,256)(15.759,-13.508)[0]{} (908.92,245.77) (910,245) \[fig1\] Simulation results ================== We have simulated Eqns. (\[sde\]) in two spatial dimensions in a system with periodic boundary conditions. The velocity autocorrelation function of the DPD particles and also the environment velocity autocorrelation function have been computed at equilibrium. In order to derive Eqn. (\[acvf02\]) we made the approximation that the density field is approximately constant. We check now that this assumption was reasonable by computing in a simulation the environment velocity correlation $C({\overline t})$, evaluating numerically the integral term in Eqn. (\[acvf02\]), and adding up the first exponential term in Eqn. (\[acvf02\]). The result is the dotted line in Fig. \[fig.check\]. Also shown is the result for the velocity correlation function $c({\overline t})$ obtained directly from the simulation (solid line). We see that both results are in quite good agreement giving confidence on Eqn. (\[acvf02\]) as a sounded starting point for theoretical analysis. Similar good agreement is obtained for all the values of $\Omega$ studied ($\Omega=0.5,8.3,25$) (974,675)(0,0) =cmr10 at 10pt (220.0,113.0) ------------------------------------------------------------------------ (220.0,113.0) ------------------------------------------------------------------------ (220.0,113.0) ------------------------------------------------------------------------ (198,113)[(0,0)\[r\][0]{}]{} (890.0,113.0) ------------------------------------------------------------------------ (220.0,221.0) ------------------------------------------------------------------------ (198,221)[(0,0)\[r\][0.2]{}]{} (890.0,221.0) ------------------------------------------------------------------------ (220.0,329.0) ------------------------------------------------------------------------ (198,329)[(0,0)\[r\][0.4]{}]{} (890.0,329.0) ------------------------------------------------------------------------ (220.0,436.0) ------------------------------------------------------------------------ (198,436)[(0,0)\[r\][0.6]{}]{} (890.0,436.0) ------------------------------------------------------------------------ (220.0,544.0) ------------------------------------------------------------------------ (198,544)[(0,0)\[r\][0.8]{}]{} (890.0,544.0) ------------------------------------------------------------------------ (220.0,652.0) ------------------------------------------------------------------------ (198,652)[(0,0)\[r\][1]{}]{} (890.0,652.0) ------------------------------------------------------------------------ (220.0,113.0) ------------------------------------------------------------------------ (220,68)[(0,0)[0]{}]{} (220.0,632.0) ------------------------------------------------------------------------ (358.0,113.0) ------------------------------------------------------------------------ (358,68)[(0,0)[0.2]{}]{} (358.0,632.0) ------------------------------------------------------------------------ (496.0,113.0) ------------------------------------------------------------------------ (496,68)[(0,0)[0.4]{}]{} (496.0,632.0) ------------------------------------------------------------------------ (634.0,113.0) ------------------------------------------------------------------------ (634,68)[(0,0)[0.6]{}]{} (634.0,632.0) ------------------------------------------------------------------------ (772.0,113.0) ------------------------------------------------------------------------ (772,68)[(0,0)[0.8]{}]{} (772.0,632.0) ------------------------------------------------------------------------ (910.0,113.0) ------------------------------------------------------------------------ (910,68)[(0,0)[1]{}]{} (910.0,632.0) ------------------------------------------------------------------------ (220.0,113.0) ------------------------------------------------------------------------ (910.0,113.0) ------------------------------------------------------------------------ (220.0,652.0) ------------------------------------------------------------------------ (45,382)[(0,0)[$c({\overline t})$]{}]{} (565,23)[(0,0)[${\overline t}$]{}]{} (220.0,113.0) ------------------------------------------------------------------------ (221,652)(1.263,-20.717)[4]{} (226,570)(1.233,-20.719)[5]{} (231,486)(1.908,-20.668)[3]{} (237,421)(2.379,-20.619)[2]{} (243,369)(2.574,-20.595)[2]{} (248,329)(3.825,-20.400)[2]{} (257.78,280.62) (262.52,260.42) (268.34,240.53) (275.74,221.14) (277,218)(7.983,-19.159)[0]{} (284.25,202.25) (288,196)(12.453,-16.604)[0]{} (295.84,185.06) (299,180)(14.676,-14.676)[0]{} (309.84,169.96) (311,169)(14.676,-14.676)[0]{} (316,164)(17.270,-11.513)[0]{} (326.13,157.24) (328,156)(17.798,-10.679)[0]{} (333,153)(18.564,-9.282)[0]{} (339,150)(19.271,-7.708)[0]{} (344.56,147.81) (350,146)(19.690,-6.563)[0]{} (356,144)(19.271,-7.708)[0]{} (364.14,140.95) (367,140)(20.473,-3.412)[0]{} (373,139)(19.271,-7.708)[0]{} (378,137)(20.473,-3.412)[0]{} (384.19,135.97) (390,135)(20.595,-2.574)[3]{} (446,128)(20.705,-1.453)[2]{} (503,124)(20.742,-0.741)[3]{} (559,122)(20.743,-0.728)[3]{} (616,120)(20.752,-0.364)[2]{} (673,119)(20.752,-0.371)[3]{} (729,118)(20.752,-0.364)[3]{} (786,117)(20.756,0.000)[3]{} (842,117)(20.752,-0.364)[2]{} (902.31,116.00) (910,116) (222.60,620.87)(0.468,-10.717)[5]{} ------------------------------------------------------------------------ (221.17,636.43)(4.000,-58.433)[2]{} ------------------------------------------------------------------------ (226.59,550.35)(0.477,-9.059)[7]{} ------------------------------------------------------------------------ (225.17,564.18)(5.000,-68.177)[2]{} ------------------------------------------------------------------------ (231.59,477.32)(0.482,-5.915)[9]{} ------------------------------------------------------------------------ (230.17,486.66)(6.000,-56.660)[2]{} ------------------------------------------------------------------------ (237.59,414.64)(0.482,-4.830)[9]{} ------------------------------------------------------------------------ (236.17,422.32)(6.000,-46.320)[2]{} ------------------------------------------------------------------------ (243.59,361.31)(0.477,-4.718)[7]{} ------------------------------------------------------------------------ (242.17,368.65)(5.000,-35.653)[2]{} ------------------------------------------------------------------------ (248.59,323.18)(0.482,-3.022)[9]{} ------------------------------------------------------------------------ (247.17,328.09)(6.000,-29.088)[2]{} ------------------------------------------------------------------------ (254.59,290.56)(0.482,-2.570)[9]{} ------------------------------------------------------------------------ (253.17,294.78)(6.000,-24.780)[2]{} ------------------------------------------------------------------------ (260.59,261.95)(0.477,-2.491)[7]{} ------------------------------------------------------------------------ (259.17,265.97)(5.000,-18.973)[2]{} ------------------------------------------------------------------------ (265.59,241.33)(0.482,-1.666)[9]{} ------------------------------------------------------------------------ (264.17,244.16)(6.000,-16.163)[2]{} ------------------------------------------------------------------------ (271.59,223.16)(0.482,-1.395)[9]{} ------------------------------------------------------------------------ (270.17,225.58)(6.000,-13.579)[2]{} ------------------------------------------------------------------------ (277,212) (277,212) (277.59,207.27)(0.477,-1.378)[7]{} ------------------------------------------------------------------------ (276.17,209.63)(5.000,-10.634)[2]{} ------------------------------------------------------------------------ (282.59,195.82)(0.482,-0.852)[9]{} ------------------------------------------------------------------------ (281.17,197.41)(6.000,-8.409)[2]{} ------------------------------------------------------------------------ (288.59,186.09)(0.482,-0.762)[9]{} ------------------------------------------------------------------------ (287.17,187.55)(6.000,-7.547)[2]{} ------------------------------------------------------------------------ (294.59,177.26)(0.477,-0.710)[7]{} ------------------------------------------------------------------------ (293.17,178.63)(5.000,-5.630)[2]{} ------------------------------------------------------------------------ (299.59,170.65)(0.482,-0.581)[9]{} ------------------------------------------------------------------------ (298.17,171.82)(6.000,-5.824)[2]{} ------------------------------------------------------------------------ (305.00,164.93)(0.599,-0.477)[7]{} ------------------------------------------------------------------------ (305.00,165.17)(4.796,-5.000)[2]{} ------------------------------------------------------------------------ (311.00,159.94)(0.627,-0.468)[5]{} ------------------------------------------------------------------------ (311.00,160.17)(3.755,-4.000)[2]{} ------------------------------------------------------------------------ (316.00,155.94)(0.774,-0.468)[5]{} ------------------------------------------------------------------------ (316.00,156.17)(4.547,-4.000)[2]{} ------------------------------------------------------------------------ (322.00,151.95)(1.132,-0.447)[3]{} ------------------------------------------------------------------------ (322.00,152.17)(4.132,-3.000)[2]{} ------------------------------------------------------------------------ (328.00,148.95)(0.909,-0.447)[3]{} ------------------------------------------------------------------------ (328.00,149.17)(3.409,-3.000)[2]{} ------------------------------------------------------------------------ (333.00,145.95)(1.132,-0.447)[3]{} ------------------------------------------------------------------------ (333.00,146.17)(4.132,-3.000)[2]{} ------------------------------------------------------------------------ (339,142.17) ------------------------------------------------------------------------ (339.00,143.17)(2.717,-2.000)[2]{} ------------------------------------------------------------------------ (344,140.67) ------------------------------------------------------------------------ (344.00,141.17)(3.000,-1.000)[2]{} ------------------------------------------------------------------------ (350,139.17) ------------------------------------------------------------------------ (350.00,140.17)(3.302,-2.000)[2]{} ------------------------------------------------------------------------ (356,137.67) ------------------------------------------------------------------------ (356.00,138.17)(2.500,-1.000)[2]{} ------------------------------------------------------------------------ (361,136.17) ------------------------------------------------------------------------ (361.00,137.17)(3.302,-2.000)[2]{} ------------------------------------------------------------------------ (367,134.67) ------------------------------------------------------------------------ (367.00,135.17)(3.000,-1.000)[2]{} ------------------------------------------------------------------------ (373,133.67) ------------------------------------------------------------------------ (373.00,134.17)(2.500,-1.000)[2]{} ------------------------------------------------------------------------ (378,132.67) ------------------------------------------------------------------------ (378.00,133.17)(3.000,-1.000)[2]{} ------------------------------------------------------------------------ (384,131.67) ------------------------------------------------------------------------ (384.00,132.17)(3.000,-1.000)[2]{} ------------------------------------------------------------------------ (390.00,130.93)(6.165,-0.477)[7]{} ------------------------------------------------------------------------ (390.00,131.17)(46.494,-5.000)[2]{} ------------------------------------------------------------------------ (446.00,125.93)(6.276,-0.477)[7]{} ------------------------------------------------------------------------ (446.00,126.17)(47.328,-5.000)[2]{} ------------------------------------------------------------------------ (503,120.17) ------------------------------------------------------------------------ (503.00,121.17)(32.546,-2.000)[2]{} ------------------------------------------------------------------------ (559,118.67) ------------------------------------------------------------------------ (559.00,119.17)(28.500,-1.000)[2]{} ------------------------------------------------------------------------ (616,117.67) ------------------------------------------------------------------------ (616.00,118.17)(28.500,-1.000)[2]{} ------------------------------------------------------------------------ (673,116.67) ------------------------------------------------------------------------ (673.00,117.17)(28.000,-1.000)[2]{} ------------------------------------------------------------------------ (786,115.67) ------------------------------------------------------------------------ (786.00,116.17)(28.000,-1.000)[2]{} ------------------------------------------------------------------------ (729.0,117.0) ------------------------------------------------------------------------ (842.0,116.0) ------------------------------------------------------------------------ For a given value of $\Omega=25$ we plot in Fig. \[fig2\] the value of $s^2 C({\overline t})$, for different values of $s$. According to the hydrodynamic prediction (\[hp3\]) this curve should be independent of the overlapping coefficient $s$ (dotted curve in Fig. \[fig2\]). We observe that the simulation results converge towards the theoretical prediction only when the overlapping coefficient is sufficiently large. This is expected from the fact that hydrodynamic behavior appears only on length scales that involve a relatively large number of particles. (974,675)(0,0) =cmr10 at 10pt (220.0,113.0) ------------------------------------------------------------------------ (220.0,113.0) ------------------------------------------------------------------------ (220.0,113.0) ------------------------------------------------------------------------ (198,113)[(0,0)\[r\][0]{}]{} (890.0,113.0) ------------------------------------------------------------------------ (220.0,221.0) ------------------------------------------------------------------------ (198,221)[(0,0)\[r\][0.2]{}]{} (890.0,221.0) ------------------------------------------------------------------------ (220.0,329.0) ------------------------------------------------------------------------ (198,329)[(0,0)\[r\][0.4]{}]{} (890.0,329.0) ------------------------------------------------------------------------ (220.0,436.0) ------------------------------------------------------------------------ (198,436)[(0,0)\[r\][0.6]{}]{} (890.0,436.0) ------------------------------------------------------------------------ (220.0,544.0) ------------------------------------------------------------------------ (198,544)[(0,0)\[r\][0.8]{}]{} (890.0,544.0) ------------------------------------------------------------------------ (220.0,652.0) ------------------------------------------------------------------------ (198,652)[(0,0)\[r\][1]{}]{} (890.0,652.0) ------------------------------------------------------------------------ (220.0,113.0) ------------------------------------------------------------------------ (220,68)[(0,0)[0]{}]{} (220.0,632.0) ------------------------------------------------------------------------ (335.0,113.0) ------------------------------------------------------------------------ (335,68)[(0,0)[0.5]{}]{} (335.0,632.0) ------------------------------------------------------------------------ (450.0,113.0) ------------------------------------------------------------------------ (450,68)[(0,0)[1]{}]{} (450.0,632.0) ------------------------------------------------------------------------ (565.0,113.0) ------------------------------------------------------------------------ (565,68)[(0,0)[1.5]{}]{} (565.0,632.0) ------------------------------------------------------------------------ (680.0,113.0) ------------------------------------------------------------------------ (680,68)[(0,0)[2]{}]{} (680.0,632.0) ------------------------------------------------------------------------ (795.0,113.0) ------------------------------------------------------------------------ (795,68)[(0,0)[2.5]{}]{} (795.0,632.0) ------------------------------------------------------------------------ (910.0,113.0) ------------------------------------------------------------------------ (910,68)[(0,0)[3]{}]{} (910.0,632.0) ------------------------------------------------------------------------ (220.0,113.0) ------------------------------------------------------------------------ (910.0,113.0) ------------------------------------------------------------------------ (220.0,652.0) ------------------------------------------------------------------------ (45,382)[(0,0)[$s^2 C({\overline t}) $]{}]{} (565,23)[(0,0)[${\overline t}$]{}]{} (220.0,113.0) ------------------------------------------------------------------------ (220,635) (220.59,592.52)(0.489,-13.251)[15]{} ------------------------------------------------------------------------ (219.17,613.76)(9.000,-206.760)[2]{} ------------------------------------------------------------------------ (229.58,392.80)(0.491,-4.300)[17]{} ------------------------------------------------------------------------ (228.17,399.90)(10.000,-75.902)[2]{} ------------------------------------------------------------------------ (239.59,315.84)(0.489,-2.417)[15]{} ------------------------------------------------------------------------ (238.17,319.92)(9.000,-37.918)[2]{} ------------------------------------------------------------------------ (248.58,277.10)(0.491,-1.381)[17]{} ------------------------------------------------------------------------ (247.17,279.55)(10.000,-24.551)[2]{} ------------------------------------------------------------------------ (258.59,250.90)(0.489,-1.135)[15]{} ------------------------------------------------------------------------ (257.17,252.95)(9.000,-17.948)[2]{} ------------------------------------------------------------------------ (267.58,232.26)(0.491,-0.704)[17]{} ------------------------------------------------------------------------ (266.17,233.63)(10.000,-12.630)[2]{} ------------------------------------------------------------------------ (277.59,218.37)(0.489,-0.669)[15]{} ------------------------------------------------------------------------ (276.17,219.69)(9.000,-10.685)[2]{} ------------------------------------------------------------------------ (286.59,206.74)(0.489,-0.553)[15]{} ------------------------------------------------------------------------ (285.17,207.87)(9.000,-8.870)[2]{} ------------------------------------------------------------------------ (295.00,197.93)(0.626,-0.488)[13]{} ------------------------------------------------------------------------ (295.00,198.17)(8.755,-8.000)[2]{} ------------------------------------------------------------------------ (305.00,189.93)(0.645,-0.485)[11]{} ------------------------------------------------------------------------ (305.00,190.17)(7.725,-7.000)[2]{} ------------------------------------------------------------------------ (314.00,182.93)(0.852,-0.482)[9]{} ------------------------------------------------------------------------ (314.00,183.17)(8.409,-6.000)[2]{} ------------------------------------------------------------------------ (324.00,176.93)(0.933,-0.477)[7]{} ------------------------------------------------------------------------ (324.00,177.17)(7.298,-5.000)[2]{} ------------------------------------------------------------------------ (333.00,171.94)(1.358,-0.468)[5]{} ------------------------------------------------------------------------ (333.00,172.17)(7.717,-4.000)[2]{} ------------------------------------------------------------------------ (343.00,167.94)(1.212,-0.468)[5]{} ------------------------------------------------------------------------ (343.00,168.17)(6.924,-4.000)[2]{} ------------------------------------------------------------------------ (352.00,163.94)(1.212,-0.468)[5]{} ------------------------------------------------------------------------ (352.00,164.17)(6.924,-4.000)[2]{} ------------------------------------------------------------------------ (361.00,159.95)(2.025,-0.447)[3]{} ------------------------------------------------------------------------ (361.00,160.17)(7.025,-3.000)[2]{} ------------------------------------------------------------------------ (371.00,156.95)(1.802,-0.447)[3]{} ------------------------------------------------------------------------ (371.00,157.17)(6.302,-3.000)[2]{} ------------------------------------------------------------------------ (380,153.17) ------------------------------------------------------------------------ (380.00,154.17)(5.641,-2.000)[2]{} ------------------------------------------------------------------------ (390,151.17) ------------------------------------------------------------------------ (390.00,152.17)(5.056,-2.000)[2]{} ------------------------------------------------------------------------ (399,149.17) ------------------------------------------------------------------------ (399.00,150.17)(5.641,-2.000)[2]{} ------------------------------------------------------------------------ (409,147.17) ------------------------------------------------------------------------ (409.00,148.17)(5.056,-2.000)[2]{} ------------------------------------------------------------------------ (418,145.17) ------------------------------------------------------------------------ (418.00,146.17)(5.056,-2.000)[2]{} ------------------------------------------------------------------------ (427,143.17) ------------------------------------------------------------------------ (427.00,144.17)(5.641,-2.000)[2]{} ------------------------------------------------------------------------ (437,141.17) ------------------------------------------------------------------------ (437.00,142.17)(5.056,-2.000)[2]{} ------------------------------------------------------------------------ (446,139.67) ------------------------------------------------------------------------ (446.00,140.17)(5.000,-1.000)[2]{} ------------------------------------------------------------------------ (456,138.67) ------------------------------------------------------------------------ (456.00,139.17)(4.500,-1.000)[2]{} ------------------------------------------------------------------------ (465,137.67) ------------------------------------------------------------------------ (465.00,138.17)(5.000,-1.000)[2]{} ------------------------------------------------------------------------ (475,136.67) ------------------------------------------------------------------------ (475.00,137.17)(4.500,-1.000)[2]{} ------------------------------------------------------------------------ (484,135.67) ------------------------------------------------------------------------ (484.00,136.17)(4.500,-1.000)[2]{} ------------------------------------------------------------------------ (493,134.67) ------------------------------------------------------------------------ (493.00,135.17)(5.000,-1.000)[2]{} ------------------------------------------------------------------------ (503,133.67) ------------------------------------------------------------------------ (503.00,134.17)(4.500,-1.000)[2]{} ------------------------------------------------------------------------ (512,132.67) ------------------------------------------------------------------------ (512.00,133.17)(5.000,-1.000)[2]{} ------------------------------------------------------------------------ (522,131.67) ------------------------------------------------------------------------ (522.00,132.17)(4.500,-1.000)[2]{} ------------------------------------------------------------------------ (541,130.67) ------------------------------------------------------------------------ (541.00,131.17)(4.500,-1.000)[2]{} ------------------------------------------------------------------------ (550,129.67) ------------------------------------------------------------------------ (550.00,130.17)(4.500,-1.000)[2]{} ------------------------------------------------------------------------ (559,128.67) ------------------------------------------------------------------------ (559.00,129.17)(5.000,-1.000)[2]{} ------------------------------------------------------------------------ (531.0,132.0) ------------------------------------------------------------------------ (578,127.67) ------------------------------------------------------------------------ (578.00,128.17)(5.000,-1.000)[2]{} ------------------------------------------------------------------------ (569.0,129.0) ------------------------------------------------------------------------ (597,126.67) ------------------------------------------------------------------------ (597.00,127.17)(5.000,-1.000)[2]{} ------------------------------------------------------------------------ (588.0,128.0) ------------------------------------------------------------------------ (616,125.67) ------------------------------------------------------------------------ (616.00,126.17)(4.500,-1.000)[2]{} ------------------------------------------------------------------------ (607.0,127.0) ------------------------------------------------------------------------ (644,124.67) ------------------------------------------------------------------------ (644.00,125.17)(5.000,-1.000)[2]{} ------------------------------------------------------------------------ (625.0,126.0) ------------------------------------------------------------------------ (663,123.67) ------------------------------------------------------------------------ (663.00,124.17)(5.000,-1.000)[2]{} ------------------------------------------------------------------------ (654.0,125.0) ------------------------------------------------------------------------ (692,122.67) ------------------------------------------------------------------------ (692.00,123.17)(4.500,-1.000)[2]{} ------------------------------------------------------------------------ (673.0,124.0) ------------------------------------------------------------------------ (720,121.67) ------------------------------------------------------------------------ (720.00,122.17)(4.500,-1.000)[2]{} ------------------------------------------------------------------------ (701.0,123.0) ------------------------------------------------------------------------ (758,120.67) ------------------------------------------------------------------------ (758.00,121.17)(4.500,-1.000)[2]{} ------------------------------------------------------------------------ (729.0,122.0) ------------------------------------------------------------------------ (795,119.67) ------------------------------------------------------------------------ (795.00,120.17)(5.000,-1.000)[2]{} ------------------------------------------------------------------------ (767.0,121.0) ------------------------------------------------------------------------ (833,118.67) ------------------------------------------------------------------------ (833.00,119.17)(4.500,-1.000)[2]{} ------------------------------------------------------------------------ (805.0,120.0) ------------------------------------------------------------------------ (880,117.67) ------------------------------------------------------------------------ (880.00,118.17)(5.000,-1.000)[2]{} ------------------------------------------------------------------------ (842.0,119.0) ------------------------------------------------------------------------ (890.0,118.0) ------------------------------------------------------------------------ (220.59,604.72)(0.489,-14.766)[15]{} ------------------------------------------------------------------------ (219.17,628.36)(9.000,-230.362)[2]{} ------------------------------------------------------------------------ (229.58,385.46)(0.491,-3.779)[17]{} ------------------------------------------------------------------------ (228.17,391.73)(10.000,-66.732)[2]{} ------------------------------------------------------------------------ (239.59,317.94)(0.489,-2.067)[15]{} ------------------------------------------------------------------------ (238.17,321.47)(9.000,-32.472)[2]{} ------------------------------------------------------------------------ (248.58,284.60)(0.491,-1.225)[17]{} ------------------------------------------------------------------------ (247.17,286.80)(10.000,-21.800)[2]{} ------------------------------------------------------------------------ (258.59,261.26)(0.489,-1.019)[15]{} ------------------------------------------------------------------------ (257.17,263.13)(9.000,-16.132)[2]{} ------------------------------------------------------------------------ (267.58,244.09)(0.491,-0.756)[17]{} ------------------------------------------------------------------------ (266.17,245.55)(10.000,-13.547)[2]{} ------------------------------------------------------------------------ (277.59,229.56)(0.489,-0.611)[15]{} ------------------------------------------------------------------------ (276.17,230.78)(9.000,-9.778)[2]{} ------------------------------------------------------------------------ (286.59,218.74)(0.489,-0.553)[15]{} ------------------------------------------------------------------------ (285.17,219.87)(9.000,-8.870)[2]{} ------------------------------------------------------------------------ (295.00,209.93)(0.626,-0.488)[13]{} ------------------------------------------------------------------------ (295.00,210.17)(8.755,-8.000)[2]{} ------------------------------------------------------------------------ (305.00,201.93)(0.645,-0.485)[11]{} ------------------------------------------------------------------------ (305.00,202.17)(7.725,-7.000)[2]{} ------------------------------------------------------------------------ (314.00,194.93)(0.852,-0.482)[9]{} ------------------------------------------------------------------------ (314.00,195.17)(8.409,-6.000)[2]{} ------------------------------------------------------------------------ (324.00,188.93)(0.933,-0.477)[7]{} ------------------------------------------------------------------------ (324.00,189.17)(7.298,-5.000)[2]{} ------------------------------------------------------------------------ (333.00,183.93)(1.044,-0.477)[7]{} ------------------------------------------------------------------------ (333.00,184.17)(8.132,-5.000)[2]{} ------------------------------------------------------------------------ (343.00,178.94)(1.212,-0.468)[5]{} ------------------------------------------------------------------------ (343.00,179.17)(6.924,-4.000)[2]{} ------------------------------------------------------------------------ (352.00,174.94)(1.212,-0.468)[5]{} ------------------------------------------------------------------------ (352.00,175.17)(6.924,-4.000)[2]{} ------------------------------------------------------------------------ (361.00,170.94)(1.358,-0.468)[5]{} ------------------------------------------------------------------------ (361.00,171.17)(7.717,-4.000)[2]{} ------------------------------------------------------------------------ (371.00,166.95)(1.802,-0.447)[3]{} ------------------------------------------------------------------------ (371.00,167.17)(6.302,-3.000)[2]{} ------------------------------------------------------------------------ (380.00,163.95)(2.025,-0.447)[3]{} ------------------------------------------------------------------------ (380.00,164.17)(7.025,-3.000)[2]{} ------------------------------------------------------------------------ (390.00,160.95)(1.802,-0.447)[3]{} ------------------------------------------------------------------------ (390.00,161.17)(6.302,-3.000)[2]{} ------------------------------------------------------------------------ (399,157.17) ------------------------------------------------------------------------ (399.00,158.17)(5.641,-2.000)[2]{} ------------------------------------------------------------------------ (409,155.17) ------------------------------------------------------------------------ (409.00,156.17)(5.056,-2.000)[2]{} ------------------------------------------------------------------------ (418,153.17) ------------------------------------------------------------------------ (418.00,154.17)(5.056,-2.000)[2]{} ------------------------------------------------------------------------ (427,151.17) ------------------------------------------------------------------------ (427.00,152.17)(5.641,-2.000)[2]{} ------------------------------------------------------------------------ (437,149.17) ------------------------------------------------------------------------ (437.00,150.17)(5.056,-2.000)[2]{} ------------------------------------------------------------------------ (446,147.17) ------------------------------------------------------------------------ (446.00,148.17)(5.641,-2.000)[2]{} ------------------------------------------------------------------------ (456,145.17) ------------------------------------------------------------------------ (456.00,146.17)(5.056,-2.000)[2]{} ------------------------------------------------------------------------ (465,143.67) ------------------------------------------------------------------------ (465.00,144.17)(5.000,-1.000)[2]{} ------------------------------------------------------------------------ (475,142.67) ------------------------------------------------------------------------ (475.00,143.17)(4.500,-1.000)[2]{} ------------------------------------------------------------------------ (484,141.17) ------------------------------------------------------------------------ (484.00,142.17)(5.056,-2.000)[2]{} ------------------------------------------------------------------------ (493,139.67) ------------------------------------------------------------------------ (493.00,140.17)(5.000,-1.000)[2]{} ------------------------------------------------------------------------ (503,138.67) ------------------------------------------------------------------------ (503.00,139.17)(4.500,-1.000)[2]{} ------------------------------------------------------------------------ (512,137.67) ------------------------------------------------------------------------ (512.00,138.17)(5.000,-1.000)[2]{} ------------------------------------------------------------------------ (522,136.67) ------------------------------------------------------------------------ (522.00,137.17)(4.500,-1.000)[2]{} ------------------------------------------------------------------------ (531,135.67) ------------------------------------------------------------------------ (531.00,136.17)(5.000,-1.000)[2]{} ------------------------------------------------------------------------ (541,134.67) ------------------------------------------------------------------------ (541.00,135.17)(4.500,-1.000)[2]{} ------------------------------------------------------------------------ (550,133.67) ------------------------------------------------------------------------ (550.00,134.17)(4.500,-1.000)[2]{} ------------------------------------------------------------------------ (569,132.67) ------------------------------------------------------------------------ (569.00,133.17)(4.500,-1.000)[2]{} ------------------------------------------------------------------------ (578,131.67) ------------------------------------------------------------------------ (578.00,132.17)(5.000,-1.000)[2]{} ------------------------------------------------------------------------ (559.0,134.0) ------------------------------------------------------------------------ (597,130.67) ------------------------------------------------------------------------ (597.00,131.17)(5.000,-1.000)[2]{} ------------------------------------------------------------------------ (607,129.67) ------------------------------------------------------------------------ (607.00,130.17)(4.500,-1.000)[2]{} ------------------------------------------------------------------------ (588.0,132.0) ------------------------------------------------------------------------ (625,128.67) ------------------------------------------------------------------------ (625.00,129.17)(5.000,-1.000)[2]{} ------------------------------------------------------------------------ (616.0,130.0) ------------------------------------------------------------------------ (644,127.67) ------------------------------------------------------------------------ (644.00,128.17)(5.000,-1.000)[2]{} ------------------------------------------------------------------------ (635.0,129.0) ------------------------------------------------------------------------ (663,126.67) ------------------------------------------------------------------------ (663.00,127.17)(5.000,-1.000)[2]{} ------------------------------------------------------------------------ (654.0,128.0) ------------------------------------------------------------------------ (682,125.67) ------------------------------------------------------------------------ (682.00,126.17)(5.000,-1.000)[2]{} ------------------------------------------------------------------------ (673.0,127.0) ------------------------------------------------------------------------ (710,124.67) ------------------------------------------------------------------------ (710.00,125.17)(5.000,-1.000)[2]{} ------------------------------------------------------------------------ (692.0,126.0) ------------------------------------------------------------------------ (739,123.67) ------------------------------------------------------------------------ (739.00,124.17)(4.500,-1.000)[2]{} ------------------------------------------------------------------------ (720.0,125.0) ------------------------------------------------------------------------ (776,122.67) ------------------------------------------------------------------------ (776.00,123.17)(5.000,-1.000)[2]{} ------------------------------------------------------------------------ (748.0,124.0) ------------------------------------------------------------------------ (814,121.67) ------------------------------------------------------------------------ (814.00,122.17)(5.000,-1.000)[2]{} ------------------------------------------------------------------------ (786.0,123.0) ------------------------------------------------------------------------ (861,120.67) ------------------------------------------------------------------------ (861.00,121.17)(5.000,-1.000)[2]{} ------------------------------------------------------------------------ (824.0,122.0) ------------------------------------------------------------------------ (871.0,121.0) ------------------------------------------------------------------------ (221.59,596.17)(0.488,-17.597)[13]{} ------------------------------------------------------------------------ (220.17,624.08)(8.000,-239.084)[2]{} ------------------------------------------------------------------------ (229.58,375.12)(0.491,-2.945)[17]{} ------------------------------------------------------------------------ (228.17,380.06)(10.000,-52.060)[2]{} ------------------------------------------------------------------------ (239.59,322.23)(0.489,-1.660)[15]{} ------------------------------------------------------------------------ (238.17,325.12)(9.000,-26.117)[2]{} ------------------------------------------------------------------------ (248.58,295.10)(0.491,-1.069)[17]{} ------------------------------------------------------------------------ (247.17,297.05)(10.000,-19.049)[2]{} ------------------------------------------------------------------------ (258.59,274.82)(0.489,-0.844)[15]{} ------------------------------------------------------------------------ (257.17,276.41)(9.000,-13.409)[2]{} ------------------------------------------------------------------------ (267.58,260.26)(0.491,-0.704)[17]{} ------------------------------------------------------------------------ (266.17,261.63)(10.000,-12.630)[2]{} ------------------------------------------------------------------------ (277.59,246.74)(0.489,-0.553)[15]{} ------------------------------------------------------------------------ (276.17,247.87)(9.000,-8.870)[2]{} ------------------------------------------------------------------------ (286.59,236.74)(0.489,-0.553)[15]{} ------------------------------------------------------------------------ (285.17,237.87)(9.000,-8.870)[2]{} ------------------------------------------------------------------------ (295.00,227.93)(0.626,-0.488)[13]{} ------------------------------------------------------------------------ (295.00,228.17)(8.755,-8.000)[2]{} ------------------------------------------------------------------------ (305.00,219.93)(0.645,-0.485)[11]{} ------------------------------------------------------------------------ (305.00,220.17)(7.725,-7.000)[2]{} ------------------------------------------------------------------------ (314.00,212.93)(0.852,-0.482)[9]{} ------------------------------------------------------------------------ (314.00,213.17)(8.409,-6.000)[2]{} ------------------------------------------------------------------------ (324.00,206.93)(0.762,-0.482)[9]{} ------------------------------------------------------------------------ (324.00,207.17)(7.547,-6.000)[2]{} ------------------------------------------------------------------------ (333.00,200.93)(1.044,-0.477)[7]{} ------------------------------------------------------------------------ (333.00,201.17)(8.132,-5.000)[2]{} ------------------------------------------------------------------------ (343.00,195.93)(0.933,-0.477)[7]{} ------------------------------------------------------------------------ (343.00,196.17)(7.298,-5.000)[2]{} ------------------------------------------------------------------------ (352.00,190.94)(1.212,-0.468)[5]{} ------------------------------------------------------------------------ (352.00,191.17)(6.924,-4.000)[2]{} ------------------------------------------------------------------------ (361.00,186.94)(1.358,-0.468)[5]{} ------------------------------------------------------------------------ (361.00,187.17)(7.717,-4.000)[2]{} ------------------------------------------------------------------------ (371.00,182.94)(1.212,-0.468)[5]{} ------------------------------------------------------------------------ (371.00,183.17)(6.924,-4.000)[2]{} ------------------------------------------------------------------------ (380.00,178.95)(2.025,-0.447)[3]{} ------------------------------------------------------------------------ (380.00,179.17)(7.025,-3.000)[2]{} ------------------------------------------------------------------------ (390.00,175.95)(1.802,-0.447)[3]{} ------------------------------------------------------------------------ (390.00,176.17)(6.302,-3.000)[2]{} ------------------------------------------------------------------------ (399.00,172.95)(2.025,-0.447)[3]{} ------------------------------------------------------------------------ (399.00,173.17)(7.025,-3.000)[2]{} ------------------------------------------------------------------------ (409.00,169.95)(1.802,-0.447)[3]{} ------------------------------------------------------------------------ (409.00,170.17)(6.302,-3.000)[2]{} ------------------------------------------------------------------------ (418,166.17) ------------------------------------------------------------------------ (418.00,167.17)(5.056,-2.000)[2]{} ------------------------------------------------------------------------ (427.00,164.95)(2.025,-0.447)[3]{} ------------------------------------------------------------------------ (427.00,165.17)(7.025,-3.000)[2]{} ------------------------------------------------------------------------ (437,161.17) ------------------------------------------------------------------------ (437.00,162.17)(5.056,-2.000)[2]{} ------------------------------------------------------------------------ (446,159.17) ------------------------------------------------------------------------ (446.00,160.17)(5.641,-2.000)[2]{} ------------------------------------------------------------------------ (456,157.17) ------------------------------------------------------------------------ (456.00,158.17)(5.056,-2.000)[2]{} ------------------------------------------------------------------------ (465,155.17) ------------------------------------------------------------------------ (465.00,156.17)(5.641,-2.000)[2]{} ------------------------------------------------------------------------ (475,153.67) ------------------------------------------------------------------------ (475.00,154.17)(4.500,-1.000)[2]{} ------------------------------------------------------------------------ (484,152.17) ------------------------------------------------------------------------ (484.00,153.17)(5.056,-2.000)[2]{} ------------------------------------------------------------------------ (493,150.67) ------------------------------------------------------------------------ (493.00,151.17)(5.000,-1.000)[2]{} ------------------------------------------------------------------------ (503,149.17) ------------------------------------------------------------------------ (503.00,150.17)(5.056,-2.000)[2]{} ------------------------------------------------------------------------ (512,147.67) ------------------------------------------------------------------------ (512.00,148.17)(5.000,-1.000)[2]{} ------------------------------------------------------------------------ (522,146.17) ------------------------------------------------------------------------ (522.00,147.17)(5.056,-2.000)[2]{} ------------------------------------------------------------------------ (531,144.67) ------------------------------------------------------------------------ (531.00,145.17)(5.000,-1.000)[2]{} ------------------------------------------------------------------------ (541,143.67) ------------------------------------------------------------------------ (541.00,144.17)(4.500,-1.000)[2]{} ------------------------------------------------------------------------ (550,142.67) ------------------------------------------------------------------------ (550.00,143.17)(4.500,-1.000)[2]{} ------------------------------------------------------------------------ (559,141.67) ------------------------------------------------------------------------ (559.00,142.17)(5.000,-1.000)[2]{} ------------------------------------------------------------------------ (569,140.67) ------------------------------------------------------------------------ (569.00,141.17)(4.500,-1.000)[2]{} ------------------------------------------------------------------------ (578,139.67) ------------------------------------------------------------------------ (578.00,140.17)(5.000,-1.000)[2]{} ------------------------------------------------------------------------ (588,138.67) ------------------------------------------------------------------------ (588.00,139.17)(4.500,-1.000)[2]{} ------------------------------------------------------------------------ (597,137.67) ------------------------------------------------------------------------ (597.00,138.17)(5.000,-1.000)[2]{} ------------------------------------------------------------------------ (616,136.67) ------------------------------------------------------------------------ (616.00,137.17)(4.500,-1.000)[2]{} ------------------------------------------------------------------------ (625,135.67) ------------------------------------------------------------------------ (625.00,136.17)(5.000,-1.000)[2]{} ------------------------------------------------------------------------ (635,134.67) ------------------------------------------------------------------------ (635.00,135.17)(4.500,-1.000)[2]{} ------------------------------------------------------------------------ (607.0,138.0) ------------------------------------------------------------------------ (654,133.67) ------------------------------------------------------------------------ (654.00,134.17)(4.500,-1.000)[2]{} ------------------------------------------------------------------------ (663,132.67) ------------------------------------------------------------------------ (663.00,133.17)(5.000,-1.000)[2]{} ------------------------------------------------------------------------ (644.0,135.0) ------------------------------------------------------------------------ (682,131.67) ------------------------------------------------------------------------ (682.00,132.17)(5.000,-1.000)[2]{} ------------------------------------------------------------------------ (673.0,133.0) ------------------------------------------------------------------------ (701,130.67) ------------------------------------------------------------------------ (701.00,131.17)(4.500,-1.000)[2]{} ------------------------------------------------------------------------ (710,129.67) ------------------------------------------------------------------------ (710.00,130.17)(5.000,-1.000)[2]{} ------------------------------------------------------------------------ (692.0,132.0) ------------------------------------------------------------------------ (739,128.67) ------------------------------------------------------------------------ (739.00,129.17)(4.500,-1.000)[2]{} ------------------------------------------------------------------------ (720.0,130.0) ------------------------------------------------------------------------ (758,127.67) ------------------------------------------------------------------------ (758.00,128.17)(4.500,-1.000)[2]{} ------------------------------------------------------------------------ (748.0,129.0) ------------------------------------------------------------------------ (786,126.67) ------------------------------------------------------------------------ (786.00,127.17)(4.500,-1.000)[2]{} ------------------------------------------------------------------------ (767.0,128.0) ------------------------------------------------------------------------ (814,125.67) ------------------------------------------------------------------------ (814.00,126.17)(5.000,-1.000)[2]{} ------------------------------------------------------------------------ (795.0,127.0) ------------------------------------------------------------------------ (852,124.67) ------------------------------------------------------------------------ (852.00,125.17)(4.500,-1.000)[2]{} ------------------------------------------------------------------------ (824.0,126.0) ------------------------------------------------------------------------ (890,123.67) ------------------------------------------------------------------------ (890.00,124.17)(4.500,-1.000)[2]{} ------------------------------------------------------------------------ (861.0,125.0) ------------------------------------------------------------------------ (899.0,124.0) ------------------------------------------------------------------------ (220,650) (220,650)(0.279,-20.754)[8]{} (222,501)(1.054,-20.729)[3]{} (226.06,421.80) (227,404)(1.479,-20.703)[2]{} (231.21,359.78) (233.65,339.17) (234,336)(2.743,-20.573)[0]{} (236.34,318.59) (240.19,298.23) (241,295)(4.070,-20.352)[0]{} (244.57,277.94) (245,276)(6.563,-19.690)[0]{} (248,267)(5.702,-19.957)[0]{} (250.55,258.08) (252,253)(9.282,-18.564)[0]{} (255,247)(6.563,-19.690)[0]{} (257.91,238.73) (259,236)(7.708,-19.271)[0]{} (261,231)(12.453,-16.604)[0]{} (264,227)(7.708,-19.271)[0]{} (267.13,220.30) (268,219)(12.453,-16.604)[0]{} (271,215)(11.513,-17.270)[0]{} (273,212)(11.513,-17.270)[0]{} (275,209)(14.676,-14.676)[0]{} (279.52,203.72) (280,203)(11.513,-17.270)[0]{} (282,200)(14.676,-14.676)[0]{} (284,198)(14.676,-14.676)[0]{} (287,195)(14.676,-14.676)[0]{} (289,193)(14.676,-14.676)[0]{} (293.96,189.03) (294,189)(14.676,-14.676)[0]{} (296,187)(14.676,-14.676)[0]{} (298,185)(19.690,-6.563)[0]{} (301,184)(14.676,-14.676)[0]{} (303,182)(14.676,-14.676)[0]{} (305,180)(18.564,-9.282)[0]{} (307,179)(17.270,-11.513)[0]{} (310.35,176.83) (312,176)(18.564,-9.282)[0]{} (314,175)(17.270,-11.513)[0]{} (317,173)(18.564,-9.282)[0]{} (319,172)(18.564,-9.282)[0]{} (321,171)(19.690,-6.563)[0]{} (324,170)(18.564,-9.282)[0]{} (326,169)(18.564,-9.282)[0]{} (328.86,167.57) (330,167)(19.690,-6.563)[0]{} (333,166)(18.564,-9.282)[0]{} (335,165)(18.564,-9.282)[0]{} (337,164)(19.690,-6.563)[0]{} (340,163)(18.564,-9.282)[0]{} (342,162)(18.564,-9.282)[0]{} (344,161)(20.756,0.000)[0]{} (348.08,160.46) (349,160)(18.564,-9.282)[0]{} (351,159)(18.564,-9.282)[0]{} (353,158)(20.756,0.000)[0]{} (356,158)(18.564,-9.282)[0]{} (358,157)(18.564,-9.282)[0]{} (360,156)(20.756,0.000)[0]{} (363,156)(18.564,-9.282)[0]{} (365,155)(18.564,-9.282)[0]{} (367.31,154.00) (370,154)(18.564,-9.282)[0]{} (372,153)(20.756,0.000)[0]{} (374,153)(18.564,-9.282)[0]{} (376,152)(20.756,0.000)[0]{} (379,152)(18.564,-9.282)[0]{} (381,151)(20.756,0.000)[0]{} (383,151)(19.690,-6.563)[0]{} (387.20,150.00) (388,150)(18.564,-9.282)[0]{} (390,149)(20.756,0.000)[0]{} (393,149)(18.564,-9.282)[0]{} (395,148)(20.756,0.000)[0]{} (397,148)(20.756,0.000)[0]{} (399,148)(19.690,-6.563)[0]{} (402,147)(20.756,0.000)[0]{} (404,147)(18.564,-9.282)[0]{} (407.08,146.00) (409,146)(20.756,0.000)[0]{} (411,146)(18.564,-9.282)[0]{} (413,145)(20.756,0.000)[0]{} (416,145)(18.564,-9.282)[0]{} (418,144)(20.756,0.000)[0]{} (420,144)(20.756,0.000)[0]{} (422,144)(19.690,-6.563)[0]{} (425,143)(20.756,0.000)[0]{} (427.21,143.00) (429,143)(19.690,-6.563)[0]{} (432,142)(20.756,0.000)[0]{} (434,142)(20.756,0.000)[0]{} (436,142)(20.756,0.000)[0]{} (439,142)(18.564,-9.282)[0]{} (441,141)(20.756,0.000)[0]{} (443,141)(20.756,0.000)[0]{} (447.43,140.19) (448,140)(20.756,0.000)[0]{} (450,140)(20.756,0.000)[0]{} (452,140)(20.756,0.000)[0]{} (455,140)(18.564,-9.282)[0]{} (457,139)(20.756,0.000)[0]{} (459,139)(20.756,0.000)[0]{} (462,139)(20.756,0.000)[0]{} (464,139)(18.564,-9.282)[0]{} (467.68,138.00) (468,138)(20.756,0.000)[0]{} (471,138)(20.756,0.000)[0]{} (473,138)(20.756,0.000)[0]{} (475,138)(19.690,-6.563)[0]{} (478,137)(20.756,0.000)[0]{} (480,137)(20.756,0.000)[0]{} (482,137)(20.756,0.000)[0]{} (485,137)(20.756,0.000)[0]{} (488.14,136.43) (489,136)(20.756,0.000)[0]{} (491,136)(20.756,0.000)[0]{} (494,136)(20.756,0.000)[0]{} (496,136)(20.756,0.000)[0]{} (498,136)(19.690,-6.563)[0]{} (501,135)(20.756,0.000)[0]{} (503,135)(20.756,0.000)[0]{} (505,135)(20.756,0.000)[0]{} (508.63,135.00) (510,135)(20.756,0.000)[0]{} (512,135)(18.564,-9.282)[0]{} (514,134)(20.756,0.000)[0]{} (517,134)(20.756,0.000)[0]{} (519,134)(20.756,0.000)[0]{} (521,134)(20.756,0.000)[0]{} (524,134)(20.756,0.000)[0]{} (526,134)(18.564,-9.282)[0]{} (528.92,133.00) (531,133)(20.756,0.000)[0]{} (533,133)(20.756,0.000)[0]{} (535,133)(20.756,0.000)[0]{} (537,133)(20.756,0.000)[0]{} (540,133)(20.756,0.000)[0]{} (542,133)(18.564,-9.282)[0]{} (544,132)(20.756,0.000)[0]{} (547,132)(20.756,0.000)[0]{} (549.44,132.00) (551,132)(20.756,0.000)[0]{} (554,132)(20.756,0.000)[0]{} (556,132)(20.756,0.000)[0]{} (558,132)(18.564,-9.282)[0]{} (560,131)(20.756,0.000)[0]{} (563,131)(20.756,0.000)[0]{} (565,131)(20.756,0.000)[0]{} (569.96,131.00) (570,131)(20.756,0.000)[0]{} (572,131)(20.756,0.000)[0]{} (574,131)(20.756,0.000)[0]{} (577,131)(18.564,-9.282)[0]{} (579,130)(20.756,0.000)[0]{} (581,130)(20.756,0.000)[0]{} (583,130)(20.756,0.000)[0]{} (586,130)(20.756,0.000)[0]{} (588,130)(20.756,0.000)[0]{} (590.48,130.00) (593,130)(20.756,0.000)[0]{} (595,130)(20.756,0.000)[0]{} (597,130)(20.756,0.000)[0]{} (600,130)(18.564,-9.282)[0]{} (602,129)(20.756,0.000)[0]{} (604,129)(20.756,0.000)[0]{} (606,129)(20.756,0.000)[0]{} (610.99,129.00) (611,129)(20.756,0.000)[0]{} (613,129)(20.756,0.000)[0]{} (616,129)(20.756,0.000)[0]{} (618,129)(20.756,0.000)[0]{} (620,129)(20.756,0.000)[0]{} (623,129)(18.564,-9.282)[0]{} (625,128)(20.756,0.000)[0]{} (627,128)(20.756,0.000)[0]{} (631.51,128.00) (632,128)(20.756,0.000)[0]{} (634,128)(20.756,0.000)[0]{} (636,128)(20.756,0.000)[0]{} (639,128)(20.756,0.000)[0]{} (641,128)(20.756,0.000)[0]{} (643,128)(20.756,0.000)[0]{} (646,128)(20.756,0.000)[0]{} (648,128)(20.756,0.000)[0]{} (650,128)(18.564,-9.282)[0]{} (652.03,127.00) (655,127)(20.756,0.000)[0]{} (657,127)(20.756,0.000)[0]{} (659,127)(20.756,0.000)[0]{} (662,127)(20.756,0.000)[0]{} (664,127)(20.756,0.000)[0]{} (666,127)(20.756,0.000)[0]{} (669,127)(20.756,0.000)[0]{} (672.79,127.00) (673,127)(20.756,0.000)[0]{} (675,127)(20.756,0.000)[0]{} (678,127)(20.756,0.000)[0]{} (680,127)(18.564,-9.282)[0]{} (682,126)(20.756,0.000)[0]{} (685,126)(20.756,0.000)[0]{} (687,126)(20.756,0.000)[0]{} (689,126)(20.756,0.000)[0]{} (693.31,126.00) (694,126)(20.756,0.000)[0]{} (696,126)(20.756,0.000)[0]{} (698,126)(20.756,0.000)[0]{} (701,126)(20.756,0.000)[0]{} (703,126)(20.756,0.000)[0]{} (705,126)(20.756,0.000)[0]{} (708,126)(20.756,0.000)[0]{} (710,126)(20.756,0.000)[0]{} (714.06,126.00) (715,126)(20.756,0.000)[0]{} (717,126)(18.564,-9.282)[0]{} (719,125)(20.756,0.000)[0]{} (721,125)(20.756,0.000)[0]{} (724,125)(20.756,0.000)[0]{} (726,125)(20.756,0.000)[0]{} (728,125)(20.756,0.000)[0]{} (731,125)(20.756,0.000)[0]{} (734.58,125.00) (735,125)(20.756,0.000)[0]{} (738,125)(20.756,0.000)[0]{} (740,125)(20.756,0.000)[0]{} (742,125)(20.756,0.000)[0]{} (744,125)(20.756,0.000)[0]{} (747,125)(20.756,0.000)[0]{} (749,125)(20.756,0.000)[0]{} (751,125)(20.756,0.000)[0]{} (755.34,125.00) (756,125)(18.564,-9.282)[0]{} (758,124)(20.756,0.000)[0]{} (761,124)(20.756,0.000)[0]{} (763,124)(20.756,0.000)[0]{} (765,124)(20.756,0.000)[0]{} (767,124)(20.756,0.000)[0]{} (770,124)(20.756,0.000)[0]{} (772,124)(20.756,0.000)[0]{} (775.86,124.00) (777,124)(20.756,0.000)[0]{} (779,124)(20.756,0.000)[0]{} (781,124)(20.756,0.000)[0]{} (784,124)(20.756,0.000)[0]{} (786,124)(20.756,0.000)[0]{} (788,124)(20.756,0.000)[0]{} (790,124)(20.756,0.000)[0]{} (793,124)(20.756,0.000)[0]{} (796.61,124.00) (797,124)(20.756,0.000)[0]{} (800,124)(20.756,0.000)[0]{} (802,124)(20.756,0.000)[0]{} (804,124)(19.690,-6.563)[0]{} (807,123)(20.756,0.000)[0]{} (809,123)(20.756,0.000)[0]{} (811,123)(20.756,0.000)[0]{} (813,123)(20.756,0.000)[0]{} (817.21,123.00) (818,123)(20.756,0.000)[0]{} (820,123)(20.756,0.000)[0]{} (823,123)(20.756,0.000)[0]{} (825,123)(20.756,0.000)[0]{} (827,123)(20.756,0.000)[0]{} (830,123)(20.756,0.000)[0]{} (832,123)(20.756,0.000)[0]{} (834,123)(20.756,0.000)[0]{} (837.96,123.00) (839,123)(20.756,0.000)[0]{} (841,123)(20.756,0.000)[0]{} (843,123)(20.756,0.000)[0]{} (846,123)(20.756,0.000)[0]{} (848,123)(20.756,0.000)[0]{} (850,123)(20.756,0.000)[0]{} (853,123)(20.756,0.000)[0]{} (855,123)(20.756,0.000)[0]{} (858.72,123.00) (859,123)(20.756,0.000)[0]{} (862,123)(18.564,-9.282)[0]{} (864,122)(20.756,0.000)[0]{} (866,122)(20.756,0.000)[0]{} (869,122)(20.756,0.000)[0]{} (871,122)(20.756,0.000)[0]{} (873,122)(20.756,0.000)[0]{} (876,122)(20.756,0.000)[0]{} (879.24,122.00) (880,122)(20.756,0.000)[0]{} (882,122)(20.756,0.000)[0]{} (885,122)(20.756,0.000)[0]{} (887,122)(20.756,0.000)[0]{} (889,122)(20.756,0.000)[0]{} (892,122)(20.756,0.000)[0]{} (894,122)(20.756,0.000)[0]{} (896,122)(20.756,0.000)[0]{} (899.99,122.00) (901,122)(20.756,0.000)[0]{} (903,122)(20.756,0.000)[0]{} (905,122)(20.756,0.000)[0]{} (908,122)(20.756,0.000)[0]{} (910,122) \[fig2\] We have investigated also the effect of the finite system size. We plot in Fig. \[fig2\] the environment velocity correlation function for different system sizes, while keeping the rest of parameters constant ($s=2.82$, $\Omega= 25$). We observe a large discrepancy between the hydrodynamic prediction and the simulation results when the box is small. This discrepancy appears at large times and is reduced when the system size increases. This effect can be understood as an self-interaction through periodic boundary conditions. According to Onsager hypothesis for the regression of fluctuations, the correlation of the velocity fluctuations decays in essentially the same way as a localized macroscopic velocity perturbation. A perturbation in the velocity field in the form of an impulse at the origin decays through the formation of a vortex ring whose center evolves diffusively as $\sqrt{t}$ [@esp95c]. In a system with periodic boundary conditions the vortex ring is reintroduced into the system producing an effective slowing down of the velocity field at the origin. Obviously, the larger is the system size, the latter the effect appears and also the smaller is the amplitude of the effect. (974,675)(0,0) =cmr10 at 10pt (220.0,113.0) ------------------------------------------------------------------------ (198,113)[(0,0)\[r\][0.0001]{}]{} (890.0,113.0) ------------------------------------------------------------------------ (220.0,173.0) ------------------------------------------------------------------------ (900.0,173.0) ------------------------------------------------------------------------ (220.0,208.0) ------------------------------------------------------------------------ (900.0,208.0) ------------------------------------------------------------------------ (220.0,233.0) ------------------------------------------------------------------------ (900.0,233.0) ------------------------------------------------------------------------ (220.0,253.0) ------------------------------------------------------------------------ (900.0,253.0) ------------------------------------------------------------------------ (220.0,268.0) ------------------------------------------------------------------------ (900.0,268.0) ------------------------------------------------------------------------ (220.0,282.0) ------------------------------------------------------------------------ (900.0,282.0) ------------------------------------------------------------------------ (220.0,293.0) ------------------------------------------------------------------------ (900.0,293.0) ------------------------------------------------------------------------ (220.0,304.0) ------------------------------------------------------------------------ (900.0,304.0) ------------------------------------------------------------------------ (220.0,313.0) ------------------------------------------------------------------------ (198,313)[(0,0)\[r\][0.001]{}]{} (890.0,313.0) ------------------------------------------------------------------------ (220.0,373.0) ------------------------------------------------------------------------ (900.0,373.0) ------------------------------------------------------------------------ (220.0,408.0) ------------------------------------------------------------------------ (900.0,408.0) ------------------------------------------------------------------------ (220.0,433.0) ------------------------------------------------------------------------ (900.0,433.0) ------------------------------------------------------------------------ (220.0,452.0) ------------------------------------------------------------------------ (900.0,452.0) ------------------------------------------------------------------------ (220.0,468.0) ------------------------------------------------------------------------ (900.0,468.0) ------------------------------------------------------------------------ (220.0,481.0) ------------------------------------------------------------------------ (900.0,481.0) ------------------------------------------------------------------------ (220.0,493.0) ------------------------------------------------------------------------ (900.0,493.0) ------------------------------------------------------------------------ (220.0,503.0) ------------------------------------------------------------------------ (900.0,503.0) ------------------------------------------------------------------------ (220.0,512.0) ------------------------------------------------------------------------ (198,512)[(0,0)\[r\][0.01]{}]{} (890.0,512.0) ------------------------------------------------------------------------ (220.0,573.0) ------------------------------------------------------------------------ (900.0,573.0) ------------------------------------------------------------------------ (220.0,608.0) ------------------------------------------------------------------------ (900.0,608.0) ------------------------------------------------------------------------ (220.0,633.0) ------------------------------------------------------------------------ (900.0,633.0) ------------------------------------------------------------------------ (220.0,652.0) ------------------------------------------------------------------------ (900.0,652.0) ------------------------------------------------------------------------ (220.0,113.0) ------------------------------------------------------------------------ (220,68)[(0,0)[0.1]{}]{} (220.0,632.0) ------------------------------------------------------------------------ (324.0,113.0) ------------------------------------------------------------------------ (324.0,642.0) ------------------------------------------------------------------------ (385.0,113.0) ------------------------------------------------------------------------ (385.0,642.0) ------------------------------------------------------------------------ (428.0,113.0) ------------------------------------------------------------------------ (428.0,642.0) ------------------------------------------------------------------------ (461.0,113.0) ------------------------------------------------------------------------ (461.0,642.0) ------------------------------------------------------------------------ (488.0,113.0) ------------------------------------------------------------------------ (488.0,642.0) ------------------------------------------------------------------------ (512.0,113.0) ------------------------------------------------------------------------ (512.0,642.0) ------------------------------------------------------------------------ (532.0,113.0) ------------------------------------------------------------------------ (532.0,642.0) ------------------------------------------------------------------------ (549.0,113.0) ------------------------------------------------------------------------ (549.0,642.0) ------------------------------------------------------------------------ (565.0,113.0) ------------------------------------------------------------------------ (565,68)[(0,0)[1]{}]{} (565.0,632.0) ------------------------------------------------------------------------ (669.0,113.0) ------------------------------------------------------------------------ (669.0,642.0) ------------------------------------------------------------------------ (730.0,113.0) ------------------------------------------------------------------------ (730.0,642.0) ------------------------------------------------------------------------ (773.0,113.0) ------------------------------------------------------------------------ (773.0,642.0) ------------------------------------------------------------------------ (806.0,113.0) ------------------------------------------------------------------------ (806.0,642.0) ------------------------------------------------------------------------ (833.0,113.0) ------------------------------------------------------------------------ (833.0,642.0) ------------------------------------------------------------------------ (857.0,113.0) ------------------------------------------------------------------------ (857.0,642.0) ------------------------------------------------------------------------ (877.0,113.0) ------------------------------------------------------------------------ (877.0,642.0) ------------------------------------------------------------------------ (894.0,113.0) ------------------------------------------------------------------------ (894.0,642.0) ------------------------------------------------------------------------ (910.0,113.0) ------------------------------------------------------------------------ (910,68)[(0,0)[10]{}]{} (910.0,632.0) ------------------------------------------------------------------------ (220.0,113.0) ------------------------------------------------------------------------ (910.0,113.0) ------------------------------------------------------------------------ (220.0,652.0) ------------------------------------------------------------------------ (45,382)[(0,0)[$ C({\overline t})$]{}]{} (565,23)[(0,0)[${\overline t}$]{}]{} (220.0,113.0) ------------------------------------------------------------------------ (220.00,636.92)(1.590,-0.491)[17]{} ------------------------------------------------------------------------ (220.00,637.17)(28.219,-10.000)[2]{} ------------------------------------------------------------------------ (251.00,626.92)(1.455,-0.494)[27]{} ------------------------------------------------------------------------ (251.00,627.17)(40.412,-15.000)[2]{} ------------------------------------------------------------------------ (294.00,611.92)(1.329,-0.493)[23]{} ------------------------------------------------------------------------ (294.00,612.17)(31.621,-13.000)[2]{} ------------------------------------------------------------------------ (328.00,598.92)(1.142,-0.492)[21]{} ------------------------------------------------------------------------ (328.00,599.17)(24.924,-12.000)[2]{} ------------------------------------------------------------------------ (355.00,586.92)(1.173,-0.491)[17]{} ------------------------------------------------------------------------ (355.00,587.17)(20.883,-10.000)[2]{} ------------------------------------------------------------------------ (378.00,576.92)(1.017,-0.491)[17]{} ------------------------------------------------------------------------ (378.00,577.17)(18.132,-10.000)[2]{} ------------------------------------------------------------------------ (398.00,566.93)(1.019,-0.489)[15]{} ------------------------------------------------------------------------ (398.00,567.17)(16.132,-9.000)[2]{} ------------------------------------------------------------------------ (416.00,557.93)(0.956,-0.488)[13]{} ------------------------------------------------------------------------ (416.00,558.17)(13.236,-8.000)[2]{} ------------------------------------------------------------------------ (431.00,549.93)(0.956,-0.488)[13]{} ------------------------------------------------------------------------ (431.00,550.17)(13.236,-8.000)[2]{} ------------------------------------------------------------------------ (446.00,541.93)(0.824,-0.488)[13]{} ------------------------------------------------------------------------ (446.00,542.17)(11.443,-8.000)[2]{} ------------------------------------------------------------------------ (459.00,533.93)(0.874,-0.485)[11]{} ------------------------------------------------------------------------ (459.00,534.17)(10.369,-7.000)[2]{} ------------------------------------------------------------------------ (471.00,526.93)(0.798,-0.485)[11]{} ------------------------------------------------------------------------ (471.00,527.17)(9.488,-7.000)[2]{} ------------------------------------------------------------------------ (482.00,519.93)(0.721,-0.485)[11]{} ------------------------------------------------------------------------ (482.00,520.17)(8.606,-7.000)[2]{} ------------------------------------------------------------------------ (492.00,512.93)(0.721,-0.485)[11]{} ------------------------------------------------------------------------ (492.00,513.17)(8.606,-7.000)[2]{} ------------------------------------------------------------------------ (502.00,505.93)(0.762,-0.482)[9]{} ------------------------------------------------------------------------ (502.00,506.17)(7.547,-6.000)[2]{} ------------------------------------------------------------------------ (511.00,499.93)(0.671,-0.482)[9]{} ------------------------------------------------------------------------ (511.00,500.17)(6.685,-6.000)[2]{} ------------------------------------------------------------------------ (519.00,493.93)(0.762,-0.482)[9]{} ------------------------------------------------------------------------ (519.00,494.17)(7.547,-6.000)[2]{} ------------------------------------------------------------------------ (528.00,487.93)(0.581,-0.482)[9]{} ------------------------------------------------------------------------ (528.00,488.17)(5.824,-6.000)[2]{} ------------------------------------------------------------------------ (535.00,481.93)(0.671,-0.482)[9]{} ------------------------------------------------------------------------ (535.00,482.17)(6.685,-6.000)[2]{} ------------------------------------------------------------------------ (543.00,475.93)(0.581,-0.482)[9]{} ------------------------------------------------------------------------ (543.00,476.17)(5.824,-6.000)[2]{} ------------------------------------------------------------------------ (550.00,469.93)(0.599,-0.477)[7]{} ------------------------------------------------------------------------ (550.00,470.17)(4.796,-5.000)[2]{} ------------------------------------------------------------------------ (556.00,464.93)(0.581,-0.482)[9]{} ------------------------------------------------------------------------ (556.00,465.17)(5.824,-6.000)[2]{} ------------------------------------------------------------------------ (563.00,458.93)(0.599,-0.477)[7]{} ------------------------------------------------------------------------ (563.00,459.17)(4.796,-5.000)[2]{} ------------------------------------------------------------------------ (569.00,453.93)(0.599,-0.477)[7]{} ------------------------------------------------------------------------ (569.00,454.17)(4.796,-5.000)[2]{} ------------------------------------------------------------------------ (575.00,448.93)(0.487,-0.477)[7]{} ------------------------------------------------------------------------ (575.00,449.17)(3.962,-5.000)[2]{} ------------------------------------------------------------------------ (580.00,443.93)(0.491,-0.482)[9]{} ------------------------------------------------------------------------ (580.00,444.17)(4.962,-6.000)[2]{} ------------------------------------------------------------------------ (586.00,437.93)(0.487,-0.477)[7]{} ------------------------------------------------------------------------ (586.00,438.17)(3.962,-5.000)[2]{} ------------------------------------------------------------------------ (591.00,432.93)(0.487,-0.477)[7]{} ------------------------------------------------------------------------ (591.00,433.17)(3.962,-5.000)[2]{} ------------------------------------------------------------------------ (596.00,427.94)(0.627,-0.468)[5]{} ------------------------------------------------------------------------ (596.00,428.17)(3.755,-4.000)[2]{} ------------------------------------------------------------------------ (601.00,423.93)(0.487,-0.477)[7]{} ------------------------------------------------------------------------ (601.00,424.17)(3.962,-5.000)[2]{} ------------------------------------------------------------------------ (606.60,417.51)(0.468,-0.627)[5]{} ------------------------------------------------------------------------ (605.17,418.75)(4.000,-3.755)[2]{} ------------------------------------------------------------------------ (610.00,413.93)(0.487,-0.477)[7]{} ------------------------------------------------------------------------ (610.00,414.17)(3.962,-5.000)[2]{} ------------------------------------------------------------------------ (615.60,407.51)(0.468,-0.627)[5]{} ------------------------------------------------------------------------ (614.17,408.75)(4.000,-3.755)[2]{} ------------------------------------------------------------------------ (619.60,402.51)(0.468,-0.627)[5]{} ------------------------------------------------------------------------ (618.17,403.75)(4.000,-3.755)[2]{} ------------------------------------------------------------------------ (623.00,398.94)(0.481,-0.468)[5]{} ------------------------------------------------------------------------ (623.00,399.17)(2.962,-4.000)[2]{} ------------------------------------------------------------------------ (627.60,393.51)(0.468,-0.627)[5]{} ------------------------------------------------------------------------ (626.17,394.75)(4.000,-3.755)[2]{} ------------------------------------------------------------------------ (631.00,389.94)(0.481,-0.468)[5]{} ------------------------------------------------------------------------ (631.00,390.17)(2.962,-4.000)[2]{} ------------------------------------------------------------------------ (635.00,385.94)(0.481,-0.468)[5]{} ------------------------------------------------------------------------ (635.00,386.17)(2.962,-4.000)[2]{} ------------------------------------------------------------------------ (639.00,381.94)(0.481,-0.468)[5]{} ------------------------------------------------------------------------ (639.00,382.17)(2.962,-4.000)[2]{} ------------------------------------------------------------------------ (643.61,375.82)(0.447,-0.909)[3]{} ------------------------------------------------------------------------ (642.17,377.41)(3.000,-3.409)[2]{} ------------------------------------------------------------------------ (646.60,371.51)(0.468,-0.627)[5]{} ------------------------------------------------------------------------ (645.17,372.75)(4.000,-3.755)[2]{} ------------------------------------------------------------------------ (650.61,366.37)(0.447,-0.685)[3]{} ------------------------------------------------------------------------ (649.17,367.69)(3.000,-2.685)[2]{} ------------------------------------------------------------------------ (653.00,363.94)(0.481,-0.468)[5]{} ------------------------------------------------------------------------ (653.00,364.17)(2.962,-4.000)[2]{} ------------------------------------------------------------------------ (657.61,358.37)(0.447,-0.685)[3]{} ------------------------------------------------------------------------ (656.17,359.69)(3.000,-2.685)[2]{} ------------------------------------------------------------------------ (660.61,354.37)(0.447,-0.685)[3]{} ------------------------------------------------------------------------ (659.17,355.69)(3.000,-2.685)[2]{} ------------------------------------------------------------------------ (663.61,350.37)(0.447,-0.685)[3]{} ------------------------------------------------------------------------ (662.17,351.69)(3.000,-2.685)[2]{} ------------------------------------------------------------------------ (666.60,346.51)(0.468,-0.627)[5]{} ------------------------------------------------------------------------ (665.17,347.75)(4.000,-3.755)[2]{} ------------------------------------------------------------------------ (670.61,340.82)(0.447,-0.909)[3]{} ------------------------------------------------------------------------ (669.17,342.41)(3.000,-3.409)[2]{} ------------------------------------------------------------------------ (673.61,335.26)(0.447,-1.132)[3]{} ------------------------------------------------------------------------ (672.17,337.13)(3.000,-4.132)[2]{} ------------------------------------------------------------------------ (676,331.17) ------------------------------------------------------------------------ (676.00,332.17)(1.000,-2.000)[2]{} ------------------------------------------------------------------------ (678.61,327.82)(0.447,-0.909)[3]{} ------------------------------------------------------------------------ (677.17,329.41)(3.000,-3.409)[2]{} ------------------------------------------------------------------------ (681.61,322.82)(0.447,-0.909)[3]{} ------------------------------------------------------------------------ (680.17,324.41)(3.000,-3.409)[2]{} ------------------------------------------------------------------------ (684,319.17) ------------------------------------------------------------------------ (684.00,320.17)(1.547,-2.000)[2]{} ------------------------------------------------------------------------ (687.61,315.82)(0.447,-0.909)[3]{} ------------------------------------------------------------------------ (686.17,317.41)(3.000,-3.409)[2]{} ------------------------------------------------------------------------ (690.17,311) ------------------------------------------------------------------------ (689.17,312.55)(2.000,-1.547)[2]{} ------------------------------------------------------------------------ (692.61,308.37)(0.447,-0.685)[3]{} ------------------------------------------------------------------------ (691.17,309.69)(3.000,-2.685)[2]{} ------------------------------------------------------------------------ (695.17,303) ------------------------------------------------------------------------ (694.17,305.13)(2.000,-2.132)[2]{} ------------------------------------------------------------------------ (697.61,300.37)(0.447,-0.685)[3]{} ------------------------------------------------------------------------ (696.17,301.69)(3.000,-2.685)[2]{} ------------------------------------------------------------------------ (700.17,296) ------------------------------------------------------------------------ (699.17,297.55)(2.000,-1.547)[2]{} ------------------------------------------------------------------------ (702.61,292.82)(0.447,-0.909)[3]{} ------------------------------------------------------------------------ (701.17,294.41)(3.000,-3.409)[2]{} ------------------------------------------------------------------------ (705.17,288) ------------------------------------------------------------------------ (704.17,289.55)(2.000,-1.547)[2]{} ------------------------------------------------------------------------ (707,286.17) ------------------------------------------------------------------------ (707.00,287.17)(1.547,-2.000)[2]{} ------------------------------------------------------------------------ (710.17,282) ------------------------------------------------------------------------ (709.17,284.13)(2.000,-2.132)[2]{} ------------------------------------------------------------------------ (712.17,279) ------------------------------------------------------------------------ (711.17,280.55)(2.000,-1.547)[2]{} ------------------------------------------------------------------------ (714.17,275) ------------------------------------------------------------------------ (713.17,277.13)(2.000,-2.132)[2]{} ------------------------------------------------------------------------ (716.61,272.37)(0.447,-0.685)[3]{} ------------------------------------------------------------------------ (715.17,273.69)(3.000,-2.685)[2]{} ------------------------------------------------------------------------ (719.17,267) ------------------------------------------------------------------------ (718.17,269.13)(2.000,-2.132)[2]{} ------------------------------------------------------------------------ (721.17,262) ------------------------------------------------------------------------ (720.17,264.72)(2.000,-2.717)[2]{} ------------------------------------------------------------------------ (723.17,257) ------------------------------------------------------------------------ (722.17,259.72)(2.000,-2.717)[2]{} ------------------------------------------------------------------------ (725.17,254) ------------------------------------------------------------------------ (724.17,255.55)(2.000,-1.547)[2]{} ------------------------------------------------------------------------ (727.17,249) ------------------------------------------------------------------------ (726.17,251.72)(2.000,-2.717)[2]{} ------------------------------------------------------------------------ (729.17,243) ------------------------------------------------------------------------ (728.17,246.30)(2.000,-3.302)[2]{} ------------------------------------------------------------------------ (731.17,238) ------------------------------------------------------------------------ (730.17,240.72)(2.000,-2.717)[2]{} ------------------------------------------------------------------------ (733.17,235) ------------------------------------------------------------------------ (732.17,236.55)(2.000,-1.547)[2]{} ------------------------------------------------------------------------ (735,235.17) ------------------------------------------------------------------------ (735.00,234.17)(1.000,2.000)[2]{} ------------------------------------------------------------------------ (737,235.17) ------------------------------------------------------------------------ (737.00,236.17)(1.000,-2.000)[2]{} ------------------------------------------------------------------------ (739.17,232) ------------------------------------------------------------------------ (738.17,233.55)(2.000,-1.547)[2]{} ------------------------------------------------------------------------ (741.17,227) ------------------------------------------------------------------------ (740.17,229.72)(2.000,-2.717)[2]{} ------------------------------------------------------------------------ (743.17,218) ------------------------------------------------------------------------ (742.17,223.06)(2.000,-5.056)[2]{} ------------------------------------------------------------------------ (745.17,211) ------------------------------------------------------------------------ (744.17,214.89)(2.000,-3.887)[2]{} ------------------------------------------------------------------------ (746.67,208) ------------------------------------------------------------------------ (746.17,209.50)(1.000,-1.500)[2]{} ------------------------------------------------------------------------ (748.17,205) ------------------------------------------------------------------------ (747.17,206.55)(2.000,-1.547)[2]{} ------------------------------------------------------------------------ (750.17,200) ------------------------------------------------------------------------ (749.17,202.72)(2.000,-2.717)[2]{} ------------------------------------------------------------------------ (754.17,197) ------------------------------------------------------------------------ (753.17,198.55)(2.000,-1.547)[2]{} ------------------------------------------------------------------------ (756,196.67) ------------------------------------------------------------------------ (756.00,196.17)(0.500,1.000)[2]{} ------------------------------------------------------------------------ (757,196.67) ------------------------------------------------------------------------ (757.00,197.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (759.17,193) ------------------------------------------------------------------------ (758.17,195.13)(2.000,-2.132)[2]{} ------------------------------------------------------------------------ (760.67,191) ------------------------------------------------------------------------ (760.17,192.00)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (762.17,184) ------------------------------------------------------------------------ (761.17,187.89)(2.000,-3.887)[2]{} ------------------------------------------------------------------------ (764.17,184) ------------------------------------------------------------------------ (763.17,184.00)(2.000,1.547)[2]{} ------------------------------------------------------------------------ (765.67,182) ------------------------------------------------------------------------ (765.17,184.50)(1.000,-2.500)[2]{} ------------------------------------------------------------------------ (767.17,178) ------------------------------------------------------------------------ (766.17,180.13)(2.000,-2.132)[2]{} ------------------------------------------------------------------------ (769,177.67) ------------------------------------------------------------------------ (769.00,177.17)(0.500,1.000)[2]{} ------------------------------------------------------------------------ (770,177.67) ------------------------------------------------------------------------ (770.00,178.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (771.67,178) ------------------------------------------------------------------------ (771.17,178.00)(1.000,1.500)[2]{} ------------------------------------------------------------------------ (773.17,177) ------------------------------------------------------------------------ (772.17,179.13)(2.000,-2.132)[2]{} ------------------------------------------------------------------------ (774.67,169) ------------------------------------------------------------------------ (774.17,173.00)(1.000,-4.000)[2]{} ------------------------------------------------------------------------ (776,167.17) ------------------------------------------------------------------------ (776.00,168.17)(1.000,-2.000)[2]{} ------------------------------------------------------------------------ (777.67,160) ------------------------------------------------------------------------ (777.17,163.50)(1.000,-3.500)[2]{} ------------------------------------------------------------------------ (779,159.67) ------------------------------------------------------------------------ (779.00,159.17)(1.000,1.000)[2]{} ------------------------------------------------------------------------ (781,160.67) ------------------------------------------------------------------------ (781.00,160.17)(0.500,1.000)[2]{} ------------------------------------------------------------------------ (782.17,155) ------------------------------------------------------------------------ (781.17,158.89)(2.000,-3.887)[2]{} ------------------------------------------------------------------------ (783.67,155) ------------------------------------------------------------------------ (783.17,155.00)(1.000,1.500)[2]{} ------------------------------------------------------------------------ (785,157.67) ------------------------------------------------------------------------ (785.00,157.17)(1.000,1.000)[2]{} ------------------------------------------------------------------------ (786.67,155) ------------------------------------------------------------------------ (786.17,157.00)(1.000,-2.000)[2]{} ------------------------------------------------------------------------ (787.67,155) ------------------------------------------------------------------------ (787.17,155.00)(1.000,1.000)[2]{} ------------------------------------------------------------------------ (789.17,157) ------------------------------------------------------------------------ (788.17,157.00)(2.000,1.547)[2]{} ------------------------------------------------------------------------ (790.67,158) ------------------------------------------------------------------------ (790.17,159.00)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (752.0,200.0) ------------------------------------------------------------------------ (793.17,153) ------------------------------------------------------------------------ (792.17,155.72)(2.000,-2.717)[2]{} ------------------------------------------------------------------------ (792.0,158.0) (795.67,148) ------------------------------------------------------------------------ (795.17,150.50)(1.000,-2.500)[2]{} ------------------------------------------------------------------------ (795.0,153.0) (798.67,148) ------------------------------------------------------------------------ (798.17,148.00)(1.000,6.000)[2]{} ------------------------------------------------------------------------ (799.67,157) ------------------------------------------------------------------------ (799.17,158.50)(1.000,-1.500)[2]{} ------------------------------------------------------------------------ (800.67,157) ------------------------------------------------------------------------ (800.17,157.00)(1.000,2.000)[2]{} ------------------------------------------------------------------------ (802.17,158) ------------------------------------------------------------------------ (801.17,159.55)(2.000,-1.547)[2]{} ------------------------------------------------------------------------ (803.67,153) ------------------------------------------------------------------------ (803.17,155.50)(1.000,-2.500)[2]{} ------------------------------------------------------------------------ (804.67,153) ------------------------------------------------------------------------ (804.17,153.00)(1.000,2.000)[2]{} ------------------------------------------------------------------------ (805.67,151) ------------------------------------------------------------------------ (805.17,154.00)(1.000,-3.000)[2]{} ------------------------------------------------------------------------ (807,151.17) ------------------------------------------------------------------------ (807.00,150.17)(1.000,2.000)[2]{} ------------------------------------------------------------------------ (809,151.67) ------------------------------------------------------------------------ (809.00,152.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (810,151.67) ------------------------------------------------------------------------ (810.00,151.17)(0.500,1.000)[2]{} ------------------------------------------------------------------------ (810.67,149) ------------------------------------------------------------------------ (810.17,151.00)(1.000,-2.000)[2]{} ------------------------------------------------------------------------ (811.67,144) ------------------------------------------------------------------------ (811.17,146.50)(1.000,-2.500)[2]{} ------------------------------------------------------------------------ (813.17,141) ------------------------------------------------------------------------ (812.17,142.55)(2.000,-1.547)[2]{} ------------------------------------------------------------------------ (814.67,136) ------------------------------------------------------------------------ (814.17,138.50)(1.000,-2.500)[2]{} ------------------------------------------------------------------------ (815.67,136) ------------------------------------------------------------------------ (815.17,136.00)(1.000,1.500)[2]{} ------------------------------------------------------------------------ (816.67,137) ------------------------------------------------------------------------ (816.17,138.00)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (817.67,137) ------------------------------------------------------------------------ (817.17,137.00)(1.000,1.500)[2]{} ------------------------------------------------------------------------ (818.67,137) ------------------------------------------------------------------------ (818.17,138.50)(1.000,-1.500)[2]{} ------------------------------------------------------------------------ (819.67,134) ------------------------------------------------------------------------ (819.17,135.50)(1.000,-1.500)[2]{} ------------------------------------------------------------------------ (820.67,134) ------------------------------------------------------------------------ (820.17,134.00)(1.000,1.000)[2]{} ------------------------------------------------------------------------ (822.17,136) ------------------------------------------------------------------------ (821.17,136.00)(2.000,1.547)[2]{} ------------------------------------------------------------------------ (823.67,139) ------------------------------------------------------------------------ (823.17,139.00)(1.000,5.000)[2]{} ------------------------------------------------------------------------ (824.67,149) ------------------------------------------------------------------------ (824.17,149.00)(1.000,3.000)[2]{} ------------------------------------------------------------------------ (826,153.67) ------------------------------------------------------------------------ (826.00,154.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (797.0,148.0) ------------------------------------------------------------------------ (827.67,150) ------------------------------------------------------------------------ (827.17,152.00)(1.000,-2.000)[2]{} ------------------------------------------------------------------------ (828.67,148) ------------------------------------------------------------------------ (828.17,149.00)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (829.67,141) ------------------------------------------------------------------------ (829.17,144.50)(1.000,-3.500)[2]{} ------------------------------------------------------------------------ (830.67,141) ------------------------------------------------------------------------ (830.17,141.00)(1.000,2.500)[2]{} ------------------------------------------------------------------------ (831.67,143) ------------------------------------------------------------------------ (831.17,144.50)(1.000,-1.500)[2]{} ------------------------------------------------------------------------ (833,141.67) ------------------------------------------------------------------------ (833.00,142.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (833.67,134) ------------------------------------------------------------------------ (833.17,138.00)(1.000,-4.000)[2]{} ------------------------------------------------------------------------ (834.67,134) ------------------------------------------------------------------------ (834.17,134.00)(1.000,4.500)[2]{} ------------------------------------------------------------------------ (835.67,139) ------------------------------------------------------------------------ (835.17,141.00)(1.000,-2.000)[2]{} ------------------------------------------------------------------------ (836.67,133) ------------------------------------------------------------------------ (836.17,136.00)(1.000,-3.000)[2]{} ------------------------------------------------------------------------ (837.67,133) ------------------------------------------------------------------------ (837.17,133.00)(1.000,2.000)[2]{} ------------------------------------------------------------------------ (838.67,133) ------------------------------------------------------------------------ (838.17,135.00)(1.000,-2.000)[2]{} ------------------------------------------------------------------------ (839.67,126) ------------------------------------------------------------------------ (839.17,129.50)(1.000,-3.500)[2]{} ------------------------------------------------------------------------ (840.67,123) ------------------------------------------------------------------------ (840.17,124.50)(1.000,-1.500)[2]{} ------------------------------------------------------------------------ (841.67,123) ------------------------------------------------------------------------ (841.17,123.00)(1.000,2.000)[2]{} ------------------------------------------------------------------------ (843,125.67) ------------------------------------------------------------------------ (843.00,126.17)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (843.67,126) ------------------------------------------------------------------------ (843.17,126.00)(1.000,1.000)[2]{} ------------------------------------------------------------------------ (844.67,119) ------------------------------------------------------------------------ (844.17,123.50)(1.000,-4.500)[2]{} ------------------------------------------------------------------------ (845.67,119) ------------------------------------------------------------------------ (845.17,119.00)(1.000,2.500)[2]{} ------------------------------------------------------------------------ (827.0,154.0) (847.0,113.0) ------------------------------------------------------------------------ (220.00,640.00) (239.95,634.29) (241,634)(19.271,-7.708)[0]{} (259.52,627.44) (261,627)(19.690,-6.563)[0]{} (270,624)(19.434,-7.288)[0]{} (279.11,620.58) (286,618)(20.136,-5.034)[0]{} (298.73,613.97) (301,613)(19.077,-8.176)[0]{} (308,610)(19.957,-5.702)[0]{} (318.03,606.48) (321,605)(19.957,-5.702)[0]{} (328,603)(19.271,-7.708)[0]{} (337.27,598.86) (339,598)(19.690,-6.563)[0]{} (345,596)(19.271,-7.708)[0]{} (350,594)(19.271,-7.708)[0]{} (356.60,591.36) (360,590)(19.271,-7.708)[0]{} (365,588)(18.564,-9.282)[0]{} (369,586)(19.271,-7.708)[0]{} (375.66,583.17) (378,582)(18.564,-9.282)[0]{} (382,580)(18.564,-9.282)[0]{} (386,578)(18.564,-9.282)[0]{} (390,576)(20.136,-5.034)[0]{} (394.54,574.73) (398,573)(18.564,-9.282)[0]{} (402,571)(17.270,-11.513)[0]{} (405,569)(20.136,-5.034)[0]{} (409,568)(17.270,-11.513)[0]{} (412.96,565.52) (416,564)(19.690,-6.563)[0]{} (419,563)(17.270,-11.513)[0]{} (422,561)(17.270,-11.513)[0]{} (425,559)(19.690,-6.563)[0]{} (428,558)(19.690,-6.563)[0]{} (431.55,556.63) (434,555)(17.270,-11.513)[0]{} (437,553)(19.690,-6.563)[0]{} (440,552)(17.270,-11.513)[0]{} (443,550)(19.690,-6.563)[0]{} (446,549)(14.676,-14.676)[0]{} (449.37,546.54) (451,546)(17.270,-11.513)[0]{} (454,544)(18.564,-9.282)[0]{} (456,543)(19.690,-6.563)[0]{} (459,542)(14.676,-14.676)[0]{} (461,540)(19.690,-6.563)[0]{} (464,539)(14.676,-14.676)[0]{} (467.09,536.46) (468,536)(19.690,-6.563)[0]{} (471,535)(14.676,-14.676)[0]{} (473,533)(18.564,-9.282)[0]{} (475,532)(18.564,-9.282)[0]{} (477,531)(17.270,-11.513)[0]{} (480,529)(20.756,0.000)[0]{} (482,529)(14.676,-14.676)[0]{} (484.75,526.62) (486,526)(14.676,-14.676)[0]{} (488,524)(18.564,-9.282)[0]{} (490,523)(18.564,-9.282)[0]{} (492,522)(18.564,-9.282)[0]{} (494,521)(14.676,-14.676)[0]{} (496,519)(18.564,-9.282)[0]{} (498,518)(18.564,-9.282)[0]{} (500,517)(20.756,0.000)[0]{} (502.37,516.63) (504,515)(14.676,-14.676)[0]{} (506,513)(14.676,-14.676)[0]{} (507,512)(18.564,-9.282)[0]{} (509,511)(18.564,-9.282)[0]{} (511,510)(18.564,-9.282)[0]{} (513,509)(14.676,-14.676)[0]{} (514,508)(14.676,-14.676)[0]{} (516,506)(18.564,-9.282)[0]{} (518,505)(20.756,0.000)[0]{} (519.01,504.99) (521,503)(18.564,-9.282)[0]{} (523,502)(14.676,-14.676)[0]{} (524,501)(18.564,-9.282)[0]{} (526,500)(18.564,-9.282)[0]{} (528,499)(14.676,-14.676)[0]{} (529,498)(18.564,-9.282)[0]{} (531,497)(14.676,-14.676)[0]{} (532,496)(18.564,-9.282)[0]{} (534,495)(20.756,0.000)[0]{} (536.08,493.92) (537,493)(9.282,-18.564)[0]{} (538,491)(18.564,-9.282)[0]{} (540,490)(14.676,-14.676)[0]{} (541,489)(18.564,-9.282)[0]{} (543,488)(14.676,-14.676)[0]{} (544,487)(14.676,-14.676)[0]{} (545,486)(18.564,-9.282)[0]{} (547,485)(14.676,-14.676)[0]{} (548,484)(18.564,9.282)[0]{} (550.83,482.52) (551,482)(14.676,-14.676)[0]{} (552,481)(18.564,-9.282)[0]{} (554,480)(14.676,-14.676)[0]{} (555,479)(14.676,-14.676)[0]{} (556,478)(18.564,-9.282)[0]{} (558,477)(14.676,-14.676)[0]{} (559,476)(14.676,-14.676)[0]{} (560,475)(14.676,-14.676)[0]{} (561,474)(18.564,9.282)[0]{} (563,475)(6.563,-19.690)[0]{} (564,472)(14.676,-14.676)[0]{} (565.44,471.00) (566,471)(14.676,-14.676)[0]{} (567,470)(18.564,-9.282)[0]{} (569,469)(14.676,-14.676)[0]{} (570,468)(14.676,-14.676)[0]{} (571,467)(14.676,-14.676)[0]{} (572,466)(14.676,-14.676)[0]{} (573,465)(18.564,9.282)[0]{} (575,466)(6.563,-19.690)[0]{} (576,463)(14.676,-14.676)[0]{} (577,462)(14.676,-14.676)[0]{} (578,461)(20.756,0.000)[0]{} (579,461)(14.676,-14.676)[0]{} (580.17,459.83) (581,459)(14.676,-14.676)[0]{} (582,458)(18.564,-9.282)[0]{} (584,457)(20.756,0.000)[0]{} (585,457)(20.756,0.000)[0]{} (586,457)(9.282,-18.564)[0]{} (587,455)(14.676,-14.676)[0]{} (588,454)(14.676,-14.676)[0]{} (589,453)(14.676,-14.676)[0]{} (590,452)(14.676,-14.676)[0]{} (591,451)(20.756,0.000)[0]{} (592,451)(14.676,-14.676)[0]{} (593,450)(14.676,-14.676)[0]{} (594,449)(14.676,-14.676)[0]{} (595.80,448.00) (596,448)(9.282,-18.564)[0]{} (597,446)(14.676,-14.676)[0]{} (598,445)(20.756,0.000)[0]{} (599,445)(14.676,-14.676)[0]{} (600,444)(14.676,-14.676)[0]{} (601,443)(14.676,-14.676)[0]{} (602,442)(14.676,-14.676)[0]{} (603,441)(14.676,-14.676)[0]{} (604,440)(14.676,-14.676)[0]{} (605,439)(14.676,14.676)[0]{} (606,440)(9.282,-18.564)[0]{} (607,438)(14.676,-14.676)[0]{} (608,437)(0.000,-20.756)[0]{} (608,436)(20.756,0.000)[0]{} (609.25,435.75) (610,435)(14.676,-14.676)[0]{} (611,434)(14.676,-14.676)[0]{} (612,433)(20.756,0.000)[0]{} (613,433)(14.676,-14.676)[0]{} (614,432)(20.756,0.000)[0]{} (615,432)(9.282,-18.564)[0]{} (616,430)(20.756,0.000)[0]{} (617,430)(0.000,-20.756)[0]{} (617,429)(20.756,0.000)[0]{} (618,429)(14.676,-14.676)[0]{} (619,428)(14.676,-14.676)[0]{} (620,427)(20.756,0.000)[0]{} (621,427)(14.676,-14.676)[0]{} (622,426)(14.676,-14.676)[0]{} (623,425)(0.000,-20.756)[0]{} (623.56,424.00) (624,424)(14.676,-14.676)[0]{} (625,423)(14.676,-14.676)[0]{} (626,422)(20.756,0.000)[0]{} (627,422)(0.000,-20.756)[0]{} (627,421)(20.756,0.000)[0]{} (628,421)(14.676,-14.676)[0]{} (629,420)(14.676,-14.676)[0]{} (630,419)(14.676,-14.676)[0]{} (631,418)(0.000,-20.756)[0]{} (637.54,410.46) (639,409)(14.676,-14.676)[0]{} (652.21,395.79) (653,395)(13.668,-15.620)[0]{} (660,387)(14.676,-14.676)[0]{} (666.37,380.63) (673,374)(12.064,-16.889)[0]{} (679.97,365.03) (684,361)(13.508,-15.759)[0]{} (693.39,349.25) (695,347)(13.287,-15.945)[0]{} (700,341)(12.064,-16.889)[0]{} (706.11,332.89) (710,329)(10.298,-18.021)[0]{} (714,322)(14.676,-14.676)[0]{} (719.07,316.90) (723,311)(10.298,-18.021)[0]{} (730.50,299.62) (731,299)(11.513,-17.270)[0]{} (735,293)(12.966,-16.207)[0]{} (742.96,283.05) (743,283)(12.966,-16.207)[0]{} (747,278)(9.282,-18.564)[0]{} (750,272)(12.966,-16.207)[0]{} (754.71,266.05) (757,263)(12.966,-16.207)[0]{} (761,258)(14.676,-14.676)[0]{} (764,255)(10.679,-17.798)[0]{} (767.24,249.60) (770,245)(14.676,-14.676)[0]{} (773,242)(12.453,-16.604)[0]{} (776,238)(17.270,-11.513)[0]{} (780.52,233.97) (782,232)(14.676,-14.676)[0]{} (785,229)(14.676,-14.676)[0]{} (788,226)(14.676,-14.676)[0]{} (791,223)(11.513,-17.270)[0]{} (794.63,218.91) (796,218)(17.270,-11.513)[0]{} (799,216)(20.756,0.000)[0]{} (801,216)(14.676,-14.676)[0]{} (804,213)(20.756,0.000)[0]{} (806,213)(19.690,-6.563)[0]{} (809,212)(18.564,-9.282)[0]{} (812.67,210.17) (813,210)(17.270,11.513)[0]{} (816,212)(9.282,-18.564)[0]{} (818,208)(14.676,-14.676)[0]{} (820,206)(14.676,-14.676)[0]{} (822,204)(19.690,-6.563)[0]{} (826.56,199.88) (827,199)(18.564,9.282)[0]{} (829,200)(14.676,-14.676)[0]{} (831,198)(6.563,-19.690)[0]{} (833,192)(20.756,0.000)[0]{} (835,192)(18.564,-9.282)[0]{} (837,191)(11.513,17.270)[0]{} (839.54,194.00) (841,194)(18.564,-9.282)[0]{} (843,193)(20.756,0.000)[0]{} (845,193)(18.564,9.282)[0]{} (847,194)(20.756,0.000)[0]{} (849,194)(14.676,-14.676)[0]{} (851,192)(20.756,0.000)[0]{} (852,192)(14.676,-14.676)[0]{} (854,190)(20.756,0.000)[0]{} (857.94,189.03) (858,189)(14.676,14.676)[0]{} (859,190)(20.756,0.000)[0]{} (861,190)(18.564,-9.282)[0]{} (863,189)(9.282,18.564)[0]{} (864,191)(20.756,0.000)[0]{} (866,191)(20.756,0.000)[0]{} (868,191)(20.756,0.000)[0]{} (869,191)(18.564,9.282)[0]{} (871,192)(11.513,-17.270)[0]{} (873.87,187.25) (874,187)(14.676,-14.676)[0]{} (876,185)(9.282,-18.564)[0]{} (877,183)(9.282,-18.564)[0]{} (879,179)(14.676,14.676)[0]{} (880,180)(14.676,-14.676)[0]{} (882,178)(20.756,0.000)[0]{} (883,178)(20.756,0.000)[0]{} (885.90,174.41) (886,174)(14.676,14.676)[0]{} (888,176)(9.282,-18.564)[0]{} (889,174)(20.756,0.000)[0]{} (890,174)(14.676,-14.676)[0]{} (892,172)(9.282,-18.564)[0]{} (893,170)(18.564,-9.282)[0]{} (895,169)(14.676,-14.676)[0]{} (896,168)(5.034,-20.136)[0]{} (898.43,164.00) (899,164)(20.756,0.000)[0]{} (900,164)(6.563,19.690)[0]{} (901,167)(11.513,-17.270)[0]{} (903,164)(9.282,-18.564)[0]{} (904,162)(20.756,0.000)[0]{} (905,162)(20.756,0.000)[0]{} (906,162)(11.513,-17.270)[0]{} (908,159)(14.676,-14.676)[0]{} (909,158)(14.676,-14.676)[0]{} (910,157) (220.00,639.08)(1.873,-0.516)[11]{} ------------------------------------------------------------------------ (220.00,639.34)(24.866,-9.000)[2]{} ------------------------------------------------------------------------ (251.00,630.09)(1.484,-0.508)[23]{} ------------------------------------------------------------------------ (251.00,630.34)(37.825,-15.000)[2]{} ------------------------------------------------------------------------ (294.00,615.08)(1.485,-0.511)[17]{} ------------------------------------------------------------------------ (294.00,615.34)(28.880,-12.000)[2]{} ------------------------------------------------------------------------ (328.00,603.08)(1.287,-0.512)[15]{} ------------------------------------------------------------------------ (328.00,603.34)(22.509,-11.000)[2]{} ------------------------------------------------------------------------ (355.00,592.08)(1.211,-0.514)[13]{} ------------------------------------------------------------------------ (355.00,592.34)(18.766,-10.000)[2]{} ------------------------------------------------------------------------ (378.00,582.08)(1.179,-0.516)[11]{} ------------------------------------------------------------------------ (378.00,582.34)(15.895,-9.000)[2]{} ------------------------------------------------------------------------ (398.00,573.08)(1.212,-0.520)[9]{} ------------------------------------------------------------------------ (398.00,573.34)(13.849,-8.000)[2]{} ------------------------------------------------------------------------ (416.00,565.08)(0.993,-0.520)[9]{} ------------------------------------------------------------------------ (416.00,565.34)(11.472,-8.000)[2]{} ------------------------------------------------------------------------ (431.00,557.08)(1.176,-0.526)[7]{} ------------------------------------------------------------------------ (431.00,557.34)(11.027,-7.000)[2]{} ------------------------------------------------------------------------ (446.00,550.08)(1.000,-0.526)[7]{} ------------------------------------------------------------------------ (446.00,550.34)(9.501,-7.000)[2]{} ------------------------------------------------------------------------ (459.00,543.08)(0.913,-0.526)[7]{} ------------------------------------------------------------------------ (459.00,543.34)(8.738,-7.000)[2]{} ------------------------------------------------------------------------ (471.00,536.07)(1.020,-0.536)[5]{} ------------------------------------------------------------------------ (471.00,536.34)(7.541,-6.000)[2]{} ------------------------------------------------------------------------ (482.00,530.07)(0.909,-0.536)[5]{} ------------------------------------------------------------------------ (482.00,530.34)(6.817,-6.000)[2]{} ------------------------------------------------------------------------ (492.00,524.07)(0.909,-0.536)[5]{} ------------------------------------------------------------------------ (492.00,524.34)(6.817,-6.000)[2]{} ------------------------------------------------------------------------ (502.00,518.06)(1.096,-0.560)[3]{} ------------------------------------------------------------------------ (502.00,518.34)(5.596,-5.000)[2]{} ------------------------------------------------------------------------ (511.00,513.07)(0.685,-0.536)[5]{} ------------------------------------------------------------------------ (511.00,513.34)(5.371,-6.000)[2]{} ------------------------------------------------------------------------ (519.00,507.06)(1.096,-0.560)[3]{} ------------------------------------------------------------------------ (519.00,507.34)(5.596,-5.000)[2]{} ------------------------------------------------------------------------ (528.00,502.06)(0.760,-0.560)[3]{} ------------------------------------------------------------------------ (528.00,502.34)(4.260,-5.000)[2]{} ------------------------------------------------------------------------ (535.00,497.06)(0.928,-0.560)[3]{} ------------------------------------------------------------------------ (535.00,497.34)(4.928,-5.000)[2]{} ------------------------------------------------------------------------ (543,490.34) ------------------------------------------------------------------------ (543.00,492.34)(3.679,-4.000)[2]{} ------------------------------------------------------------------------ (550.00,488.06)(0.592,-0.560)[3]{} ------------------------------------------------------------------------ (550.00,488.34)(3.592,-5.000)[2]{} ------------------------------------------------------------------------ (556,481.34) ------------------------------------------------------------------------ (556.00,483.34)(3.679,-4.000)[2]{} ------------------------------------------------------------------------ (563.00,479.06)(0.592,-0.560)[3]{} ------------------------------------------------------------------------ (563.00,479.34)(3.592,-5.000)[2]{} ------------------------------------------------------------------------ (569,472.34) ------------------------------------------------------------------------ (569.00,474.34)(3.094,-4.000)[2]{} ------------------------------------------------------------------------ (575,468.34) ------------------------------------------------------------------------ (575.00,470.34)(2.509,-4.000)[2]{} ------------------------------------------------------------------------ (580,464.34) ------------------------------------------------------------------------ (580.00,466.34)(3.094,-4.000)[2]{} ------------------------------------------------------------------------ (586,460.34) ------------------------------------------------------------------------ (586.00,462.34)(2.509,-4.000)[2]{} ------------------------------------------------------------------------ (591,456.34) ------------------------------------------------------------------------ (591.00,458.34)(2.509,-4.000)[2]{} ------------------------------------------------------------------------ (596,452.84) ------------------------------------------------------------------------ (596.00,454.34)(2.500,-3.000)[2]{} ------------------------------------------------------------------------ (601,449.34) ------------------------------------------------------------------------ (601.00,451.34)(2.509,-4.000)[2]{} ------------------------------------------------------------------------ (606,445.34) ------------------------------------------------------------------------ (606.00,447.34)(2.000,-4.000)[2]{} ------------------------------------------------------------------------ (610,441.84) ------------------------------------------------------------------------ (610.00,443.34)(2.500,-3.000)[2]{} ------------------------------------------------------------------------ (615,438.34) ------------------------------------------------------------------------ (615.00,440.34)(2.000,-4.000)[2]{} ------------------------------------------------------------------------ (619,434.84) ------------------------------------------------------------------------ (619.00,436.34)(2.000,-3.000)[2]{} ------------------------------------------------------------------------ (623,431.34) ------------------------------------------------------------------------ (623.00,433.34)(2.000,-4.000)[2]{} ------------------------------------------------------------------------ (627,427.84) ------------------------------------------------------------------------ (627.00,429.34)(2.000,-3.000)[2]{} ------------------------------------------------------------------------ (631,424.84) ------------------------------------------------------------------------ (631.00,426.34)(2.000,-3.000)[2]{} ------------------------------------------------------------------------ (635,421.34) ------------------------------------------------------------------------ (635.00,423.34)(2.000,-4.000)[2]{} ------------------------------------------------------------------------ (639,417.84) ------------------------------------------------------------------------ (639.00,419.34)(2.000,-3.000)[2]{} ------------------------------------------------------------------------ (643,414.84) ------------------------------------------------------------------------ (643.00,416.34)(1.500,-3.000)[2]{} ------------------------------------------------------------------------ (646,411.84) ------------------------------------------------------------------------ (646.00,413.34)(2.000,-3.000)[2]{} ------------------------------------------------------------------------ (650,408.84) ------------------------------------------------------------------------ (650.00,410.34)(1.500,-3.000)[2]{} ------------------------------------------------------------------------ (653,405.84) ------------------------------------------------------------------------ (653.00,407.34)(2.000,-3.000)[2]{} ------------------------------------------------------------------------ (657,402.84) ------------------------------------------------------------------------ (657.00,404.34)(1.500,-3.000)[2]{} ------------------------------------------------------------------------ (660,400.34) ------------------------------------------------------------------------ (660.00,401.34)(1.500,-2.000)[2]{} ------------------------------------------------------------------------ (663,397.84) ------------------------------------------------------------------------ (663.00,399.34)(1.500,-3.000)[2]{} ------------------------------------------------------------------------ (666,394.84) ------------------------------------------------------------------------ (666.00,396.34)(2.000,-3.000)[2]{} ------------------------------------------------------------------------ (670,391.84) ------------------------------------------------------------------------ (670.00,393.34)(1.500,-3.000)[2]{} ------------------------------------------------------------------------ (673,388.84) ------------------------------------------------------------------------ (673.00,390.34)(1.500,-3.000)[2]{} ------------------------------------------------------------------------ (676,386.34) ------------------------------------------------------------------------ (676.00,387.34)(1.000,-2.000)[2]{} ------------------------------------------------------------------------ (678,383.84) ------------------------------------------------------------------------ (678.00,385.34)(1.500,-3.000)[2]{} ------------------------------------------------------------------------ (681,380.84) ------------------------------------------------------------------------ (681.00,382.34)(1.500,-3.000)[2]{} ------------------------------------------------------------------------ (684,377.84) ------------------------------------------------------------------------ (684.00,379.34)(1.500,-3.000)[2]{} ------------------------------------------------------------------------ (687,374.84) ------------------------------------------------------------------------ (687.00,376.34)(1.500,-3.000)[2]{} ------------------------------------------------------------------------ (689.34,372) ------------------------------------------------------------------------ (688.34,373.50)(2.000,-1.500)[2]{} ------------------------------------------------------------------------ (692,369.34) ------------------------------------------------------------------------ (692.00,370.34)(1.500,-2.000)[2]{} ------------------------------------------------------------------------ (694.34,367) ------------------------------------------------------------------------ (693.34,368.50)(2.000,-1.500)[2]{} ------------------------------------------------------------------------ (697,364.34) ------------------------------------------------------------------------ (697.00,365.34)(1.500,-2.000)[2]{} ------------------------------------------------------------------------ (699.34,362) ------------------------------------------------------------------------ (698.34,363.50)(2.000,-1.500)[2]{} ------------------------------------------------------------------------ (702,359.34) ------------------------------------------------------------------------ (702.00,360.34)(1.500,-2.000)[2]{} ------------------------------------------------------------------------ (704.34,357) ------------------------------------------------------------------------ (703.34,358.50)(2.000,-1.500)[2]{} ------------------------------------------------------------------------ (707,354.34) ------------------------------------------------------------------------ (707.00,355.34)(1.500,-2.000)[2]{} ------------------------------------------------------------------------ (710,352.34) ------------------------------------------------------------------------ (710.00,353.34)(1.000,-2.000)[2]{} ------------------------------------------------------------------------ (711.34,350) ------------------------------------------------------------------------ (710.34,351.50)(2.000,-1.500)[2]{} ------------------------------------------------------------------------ (713.34,347) ------------------------------------------------------------------------ (712.34,348.50)(2.000,-1.500)[2]{} ------------------------------------------------------------------------ (716,344.34) ------------------------------------------------------------------------ (716.00,345.34)(1.500,-2.000)[2]{} ------------------------------------------------------------------------ (718.34,342) ------------------------------------------------------------------------ (717.34,343.50)(2.000,-1.500)[2]{} ------------------------------------------------------------------------ (721,339.34) ------------------------------------------------------------------------ (721.00,340.34)(1.000,-2.000)[2]{} ------------------------------------------------------------------------ (722.34,337) ------------------------------------------------------------------------ (721.34,338.50)(2.000,-1.500)[2]{} ------------------------------------------------------------------------ (725,334.34) ------------------------------------------------------------------------ (725.00,335.34)(1.000,-2.000)[2]{} ------------------------------------------------------------------------ (727,332.34) ------------------------------------------------------------------------ (727.00,333.34)(1.000,-2.000)[2]{} ------------------------------------------------------------------------ (728.34,330) ------------------------------------------------------------------------ (727.34,331.50)(2.000,-1.500)[2]{} ------------------------------------------------------------------------ (731,327.84) ------------------------------------------------------------------------ (731.00,328.34)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (732.34,326) ------------------------------------------------------------------------ (731.34,327.50)(2.000,-1.500)[2]{} ------------------------------------------------------------------------ (735,323.34) ------------------------------------------------------------------------ (735.00,324.34)(1.000,-2.000)[2]{} ------------------------------------------------------------------------ (737,321.34) ------------------------------------------------------------------------ (737.00,322.34)(1.000,-2.000)[2]{} ------------------------------------------------------------------------ (739,319.34) ------------------------------------------------------------------------ (739.00,320.34)(1.000,-2.000)[2]{} ------------------------------------------------------------------------ (741,317.34) ------------------------------------------------------------------------ (741.00,318.34)(1.000,-2.000)[2]{} ------------------------------------------------------------------------ (743,315.34) ------------------------------------------------------------------------ (743.00,316.34)(1.000,-2.000)[2]{} ------------------------------------------------------------------------ (745,313.34) ------------------------------------------------------------------------ (745.00,314.34)(1.000,-2.000)[2]{} ------------------------------------------------------------------------ (745.84,312) ------------------------------------------------------------------------ (745.34,313.00)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (748,309.84) ------------------------------------------------------------------------ (748.00,310.34)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (750,308.34) ------------------------------------------------------------------------ (750.00,309.34)(1.000,-2.000)[2]{} ------------------------------------------------------------------------ (751.34,306) ------------------------------------------------------------------------ (750.34,307.50)(2.000,-1.500)[2]{} ------------------------------------------------------------------------ (754,303.34) ------------------------------------------------------------------------ (754.00,304.34)(1.000,-2.000)[2]{} ------------------------------------------------------------------------ (754.84,301) ------------------------------------------------------------------------ (754.34,302.50)(1.000,-1.500)[2]{} ------------------------------------------------------------------------ (756.34,298) ------------------------------------------------------------------------ (755.34,299.50)(2.000,-1.500)[2]{} ------------------------------------------------------------------------ (759,295.84) ------------------------------------------------------------------------ (759.00,296.34)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (759.84,294) ------------------------------------------------------------------------ (759.34,295.50)(1.000,-1.500)[2]{} ------------------------------------------------------------------------ (762,291.34) ------------------------------------------------------------------------ (762.00,292.34)(1.000,-2.000)[2]{} ------------------------------------------------------------------------ (764,289.34) ------------------------------------------------------------------------ (764.00,290.34)(1.000,-2.000)[2]{} ------------------------------------------------------------------------ (764.84,287) ------------------------------------------------------------------------ (764.34,288.50)(1.000,-1.500)[2]{} ------------------------------------------------------------------------ (767,284.84) ------------------------------------------------------------------------ (767.00,285.34)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (767.84,284) ------------------------------------------------------------------------ (767.34,285.00)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (770,281.84) ------------------------------------------------------------------------ (770.00,282.34)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (772,280.84) ------------------------------------------------------------------------ (772.00,281.34)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (773,279.34) ------------------------------------------------------------------------ (773.00,280.34)(1.000,-2.000)[2]{} ------------------------------------------------------------------------ (773.84,276) ------------------------------------------------------------------------ (773.34,278.00)(1.000,-2.000)[2]{} ------------------------------------------------------------------------ (776,273.84) ------------------------------------------------------------------------ (776.00,274.34)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (776.84,271) ------------------------------------------------------------------------ (776.34,273.00)(1.000,-2.000)[2]{} ------------------------------------------------------------------------ (778.34,268) ------------------------------------------------------------------------ (777.34,269.50)(2.000,-1.500)[2]{} ------------------------------------------------------------------------ (779.84,265) ------------------------------------------------------------------------ (779.34,266.50)(1.000,-1.500)[2]{} ------------------------------------------------------------------------ (781.34,262) ------------------------------------------------------------------------ (780.34,263.50)(2.000,-1.500)[2]{} ------------------------------------------------------------------------ (784,259.84) ------------------------------------------------------------------------ (784.00,260.34)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (785,258.84) ------------------------------------------------------------------------ (785.00,259.34)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (787,257.84) ------------------------------------------------------------------------ (787.00,258.34)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (789,256.34) ------------------------------------------------------------------------ (789.00,257.34)(1.000,-2.000)[2]{} ------------------------------------------------------------------------ (791,254.84) ------------------------------------------------------------------------ (791.00,255.34)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (792,253.84) ------------------------------------------------------------------------ (792.00,254.34)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (793,252.84) ------------------------------------------------------------------------ (793.00,253.34)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (788.0,259.0) (794.84,251) ------------------------------------------------------------------------ (794.34,252.50)(1.000,-1.500)[2]{} ------------------------------------------------------------------------ (797,248.34) ------------------------------------------------------------------------ (797.00,249.34)(1.000,-2.000)[2]{} ------------------------------------------------------------------------ (797.84,247) ------------------------------------------------------------------------ (797.34,248.00)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (795.0,254.0) (801,244.84) ------------------------------------------------------------------------ (801.00,245.34)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (801.34,243) ------------------------------------------------------------------------ (800.34,244.50)(2.000,-1.500)[2]{} ------------------------------------------------------------------------ (802.84,241) ------------------------------------------------------------------------ (802.34,242.00)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (805,238.84) ------------------------------------------------------------------------ (805.00,239.34)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (806,237.84) ------------------------------------------------------------------------ (806.00,238.34)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (807,236.34) ------------------------------------------------------------------------ (807.00,237.34)(1.000,-2.000)[2]{} ------------------------------------------------------------------------ (809,234.84) ------------------------------------------------------------------------ (809.00,235.34)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (808.84,233) ------------------------------------------------------------------------ (808.34,234.50)(1.000,-1.500)[2]{} ------------------------------------------------------------------------ (811,230.84) ------------------------------------------------------------------------ (811.00,231.34)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (812,230.84) ------------------------------------------------------------------------ (812.00,230.34)(0.500,1.000)[2]{} ------------------------------------------------------------------------ (813,230.84) ------------------------------------------------------------------------ (813.00,231.34)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (813.84,230) ------------------------------------------------------------------------ (813.34,231.00)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (816,227.84) ------------------------------------------------------------------------ (816.00,228.34)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (815.84,225) ------------------------------------------------------------------------ (815.34,227.00)(1.000,-2.000)[2]{} ------------------------------------------------------------------------ (816.84,225) ------------------------------------------------------------------------ (816.34,225.00)(1.000,1.500)[2]{} ------------------------------------------------------------------------ (817.84,226) ------------------------------------------------------------------------ (817.34,227.00)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (818.84,224) ------------------------------------------------------------------------ (818.34,225.00)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (821,222.84) ------------------------------------------------------------------------ (821.00,222.34)(0.500,1.000)[2]{} ------------------------------------------------------------------------ (822,222.84) ------------------------------------------------------------------------ (822.00,223.34)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (824,221.84) ------------------------------------------------------------------------ (824.00,222.34)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (825,221.84) ------------------------------------------------------------------------ (825.00,221.34)(0.500,1.000)[2]{} ------------------------------------------------------------------------ (824.84,222) ------------------------------------------------------------------------ (824.34,223.00)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (827,219.84) ------------------------------------------------------------------------ (827.00,220.34)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (828,218.84) ------------------------------------------------------------------------ (828.00,219.34)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (827.84,218) ------------------------------------------------------------------------ (827.34,219.00)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (830,215.84) ------------------------------------------------------------------------ (830.00,216.34)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (829.84,214) ------------------------------------------------------------------------ (829.34,215.50)(1.000,-1.500)[2]{} ------------------------------------------------------------------------ (830.84,212) ------------------------------------------------------------------------ (830.34,213.00)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (831.84,212) ------------------------------------------------------------------------ (831.34,212.00)(1.000,1.500)[2]{} ------------------------------------------------------------------------ (832.84,212) ------------------------------------------------------------------------ (832.34,213.50)(1.000,-1.500)[2]{} ------------------------------------------------------------------------ (833.84,210) ------------------------------------------------------------------------ (833.34,211.00)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (800.0,247.0) (838,208.84) ------------------------------------------------------------------------ (838.00,208.34)(0.500,1.000)[2]{} ------------------------------------------------------------------------ (836.0,210.0) (840,208.84) ------------------------------------------------------------------------ (840.00,209.34)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (839.0,211.0) (842,208.84) ------------------------------------------------------------------------ (842.00,208.34)(0.500,1.000)[2]{} ------------------------------------------------------------------------ (841.84,208) ------------------------------------------------------------------------ (841.34,209.50)(1.000,-1.500)[2]{} ------------------------------------------------------------------------ (841.0,210.0) (843.84,206) ------------------------------------------------------------------------ (843.34,207.00)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (846,204.84) ------------------------------------------------------------------------ (846.00,204.34)(0.500,1.000)[2]{} ------------------------------------------------------------------------ (847,204.84) ------------------------------------------------------------------------ (847.00,205.34)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (848,204.84) ------------------------------------------------------------------------ (848.00,204.34)(0.500,1.000)[2]{} ------------------------------------------------------------------------ (844.0,208.0) (848.84,205) ------------------------------------------------------------------------ (848.34,206.00)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (849.0,207.0) (849.84,201) ------------------------------------------------------------------------ (849.34,202.50)(1.000,-1.500)[2]{} ------------------------------------------------------------------------ (850.84,199) ------------------------------------------------------------------------ (850.34,200.00)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (851.84,196) ------------------------------------------------------------------------ (851.34,197.50)(1.000,-1.500)[2]{} ------------------------------------------------------------------------ (852.84,193) ------------------------------------------------------------------------ (852.34,194.50)(1.000,-1.500)[2]{} ------------------------------------------------------------------------ (853.84,193) ------------------------------------------------------------------------ (853.34,193.00)(1.000,1.000)[2]{} ------------------------------------------------------------------------ (856,193.84) ------------------------------------------------------------------------ (856.00,193.34)(0.500,1.000)[2]{} ------------------------------------------------------------------------ (855.84,193) ------------------------------------------------------------------------ (855.34,194.50)(1.000,-1.500)[2]{} ------------------------------------------------------------------------ (856.84,193) ------------------------------------------------------------------------ (856.34,193.00)(1.000,1.500)[2]{} ------------------------------------------------------------------------ (851.0,204.0) (859,192.84) ------------------------------------------------------------------------ (859.00,193.34)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (858.84,194) ------------------------------------------------------------------------ (858.34,194.00)(1.000,1.000)[2]{} ------------------------------------------------------------------------ (861,194.84) ------------------------------------------------------------------------ (861.00,194.34)(0.500,1.000)[2]{} ------------------------------------------------------------------------ (862,195.84) ------------------------------------------------------------------------ (862.00,195.34)(0.500,1.000)[2]{} ------------------------------------------------------------------------ (861.84,198) ------------------------------------------------------------------------ (861.34,198.00)(1.000,2.500)[2]{} ------------------------------------------------------------------------ (859.0,195.0) (864,203) (862.84,203) ------------------------------------------------------------------------ (862.34,203.00)(1.000,1.000)[2]{} ------------------------------------------------------------------------ (863.84,201) ------------------------------------------------------------------------ (863.34,203.00)(1.000,-2.000)[2]{} ------------------------------------------------------------------------ (866,199.84) ------------------------------------------------------------------------ (866.00,199.34)(0.500,1.000)[2]{} ------------------------------------------------------------------------ (867,199.84) ------------------------------------------------------------------------ (867.00,200.34)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (868.0,201.0) (869.0,201.0) (870,199.84) ------------------------------------------------------------------------ (870.00,200.34)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (869.84,201) ------------------------------------------------------------------------ (869.34,201.00)(1.000,1.000)[2]{} ------------------------------------------------------------------------ (872,200.84) ------------------------------------------------------------------------ (872.00,201.34)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (869.0,202.0) (873,202) (873,200.84) ------------------------------------------------------------------------ (873.00,200.34)(0.500,1.000)[2]{} ------------------------------------------------------------------------ (875,201.84) ------------------------------------------------------------------------ (875.00,201.34)(0.500,1.000)[2]{} ------------------------------------------------------------------------ (874.0,203.0) (876,199.84) ------------------------------------------------------------------------ (876.00,200.34)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (877,198.84) ------------------------------------------------------------------------ (877.00,199.34)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (876.84,198) ------------------------------------------------------------------------ (876.34,199.00)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (877.84,198) ------------------------------------------------------------------------ (877.34,198.00)(1.000,1.500)[2]{} ------------------------------------------------------------------------ (876.0,202.0) (220,641)(19.932,-5.787)[2]{} (251,632)(19.736,-6.426)[2]{} (294,618)(19.572,-6.908)[2]{} (338.29,602.19) (355,596)(19.034,-8.276)[2]{} (395.98,578.81) (414.62,569.69) (416,569)(18.808,-8.777)[0]{} (433.41,560.87) (452.04,551.75) (470.50,542.25) (471,542)(18.221,-9.939)[0]{} (488.57,532.06) (492,530)(18.564,-9.282)[0]{} (506.88,522.29) (511,520)(17.601,-11.000)[0]{} (524.77,511.79) (528,510)(16.889,-12.064)[0]{} (542.57,501.22) (543,501)(18.021,-10.298)[0]{} (550,497)(15.945,-13.287)[0]{} (559.82,489.82) (563,488)(17.270,-11.513)[0]{} (569,484)(17.270,-11.513)[0]{} (577.09,478.33) (580,476)(17.270,-11.513)[0]{} (586,472)(16.207,-12.966)[0]{} (593.66,465.87) (596,464)(17.798,-10.679)[0]{} (601,461)(17.798,-10.679)[0]{} (606,458)(16.604,-12.453)[0]{} (610.94,454.43) (615,452)(16.604,-12.453)[0]{} (619,449)(16.604,-12.453)[0]{} (623,446)(16.604,-12.453)[0]{} (627.82,442.39) (631,440)(16.604,-12.453)[0]{} (635,437)(16.604,-12.453)[0]{} (639,434)(16.604,-12.453)[0]{} (644.26,429.74) (646,428)(18.564,-9.282)[0]{} (650,426)(14.676,-14.676)[0]{} (653,423)(16.604,-12.453)[0]{} (657,420)(17.270,-11.513)[0]{} (660.69,417.31) (663,415)(17.270,-11.513)[0]{} (666,413)(18.564,-9.282)[0]{} (670,411)(14.676,-14.676)[0]{} (673,408)(17.270,-11.513)[0]{} (676.87,404.70) (678,403)(17.270,-11.513)[0]{} (681,401)(17.270,-11.513)[0]{} (684,399)(19.690,-6.563)[0]{} (687,398)(17.270,-11.513)[0]{} (690,396)(11.513,-17.270)[0]{} (692.94,392.38) (695,391)(14.676,-14.676)[0]{} (697,389)(17.270,-11.513)[0]{} (700,387)(18.564,-9.282)[0]{} (702,386)(14.676,-14.676)[0]{} (705,383)(14.676,-14.676)[0]{} (709.40,380.20) (710,380)(11.513,-17.270)[0]{} (712,377)(18.564,-9.282)[0]{} (714,376)(11.513,-17.270)[0]{} (716,373)(19.690,-6.563)[0]{} (719,372)(14.676,-14.676)[0]{} (721,370)(11.513,-17.270)[0]{} (723.97,366.52) (725,366)(14.676,-14.676)[0]{} (727,364)(11.513,-17.270)[0]{} (729,361)(18.564,-9.282)[0]{} (731,360)(14.676,-14.676)[0]{} (733,358)(14.676,-14.676)[0]{} (735,356)(18.564,-9.282)[0]{} (738.69,352.47) (739,352)(14.676,-14.676)[0]{} (741,350)(14.676,-14.676)[0]{} (743,348)(18.564,-9.282)[0]{} (745,347)(18.564,-9.282)[0]{} (747,346)(9.282,-18.564)[0]{} (748,344)(14.676,-14.676)[0]{} (750,342)(18.564,-9.282)[0]{} (752,341)(18.564,-9.282)[0]{} (754.47,339.77) (756,339)(9.282,-18.564)[0]{} (757,337)(11.513,-17.270)[0]{} (759,334)(14.676,-14.676)[0]{} (761,332)(14.676,-14.676)[0]{} (762,331)(14.676,-14.676)[0]{} (764,329)(14.676,-14.676)[0]{} (766,327)(14.676,-14.676)[0]{} (768.34,324.66) (769,324)(14.676,-14.676)[0]{} (770,323)(18.564,-9.282)[0]{} (772,322)(9.282,-18.564)[0]{} (773,320)(11.513,-17.270)[0]{} (775,317)(9.282,-18.564)[0]{} (776,315)(20.756,0.000)[0]{} (778,315)(14.676,-14.676)[0]{} (779,314)(18.564,-9.282)[0]{} (781,313)(14.676,-14.676)[0]{} (783.02,312.00) (784,312)(14.676,-14.676)[0]{} (785,311)(18.564,9.282)[0]{} (787,312)(9.282,-18.564)[0]{} (788,310)(20.756,0.000)[0]{} (789,310)(18.564,-9.282)[0]{} (791,309)(20.756,0.000)[0]{} (792,309)(20.756,0.000)[0]{} (793,309)(11.513,-17.270)[0]{} (795,306)(9.282,-18.564)[0]{} (796,304)(9.282,-18.564)[0]{} (797.52,301.74) (799,301)(14.676,14.676)[0]{} (800,302)(20.756,0.000)[0]{} (801,302)(20.756,0.000)[0]{} (802,302)(20.756,0.000)[0]{} (804,302)(14.676,-14.676)[0]{} (805,301)(20.756,0.000)[0]{} (806,301)(14.676,14.676)[0]{} (807,302)(20.756,0.000)[0]{} (809,302)(20.756,0.000)[0]{} (810,302)(14.676,-14.676)[0]{} (811,301)(20.756,0.000)[0]{} (812,301)(6.563,-19.690)[0]{} (814.14,297.43) (815,297)(14.676,14.676)[0]{} (816,298)(20.756,0.000)[0]{} (817,298)(20.756,0.000)[0]{} (818,298)(20.756,0.000)[0]{} (819,298)(20.756,0.000)[0]{} (820,298)(14.676,-14.676)[0]{} (821,297)(14.676,-14.676)[0]{} (822,296)(18.564,-9.282)[0]{} (824,295)(20.756,0.000)[0]{} (825,295)(20.756,0.000)[0]{} (826,295)(14.676,-14.676)[0]{} (827,294)(9.282,-18.564)[0]{} (828,292)(14.676,-14.676)[0]{} (829,291)(14.676,14.676)[0]{} (830.59,291.41) (831,291)(9.282,18.564)[0]{} (832,293)(20.756,0.000)[0]{} (833,293)(14.676,-14.676)[0]{} (834,292)(20.756,0.000)[0]{} (835,292)(9.282,-18.564)[0]{} (836,290)(9.282,18.564)[0]{} (837,292)(14.676,14.676)[0]{} (838,293)(14.676,-14.676)[0]{} (839,292)(14.676,-14.676)[0]{} (840,291)(9.282,18.564)[0]{} (841,293)(14.676,-14.676)[0]{} (842.97,290.06) (843,290)(14.676,-14.676)[0]{} (844,289)(9.282,-18.564)[0]{} (845,287)(14.676,-14.676)[0]{} (846,286)(20.756,0.000)[0]{} (847,286)(14.676,-14.676)[0]{} (848,285)(20.756,0.000)[0]{} (849,285)(20.756,0.000)[0]{} (850,285)(14.676,14.676)[0]{} (851,286)(0.000,20.756)[0]{} (851,287)(20.756,0.000)[0]{} (852,287)(20.756,0.000)[0]{} (853,287)(20.756,0.000)[0]{} (854,287)(20.756,0.000)[0]{} (855,287)(20.756,0.000)[0]{} (856,287)(9.282,-18.564)[0]{} (857,285)(20.756,0.000)[0]{} (858.39,284.61) (859,284)(0.000,-20.756)[0]{} (859,282)(14.676,-14.676)[0]{} (860,281)(14.676,-14.676)[0]{} (861,280)(9.282,-18.564)[0]{} (862,278)(14.676,14.676)[0]{} (863,279)(14.676,-14.676)[0]{} (864,278)(0.000,20.756)[0]{} (864,279)(9.282,-18.564)[0]{} (865,277)(9.282,-18.564)[0]{} (866,275)(20.756,0.000)[0]{} (867,275)(14.676,-14.676)[0]{} (868,274)(14.676,-14.676)[0]{} (869.00,272.29) (869,269)(14.676,14.676)[0]{} (870,270)(14.676,14.676)[0]{} (871,271)(14.676,14.676)[0]{} (872,272)(20.756,0.000)[0]{} (873,272)(0.000,-20.756)[0]{} (873,269)(14.676,-14.676)[0]{} (874,268)(14.676,-14.676)[0]{} (875,267)(14.676,-14.676)[0]{} (876,266)(0.000,-20.756)[0]{} (876.96,261.14) (877,261)(5.034,-20.136)[0]{} (878,257)(9.282,18.564)[0]{} (879,259)(9.282,18.564)[0]{} (880,261) (220,633) (220,628.01) ------------------------------------------------------------------------ (220.00,630.51)(7.000,-5.000)[2]{} ------------------------------------------------------------------------ (234,623.51) ------------------------------------------------------------------------ (234.00,625.51)(6.500,-4.000)[2]{} ------------------------------------------------------------------------ (247,619.01) ------------------------------------------------------------------------ (247.00,621.51)(6.000,-5.000)[2]{} ------------------------------------------------------------------------ (259,614.51) ------------------------------------------------------------------------ (259.00,616.51)(5.500,-4.000)[2]{} ------------------------------------------------------------------------ (270,610.51) ------------------------------------------------------------------------ (270.00,612.51)(5.500,-4.000)[2]{} ------------------------------------------------------------------------ (281,606.51) ------------------------------------------------------------------------ (281.00,608.51)(4.500,-4.000)[2]{} ------------------------------------------------------------------------ (290,603.01) ------------------------------------------------------------------------ (290.00,604.51)(5.000,-3.000)[2]{} ------------------------------------------------------------------------ (300,599.51) ------------------------------------------------------------------------ (300.00,601.51)(4.000,-4.000)[2]{} ------------------------------------------------------------------------ (308,596.01) ------------------------------------------------------------------------ (308.00,597.51)(4.000,-3.000)[2]{} ------------------------------------------------------------------------ (316,593.01) ------------------------------------------------------------------------ (316.00,594.51)(4.000,-3.000)[2]{} ------------------------------------------------------------------------ (324,590.01) ------------------------------------------------------------------------ (324.00,591.51)(3.500,-3.000)[2]{} ------------------------------------------------------------------------ (331,587.01) ------------------------------------------------------------------------ (331.00,588.51)(3.500,-3.000)[2]{} ------------------------------------------------------------------------ (338,584.01) ------------------------------------------------------------------------ (338.00,585.51)(3.500,-3.000)[2]{} ------------------------------------------------------------------------ (345,581.01) ------------------------------------------------------------------------ (345.00,582.51)(3.000,-3.000)[2]{} ------------------------------------------------------------------------ (351,578.01) ------------------------------------------------------------------------ (351.00,579.51)(3.000,-3.000)[2]{} ------------------------------------------------------------------------ (357,575.01) ------------------------------------------------------------------------ (357.00,576.51)(3.000,-3.000)[2]{} ------------------------------------------------------------------------ (363,572.51) ------------------------------------------------------------------------ (363.00,573.51)(3.000,-2.000)[2]{} ------------------------------------------------------------------------ (369,570.01) ------------------------------------------------------------------------ (369.00,571.51)(2.500,-3.000)[2]{} ------------------------------------------------------------------------ (374,567.51) ------------------------------------------------------------------------ (374.00,568.51)(3.000,-2.000)[2]{} ------------------------------------------------------------------------ (380,565.51) ------------------------------------------------------------------------ (380.00,566.51)(2.500,-2.000)[2]{} ------------------------------------------------------------------------ (385,563.01) ------------------------------------------------------------------------ (385.00,564.51)(2.500,-3.000)[2]{} ------------------------------------------------------------------------ (390,560.51) ------------------------------------------------------------------------ (390.00,561.51)(2.000,-2.000)[2]{} ------------------------------------------------------------------------ (394,558.51) ------------------------------------------------------------------------ (394.00,559.51)(2.500,-2.000)[2]{} ------------------------------------------------------------------------ (399,556.51) ------------------------------------------------------------------------ (399.00,557.51)(2.000,-2.000)[2]{} ------------------------------------------------------------------------ (403,554.51) ------------------------------------------------------------------------ (403.00,555.51)(2.500,-2.000)[2]{} ------------------------------------------------------------------------ (408,552.51) ------------------------------------------------------------------------ (408.00,553.51)(2.000,-2.000)[2]{} ------------------------------------------------------------------------ (412,550.51) ------------------------------------------------------------------------ (412.00,551.51)(2.000,-2.000)[2]{} ------------------------------------------------------------------------ (416,548.51) ------------------------------------------------------------------------ (416.00,549.51)(2.000,-2.000)[2]{} ------------------------------------------------------------------------ (420,546.51) ------------------------------------------------------------------------ (420.00,547.51)(2.000,-2.000)[2]{} ------------------------------------------------------------------------ (424,544.51) ------------------------------------------------------------------------ (424.00,545.51)(2.000,-2.000)[2]{} ------------------------------------------------------------------------ (428,542.51) ------------------------------------------------------------------------ (428.00,543.51)(1.500,-2.000)[2]{} ------------------------------------------------------------------------ (431,540.51) ------------------------------------------------------------------------ (431.00,541.51)(2.000,-2.000)[2]{} ------------------------------------------------------------------------ (435,538.51) ------------------------------------------------------------------------ (435.00,539.51)(2.000,-2.000)[2]{} ------------------------------------------------------------------------ (439,537.01) ------------------------------------------------------------------------ (439.00,537.51)(1.500,-1.000)[2]{} ------------------------------------------------------------------------ (442,535.51) ------------------------------------------------------------------------ (442.00,536.51)(1.500,-2.000)[2]{} ------------------------------------------------------------------------ (445,533.51) ------------------------------------------------------------------------ (445.00,534.51)(2.000,-2.000)[2]{} ------------------------------------------------------------------------ (449,532.01) ------------------------------------------------------------------------ (449.00,532.51)(1.500,-1.000)[2]{} ------------------------------------------------------------------------ (452,530.51) ------------------------------------------------------------------------ (452.00,531.51)(1.500,-2.000)[2]{} ------------------------------------------------------------------------ (455,528.51) ------------------------------------------------------------------------ (455.00,529.51)(1.500,-2.000)[2]{} ------------------------------------------------------------------------ (458,527.01) ------------------------------------------------------------------------ (458.00,527.51)(1.500,-1.000)[2]{} ------------------------------------------------------------------------ (461,525.51) ------------------------------------------------------------------------ (461.00,526.51)(1.500,-2.000)[2]{} ------------------------------------------------------------------------ (464,524.01) ------------------------------------------------------------------------ (464.00,524.51)(1.500,-1.000)[2]{} ------------------------------------------------------------------------ (467,522.51) ------------------------------------------------------------------------ (467.00,523.51)(1.500,-2.000)[2]{} ------------------------------------------------------------------------ (470,521.01) ------------------------------------------------------------------------ (470.00,521.51)(1.500,-1.000)[2]{} ------------------------------------------------------------------------ (473,520.01) ------------------------------------------------------------------------ (473.00,520.51)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (475,518.51) ------------------------------------------------------------------------ (475.00,519.51)(1.500,-2.000)[2]{} ------------------------------------------------------------------------ (478,517.01) ------------------------------------------------------------------------ (478.00,517.51)(1.500,-1.000)[2]{} ------------------------------------------------------------------------ (481,515.51) ------------------------------------------------------------------------ (481.00,516.51)(1.000,-2.000)[2]{} ------------------------------------------------------------------------ (483,514.01) ------------------------------------------------------------------------ (483.00,514.51)(1.500,-1.000)[2]{} ------------------------------------------------------------------------ (486,513.01) ------------------------------------------------------------------------ (486.00,513.51)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (488,511.51) ------------------------------------------------------------------------ (488.00,512.51)(1.500,-2.000)[2]{} ------------------------------------------------------------------------ (491,510.01) ------------------------------------------------------------------------ (491.00,510.51)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (493,509.01) ------------------------------------------------------------------------ (493.00,509.51)(1.500,-1.000)[2]{} ------------------------------------------------------------------------ (496,508.01) ------------------------------------------------------------------------ (496.00,508.51)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (498,506.51) ------------------------------------------------------------------------ (498.00,507.51)(1.000,-2.000)[2]{} ------------------------------------------------------------------------ (500,505.01) ------------------------------------------------------------------------ (500.00,505.51)(1.500,-1.000)[2]{} ------------------------------------------------------------------------ (503,504.01) ------------------------------------------------------------------------ (503.00,504.51)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (505,503.01) ------------------------------------------------------------------------ (505.00,503.51)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (507,502.01) ------------------------------------------------------------------------ (507.00,502.51)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (509,500.51) ------------------------------------------------------------------------ (509.00,501.51)(1.500,-2.000)[2]{} ------------------------------------------------------------------------ (512,499.01) ------------------------------------------------------------------------ (512.00,499.51)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (514,498.01) ------------------------------------------------------------------------ (514.00,498.51)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (516,497.01) ------------------------------------------------------------------------ (516.00,497.51)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (518,496.01) ------------------------------------------------------------------------ (518.00,496.51)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (520,495.01) ------------------------------------------------------------------------ (520.00,495.51)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (522,494.01) ------------------------------------------------------------------------ (522.00,494.51)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (524,493.01) ------------------------------------------------------------------------ (524.00,493.51)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (526,492.01) ------------------------------------------------------------------------ (526.00,492.51)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (528,491.01) ------------------------------------------------------------------------ (528.00,491.51)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (530,489.51) ------------------------------------------------------------------------ (530.00,490.51)(1.000,-2.000)[2]{} ------------------------------------------------------------------------ (532,488.01) ------------------------------------------------------------------------ (532.00,488.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (533,487.01) ------------------------------------------------------------------------ (533.00,487.51)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (535,486.01) ------------------------------------------------------------------------ (535.00,486.51)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (537,485.01) ------------------------------------------------------------------------ (537.00,485.51)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (539,484.01) ------------------------------------------------------------------------ (539.00,484.51)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (542,483.01) ------------------------------------------------------------------------ (542.00,483.51)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (544,482.01) ------------------------------------------------------------------------ (544.00,482.51)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (546,481.01) ------------------------------------------------------------------------ (546.00,481.51)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (548,480.01) ------------------------------------------------------------------------ (548.00,480.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (549,479.01) ------------------------------------------------------------------------ (549.00,479.51)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (551,478.01) ------------------------------------------------------------------------ (551.00,478.51)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (553,477.01) ------------------------------------------------------------------------ (553.00,477.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (554,476.01) ------------------------------------------------------------------------ (554.00,476.51)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (556,475.01) ------------------------------------------------------------------------ (556.00,475.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (557,474.01) ------------------------------------------------------------------------ (557.00,474.51)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (559,473.01) ------------------------------------------------------------------------ (559.00,473.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (541.0,486.0) (562,472.01) ------------------------------------------------------------------------ (562.00,472.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (563,471.01) ------------------------------------------------------------------------ (563.00,471.51)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (565,470.01) ------------------------------------------------------------------------ (565.00,470.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (566,469.01) ------------------------------------------------------------------------ (566.00,469.51)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (568,468.01) ------------------------------------------------------------------------ (568.00,468.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (560.0,475.0) (571,467.01) ------------------------------------------------------------------------ (571.00,467.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (572,466.01) ------------------------------------------------------------------------ (572.00,466.51)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (574,465.01) ------------------------------------------------------------------------ (574.00,465.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (575,464.01) ------------------------------------------------------------------------ (575.00,464.51)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (569.0,470.0) (578,463.01) ------------------------------------------------------------------------ (578.00,463.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (579,462.01) ------------------------------------------------------------------------ (579.00,462.51)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (581,461.01) ------------------------------------------------------------------------ (581.00,461.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (577.0,466.0) (583,460.01) ------------------------------------------------------------------------ (583.00,460.51)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (585,459.01) ------------------------------------------------------------------------ (585.00,459.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (586,458.01) ------------------------------------------------------------------------ (586.00,458.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (582.0,463.0) (589,457.01) ------------------------------------------------------------------------ (589.00,457.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (590,456.01) ------------------------------------------------------------------------ (590.00,456.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (591,455.01) ------------------------------------------------------------------------ (591.00,455.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (587.0,460.0) (594,454.01) ------------------------------------------------------------------------ (594.00,454.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (595,453.01) ------------------------------------------------------------------------ (595.00,453.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (592.0,457.0) (597,452.01) ------------------------------------------------------------------------ (597.00,452.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (598,451.01) ------------------------------------------------------------------------ (598.00,451.51)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (596.0,455.0) (601,450.01) ------------------------------------------------------------------------ (601.00,450.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (602,449.01) ------------------------------------------------------------------------ (602.00,449.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (600.0,453.0) (604,448.01) ------------------------------------------------------------------------ (604.00,448.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (605,447.01) ------------------------------------------------------------------------ (605.00,447.51)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (603.0,451.0) (608,446.01) ------------------------------------------------------------------------ (608.00,446.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (609,445.01) ------------------------------------------------------------------------ (609.00,445.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (607.0,449.0) (611,444.01) ------------------------------------------------------------------------ (611.00,444.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (612,443.01) ------------------------------------------------------------------------ (612.00,443.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (610.0,447.0) (614,442.01) ------------------------------------------------------------------------ (614.00,442.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (615,441.01) ------------------------------------------------------------------------ (615.00,441.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (613.0,445.0) (618,440.01) ------------------------------------------------------------------------ (618.00,440.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (616.0,443.0) (620,439.01) ------------------------------------------------------------------------ (620.00,439.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (621,438.01) ------------------------------------------------------------------------ (621.00,438.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (619.0,442.0) (623,437.01) ------------------------------------------------------------------------ (623.00,437.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (622.0,440.0) (625,436.01) ------------------------------------------------------------------------ (625.00,436.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (624.0,439.0) (627,435.01) ------------------------------------------------------------------------ (627.00,435.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (628,434.01) ------------------------------------------------------------------------ (628.00,434.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (626.0,438.0) (630,433.01) ------------------------------------------------------------------------ (630.00,433.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (629.0,436.0) (632,432.01) ------------------------------------------------------------------------ (632.00,432.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (631.0,435.0) (633.0,434.0) (634,430.01) ------------------------------------------------------------------------ (634.00,430.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (634.0,433.0) (636,429.01) ------------------------------------------------------------------------ (636.00,429.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (635.0,432.0) (638,428.01) ------------------------------------------------------------------------ (638.00,428.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (637.0,431.0) (640,427.01) ------------------------------------------------------------------------ (640.00,427.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (639.0,430.0) (642,426.01) ------------------------------------------------------------------------ (642.00,426.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (641.0,429.0) (644,425.01) ------------------------------------------------------------------------ (644.00,425.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (643.0,428.0) (645,427) (645,424.01) ------------------------------------------------------------------------ (645.00,424.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (646,423.01) ------------------------------------------------------------------------ (646.00,423.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (648,422.01) ------------------------------------------------------------------------ (648.00,422.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (647.0,425.0) (650,421.01) ------------------------------------------------------------------------ (650.00,421.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (649.0,424.0) (651,423) (651,420.01) ------------------------------------------------------------------------ (651.00,420.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (653,419.01) ------------------------------------------------------------------------ (653.00,419.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (652.0,422.0) (655,418.01) ------------------------------------------------------------------------ (655.00,418.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (654.0,421.0) (656,420) (657,417.01) ------------------------------------------------------------------------ (657.00,417.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (656.0,420.0) (659,416.01) ------------------------------------------------------------------------ (659.00,416.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (658.0,419.0) (660,418) (660,415.01) ------------------------------------------------------------------------ (660.00,415.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (662,414.01) ------------------------------------------------------------------------ (662.00,414.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (661.0,417.0) (663.0,416.0) (664.0,415.0) (665,412.01) ------------------------------------------------------------------------ (665.00,412.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (664.0,415.0) (666.0,414.0) (667.0,413.0) (669,410.01) ------------------------------------------------------------------------ (669.00,410.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (667.0,413.0) (670,412) (670,409.01) ------------------------------------------------------------------------ (670.00,409.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (672,408.01) ------------------------------------------------------------------------ (672.00,408.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (671.0,411.0) (673,410) (673,407.01) ------------------------------------------------------------------------ (673.00,407.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (675,406.01) ------------------------------------------------------------------------ (675.00,406.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (674.0,409.0) (677,405.01) ------------------------------------------------------------------------ (677.00,405.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (676.0,408.0) (678,407) (678,404.01) ------------------------------------------------------------------------ (678.00,404.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (680,403.01) ------------------------------------------------------------------------ (680.00,403.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (679.0,406.0) (681.0,405.0) (682.0,404.0) (683,401.01) ------------------------------------------------------------------------ (683.00,401.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (682.0,404.0) (684,403) (685,400.01) ------------------------------------------------------------------------ (685.00,400.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (684.0,403.0) (686.0,402.0) (687.0,401.0) (688,398.01) ------------------------------------------------------------------------ (688.00,398.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (687.0,401.0) (689.0,400.0) (690.0,399.0) (690.0,399.0) (692.0,398.0) (693,395.01) ------------------------------------------------------------------------ (693.00,395.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (692.0,398.0) (694,397) (695,394.01) ------------------------------------------------------------------------ (695.00,394.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (694.0,397.0) (696,396) (696.0,396.0) (697.0,395.0) (698,392.01) ------------------------------------------------------------------------ (698.00,392.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (697.0,395.0) (699,394) (699.0,394.0) (700.0,393.0) (700.0,393.0) (702.0,392.0) (702.0,392.0) (703.0,391.0) (703.0,391.0) (705.0,390.0) (706,387.01) ------------------------------------------------------------------------ (706.00,387.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (705.0,390.0) (708,386.01) ------------------------------------------------------------------------ (708.00,386.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (707.0,389.0) (709,388) (709,385.01) ------------------------------------------------------------------------ (709.00,385.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (710,387) (711,384.01) ------------------------------------------------------------------------ (711.00,384.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (710.0,387.0) (712,386) (712.0,386.0) (713.0,385.0) (714,382.01) ------------------------------------------------------------------------ (714.00,382.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (713.0,385.0) (715,384) (716,381.01) ------------------------------------------------------------------------ (716.00,381.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (715.0,384.0) (717,383) (717.0,383.0) (718.0,382.0) (719,379.01) ------------------------------------------------------------------------ (719.00,379.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (718.0,382.0) (720,381) (720.0,381.0) (721.0,380.0) (722,377.01) ------------------------------------------------------------------------ (722.00,377.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (721.0,380.0) (723,379) (724,376.01) ------------------------------------------------------------------------ (724.00,376.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (723.0,379.0) (725,378) (725.0,378.0) (726.0,377.0) (727,374.01) ------------------------------------------------------------------------ (727.00,374.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (726.0,377.0) (728,376) (729,373.01) ------------------------------------------------------------------------ (729.00,373.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (728.0,376.0) (730,375) (730.0,375.0) (731.0,374.0) (732,371.01) ------------------------------------------------------------------------ (732.00,371.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (731.0,374.0) (733,373) (733.0,373.0) (734.0,372.0) (735,369.01) ------------------------------------------------------------------------ (735.00,369.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (734.0,372.0) (736,371) (736.0,371.0) (737.0,370.0) (738,367.01) ------------------------------------------------------------------------ (738.00,367.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (737.0,370.0) (739,369) (740,366.01) ------------------------------------------------------------------------ (740.00,366.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (739.0,369.0) (741,368) (741.0,368.0) (742.0,367.0) (743,364.01) ------------------------------------------------------------------------ (743.00,364.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (742.0,367.0) (744,366) (744.0,366.0) (745.0,365.0) (745.0,365.0) (747.0,364.0) (747.0,364.0) (748.0,363.0) (748.0,363.0) (750.0,362.0) (750.0,362.0) (751.0,361.0) (751.0,361.0) (753.0,360.0) (754,357.01) ------------------------------------------------------------------------ (754.00,357.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (753.0,360.0) (755,359) (756,356.01) ------------------------------------------------------------------------ (756.00,356.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (755.0,359.0) (757,358) (757,358) (757,355.01) ------------------------------------------------------------------------ (757.00,355.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (758,357) (758.0,357.0) (759.0,356.0) (759.0,356.0) (761.0,355.0) (762,352.01) ------------------------------------------------------------------------ (762.00,352.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (761.0,355.0) (763,354) (763,354) (763.0,354.0) (764.0,353.0) (765,350.01) ------------------------------------------------------------------------ (765.00,350.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (764.0,353.0) (766,352) (766.0,352.0) (767.0,351.0) (767.0,351.0) (769.0,350.0) (769.0,350.0) (770.0,349.0) (770.0,349.0) (772.0,348.0) (772.0,348.0) (773.0,347.0) (773.0,347.0) (775.0,346.0) (775.0,346.0) (776.0,345.0) (776.0,345.0) (778.0,344.0) (779,341.01) ------------------------------------------------------------------------ (779.00,341.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (778.0,344.0) (780,343) (780,343) (780.0,343.0) (781.0,342.0) (782,339.01) ------------------------------------------------------------------------ (782.00,339.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (781.0,342.0) (783,341) (783.0,341.0) (784.0,340.0) (784.0,340.0) (786.0,339.0) (786.0,339.0) (787.0,338.0) (787.0,338.0) (789.0,337.0) (789.0,337.0) (790.0,336.0) (790.0,336.0) (792.0,335.0) (793,332.01) ------------------------------------------------------------------------ (793.00,332.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (792.0,335.0) (794,334) (794,334) (794.0,334.0) (795.0,333.0) (796,330.01) ------------------------------------------------------------------------ (796.00,330.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (795.0,333.0) (797,332) (797,332) (797.0,332.0) (798.0,331.0) (799,328.01) ------------------------------------------------------------------------ (799.00,328.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (798.0,331.0) (800,330) (800,330) (800.0,330.0) (801.0,329.0) (801.0,329.0) (803.0,328.0) (803.0,328.0) (804.0,327.0) (804.0,327.0) (806.0,326.0) (806.0,326.0) (807.0,325.0) (807.0,325.0) (809.0,324.0) (809.0,324.0) (810.0,323.0) (810.0,323.0) (812.0,322.0) (812.0,322.0) (813.0,321.0) (813.0,321.0) (815.0,320.0) (816,317.01) ------------------------------------------------------------------------ (816.00,317.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (815.0,320.0) (817,319) (817,319) (817,319) (817.0,319.0) (818.0,318.0) (818.0,318.0) (819.0,317.0) (819.0,317.0) (821.0,316.0) (822,313.01) ------------------------------------------------------------------------ (822.00,313.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (821.0,316.0) (823,315) (823,315) (823,315) (823.0,315.0) (824.0,314.0) (825,311.01) ------------------------------------------------------------------------ (825.00,311.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (824.0,314.0) (826,313) (826,313) (826,313) (826.0,313.0) (827.0,312.0) (828,309.01) ------------------------------------------------------------------------ (828.00,309.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (827.0,312.0) (829,311) (829,311) (829,311) (829.0,311.0) (830.0,310.0) (830.0,310.0) (831.0,309.0) (831.0,309.0) (833.0,308.0) (833.0,308.0) (834.0,307.0) (834.0,307.0) (836.0,306.0) (837,303.01) ------------------------------------------------------------------------ (837.00,303.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (836.0,306.0) (838,305) (838,305) (838,305) (838.0,305.0) (839.0,304.0) (840,301.01) ------------------------------------------------------------------------ (840.00,301.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (839.0,304.0) (841,303) (841,303) (841,303) (841,303) (841.0,303.0) (842.0,302.0) (843,299.01) ------------------------------------------------------------------------ (843.00,299.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (842.0,302.0) (844,301) (844,301) (844,301) (844.0,301.0) (845.0,300.0) (846,297.01) ------------------------------------------------------------------------ (846.00,297.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (845.0,300.0) (847,299) (847,299) (847,299) (847.0,299.0) (848.0,298.0) (849,295.01) ------------------------------------------------------------------------ (849.00,295.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (848.0,298.0) (850,297) (850,297) (850,297) (850,297) (850.0,297.0) (851.0,296.0) (851.0,296.0) (852.0,295.0) (852.0,295.0) (854.0,294.0) (854.0,294.0) (855.0,293.0) (855.0,293.0) (857.0,292.0) (857.0,292.0) (858.0,291.0) (858.0,291.0) (860.0,290.0) (860.0,290.0) (861.0,289.0) (861.0,289.0) (863.0,288.0) (863.0,288.0) (864.0,287.0) (864.0,287.0) (866.0,286.0) (866.0,286.0) (867.0,285.0) (867.0,285.0) (869.0,284.0) (869.0,284.0) (870.0,283.0) (870.0,283.0) (872.0,282.0) (872.0,282.0) (873.0,281.0) (873.0,281.0) (875.0,280.0) (875.0,280.0) (876.0,279.0) (876.0,279.0) (878.0,278.0) (878.0,278.0) (879.0,277.0) (879.0,277.0) (881.0,276.0) (881.0,276.0) (882.0,275.0) (882.0,275.0) (884.0,274.0) (884.0,274.0) (885.0,273.0) (885.0,273.0) (887.0,272.0) (887.0,272.0) (888.0,271.0) (889,268.01) ------------------------------------------------------------------------ (889.00,268.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (888.0,271.0) (890,270) (890,270) (890,270) (890,270) (890.0,270.0) (891.0,269.0) (892,266.01) ------------------------------------------------------------------------ (892.00,266.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (891.0,269.0) (893,268) (893,268) (893,268) (893,268) (893,268) (893.0,268.0) (894.0,267.0) (895,264.01) ------------------------------------------------------------------------ (895.00,264.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (894.0,267.0) (896,266) (896,266) (896,266) (896,266) (896,266) (896.0,266.0) (897.0,265.0) (897.0,265.0) (898.0,264.0) (898.0,264.0) (900.0,263.0) (901,260.01) ------------------------------------------------------------------------ (901.00,260.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (900.0,263.0) (902,262) (902,262) (902,262) (902,262) (902,262) (902,262) (902.0,262.0) (903.0,261.0) (904,258.01) ------------------------------------------------------------------------ (904.00,258.51)(0.500,-1.000)[2]{} ------------------------------------------------------------------------ (903.0,261.0) (905,260) (905,260) (905,260) (905,260) (905,260) (905,260) (905.0,260.0) (906.0,259.0) (906.0,259.0) (907.0,258.0) (907.0,258.0) (909.0,257.0) (909.0,257.0) \[figbox\] Next, we study the effect of $\Omega$ on the velocity autocorrelation function of the DPD particles. In Fig. \[fig.omega\] we show the velocity correlation function for given $s=2.82, \mu=10.0$ and three different values of $\Omega=0.5,8.3,25$. We also plot the corresponding exponential terms in Eqn. (\[acvf02\]). For small $\Omega$ (small friction or high temperature) the decay of the vaf is very accurately given by the exponential term. As long as $\Omega$ increases, discrepancies from the exponential behavior are observed. This discrepancies are due to the effect of the collective term in Eqn. (\[acvf02\]). (974,675)(0,0) =cmr10 at 10pt (220.0,113.0) ------------------------------------------------------------------------ (900.0,113.0) ------------------------------------------------------------------------ (220.0,138.0) ------------------------------------------------------------------------ (900.0,138.0) ------------------------------------------------------------------------ (220.0,157.0) ------------------------------------------------------------------------ (900.0,157.0) ------------------------------------------------------------------------ (220.0,173.0) ------------------------------------------------------------------------ (900.0,173.0) ------------------------------------------------------------------------ (220.0,186.0) ------------------------------------------------------------------------ (900.0,186.0) ------------------------------------------------------------------------ (220.0,198.0) ------------------------------------------------------------------------ (900.0,198.0) ------------------------------------------------------------------------ (220.0,208.0) ------------------------------------------------------------------------ (900.0,208.0) ------------------------------------------------------------------------ (220.0,217.0) ------------------------------------------------------------------------ (198,217)[(0,0)\[r\][0.01]{}]{} (890.0,217.0) ------------------------------------------------------------------------ (220.0,278.0) ------------------------------------------------------------------------ (900.0,278.0) ------------------------------------------------------------------------ (220.0,313.0) ------------------------------------------------------------------------ (900.0,313.0) ------------------------------------------------------------------------ (220.0,338.0) ------------------------------------------------------------------------ (900.0,338.0) ------------------------------------------------------------------------ (220.0,357.0) ------------------------------------------------------------------------ (900.0,357.0) ------------------------------------------------------------------------ (220.0,373.0) ------------------------------------------------------------------------ (900.0,373.0) ------------------------------------------------------------------------ (220.0,386.0) ------------------------------------------------------------------------ (900.0,386.0) ------------------------------------------------------------------------ (220.0,398.0) ------------------------------------------------------------------------ (900.0,398.0) ------------------------------------------------------------------------ (220.0,408.0) ------------------------------------------------------------------------ (900.0,408.0) ------------------------------------------------------------------------ (220.0,417.0) ------------------------------------------------------------------------ (198,417)[(0,0)\[r\][0.1]{}]{} (890.0,417.0) ------------------------------------------------------------------------ (220.0,477.0) ------------------------------------------------------------------------ (900.0,477.0) ------------------------------------------------------------------------ (220.0,512.0) ------------------------------------------------------------------------ (900.0,512.0) ------------------------------------------------------------------------ (220.0,537.0) ------------------------------------------------------------------------ (900.0,537.0) ------------------------------------------------------------------------ (220.0,557.0) ------------------------------------------------------------------------ (900.0,557.0) ------------------------------------------------------------------------ (220.0,573.0) ------------------------------------------------------------------------ (900.0,573.0) ------------------------------------------------------------------------ (220.0,586.0) ------------------------------------------------------------------------ (900.0,586.0) ------------------------------------------------------------------------ (220.0,597.0) ------------------------------------------------------------------------ (900.0,597.0) ------------------------------------------------------------------------ (220.0,608.0) ------------------------------------------------------------------------ (900.0,608.0) ------------------------------------------------------------------------ (220.0,617.0) ------------------------------------------------------------------------ (198,617)[(0,0)\[r\][1]{}]{} (890.0,617.0) ------------------------------------------------------------------------ (220.0,113.0) ------------------------------------------------------------------------ (220,68)[(0,0)[0.01]{}]{} (220.0,632.0) ------------------------------------------------------------------------ (289.0,113.0) ------------------------------------------------------------------------ (289.0,642.0) ------------------------------------------------------------------------ (330.0,113.0) ------------------------------------------------------------------------ (330.0,642.0) ------------------------------------------------------------------------ (358.0,113.0) ------------------------------------------------------------------------ (358.0,642.0) ------------------------------------------------------------------------ (381.0,113.0) ------------------------------------------------------------------------ (381.0,642.0) ------------------------------------------------------------------------ (399.0,113.0) ------------------------------------------------------------------------ (399.0,642.0) ------------------------------------------------------------------------ (414.0,113.0) ------------------------------------------------------------------------ (414.0,642.0) ------------------------------------------------------------------------ (428.0,113.0) ------------------------------------------------------------------------ (428.0,642.0) ------------------------------------------------------------------------ (439.0,113.0) ------------------------------------------------------------------------ (439.0,642.0) ------------------------------------------------------------------------ (450.0,113.0) ------------------------------------------------------------------------ (450,68)[(0,0)[0.1]{}]{} (450.0,632.0) ------------------------------------------------------------------------ (519.0,113.0) ------------------------------------------------------------------------ (519.0,642.0) ------------------------------------------------------------------------ (560.0,113.0) ------------------------------------------------------------------------ (560.0,642.0) ------------------------------------------------------------------------ (588.0,113.0) ------------------------------------------------------------------------ (588.0,642.0) ------------------------------------------------------------------------ (611.0,113.0) ------------------------------------------------------------------------ (611.0,642.0) ------------------------------------------------------------------------ (629.0,113.0) ------------------------------------------------------------------------ (629.0,642.0) ------------------------------------------------------------------------ (644.0,113.0) ------------------------------------------------------------------------ (644.0,642.0) ------------------------------------------------------------------------ (658.0,113.0) ------------------------------------------------------------------------ (658.0,642.0) ------------------------------------------------------------------------ (669.0,113.0) ------------------------------------------------------------------------ (669.0,642.0) ------------------------------------------------------------------------ (680.0,113.0) ------------------------------------------------------------------------ (680,68)[(0,0)[1]{}]{} (680.0,632.0) ------------------------------------------------------------------------ (749.0,113.0) ------------------------------------------------------------------------ (749.0,642.0) ------------------------------------------------------------------------ (790.0,113.0) ------------------------------------------------------------------------ (790.0,642.0) ------------------------------------------------------------------------ (818.0,113.0) ------------------------------------------------------------------------ (818.0,642.0) ------------------------------------------------------------------------ (841.0,113.0) ------------------------------------------------------------------------ (841.0,642.0) ------------------------------------------------------------------------ (859.0,113.0) ------------------------------------------------------------------------ (859.0,642.0) ------------------------------------------------------------------------ (874.0,113.0) ------------------------------------------------------------------------ (874.0,642.0) ------------------------------------------------------------------------ (888.0,113.0) ------------------------------------------------------------------------ (888.0,642.0) ------------------------------------------------------------------------ (899.0,113.0) ------------------------------------------------------------------------ (899.0,642.0) ------------------------------------------------------------------------ (910.0,113.0) ------------------------------------------------------------------------ (910,68)[(0,0)[10]{}]{} (910.0,632.0) ------------------------------------------------------------------------ (220.0,113.0) ------------------------------------------------------------------------ (910.0,113.0) ------------------------------------------------------------------------ (220.0,652.0) ------------------------------------------------------------------------ (45,382)[(0,0)[$c({\overline t})$]{}]{} (565,23)[(0,0)[${\overline t}$]{}]{} (220.0,113.0) ------------------------------------------------------------------------ (269,585) (310,568) (339,552) (361,538) (379,524) (395,510) (408,498) (420,486) (430,475) (440,465) (448,455) (456,445) (464,436) (471,428) (477,420) (483,412) (489,405) (494,398) (499,392) (504,385) (509,379) (513,374) (518,368) (522,363) (526,358) (529,353) (533,349) (537,344) (540,340) (543,336) (546,332) (549,328) (552,324) (555,321) (558,318) (561,314) (564,311) (566,308) (569,305) (571,301) (574,298) (576,296) (578,293) (580,290) (583,288) (585,285) (587,282) (589,280) (591,279) (593,275) (595,273) (597,271) (599,269) (600,267) (602,265) (604,263) (606,261) (607,259) (609,258) (611,254) (612,252) (614,250) (616,249) (617,247) (619,245) (620,244) (622,243) (623,241) (625,240) (626,238) (627,237) (629,235) (630,234) (631,232) (633,230) (634,229) (635,227) (637,226) (638,225) (639,223) (640,221) (642,220) (643,218) (644,217) (645,216) (646,214) (647,213) (649,212) (650,211) (651,210) (652,208) (653,207) (654,205) (655,204) (656,203) (657,201) (658,200) (659,198) (660,199) (661,196) (662,194) (663,193) (664,192) (665,190) (666,189) (667,188) (668,187) (669,187) (670,188) (671,185) (671,184) (672,183) (673,181) (674,180) (675,179) (676,178) (677,177) (678,176) (678,176) (679,175) (680,174) (681,173) (682,172) (682,171) (683,170) (684,170) (685,169) (686,168) (686,166) (687,165) (688,164) (689,162) (689,160) (690,159) (691,158) (692,157) (692,156) (693,156) (694,157) (694,155) (695,154) (696,153) (697,151) (697,150) (698,150) (699,149) (699,149) (700,149) (701,147) (701,148) (702,146) (703,145) (703,143) (704,142) (705,141) (705,140) (706,140) (706,139) (707,137) (708,136) (708,134) (709,132) (710,130) (710,130) (711,129) (711,128) (712,128) (713,127) (713,128) (714,126) (714,125) (715,125) (716,125) (716,124) (717,123) (717,122) (718,121) (718,120) (719,120) (719,117) (720,117) (721,117) (721,117) (722,117) (722,117) (723,117) (723,118) (724,117) (361,615)[(0,0)[$+$]{}]{} (430,613)[(0,0)[$\circ$]{}]{} (471,611)[(0,0)[$\circ$]{}]{} (499,610)[(0,0)[$\circ$]{}]{} (522,608)[(0,0)[$\circ$]{}]{} (540,606)[(0,0)[$\circ$]{}]{} (555,604)[(0,0)[$\circ$]{}]{} (569,603)[(0,0)[$\circ$]{}]{} (580,601)[(0,0)[$\circ$]{}]{} (591,599)[(0,0)[$\circ$]{}]{} (600,598)[(0,0)[$\circ$]{}]{} (609,596)[(0,0)[$\circ$]{}]{} (617,594)[(0,0)[$\circ$]{}]{} (625,592)[(0,0)[$\circ$]{}]{} (631,591)[(0,0)[$\circ$]{}]{} (638,589)[(0,0)[$\circ$]{}]{} (644,587)[(0,0)[$\circ$]{}]{} (650,585)[(0,0)[$\circ$]{}]{} (655,584)[(0,0)[$\circ$]{}]{} (660,582)[(0,0)[$\circ$]{}]{} (665,580)[(0,0)[$\circ$]{}]{} (670,578)[(0,0)[$\circ$]{}]{} (674,577)[(0,0)[$\circ$]{}]{} (678,575)[(0,0)[$\circ$]{}]{} (682,573)[(0,0)[$\circ$]{}]{} (686,572)[(0,0)[$\circ$]{}]{} (690,570)[(0,0)[$\circ$]{}]{} (694,568)[(0,0)[$\circ$]{}]{} (697,566)[(0,0)[$\circ$]{}]{} (701,565)[(0,0)[$\circ$]{}]{} (704,563)[(0,0)[$\circ$]{}]{} (707,561)[(0,0)[$\circ$]{}]{} (710,559)[(0,0)[$\circ$]{}]{} (713,558)[(0,0)[$\circ$]{}]{} (716,556)[(0,0)[$\circ$]{}]{} (719,554)[(0,0)[$\circ$]{}]{} (722,552)[(0,0)[$\circ$]{}]{} (724,551)[(0,0)[$\circ$]{}]{} (727,549)[(0,0)[$\circ$]{}]{} (729,547)[(0,0)[$\circ$]{}]{} (732,545)[(0,0)[$\circ$]{}]{} (734,544)[(0,0)[$\circ$]{}]{} (737,542)[(0,0)[$\circ$]{}]{} (739,540)[(0,0)[$\circ$]{}]{} (741,538)[(0,0)[$\circ$]{}]{} (743,537)[(0,0)[$\circ$]{}]{} (746,535)[(0,0)[$\circ$]{}]{} (748,533)[(0,0)[$\circ$]{}]{} (750,532)[(0,0)[$\circ$]{}]{} (752,530)[(0,0)[$\circ$]{}]{} (754,528)[(0,0)[$\circ$]{}]{} (756,526)[(0,0)[$\circ$]{}]{} (758,525)[(0,0)[$\circ$]{}]{} (759,523)[(0,0)[$\circ$]{}]{} (761,521)[(0,0)[$\circ$]{}]{} (763,519)[(0,0)[$\circ$]{}]{} (765,518)[(0,0)[$\circ$]{}]{} (767,516)[(0,0)[$\circ$]{}]{} (768,514)[(0,0)[$\circ$]{}]{} (770,512)[(0,0)[$\circ$]{}]{} (772,511)[(0,0)[$\circ$]{}]{} (773,509)[(0,0)[$\circ$]{}]{} (775,507)[(0,0)[$\circ$]{}]{} (776,505)[(0,0)[$\circ$]{}]{} (778,504)[(0,0)[$\circ$]{}]{} (779,502)[(0,0)[$\circ$]{}]{} (781,500)[(0,0)[$\circ$]{}]{} (782,498)[(0,0)[$\circ$]{}]{} (784,496)[(0,0)[$\circ$]{}]{} (785,495)[(0,0)[$\circ$]{}]{} (787,493)[(0,0)[$\circ$]{}]{} (788,491)[(0,0)[$\circ$]{}]{} (790,489)[(0,0)[$\circ$]{}]{} (791,488)[(0,0)[$\circ$]{}]{} (792,486)[(0,0)[$\circ$]{}]{} (794,484)[(0,0)[$\circ$]{}]{} (795,482)[(0,0)[$\circ$]{}]{} (796,481)[(0,0)[$\circ$]{}]{} (797,479)[(0,0)[$\circ$]{}]{} (799,477)[(0,0)[$\circ$]{}]{} (800,475)[(0,0)[$\circ$]{}]{} (801,474)[(0,0)[$\circ$]{}]{} (802,472)[(0,0)[$\circ$]{}]{} (804,470)[(0,0)[$\circ$]{}]{} (805,468)[(0,0)[$\circ$]{}]{} (806,466)[(0,0)[$\circ$]{}]{} (807,465)[(0,0)[$\circ$]{}]{} (808,463)[(0,0)[$\circ$]{}]{} (809,461)[(0,0)[$\circ$]{}]{} (810,459)[(0,0)[$\circ$]{}]{} (812,458)[(0,0)[$\circ$]{}]{} (813,456)[(0,0)[$\circ$]{}]{} (814,454)[(0,0)[$\circ$]{}]{} (815,452)[(0,0)[$\circ$]{}]{} (816,450)[(0,0)[$\circ$]{}]{} (817,449)[(0,0)[$\circ$]{}]{} (818,447)[(0,0)[$\circ$]{}]{} (819,445)[(0,0)[$\circ$]{}]{} (820,443)[(0,0)[$\circ$]{}]{} (821,441)[(0,0)[$\circ$]{}]{} (822,440)[(0,0)[$\circ$]{}]{} (823,438)[(0,0)[$\circ$]{}]{} (824,436)[(0,0)[$\circ$]{}]{} (825,434)[(0,0)[$\circ$]{}]{} (826,433)[(0,0)[$\circ$]{}]{} (827,431)[(0,0)[$\circ$]{}]{} (828,429)[(0,0)[$\circ$]{}]{} (829,427)[(0,0)[$\circ$]{}]{} (830,425)[(0,0)[$\circ$]{}]{} (830,424)[(0,0)[$\circ$]{}]{} (831,422)[(0,0)[$\circ$]{}]{} (832,420)[(0,0)[$\circ$]{}]{} (833,418)[(0,0)[$\circ$]{}]{} (834,417)[(0,0)[$\circ$]{}]{} (835,415)[(0,0)[$\circ$]{}]{} (836,413)[(0,0)[$\circ$]{}]{} (837,411)[(0,0)[$\circ$]{}]{} (837,410)[(0,0)[$\circ$]{}]{} (838,408)[(0,0)[$\circ$]{}]{} (839,406)[(0,0)[$\circ$]{}]{} (840,404)[(0,0)[$\circ$]{}]{} (841,403)[(0,0)[$\circ$]{}]{} (842,401)[(0,0)[$\circ$]{}]{} (842,399)[(0,0)[$\circ$]{}]{} (843,397)[(0,0)[$\circ$]{}]{} (844,395)[(0,0)[$\circ$]{}]{} (845,394)[(0,0)[$\circ$]{}]{} (846,392)[(0,0)[$\circ$]{}]{} (846,390)[(0,0)[$\circ$]{}]{} (847,388)[(0,0)[$\circ$]{}]{} (848,386)[(0,0)[$\circ$]{}]{} (849,384)[(0,0)[$\circ$]{}]{} (849,383)[(0,0)[$\circ$]{}]{} (850,381)[(0,0)[$\circ$]{}]{} (851,379)[(0,0)[$\circ$]{}]{} (852,377)[(0,0)[$\circ$]{}]{} (852,375)[(0,0)[$\circ$]{}]{} (853,373)[(0,0)[$\circ$]{}]{} (854,371)[(0,0)[$\circ$]{}]{} (855,370)[(0,0)[$\circ$]{}]{} (855,368)[(0,0)[$\circ$]{}]{} (856,366)[(0,0)[$\circ$]{}]{} (857,364)[(0,0)[$\circ$]{}]{} (857,362)[(0,0)[$\circ$]{}]{} (858,360)[(0,0)[$\circ$]{}]{} (859,358)[(0,0)[$\circ$]{}]{} (859,357)[(0,0)[$\circ$]{}]{} (860,355)[(0,0)[$\circ$]{}]{} (861,353)[(0,0)[$\circ$]{}]{} (861,351)[(0,0)[$\circ$]{}]{} (862,349)[(0,0)[$\circ$]{}]{} (863,347)[(0,0)[$\circ$]{}]{} (863,345)[(0,0)[$\circ$]{}]{} (864,343)[(0,0)[$\circ$]{}]{} (865,341)[(0,0)[$\circ$]{}]{} (865,339)[(0,0)[$\circ$]{}]{} (866,337)[(0,0)[$\circ$]{}]{} (867,335)[(0,0)[$\circ$]{}]{} (867,333)[(0,0)[$\circ$]{}]{} (868,331)[(0,0)[$\circ$]{}]{} (869,329)[(0,0)[$\circ$]{}]{} (869,327)[(0,0)[$\circ$]{}]{} (870,325)[(0,0)[$\circ$]{}]{} (870,323)[(0,0)[$\circ$]{}]{} (871,321)[(0,0)[$\circ$]{}]{} (872,319)[(0,0)[$\circ$]{}]{} (872,317)[(0,0)[$\circ$]{}]{} (873,315)[(0,0)[$\circ$]{}]{} (873,313)[(0,0)[$\circ$]{}]{} (874,312)[(0,0)[$\circ$]{}]{} (875,310)[(0,0)[$\circ$]{}]{} (875,308)[(0,0)[$\circ$]{}]{} (876,305)[(0,0)[$\circ$]{}]{} (876,303)[(0,0)[$\circ$]{}]{} (877,301)[(0,0)[$\circ$]{}]{} (877,299)[(0,0)[$\circ$]{}]{} (878,297)[(0,0)[$\circ$]{}]{} (879,295)[(0,0)[$\circ$]{}]{} (879,293)[(0,0)[$\circ$]{}]{} (880,291)[(0,0)[$\circ$]{}]{} (880,289)[(0,0)[$\circ$]{}]{} (881,287)[(0,0)[$\circ$]{}]{} (881,285)[(0,0)[$\circ$]{}]{} (882,283)[(0,0)[$\circ$]{}]{} (882,281)[(0,0)[$\circ$]{}]{} (883,279)[(0,0)[$\circ$]{}]{} (883,277)[(0,0)[$\circ$]{}]{} (884,274)[(0,0)[$\circ$]{}]{} (885,272)[(0,0)[$\circ$]{}]{} (885,270)[(0,0)[$\circ$]{}]{} (886,268)[(0,0)[$\circ$]{}]{} (886,266)[(0,0)[$\circ$]{}]{} (887,263)[(0,0)[$\circ$]{}]{} (887,261)[(0,0)[$\circ$]{}]{} (888,259)[(0,0)[$\circ$]{}]{} (888,257)[(0,0)[$\circ$]{}]{} (889,255)[(0,0)[$\circ$]{}]{} (889,253)[(0,0)[$\circ$]{}]{} (890,250)[(0,0)[$\circ$]{}]{} (890,248)[(0,0)[$\circ$]{}]{} (891,246)[(0,0)[$\circ$]{}]{} (891,244)[(0,0)[$\circ$]{}]{} (892,241)[(0,0)[$\circ$]{}]{} (892,239)[(0,0)[$\circ$]{}]{} (893,237)[(0,0)[$\circ$]{}]{} (893,234)[(0,0)[$\circ$]{}]{} (894,232)[(0,0)[$\circ$]{}]{} (894,230)[(0,0)[$\circ$]{}]{} (895,228)[(0,0)[$\circ$]{}]{} (895,225)[(0,0)[$\circ$]{}]{} (896,223)[(0,0)[$\circ$]{}]{} (896,220)[(0,0)[$\circ$]{}]{} (896,218)[(0,0)[$\circ$]{}]{} (897,215)[(0,0)[$\circ$]{}]{} (897,213)[(0,0)[$\circ$]{}]{} (898,210)[(0,0)[$\circ$]{}]{} (898,207)[(0,0)[$\circ$]{}]{} (899,205)[(0,0)[$\circ$]{}]{} (899,202)[(0,0)[$\circ$]{}]{} (900,199)[(0,0)[$\circ$]{}]{} (900,197)[(0,0)[$\circ$]{}]{} (901,194)[(0,0)[$\circ$]{}]{} (901,191)[(0,0)[$\circ$]{}]{} (901,189)[(0,0)[$\circ$]{}]{} (902,186)[(0,0)[$\circ$]{}]{} (902,183)[(0,0)[$\circ$]{}]{} (903,181)[(0,0)[$\circ$]{}]{} (903,179)[(0,0)[$\circ$]{}]{} (904,176)[(0,0)[$\circ$]{}]{} (904,173)[(0,0)[$\circ$]{}]{} (905,171)[(0,0)[$\circ$]{}]{} (905,169)[(0,0)[$\circ$]{}]{} (905,166)[(0,0)[$\circ$]{}]{} (906,164)[(0,0)[$\circ$]{}]{} (906,161)[(0,0)[$\circ$]{}]{} (907,158)[(0,0)[$\circ$]{}]{} (907,156)[(0,0)[$\circ$]{}]{} (908,155)[(0,0)[$\circ$]{}]{} (908,153)[(0,0)[$\circ$]{}]{} (908,151)[(0,0)[$\circ$]{}]{} (909,149)[(0,0)[$\circ$]{}]{} (909,147)[(0,0)[$\circ$]{}]{} (910,146)[(0,0)[$\circ$]{}]{} (361,589) (430,562) (471,537) (499,514) (522,492) (540,471) (555,452) (569,433) (580,416) (591,400) (600,385) (609,371) (617,357) (625,345) (631,334) (638,322) (644,312) (650,302) (655,293) (660,283) (665,276) (670,267) (674,258) (678,250) (682,244) (686,238) (690,231) (694,225) (697,217) (701,209) (704,201) (707,192) (710,182) (713,174) (716,170) (719,166) (722,162) (724,160) (727,155) (729,151) (732,148) (734,146) (737,140) (739,133) (741,131) (743,117) (746,115) (761,116) (763,119) (765,115) (767,113) (220,597) (220,595.67) ------------------------------------------------------------------------ (220.00,596.17)(3.500,-1.000)[2]{} ------------------------------------------------------------------------ (227,594.17) ------------------------------------------------------------------------ (227.00,595.17)(3.887,-2.000)[2]{} ------------------------------------------------------------------------ (234,592.17) ------------------------------------------------------------------------ (234.00,593.17)(3.887,-2.000)[2]{} ------------------------------------------------------------------------ (241,590.17) ------------------------------------------------------------------------ (241.00,591.17)(3.887,-2.000)[2]{} ------------------------------------------------------------------------ (248,588.17) ------------------------------------------------------------------------ (248.00,589.17)(3.887,-2.000)[2]{} ------------------------------------------------------------------------ (255,586.17) ------------------------------------------------------------------------ (255.00,587.17)(3.887,-2.000)[2]{} ------------------------------------------------------------------------ (262,584.17) ------------------------------------------------------------------------ (262.00,585.17)(3.887,-2.000)[2]{} ------------------------------------------------------------------------ (269.00,582.95)(1.355,-0.447)[3]{} ------------------------------------------------------------------------ (269.00,583.17)(4.855,-3.000)[2]{} ------------------------------------------------------------------------ (276.00,579.95)(1.355,-0.447)[3]{} ------------------------------------------------------------------------ (276.00,580.17)(4.855,-3.000)[2]{} ------------------------------------------------------------------------ (283.00,576.95)(1.355,-0.447)[3]{} ------------------------------------------------------------------------ (283.00,577.17)(4.855,-3.000)[2]{} ------------------------------------------------------------------------ (290.00,573.95)(1.355,-0.447)[3]{} ------------------------------------------------------------------------ (290.00,574.17)(4.855,-3.000)[2]{} ------------------------------------------------------------------------ (297.00,570.95)(1.355,-0.447)[3]{} ------------------------------------------------------------------------ (297.00,571.17)(4.855,-3.000)[2]{} ------------------------------------------------------------------------ (304.00,567.94)(0.920,-0.468)[5]{} ------------------------------------------------------------------------ (304.00,568.17)(5.340,-4.000)[2]{} ------------------------------------------------------------------------ (311.00,563.94)(0.920,-0.468)[5]{} ------------------------------------------------------------------------ (311.00,564.17)(5.340,-4.000)[2]{} ------------------------------------------------------------------------ (318.00,559.94)(0.920,-0.468)[5]{} ------------------------------------------------------------------------ (318.00,560.17)(5.340,-4.000)[2]{} ------------------------------------------------------------------------ (325.00,555.94)(0.920,-0.468)[5]{} ------------------------------------------------------------------------ (325.00,556.17)(5.340,-4.000)[2]{} ------------------------------------------------------------------------ (332.00,551.93)(0.599,-0.477)[7]{} ------------------------------------------------------------------------ (332.00,552.17)(4.796,-5.000)[2]{} ------------------------------------------------------------------------ (338.00,546.93)(0.710,-0.477)[7]{} ------------------------------------------------------------------------ (338.00,547.17)(5.630,-5.000)[2]{} ------------------------------------------------------------------------ (345.00,541.93)(0.581,-0.482)[9]{} ------------------------------------------------------------------------ (345.00,542.17)(5.824,-6.000)[2]{} ------------------------------------------------------------------------ (352.00,535.93)(0.581,-0.482)[9]{} ------------------------------------------------------------------------ (352.00,536.17)(5.824,-6.000)[2]{} ------------------------------------------------------------------------ (359.00,529.93)(0.581,-0.482)[9]{} ------------------------------------------------------------------------ (359.00,530.17)(5.824,-6.000)[2]{} ------------------------------------------------------------------------ (366.00,523.93)(0.492,-0.485)[11]{} ------------------------------------------------------------------------ (366.00,524.17)(5.962,-7.000)[2]{} ------------------------------------------------------------------------ (373.00,516.93)(0.492,-0.485)[11]{} ------------------------------------------------------------------------ (373.00,517.17)(5.962,-7.000)[2]{} ------------------------------------------------------------------------ (380.59,508.69)(0.485,-0.569)[11]{} ------------------------------------------------------------------------ (379.17,509.84)(7.000,-6.844)[2]{} ------------------------------------------------------------------------ (387.59,500.69)(0.485,-0.569)[11]{} ------------------------------------------------------------------------ (386.17,501.84)(7.000,-6.844)[2]{} ------------------------------------------------------------------------ (394.59,492.45)(0.485,-0.645)[11]{} ------------------------------------------------------------------------ (393.17,493.73)(7.000,-7.725)[2]{} ------------------------------------------------------------------------ (401.59,483.21)(0.485,-0.721)[11]{} ------------------------------------------------------------------------ (400.17,484.61)(7.000,-8.606)[2]{} ------------------------------------------------------------------------ (408.59,473.21)(0.485,-0.721)[11]{} ------------------------------------------------------------------------ (407.17,474.61)(7.000,-8.606)[2]{} ------------------------------------------------------------------------ (415.59,462.98)(0.485,-0.798)[11]{} ------------------------------------------------------------------------ (414.17,464.49)(7.000,-9.488)[2]{} ------------------------------------------------------------------------ (422.59,451.74)(0.485,-0.874)[11]{} ------------------------------------------------------------------------ (421.17,453.37)(7.000,-10.369)[2]{} ------------------------------------------------------------------------ (429.59,439.50)(0.485,-0.950)[11]{} ------------------------------------------------------------------------ (428.17,441.25)(7.000,-11.251)[2]{} ------------------------------------------------------------------------ (436.59,426.50)(0.485,-0.950)[11]{} ------------------------------------------------------------------------ (435.17,428.25)(7.000,-11.251)[2]{} ------------------------------------------------------------------------ (443.59,413.03)(0.485,-1.103)[11]{} ------------------------------------------------------------------------ (442.17,415.01)(7.000,-13.013)[2]{} ------------------------------------------------------------------------ (450.59,397.79)(0.485,-1.179)[11]{} ------------------------------------------------------------------------ (449.17,399.89)(7.000,-13.895)[2]{} ------------------------------------------------------------------------ (457.59,381.79)(0.485,-1.179)[11]{} ------------------------------------------------------------------------ (456.17,383.89)(7.000,-13.895)[2]{} ------------------------------------------------------------------------ (464.59,365.32)(0.485,-1.332)[11]{} ------------------------------------------------------------------------ (463.17,367.66)(7.000,-15.658)[2]{} ------------------------------------------------------------------------ (471.59,346.84)(0.485,-1.484)[11]{} ------------------------------------------------------------------------ (470.17,349.42)(7.000,-17.420)[2]{} ------------------------------------------------------------------------ (478.59,326.84)(0.485,-1.484)[11]{} ------------------------------------------------------------------------ (477.17,329.42)(7.000,-17.420)[2]{} ------------------------------------------------------------------------ (485.59,306.13)(0.485,-1.713)[11]{} ------------------------------------------------------------------------ (484.17,309.06)(7.000,-20.065)[2]{} ------------------------------------------------------------------------ (492.59,283.13)(0.485,-1.713)[11]{} ------------------------------------------------------------------------ (491.17,286.06)(7.000,-20.065)[2]{} ------------------------------------------------------------------------ (499.59,259.42)(0.485,-1.942)[11]{} ------------------------------------------------------------------------ (498.17,262.71)(7.000,-22.709)[2]{} ------------------------------------------------------------------------ (506.59,233.18)(0.485,-2.018)[11]{} ------------------------------------------------------------------------ (505.17,236.59)(7.000,-23.590)[2]{} ------------------------------------------------------------------------ (513.59,205.47)(0.485,-2.247)[11]{} ------------------------------------------------------------------------ (512.17,209.23)(7.000,-26.234)[2]{} ------------------------------------------------------------------------ (520.59,175.23)(0.485,-2.323)[11]{} ------------------------------------------------------------------------ (519.17,179.12)(7.000,-27.116)[2]{} ------------------------------------------------------------------------ (527.59,143.52)(0.485,-2.552)[11]{} ------------------------------------------------------------------------ (526.17,147.76)(7.000,-29.760)[2]{} ------------------------------------------------------------------------ (533.67,113) ------------------------------------------------------------------------ (533.17,115.50)(1.000,-2.500)[2]{} ------------------------------------------------------------------------ (220,616) (332,614.67) ------------------------------------------------------------------------ (332.00,615.17)(3.000,-1.000)[2]{} ------------------------------------------------------------------------ (220.0,616.0) ------------------------------------------------------------------------ (387,613.67) ------------------------------------------------------------------------ (387.00,614.17)(3.500,-1.000)[2]{} ------------------------------------------------------------------------ (338.0,615.0) ------------------------------------------------------------------------ (422,612.67) ------------------------------------------------------------------------ (422.00,613.17)(3.500,-1.000)[2]{} ------------------------------------------------------------------------ (394.0,614.0) ------------------------------------------------------------------------ (443,611.67) ------------------------------------------------------------------------ (443.00,612.17)(3.500,-1.000)[2]{} ------------------------------------------------------------------------ (429.0,613.0) ------------------------------------------------------------------------ (464,610.67) ------------------------------------------------------------------------ (464.00,611.17)(3.500,-1.000)[2]{} ------------------------------------------------------------------------ (450.0,612.0) ------------------------------------------------------------------------ (485,609.67) ------------------------------------------------------------------------ (485.00,610.17)(3.500,-1.000)[2]{} ------------------------------------------------------------------------ (471.0,611.0) ------------------------------------------------------------------------ (499,608.67) ------------------------------------------------------------------------ (499.00,609.17)(3.500,-1.000)[2]{} ------------------------------------------------------------------------ (492.0,610.0) ------------------------------------------------------------------------ (513,607.67) ------------------------------------------------------------------------ (513.00,608.17)(3.500,-1.000)[2]{} ------------------------------------------------------------------------ (520,606.67) ------------------------------------------------------------------------ (520.00,607.17)(3.500,-1.000)[2]{} ------------------------------------------------------------------------ (506.0,609.0) ------------------------------------------------------------------------ (534,605.67) ------------------------------------------------------------------------ (534.00,606.17)(3.500,-1.000)[2]{} ------------------------------------------------------------------------ (541,604.67) ------------------------------------------------------------------------ (541.00,605.17)(3.500,-1.000)[2]{} ------------------------------------------------------------------------ (548,603.67) ------------------------------------------------------------------------ (548.00,604.17)(3.500,-1.000)[2]{} ------------------------------------------------------------------------ (527.0,607.0) ------------------------------------------------------------------------ (562,602.67) ------------------------------------------------------------------------ (562.00,603.17)(3.000,-1.000)[2]{} ------------------------------------------------------------------------ (568,601.67) ------------------------------------------------------------------------ (568.00,602.17)(3.500,-1.000)[2]{} ------------------------------------------------------------------------ (575,600.67) ------------------------------------------------------------------------ (575.00,601.17)(3.500,-1.000)[2]{} ------------------------------------------------------------------------ (582,599.17) ------------------------------------------------------------------------ (582.00,600.17)(3.887,-2.000)[2]{} ------------------------------------------------------------------------ (589,597.67) ------------------------------------------------------------------------ (589.00,598.17)(3.500,-1.000)[2]{} ------------------------------------------------------------------------ (596,596.67) ------------------------------------------------------------------------ (596.00,597.17)(3.500,-1.000)[2]{} ------------------------------------------------------------------------ (603,595.17) ------------------------------------------------------------------------ (603.00,596.17)(3.887,-2.000)[2]{} ------------------------------------------------------------------------ (610,593.67) ------------------------------------------------------------------------ (610.00,594.17)(3.500,-1.000)[2]{} ------------------------------------------------------------------------ (617,592.17) ------------------------------------------------------------------------ (617.00,593.17)(3.887,-2.000)[2]{} ------------------------------------------------------------------------ (624,590.17) ------------------------------------------------------------------------ (624.00,591.17)(3.887,-2.000)[2]{} ------------------------------------------------------------------------ (631,588.17) ------------------------------------------------------------------------ (631.00,589.17)(3.887,-2.000)[2]{} ------------------------------------------------------------------------ (638,586.17) ------------------------------------------------------------------------ (638.00,587.17)(3.887,-2.000)[2]{} ------------------------------------------------------------------------ (645,584.17) ------------------------------------------------------------------------ (645.00,585.17)(3.887,-2.000)[2]{} ------------------------------------------------------------------------ (652,582.17) ------------------------------------------------------------------------ (652.00,583.17)(3.887,-2.000)[2]{} ------------------------------------------------------------------------ (659.00,580.95)(1.355,-0.447)[3]{} ------------------------------------------------------------------------ (659.00,581.17)(4.855,-3.000)[2]{} ------------------------------------------------------------------------ (666.00,577.95)(1.355,-0.447)[3]{} ------------------------------------------------------------------------ (666.00,578.17)(4.855,-3.000)[2]{} ------------------------------------------------------------------------ (673.00,574.95)(1.355,-0.447)[3]{} ------------------------------------------------------------------------ (673.00,575.17)(4.855,-3.000)[2]{} ------------------------------------------------------------------------ (680.00,571.95)(1.355,-0.447)[3]{} ------------------------------------------------------------------------ (680.00,572.17)(4.855,-3.000)[2]{} ------------------------------------------------------------------------ (687.00,568.95)(1.355,-0.447)[3]{} ------------------------------------------------------------------------ (687.00,569.17)(4.855,-3.000)[2]{} ------------------------------------------------------------------------ (694.00,565.94)(0.920,-0.468)[5]{} ------------------------------------------------------------------------ (694.00,566.17)(5.340,-4.000)[2]{} ------------------------------------------------------------------------ (701.00,561.95)(1.355,-0.447)[3]{} ------------------------------------------------------------------------ (701.00,562.17)(4.855,-3.000)[2]{} ------------------------------------------------------------------------ (708.00,558.93)(0.710,-0.477)[7]{} ------------------------------------------------------------------------ (708.00,559.17)(5.630,-5.000)[2]{} ------------------------------------------------------------------------ (715.00,553.94)(0.920,-0.468)[5]{} ------------------------------------------------------------------------ (715.00,554.17)(5.340,-4.000)[2]{} ------------------------------------------------------------------------ (722.00,549.93)(0.710,-0.477)[7]{} ------------------------------------------------------------------------ (722.00,550.17)(5.630,-5.000)[2]{} ------------------------------------------------------------------------ (729.00,544.93)(0.710,-0.477)[7]{} ------------------------------------------------------------------------ (729.00,545.17)(5.630,-5.000)[2]{} ------------------------------------------------------------------------ (736.00,539.93)(0.710,-0.477)[7]{} ------------------------------------------------------------------------ (736.00,540.17)(5.630,-5.000)[2]{} ------------------------------------------------------------------------ (743.00,534.93)(0.581,-0.482)[9]{} ------------------------------------------------------------------------ (743.00,535.17)(5.824,-6.000)[2]{} ------------------------------------------------------------------------ (750.00,528.93)(0.492,-0.485)[11]{} ------------------------------------------------------------------------ (750.00,529.17)(5.962,-7.000)[2]{} ------------------------------------------------------------------------ (757.00,521.93)(0.581,-0.482)[9]{} ------------------------------------------------------------------------ (757.00,522.17)(5.824,-6.000)[2]{} ------------------------------------------------------------------------ (764.59,514.69)(0.485,-0.569)[11]{} ------------------------------------------------------------------------ (763.17,515.84)(7.000,-6.844)[2]{} ------------------------------------------------------------------------ (771.00,507.93)(0.492,-0.485)[11]{} ------------------------------------------------------------------------ (771.00,508.17)(5.962,-7.000)[2]{} ------------------------------------------------------------------------ (778.59,499.45)(0.485,-0.645)[11]{} ------------------------------------------------------------------------ (777.17,500.73)(7.000,-7.725)[2]{} ------------------------------------------------------------------------ (785.59,490.45)(0.485,-0.645)[11]{} ------------------------------------------------------------------------ (784.17,491.73)(7.000,-7.725)[2]{} ------------------------------------------------------------------------ (792.59,481.09)(0.482,-0.762)[9]{} ------------------------------------------------------------------------ (791.17,482.55)(6.000,-7.547)[2]{} ------------------------------------------------------------------------ (798.59,472.21)(0.485,-0.721)[11]{} ------------------------------------------------------------------------ (797.17,473.61)(7.000,-8.606)[2]{} ------------------------------------------------------------------------ (805.59,461.98)(0.485,-0.798)[11]{} ------------------------------------------------------------------------ (804.17,463.49)(7.000,-9.488)[2]{} ------------------------------------------------------------------------ (812.59,450.74)(0.485,-0.874)[11]{} ------------------------------------------------------------------------ (811.17,452.37)(7.000,-10.369)[2]{} ------------------------------------------------------------------------ (819.59,438.50)(0.485,-0.950)[11]{} ------------------------------------------------------------------------ (818.17,440.25)(7.000,-11.251)[2]{} ------------------------------------------------------------------------ (826.59,425.50)(0.485,-0.950)[11]{} ------------------------------------------------------------------------ (825.17,427.25)(7.000,-11.251)[2]{} ------------------------------------------------------------------------ (833.59,412.03)(0.485,-1.103)[11]{} ------------------------------------------------------------------------ (832.17,414.01)(7.000,-13.013)[2]{} ------------------------------------------------------------------------ (840.59,396.79)(0.485,-1.179)[11]{} ------------------------------------------------------------------------ (839.17,398.89)(7.000,-13.895)[2]{} ------------------------------------------------------------------------ (847.59,380.79)(0.485,-1.179)[11]{} ------------------------------------------------------------------------ (846.17,382.89)(7.000,-13.895)[2]{} ------------------------------------------------------------------------ (854.59,364.32)(0.485,-1.332)[11]{} ------------------------------------------------------------------------ (853.17,366.66)(7.000,-15.658)[2]{} ------------------------------------------------------------------------ (861.59,346.08)(0.485,-1.408)[11]{} ------------------------------------------------------------------------ (860.17,348.54)(7.000,-16.539)[2]{} ------------------------------------------------------------------------ (868.59,326.60)(0.485,-1.560)[11]{} ------------------------------------------------------------------------ (867.17,329.30)(7.000,-18.302)[2]{} ------------------------------------------------------------------------ (875.59,305.37)(0.485,-1.637)[11]{} ------------------------------------------------------------------------ (874.17,308.18)(7.000,-19.183)[2]{} ------------------------------------------------------------------------ (882.59,282.89)(0.485,-1.789)[11]{} ------------------------------------------------------------------------ (881.17,285.95)(7.000,-20.946)[2]{} ------------------------------------------------------------------------ (889.59,258.65)(0.485,-1.865)[11]{} ------------------------------------------------------------------------ (888.17,261.83)(7.000,-21.827)[2]{} ------------------------------------------------------------------------ (896.59,232.94)(0.485,-2.094)[11]{} ------------------------------------------------------------------------ (895.17,236.47)(7.000,-24.472)[2]{} ------------------------------------------------------------------------ (903.59,204.71)(0.485,-2.171)[11]{} ------------------------------------------------------------------------ (902.17,208.35)(7.000,-25.353)[2]{} ------------------------------------------------------------------------ (555.0,604.0) ------------------------------------------------------------------------ (220,610) (227,608.67) ------------------------------------------------------------------------ (227.00,609.17)(3.500,-1.000)[2]{} ------------------------------------------------------------------------ (220.0,610.0) ------------------------------------------------------------------------ (241,607.67) ------------------------------------------------------------------------ (241.00,608.17)(3.500,-1.000)[2]{} ------------------------------------------------------------------------ (248,606.67) ------------------------------------------------------------------------ (248.00,607.17)(3.500,-1.000)[2]{} ------------------------------------------------------------------------ (255,605.67) ------------------------------------------------------------------------ (255.00,606.17)(3.500,-1.000)[2]{} ------------------------------------------------------------------------ (234.0,609.0) ------------------------------------------------------------------------ (269,604.67) ------------------------------------------------------------------------ (269.00,605.17)(3.500,-1.000)[2]{} ------------------------------------------------------------------------ (276,603.67) ------------------------------------------------------------------------ (276.00,604.17)(3.500,-1.000)[2]{} ------------------------------------------------------------------------ (283,602.67) ------------------------------------------------------------------------ (283.00,603.17)(3.500,-1.000)[2]{} ------------------------------------------------------------------------ (290,601.67) ------------------------------------------------------------------------ (290.00,602.17)(3.500,-1.000)[2]{} ------------------------------------------------------------------------ (297,600.67) ------------------------------------------------------------------------ (297.00,601.17)(3.500,-1.000)[2]{} ------------------------------------------------------------------------ (304,599.67) ------------------------------------------------------------------------ (304.00,600.17)(3.500,-1.000)[2]{} ------------------------------------------------------------------------ (311,598.17) ------------------------------------------------------------------------ (311.00,599.17)(3.887,-2.000)[2]{} ------------------------------------------------------------------------ (318,596.67) ------------------------------------------------------------------------ (318.00,597.17)(3.500,-1.000)[2]{} ------------------------------------------------------------------------ (325,595.17) ------------------------------------------------------------------------ (325.00,596.17)(3.887,-2.000)[2]{} ------------------------------------------------------------------------ (332,593.67) ------------------------------------------------------------------------ (332.00,594.17)(3.000,-1.000)[2]{} ------------------------------------------------------------------------ (338,592.17) ------------------------------------------------------------------------ (338.00,593.17)(3.887,-2.000)[2]{} ------------------------------------------------------------------------ (345,590.17) ------------------------------------------------------------------------ (345.00,591.17)(3.887,-2.000)[2]{} ------------------------------------------------------------------------ (352,588.17) ------------------------------------------------------------------------ (352.00,589.17)(3.887,-2.000)[2]{} ------------------------------------------------------------------------ (359,586.17) ------------------------------------------------------------------------ (359.00,587.17)(3.887,-2.000)[2]{} ------------------------------------------------------------------------ (366,584.17) ------------------------------------------------------------------------ (366.00,585.17)(3.887,-2.000)[2]{} ------------------------------------------------------------------------ (373,582.17) ------------------------------------------------------------------------ (373.00,583.17)(3.887,-2.000)[2]{} ------------------------------------------------------------------------ (380.00,580.95)(1.355,-0.447)[3]{} ------------------------------------------------------------------------ (380.00,581.17)(4.855,-3.000)[2]{} ------------------------------------------------------------------------ (387.00,577.95)(1.355,-0.447)[3]{} ------------------------------------------------------------------------ (387.00,578.17)(4.855,-3.000)[2]{} ------------------------------------------------------------------------ (394.00,574.95)(1.355,-0.447)[3]{} ------------------------------------------------------------------------ (394.00,575.17)(4.855,-3.000)[2]{} ------------------------------------------------------------------------ (401.00,571.95)(1.355,-0.447)[3]{} ------------------------------------------------------------------------ (401.00,572.17)(4.855,-3.000)[2]{} ------------------------------------------------------------------------ (408.00,568.95)(1.355,-0.447)[3]{} ------------------------------------------------------------------------ (408.00,569.17)(4.855,-3.000)[2]{} ------------------------------------------------------------------------ (415.00,565.94)(0.920,-0.468)[5]{} ------------------------------------------------------------------------ (415.00,566.17)(5.340,-4.000)[2]{} ------------------------------------------------------------------------ (422.00,561.94)(0.920,-0.468)[5]{} ------------------------------------------------------------------------ (422.00,562.17)(5.340,-4.000)[2]{} ------------------------------------------------------------------------ (429.00,557.94)(0.920,-0.468)[5]{} ------------------------------------------------------------------------ (429.00,558.17)(5.340,-4.000)[2]{} ------------------------------------------------------------------------ (436.00,553.93)(0.710,-0.477)[7]{} ------------------------------------------------------------------------ (436.00,554.17)(5.630,-5.000)[2]{} ------------------------------------------------------------------------ (443.00,548.93)(0.710,-0.477)[7]{} ------------------------------------------------------------------------ (443.00,549.17)(5.630,-5.000)[2]{} ------------------------------------------------------------------------ (450.00,543.93)(0.710,-0.477)[7]{} ------------------------------------------------------------------------ (450.00,544.17)(5.630,-5.000)[2]{} ------------------------------------------------------------------------ (457.00,538.93)(0.581,-0.482)[9]{} ------------------------------------------------------------------------ (457.00,539.17)(5.824,-6.000)[2]{} ------------------------------------------------------------------------ (464.00,532.93)(0.581,-0.482)[9]{} ------------------------------------------------------------------------ (464.00,533.17)(5.824,-6.000)[2]{} ------------------------------------------------------------------------ (471.00,526.93)(0.581,-0.482)[9]{} ------------------------------------------------------------------------ (471.00,527.17)(5.824,-6.000)[2]{} ------------------------------------------------------------------------ (478.00,520.93)(0.492,-0.485)[11]{} ------------------------------------------------------------------------ (478.00,521.17)(5.962,-7.000)[2]{} ------------------------------------------------------------------------ (485.00,513.93)(0.492,-0.485)[11]{} ------------------------------------------------------------------------ (485.00,514.17)(5.962,-7.000)[2]{} ------------------------------------------------------------------------ (492.59,505.69)(0.485,-0.569)[11]{} ------------------------------------------------------------------------ (491.17,506.84)(7.000,-6.844)[2]{} ------------------------------------------------------------------------ (499.59,497.45)(0.485,-0.645)[11]{} ------------------------------------------------------------------------ (498.17,498.73)(7.000,-7.725)[2]{} ------------------------------------------------------------------------ (506.59,488.45)(0.485,-0.645)[11]{} ------------------------------------------------------------------------ (505.17,489.73)(7.000,-7.725)[2]{} ------------------------------------------------------------------------ (513.59,479.21)(0.485,-0.721)[11]{} ------------------------------------------------------------------------ (512.17,480.61)(7.000,-8.606)[2]{} ------------------------------------------------------------------------ (520.59,469.21)(0.485,-0.721)[11]{} ------------------------------------------------------------------------ (519.17,470.61)(7.000,-8.606)[2]{} ------------------------------------------------------------------------ (527.59,458.98)(0.485,-0.798)[11]{} ------------------------------------------------------------------------ (526.17,460.49)(7.000,-9.488)[2]{} ------------------------------------------------------------------------ (534.59,447.74)(0.485,-0.874)[11]{} ------------------------------------------------------------------------ (533.17,449.37)(7.000,-10.369)[2]{} ------------------------------------------------------------------------ (541.59,435.50)(0.485,-0.950)[11]{} ------------------------------------------------------------------------ (540.17,437.25)(7.000,-11.251)[2]{} ------------------------------------------------------------------------ (548.59,422.26)(0.485,-1.026)[11]{} ------------------------------------------------------------------------ (547.17,424.13)(7.000,-12.132)[2]{} ------------------------------------------------------------------------ (555.59,408.03)(0.485,-1.103)[11]{} ------------------------------------------------------------------------ (554.17,410.01)(7.000,-13.013)[2]{} ------------------------------------------------------------------------ (562.59,392.16)(0.482,-1.395)[9]{} ------------------------------------------------------------------------ (561.17,394.58)(6.000,-13.579)[2]{} ------------------------------------------------------------------------ (568.59,376.55)(0.485,-1.255)[11]{} ------------------------------------------------------------------------ (567.17,378.78)(7.000,-14.776)[2]{} ------------------------------------------------------------------------ (575.59,359.08)(0.485,-1.408)[11]{} ------------------------------------------------------------------------ (574.17,361.54)(7.000,-16.539)[2]{} ------------------------------------------------------------------------ (582.59,340.08)(0.485,-1.408)[11]{} ------------------------------------------------------------------------ (581.17,342.54)(7.000,-16.539)[2]{} ------------------------------------------------------------------------ (589.59,320.60)(0.485,-1.560)[11]{} ------------------------------------------------------------------------ (588.17,323.30)(7.000,-18.302)[2]{} ------------------------------------------------------------------------ (596.59,299.13)(0.485,-1.713)[11]{} ------------------------------------------------------------------------ (595.17,302.06)(7.000,-20.065)[2]{} ------------------------------------------------------------------------ (603.59,275.89)(0.485,-1.789)[11]{} ------------------------------------------------------------------------ (602.17,278.95)(7.000,-20.946)[2]{} ------------------------------------------------------------------------ (610.59,251.42)(0.485,-1.942)[11]{} ------------------------------------------------------------------------ (609.17,254.71)(7.000,-22.709)[2]{} ------------------------------------------------------------------------ (617.59,224.94)(0.485,-2.094)[11]{} ------------------------------------------------------------------------ (616.17,228.47)(7.000,-24.472)[2]{} ------------------------------------------------------------------------ (624.59,196.47)(0.485,-2.247)[11]{} ------------------------------------------------------------------------ (623.17,200.23)(7.000,-26.234)[2]{} ------------------------------------------------------------------------ (631.59,165.99)(0.485,-2.399)[11]{} ------------------------------------------------------------------------ (630.17,170.00)(7.000,-27.997)[2]{} ------------------------------------------------------------------------ (638.59,133.56)(0.482,-2.570)[9]{} ------------------------------------------------------------------------ (637.17,137.78)(6.000,-24.780)[2]{} ------------------------------------------------------------------------ (262.0,606.0) ------------------------------------------------------------------------ \[fig.omega\] In Fig. \[fig.sim\] it is shown that for large $\Omega$, at large times the environment and particle velocities coincide according to Eqn. (\[highomeg\]). (974,675)(0,0) =cmr10 at 10pt (176.0,113.0) ------------------------------------------------------------------------ (176.0,113.0) ------------------------------------------------------------------------ (176.0,113.0) ------------------------------------------------------------------------ (154,113)[(0,0)\[r\][0]{}]{} (890.0,113.0) ------------------------------------------------------------------------ (176.0,185.0) ------------------------------------------------------------------------ (154,185)[(0,0)\[r\][0.02]{}]{} (890.0,185.0) ------------------------------------------------------------------------ (176.0,257.0) ------------------------------------------------------------------------ (154,257)[(0,0)\[r\][0.04]{}]{} (890.0,257.0) ------------------------------------------------------------------------ (176.0,329.0) ------------------------------------------------------------------------ (154,329)[(0,0)\[r\][0.06]{}]{} (890.0,329.0) ------------------------------------------------------------------------ (176.0,400.0) ------------------------------------------------------------------------ (154,400)[(0,0)\[r\][0.08]{}]{} (890.0,400.0) ------------------------------------------------------------------------ (176.0,472.0) ------------------------------------------------------------------------ (154,472)[(0,0)\[r\][0.1]{}]{} (890.0,472.0) ------------------------------------------------------------------------ (176.0,544.0) ------------------------------------------------------------------------ (154,544)[(0,0)\[r\][0.12]{}]{} (890.0,544.0) ------------------------------------------------------------------------ (176.0,616.0) ------------------------------------------------------------------------ (154,616)[(0,0)\[r\][0.14]{}]{} (890.0,616.0) ------------------------------------------------------------------------ (176.0,113.0) ------------------------------------------------------------------------ (176,68)[(0,0)[0]{}]{} (176.0,632.0) ------------------------------------------------------------------------ (298.0,113.0) ------------------------------------------------------------------------ (298,68)[(0,0)[0.5]{}]{} (298.0,632.0) ------------------------------------------------------------------------ (421.0,113.0) ------------------------------------------------------------------------ (421,68)[(0,0)[1]{}]{} (421.0,632.0) ------------------------------------------------------------------------ (543.0,113.0) ------------------------------------------------------------------------ (543,68)[(0,0)[1.5]{}]{} (543.0,632.0) ------------------------------------------------------------------------ (665.0,113.0) ------------------------------------------------------------------------ (665,68)[(0,0)[2]{}]{} (665.0,632.0) ------------------------------------------------------------------------ (788.0,113.0) ------------------------------------------------------------------------ (788,68)[(0,0)[2.5]{}]{} (788.0,632.0) ------------------------------------------------------------------------ (910.0,113.0) ------------------------------------------------------------------------ (910,68)[(0,0)[3]{}]{} (910.0,632.0) ------------------------------------------------------------------------ (176.0,113.0) ------------------------------------------------------------------------ (910.0,113.0) ------------------------------------------------------------------------ (176.0,652.0) ------------------------------------------------------------------------ (543,23)[(0,0)[${\overline t}$]{}]{} (176.0,113.0) ------------------------------------------------------------------------ (200.67,610) ------------------------------------------------------------------------ (200.17,631.00)(1.000,-21.000)[2]{} ------------------------------------------------------------------------ (202.17,561) ------------------------------------------------------------------------ (201.17,589.45)(2.000,-28.452)[2]{} ------------------------------------------------------------------------ (204.17,519) ------------------------------------------------------------------------ (203.17,543.36)(2.000,-24.358)[2]{} ------------------------------------------------------------------------ (206.17,483) ------------------------------------------------------------------------ (205.17,503.85)(2.000,-20.848)[2]{} ------------------------------------------------------------------------ (208.17,452) ------------------------------------------------------------------------ (207.17,469.92)(2.000,-17.924)[2]{} ------------------------------------------------------------------------ (210.17,425) ------------------------------------------------------------------------ (209.17,440.58)(2.000,-15.584)[2]{} ------------------------------------------------------------------------ (212.17,402) ------------------------------------------------------------------------ (211.17,415.24)(2.000,-13.245)[2]{} ------------------------------------------------------------------------ (214.17,381) ------------------------------------------------------------------------ (213.17,393.08)(2.000,-12.075)[2]{} ------------------------------------------------------------------------ (216.17,362) ------------------------------------------------------------------------ (215.17,372.91)(2.000,-10.905)[2]{} ------------------------------------------------------------------------ (218.17,346) ------------------------------------------------------------------------ (217.17,355.15)(2.000,-9.151)[2]{} ------------------------------------------------------------------------ (220.17,331) ------------------------------------------------------------------------ (219.17,339.57)(2.000,-8.566)[2]{} ------------------------------------------------------------------------ (222.17,318) ------------------------------------------------------------------------ (221.17,325.40)(2.000,-7.396)[2]{} ------------------------------------------------------------------------ (224.17,306) ------------------------------------------------------------------------ (223.17,312.81)(2.000,-6.811)[2]{} ------------------------------------------------------------------------ (226.17,295) ------------------------------------------------------------------------ (225.17,301.23)(2.000,-6.226)[2]{} ------------------------------------------------------------------------ (228.17,285) ------------------------------------------------------------------------ (227.17,290.64)(2.000,-5.641)[2]{} ------------------------------------------------------------------------ (230.17,276) ------------------------------------------------------------------------ (229.17,281.06)(2.000,-5.056)[2]{} ------------------------------------------------------------------------ (232.17,268) ------------------------------------------------------------------------ (231.17,272.47)(2.000,-4.472)[2]{} ------------------------------------------------------------------------ (234.17,261) ------------------------------------------------------------------------ (233.17,264.89)(2.000,-3.887)[2]{} ------------------------------------------------------------------------ (236.17,254) ------------------------------------------------------------------------ (235.17,257.89)(2.000,-3.887)[2]{} ------------------------------------------------------------------------ (238.17,247) ------------------------------------------------------------------------ (237.17,250.89)(2.000,-3.887)[2]{} ------------------------------------------------------------------------ (240.17,242) ------------------------------------------------------------------------ (239.17,244.72)(2.000,-2.717)[2]{} ------------------------------------------------------------------------ (242.17,236) ------------------------------------------------------------------------ (241.17,239.30)(2.000,-3.302)[2]{} ------------------------------------------------------------------------ (244.17,232) ------------------------------------------------------------------------ (243.17,234.13)(2.000,-2.132)[2]{} ------------------------------------------------------------------------ (246.17,227) ------------------------------------------------------------------------ (245.17,229.72)(2.000,-2.717)[2]{} ------------------------------------------------------------------------ (248.17,223) ------------------------------------------------------------------------ (247.17,225.13)(2.000,-2.132)[2]{} ------------------------------------------------------------------------ (250.17,218) ------------------------------------------------------------------------ (249.17,220.72)(2.000,-2.717)[2]{} ------------------------------------------------------------------------ (252.17,215) ------------------------------------------------------------------------ (251.17,216.55)(2.000,-1.547)[2]{} ------------------------------------------------------------------------ (254.17,212) ------------------------------------------------------------------------ (253.17,213.55)(2.000,-1.547)[2]{} ------------------------------------------------------------------------ (256.17,208) ------------------------------------------------------------------------ (255.17,210.13)(2.000,-2.132)[2]{} ------------------------------------------------------------------------ (258.17,204) ------------------------------------------------------------------------ (257.17,206.13)(2.000,-2.132)[2]{} ------------------------------------------------------------------------ (260,202.17) ------------------------------------------------------------------------ (260.00,203.17)(1.000,-2.000)[2]{} ------------------------------------------------------------------------ (262.17,199) ------------------------------------------------------------------------ (261.17,200.55)(2.000,-1.547)[2]{} ------------------------------------------------------------------------ (264.17,196) ------------------------------------------------------------------------ (263.17,197.55)(2.000,-1.547)[2]{} ------------------------------------------------------------------------ (266,194.17) ------------------------------------------------------------------------ (266.00,195.17)(1.000,-2.000)[2]{} ------------------------------------------------------------------------ (268.17,191) ------------------------------------------------------------------------ (267.17,192.55)(2.000,-1.547)[2]{} ------------------------------------------------------------------------ (270,189.17) ------------------------------------------------------------------------ (270.00,190.17)(1.000,-2.000)[2]{} ------------------------------------------------------------------------ (272,187.17) ------------------------------------------------------------------------ (272.00,188.17)(1.000,-2.000)[2]{} ------------------------------------------------------------------------ (274,185.67) ------------------------------------------------------------------------ (274.00,186.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (276.17,183) ------------------------------------------------------------------------ (275.17,184.55)(2.000,-1.547)[2]{} ------------------------------------------------------------------------ (278,181.17) ------------------------------------------------------------------------ (278.00,182.17)(1.000,-2.000)[2]{} ------------------------------------------------------------------------ (280,179.67) ------------------------------------------------------------------------ (280.00,180.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (282,178.17) ------------------------------------------------------------------------ (282.00,179.17)(1.000,-2.000)[2]{} ------------------------------------------------------------------------ (284,176.67) ------------------------------------------------------------------------ (284.00,177.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (286,175.17) ------------------------------------------------------------------------ (286.00,176.17)(1.000,-2.000)[2]{} ------------------------------------------------------------------------ (288,173.67) ------------------------------------------------------------------------ (288.00,174.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (290,172.17) ------------------------------------------------------------------------ (290.00,173.17)(1.000,-2.000)[2]{} ------------------------------------------------------------------------ (292,170.67) ------------------------------------------------------------------------ (292.00,171.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (294,169.67) ------------------------------------------------------------------------ (294.00,170.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (296,168.17) ------------------------------------------------------------------------ (296.00,169.17)(1.000,-2.000)[2]{} ------------------------------------------------------------------------ (298,166.67) ------------------------------------------------------------------------ (298.00,167.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (300,165.67) ------------------------------------------------------------------------ (300.00,166.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (302,164.67) ------------------------------------------------------------------------ (302.00,165.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (304,163.17) ------------------------------------------------------------------------ (304.00,164.17)(1.000,-2.000)[2]{} ------------------------------------------------------------------------ (308,161.67) ------------------------------------------------------------------------ (308.00,162.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (310,160.67) ------------------------------------------------------------------------ (310.00,161.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (312,159.67) ------------------------------------------------------------------------ (312.00,160.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (306.0,163.0) ------------------------------------------------------------------------ (316,158.67) ------------------------------------------------------------------------ (316.00,159.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (318,157.67) ------------------------------------------------------------------------ (318.00,158.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (320,156.67) ------------------------------------------------------------------------ (320.00,157.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (322,155.67) ------------------------------------------------------------------------ (322.00,156.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (324,154.67) ------------------------------------------------------------------------ (324.00,155.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (314.0,160.0) ------------------------------------------------------------------------ (328,153.67) ------------------------------------------------------------------------ (328.00,154.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (330,152.67) ------------------------------------------------------------------------ (330.00,153.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (326.0,155.0) ------------------------------------------------------------------------ (334,151.67) ------------------------------------------------------------------------ (334.00,152.17)(1.500,-1.000)[2]{} ------------------------------------------------------------------------ (337,150.67) ------------------------------------------------------------------------ (337.00,151.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (332.0,153.0) ------------------------------------------------------------------------ (341,149.67) ------------------------------------------------------------------------ (341.00,150.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (343,148.67) ------------------------------------------------------------------------ (343.00,149.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (339.0,151.0) ------------------------------------------------------------------------ (347,147.67) ------------------------------------------------------------------------ (347.00,148.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (345.0,149.0) ------------------------------------------------------------------------ (351,146.67) ------------------------------------------------------------------------ (351.00,147.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (349.0,148.0) ------------------------------------------------------------------------ (355,145.67) ------------------------------------------------------------------------ (355.00,146.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (353.0,147.0) ------------------------------------------------------------------------ (359,144.67) ------------------------------------------------------------------------ (359.00,145.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (357.0,146.0) ------------------------------------------------------------------------ (363,143.67) ------------------------------------------------------------------------ (363.00,144.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (361.0,145.0) ------------------------------------------------------------------------ (367,142.67) ------------------------------------------------------------------------ (367.00,143.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (365.0,144.0) ------------------------------------------------------------------------ (371,141.67) ------------------------------------------------------------------------ (371.00,142.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (369.0,143.0) ------------------------------------------------------------------------ (377,140.67) ------------------------------------------------------------------------ (377.00,141.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (373.0,142.0) ------------------------------------------------------------------------ (381,139.67) ------------------------------------------------------------------------ (381.00,140.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (379.0,141.0) ------------------------------------------------------------------------ (385,138.67) ------------------------------------------------------------------------ (385.00,139.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (383.0,140.0) ------------------------------------------------------------------------ (391,137.67) ------------------------------------------------------------------------ (391.00,138.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (387.0,139.0) ------------------------------------------------------------------------ (401,136.67) ------------------------------------------------------------------------ (401.00,137.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (393.0,138.0) ------------------------------------------------------------------------ (405,135.67) ------------------------------------------------------------------------ (405.00,136.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (403.0,137.0) ------------------------------------------------------------------------ (411,134.67) ------------------------------------------------------------------------ (411.00,135.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (407.0,136.0) ------------------------------------------------------------------------ (423,133.67) ------------------------------------------------------------------------ (423.00,134.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (413.0,135.0) ------------------------------------------------------------------------ (433,132.67) ------------------------------------------------------------------------ (433.00,133.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (425.0,134.0) ------------------------------------------------------------------------ (439,131.67) ------------------------------------------------------------------------ (439.00,132.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (435.0,133.0) ------------------------------------------------------------------------ (445,130.67) ------------------------------------------------------------------------ (445.00,131.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (441.0,132.0) ------------------------------------------------------------------------ (459,129.67) ------------------------------------------------------------------------ (459.00,130.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (447.0,131.0) ------------------------------------------------------------------------ (467,128.67) ------------------------------------------------------------------------ (467.00,129.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (461.0,130.0) ------------------------------------------------------------------------ (483,127.67) ------------------------------------------------------------------------ (483.00,128.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (469.0,129.0) ------------------------------------------------------------------------ (493,126.67) ------------------------------------------------------------------------ (493.00,127.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (485.0,128.0) ------------------------------------------------------------------------ (501,125.67) ------------------------------------------------------------------------ (501.00,126.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (495.0,127.0) ------------------------------------------------------------------------ (519,124.67) ------------------------------------------------------------------------ (519.00,125.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (503.0,126.0) ------------------------------------------------------------------------ (537,123.67) ------------------------------------------------------------------------ (537.00,124.17)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (521.0,125.0) ------------------------------------------------------------------------ (557,122.67) ------------------------------------------------------------------------ (557.00,123.17)(10.000,-1.000)[2]{} ------------------------------------------------------------------------ (577,121.67) ------------------------------------------------------------------------ (577.00,122.17)(10.000,-1.000)[2]{} ------------------------------------------------------------------------ (597,120.67) ------------------------------------------------------------------------ (597.00,121.17)(10.000,-1.000)[2]{} ------------------------------------------------------------------------ (617,119.67) ------------------------------------------------------------------------ (617.00,120.17)(10.000,-1.000)[2]{} ------------------------------------------------------------------------ (539.0,124.0) ------------------------------------------------------------------------ (678,118.67) ------------------------------------------------------------------------ (678.00,119.17)(10.000,-1.000)[2]{} ------------------------------------------------------------------------ (698,117.67) ------------------------------------------------------------------------ (698.00,118.17)(10.000,-1.000)[2]{} ------------------------------------------------------------------------ (637.0,120.0) ------------------------------------------------------------------------ (738,116.67) ------------------------------------------------------------------------ (738.00,117.17)(10.000,-1.000)[2]{} ------------------------------------------------------------------------ (718.0,118.0) ------------------------------------------------------------------------ (818,115.67) ------------------------------------------------------------------------ (818.00,116.17)(10.000,-1.000)[2]{} ------------------------------------------------------------------------ (758.0,117.0) ------------------------------------------------------------------------ (838.0,116.0) ------------------------------------------------------------------------ (176,558) (175.34,494) ------------------------------------------------------------------------ (174.34,526.00)(2.000,-32.000)[2]{} ------------------------------------------------------------------------ (177.34,445) ------------------------------------------------------------------------ (176.34,469.50)(2.000,-24.500)[2]{} ------------------------------------------------------------------------ (179.34,407) ------------------------------------------------------------------------ (178.34,426.00)(2.000,-19.000)[2]{} ------------------------------------------------------------------------ (181.34,378) ------------------------------------------------------------------------ (180.34,392.50)(2.000,-14.500)[2]{} ------------------------------------------------------------------------ (183.34,354) ------------------------------------------------------------------------ (182.34,366.00)(2.000,-12.000)[2]{} ------------------------------------------------------------------------ (185.34,336) ------------------------------------------------------------------------ (184.34,345.00)(2.000,-9.000)[2]{} ------------------------------------------------------------------------ (187.34,320) ------------------------------------------------------------------------ (186.34,328.00)(2.000,-8.000)[2]{} ------------------------------------------------------------------------ (189.34,307) ------------------------------------------------------------------------ (188.34,313.50)(2.000,-6.500)[2]{} ------------------------------------------------------------------------ (191.34,296) ------------------------------------------------------------------------ (190.34,301.50)(2.000,-5.500)[2]{} ------------------------------------------------------------------------ (193.34,287) ------------------------------------------------------------------------ (192.34,291.50)(2.000,-4.500)[2]{} ------------------------------------------------------------------------ (195.34,278) ------------------------------------------------------------------------ (194.34,282.50)(2.000,-4.500)[2]{} ------------------------------------------------------------------------ (197.34,271) ------------------------------------------------------------------------ (196.34,274.50)(2.000,-3.500)[2]{} ------------------------------------------------------------------------ (199.34,264) ------------------------------------------------------------------------ (198.34,267.50)(2.000,-3.500)[2]{} ------------------------------------------------------------------------ (201.34,258) ------------------------------------------------------------------------ (200.34,261.00)(2.000,-3.000)[2]{} ------------------------------------------------------------------------ (203.34,253) ------------------------------------------------------------------------ (202.34,255.50)(2.000,-2.500)[2]{} ------------------------------------------------------------------------ (205.34,248) ------------------------------------------------------------------------ (204.34,250.50)(2.000,-2.500)[2]{} ------------------------------------------------------------------------ (207.34,243) ------------------------------------------------------------------------ (206.34,245.50)(2.000,-2.500)[2]{} ------------------------------------------------------------------------ (209.34,239) ------------------------------------------------------------------------ (208.34,241.00)(2.000,-2.000)[2]{} ------------------------------------------------------------------------ (211.34,235) ------------------------------------------------------------------------ (210.34,237.00)(2.000,-2.000)[2]{} ------------------------------------------------------------------------ (213.34,232) ------------------------------------------------------------------------ (212.34,233.50)(2.000,-1.500)[2]{} ------------------------------------------------------------------------ (215.34,227) ------------------------------------------------------------------------ (214.34,229.50)(2.000,-2.500)[2]{} ------------------------------------------------------------------------ (217.34,224) ------------------------------------------------------------------------ (216.34,225.50)(2.000,-1.500)[2]{} ------------------------------------------------------------------------ (219.34,221) ------------------------------------------------------------------------ (218.34,222.50)(2.000,-1.500)[2]{} ------------------------------------------------------------------------ (221.34,218) ------------------------------------------------------------------------ (220.34,219.50)(2.000,-1.500)[2]{} ------------------------------------------------------------------------ (223.34,215) ------------------------------------------------------------------------ (222.34,216.50)(2.000,-1.500)[2]{} ------------------------------------------------------------------------ (225.34,212) ------------------------------------------------------------------------ (224.34,213.50)(2.000,-1.500)[2]{} ------------------------------------------------------------------------ (228,209.34) ------------------------------------------------------------------------ (228.00,210.34)(1.000,-2.000)[2]{} ------------------------------------------------------------------------ (229.34,207) ------------------------------------------------------------------------ (228.34,208.50)(2.000,-1.500)[2]{} ------------------------------------------------------------------------ (232,204.34) ------------------------------------------------------------------------ (232.00,205.34)(1.000,-2.000)[2]{} ------------------------------------------------------------------------ (234,202.34) ------------------------------------------------------------------------ (234.00,203.34)(1.000,-2.000)[2]{} ------------------------------------------------------------------------ (236,200.34) ------------------------------------------------------------------------ (236.00,201.34)(1.000,-2.000)[2]{} ------------------------------------------------------------------------ (238,198.34) ------------------------------------------------------------------------ (238.00,199.34)(1.000,-2.000)[2]{} ------------------------------------------------------------------------ (240,196.34) ------------------------------------------------------------------------ (240.00,197.34)(1.000,-2.000)[2]{} ------------------------------------------------------------------------ (242,194.34) ------------------------------------------------------------------------ (242.00,195.34)(1.000,-2.000)[2]{} ------------------------------------------------------------------------ (244,192.34) ------------------------------------------------------------------------ (244.00,193.34)(1.000,-2.000)[2]{} ------------------------------------------------------------------------ (246,190.34) ------------------------------------------------------------------------ (246.00,191.34)(1.000,-2.000)[2]{} ------------------------------------------------------------------------ (248,188.84) ------------------------------------------------------------------------ (248.00,189.34)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (250,187.34) ------------------------------------------------------------------------ (250.00,188.34)(1.000,-2.000)[2]{} ------------------------------------------------------------------------ (252,185.84) ------------------------------------------------------------------------ (252.00,186.34)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (254,184.84) ------------------------------------------------------------------------ (254.00,185.34)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (256,183.34) ------------------------------------------------------------------------ (256.00,184.34)(1.000,-2.000)[2]{} ------------------------------------------------------------------------ (258,181.34) ------------------------------------------------------------------------ (258.00,182.34)(1.000,-2.000)[2]{} ------------------------------------------------------------------------ (260,179.84) ------------------------------------------------------------------------ (260.00,180.34)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (262,178.84) ------------------------------------------------------------------------ (262.00,179.34)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (264,177.34) ------------------------------------------------------------------------ (264.00,178.34)(1.000,-2.000)[2]{} ------------------------------------------------------------------------ (266,175.84) ------------------------------------------------------------------------ (266.00,176.34)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (268,174.84) ------------------------------------------------------------------------ (268.00,175.34)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (270,173.84) ------------------------------------------------------------------------ (270.00,174.34)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (272,172.84) ------------------------------------------------------------------------ (272.00,173.34)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (274,171.84) ------------------------------------------------------------------------ (274.00,172.34)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (276,170.84) ------------------------------------------------------------------------ (276.00,171.34)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (278,169.84) ------------------------------------------------------------------------ (278.00,170.34)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (280,168.84) ------------------------------------------------------------------------ (280.00,169.34)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (282,167.84) ------------------------------------------------------------------------ (282.00,168.34)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (284,166.84) ------------------------------------------------------------------------ (284.00,167.34)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (286,165.84) ------------------------------------------------------------------------ (286.00,166.34)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (288,164.84) ------------------------------------------------------------------------ (288.00,165.34)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (290,163.84) ------------------------------------------------------------------------ (290.00,164.34)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (292,162.84) ------------------------------------------------------------------------ (292.00,163.34)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (296,161.34) ------------------------------------------------------------------------ (296.00,162.34)(1.000,-2.000)[2]{} ------------------------------------------------------------------------ (294.0,164.0) (300,159.84) ------------------------------------------------------------------------ (300.00,160.34)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (302,158.84) ------------------------------------------------------------------------ (302.00,159.34)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (304,157.84) ------------------------------------------------------------------------ (304.00,158.34)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (298.0,162.0) (308,156.84) ------------------------------------------------------------------------ (308.00,157.34)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (310,155.84) ------------------------------------------------------------------------ (310.00,156.34)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (306.0,159.0) (314,154.84) ------------------------------------------------------------------------ (314.00,155.34)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (316,153.84) ------------------------------------------------------------------------ (316.00,154.34)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (312.0,157.0) (320,152.84) ------------------------------------------------------------------------ (320.00,153.34)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (318.0,155.0) (324,151.84) ------------------------------------------------------------------------ (324.00,152.34)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (322.0,154.0) (328,150.84) ------------------------------------------------------------------------ (328.00,151.34)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (330,149.84) ------------------------------------------------------------------------ (330.00,150.34)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (326.0,153.0) (337,148.84) ------------------------------------------------------------------------ (337.00,149.34)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (339,147.84) ------------------------------------------------------------------------ (339.00,148.34)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (332.0,151.0) ------------------------------------------------------------------------ (343,146.84) ------------------------------------------------------------------------ (343.00,147.34)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (341.0,149.0) (347,145.84) ------------------------------------------------------------------------ (347.00,146.34)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (345.0,148.0) (353,144.84) ------------------------------------------------------------------------ (353.00,145.34)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (349.0,147.0) ------------------------------------------------------------------------ (357,143.84) ------------------------------------------------------------------------ (357.00,144.34)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (355.0,146.0) (363,142.84) ------------------------------------------------------------------------ (363.00,143.34)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (359.0,145.0) ------------------------------------------------------------------------ (367,141.84) ------------------------------------------------------------------------ (367.00,142.34)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (365.0,144.0) (373,140.84) ------------------------------------------------------------------------ (373.00,141.34)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (369.0,143.0) ------------------------------------------------------------------------ (379,139.84) ------------------------------------------------------------------------ (379.00,140.34)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (375.0,142.0) ------------------------------------------------------------------------ (385,138.84) ------------------------------------------------------------------------ (385.00,139.34)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (381.0,141.0) ------------------------------------------------------------------------ (391,137.84) ------------------------------------------------------------------------ (391.00,138.34)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (387.0,140.0) ------------------------------------------------------------------------ (397,136.84) ------------------------------------------------------------------------ (397.00,137.34)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (393.0,139.0) ------------------------------------------------------------------------ (405,135.84) ------------------------------------------------------------------------ (405.00,136.34)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (399.0,138.0) ------------------------------------------------------------------------ (411,134.84) ------------------------------------------------------------------------ (411.00,135.34)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (407.0,137.0) ------------------------------------------------------------------------ (419,133.84) ------------------------------------------------------------------------ (419.00,134.34)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (413.0,136.0) ------------------------------------------------------------------------ (427,132.84) ------------------------------------------------------------------------ (427.00,133.34)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (421.0,135.0) ------------------------------------------------------------------------ (437,131.84) ------------------------------------------------------------------------ (437.00,132.34)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (429.0,134.0) ------------------------------------------------------------------------ (447,130.84) ------------------------------------------------------------------------ (447.00,131.34)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (439.0,133.0) ------------------------------------------------------------------------ (459,129.84) ------------------------------------------------------------------------ (459.00,130.34)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (449.0,132.0) ------------------------------------------------------------------------ (469,128.84) ------------------------------------------------------------------------ (469.00,129.34)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (461.0,131.0) ------------------------------------------------------------------------ (481,127.84) ------------------------------------------------------------------------ (481.00,128.34)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (471.0,130.0) ------------------------------------------------------------------------ (493,126.84) ------------------------------------------------------------------------ (493.00,127.34)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (495,126.84) ------------------------------------------------------------------------ (495.00,126.34)(1.000,1.000)[2]{} ------------------------------------------------------------------------ (497,126.84) ------------------------------------------------------------------------ (497.00,127.34)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (483.0,129.0) ------------------------------------------------------------------------ (509,125.84) ------------------------------------------------------------------------ (509.00,126.34)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (499.0,128.0) ------------------------------------------------------------------------ (527,124.84) ------------------------------------------------------------------------ (527.00,125.34)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (511.0,127.0) ------------------------------------------------------------------------ (547,123.84) ------------------------------------------------------------------------ (547.00,124.34)(1.000,-1.000)[2]{} ------------------------------------------------------------------------ (529.0,126.0) ------------------------------------------------------------------------ (557,122.84) ------------------------------------------------------------------------ (557.00,123.34)(10.000,-1.000)[2]{} ------------------------------------------------------------------------ (577,121.84) ------------------------------------------------------------------------ (577.00,122.34)(10.000,-1.000)[2]{} ------------------------------------------------------------------------ (597,120.84) ------------------------------------------------------------------------ (597.00,121.34)(10.000,-1.000)[2]{} ------------------------------------------------------------------------ (617,119.84) ------------------------------------------------------------------------ (617.00,120.34)(10.000,-1.000)[2]{} ------------------------------------------------------------------------ (549.0,125.0) ------------------------------------------------------------------------ (658,118.84) ------------------------------------------------------------------------ (658.00,119.34)(10.000,-1.000)[2]{} ------------------------------------------------------------------------ (637.0,121.0) ------------------------------------------------------------------------ (698,117.84) ------------------------------------------------------------------------ (698.00,118.34)(10.000,-1.000)[2]{} ------------------------------------------------------------------------ (678.0,120.0) ------------------------------------------------------------------------ (738,116.84) ------------------------------------------------------------------------ (738.00,117.34)(10.000,-1.000)[2]{} ------------------------------------------------------------------------ (718.0,119.0) ------------------------------------------------------------------------ (798,115.84) ------------------------------------------------------------------------ (798.00,116.34)(10.000,-1.000)[2]{} ------------------------------------------------------------------------ (758.0,118.0) ------------------------------------------------------------------------ (878,114.84) ------------------------------------------------------------------------ (878.00,115.34)(10.000,-1.000)[2]{} ------------------------------------------------------------------------ (818.0,117.0) ------------------------------------------------------------------------ (898.0,116.0) ------------------------------------------------------------------------ (195,652)(0.389,-20.752)[8]{} (198,492)(0.825,-20.739)[10]{} (206,291)(1.525,-20.699)[4]{} (213,196)(3.261,-20.498)[3]{} (226.37,135.28) (228,131)(12.743,-16.383)[0]{} (239.05,119.47) (243,117)(19.957,-5.702)[0]{} (250,115)(20.595,-2.574)[0]{} (258.74,113.89) (265,113)(20.756,0.000)[0]{} (279.44,113.00) (280,113)(20.756,0.000)[0]{} (287,113)(20.756,0.000)[0]{} (300.19,113.00) (302,113)(20.756,0.000)[0]{} (309,113)(20.756,0.000)[0]{} (320.95,113.00) (324,113)(20.756,0.000)[0]{} (332,113)(20.756,0.000)[0]{} (341.70,113.00) (347,113)(20.756,0.000)[0]{} (354,113)(20.756,0.000)[0]{} (362.46,113.00) (369,113)(20.756,0.000)[0]{} (383.21,113.00) (384,113)(20.756,0.000)[0]{} (391,113)(20.756,0.000)[0]{} (403.97,113.00) (406,113)(20.756,0.000)[0]{} (413,113)(20.756,0.000)[0]{} (424.72,113.00) (428,113)(20.756,0.000)[0]{} (435,113)(20.756,0.000)[0]{} (445.48,113.00) (450,113)(20.756,0.000)[0]{} (458,113)(20.756,0.000)[0]{} (466.24,113.00) (473,113)(20.756,0.000)[0]{} (486.99,113.00) (487,113)(20.756,0.000)[0]{} (495,113)(20.756,0.000)[0]{} (507.75,113.00) (510,113)(20.756,0.000)[0]{} (517,113)(20.756,0.000)[0]{} (528.50,113.00) (532,113)(20.756,0.000)[0]{} (539,113)(20.756,0.000)[0]{} (549.26,113.00) (554,113)(20.756,0.000)[0]{} (562,113)(20.756,0.000)[0]{} (570.01,113.00) (576,113)(20.756,0.000)[0]{} (590.77,113.00) (591,113)(20.756,0.000)[0]{} (599,113)(20.756,0.000)[0]{} (611.52,113.00) (613,113)(20.756,0.000)[0]{} (621,113)(20.756,0.000)[0]{} (632.28,113.00) (636,113)(20.756,0.000)[0]{} (643,113)(20.756,0.000)[0]{} (653.04,113.00) (658,113)(20.756,0.000)[0]{} (665,113)(20.756,0.000)[0]{} (673.79,113.00) (680,113)(20.756,0.000)[0]{} (694.55,113.00) (695,113)(20.756,0.000)[0]{} (702,113)(20.756,0.000)[0]{} (715.30,113.00) (717,113)(20.756,0.000)[0]{} (725,113)(20.756,0.000)[0]{} (736.06,113.00) (739,113)(20.756,0.000)[0]{} (747,113)(20.756,0.000)[0]{} (756.81,113.00) (762,113)(20.756,0.000)[0]{} (769,113)(20.756,0.000)[0]{} (777.57,113.00) (784,113)(20.756,0.000)[0]{} (798.32,113.00) (799,113)(20.756,0.000)[0]{} (806,113)(20.756,0.000)[0]{} (819.08,113.00) (821,113)(20.756,0.000)[0]{} (828,113)(20.756,0.000)[0]{} (839.83,113.00) (843,113)(20.756,0.000)[0]{} (851,113)(20.756,0.000)[0]{} (860.59,113.00) (866,113)(20.756,0.000)[0]{} (873,113)(20.756,0.000)[0]{} (881.35,113.00) (888,113)(20.756,0.000)[0]{} (902.10,113.00) (903,113)(20.756,0.000)[0]{} (910,113) \[fig.sim\] Summary and discussion ====================== We have presented a theory for Dissipative Particle Dynamics that allows to understand the different dynamical regimes displayed by the model. The theory is based on the physical picture that DPD particles behave like Brownian particles in a non-equilibrium environment due to the rest of DPD particles. An explicit expression for the velocity autocorrelation function is derived in which the Brownian exponential behavior is corrected by the presence of collective effects. By using dimensionless variables it is possible to asses the range of parameters for which the collective effects are important. Three dimensionless parameters appear in the model, $s,\Omega,\mu$, and they characterize the different dynamical regimes in the system. The relevance of precisely these dimensionless groups is motivated by the theory presented (in Ref. [@pag98] other dimensionless groups are introduced). Two dynamical regimes are identified, the mean field regime and the collective behavior regime. The transition between both is governed essentially by the dimensionless friction $\Omega$ with important effects of $s$ and $\mu$. Mean field behavior appears for small friction $\Omega$ or large overlapping $s$. This is physically reasonable: For small friction, the dynamics of the environment of a given particle hardly affects the behavior of this particle. For large overlapping the non-equilibrium collective effects are smeared out over large regions. The mean field approximation is closely related to the molecular chaos assumption made in kinetic theory. Actually, it is possible to compute the diffusion coefficient of the DPD particles by using the mean field approximation for the vaf into the usual Green-Kubo formula for the diffusion coefficient [@gro97]. The result is precisely the prediction for the diffusion coefficient given by kinetic theory [@mar97] (with due account of the normalization given in Eqn. (\[norm\])), i.e. $$D = \int_0^\infty \frac{1}{d}\langle {\bf v}_i(t)\!\cdot\!{\bf v}_i(0)\rangle = \frac{d}{\gamma}\frac{k_BT}{m}. \label{dif}$$ When the dimensionless friction $\Omega$ is high, the vaf does not decay in an exponential way because it is dominated by the collective dynamics. We have presented a theoretical prediction for the collective part of the vaf by assuming that the correlation function of the environment velocity reflects an underlying hydrodynamic behavior. Such a behavior is expected (and observed in the simulations) when the overlapping coefficient $s$ is large enough. In this case, the collective effect is small but well described by hydrodynamics. The fact that hydrodynamics governs the dynamics of the velocity of the particles has two important consequences. The first one is that finite size effects are important. Hydrodynamic self-interaction through the periodic boundary conditions exists and large box sizes must be considered in order to render this effect small. This self-interaction occurs not only through the sound mode but also through the shear mode [@web]. The second consequence is the appearance of the celebrated long-time tails in the vaf. These algebraic tails occur at very long times (for which the vaf has decayed to a factor $10^{-3}$ from its original value) and are difficult to observe in our simulations due to the statistical noise. Nevertheless, we have provided sufficient numerical evidence for the hydrodynamic behavior at smaller times and we expect the presence of long time tails at large times for sufficiently large system sizes. The $t^{-1}$ dependence of the vaf, when introduced into the Green-Kubo formula (\[dif\]) leads to a logarithmically divergent diffusion coefficient. It is apparent that this “small” divergence will be hardly observable in any simulation with a finite box size. In the regimes for which collective effects are important, we expect that deviations from the predictions of kinetic theory occur not only for the diffusion coefficient, but also for the rest of the transport coefficients of the DPD fluid [@pag98]. It is actually possible to use the theory presented in this paper in order to compute the transport coefficients for the DPD fluid expressed in the form of Green-Kubo formulae [@esp95b]. One obtains then a set of recursive equations in which the transport coefficients are expressed in terms of the transport coefficients used in the hydrodynamic assumption. We do not follow this rather cumbersome numerical procedure here because it does not provide any new physical insight. Dissipative Particle Dynamics is designed to simulate hydrodynamic problems. Actually, one would like to have the DPD particles moving in such a way that they follow accurately the flow field intended to be modeled. We see that this will happen whenever the friction is sufficiently large (in this case the velocity of a DPD particle is slaved by its environment) and the overlapping is sufficiently large (in such a way that the environment velocity moves hydrodynamically). In this regime, the dynamics of the particles is mainly collective and kinetic theory gives inaccurate values for the transport coefficients [@pag98]. Acknowledgments {#acknowledgments .unnumbered} =============== We are grateful to M.H. Ernst, P. Warren, M. Ripoll, I. Zúñiga, and M. Revenga for stimulating discussions. Appendix I {#appendix-i .unnumbered} ========== We derive the equilibrium value for the environment velocity correlation function given in Eqn. (\[evt0\]). From the definition of ${\bf V}_i$ in Eqn. (\[vhidro\]) we calculate the mean value by using the canonical ensemble $$\begin{aligned} \frac{1}{d}\langle {\bf V}_i(0)\!\cdot\! {\bf V}_i(0) \rangle &=&d\langle(\sum_{j\neq i}\omega_{ij}{\bf e}_{ij}{\bf e}_{ij} \!\cdot\!{\bf v}_j)(\sum_{k\neq i}\omega_{ik}{\bf e}_{ik}{\bf e}_{ik} \!\cdot\!{\bf v}_k) \rangle \nonumber\\ &=& \frac{d k_B T (N-1)}{m}\frac{\int d{\bf r }_id{\bf r }_j \omega(r_{ij})}{\int d{\bf r }_id{\bf r }_j} \nonumber\\ &=&\frac{d k_B T}{ m} \frac{N-1}{N}\frac{3}{2 \pi r_c^2 n} \nonumber\\ &\approx&\frac{d k_B T}{ m} \frac{3}{2 \pi s^2}, \label{ve0}\end{aligned}$$ for large $N$ (number of DPD particles in the system) and $n$ the number density. Appendix II {#appendix-ii .unnumbered} =========== On simple symmetry grounds, the tensor ${\bf \omega}({\bf k})$ has the form $${\bf \omega}({\bf k}) = \int d{\bf r}\omega(r){\bf \hat{r}}{\bf \hat{r}} \exp\{-i {\bf k }{\bf r}\} = a(kr_c){\bf 1}+ b(kr_c){\bf \hat{k}}{\bf \hat{k}}$$ In order to calculate the functions $a(kr_c)$ and $b(kr_c)$ we double contract ${\bf \omega}({\bf k})$ with the dyadic ${\bf\hat{k}}{\bf \hat{k}}$ and also take its trace. This leads to $$\begin{aligned} {\bf \hat{k}^T}{\bf \omega}(k){\bf \hat{k}} &=& a(kr_c)+b(kr_c) \nonumber\\ &=& 2 \pi \int_0^\infty dr \omega(r) \left(\frac{J_{1}(kr)}{k}- r J_2(kr)\right), \nonumber\\ {\rm tr}({\bf \omega}(k)) &=& 2 a(kr_c)+ b(kr_c) \nonumber\\ &=& 2 \pi \int_0^\infty r \omega(r)J_{o}(kr).\end{aligned}$$ The integrals are given in terms of generalized hypergeometric functions $_p F_q\{{\bf a},{\bf b},z\}$ [@mathematica] and Bessel functions $$_p F_q\{{\bf a},{\bf b},z\}= \sum_{k=0}^{\infty}\frac{(a_1)_k...(a_p)_k}{(b_1)_k...(b_q)_k}\frac{z^k}{k! }$$ where $(m)_k=m(m+1)...(m+k-1)$. More precisely, $$\begin{aligned} a(kr_c)+b(kr_c) &=& -\frac{1}{n} {_1 F_2 }\{( \frac{3}{2} ),(2,\frac{5}{2}),-\frac{(k r_c)^2}{4}\} \nonumber\\ &-& \frac{6}{n (k r_c)^2}+ \frac{6}{n (k r_c)^2}J_{0}(k r_c) + \frac{6}{n k r_c} J_{1}(k r_c) \nonumber\\ &+&\frac{6 (k r_c)^2}{n} {_1 F_2}\{ (\frac{5}{2}),(3,\frac{7}{2}),-\frac{(k r_c)^2}{4}\} \nonumber\\ 2 a(kr_c)+ b(kr_c) &=& \frac{6 J_{1} (kr_c)}{k r_c n} -\frac{2}{n} {_1 F_2 }\{(\frac{3}{2}),(1,\frac{5}{2}),-\frac{(k r_c)^2}{4}\}\end{aligned}$$ The solution of this system of two equations provides the values for $a(kr_c)$ and $b(kr_c)$. [10]{} P.J. Hoogerbrugge and J.M.V.A. Koelman, Europhys. Lett. [**19**]{}, 155 (1992). J.M.V.A. Koelman and P.J. Hoogerbrugge, Europhys. Lett. [**21**]{}, 369 (1993). P. Espa[ñ]{}ol and P. Warren, Europhys. Lett. [**30**]{}, 191 (1995). A.G. Schlijper, P.J. Hoogerbrugge, and C.W. Manke, J. Rheol. [**39**]{}, 567 (1995). E.S. Boek, P.V. Coveney, H.N.W. Lekkerkerker, and P. van der Schoot, Phys. Rev. E [**55**]{}, 3124 (1997). E.S. Boek, P.V. Coveney, and H.N.W. Lekkerkerker, J. Phys.: Condens. Matter [**8**]{}, 9509 (1997). P.V. Coveney and K. Novik, Phys. Rev. E [**54**]{}, 5134 (1996). R.D. Groot and P.B. Warren, J. Chem. Phys. [**107**]{}, 4423 (1997). See the Proceedings of the 1996 and 1998 Conferences on Discrete Models for Fluid Mechanics, B.M. Boghosian, F.J. Alexander and P.V. Coveney (eds), in special issues of Int. J. Mod. Phys. C [**8**]{} (1997) and Int. J. Mod. Phys. C [**??**]{} (1999). I. Pagonabarraga, M.H.J. Hagen, and D. Frenkel, Europhys. Lett. [**42**]{}, 377 (1998). C. Marsh, G. Backx, and M.H. Ernst, Europhys. Lett. [**38**]{}, 411 (1997). C. Marsh, G. Backx, and M.H. Ernst, Phys. Rev. E [**56**]{}, 1976 (1997). C.W. Gardiner, [*Handbook of Stochastic Methods*]{}, (Springer Verlag, Berlin, 1983). J.P. Boon and S. Yip, [*Molecular hydrodynamics*]{} (McGraw-Hill, New York, 1980). M.Revenga, I. Zúñiga, P. Español, and I. Pagonabarraga, to appear in I. J. Mod. Phys. C . P. Español, Phys. Rev. E,57, 2930 (1998). P. Español, Physica A 214, 185 (1995). For an animation in real space showing that the sound and shear modes both contribute to the self-interaction with similar time-scales, see [**:http://www.fisfun.uned.es/ pep/H-modes.html**]{}. P. Espa[ñ]{}ol, Phys. Rev. E, [**52**]{}, 1734 (1995). S. Wolfram, [*Mathematica, A system for doing mathematics by computer*]{}, (Addison-Wesley Publishing Company, Inc. 1991).

ArXiv

--- abstract: 'We present a magneto-infrared spectroscopy study on a newly identified three-dimensional (3D) Dirac semimetal ZrTe$_5$. We observe clear transitions between Landau levels and their further splitting under magnetic field. Both the sequence of transitions and their field dependence follow quantitatively the relation expected for 3D *massless* Dirac fermions. The measurement also reveals an exceptionally low magnetic field needed to drive the compound into its quantum limit, demonstrating that ZrTe$_5$ is an extremely clean system and ideal platform for studying 3D Dirac fermions. The splitting of the Landau levels provides a direct and bulk spectroscopic evidence that a relatively weak magnetic field can produce a sizeable Zeeman effect on the 3D Dirac fermions, which lifts the spin degeneracy of Landau levels. Our analysis indicates that the compound evolves from a Dirac semimetal into a topological line-node semimetal under current magnetic field configuration.' author: - 'R. Y. Chen$^{\dagger}$' - 'Z. G. Chen$^{\dagger}$' - 'X.-Y. Song' - 'J. A. Schneeloch' - 'G. D. Gu' - 'F. Wang' - 'N. L. Wang' bibliography: - 'ZrTe5.bib' title: ' Magneto-infrared spectroscopy of Landau levels and Zeeman splitting of three-dimensional massless Dirac Fermions in ZrTe$_5$ ' --- 3D topological Dirac/Weyl semimetals are new kinds of topological materials that possess linear band dispersion in the bulk along all three momentum directions[@PhysRevB.85.195320; @PhysRevB.88.125427; @PhysRevB.83.205101; @Liu2014-Science; @Liu2014-NatMater; @Borisenko2014; @Neupane2014]. Their low-energy quasiparticles are the condensed matter realization of Dirac and Weyl fermions in relativistic high energy physics[@Young2012; @Kharzeev2014]. These materials are expected to host many unusual phenomena[@He2014a; @Liang2015c; @Xu2015-Science], in particular the chiral and axial anomaly associated with Weyl fermions[@PhysRevB.83.205101; @Chen2013a; @Hosur2014; @Potter2014]. It is well known that the Dirac nodes are protected by both time-reversal and space inversion symmetry. Since magnetic field breaks the time-reversal symmetry, a Dirac node may be split into a pair of Weyl nodes along the magnetic field direction in the momentum space [@PhysRevLett.107.127205; @PhysRevB.84.235126; @Gorbar-MagneticDiracWeyl] or transformed into line-nodes[@PhysRevB.84.235126; @PhysRevB.90.115111]. Therefore, a Dirac semimetal can be considered as a parent compound to realize other topological variant quantum states. However, past 3D Dirac semimetal materials (e.g. Cd$_3$As$_2$) suffer from the problem of large residual carrier density which requires very high magnetic field (e.g. above 60 Tesla) to drive them to their quantum limit[@PhysRevLett.114.117201; @JCao]. This makes it extremely difficult to explore the transformation from Dirac to Weyl or line-node semimetals. Up to now, there are no direct evidences of such transformations. ZrTe$_5$ appears to be a new topological 3D Dirac material that exhibits novel and interesting properties. The compound crystallizes in the layered orthorhombic crystal structure, with prismatic ZrTe$_6$ chains running along the crystallographic $a$-axis and linked along the $c$-axis via zigzag chains of Te atoms to form two-dimensional (2D) layers. Those layers stack along the $b$-axis. A recent *ab initio* calculation suggests that bulk ZrTe$_5$ locates close to the phase boundary between weak and strong topological insulators [@PhysRevX.4.011002]. However, more recent transport and ARPES experiments identify it to be a 3D Dirac semimetal with only one Dirac node at the $\Gamma$ point [@LiQ]. Interestingly, a chiral magnetic effect associated with the transformation from a Dirac to Weyl semimetal was observed on ZrTe$_5$ through a magneto-transport measurement [@LiQ]. Our recent optical spectroscopy measurement at zero field revealed clearly a linear energy dependence of optical conductivity, being another hallmark of 3D massless Dirac fermions [@RYChen]. In this letter, we present magneto-infrared spectroscopy study on ZrTe$_5$ single crystals. We observe clear transitions between Landau levels and their further splitting under magnetic field. Both the sequence of transitions and their field dependence follow quantitatively the relation expected for 3D *massless* Dirac fermions. Furthermore, the measurement reveals an exceptionally low magnetic field (about 1 Tesla) needed to drive the compound into its quantum limit. Both facts demonstrate that ZrTe$_5$ is an extremely clean system and ideal platform for studying 3D Dirac fermions. The presence of further splitting of Landau levels, which has never been observed in 2D massless Dirac fermions, e.g. graphene, provides direct evidence for the lifting of spin degeneracy of Landau levels, an effect being linked to the transformation from a Dirac semimetal to a line-node or Weyl semimetal. Our theoretical analysis indicates that the former one is more likely realized in the present magnetic field configuration. Figure \[Fig:RB\] shows the reflectance spectra under different magnetic field $R(B)$ renormalized by the zero field reflectance $R(0)$ in the far- and mid-infrared region. For the lowest magnetic field (1 $T$), a series of peaks could be clearly resolved, which keep growing more pronounced and shift to higher energies when the field strength $B$ increases. In optical reflectance measurement, such peak features usually come from the interband transitions. Since these sharp peaks emerge in the reflectivity only by applying magnetic field, it is natural to connect them to the Landau quantization of 3D Dirac electrons. Thus the peaks should stem from electronic transitions connecting different Landau levels. Significantly, the first broad peak, which appears at the lowest energy, gradually split into four narrow peaks as $B$ increases. This character is quite intriguing and has never been observed ever before, which will be explained in detail later. For the sake of convenience, they are marked by the numbers “1, 2, 3, 4” respectively at the top of Fig. \[Fig:RB\]. However, the splittings of other peaks located at higher energies are rather vague in this plot. ![ The relative reflectivity of ZrTe$_5$ under magnetic field, as a function of energy. The spectrum are shifted upward by equal interval corresponding to different $B$.[]{data-label="Fig:RB"}](fig1.eps "fig:"){width="9cm"}\ In order to verify our speculation on the origin of the emerging peaks and capture the underlying physics of 3D Dirac semimetal, we examine the sequence of peaks observed at low field. For a 3D system, the band structure would transform into a set of 1D Landau levels by applying strong enough magnetic field, which are only dispersive along the field direction. Theoretical calculation on an isolated Weyl point has suggested that the magneto-optical conductivity is constituted of a series of asymmetric peaks lying on top of a linear background [@Hosur2012; @Ashby2013]. Especially, the peaks associated with allowed interband transitions in the Landau level structure occur at $\omega\propto \sqrt{n}+\sqrt{n+1}$, corresponding to transition from $L_{-n}$ to $L_{n+1}$ or from $L_{-(n+1)}$ to $L_n$, where $L_n$ represents for the $n$th Landau level. This conclusion applies to 3D massless Dirac fermion as well, because massless Dirac fermion can be thought as two sets of Weyl fermions with opposite chirality. A linear rising optical conductivity has been revealed in our previous zero field spectroscopic experiment on ZrTe$_5$ single crystal [@RYChen] which already provides a strong evidence for 3D massless Dirac or Weyl femions. We blew up the results of $B=2$ T as displayed in Fig. \[Fig:LL\], in which six peaks could be clearly resolved. The positions of these peaks are identified to be about $202$, $480$, $628$, $748$, $856$, $937$ [cm$^{-1}$]{}in sequence. The energy ratios of the peaks observed here can be approximately reduce to 1 : 1+$\sqrt2$ : $\sqrt2$+$\sqrt3$ : $\sqrt3$+$\sqrt4$ : $\sqrt4$+$\sqrt5$ : $\sqrt5$+$\sqrt6$, in nearly perfect accordance with the predicted massless Dirac semimetal behavior. This results in a linear dependence of the transition energy between Landau levels on $\sqrt{n}+\sqrt{n+1}$, as shown in the inset of the figure. From this observation, the first peak can be unambiguously determined to correspond to $n$=0, ascribed to transitions from $L_0$ to $L_1$ and $L_{-1}$ to $L_0$. Only when the chemical potential lies in between $L_1$ and $L_{-1}$, can this transition be clearly visible.[@Ashby2013] This is quite exciting because it demonstrates that the quantum limit could be easily approached by magnetic field as low as $1$ T, where the $n=0$ peak is distinctively observed. As a contrast, the quantum limit of the well-known Dirac semimetal Cd$_3$As$_2$ can not be reached with magnetic field lower than $65$ T [@PhysRevLett.114.117201; @JCao]. It indicates that ZrTe$_5$ compound is extremely close to an ideal Dirac semimetal, with the chemical potential lying in the vicinity of Dirac point, and meanwhile it is an extraordinarily clean system. We anticipate that our finding of easy access of quantum limit will motivate many other experimental probes on this compound. In Supplemental Information [@SI] we perform more detailed analysis of this peak sequence and obtain estimates of the average $ac$-plane Fermi velocity $v_{\perp}=\sqrt{v_a v_c} \sim 4.84\times 10^5$ (m/s) and a vanishingly small Dirac mass $|m|\sim 2$ (cm$^{-1}$). ![ The wave number dependent relative reflectivity $R(B)/R(0)$ under magnetic field of 2 T. $n=0\ldots 5$ point to six different peaks in sequence. The inset shows the linear dependence of the transition energies between Landau levels on $\sqrt{n}+\sqrt{n+1}$. The dash line is a guide to the eyes.[]{data-label="Fig:LL"}](fig2.eps "fig:"){width="7cm"}\ In Fig. \[Fig:RB\], where $R(B)/R(0)$ was shifted by equal interval with regard to increasing magnetic field, it is noted that the peak positions evolve in a way much alike the parabolic fashion as $B$. To further illustrate the characteristic features of the Landau levels, we plot $R(B)/R(0)$ in a pseudo-color photograph as a function of $\sqrt{B}$. It is clearly seen in Fig. \[Fig:mag\] that the wave numbers of the peaks are basically linear proportional to $\sqrt{B}$. The dashed lines are instructions for eyes, whose intercepts at $0$ T are all absolute zero. For a single massless 3D Dirac node, the $n$th Landau levels caused by magnetic field are dispersive only along the field direction (see Supplementary Information[@SI] for more details), with doubly degenerate $n\neq 0$ levels $E_n(k_{\parallel}) = \text{sgn}(n)\cdot \sqrt{2 v_{\perp}^2 eB\hbar\cdot |n|+\hbar^2 v_{\parallel}^2 k_{\parallel}^2}$, and $E_0(k_{\parallel})=\pm \hbar v_{\parallel} k_{\parallel}$ for $n=0$. If neglecting the dispersion along the magnetic field direction, then $E_n\propto \sqrt{B}$. As a comparison, for free electron systems, the magnetic induced Landau level obeys $E_n=(n+\frac{1}{2})\hbar\omega_c$, where $\omega_c$ is the the cyclotron angular velocity and proportional to $B$ instead of $\sqrt{B}$. Additionally, the Landau level energy of massive 3D Dirac fermions in topological insulator Bi$_2$Se$_3$ is reported to be in linear scale with $B$ as well[@Bi2Se3]. Therefore, the $\sqrt{B}$ dependence in Fig. \[Fig:mag\] intensively implies again the characteristic property of 3D massless Dirac fermions in ZrTe$_5$ under magnetic field. ![ The pseudo-colors photograph of the relative reflectivity $R(B)/R(0)$ as functions of wave number and $\sqrt{B}$. The dashed lines are linear fittings of the peak energies dependent on $\sqrt{B}$.[]{data-label="Fig:mag"}](fig3.eps "fig:"){width="8cm"}\ In addition to the sequence of peaks at low field, the peak splitting shown in Fig. \[Fig:RB\] could also be well resolved in Fig. \[Fig:mag\]. The $n=0$ peak evolves into four peaks at high magnetic field, with two very pronounced ones at lower energies and two relatively weak ones at higher energies. They are labeled as “1, 2, 3, 4” respectively in accordance with Fig. \[Fig:RB\] The splittings of the rest peaks are, although too vague to be precisely identified, but for certain to exist. The $n=1$ peak, arising from transitions from $L_{-2}$ to $L_1$ and $L_{-1}$ to $L_2$, seems to split into 3 peaks. Such splitting has never been observed in 2D massless Dirac fermion system, for example graphene[@PhysRevLett.97.266405]. ![ Two proposed scenarios. The left panel (a) shows the splitting of Landau levels simply caused by Zeeman effect. The strong spin-orbit coupling could lead to a mixing of spin up and down components. So the split “+” (solid lines) and “-” (dashed lines) levels do not have pure spin up and spin down components. This would allow inter Landau level transitions to occur with opposite signs, however, weak in intensity. The purple dashed line is the Fermi energy $E_F$. Zeeman field without orbital effect will transform 3D Dirac node into “line-nodes” in this case[@PhysRevB.84.235126]. The right panel (b) represents for the Landau levels of a Weyl semimetal induced by the magnetic field. The crossing points of $n=0$ Landau levels with the horizontal axis are the would-be Weyl points, if the orbital effect of magnetic field is ignored. The solid lines and dashed lines represent the two sets of Landau levels from the two Weyl nodes with opposite chirality. Transitions between Landau levels of opposite chirality will have weaker intensity. In both panels, the red solid arrows represent for transitions between Landau levels of the same spin/chirality, whereas the green dashed ones indicate that of different spin/chirality. []{data-label="Fig:zeeman"}](fig4.eps "fig:"){width="8cm"}\ We now explore the underlying mechanism for the splitting. We will show that this splitting can be naturally explained by the Zeeman effects of magnetic field on 3D Dirac fermions. The effect of Zeeman field on Dirac semimetals has been thoroughly studied by Burkov *et al.*[@PhysRevB.84.235126]. It was pointed out that Zeeman field may split the Dirac node into two Weyl nodes, or transform the Dirac node into “line-nodes”[@PhysRevB.84.235126]. Further including orbital effects of magnetic field will generate Zeeman-split Landau levels, schematically shown in Fig. \[Fig:zeeman\]. Here we will briefly explain the two possible scenarios and leave the details in the Supplementary Information[@SI]. The first scenario we consider is the “line-nodes” picture, with the resulting Landau level structure depicted in Fig. \[Fig:zeeman\](a). The doubly degenerate Landau levels $E_n(k_{\parallel})$ for $n\neq 0$ from 3D Dirac fermions will be split into $E_{n,\pm}\sim E_n(k_{\parallel})\pm \bar{g}\mu_B B/2$, where $\bar{g}$ is the average Landé $g$-factor for the conduction and valence bands of 3D Dirac fermion. The $n=0$ Landau levels $E_0(k_{\parallel})=\pm \hbar v_{\parallel} k_{\parallel}$ will mix around $k_{\parallel}=0$ and open a gap of size $\bar{g}\mu_B B$ there, and become $E_{0,\pm}\sim \pm \sqrt{\hbar^2 v_{\parallel}^2 k_{\parallel}^2+(\bar{g}\mu_B B/2)^2}$. The split Landau levels are labeled by “spin” indices ‘$+$’ and ‘$-$’ in Fig. \[Fig:zeeman\](a), which indicate that the states are of spin up or spin down at $k_{\parallel}=0$. However with $k_{\parallel}\neq 0$ the split Landau levels are not purely spin up or spin down due to strong spin-orbit coupling. Optical transitions between levels of the same “spin” indices can happen at $k_{\parallel}=0$, and produce strong peaks in optical conductivity (thus reflectivity)[@Ashby2013], but transitions between levels of opposite spin indices will be suppressed at $k_{\parallel}=0$, leading to weak and broad peaks. Therefore the original $n=0$ peak will split into two strong peaks “1” from $L_{0,-}$ to $L_{1,-}$, and “2” from $L_{-1,+}$ to $L_{0,+}$, and two weak peaks “3” from $L_{0,-}$ to $L_{1,+}$, and “4” from $L_{-1,-}$ to $L_{0,+}$. The peak “1” and peak “2” can have different energy if the chemical potential is not in the gap between $L_{0,-}$ and $L_{0,+}$ as depicted in Fig. \[Fig:zeeman\](a), or if the conduction and valence bands of 3D Dirac fermion have different $g$-factor (see Supplementary Information[@SI]). In any case the splitting between peak “3” and peak “1”, and between peak “4” and peak “2” will be about $\bar{g}\mu_B B$. Based on previous *ab initio* results[@PhysRevX.4.011002] on ZrTe$_5$ and the experimental setup we conclude that this scenario is the most likely explanation of our observation (see Supplementary Information[@SI] for details). The second and more interesting scenario is the “Weyl nodes” picture illustrated in Fig. \[Fig:zeeman\](b). In this case the Zeeman field effectively shifts the wave vector parallel to field by $\pm \bar{g}\mu_B B/2\hbar v_{\parallel}$, where the $\pm$ sign depends on the chirality of Weyl fermions. The degenerate $n\neq 0$ Landau levels become $E_{n,\pm}(k_{\parallel})\sim E_n(k_{\parallel}\mp \bar{g}\mu_B B/2\hbar v_{\parallel})$, and the $n=0$ Landau levels become $E_{0,\pm}(k_{\parallel})\sim \pm \hbar v_{\parallel} k_{\parallel}-\bar{g}\mu_B B/2$. In this scenario transitions between Landau levels of the same (different) chirality will have strong (weak) intensity. The original $n=0$ peak will also split into two strong peaks “1” from $L_{0,-s}$ to $L_{1,-s}$, and “2” from $L_{-1,-s}$ to $L_{0,-s}$, and two weak peaks “3” from $L_{0,-s}$ to $L_{1,s}$, and “4” from $L_{-1,s}$ to $L_{0,-s}$, where $s=\pm$. The peak “1” and peak “2” can have different energy if the chemical potential is not at the charge neutrality as depicted in Fig. \[Fig:zeeman\](b). However the splitting between peak “3” and “1” will in general not be linear in $B$, unless the conduction and valence bands have very different $g$-factors. According to our analysis[@SI] this scenario is more likely to happen when the field is applied along crystal $c$-direction. ![ The magnetic field dependent energy difference of $E_3$-$E_1$ and $E_4$-$E_2$. They represent the Landau level splittings for the conduction and valence bands, respectively. []{data-label="Fig:linear"}](fig5.eps "fig:"){width="6cm"}\ From the peak positions of “1”, “2”, “3” and “4”, we can immediately obtain the dependence of the split energy as a function of magnetic field. Figure \[Fig:linear\] displays the magnetic field dependence of $E_4-E_2$ above 6 T and $E_3$-$E_1$ above $13$ T, respectively. The energy positions for peak “3” could not be well resolved below 12 T, so the energy difference of $E_3-E_1$ is plotted only at high magnetic field. Obviously, both $E_4-E_2$ and $E_3$-$E_1$ exhibit good linear dependence, in better agreement with the first “line-nodes” scenario. The energy splitting is roughly 10.5 cm$^{-1}$/T ($\sim$1.3 meV/T) for $E_4-E_2$ and 7.4 cm$^{-1}$/T ($\sim$0.92 meV/T) for $E_3-E_1$. According to previous discussions, this leads to estimates of the average $g$-factor being $22.5$ or $15.8$, from the $E_4 - E_2$ or $E_3 - E_1$ splittings respectively. It was known that the $g$-factor reaches about 37 for Cd$_3$As$_2$[@Jeon2014], which is even bigger than our estimated $g$-factor for the ZrTe$_5$ compound. We do not yet have a good explanation for the discrepancy between $E_4 - E_2$ and $E_3 - E_1$ splittings. Based on our analysis in Supplementary Information[@SI], this could be due to the broad nature of the weaker peaks, and the current analysis may overestimate the $g$-factors for this reason. The transitions associated with $L_{-2}\rightarrow L_{1}$ and $L_{-1}\rightarrow L_2$ are much more complex. Analogy to the transitions between $L_0$ and $L_{\pm 1}$, the selection rule permissive ones (indicated by red solid arrows) should be more pronounced than those between opposite spin orientations (the dashed green ones). Considering possible different splittings of the valence and conduction bands, the $n=1$ peak appeared in $R(B)/R(0)$ is supposed to contains at least three components. In summary, by performing magneto-optical measurement on the single crystalline 3D massless Dirac semimetal ZrTe$_5$, we have clearly observed the magnetic field induced Landau levels, evidenced by regular organized peaks shown in the renormalized reflectivity $R(B)/R(0)$. Particularly, the first peak is identified to be originated from the transitions between the zeroth and first Landau Levels, which reveals the Fermi energy lies very close to the Dirac point. The appearance of the first peak under magnetic field as low as $1$ T demonstrates an exceptionally low quantum limit of ZrTe$_5$ compared to other 3D Dirac semimetals, which provided an elegant platform to explore more intriguing non-trivial quantum phenomena. Of most importance, the fourfold splitting of the first peak yield direct and clear evidence for the release of spin degeneracy of Landau level, hence the transformation from a Dirac semimetal into line-node or Weyl semimetal. Our theoretical modeling and analysis indicate that the former one is more likely realized in the present magnetic field configuration. $^{\dagger}$ These authors contributed equally to this work. We acknowledge very helpful discussions with Z. Fang, H. M. Weng, X. C. Xie, D. H. Lee, L. Fu, Q. Li, X. Dai, H. W. Liu. This work was supported by the National Science Foundation of China (11120101003, 11327806, 11374018), and the 973 project of the Ministry of Science and Technology of China (2011CB921701, 2012CB821403, 2014CB920902). Work at Brookhaven is supported by the Office of Basic Energy Sciences, Division of Materials Sciences and Engineering, U.S. Department of Energy under Contract No. DE-SC00112704.

ArXiv

--- author: - | S. Maschio\ Dipartimento di Matematica, Università di Padova\ Via Trieste, Padova\ [samuele.maschio@math.unipd.it]{}\ T. Streicher\ Fachbereich 4 Mathematik, TU Darmstadt\ Schlo[ß]{}gartenstr. 7, D-64289 Darmstadt, Germany\ [streicher@mathematik.tu-darmstadt.de]{} title: | Models of Intuitionistic Set Theory\ in Subtoposes of Nested Realizability Toposes --- Introduction {#introduction .unnumbered} ============ Given a *partial combinatory algebra* (pca) ${{\mathcal{A}}}$ (see e.g.  [@VOO08]) together with a subpca ${{\mathcal{A}}}_\#$ of ${{\mathcal{A}}}$ we will construct the *nested realizability topos* ${{\mathbf{RT}}}({{\mathcal{A}}},{{\mathcal{A}}}_\#)$ as described in [@BVO02] (without giving it a proper name there). It is well known (from e.g. [@VOO08]) that ${{\mathbf{RT}}}({{\mathcal{A}}},{{\mathcal{A}}}_\#)$ appears as the exact/regular completion of its subcategory ${{\mathbf{Asm}}}({{\mathcal{A}}},{{\mathcal{A}}}_\#)$ of assemblies. In [@BVO02] the authors considered two complementary subtoposes of ${{\mathbf{RT}}}({{\mathcal{A}}},{{\mathcal{A}}}_\#)$, namely the *relative realizability topos* ${{\mathbf{RT}}}_r({{\mathcal{A}}},{{\mathcal{A}}}_\#)$ and the *modified relative realizability topos* ${{\mathbf{RT}}}_m({{\mathcal{A}}},{{\mathcal{A}}}_\#)$, respectively. Within nested realizability toposes we will identify a class of *small maps* giving rise to a model of intuitionistic set theory ${\mathsf{IZF}}$ (see [@FRI73; @MCC84]) as described in [@JM95]. For this purpose we first identify a class of display maps in ${{\mathbf{Asm}}}({{\mathcal{A}}},{{\mathcal{A}}}_\#)$ which using a result of [@VDBM08] gives rise to the desired class of small maps in the exact/regular completion ${{\mathbf{RT}}}({{\mathcal{A}}},{{\mathcal{A}}}_\#)$ of ${{\mathbf{Asm}}}({{\mathcal{A}}},{{\mathcal{A}}}_\#)$. For showing that the subtoposes ${{\mathbf{RT}}}_r({{\mathcal{A}}},{{\mathcal{A}}}_\#)$ and ${{\mathbf{RT}}}_m({{\mathcal{A}}},{{\mathcal{A}}}_\#)$ also give rise to models of ${\mathsf{IZF}}$ we will prove the following general result. If ${{\mathcal{E}}}$ is a topos with a class ${{\mathcal{S}}}$ of small maps and ${{\mathcal{F}}}$ is a subtopos of ${{\mathcal{E}}}$ then there is a class ${{\mathcal{S}}}_{{\mathcal{F}}}$ of small maps in ${{\mathcal{F}}}$ which is obtained by closing sheafifications of maps in ${{\mathcal{S}}}$ under quotients in ${{\mathcal{F}}}$. As explained in subsections \[modreal\] and \[HerbTop\] below this covers also the Modified Realizability topos as studied in [@VOO97] and the more recent Herbrand topos of van den Berg. Nested Realizability Toposes and some of their Subtoposes ========================================================= Given a pca ${{\mathcal{A}}}$ in an elementary topos $\mathscr{S}$ we may construct the realizability topos ${{\mathbf{RT}}}_{\mathscr{S}}({{\mathcal{A}}})$ relative to ${{\mathcal{S}}}$ as described in [@VOO08]. If $\mathscr{S}$ is the Sierpiński topos ${\mathbf{Set}}^{\mathbf{2}^{{\mathsf{op}}}}$ then a “nested pca”, i.e. a pca ${{\mathcal{A}}}$ together with a subpca ${{\mathcal{A}}}_\#$ gives rise to a pca internal to ${\mathbf{Set}}^{\mathbf{2}^{{\mathsf{op}}}}$ from which one may construct the “nested realizability topos” ${{\mathbf{RT}}}({{\mathcal{A}}},{{\mathcal{A}}}_\#)$ as described in [@BVO02; @VOO08].[^1] Within ${{\mathbf{RT}}}({{\mathcal{A}}},{{\mathcal{A}}}_\#)$ there is a unique nontrivial subterminal object $u$ giving rise to the *open* subtopos induced by the closure operator $u \to (-)$ and the complementary subtopos induced by the closure operator $u \vee (-)$ as described in [@BVO02]. Next we will give more elementary descriptions of ${{\mathbf{RT}}}({{\mathcal{A}}},{{\mathcal{A}}}_\#)$ and the above mentioned subtoposes. The Nested Realizability Topos ${{\mathbf{RT}}}({{\mathcal{A}}},{{\mathcal{A}}}_\#)$ ------------------------------------------------------------------------------------ Let ${{\mathcal{A}}}$ be a pca whose partial application is denoted by juxtaposition and ${{\mathcal{A}}}_\#$ be a subpca of ${{\mathcal{A}}}$, i.e. ${{\mathcal{A}}}_\#$ is a subset of ${{\mathcal{A}}}$ closed under application and there are elements $k$ and $s$ of ${{\mathcal{A}}}_\sharp$ such that for all $x,y,z \in {{\mathcal{A}}}$ it holds that $kxy = x$, $sxyz \simeq xz(yz)$ and $sxy$ is always defined. We write $i$ for $skk$ and $\bar{k}$ for $ki$ which, obviously, satisfy the equations $ix = x$ and $\bar{k}xy = y$, respectively. We write $p$, $p_0$ and $p_1$ for elements of ${{\mathcal{A}}}$ such that $px_0x_1$ is always defined and $p_i(px_0x_1) = x_i$ for $i = 0,1$. For every natural number $n$ we write $\underline{n}$ for the corresponding numeral as defined in [@VOO08]. Notice that $k$, $\bar{k}$, $p$, $p_0$, $p_1$ and the numerals $\underline{n}$ are all elements of ${{\mathcal{A}}}_\#$. Since subsets of ${{\mathcal{A}}}$ are the propositions of the realizability topos ${{\mathbf{RT}}}({{\mathcal{A}}})$ it is useful to fix some notation for the propositional connectives 1. $A \rightarrow B = \left\{ a \in {{\mathcal{A}}}\mid ax \in B \textrm{ for all } x \in A \right\}$ 2. $A \wedge B = \left\{ pxy \mid x\in A, y \in B \right\}$ 3. $A \vee B = \left(\{ k \} \wedge A\right) \cup \left(\{ \bar{k} \} \wedge B \right)$ Propositions of the nested realizability topos ${{\mathbf{RT}}}({{\mathcal{A}}},{{\mathcal{A}}}_\#)$ will be pairs $A = (A_p,A_a) \in {\mathcal{P}}({{\mathcal{A}}}) \times {\mathcal{P}}({{\mathcal{A}}}_\#)$ such that $A_a \subseteq A_p$ where we call $A_p$ and $A_a$ the set of *potential* and *actual* realizers, respectively. We write $\Sigma({{\mathcal{A}}},{{\mathcal{A}}}_\#)$ for the set of these propositions. The above notation for propositional connectives is adapted to the current class of propositions as follows 1. $A \rightarrow B = \left( A_p \rightarrow B_p, {{\mathcal{A}}}_\# \cap (A_p \rightarrow B_p) \cap (A_a \rightarrow B_a)\right)$ 2. $A \wedge B = (A_p \wedge B_p, A_a \wedge B_a)$ 3. $A \vee B = (A_p \vee B_p, A_a \vee B_a)$ For the realizability tripos $\mathscr{P}({{\mathcal{A}}})$ induced by the pca ${{\mathcal{A}}}$ see [@VOO08]. The nested realizability tripos ${{\mathscr{P}}}({{\mathcal{A}}},{{\mathcal{A}}}_\#)$ over ${\mathbf{Set}}$ induced by the nested pca ${{\mathcal{A}}}_\# \subseteq {{\mathcal{A}}}$ is defined as follows. For a set $I$ the fibre ${{\mathscr{P}}}({{\mathcal{A}}},{{\mathcal{A}}}_\#)(I)$ is given by the set $\Sigma({{\mathcal{A}}},{{\mathcal{A}}}_\#)^I$ preordered by the relation $\vdash_I$ defined as $$\phi \vdash_{I} \psi \qquad \textrm{ if and only if } \qquad \bigcap_{i\in I}(\phi(i) \rightarrow \psi(i))_{a} \neq \emptyset$$ for $\phi,\psi \in {{\mathscr{P}}}({{\mathcal{A}}},{{\mathcal{A}}}_\#)(I)$. For $u : J \to I$ reindexing along $u$ is given by precomposition with $u$ and denoted as $u^*$. The fibres are preHeyting algebras where the propositional connectives are given by applying the operations $\rightarrow$, $\wedge$ and $\vee$ pointwise. It is easy to check that $u^*$ commutes with the propositional connectives in the fibres. For a map $u : J \to I$, the reindexing $u^*$ has left and right adjoints $\exists_u$ and $\forall_u$, respectively, given by $$\exists_u(\phi)(i) = \left(\bigcup_{u(j)=i} \phi_p(j), \bigcup_{u(j)=i}\phi_a(j)\right)$$ $$\forall_u(\phi)(i) = \left(\bigcap_{j \in J}(Eq(u(j),i) \rightarrow \phi(j))_p, \bigcap_{j \in J}(Eq(u(j),i) \rightarrow \phi(j))_a \right)$$ where $Eq(x,y) = \left(\left\{a \in {{\mathcal{A}}}\mid x = y\right\},\left\{a \in {{\mathcal{A}}}_\# \mid x = y\right\}\right)$. It is straightforward to check that the so defined quantifiers satisfy the respective Beck-Chevalley conditions. The identity on $\Sigma({{\mathcal{A}}},{{\mathcal{A}}}_\#)$ gives rise to a generic family and, therefore, the fibered preorder ${{\mathscr{P}}}({{\mathcal{A}}},{{\mathcal{A}}}_\#)$ is actually a tripos in the sense of [@HJP80]. We write ${{\mathbf{RT}}}({{\mathcal{A}}},{{\mathcal{A}}}_\#)$ for the ensuing topos. Some Subtoposes of ${{\mathbf{RT}}}({{\mathcal{A}}},{{\mathcal{A}}}_\#)$ {#subtopex} ------------------------------------------------------------------------ In ${{\mathbf{RT}}}({{\mathcal{A}}},{{\mathcal{A}}}_\#)$ there is a nontrivial subterminal $u = ({{\mathcal{A}}},\emptyset)$ giving rise to two complementary subtoposes induced by the closure operators $o_u(p) = u \to p$ and $c_u(p) = u \vee p$ as in [@BVO02]. We denote the open subtopos induced by $o_u$ by ${{\mathbf{RT}}}_r({{\mathcal{A}}},{{\mathcal{A}}}_\#)$ and the complementary subtopos induced by $c_u$ by ${{\mathbf{RT}}}_m({{\mathcal{A}}},{{\mathcal{A}}}_\#)$. In [@BVO02] these two subtoposes are referred to as the *relative* and the *modified relative realizability* topos, respectively. For sake of concreteness and later reference in the following two subsections we give an elementary and explicit construction of triposes inducing ${{\mathbf{RT}}}_r({{\mathcal{A}}},{{\mathcal{A}}}_\#)$ and ${{\mathbf{RT}}}_m({{\mathcal{A}}},{{\mathcal{A}}}_\#)$, respectively. ### The Relative Realizability Topos ${{\mathbf{RT}}}_r({{\mathcal{A}}},{{\mathcal{A}}}_\#)$ is induced by the tripos ${{\mathscr{P}}}_r({{\mathcal{A}}},{{\mathcal{A}}}_\#)$ over ${\mathbf{Set}}$ which we describe next. Let $\Sigma_r({{\mathcal{A}}},{{\mathcal{A}}}_\#) = {\mathcal{P}}({{\mathcal{A}}})$. The fibre of ${{\mathscr{P}}}_r({{\mathcal{A}}},{{\mathcal{A}}}_\#)$ over $I$ is given by the preorder $\left({\mathcal{P}}({{\mathcal{A}}})^{I},\vdash^r_I\right)$ where $$\phi \vdash_I^r \psi \qquad \textrm{ if and only if } \qquad {{\mathcal{A}}}_\# \cap \bigcap_{i\in I} (\phi(i) \rightarrow \psi(i)) \neq \emptyset$$ and as usual reindexing is given by precomposition. At first sight this tripos looks like the tripos ${{\mathscr{P}}}({{\mathcal{A}}})$ inducing the realizability topos ${{\mathbf{RT}}}({{\mathcal{A}}})$ but notice that entailment in the fibres is defined in a more restrictive way, namely by requiring that the entailment be realized by an element of ${{\mathcal{A}}}_\#$ and not just an element of ${{\mathcal{A}}}$. Nevertheless, the propositional connectives, quantifiers and the generic family of ${{\mathbf{RT}}}_r({{\mathcal{A}}},{{\mathcal{A}}}_\#)$ can be constructed according to the same recipes as for ${{\mathscr{P}}}({{\mathcal{A}}})$ (see [@VOO08]). There is an obvious logical morphism from ${{\mathscr{P}}}_r({{\mathcal{A}}},{{\mathcal{A}}}_\#)$ to ${{\mathscr{P}}}({{\mathcal{A}}})$ which is the identity on objects. But there is also an injective geometric morphism from ${{\mathscr{P}}}_r({{\mathcal{A}}},{{\mathcal{A}}}_\#)$ to ${{\mathscr{P}}}({{\mathcal{A}}},{{\mathcal{A}}}_\#)$ sending a family $\phi \in {\mathcal{P}}({{\mathcal{A}}})^I$ to the family $\lambda i{:}I. (\phi(i),{{\mathcal{A}}}_\# \cap \phi(i))$. These morphisms between triposes over ${\mathbf{Set}}$ extend to morphisms between the associated toposes as described in [@VOO08]. ### The Modified Relative Realizability Topos ${{\mathbf{RT}}}_m({{\mathcal{A}}},{{\mathcal{A}}}_\#)$ {#modreal} is induced by the tripos ${{\mathscr{P}}}_m({{\mathcal{A}}},{{\mathcal{A}}}_\#)$ over ${\mathbf{Set}}$ which is obtained from ${{\mathscr{P}}}({{\mathcal{A}}},{{\mathcal{A}}}_\#)$ by restricting the fibre ${{\mathscr{P}}}({{\mathcal{A}}},{{\mathcal{A}}}_\#)(I)$ to the set of all $\phi \in \Sigma({{\mathcal{A}}},{{\mathcal{A}}}_\#)^I$ with ${{\mathcal{A}}}_\# \cap \bigcap_{i \in I} \phi_p(i) \neq \emptyset$. The logical structure is essentially inherited from ${{\mathscr{P}}}({{\mathcal{A}}},{{\mathcal{A}}}_\#)$ though now and then one has to insert the closure operator $c_U$ in order to stay within ${{\mathscr{P}}}_m({{\mathcal{A}}},{{\mathcal{A}}}_\#)$. A generic family for ${{\mathscr{P}}}_m({{\mathcal{A}}},{{\mathcal{A}}}_\#)$ is given by the identity on $\Sigma_m({{\mathcal{A}}},{{\mathcal{A}}}_\#) = \{ u \vee p \mid p \in \Sigma({{\mathcal{A}}},{{\mathcal{A}}}_\#) \}$. The obvious inclusion of ${{\mathscr{P}}}_m({{\mathcal{A}}},{{\mathcal{A}}}_\#)$ into ${{\mathscr{P}}}({{\mathcal{A}}},{{\mathcal{A}}}_\#)$ gives rise to the inclusion of ${{\mathbf{RT}}}_m({{\mathcal{A}}},{{\mathcal{A}}}_\#)$ into ${{\mathbf{RT}}}({{\mathcal{A}}},{{\mathcal{A}}}_\#)$. Notice that in case ${{\mathcal{A}}}= {{\mathcal{A}}}_\#$ we obtain the *modified realizability* topos as described in [@VOO97; @VOO08] for the case where ${{\mathcal{A}}}$ is the first Kleene algebra $\mathcal{K}_1$. Another prominent example of a modified relative realizability model can be found in a paper by J. R. Moschovakis [@MOS] from 1971 where she constructed a model for a theory $\mathrm{INT}$ of Brouwerian intuitionism validating the proposition that all functions on natural numbers are not not recursive, i.e.that there are no non-recursive functions on the natural numbers. Of course, the model of [@MOS] was not constructed in topos-theoretic terms but it is equivalent to the interpretation of the system considered in *loc.cit.* in the topos ${{\mathbf{RT}}}_m(\mathcal{K}_2,\mathcal{K}_2^{\mathit{rec}})$ where $\mathcal{K}_2$ is the second Kleene algebra whose underlying set is Baire space ${\mathbb{N}}^{\mathbb{N}}$ and $\mathcal{K}_2^{\mathit{rec}})$ is the sub-pca of recursive sequences of natural numbers. The ensuing interpretation of $\mathrm{INT}$ was called *$G$-realizability* in *loc.cit.* ### The Herbrand Realizability Topos {#HerbTop} As shown by J. van Oosten, see Lemma 3.2 of [@JohGl], B. van den Berg’s *Herbrand realizability topos* over a pca ${{\mathcal{A}}}$ arises as a subtopos of ${{\mathbf{RT}}}({{\mathcal{A}}},{{\mathcal{A}}})$ induced by some closure operator on ${{\mathscr{P}}}({{\mathcal{A}}},{{\mathcal{A}}})$. Moreover, as shown in *loc.cit.* it is disjoint from the open subtopos ${{\mathbf{RT}}}_r({{\mathcal{A}}},{{\mathcal{A}}})$ equivalent to ${{\mathbf{RT}}}({{\mathcal{A}}})$. Assemblies induced by ${{\mathscr{P}}}({{\mathcal{A}}},{{\mathcal{A}}}_\#)$ --------------------------------------------------------------------------- As described in [@VOO08] for every tripos ${{\mathscr{P}}}$ (over ${\mathbf{Set}}$) one may consider the full subcategory ${{\mathbf{Asm}}}({{\mathscr{P}}})$ of *assemblies* in ${\mathbf{Set}}({{\mathscr{P}}})$, i.e. subobjects of objects of the form $\Delta(S)$ where $S \in {\mathbf{Set}}$ and $\Delta : {\mathbf{Set}}\to {\mathbf{Set}}({{\mathscr{P}}})$ is the *constant objects* functor sending a set $S$ to $(S,\exists_{\delta_S}(\top_S))$.[^2] One can show that the category ${{\mathbf{Asm}}}({{\mathscr{P}}}({{\mathcal{A}}},{{\mathcal{A}}}_\#))$ is equivalent to the category ${{\mathbf{Asm}}}({{\mathcal{A}}},{{\mathcal{A}}}_\#)$ whose objects are pairs $X = (|X|,E_X)$ where $|X|$ is a set and $E_X : |X| \to \Sigma({{\mathcal{A}}},{{\mathcal{A}}}_\#)$ with $E_X(x)_p \neq \emptyset$ for all $x \in |X|$. An arrow from $X$ to $Y$ is a function $f : |X| \rightarrow |Y|$ such that $E_X \vdash_{|X|} f^*E_Y$. As follows from [@VOO08] Cor. 2.4.5 the topos ${{\mathbf{RT}}}({{\mathcal{A}}},{{\mathcal{A}}}_\#)$ appears as the exact/regular completion of ${{\mathbf{Asm}}}({{\mathcal{A}}},{{\mathcal{A}}}_\#)$. For further reference we note the following \[AsmplcH\] ${{\mathbf{Asm}}}({{\mathcal{A}}},{{\mathcal{A}}}_\#)$ is a locally cartesian closed Heyting category with stable and disjoint finite sums with a generic monomorphism $\top : {{\mathit{Tr}}}{\rightarrowtail}{{\mathit{Prop}}}$.[^3] The locally cartesian closed structure is constructed as in the case of ${{\mathbf{Asm}}}({{\mathcal{A}}})$, i.e. assemblies within ${{\mathbf{RT}}}({{\mathcal{A}}})$ where ${{\mathcal{A}}}$ is a pca. Similarly, one shows that ${{\mathbf{Asm}}}({{\mathcal{A}}},{{\mathcal{A}}}_\#)$ is a Heyting category and it has stable and disjoint finite sums. Finally we exhibit a generic mono $\top : {{\mathit{Tr}}}{\rightarrowtail}{{\mathit{Prop}}}$. The object ${{\mathit{Prop}}}$ is defined as $\Delta(\Sigma({{\mathcal{A}}},{{\mathcal{A}}}_\#))$. The underlying set of ${{\mathit{Tr}}}$ is the subset of $\Sigma({{\mathcal{A}}},{{\mathcal{A}}}_\#)$ consisting of those pairs $A = (A_p,A_a)$ where $A_p \neq \emptyset$ and $E_{{\mathit{Tr}}}(A) = A$. Notice, however, that in general ${{\mathbf{Asm}}}({{\mathcal{A}}},{{\mathcal{A}}}_\#)$ is not well-pointed. Some Facts about Small Maps =========================== A Heyting category is a regular category ${{\mathcal{C}}}$ where for all $f : Y \to X$ in ${{\mathcal{C}}}$ the pullback functor $f^{-1} : {{\mathsf{Sub}}}_{{\mathcal{C}}}(X) \to {{\mathsf{Sub}}}_{{\mathcal{C}}}(Y)$ has a right adjoint $\forall_f$. It is a Heyting pretopos iff, moreover, it has stable disjoint finite sums and every equivalence relation is effective (i.e.  appears as kernel pair of its coequalizer). \[Sdef\] Let ${{\mathcal{C}}}$ be a locally cartesian Heyting category with stable and disjoint finite sums and a natural numbers object $N$. For a class ${{\mathcal{S}}}$ of maps in ${{\mathcal{C}}}$ we consider the following properties. - (Pullback Stability) For a pullback square D & \^q & B\ \^g & & \_f\ C & \_p & A in ${{\mathcal{C}}}$ from $f \in {{\mathcal{S}}}$ it follows that $g \in {{\mathcal{S}}}$. - (Descent) If in a pullback square as above $p$ is a cover, i.e. a regular epimorphism, then $f \in {{\mathcal{S}}}$ whenever $g \in {{\mathcal{S}}}$. - (Sums) If $f$ and $g$ are in ${{\mathcal{S}}}$ then $f + g$ is in ${{\mathcal{S}}}$. - (Finiteness) The maps $0 \to 1$, $1 \to 1$ and $1 + 1 \to 1$ are in ${{\mathcal{S}}}$. - (Composition) Maps in ${{\mathcal{S}}}$ are closed under composition. - (Quotient) If $f \circ e$ is in ${{\mathcal{S}}}$ and $e$ is a cover then $f$ is in ${{\mathcal{S}}}$. - (Collection) Any arrows $p : Y \to X$ and $f : X \to A$ where $p$ is a cover and $f \in {{\mathcal{S}}}$ fit into a quasipullback[^4] Z & & Y & \^p & X\ \^g & & & & \_f\ B & & \_h & & A where $g \in {{\mathcal{S}}}$ and $h$ is a cover. - (Representability) There is a universal family $\pi : E \to U$ in ${{\mathcal{S}}}$ such that every $f : Y \to X$ in ${{\mathcal{S}}}$ fits into a diagram Y & & Y\^& & E\ \^f & & \^[f\^]{} & & \_\ X & & X\^& & U\ where the left square is a quasipullback and the right square is a pullback. - (Infinity) The terminal projection $N \to 1$ is in ${{\mathcal{S}}}$. - (Separation) All monomorphisms are in ${{\mathcal{S}}}$. A class ${{\mathcal{S}}}$ of maps in ${{\mathcal{C}}}$ validating properties [**(A0)**]{}–[**(A9)**]{} is called a class of small maps. The following theorem will be essential later on. \[bennoextended\] Let ${{\mathcal{C}}}$ be a Heyting category with stable and disjoint finite sums and ${{\mathcal{S}}}$ be a class of small maps in ${{\mathcal{C}}}$. Let $\bar{{{\mathcal{C}}}}$ be the exact/regular completion of ${{\mathcal{C}}}$ and $\bar{{{\mathcal{S}}}}$ the class of maps $f$ in $\bar{{{\mathcal{C}}}}$ which fit into a quasipullback & &\ \^g & & \_f\ & & with $g$ in the subcategory ${{\mathcal{C}}}$ of $\bar{{{\mathcal{C}}}}$. Then $\bar{{{\mathcal{S}}}}$ is a class of small maps within the Heyting pretopos $\bar{{{\mathcal{C}}}}$. That $\bar{{{\mathcal{S}}}}$ validates conditions [**(A0)**]{}–[**(A8)**]{} follows from Lemma 5.8 and Propositions 6.2 and 6.21 in [@VDBM08]. Condition [**(A9)**]{} holds for $\bar{{{\mathcal{S}}}}$ in $\bar{{{\mathcal{C}}}}$ for the following reason. Let $m : B {\rightarrowtail}A$ be a mono in $\bar{{{\mathcal{C}}}}$. Since $\bar{{{\mathcal{C}}}}$ is the exact completion of ${{\mathcal{C}}}$ there is a cover $p : X {\rightarrowtriangle}A$ with $X$ in ${{\mathcal{C}}}$. Then for the pullback Y & \^q & B\ \^n & & \_m\ X & \_p & A in $\bar{{{\mathcal{C}}}}$ we know that $q$ is a cover and $n$ is a mono. It follows from Lemma 2.4.4 of [@VOO08] that $Y$ is isomorphic to an object in ${{\mathcal{C}}}$. Small Maps in Nested Realizability Toposes ========================================== We will first identify within ${{\mathbf{Asm}}}({{\mathcal{A}}},{{\mathcal{A}}}_\#)$ a class ${{\mathcal{S}}}$ of small maps so that we can apply Theorem \[bennoextended\] to it in order to obtain a class $\bar{{{\mathcal{S}}}}$ of small maps on ${{\mathbf{RT}}}({{\mathcal{A}}},{{\mathcal{A}}}_\#)$ which is known to arise as the exact/regular completion of ${{\mathbf{Asm}}}({{\mathcal{A}}},{{\mathcal{A}}}_\#)$ (see section 2.4 of [@VOO08] for more details). However, for showing that $\bar{{{\mathcal{S}}}}$ is closed under power types we have to appeal to Lemma 27 of [@VDBM11] guaranteeing that if ${{\mathbf{Asm}}}({{\mathcal{A}}},{{\mathcal{A}}}_\#)$ has *weak* power types under which ${{\mathcal{S}}}$ is closed then ${{\mathbf{RT}}}({{\mathcal{A}}},{{\mathcal{A}}}_\#)$ has power objects under which $\bar{{{\mathcal{S}}}}$ is closed. Small maps in ${{\mathbf{Asm}}}({{\mathcal{A}}},{{\mathcal{A}}}_\#)$ -------------------------------------------------------------------- For constructing a class of small maps in ${{\mathbf{Asm}}}({{\mathcal{A}}},{{\mathcal{A}}}_\#)$ let us first choose a *strongly inaccessible* cardinal $\kappa$ exceeding the cardinality of ${{\mathcal{A}}}$. \[Sclos1\] Let ${{\mathcal{S}}}$ be the class of all maps $f:Y \to X$ in ${{\mathbf{Asm}}}({{\mathcal{A}}},{{\mathcal{A}}}_\#)$ such that ${{\mathsf{card}}}\left(f^{-1}(x)\right) < \kappa$ for all $x \in |X|$. Then ${{\mathcal{S}}}$ is a class of small maps in ${{\mathbf{Asm}}}({{\mathcal{A}}},{{\mathcal{A}}}_\#)$ in the sense of Def. \[Sdef\]. Conditions [**(A0)**]{} and [**(A1)**]{} follow from the fact that the forgetful functor from ${{\mathbf{Asm}}}({{\mathcal{A}}},{{\mathcal{A}}}_\#)$ to ${\mathbf{Set}}$ preserves finite limits and covers. Since the forgetful functor from ${{\mathbf{Asm}}}({{\mathcal{A}}},{{\mathcal{A}}}_\#)$ to ${\mathbf{Set}}$ preserves finite sums condition [**(A2)**]{} holds. Since $\kappa$ is infinite all maps in ${{\mathbf{Asm}}}({{\mathcal{A}}},{{\mathcal{A}}}_\#)$ with finite fibres are in ${{\mathcal{S}}}$. For this reason [**(A3)**]{} and [**(A9)**]{} trivially hold. Condition [**(A4)**]{} holds since $\kappa$ is regular. For [**(A5)**]{} suppose $f \circ e$ is in ${{\mathcal{S}}}$ and $e$ is a cover. Then the fibres of $f$ have cardinalities $< \kappa$ since by assumption the fibres of $f \circ e$ have cardinalities $< \kappa$ and the underlying map of $e$ is onto. Condition [**(A8)**]{} holds since $\kappa$ exceeds the cardinality of ${\mathbb{N}}$. For showing that [**(A6)**]{} holds suppose $p : Y \to X$ is a cover and $f : X \to A$ is in ${{\mathcal{S}}}$. Since $p$ is a cover the underlying map of $p$ (also denoted by $p$) is onto and there exists $a \in {{\mathcal{A}}}_\#$ such that for all $x \in |X|$ it holds that 1. if $b \in E_X(x)_p$ then $ab{\downarrow}$ and $ab \in E_Y(y_{x,b})$ for some $y_{x,b} \in p^{-1}(x)$ and 2. if $b \in E_X(x)_a$ then $ab \in E_Y(y_{x,b})_a$. Let $Z$ be the object of ${{\mathbf{Asm}}}({{\mathcal{A}}},{{\mathcal{A}}}_\#)$ whose underlying set $|Z| = \{ y_{x,b} \mid x \in |X|, b \in E_X(x)_p \}$ and $E_Z(y) = E_Y(y)$ for $y \in |Z|$. Let $i : Z {\hookrightarrow}Y$ be the obvious inclusion of $Z$ into $Y$. Then the rectangle Z & \^i & Y & \^p & X\ & & & & \_f\ A & & & & A is a quasipullback since $p \circ i$ is a cover. Since the fibres of $p \circ i$ have cardinality $\leq {{\mathsf{card}}}({{\mathcal{A}}}) < \kappa$ the map $p \circ i$ is in ${{\mathcal{S}}}$. Thus, by [**(A4)**]{} the map $f \circ p \circ i : Z \to A$ is in ${{\mathcal{S}}}$, too. Condition [**(A7)**]{} holds in a very strong sense because we can exhibit a generic map $\pi : E \to U$ in ${{\mathcal{S}}}$, i.e. $\pi \in {{\mathcal{S}}}$ and all maps in ${{\mathcal{S}}}$ can be obtained as pullbacks of the generic map $\pi$. The codomain $U$ of $\pi$ is given by $$\Delta\left(\left\{ X \in {{\mathbf{Asm}}}({{\mathcal{A}}},{{\mathcal{A}}}_\#) \mid |X| \subseteq \kappa, {{\mathsf{card}}}(|X|) < \kappa\right\}\right)$$ and its domain $E$ has underlying set $$|E| = \left\{ (X,x) \mid X \in |U|, x \in |X| \right\}$$ and whose existence predicate is given by $E_E(X,x) = E_X(x)$. The map $\pi : E \to U$ is given by projection on the first component, i.e. $\pi(X,x) = X$. Obviously, the map $\pi$ has fibres of cardinality $< \kappa$ and we leave it as a straighforward exercise for the reader to show that every map in ${{\mathcal{S}}}$ can actually be obtained as pullback of $\pi$. It is easy to check that the class ${{\mathcal{S}}}$ in ${{\mathbf{Asm}}}({{\mathcal{A}}},{{\mathcal{A}}}_\#)$ is closed under dependent products, i.e. $\Pi_f g \in {{\mathcal{S}}}$ whenever $f$ and $g$ are in ${{\mathcal{S}}}$. As a consequence for $a : A \to I$ and $b : B \to I$ in ${{\mathcal{S}}}$ their exponential in the fibre over $I$, i.e. $a \to_I b = \Pi_a a^* b$, is in ${{\mathcal{S}}}$, too. Moreover, the generic mono $\top : {{\mathit{Tr}}}{\rightarrowtail}{{\mathit{Prop}}}$ constructed in Theorem \[AsmplcH\] like all monos is also an element of ${{\mathcal{S}}}$. Moreover, the terminal projection ${{\mathit{Prop}}}\to 1$ is in ${{\mathcal{S}}}$, too, since the underlying set of ${{\mathit{Prop}}}$ has cardinality $< \kappa$. Accordingly, the object ${{\mathit{Tr}}}$ is small, too. For every object $X$ in ${{\mathbf{Asm}}}({{\mathcal{A}}},{{\mathcal{A}}}_\#)$ we may construct a *weak power object* $\ni^w_X {\rightarrowtail}{{\mathit{Prop}}}^X {\times} X$ as follows \^w\_X & &\ & & \_\ \^X X & \_& where ${\mathsf{ev}}: {{\mathit{Prop}}}^X \times X \to {{\mathit{Prop}}}$ is the evaluation map. If $X$ is small, i.e. $X \to 1$ is in ${{\mathcal{S}}}$, i.e. ${{\mathsf{card}}}(X) < \kappa$, then ${{\mathit{Prop}}}^X$ is small, too, since ${{\mathsf{card}}}\left({{\mathit{Prop}}}^X\right) \leq {{\mathsf{card}}}({{\mathit{Prop}}})^{{{\mathsf{card}}}(X)} < \kappa$ because $\kappa$ is inaccessible and ${{\mathsf{card}}}({{\mathit{Prop}}}) , {{\mathsf{card}}}(X) < \kappa$. Notice that this construction of weak power objects also works in all slices. For future reference we summarize these considerations in the following \[weakpowerAsm\] The category ${{\mathbf{Asm}}}({{\mathcal{A}}},{{\mathcal{A}}}_\#)$ has weak power objects and ${{\mathcal{S}}}$ is closed under weak power objects. Small maps in ${{\mathbf{RT}}}({{\mathcal{A}}},{{\mathcal{A}}}_\#)$ ------------------------------------------------------------------- It is well known from [@VOO08] (section 2.4) that ${{\mathbf{RT}}}({{\mathcal{A}}},{{\mathcal{A}}}_\#)$ is the exact/regular completion of ${{\mathbf{Asm}}}({{\mathcal{A}}},{{\mathcal{A}}}_\#)$. Let $\bar{{{\mathcal{S}}}}$ be the class of maps defined in Theorem \[bennoextended\]. Now we can show easily that \[smallnesttop\] $\bar{{{\mathcal{S}}}}$ is a class of small maps in ${{\mathbf{RT}}}({{\mathcal{A}}},{{\mathcal{A}}}_\#)$ which is also closed under power objects and thus also under exponentiation. It is an immediate consequence of Theorem \[bennoextended\] and Theorem \[Sclos1\] that $\bar{{{\mathcal{S}}}}$ is a class of small maps in ${{\mathbf{RT}}}({{\mathcal{A}}},{{\mathcal{A}}}_\#)$. From Lemma 27 of [@VDBM11] and our Theorem \[weakpowerAsm\] it follows that $\bar{{{\mathcal{S}}}}$ is also closed under power objects. It is well known that closure under powerobjects and subobjects entails closure under exponentiation. As pointed out by J. van Oosten in private communication there is a logical functor $F : {{\mathbf{RT}}}({{\mathcal{A}}},{{\mathcal{A}}}_\sharp) \to {{\mathbf{RT}}}({{\mathcal{A}}})$ which just “forgets the actual realizers”. Already in [@JM95] there has been identified for every strongly inaccessible cardinal a class of small maps in ${{\mathbf{RT}}}({{\mathcal{A}}})$ from which our class of small maps in ${{\mathbf{RT}}}({{\mathcal{A}}},{{\mathcal{A}}}_\#)$ can be obtained as inverse image under $F$. A Model of ${\mathsf{IZF}}$ in ${{\mathbf{RT}}}({{\mathcal{A}}},{{\mathcal{A}}}_\#)$ ------------------------------------------------------------------------------------ It follows from the previous Theorem \[smallnesttop\] and Theorem 5.6 of [@JM95] that the class $\bar{{{\mathcal{S}}}}$ of small maps in ${{\mathbf{RT}}}({{\mathcal{A}}},{{\mathcal{A}}}_\#)$ gives rise to an “initial $\mathsf{ZF}$-algebra” within ${{\mathbf{RT}}}({{\mathcal{A}}},{{\mathcal{A}}}_\#)$. Accordingly, the nested realizability topos ${{\mathbf{RT}}}({{\mathcal{A}}},{{\mathcal{A}}}_\#)$ hosts a model of ${\mathsf{IZF}}$. It is an open question (raised by J. van Oosten) whether the above mentioned logical functor $F : {{\mathbf{RT}}}({{\mathcal{A}}},{{\mathcal{A}}}_\#) \to {{\mathbf{RT}}}({{\mathcal{A}}})$ preserves the initial $\mathsf{ZF}$-algebras arising from the respective classes of small maps. Small Maps for Subtoposes of ${{\mathbf{RT}}}({{\mathcal{A}}},{{\mathcal{A}}}_\#)$ ================================================================================== In the previous section we have endowed the nested realizability topos ${{\mathbf{RT}}}({{\mathcal{A}}},{{\mathcal{A}}}_\#)$ with a class $\bar{{{\mathcal{S}}}}$ of small maps in such a way that it gives rise to a model of ${\mathsf{IZF}}$ in the sense of Algebraic Set Theory as described in [@JM95]. In this section we show how to extend this result to subtoposes of ${{\mathbf{RT}}}({{\mathcal{A}}},{{\mathcal{A}}}_\#)$. Transferring Classes of Small Maps to Subtoposes ------------------------------------------------ Let ${{\mathcal{E}}}$ be an elementary topos and ${{\mathcal{S}}}$ a class of small maps in ${{\mathcal{E}}}$. Let ${{\mathsf{a}}}\dashv {{\mathsf{i}}}: {{\mathcal{F}}}{\hookrightarrow}{{\mathcal{E}}}$ be a subtopos of ${{\mathcal{E}}}$. W.l.o.g. we assume that ${{\mathcal{F}}}$ is closed under isomorphisms in ${{\mathcal{E}}}$ and that ${{\mathsf{a}}}f = f$ for $f \in {{\mathcal{F}}}$. We want to endow ${{\mathcal{F}}}$ with a class ${{\mathcal{S}}}_{{\mathcal{F}}}$ of small maps such that ${{\mathsf{a}}}: {{\mathcal{E}}}\to {{\mathcal{F}}}$ sends ${{\mathcal{S}}}$ to ${{\mathcal{S}}}_{{\mathcal{F}}}$. Thus, it is tempting to define ${{\mathcal{S}}}_{{\mathcal{F}}}$ as ${{\mathsf{a}}}{{\mathcal{F}}}$ by which we denote the closure under isomorphism in ${{\mathcal{F}}}$ of the image of ${{\mathsf{a}}}$. But then there are problems with condition [**(A5)**]{} because epimorphisms in ${{\mathcal{F}}}$ need not be epic in ${{\mathcal{E}}}$. In order to overcome this problem we define ${{\mathcal{S}}}_{{\mathcal{F}}}$ as follows \[SFdf\] Let ${{\mathcal{S}}}_{{\mathcal{F}}}$ be the class of all maps $f : B \to A$ in ${{\mathcal{F}}}$ fitting into a quasipullback Y & & B\ \^[g]{} & & \_f\ X & \_e & A in ${{\mathcal{F}}}$ for some $g : Y \to X$ in ${{\mathcal{S}}}$, i.e. $e^*f$ is a quotient of some ${{\mathsf{a}}}g$ in ${{\mathcal{F}}}/{{\mathsf{a}}}X$. The following little observation will be used later on. \[Slemepi\] The epis in ${{\mathcal{F}}}$ are precisely the sheafifications of epis in ${{\mathcal{E}}}$. First recall that epis in toposes are regular. Thus, since ${{\mathsf{a}}}$ is a left adjoint it preserves regular epis. For the converse direction suppose $e$ is an epi in ${{\mathcal{F}}}$. Consider its factorization $e = m \circ p$ in ${{\mathcal{E}}}$ where $m$ is monic and $p$ is an epi in ${{\mathcal{E}}}$. Then $e = {{\mathsf{a}}}(m \circ p) = {{\mathsf{a}}}m \circ {{\mathsf{a}}}p$ in ${{\mathcal{F}}}$. Since ${{\mathsf{a}}}$ preserves monos and epis and $e$ is epic in ${{\mathcal{F}}}$ it follows that ${{\mathsf{a}}}m$ is an iso. Now we are ready to prove the main theorem of this subsection. \[smallsub\] Suppose ${{\mathcal{E}}}$ is a topos with a natural numbers object $N$ and ${{\mathcal{S}}}$ is a class of small maps in ${{\mathcal{E}}}$ closed under power objects. If ${{\mathsf{a}}}\dashv {{\mathsf{i}}}: {{\mathcal{F}}}{\hookrightarrow}{{\mathcal{E}}}$ is a subtopos then ${{\mathcal{S}}}_{{\mathcal{F}}}$ as specified in Def. \[SFdf\] is a class of small maps in ${{\mathcal{F}}}$ which is closed under power objects. We will often (implicitly) use the fact that pullbacks in ${{\mathcal{F}}}$ preserve epis and maps in ${{\mathsf{a}}}{{\mathcal{S}}}$. This ensures for example that quasipullbacks of the form as considered in Def. \[SFdf\] are preserved by pullbacks along morphisms in ${{\mathcal{F}}}$. Accordingly, it follows that ${{\mathcal{S}}}_{{\mathcal{F}}}$ is closed under pullbacks in ${{\mathcal{F}}}$, i.e. validates condition [**(A0)**]{}. For showing that ${{\mathcal{S}}}_{{\mathcal{F}}}$ validates ${\bf (A1)}$ suppose that B && D\ \^f & & \_g\ A &\_[p]{} & C\ is a pullback in ${{\mathcal{F}}}$ where $f$ is in ${{\mathcal{S}}}_{{\mathcal{F}}}$ and $p$ is a cover in ${{\mathcal{F}}}$. Since $f$ is in ${{\mathcal{S}}}_{{\mathcal{F}}}$ it fits into a quasipullback Y & & B\ \^[h]{} & & \_f\ X & \_e & A where $h$ is in ${{\mathcal{S}}}$ and $e$ is a cover in ${{\mathcal{F}}}$. Since quasipullbacks are closed under composition it follows that Y & & B & & D\ \^[h]{} & & \_f & & \_g\ X & \_e & A & \_[p]{} & C is a quasipullback. Thus, since $p \circ e$ is epic, it follows that $g$ is in ${{\mathcal{S}}}_{{\mathcal{F}}}$ as desired. That ${{\mathcal{S}}}_{{\mathcal{F}}}$ validates condition [**(A2)**]{} is immediate from the facts that condition [**(A2)**]{} holds for ${{\mathcal{S}}}$, that ${{\mathsf{a}}}$ preserves $+$ and that $+$ preserves quasipullbacks. That ${{\mathcal{S}}}_{{\mathcal{F}}}$ validates condition [**(A3)**]{} is immediate from the fact that that ${{\mathsf{a}}}$ preserves colimits and finite limits. That ${{\mathcal{S}}}_{{\mathcal{F}}}$ validates [**(A4)**]{}, i.e. that ${{\mathcal{S}}}_{{\mathcal{F}}}$ is closed under composition, can be shown by adapting the proof of the analogous Lemma 2.15 of [@VDBM08]. Obviously, ${{\mathcal{S}}}_{{\mathcal{F}}}$ validates condition [**(A5)**]{} by its very definition since quasipullbacks are closed under horizontal composition. The proof that ${{\mathcal{S}}}_{{\mathcal{F}}}$ validates condition [**(A6)**]{} is analogous to the proof of case [**(A7)**]{} of Proposition 2.14 of [@VDBM08]. It is easy to check that [**(A7)**]{} holds for ${{\mathcal{S}}}_{{\mathcal{F}}}$. Let $\pi$ be a universal family for ${{\mathcal{S}}}$ then its sheafification ${{\mathsf{a}}}\pi$ is universal for ${{\mathcal{S}}}_{{\mathcal{F}}}$ which can be seen by applying ${{\mathsf{a}}}$ to the respective diagram in the formulation of [**(A7)**]{} and using the fact that quasipullbacks are closed under horizontal composition. Condition [**(A8)**]{} holds for ${{\mathcal{S}}}_{{\mathcal{F}}}$ since sheafification preserves natural numbers objects. Condition [**(A9)**]{} holds for ${{\mathcal{S}}}_{{\mathcal{F}}}$ since if $m$ is a mono in ${{\mathcal{F}}}$ then it is also a mono in ${{\mathcal{E}}}$ and thus by [**(A9)**]{} for ${{\mathcal{S}}}$ we have $m \cong {{\mathsf{a}}}m$ is in ${{\mathcal{S}}}_{{\mathcal{F}}}$. For showing that ${{\mathcal{S}}}_{{\mathcal{F}}}$ is closed under power objects one may adapt the proof of Proposition 6.6 from [@VDBM08] proving an analogous result. Small Maps in Subtoposes of ${{\mathbf{RT}}}({{\mathcal{A}}},{{\mathcal{A}}}_\#)$ --------------------------------------------------------------------------------- As a consequence of Theorem \[smallsub\] we obtain the following result. \[subnestedthm\] Let ${{\mathcal{S}}}$ be the class of small maps in ${{\mathbf{Asm}}}({{\mathcal{A}}},{{\mathcal{A}}}_\#)$ as introduced in Theorem \[Sclos1\] and $\bar{{{\mathcal{S}}}}$ be the class of small maps in ${{\mathbf{RT}}}({{\mathcal{A}}},{{\mathcal{A}}}_\#)$ as introduced in Theorem \[bennoextended\]. Suppose ${{\mathsf{a}}}\dashv {{\mathsf{i}}}: {{\mathcal{E}}}{\hookrightarrow}{{\mathbf{RT}}}({{\mathcal{A}}},{{\mathcal{A}}}_\#)$ is a subtopos of ${{\mathbf{RT}}}({{\mathcal{A}}},{{\mathcal{A}}}_\#)$ induced by a closure operator $j$ on ${{\mathscr{P}}}({{\mathcal{A}}},{{\mathcal{A}}}_\#)$. Then $\bar{{{\mathcal{S}}}}_{{\mathcal{E}}}$ as introduced in Theorem \[smallsub\] is a class of small maps in ${{\mathcal{E}}}$ closed under power objects and exponentiation. From Theorem \[smallnesttop\] we know that $\bar{{{\mathcal{S}}}}$ is a class of small maps closed under power objects. Thus, we can apply Theorem \[smallsub\] from which it follows that $\bar{{{\mathcal{S}}}}_{{\mathcal{E}}}$ is a class of small maps in ${{\mathcal{E}}}$ which is closed under power objects and, accordingly, also under exponentiation. This result applies in particular to the subtoposes of ${{\mathbf{RT}}}({{\mathcal{A}}},{{\mathcal{A}}}_\#)$ as considered in subsection \[subtopex\] and thus covers most of the examples considered in van Oosten’s book [@VOO08]. Models of ${\mathsf{IZF}}$ in Subtoposes of ${{\mathbf{RT}}}({{\mathcal{A}}},{{\mathcal{A}}}_\#)$ ------------------------------------------------------------------------------------------------- From the main result of [@JM95] and our Theorem \[subnestedthm\] it follows that most of the toposes considered in [@VOO08] host models of ${\mathsf{IZF}}$. \[MainThm\] There exist internal models for ${\mathsf{IZF}}$ in subtoposes of ${{\mathbf{RT}}}({{\mathcal{A}}},{{\mathcal{A}}}_\#)$ induced by local operators on ${{\mathscr{P}}}({{\mathcal{A}}},{{\mathcal{A}}}_\#)$. In case ${{\mathcal{A}}}= {{\mathcal{A}}}_\#$ due to [@KGVO05] we reobtain the realizability model for ${\mathsf{IZF}}$ as initially introduced by H. Friedman in [@FRI73], G. Rosolini in [@ROS82] and D. C. McCarty in [@MCC84]. In case ${{\mathcal{A}}}= {{\mathcal{A}}}_\# = {{\mathcal{K}}}_1$, the first Kleene algebra (corresponding to number realizability), from Theorem \[MainThm\] it follows that the modified realizability topos ${{\mathbf{Mod}}}= {{\mathbf{Mod}}}({{\mathcal{K}}}_1) = {{\mathbf{RT}}}_m({{\mathcal{K}}}_1,{{\mathcal{K}}}_1)$ from [@VOO97] hosts a model of ${\mathsf{IZF}}$. Thus, in ${\mathsf{IZF}}$ one cannot derive Markov’s Principle from Church’s Thesis. Conclusion ========== Relying on the main result of [@JM95] we have shown that relative realizability toposes and modified relative realizability toposes host models of ${\mathsf{IZF}}$. In the unnested case, i.e. ${{\mathcal{A}}}= {{\mathcal{A}}}_\#$ we reobtain the well known realizability models for ${\mathsf{IZF}}$ and a modified realizability model for ${\mathsf{IZF}}$ which to our knowledge cannot be found in the existing literature. Moreover, as pointed out to us by B. van den Berg our results also show that his recent Herbrand Realizability topos hosts a model of ${\mathsf{IZF}}$. We have obtained these new models for ${\mathsf{IZF}}$ in a quite uniform way using the methods of Algebraic Set Theory. Of course, one could define in each single case these models of ${\mathsf{IZF}}$ in a much more traditional and direct way. Using an appropriate adaptation of the results in [@KGVO05] one can presumably show that these “hand made” models are equivalent to the ones we have obtained in this paper by more abstract and general means. [11]{} B. van den Berg. Categorical semantics of constructive set theory. , TU Darmstadt, 2011. B. van den Berg, I. Moerdijk. Aspects of predicative Algebraic Set Theory I: Exact Completion. , 2008. B. van den Berg, I. Moerdijk. Aspects of predicative Algebraic Set Theory II: Realizability. , 2011. B. van den Berg, I. Moerdijk. A unified approach to algebraic set theory. , 2009. L. Birkedal, J. van Oosten. Relative and modified relative realizability. , 2002. H. Friedman. Some applications of Kleene’s methods for intuitionistic systems. , 1973. M. Hyland, P. T. Johnstone, A. M. Pitts. Tripos theory. , 1980. P. T. Johnstone. Sketches of an elephant: a topos theory compendium, vol.1. , 2002. P. T. Johnstone. Sketches of an Elephant: a Topos Theory Compendium, vol.2. , 2002. P. T. Johnstone. The Gleason cover of a realizability topos. , 2013. A. Joyal, I. Moerdijk. Algebraic Set Theory. , 1995. C. Kouwenhowen-Gentil, J. van Oosten. Algebraic set theory and the effective topos. , 2005. D. C. McCarty. Realizability and recursive mathematics. , 1984. J. R. Moschovakis Can there be no non-recursive functions? , 1971. J. van Oosten. The modified realizability topos. , 1997. J. van Oosten. Realizability: an introduction to its categorical side. , 2008. A. M. Pitts. The theory of triposes. , Univ. of Cambrdige, 1981. A. M. Pitts. Tripos theory in retrospect. , 1999. G. Rosolini. Un modello per la teoria intuizionista degli insiemi. , 1982. [^1]: In [@BVO02] they do not give a name to this topos and, moreover, write ${{\mathbf{RT}}}({{\mathcal{A}}},{{\mathcal{A}}}_\#)$ for the relative realizability subtopos of the nested realizability topos. [^2]: In [@VOO08] the constant objects functor is denoted by $\nabla$ because in case of realizability triposes it is right adjoint to the global elements functor $\Gamma$. However, in case of triposes induced by a complete Heyting algebra the constant objects functor is left adjoint to $\Gamma$. However, there are also triposes where the constant objects functor is neither left nor right adjoint to $\Gamma$. We prefer the notation $\Delta$ since ${{\mathit{eq}}}_S = \exists_{\delta_S}(\top_S)$ is the (Lawvere) equality predicate on the set $S$ in the sense of the tripos ${{\mathscr{P}}}$. [^3]: “generic” means that all monos can be obtained as pullbacks of $\top : {{\mathit{Tr}}}{\rightarrowtail}{{\mathit{Prop}}}$ but we may have $f^*\top \cong g^*\top$ for different $f$ and $g$ [^4]: A square is a quasipullback if the mediating arrow to the pullback square is a cover.

ArXiv

--- abstract: 'Pulsed $\gamma$-ray emission from millisecond pulsars (MSPs) has been detected by the sensitive [*Fermi*]{}, which sheds light on studies of the emission region and mechanism. In particular, the specific patterns of radio and $\gamma$-ray emission from PSR J0101-6422 challenge the popular pulsar models, e.g. outer gap and two-pole caustic (TPC) models. Using the three dimension (3D) annular gap model, we have jointly simulated radio and $\gamma$-ray light curves for three representative MSPs (PSR J0034-0534, PSR J0101-6422 and PSR J0437-4715) with distinct radio phase lags and present the best simulated results for these MSPs, particularly for PSR J0101-6422 with complex radio and $\gamma$-ray pulse profiles and for PSR J0437-4715 with a radio interpulse. It is found that both the $\gamma$-ray and radio emission originate from the annular gap region located in only one magnetic pole, and the radio emission region is not primarily lower than the $\gamma$-ray region in most cases. In addition, the annular gap model with a small magnetic inclination angle instead of “orthogonal rotator” can account for MSPs’ radio interpulse with a large phase separation from the main pulse. The annular gap model is a self-consistent model not only for young pulsars but also MSPs, and multi-wavelength light curves can be fundamentally explained by this model.' author: - 'Y.  J.  Du,  G.  J.  Qiao  and  D.  Chen' title: 'Radio and Gamma-ray Pulsed Emission from Millisecond Pulsars' --- Introduction ============ MSPs are a population of old neutron stars with short spin period $P_0 \lesssim 20$ ms and small characteristic magnetic field $B_0 < 10^9$ G (actually small period derivative $\dot{P} \lesssim 10^{-17}\,{\rm s~s^{-1}}$). They are believed to be formed from the recycling (accretion spin-up ) process in binaries [@msp]. Furthermore, there is another possible formation channel for MSPs (even for sub-millisecond pulsars), that is accretion-induced collapse of white dwarfs in a binary [@nomoto91; @smsp]. Because of the high stability of their rotations, MSPs have great potential for application: autonomous deep-space navigation, pulsar-based time scale, low-frequency gravitational wave detection and so on. When the first MSP, PSR B1937+21, had been discovered in 1982 [@backer82], [@usov83] soon predicted that this pulsar could emit $\gamma$-rays via synchrotron radiation on the order of 100 GeV. [@bhat91] subsequently calculated the $\gamma$-ray luminosity of MSPs, estimated their contribution to the diffuse $\gamma$-ray background of the Milky Way, and finally discussed detectability of MSPs as point $\gamma$-ray sources. Using the data of [*Energetic Gamma Ray Experiment*]{}, [@kuiper00] showed circumstantial evidence for the likely detection of pulsed $\gamma$-ray emission from a MSP, PSR J0218+4232, which was regarded as a $\gamma$-ray pulsar candidate for a long time. MSPs’ $\gamma$-ray emission was not observationally confirmed until [*Fermi*]{} with a sensitive Large Area Telescope (LAT) launched in June of 2008. Using eight-month data of [*Fermi*]{} LAT, eight $\gamma$-ray MSPs have already been detected [@msp-sci]. This number has grown to more than 40 up to date [@guillemot], and still increases. From observations, MSPs are analogous to young pulsars, which have multi-wavelength pulsed emission from radio to $\gamma$-ray band. Do MSPs and young pulsars share a simple model that contains similar emission region and acceleration mechanism to self-consistently explain their multi-wavelength emission? More and more multi-wavelength data with high precision give us opportunities to obtain remarkable insights of the magnetospheric physics. The multi-wavelength study is a key method to discriminate the various pulsar non-thermal emission models for both MSPs and young pulsars. Initially aiming to explain the high-energy pulsed emission from young pulsars, four traditional magnetospheric gap models have been suggested to study pulsed high energy emission of pulsars: the polar cap model [@1994ApJ...429..325D], the outer gap model [@1986ApJ...300..500C; @1995ApJ...438..314R; @1997ApJ...487..370Z], TPC/slot gap model [@2003ApJ...598.1201D; @harding08], and the annular gap model [@qiao04; @qiao07; @AG10]. To distinguish these pulsar models, the most important issues are the inducements of acceleration electric field region and related emission mechanisms to emanate high-energy photons [@vela; @crab]. One of the key discrepancies of the mentioned emission models is one-pole or two-pole emission pattern which depends on two corresponding geometry parameters: magnetic inclination angle $\alpha$ and viewing angle $\zeta$. @bulik00 adopted the polar-cap model to calculate the $\gamma$-ray emission from MSPs. They pointed out that curvature radiation of primary particles contributed to the MeV-to-GeV band, while the synchrotron radiation arising from pairs dominated only below 1 MeV. [@harding05] developed the pair-starved polar cap model and obtain similar spectral conclusion for high-energy emission from MSPs as above. In this model the accelerating field is not screened and the entire open volume is available for particle acceleration and emission of gamma rays. [@zhang03] used the outer gap model with multi-pole magnetic field to model the X-ray and $\gamma$-ray spectra for four MSPs, and the predicted results basically agree with the observations [@harding05]. Along with radio observations supplying us with excellent timing solutions for [*Fermi*]{} MSPs, the derived $\gamma$-ray and radio light curves with high signal-to-noise allow us to do joint simulation which can justify the pulsar emission models. Recently, @venter09 simulated both radio and $\gamma$-ray light curves for MSPs in the pair-starved polar cap, TPC and outer gap models, and they found that most of their simulated light curves are well explained by the TPC and outer gap models. They especially simulated light curves for a minor group of MSPs with phase-aligned radio and $\gamma$-ray pulse profiles [@venter12]. In addition, [@johnson12] also used the geometric slot gap, outer gap model or pair-starved polar cap model to fit $\gamma$-ray and radio light curves for three MSPs. A MSP PSR J0101-6422 with complex radio and $\gamma$-ray light curves, challenge the popular TPC and outer gap models [@kerr12]. It is found that neither of the two models can faithfully reproduce the observed light curves and phase lags. For such a complex radio profile the simple beam model they used may have been insufficient. In this paper, we use a 3D annular gap model to study both radio and $\gamma$-ray light curves for three MSPs which stand for the relevant types of MSPs with different radio lags. In §2, the annular gap and core gap will be simply introduced and the acceleration electric field in the annular gap is calculated. In §3, we jointly simulate radio and $\gamma$-ray band light curves for PSR J0034-0534, J0101-6422 and J0437-4715. The radio phase lags are identified and the reasons for them explored. The conclusions and discussions are shown in §4. The Annular Gap and Core Gap for MSPs ===================================== The Definition of Annular Gap and Core Gap ------------------------------------------ In a pulsar magnetosphere [@GJ69; @RS75], the critical field lines[^1] divide a polar cap into two distinct parts: the annular gap region and the core gap region [See Figure 1 of @crab]. The annular gap is constrained between the critical and last open field lines, and the core gap is around the magnetic axis and within critical field lines [@AG10]. The size of the polar cap decreases with increasing spin period, but can be quite large for MSPs and young pulsars. The annular gap width is correspondingly larger for short spin-period pulsars, and it varies with the magnetic azimuth $\psi$. For an anti-parallel rotator and $\psi=0^\circ$, the radii of the annular polar region is $r_{\rm ann}=r_{\rm pc}-r_{\rm core} = 0.26 R(\Omega R/c)^{1/2}$ [@AG10], here $R$ is a pulsar’s radius, and $\Omega$ is its angular spin frequency. Combining advantages of the outer gap and TPC models, [@qiao04; @qiao07] originally suggested the annular gap model, which has been further developed by [@AG10; @vela; @crab]. The site for generation of high energy photons is mainly located in the vicinity of the null charge surface[^2]. Being consistent with the physically calculated spectra [@vela], the Gaussian emissivities are numerically assumed when simulating light curves. The key emission geometry parameters $\alpha$ and $\zeta$ are not convincingly confirmed so far, we adopt these two values from related literature if they exist. If there are not any reliable values, we just use some random values according to theories of MSPs’ magnetic field evolution. By hypothesizing reasonable emissivities and the magnetic inclination angle, this model can make wide fan-like emission beams, thus well reproduce the relevant light curves cut by a suitable viewing angle. Acceleration Electric Field --------------------------- Co-rotating charge-separated plasma is filled in a pulsar magnetosphere [@GJ69; @RS75]. When reaching some regions near the light cylinder, charged particles can not exceed the speed of light ($c$), thus escape from the magnetosphere. To compensate the escaping particles, the pulsar has to supply sufficient charged particles to its magnetosphere. This dynamic process continuously happens, thus a huge acceleration electric field is generated in the magnetosphere. This is the general mechanism for acceleration electric field ($E_{\parallel}$), which is suitable for both young pulsars [@vela; @crab] and MSPs. The charged particles with opposite signs are simultaneously exporting from the annular gap and core gap, and satisfies the condition of circuit closure in the whole magnetosphere. We assume the flowing-out particles’ charge density equals to the local Goldreich-Julian density [@GJ69] at a radial distance of $r \sim R_{\rm LC}$, here $R_{\rm LC}=\frac{cP}{2\pi}$ is a radius of light cylinder. The detailed calculation method and formulae are presented in [@vela]. The acceleration electric fields of typical MSPs are shown in Figure \[ELEC\] using four sets of surface magnetic field $B$ and spin period $P$. For MSPs with large or small values of $P$ and $B$, we found that, in all cases, the electric field in the inner region of the annular gap is sufficiently high ($E_\parallel \gtrsim 10^6\,{\rm V\, cm^{-1}}$). The charged particles accelerated in the annular gap or core gap are flowing out along a field line in a quasi-steady state. Using the derived acceleration electric field, we can obtain the Lorentz factor $\Gamma_{\rm p}$ of primary particles from the balance of acceleration and curvature radiation reaction $$\Gamma_{\rm p} = (\frac{3\rho^2 E_{\parallel}}{2e})^{\frac{1}{4}} = 2.36\times 10^7{\rho_7}^{0.5} E_{\parallel,\, 6}^{0.25}, \nonumber \label{gam_p}$$ where $e$ is the charge of an electron, $\rho_7$ the curvature radius in units of $10^7$cm and $E_{\parallel,\, 6}$ the acceleration electric field in units of $10^6\, \rm V\, cm^{-1}$. The primary particles are accelerated to ultra-relativistic energy with typical Lorentz factors of $\Gamma_{\rm p} \sim 10^6 - 10^7$, because of the huge acceleration electric field in the annular gap. Since a lot of $\gamma$-ray photons are generated by the primary particles via curvature radiation and inverse Compton scattering processes, abundant ${\rm e}^\pm$ pairs are subsequently created through two-photon annihilation and photon-magnetic-absorption ($\gamma$-B) processes. Simulating Radio and $\gamma$-ray Light Curves for MSPs ======================================================= Thanks to the [*Fermi*]{}-LAT, we now know that some MSPs are muti-wavelength emitters which have detectable radio and $\gamma$-ray pulsed emission. According to the observations of phase lag ($\Delta$) between radio peak and $\gamma$-ray peak [@catalog; @kerr12], MSPs can be divided into four classes. PSR J0034-0534 represents a class of MSPs which has phase-aligned light curves ($\Delta \sim 0$); PSR J0101-6422 stands for another class which has moderate radio phase lag $\Delta \sim 0.2 - 0.3$ with quite complex radio or $\gamma$-ray light curves; PSR J0437-4715 stands for a third class which has larger radio lag $\Delta \sim 0.43$ and PSR J1744-1134 is a fourth class of MSPs whose $\gamma$-ray peak precedes the radio peak [@catalog], we will model this MSP in future when high signal-to-noise $\gamma$-ray light curves are available. We process [*Fermi*]{} Pass 7 data to derive the observed light curves for three MSPs according to the radio timing solutions of MSPs from Fermi Science Support Center (FSSC)[^3]. We select events with energies of $> 0.1$GeV within 2$^\circ$ of each MSP’s position and with zenith angles smaller than 105$^\circ$. The key filter conditions for good time interval are rock angle $<52^\circ$ and angsep(RA$_{\rm MSP}$, DEC$_{\rm MSP}$, RA$_{\rm SUN}$, DEC$_{\rm SUN}$)$ < 5^\circ$, where RA and DEC are right ascension and declination respectively. Then we use tempo2 [@hobbs06] with [*Fermi*]{} plug-in to obtain the spin phase for each photon. Finally we obtain the high signal-to-noise $\gamma$-ray light curves for the three MSPs (see red lines in Figure 2, 3, 4). A convincing model should have simple clear emission geometric picture with reasonable input parameters, which can not only reproduce multi-wavelength light curves for young pulsars but also for MSPs. In this paper, we will jointly simulate radio and $\gamma$-ray light curves for PSR J0034-0534, PSR J0101-6422 and PSR J0437-4715. We briefly introduce the simulation method here. As shown in table 1, $\alpha$ and $\zeta$ are “first-rank” parameters, which are primarily adopted from the literature if they exist. When there are not any convincing values, we tend to use reasonable values of $\alpha$ from relevant theory on magnetic field evolution of pulsars [@ruderman91] and $\zeta$ is adopted randomly according to the simulated emission pattern (photon sky-map). When $\alpha$ and $\zeta$ are fixed, several other model parameters are carefully adjusted for the emission regions until the observed light curve of the corresponding band can be reproduced. The model parameters for three MSPs are listed in table 1. [lccccccccc]{} Band & $\kappa$ & $ \lambda$ & $\epsilon$ & $\sigma_{\rm A}$ & $\sigma_{\rm \theta,\,A}$ & $\sigma_{\rm C}$ & $\sigma_{\rm \theta,\,C1}$ & $\sigma_{\rm \theta,\,C2}$\ \ $>0.1$GeV & 0.75 & 0.85 & 0.8 & 0.32 & 0.007 & 0.15 & 0.006 & 0.0052\ Radio & – & – & 0.58 & – & 0.013 & – & 0.00065 & 0.00053\ \ $>0.1$GeV & 0.50 & 0.85 & 0.8 & 0.3 & 0.009 & 0.15 & 0.006 & 0.006\ Radio & – & – & 1.5 & – & 0.006 & – & 0.0075 & 0.0078\ \ $>0.1$GeV & 0.1 & 0.1 & 1.3 & 0.03 & 0.002 & 0.02 & 0.001 & 0.001\ Radio & – & – & 1.2 & – & 0.0004 & – & 0.0002 & 0.0002\ [ $\alpha$ and $\zeta$ are magnetic inclination angles and viewing angles; $\kappa$ and $\lambda$ are two geometry parameters to determine the peak altitude in the annular gap; $\epsilon$ is a parameter for the peak altitude in the core gap; $\sigma_{\rm A}$ and $\sigma_{\rm C}$ are length scales for the emission region on each open field line in the annular gap and the core gap in units of $R_{\rm LC}$, respectively; $\sigma_{\rm \theta,\,A}$ is the transverse bunch scale for field lines in the annular gap; $\sigma_{\rm \theta,\,C1}$ and $\sigma_{\rm \theta,\,C2}$ are the bunch scale for field lines of $-180^\circ<\psi_{\rm s}<90^\circ$ and $90^\circ<\psi_{\rm s}<180^\circ$ in the core gap, respectively. The detailed description of these symbols can be found in [@vela].]{} PSR J0034-0534 --------------  PSR J0034-0534 is the ninth $\gamma$-ray MSP detected by the [ *Fermi*]{}-LAT [@abdo10], and it has strong $\gamma$-ray and radio pulsed emission with phase-aligned light curves. To reveal the emission region of this MSP, we use the annular gap model to jointly model the radio and $\gamma$-ray light curves. The simulation method is the same as described in §3.1 of [@vela]. The key idea is to project all radiation intensities in both the annular gap and core gap to the “non-rotating” sky, considering some physical effects (e.g. aberration effect and retardation effect). Here we use numerical emissivities to speed up the calculations, since the assumed emissivities are consistent with the physically calculated spectra, as shown in Figure 8 of [@vela]. From Figure 2, we find that both radio and $\gamma$-ray emission are mainly generated in the annular gap region co-located at intermediate altitudes $r \sim 0.24 - 0.56 R_{\rm LC}$, which leads to the phase-aligned light curves. [@abdo10] used the TPC and outer gap geometric models with $\alpha=30^\circ$ and $\zeta=70^\circ$ to obtain the light curves for PSR J0034-0534, and they derived similar conclusions that radio emission region extends from $0.6 R_{\rm LC}$ to $0.8 R_{\rm LC}$ and $\gamma$-ray region extends from $0.12 R_{\rm LC}$ to $0.9 R_{\rm LC}$. It is found that this MSP has a larger transverse emission region for radio emission. Moreover, [@venter12] developed the traditional outer gap and TPC model, adopting the similar idea of numerically assumed emissivity of piecewise-function, and derived the phase-aligned radio and $\gamma$-ray light curves for three MSPs including PSR J0034-0534. It seems that PSR J0034-0534 has off-peak pulsed $\gamma$-ray emission up to 100% duty cycle [@ackermann11], which is not reproduced by the annular gap model. This might be due to the lack of knowledge on emission geometry and magnetic field configuration, and we will further develop our model to study this MSP in detail in the future.  PSR J0101-6422 -------------- The observed light curves of PSR J0101-6422 have complex features: the $\gamma$-ray profile is likely to have three peaks; while the radio profile contains three peaks that occupy nearly the whole rotation phase (see left panels of Figure 3). To well model both light curves, we consider both cases of single-pole and two-pole emission-picture with many attempts on a large number of parameter space. The light curves of this MSP favor a single-pole emission picture with $\alpha=30^\circ$ and $\zeta=48^\circ$, and the results are shown in Figure 3. For a viewing angle of $\zeta=48^\circ$, the radio interpeak originates from the core gap with high altitudes of $0.3 R_{\rm LC}$ to $0.7 R_{\rm LC}$; while the other two peaks with a bridge (at the phases of $\sim -0.3$ and $\sim 0.45$) originate from the annular gap with altitudes of $0.45 R_{\rm LC}$ to $0.78 R_{\rm LC}$. The $\gamma$-ray profile is similar, the interpeak (at phase of $\sim 0$) mainly originates from the core gap with high altitudes of $0.32 R_{\rm LC}$ to $0.6 R_{\rm LC}$, while the other two peaks originate from the annular gap with altitudes of $0.08 R_{\rm LC}$ to $0.4 R_{\rm LC}$. According to the annular gap model, we note that the radio emission from PSR J0101-6422 is quite asymmetric in magnetic azimuth. This is possibly due to the special physical coherence condition and propagation effects in the pulsar magnetosphere. We add some discussions on radio emission in §4. PSR J0437-4715 -------------- PSR J0437-4715 is a very close MSP with a distance of 0.16kpc to the Earth [@manchester05], and has multi-wavelength emission. [@chen98] suggested that PSR J0437-4715 was an aligned rotator. We therefore adopted a relatively small magnetic inclination angle as $\alpha=25^\circ$ and a reasonable viewing angle as $\zeta=42^\circ$ from the high-precision radio timing observations [@van01; @hotan06]. [@bogdanov07] also used the value of $\zeta=42^\circ$ to successfully model the thermal X-ray pulsations of this MSP. We apply our annular gap model to jointly simulate its radio and $\gamma$-ray light curves, and the results are shown in Figure 4. We emphasize that the radio interpulse can be reproduced by our model, although it does not precisely match the observations. The radio emission including the main peak and interpulse originate from a much higher and narrower region in the annular gap region with high altitudes of $0.48 R_{\rm LC}$ to $0.57 R_{\rm LC}$; while the $\gamma$-ray emission is generated in the annular gap region with lower altitudes of $0.064 R_{\rm LC}$ to $0.15 R_{\rm LC}$ located in the same magnetic pole. This leads to a large radio lag of $\Delta \sim 0.43$. Conclusions and Discussions =========================== Pulsed $\gamma$-ray emission from MSPs has been observed by the sensitive [*Fermi*]{}-LAT. Particularly, the specific pattern of radio and $\gamma$-ray emission from the PSR J0101-6422 challenges the outer gap and TPC models. A convincing model should apply not only to young pulsars but also to MSPs. In this paper, we used the annular gap model to jointly model the radio and $\gamma$-ray light curves for three representative MSPs PSR J0034-0534, PSR J0101-6422 and PSR J0437-4715 with distinct radio phase lags. For PSR J0034-0534 with phase-aligned radio and $\gamma$-ray light curves, both bands are mainly generated in the annular gap region co-located at intermediate altitudes. For PSR J0101-6422 with complex radio and $\gamma$-ray pulse profiles, we presented the best simulated results for this type of MSPs. The radio interpulse originate from the core gap at higher altitudes; while the other two radio peaks with a bridge originate from the annular gap region. The interpeak originates from the core gap region and the other two peak from the annular gap region. For PSR J0437-4715 with a large radio lag, the radio emission (including the interpulse) originates from a much higher and narrower region in the annular gap region, and the $\gamma$-ray emission has lower altitudes. From simulations of these MSPs, the annular gap model favors a single-pole emission pattern with small inclination angles ($\alpha \lesssim 35^{\rm \circ}$) for MSPs. This result is compatible with theories of magnetic field evolution of MSPs in binaries: some recycled pulsars tend to have aligning magnetic filed moment, i.e. small magnetic inclination angle $\alpha$ [@ruderman91; @chen98]. [@lamb09] presented a concrete discussion on the $\alpha$ evolution of MSPs while they were recycling in low mass X-ray binaries, and they note that the strong interactions between spinning superfluid neutrons and magnetized superconducting protons in a pulsar’s core force the spin axis to change. A MSP would be an aligned rotator ($\alpha \sim 0^\circ$) if the star’s north and south magnetic poles are forced toward opposite spin poles by the accretion disk, or would be an orthogonal rotator ($\alpha \sim 90^\circ$) if both of the star’s magnetic poles are forced toward the same spin pole. Moreover, it is certainly unclear what happens when a MSP’s recycling process finishes, [@young10] analyzed the new pulse width data of normal pulsars, and found that the spin and magnetic axes would align when they spin down due to dipole radiation and particle outflowing. Several MSPs are simply assumed to be nearly orthogonal rotators because they have a radio interpulse separated by a large phase of $\gtrsim 180^\circ$ from its main pulse [@chen98]. [@guillemot2] studied multi-wavelength light curves for two MSPs (PSR B1937+21 and PSR B1957+20), and found that fitting the radio polarization data of PSR B1937+21 favors an orthogonal rotator. As with most RVM fits the confidence area is large but the orthogonal configuration is further supported by the altitude-limited TPC and outer gap models. However, this is not universally true, at least in the case of the annular gap model. The radio light curve of MSPs (e.g. PSR J0437-4715) with an interpulse can be well explained by the annular gap model with a small magnetic inclination angle. By simulating light curves for MSPs in the annular gap model, we found that the radio emission mainly originates from the high-altitude narrow region in the annular gap region. The radio emission pattern (photon sky-map) is patch-like in our model. The radio emissivities on each field line (in the annular gap or core gap) vary slightly (nearly uniform), but the case for $\gamma$-ray light-curve simulation is quite different, they vary a lot. High energy ($\gamma$-ray and X-ray) emission is generated by non-coherent radiation from relativistic primary particles and pairs, while radio emission is suggested to be generated by coherent radiation due to two-stream instability of outward and inward pairs [@RS75]. Here we focus on studying radio and $\gamma$-ray light curves for MSPs, the concrete emission mechanism including polarization, spectral properties and long-term stabilities of radio lags are however needed to further study, considering the coherent condition and propagation effects. [@han98] systematically studied the radio circular polarization for pulsar integrated pulse profiles, and found that sense reversals of circular polarization are observed across the conal emission in some cases, unrestricted to core components. The polarization property of high-altitude radio emission from both annular and core gaps is a valuable subject to be investigated in future. We will keep on improving our model to present better simulated light curves, especially for the phases of leading wing of peak 1, trailing wing of peak 2 and off-peak pulses. In this paper, we simulated radio and gamma-ray light curves for 3 MSPs. When high signal-to-noise data at other wavelengths is available in the future, we will re-simulate light curves and fit the multi-wavelength phase-resolved spectra. In sum, the annular gap model is a self-consistent model not only for young pulsars [@vela; @crab], but also for MSPs, and multi-wavelength light curves can be well explained by this model. The authors are very grateful to the referee for the insightful and constructive comments. We would like to appreciate Matthew Kerr very much for giving us the ephemeris and radio profile of PSR J0101-6422. YJD is supported by China Postdoctoral Science Foundation (Grant No.: 2012M510047), and partially supported by our institute project of “Five Key Foster Directions” (Grant No.: Y22116EA2S). GJQ is supported by the National Basic Research Program of China (2012CB821800) and National Natural Science Foundation of China (10833003). DC is supported by National Natural Science Foundation of China (10803006, 11010250) and Advance Research Projects of Space Science (Grant No.: XDA04070000). [*Facilities: Fermi*]{} Abdo, A. A., Ackermann, M., Ajello, M., et al. 2009, Science, 325, 848 Abdo, A. A., et al. 2010a, , 187, 460 Abdo, A. A., Ackermann, M., Ajello, M., et al. 2010b, , 712, 957 Ackermann, M., Ajello, M., Baldini, L., et al. 2011, , 726, 35 Backer, D. C., Kulkarni, S. R., Heiles, C., Davis, M. M., & Goss, W. M. 1982, , 300, 615 Bhattacharya, D., & van den Heuvel, E. P. J. 1991, , 203, 1 Bhattacharya, D., & Srinivasan, G. 1991, Journal of Astrophysics and Astronomy, 12, 17 Bogdanov, S., Rybicki, G. B., & Grindlay, J. E. 2007, , 670, 668 Bulik, T., Rudak, B., & Dyks, J. 2000, , 317, 97 Cheng, K. S., Ho, C., & Ruderman, M. 1986, , 300, 500 Chen, K., Ruderman, M., & Zhu, T. 1998, , 493, 397 Daugherty, J. K., & Harding, A. K. 1994, , 429, 325 Du, Y. J., Xu, R. X., Qiao, G. J., & Han, J. L. 2009, , 399, 1587 Du, Y. J., Qiao, G. J., Han, J. L., Lee, K. J., Xu, R. X. 2010, , 406, 2671 Du, Y. J., Han, J. L., Qiao, G. J., & Chou, C. K. 2011, , 731, 2 Du, Y. J., Qiao, G. J., & Wang, W. 2011, , 748, 84 Dyks, J., & Rudak, B. 2003, , 598, 1201 Goldreich, P., & Julian, W. H. 1969, , 157, 869 Guillemot, L., for the Fermi LAT Collaboration 2012a, arXiv:1210.5341 Guillemot, L., Johnson, T. J., Venter, C., et al. 2012b, , 744, 33 Han, J. L., Manchester, R. N., Xu, R. X., & Qiao, G. J. 1998, , 300, 373 Hao, L.-F., Wang, M., & Yang, J. 2010, Research in Astronomy and Astrophysics, 10, 805 Harding, A. K., Usov, V. V., & Muslimov, A. G. 2005, , 622, 531 Harding, A. K., Stern, J. V., Dyks, J., & Frackowiak, M. 2008, , 680, 1378 Hobbs, G. B., Edwards, R. T., & Manchester, R. N. 2006, , 369, 655 Hotan, A. W., Bailes, M., & Ord, S. M. 2006, , 369, 1502 Johnson, T. J., Harding, A. K., Venter, C., et al. 2012, arXiv:1210.1504 Kerr, M., Camilo, F., Johnson, T. J., et al. 2012, , 748, L2 Kuiper, L., Hermsen, W., Verbunt, F., et al. 2000, , 359, 615 Lamb, F. K., Boutloukos, S., Van Wassenhove, S., et al. 2009, , 706, 417 Manchester, R. N., Hobbs, G. B., Teoh, A., & Hobbs, M. 2005, , 129, 1993 Nomoto, K., & Kondo, Y. 1991, ApJ, 367, L19 Qiao, G. J., Lee, K. J., Wang, H. G., Xu, R. X., & Han, J. L. 2004, ApJ, 606, L49 Qiao, G. J., Lee, K. J., Zhang, B., Wang, H. G., & Xu, R. X. 2007, Chin. J. Astron. Astrophys., 7, 496 Romani, R. W., & Yadigaroglu, I.-A. 1995, , 438, 314 Ruderman, M. A., & Sutherland, P. G. 1975, ApJ, 196, 51 Ruderman, M. 1991, , 366, 261 Usov, V. V. 1983, , 305, 409 van Straten, W., Bailes, M., Britton, M., et al. 2001, , 412, 158 Venter, C., Harding, A. K., & Guillemot, L. 2009, , 707, 800 Venter, C., Johnson, T. J., & Harding, A. K. 2012, , 744, 34 Young, M. D. T., Chan, L. S., Burman, R. R., & Blair, D. G. 2010, MNRAS, 402, 1317 Zhang, L., & Cheng, K. S. 1997, , 487, 370 Zhang, L., & Cheng, K. S. 2003, , 398, 639 [^1]: They are defined as a set of special field lines that satisfy the condition of $\mathbf{\Omega \cdot B}=0$ at the light cylinder. [^2]: It is defined as a surface where the Goldreich-Julian (GJ) charge density $\rho_{\rm GJ}$ [@GJ69] is zero. [^3]: http://fermi.gsfc.nasa.gov/ssc/data/access/lat/ephems/

ArXiv

--- abstract: | The fastest algorithms for edge coloring run in time $2^m n^{O(1)}$, where $m$ and $n$ are the number of edges and vertices of the input graph, respectively. For dense graphs, this bound becomes $2^{\Theta(n^2)}$. This is a somewhat unique situation, since most of the studied graph problems admit algorithms running in time $2^{O(n\log n)}$. It is a notorious open problem to either show an algorithm for edge coloring running in time $2^{o(n^2)}$ or to refute it, assuming Exponential Time Hypothesis (ETH) or other well established assumption. We notice that the same question can be asked for list edge coloring, a well-studied generalization of edge coloring where every edge comes with a set (often called a [*list*]{}) of allowed colors. Our main result states that list edge coloring for simple graphs does not admit an algorithm running in time $2^{o(n^2)}$, unless ETH fails. Interestingly, the algorithm for edge coloring running in time $2^m n^{O(1)}$ generalizes to the list version without any asymptotic slow-down. Thus, our lower bound is essentially tight. This also means that in order to design an algorithm running in time $2^{o(n^2)}$ for edge coloring, one has to exploit its special features compared to the list version. author: - 'Łukasz Kowalik[^1]' - Arkadiusz Socała bibliography: - 'listedgecol.bib' title: 'Tight Lower Bounds for List Edge Coloring[^2]' --- [20]{}(0.7, 12.98) {width="40px"} [20]{}(0.46, 13.28) {width="60px"} Introduction {#sect:intro} ============ An edge coloring of a graph $G=(V,E)$ is a function $c:E\rightarrow\mathbb{N}$ which has different values (called colors) on incident edges. This is one of the most basic graph concepts with plethora of results, including classical theorems of Vizing, Shannon and Kőnig. In the decision problem [[[[Edge Coloring]{}]{}]{}]{}we are given a simple graph $G$ and an integer $k$. We ask if $G$ can be edge colored using only $k$ colors. This is an NP-complete problem, as shown by Holyer [@holyer], similarly as many other natural graph decision problems like [Clique]{}, [Vertex Coloring]{}, [Hamiltonicity]{} or [Subgraph Isomorphism]{}. However, there is an intriguing difference between our understanding of [[[[Edge Coloring]{}]{}]{}]{}and most of the studied graph problems, including the four ones mentioned above. Namely, the latter ones admit algorithms running in time $2^{O(n\log n)}$, and often even $2^{O(n)}$ for an $n$-vertex input graph, while it is not known whether [[[[Edge Coloring]{}]{}]{}]{}can be solved in time $2^{o(n^2)}$. Indeed, the fastest known algorithm for edge coloring is obtained by applying the vertex coloring algorithm of Björklund, Husfeldt and Koivisto [@BjorklundHK09] to the line graph of the input graph. As a result, we get an edge coloring algorithm which, for any graph with $m$ edges and $n$ vertices, runs in time $2^{m}n^{O(1)}$ and exponential space, which is $2^{\Theta(n^2)}$ for dense graphs. The only progress towards a tailor-made approach for edge coloring is the more recent algorithm of Björklund, Husfeldt, Kaski and Koivisto [@bhkk:narrow] which still runs in time $2^{m}n^{O(1)}$ but uses only polynomial space. In this context it is natural to ask for a lower bound. Clearly, any superpolynomial lower bound would imply P$\ne$NP. However, a more feasible goal is to prove a meaningful lower bound under the assumption of a well established conjecture, like Exponential Time Hypothesis (ETH, see Section \[sect:pre\] for a precise formulation). The reduction of Holyer, combined with standard tools (see Section \[sect:pre\]) proves that [[[[Edge Coloring]{}]{}]{}]{}does not admit an algorithm in time $2^{o(m)}$ or $2^{o(n)}$ . At the open problem session of Dagstuhl Seminar 08431 in 2008 [@dagstuhl08] it was asked to exclude $2^{O(n)}$ algorithms, assuming ETH. Despite considerable progress in ETH-based lower bounds in recent years [@CyganFGKMPS17; @CyganPP16; @marx-beat] this problem stays unsolved [@dagstuhl17]. List edge coloring is a generalization of edge coloring. An *edge list assignment* $L:E(G)\rightarrow 2^\mathbb{N}$ is a function that assigns to each edge $e$ of $G$ a set (often called a *list*) $L(e)$ of allowed colors. A function $c:E(G) \rightarrow \mathbb{N}$ is a *list edge coloring* of $(G,L)$ if $c(e) \in L(e)$ for every $e \in E(G)$, and $c(e) \neq c(f)$ for every pair of incident edges $e, f \in E(G)$. The notion of list edge coloring is also a frequent topic of research. For example, it is conjectured that if $G$ can be edge colored in $k$ colors for some $k$, then it can be list edge colored for any edge list assignment with all lists of size at least $k$. This conjecture has been proved in some classes of graphs like bipartite graphs [@galvin] or planar graphs of maximum degree at least 12 [@boro:edge-list]. In this work, we study the computational complexity of list edge coloring. The basic decision problem, [[[[List Edge Coloring in Simple Graphs]{}]{}]{}]{}, asks if for a given simple graph $G$ with edge list assignment $L$ there is a list edge coloring of $(G,L)$. Its more general variant, called [[[[List Edge Coloring in Multigraphs]{}]{}]{}]{}asks the same question but the input graph does not need to be simple, i.e., it can contain parallel edges. Although the problem seems much more general than [[[[Edge Coloring]{}]{}]{}]{}, the two best known algorithms [@BjorklundHK09; @bhkk:narrow] that decide if a given graph admits an edge coloring in $k$ colors solve [[[[List Edge Coloring in Multigraphs]{}]{}]{}]{}(and hence also [[[[List Edge Coloring in Simple Graphs]{}]{}]{}]{}) within the same time bound, i.e., $2^{m}m^{O(1)} + O(L)$, where $L$ is the total length of all lists, after only minor modifications (see Proposition 3 in [@BjorklundHK09]). Multigraphs do not admit any upper bound on the number of edges, hence this time complexity does not translate to a function on $n$. We show that this is not an accident, because satisfiability of any sufficiently sparse 3-CNF-SAT formula can be efficiently encoded as a list edge coloring instance with a bounded number of vertices. This gives the following result. [theorem]{}[thmulti]{} \[th:multi\]If there is a function $f:\mathbb{N}\rightarrow\mathbb{N}$ such that [[[[List Edge Coloring in Multigraphs]{}]{}]{}]{}can be solved in time $f(n)\cdot m^{O(1)}$ for any input graph on $n$ vertices and $m$ edges, then $P=NP$. For simple graphs $m=O(n^2)$ and hence [[[[List Edge Coloring in Simple Graphs]{}]{}]{}]{}admits an algorithm running in time $2^{O(n^2)}$. Our main result states that this bound is essentially optimal, assuming ETH. [theorem]{}[thsimple]{} \[th:simple\]If there is an algorithm for [[[[List Edge Coloring in Simple Graphs]{}]{}]{}]{}that runs in time $2^{o(n^2)}$, then Exponential Time Hypothesis fails. Our results have twofold consequences for the [[[[Edge Coloring]{}]{}]{}]{}problem. First, one may hope that our reductions can inspire a reduction for [[[[Edge Coloring]{}]{}]{}]{}. However, it is possible that such a reduction does not exist and researchers may still try to get an algorithm for [[[[Edge Coloring]{}]{}]{}]{}running in time $2^{o(n^2)}$. Then we offer a simple way of verifying if a new idea works: if it applies to the list version as well, there is no hope for it. Preliminaries {#sect:pre} ============= For an integer $k$, we denote $[k]=\{0,\ldots,k-1\}$. If $I$ and $J$ are instances of decision problems $P$ and $R$, respectively, then we say that $I$ and $J$ are [*equivalent*]{} if either both $I$ and $J$ are YES-instances of respective problems, or both are NO-instances. A clause in a CNF-formula is represented by the set of its literals. For two subsets of vertices $A$, $B$ of a graph $G=(V,E)$ by $E(A,B)$ we denote the set of edges with one endpoint in $A$ and the other in $B$. #### Exponential-Time Hypothesis. {#exponential-time-hypothesis. .unnumbered} The Exponential Time Hypothesis (ETH) of Impagliazzo et al. [@eth] states that there exists a constant $c > 0$, such that there is no algorithm solving [[[[$3$-SAT]{}]{}]{}]{}in time $O(2^{cn})$. During the recent years, ETH became the central conjecture used for proving tight bounds on the complexity of various problems. One of the most important results connected to ETH is the [*[Sparsification Lemma]{}*]{} [@seth], which essentially gives a (many-one) reduction from an arbitrary instance of to an instance where the number of clauses is linear in the number of variables. The following well-known corollary can be derived by combining ETH with the Sparsification Lemma. \[thm:eth-main\] Unless ETH fails, there is no algorithm for [[[[$3$-SAT]{}]{}]{}]{}that runs in time $2^{o(n+m)}$, where $n,m$ denote the numbers of variables and clauses, respectively. We need the following regularization result of Tovey [@Tovey84]. Following Tovey, by [[[[$(3,\!4)$-SAT]{}]{}]{}]{}we call the variant of [[[[$3$-SAT]{}]{}]{}]{}where each clause of the input formula contains exactly $3$ different variables, and each variable occurs in at most $4$ clauses. \[lem:tovey\] Given a [[[[$3$-SAT]{}]{}]{}]{}formula $\varphi$ with $n$ variables and $m$ clauses one can transform it in polynomial time into an equivalent [[[[$(3,\!4)$-SAT]{}]{}]{}]{}instance $\varphi'$ with $O(n+m)$ variables and clauses. \[cor:eth-3,4-sat\] Unless ETH fails, there is no algorithm for [[[[$(3,\!4)$-SAT]{}]{}]{}]{}that runs in time $2^{o(n)}$, where $n$ denotes the number of variables of the input formula. Hardness of [[[[List Edge Coloring]{}]{}]{}]{}in Multigraphs {#sect:multi} ============================================================ In order to prove Theorems \[th:multi\] and \[th:simple\] we show reductions from [[[[$(3,\!4)$-SAT]{}]{}]{}]{}to [[[[List Edge Coloring]{}]{}]{}]{}with strong bounds on the number of vertices in the output instance. The basic idea of both our reductions is to use two colors, denoted by $x_i$ and $\neg x_i$ for every variable $x_i$ so that in every coloring of the out graph the edges colored in $x_i$ or $\neg x_i$ form a single path with alternating colors. Then colors at the edges of this path of fixed parity can encode the value of $x_i$ in a satisfying boolean assignment. Moreover, testing a clause $C=\ell_1\vee \ell_2\vee\ell_3$ can be done very easily: it suffices to add an edge with the list $\{\ell_1,\ell_2,\ell_3\}$. However this edge can belong to the alternating path of at most one of the three variables in $C$, and we add two more parallel edges which become elements of the two other alternating paths. Unfortunately, in order to get similar phenomenon in simple graphs, we need to introduce a complicated gadget. \[lem:reduction-multi\] For any instance $\varphi$ of [[[[$(3,\!4)$-SAT]{}]{}]{}]{}with $n$ variables there is an equivalent instance $(G,L)$ of [[[[List Edge Coloring in Multigraphs]{}]{}]{}]{}with 21 vertices and $O(n)$ edges. Moreover, the instance $(G,L)$ can be constructed in polynomial time. In what follows, we prove Lemma \[lem:reduction-multi\]. Let ${{\ensuremath{\rm{vrb}}}}(\varphi)$ and ${{\ensuremath{\rm{cls}}}}(\varphi)$ be the sets of variables and clauses of $\varphi$, respectively. W.l.o.g. assume ${{\ensuremath{\rm{vrb}}}}(\varphi)=\{x_0,\ldots,x_{n-1}\}$. We construct an auxiliary graph $G_\varphi$ with $V(G_\varphi)={{\ensuremath{\rm{cls}}}}(\varphi)$ and such that two clauses $C_1,C_2\in {{\ensuremath{\rm{cls}}}}(\varphi)$ are adjacent in $G_\varphi$ iff $C_1\cap C_2 \ne \emptyset$. Since every clause has three variables and each variable can belong to at most three other clauses, it follows that the maximum degree of $G_\varphi$ is at most $9$. Let $g:{{\ensuremath{\rm{cls}}}}(\varphi)\rightarrow[10]$ be the greedy vertex coloring of $G_\varphi$ in 10 colors, which can be found in linear time in a standard way. For $i\in [10]$, let ${{\ensuremath{\mathcal{C}}}}_i=g^{-1}(i)$. Let us describe the output instance $(G,L)$. We put $V(G)=\{v_0,\ldots,v_{20}\}$. The edges of $G$ join only vertices of consecutive indices. For every $r\in [10]$, for every clause $C\in {{\ensuremath{\mathcal{C}}}}_r$ we add three new edges with endpoints $v_{2r}$ and $v_{2r+1}$. The first of this edges, denoted by $e_C^1$, gets list $C$, i.e., the three literals of clause $C$. Let $x_i$, $x_j$ and $x_k$ be the three variables that appear in $C$. Then, the two remaining edges, $e_C^2$ and $e_C^3$, get identical lists of $\{x_i,\neg x_i,x_j,\neg x_j,x_k,\neg x_k\}$. Moreover, for every $r\in [10]$ and for every variable $x_i$ that does not appear in any of the clauses of ${{\ensuremath{\mathcal{C}}}}_r$, we add a new edge $v_{2r}v_{2r+1}$ with list $\{x_i,\neg x_i\}$. Finally, for every $r\in [10]$ and for every variable $x_i\in {{\ensuremath{\rm{vrb}}}}(\varphi)$ we add a single new edge $v_{2r+1}v_{2r+2}$ with list $\{x_i,\neg x_i\}$. This finishes the description of the output instance. See Fig. \[fig:ex-multi\] for an example. (v0) at (0, 2) [$v_0$]{}; (v1) at (5,2) [$v_1$]{}; (v2) at (10,2) [$v_2$]{}; (v3) at (15,2) [$v_3$]{}; (v0) – +(0,1cm) edge \[out=80,in=100\] node \[above\] [$x_1, x_2, \neg x_3$]{} (\[yshift=1cm\]v1); (v1) – +(0,1cm); (v0) edge \[bend left=50\] node \[above,pos=0.5\] [$x_1, \neg x_1, x_2, \neg x_2, x_3, \neg x_3$]{} (v1); (v0) edge node \[above,pos=0.5\] [$x_1, \neg x_1, x_2, \neg x_2, x_3, \neg x_3$]{} (v1); (v0) edge \[bend right=50\] node \[above,pos=0.5\] [$x_4,\neg x_4$]{} (v1); (v0) – +(0,-1cm) edge \[out=-80,in=-100\] node \[above\] [$x_5, \neg x_5$]{} (\[yshift=-1cm\]v1); (v1) – +(0,-1cm); (v1) – +(0,1cm) edge \[out=80,in=100\] node \[above\] [$x_1, \neg x_1$]{} (\[yshift=1cm\]v2); (v2) – +(0,1cm); (v1) edge \[bend left=50\] node \[above,pos=0.5\] [$x_2,\neg x_2$]{} (v2); (v1) edge node \[above,pos=0.5\] [$x_3,\neg x_3$]{} (v2); (v1) edge \[bend right=50\] node \[above,pos=0.5\] [$x_4,\neg x_4$]{} (v2); (v1) – +(0,-1cm) edge \[out=-80,in=-100\] node \[above\] [$x_5, \neg x_5$]{} (\[yshift=-1cm\]v2); (v2) – +(0,-1cm); (v2) – +(0,1cm) edge \[out=80,in=100\] node \[above\] [$\neg x_2, \neg x_4, \neg x_5$]{} (\[yshift=1cm\]v3); (v3) – +(0,1cm); (v2) edge \[bend left=50\] node \[above,pos=0.5\] [$x_2, \neg x_2, x_4, \neg x_4, x_5, \neg x_5$]{} (v3); (v2) edge node \[above,pos=0.5\] [$x_2, \neg x_2, x_4, \neg x_4, x_5, \neg x_5$]{} (v3); (v2) edge \[bend right=50\] node \[above,pos=0.5\] [$x_1,\neg x_1$]{} (v3); (v2) – +(0,-1cm) edge \[out=-80,in=-100\] node \[above\] [$x_3, \neg x_3$]{} (\[yshift=-1cm\]v3); (v3) – +(0,-1cm); In what follows, edges of the form $v_{2r}v_{2r+1}$ are called [*positive*]{} and edges of the form $v_{2r+1}v_{2r+2}$ are called [*negative*]{}. \[claim:variable-path\] For every list edge coloring $c$ of $(G,L)$, for every $i\in [n]$, the edges in $c^{-1}(\{x_i,\neg x_i\})$ form a path $v_0,v_1,\ldots,v_{20}$. For every $r\in[10]$, there is exactly one edge $v_{2r+1}v_{2r+2}$ with list containing $x_i$ or $\neg x_i$, namely with list $\{x_i,\neg x_i\}$. It follows that these 10 edges belong to $c^{-1}(\{x_i,\neg x_i\})$. It suffices to prove that for every $r\in[10]$ there is also exactly one edge $v_{2r}v_{2r+1}$ in $c^{-1}(\{x_i,\neg x_i\})$. This is clear when $x_i$ does not appear in any of the clauses of ${{\ensuremath{\mathcal{C}}}}_r$, because then there is exactly one edge $v_{2r+1}v_{2r+2}$ with list containing $x_i$ or $\neg x_i$, namely with list $\{x_i,\neg x_i\}$. Otherwise, let $C=\{\ell_i,\ell_j,\ell_k\}$ be the clause of ${{\ensuremath{\mathcal{C}}}}_r$ where $\ell_i\in\{x_i,\neg x_i\}$. Let $\ell_j\in\{x_j,\neg x_j\}$, $\ell_k\in\{x_k,\neg x_k\}$. Then there are exactly three edges $e_C^1$, $e_C^2$, $e_C^3$ incident to $v_{2r}$ and $v_{2r+1}$ and with list containing one of literals in the set $\{x_i,\neg x_i,x_j,\neg x_j,x_k,\neg x_k\}$. Indeed, $L(e_C^1)=\{\ell_i,\ell_j,\ell_k\}$, and $L(e_C^2)=L(e_C^3)=\{x_i,\neg x_i,x_j,\neg x_j,x_k,\neg x_k\}$. However, we have already proved that for every $q\in\{i,j,k\}$, one of the edges with endpoints $v_{2r+1}$ and $v_{2r+2}$ is colored with $x_q$ or $\neg x_q$. Hence, since every color class is a matching, for every $q\in\{i,j,k\}$, at most one of the edges in $\{e_C^1,e_C^2,e_C^3\}$ is colored with $x_q$ or $\neg x_q$. However, lists of $e_C^1$, $e_C^2$, $e_C^3$ contain only colors of the form $x_q$ or $\neg x_q$ for $q\in\{i,j,k\}$. It follows that for every $q\in\{i,j,k\}$ exactly one of the edges in $\{e_C^1,e_C^2,e_C^3\}$ is colored with $x_q$ or $\neg x_q$. In particular there is exactly one edge $v_{2r}v_{2r+1}$ in $c^{-1}(\{x_i,\neg x_i\})$. Since $c$ is an edge coloring, the path from the claim above is colored either by $x_i,\neg x_i, x_i, \neg x_i,\ldots$, or by $\neg x_i, x_i, \neg x_i,x_i,\ldots$. This implies the following claim. \[claim:one-value-multi\] For every list edge coloring $c$ of $(G,L)$, for every $i\in [n]$, we have $|c^{-1}(x_i)|=|c^{-1}(\neg x_i)|=10$ and either all edges in $c^{-1}(x_i)$ are positive and all edges in $c^{-1}(\neg x_i)$ are negative or all edges in $c^{-1}(x_i)$ are negative and all edges in $c^{-1}(\neg x_i)$ are positive. Now we are ready to prove that $\varphi$ and $(G,L)$ are equivalent. Assume $c$ is a list edge coloring of $(G,L)$. Define a boolean assignment $f:{{\ensuremath{\rm{vrb}}}}(\varphi)\rightarrow\{T,F\}$ by setting $x_i$ to $T$ iff all edges in $c^{-1}(x_i)$ are positive. Now consider an arbitrary clause $C$. By construction, there is a positive edge $e$ with $L(e)=C$. If $c(e)=x_q$ for some variable $x_q$ then by Claim \[claim:one-value-multi\] all edges in $c^{-1}(x_q)$ are positive, and hence $f(x_q)=T$. Since $c(e)\in L(e)$ we have $x_q \in C$, so $C$ is satisfied. If $c(e)=\neg x_q$ for some variable $x_q$ then by Claim \[claim:one-value-multi\] all edges in $c^{-1}(x_q)$ are negative and hence $f(x_q)=F$. Again, since $c(e)\in L(e)$ we have $\neg x_q \in C$, so $C$ is satisfied. Assume $\varphi$ is satisfiable and let $f:{{\ensuremath{\rm{vrb}}}}(\varphi)\rightarrow\{T,F\}$ be a satisfying assignment. We define a list edge coloring $c$ of $(G,L)$ as follows. Recall that for every $r\in[10]$, and for every clause $C\in{{\ensuremath{\mathcal{C}}}}_r$ there is an edge $e_C^1$ with $L(e_C^1)=C$ and edges $e_C^2$, $e_C^3$ with $L(e_C^2)=L(e_C^3)=\{x_i,\neg x_i,x_j,\neg x_j,x_k,\neg x_k\}$, where $x_i$, $x_j$ and $x_k$ are the three variables that appear in $C$. We color $e_C^1$ with any of the satisfied literals of $C$. By symmetry assume $c(e_C^1)\in\{x_i,\neg x_i\}$. Then we color $e_C^2$ with $x_j$ if $f(x_j)=T$ and with $\neg x_j$ otherwise. Similarly, we color $e_C^3$ with $x_k$ if $f(x_k)=T$ and with $\neg x_k$ otherwise. Each of the remaining positive edges $e$ of $G$ has its list equal $\{x_i,\neg x_i\}$ for some $x_i\in{{\ensuremath{\rm{vrb}}}}(\varphi)$. We color $e$ with $x_i$ if $f(x_i)=T$ and with $\neg x_i$ otherwise. It follows that every positive edge is colored with a satisfied literal. Every negative edge $\tilde{e}$ has its list equal to $\{x_i,\neg x_i\}$ for some $x_i\in{{\ensuremath{\rm{vrb}}}}(\varphi)$. We color $\tilde{e}$ with $x_i$ when $f(x_i)=F$ and with $\neg x_i$ when $f(x_i)=T$. It follows that every negative edge is colored with an unsatisfied literal. Let us show that $c$ does not color incident edges with the same color. Since the lists of parallel negative edges are disjoint, in our coloring there are no parallel negative edges of the same color. Assume there are two parallel positive edges of the form $v_{2r}v_{2r+1}$ of the same color $\ell$, for some $r\in[10]$. Then the variable of $\ell$ belongs to a clause in ${{\ensuremath{\mathcal{C}}}}_r$, for otherwise there is exactly one edge with endpoints $v_{2r}v_{2r+1}$ and with list containing $\ell$. However, since ${{\ensuremath{\mathcal{C}}}}_r$ is independent in $G_\varphi$, there is exactly one such clause $C$ in ${{\ensuremath{\mathcal{C}}}}_r$. It follows that the two parallel edges are among the three edges $e_C^1, e_C^2, e_C^3$. However, these three edges got different colors, a contradiction. If two edges are incident but not parallel, one of them is positive and the other negative. The former is colored with a satisfied literal and the latter with an unsatisfied literal, so they are colored differently. Hence $c$ is a proper list edge coloring, as required. This ends the proof of Lemma \[lem:reduction-multi\]. Theorem \[th:multi\] follows immediately from Lemmas \[lem:tovey\] and \[lem:reduction-multi\] and the NP-hardness of [[[[$3$-SAT]{}]{}]{}]{}. Hardness of [[[[List Edge Coloring]{}]{}]{}]{}in Simple Graphs {#sect:simple} ============================================================== This section is devoted to the proof of the following lemma. \[lem:reduction-simple\] For any instance $\varphi$ of [[[[$(3,\!4)$-SAT]{}]{}]{}]{}with $n$ variables there is an equivalent instance $(G,L)$ of [[[[List Edge Coloring in Simple Graphs]{}]{}]{}]{}with $O(\sqrt{n})$ vertices. Moreover, the instance $(G,L)$ can be constructed in polynomial time. Intuition --------- The general idea is to follow the approach of Lemma \[lem:reduction-multi\] and replace the edges with multiplicity $O(n)$ with bipartite graphs with $O(\sqrt{n})$ vertices and $O(n)$ edges. It seems that using only one such graph instead of every bunch of parallel edges with common endpoints is not enough to get a simple graph (though it suffices to reduce the multiplicity to three). In our construction, for every $r\in[10]$, we replace every two consecutive bunches of parallel edges between $v_{2r}$, $v_{2r+1}$, and $v_{2r+2}$ from the construction in Lemma \[lem:reduction-multi\] by seven layers $L_i$, $i=6r+1,\ldots,6r+7$, each of $O(\sqrt{n})$ vertices, with some edges joining both consecutive and non-consecutive layers. The subgraph induced by $\bigcup_{i=6r+1}^{6r+7}L_i$ is called the $r$-th [*clause verifying gadget*]{} $G_r$. (Note that the layers $L_i$ for $i\equiv 1\pmod 6$ are shared between consecutive gadgets.) Analogously as in Lemma \[lem:reduction-multi\], the role of $G_r$ is to check whether all clauses in ${{\ensuremath{\mathcal{C}}}}_r$ are satisfied. We add also two additional layers $L_0$ and $L_{62}$ which make some of our arguments simpler. Construction ------------ It will be convenient to assume that $\sqrt{n}\in\mathbb{N}$. We do not lose on generality because otherwise we just add $n^+=({\ensuremath{\left\lceil{\sqrt{n}}\right\rceil}}+1)^2-n$ variables $y_1,y_2,\ldots,y_{n^+}$ and clauses $$\{y_1,y_2,y_3\},\{y_2,y_3,y_4\},\ldots,\{y_{n^+-2},y_{n^+-1},y_{n^+}\}.$$ Note that $n^+\ge 3$, $n^+\le(\sqrt{n}+2)^2-n=4\sqrt{n}+4$ and $\sqrt{n+n^+}={\ensuremath{\left\lceil{\sqrt{n}}\right\rceil}}+1\in\mathbb{N}$. Hence we added only $O(\sqrt{n})$ variables and clauses, and the resulting formula is still a [[[[$(3,\!4)$-SAT]{}]{}]{}]{}instance. We begin as in Lemma \[lem:reduction-multi\], by building the graph $G_\varphi$, finding its greedy coloring $g$ which partitions the clause set into 10 color classes ${{\ensuremath{\mathcal{C}}}}_r$, $r\in[10]$. Let us build the instance $(G,L)$ step by step. Add two sets of vertices (called [*layers*]{}) $L_i=\{v^i_j \mid j\in [\sqrt{n}]\}$, $i=0,1$. Then add all possible $n$ edges between $L_0$ and $L_1$ forming a complete bipartite graph. Map the $n$ variables to the $n$ edges in a $1-1$ way. For every $i\in[n]$, set the list of the edge assigned to $x_i$ to $\{x_i,\neg x_i\}$. The vertex set $V(G)$ contains further 60 layers of vertices $L_i$, $i=\{2,\ldots,61\}$, where $L_{i} = \{v^{i}_j \mid j\in [6\sqrt{n}+3]\}$. Finally, $L_{62} = \{v^{62}_j \mid j\in [\sqrt{n}+1]\}$. Denote also $L_{-1} = L_{63} = \emptyset$. In what follows we add the remaining edges of $G$. Whenever we add edges between $L_i$ and $L_{i-1}$, for every $j<i$ all the edges of the output graph between $L_j$ and $L_{j-1}$ are already added. We will make sure to keep the following invariants satisfied during the process of construction (note that they hold for the part constructed so far). \[inv:unique\] For every $i\in[62]$, for every variable $x_j\in {{\ensuremath{\rm{vrb}}}}(\varphi)$ there is at most one edge $uv\in E(L_i, L_{i+1})$ such that $\{x_j,\neg x_j\} \cap L(uv)\ne \emptyset$. Moreover, after finishing of adding edges between $L_i$ and $L_{i+1}$, there is exactly one such edge. Using the notation from Invariant \[inv:unique\], if the edge $uv$ exists, we can denote $v^+_{i,j}=u$ and $v^-_{i+1,j}=v$. \[inv:flow\] For every $i\in\{1,\ldots,62\}$, for every variable $x_j\in {{\ensuremath{\rm{vrb}}}}(\varphi)$ we have that $v^-_{i,j}=v^+_{i,j}$, unless $v^-_{i,j}$ or $v^+_{i,j}$ is undefined. Moreover, the equality holds after finishing of adding edges between $L_i$ and $L_{i+1}$. Thanks to Invariant \[inv:flow\], after finishing of adding edges between $L_i$ and $L_{i+1}$, we can just define $v_{i,j}:=v^-_{i,j}=v^+_{i,j}$ for $i\in\{1,\ldots,61\}$. We also put $v_{0,j}=v^+_{0,j}$ and $v_{62,j}=v^-_{62,j}$. In our construction we will use some additional colors apart from the literals. However, the following invariant holds. \[inv:lists\] For every edge $e$ of $G$, the list $L(e)$ contains at least one literal. For every $i\in[62]$ for every vertex $v\in L_i$ let $\deg^-(v)=|E(L_{i-1}),\{v\}|$ and $\deg^+(v)=|E(L_{i+1}),\{v\}|$. \[inv:indegree\] For every $i\in[62]$ for vertex $v\in L_i$ we have $\deg^-(v) \le \sqrt{n}$. \[inv:jump\] For every $i\in[62]$, for vertex $v\in L_i$ there are at most $\sqrt{n}$ edges from $v$ to layers $L_j$ for $j>i+1$. By Invariant \[inv:flow\] and Invariant \[inv:lists\], for every vertex $v\in V(G)$ it holds that $\deg^+(v)\le \deg^-(v)$. Hence Invariant \[inv:indegree\] gives the claim below. \[cl:outdegree\] For every $i\in[62]$ for every vertex $v\in L_i$ we have $\deg^+(v) \le \sqrt{n}$. Invariants \[inv:unique\] and \[inv:lists\] immediately imply the following. \[cl:n-edges\] For every $i\in[62]$, we have $|E(L_i,L_{i+1})| \le n$. Let us fix $r\in[10]$. We add the edges of the $r$-th [*clause verifying gadget*]{} $G_r$. Although $G$ is undirected, we will say that an edge $uv$ between $L_i$ and $L_j$ for $i<j$ is [*from $u$ to $v$*]{} and [*from $L_i$ to $L_j$*]{}. Below we [*describe*]{} the edges in $G_r$ in the order which is convenient for the exposition. However, the [*algorithm*]{} adds the edges between layers in the left-to-right order, i.e., for $i<j$, edges to $L_i$ are added before edges to $L_j$. 1. Edges to $L_{\ell}$ for $\ell=6r+2,6r+4,6r+6$. For every clause $C\in{{\ensuremath{\mathcal{C}}}}_r$ we do the following. Let $x_{i_1}, x_{i_2}, x_{i_3}$ be the three different variables that appear in the literals of $C$. Let $v_j = v^-_{\ell-1,i_j}$ for $j=1,2,3$. Note that vertices $v_1$, $v_2$, $v_3$ need not be distinct. By Claim \[cl:outdegree\], $|N(v_j)\cap L_{\ell}|\le \sqrt{n}$ for $j=1,2,3$. Let $S=\{v \in L_{\ell} \mid \deg^-(v) = \sqrt{n}\}$. By Claim \[cl:n-edges\], there is $|S|\le\sqrt{n}$. Hence, for $j=1,2,3$ we have $|L_{\ell} \setminus (N(\{v_j\}) \cup S)| \ge 4\sqrt{n}+3$ and we can pick a vertex $w_j\in L_{\ell}$ that has at most $\sqrt{n}-1$ edges from $L_{\ell-1}$, is not adjacent to $v_j$, and is different than $w_{j'}$ for each $j'<j$. If $\ell=6k+6$ we additionally require that for every $j=1,2,3$, the vertex $w_j$ is not adjacent to $v^-_{6r+2,i_j}$ or $v^-_{6r+4,i_j}$. By Invariant \[inv:jump\] this eliminates at most $2\sqrt{n}$ more candidates, so it is still possible to choose all the $w_j$’s. For each $j=1,2,3$, we add an edge $v_jw_j$ with $L(v_jw_j)=\{x_{i_j},\neg x_{i_j}\}$. Moreover, if $\ell=6k+6$, for every $j=1,2,3$ we add an edge $v^-_{6r+2,i_j}w_j$ with list $\{x_{i_j},\neg x_{i_j},a_{i,j}\}$ and an edge $v^-_{6r+4,i_j}w_j$ with list $\{x_{i_j},\neg x_{i_j},b_{i,j}\}$. The conditions used to choose $w_1$, $w_2$ and $w_3$ guarantee that we do not introduce parallel edges. For every variable $x_i$ that is not present in any of the clauses of ${{\ensuremath{\mathcal{C}}}}_r$ we find a vertex $w\in L_{\ell}$ that has at most $\sqrt{n}-1$ edges from $L_{\ell-1}$ and is not adjacent to $v^-_{\ell-1,i}$. Again, this is possible because there are at most $2\sqrt{n}$ vertices in $L_{\ell}$ that violate any of these constraints. We add an edge $v^-_{\ell-1,i}w$ with $L(v^-_{\ell-1,i}w)=\{x_{i},\neg x_{i}\}$. Note that all invariants are satisfied: for Invariant \[inv:unique\] it follows from the fact that ${{\ensuremath{\mathcal{C}}}}_r$ is independent in $G_\varphi$, while invariants \[inv:flow\], \[inv:lists\], \[inv:indegree\] follow immediately from the construction. Invariant \[inv:jump\] stays satisfied after adding $v^-_{6r+2,i_j}w_j$ because for every variable $x_k$ such that $v^-_{6r+2,k}=v^-_{6r+2,i_j}$ we add at most one edge from $v^-_{6r+2,i_j}$ to $L_{6r+6}$, and the number of such variables is equal to $\deg^-(v^-_{6r+2,i_j})$, which is at most $\sqrt{n}$ by Invariant \[inv:indegree\] (analogous argument applies to adding the edge $v^-_{6r+4,i_j}w_j$). 2. Edges to $L_{\ell}$ for $\ell=6r+3,6r+5,6r+7$. For every clause $C\in{{\ensuremath{\mathcal{C}}}}_r$ we do the following. Let $C=\{\ell_1,\ell_2,\ell_3\}$ and let $x_{i_j}$ be the variable from the literal of $\ell_j$, for $j=1,2,3$. Let $w_j = v^-_{\ell-1,i_j}$ for $j=1,2,3$. By Claim \[cl:outdegree\], $|N(\{w_1,w_2,w_3\}\cap L_{\ell})|\le 3\sqrt{n}$. Also, there are at most $\sqrt{n}+2$ vertices in $L_{\ell}$ with at least $\sqrt{n}-2$ edges from $L_{\ell-1}$. Indeed, otherwise $|E(L_{\ell-1},L_{\ell})|\ge n + \sqrt{n} - 6$ and either $n \le 36$ (and the lemma is trivial) or there is a contradiction with Claim \[cl:n-edges\]. Hence, we can find a vertex $z_{\ell,C}\in L_{\ell}$ that has at most $\sqrt{n}-3$ edges to $L_{\ell-1}$ and is not adjacent to $\{w_1,w_2,w_3\}$. If $\ell=6k+7$ we additionally require that the vertex $z_{6k+7,C}$ is not adjacent to $z_{6k+3,C}$ or $z_{6k+5,C}$. By Invariant \[inv:jump\] this eliminates at most $2\sqrt{n}$ more candidates, so it is still possible to choose vertex $z_{6k+7,C}$. For each $j=1,2,3$, we add an edge $w_jz_{\ell,C}$. We put $L(w_jz_{6r+3,C})=\{x_{i_j},\neg x_{i_j},a_{i_j}\}$, $L(w_jz_{6r+5,C})=\{x_{i_j},\neg x_{i_j},b_{i_j}\}$, and $L(w_jz_{6r+7,C})=\{\ell_j,c_{C},d_{C}\}$. (The colors $a_{i_j},b_{i_j},c_{C},d_{C}$ are not literals — these are new auxiliary colors; each variable $x_i$ has its own distinct auxiliary colors $a_i, b_i$, and each clause $C$ has its own auxiliary colors $c_C, d_C$.) We add edges $z_{6r+3,C}z_{6r+7,C}$ and $z_{6r+5,C}z_{6r+7,C}$, both with lists $\{x_{i_1},\neg x_{i_1},x_{i_2},\neg x_{i_2},x_{i_3},\neg x_{i_3}\}$. For every variable $x_i$ that is not present in any of the clauses of ${{\ensuremath{\mathcal{C}}}}_r$ we proceed analogously as in Step 1. The invariants hold for the similar reasons as before. In particular, Invariant \[inv:jump\] stays satisfied after adding $z_{6r+3,C}z_{6r+7,C}$ because for every clause $C'$ such that $z_{6r+3,C'}=z_{6r+3,C}$ we add exactly one edge from $z_{6r+3,C}$ to $L_{6r+7}$, and the number of such clauses is bounded by $\deg^-(z_{6r+3,C})/3$, which is at most $\sqrt{n}/3$ by Invariant \[inv:indegree\] (analogous argument applies to adding the edge $z_{6r+5,C}z_{6r+7,C}$). Finally, we add edges between $L_{61}$ and $L_{62}$. For every variable $x_i$ we find a vertex $w\in L_{62}$ that is not adjacent to $v^-_{61,i}$, which is possible because $\deg^+(v^-_{61,i}) \le \sqrt{n}$. We add an edge $v^-_{61,i}w$ with $L(v^-_{61,i}w)=\{x_{i},\neg x_{i}\}$. The following claims follow directly from the construction. \[cl:struct1\] For every $r\in[10]$, for every clause $C\in{{\ensuremath{\mathcal{C}}}}_r$ with variables $x_{i_1}, x_{i_2}, x_{i_3}$, and for each $\ell=6r+3,6r+5,6r+7$ we have $v_{\ell,i_1}=v_{\ell,i_2}=v_{\ell,i_3}=z_{\ell,C}$. Moreover, for each $\ell=6r+3,6r+5,6r+7$ and $j=1,2,3$ we have $L(z_{\ell,C}v_{\ell+1,i_j})=\{x_{i_j},\neg x_{i_j}\}$. \[cl:struct2\] For every edge $uv\in E(G)$, where $u\in L_j, v\in L_k$, if $\{x_i,\neg x_i\}\cap L(uv)\ne\emptyset$, then $u=v_{j,i}$ and $u=v_{k,i}$. (v11) at (0,4) [$v_{6r+1,i}$]{}; (v12) at (0,2) [$v_{6r+1,j}$]{}; (v13) at (0,0) [$v_{6r+1,k}$]{}; (v21) at (3,4) [$v_{6r+2,i}$]{}; (v22) at (3,2) [$v_{6r+2,j}$]{}; (v23) at (3,0) [$v_{6r+2,k}$]{}; (v3) at (6,2) [$v_{6r+3,i}$]{}; (v41) at (9,4) [$v_{6r+4,i}$]{}; (v42) at (9,2) [$v_{6r+4,j}$]{}; (v43) at (9,0) [$v_{6r+4,k}$]{}; (v5) at (12,2) [$v_{6r+5,i}$]{}; (v61) at (15,4) [$v_{6r+6,i}$]{}; (v62) at (15,2) [$v_{6r+6,j}$]{}; (v63) at (15,0) [$v_{6r+6,k}$]{}; (v7) at (18,2) [$v_{6r+7,i}$]{}; (v11) edge node \[above,pos=0.5\] [$x_i,\neg x_i$]{} (v21); (v12) edge node \[above,pos=0.5\] [$x_j,\neg x_j$]{} (v22); (v13) edge node \[above,pos=0.5\] [$x_k,\neg x_k$]{} (v23); (v21) edge node \[above,pos=0.5,sloped\] [$x_i,\neg x_i,a_i$]{} (v3); (v22) edge node \[above,pos=0.5\] [$x_j,\neg x_j,a_j$]{} (v3); (v23) edge node \[above,pos=0.5,sloped\] [$x_k,\neg x_k,a_k$]{} (v3); (v3) edge node \[above,pos=0.5,sloped\] [$x_i,\neg x_i$]{} (v41); (v3) edge node \[above,pos=0.5\] [$x_j,\neg x_j$]{} (v42); (v3) edge node \[above,pos=0.5,sloped\] [$x_k,\neg x_k$]{} (v43); (v41) edge node \[above,pos=0.5,sloped\] [$x_i,\neg x_i,b_i$]{} (v5); (v42) edge node \[above,pos=0.5\] [$x_j,\neg x_j,b_j$]{} (v5); (v43) edge node \[above,pos=0.5,sloped\] [$x_k,\neg x_k,b_k$]{} (v5); (v5) edge node \[above,pos=0.5,sloped\] [$x_i,\neg x_i$]{} (v61); (v5) edge node \[above,pos=0.5\] [$x_j,\neg x_j$]{} (v62); (v5) edge node \[above,pos=0.5,sloped\] [$x_k,\neg x_k$]{} (v63); (v61) edge node \[above,pos=0.5,sloped\] [$x_i,c_C,d_C$]{} (v7); (v62) edge node \[above,pos=0.5\] [$\neg x_j,c_C,d_C$]{} (v7); (v63) edge node \[above,pos=0.5,sloped\] [$x_k,c_C,d_C$]{} (v7); (v3) – +(0,2cm) edge \[out=90,in=90\] node \[above,sloped\] [$x_i,\neg x_i,x_j,\neg x_j,x_k,\neg x_k$]{} (\[yshift=2cm\]v7); (v7) – +(0,2cm); (v5) – +(0,1.5cm) edge \[out=80,in=100\] node \[above\] [$x_i,\neg x_i,x_j,\neg x_j,x_k,\neg x_k$]{} (\[yshift=1.5cm\]v7); (v7) – +(0,1.5cm); (v21) edge \[bend left\] node \[above\] [$x_i,\neg x_i,a_i$]{} (v61); (v41) edge \[bend left\] node \[above,pos=0.2,sloped\] [$x_i,\neg x_i,b_i$]{} (v61); (v22) edge \[bend right=50\] node \[below\] [$x_j,\neg x_j,a_j$]{} (v62); (v42) edge \[bend right\] node \[below\] [$x_j,\neg x_j,b_j$]{} (v62); (v23) edge \[bend right=40\] node \[below\] [$x_j,\neg x_j,a_k$]{} (v63); (v43) edge \[bend right\] node \[below\] [$x_j,\neg x_j,b_k$]{}(v63); This finishes the description of the output instance. Since $G$ contains $O(1)$ layers, each with $O(\sqrt{n})$ vertices, it follows that $|V(G)|=O(\sqrt{n})$, as required. See Fig \[fig:gadget\] for an illustration of edges representing a single clause within a clause verifying gadget. Structure of coloring --------------------- Similarly as for multigraphs the crux of the equivalence between instances is the following claim. \[claim:variable-path-simple\] For every list edge coloring $c$ of $(G,L)$, for every $i\in [n]$, the edges in $c^{-1}(\{x_i,\neg x_i\})$ form a path $P_i$ from $L_0$ to $L_{62}$. Moreover, if $P_i$ contains an edge $v_{6r+6,i}v_{6r+7,i}$ for some $r\in[10]$, then this edge in preceded by an even number of edges on $P_i$. Fix $i\in[n]$. For convenience, denote $E_i=c^{-1}(\{x_i,\neg x_i\})$. By Invariant \[inv:unique\] there is exactly one edge between $L_0$ and $L_1$ that has $x_i$ or $\neg x_i$ on its list, namely $v_{0,i}v_{1,i}$. Similarly, there is exactly one edge between $L_{61}$ and $L_{62}$ that has $x_i$ or $\neg x_i$ on its list, namely $v_{61,i}v_{62,i}$. Since $L(v_{0,i}v_{1,i})=L(v_{61,i}v_{62,i})=\{x_i,\neg x_i\}$, we know that $v_{0,i}v_{1,i},v_{61,i}v_{62,i}\in E_i$, and these are the only edges of $E_i$ in $E(L_0, L_1) \cup E(L_{61}, L_{62})$. Observe that edges between non-consecutive layers never leave the clause verifying gadgets. Hence, for the first part of the claim, it suffices to show that for every $r\in[10]$, the edges in $E_i \cap E(G_r)$ form a path between $v_{6r+1,i}$ and $v_{6r+7,i}$. In fact, by Claim \[cl:struct2\] it suffices to show that $E_i \cap E(G_r)$ [*contains*]{} a path between $v_{6r+1,i}$ and $v_{6r+7,i}$ that visits all the vertices $\{v_{6r+j,i}\mid j=1,\ldots,7\}$. To this end, fix $r\in[10]$. First assume that $x_i$ does not appear in any clause of ${{\ensuremath{\mathcal{C}}}}_r$. Then $G_r$ contains the path $v_{6r+1,i},v_{6r+2,i},\ldots, v_{6r+7,i}$, where each edge has the list $\{x_i,\neg x_i\}$. It immediately implies that all edges of this path are in $E_i \cap E(G_r)$, as required. Now let us assume that $x_i$ appears in a clause $C\in{{\ensuremath{\mathcal{C}}}}_r$. Let $C=\{\ell_i, \ell_j, \ell_k\}$ and assume that the literal $\ell_i$ contains $x_i$, the literal $\ell_j$ contains a variable $x_j$, and the literal $\ell_k$ contains a variable $x_k$. Observe that for $j=1,3,5$ we have $v_{6r+j,i}v_{6r+j+1,i} \in E_i$ because these edges have their lists equal to $\{x_i,\neg x_i\}$. Note also that $\Delta(E_i)\le 2$ because $E_i$ is a union of two matchings (colors). We consider three subcases. 1. Assume $v_{6r+3,i}v_{6r+7,i} \in E_i$. Since $\Delta(E_i)\le 2$ and $v_{6r+3,i}v_{6r+4,i} \in E_i$ we know that $v_{6r+2,i}v_{6r+3,i}\not\in E_i$, and as a consequence, $c(v_{6r+2,i}v_{6r+3,i})=a_{i}$. Hence $c(v_{6r+2,i}v_{6r+6,i})\ne a_{i}$, which implies that $v_{6r+2,i}v_{6r+6,i}\in E_i$. Then, since $\Delta(E_i)\le 2$ and $v_{6r+5,i}v_{6r+6,i} \in E_i$ we know that $v_{6r+4,i}v_{6r+6,i}\not\in E_i$, and as a consequence, $c(v_{6r+4,i}v_{6r+6,i})=b_{i}$. Hence $c(v_{6r+4,i}v_{6r+5,i})\ne b_{i}$, which implies that $v_{6r+4,i}v_{6r+5,i}\in E_i$. Thus, we have shown that $E_i$ contains the path $v_{6r+1,i},v_{6r+2,i},v_{6r+6,i},v_{6r+5,i},v_{6r+4,i},v_{6r+3,i},v_{6r+7,i}$, as required. 2. Assume $v_{6r+5,i}v_{6r+7,i} \in E_i$. Since $\Delta(E_i)\le 2$ and $v_{6r+5,i}v_{6r+6,i} \in E_i$ we know that $v_{6r+4,i}v_{6r+5,i}\not\in E_i$, and as a consequence, $c(v_{6r+4,i}v_{6r+5,i})=b_{i}$. Hence $c(v_{6r+4,i}v_{6r+6,i})\ne b_{i}$, which implies that $v_{6r+4,i}v_{6r+6,i}\in E_i$. Then, since $\Delta(E_i)\le 2$ and $v_{6r+5,i}v_{6r+6,i} \in E_i$ we know that $v_{6r+2,i}v_{6r+6,i}\not\in E_i$, and as a consequence, $c(v_{6r+2,i}v_{6r+6,i})=a_{i}$. Hence $c(v_{6r+2,i}v_{6r+3,i})\ne a_{i}$, which implies that $v_{6r+2,i}v_{6r+3,i}\in E_i$. Thus, we have shown that $E_i$ contains the path $v_{6r+1,i},v_{6r+2,i},v_{6r+3,i},v_{6r+4,i},v_{6r+6,i},v_{6r+5,i},v_{6r+7,i}$, as required. 3. Assume $v_{6r+3,i}v_{6r+7,i},v_{6r+5,i}v_{6r+7,i} \not\in E_i$. Since $L(v_{6r+3,i}v_{6r+7,i}) = L(v_{6r+5,i}v_{6r+7,i}) = \{x_i, \neg x_i, x_j, \neg x_j, x_k, \neg x_k\}$ we infer that $v_{6r+3,i}v_{6r+7,i},$ $v_{6r+5,i}v_{6r+7,i} \in E_j \cup E_k$. By Claim \[cl:struct1\] we know that $v_{6r+7,i}=v_{6r+7,j}=v_{6r+7,k}$, $v_{6r+7,i}v_{6r+8,j}\in E_j$ and $v_{6r+7,i}v_{6r+8,k}\in E_k$. Since $\Delta(E_j)\le 2$ and $\Delta(E_k)\le 2$, we get that $v_{6r+3,i}v_{6r+7,i}\in E_j$ and $v_{6r+5,i}v_{6r+7,i} \in E_k$ or vice versa. In any case, $v_{6k+6,j},v_{6k+7,i}\not\in E_j$, and $v_{6k+6,k},v_{6k+7,i}\not\in E_k$. Recall that $L(v_{6k+6,j},v_{6k+7,i}) = \{\ell_j,c_C,d_C\}$ and $L(v_{6k+6,k},v_{6k+7,i}) = \{\ell_k,c_C,d_C\}$. It follows that $c(\{v_{6k+6,j}v_{6k+7,i}, v_{6k+6,k}v_{6k+7,i}\})=\{c_C,d_C\}$. Then $c(v_{6k+6,i},v_{6k+7,i})\not\in\{c_C,d_C\}$. Since $L(v_{6k+6,i},v_{6k+7,i}) = \{\ell_i,c_C,d_C\}$, we get that $v_{6k+6,i},v_{6k+7,i} \in E_i$. Then, since $\Delta(E_i)\le 2$ and $v_{6r+5,i}v_{6r+6,i} \in E_i$ we know that $v_{6r+2,i}v_{6r+6,i},v_{6r+4,i}v_{6r+6,i}\not\in E_i$, and as a consequence, $c(v_{6r+2,i}v_{6r+6,i})=a_{i}$ and $c(v_{6r+4,i}v_{6r+6,i})=b_{i}$. Hence $c(v_{6r+2,i}v_{6r+3,i})\ne a_{i}$, and $c(v_{6r+4,i}v_{6r+5,i})\ne b_{i}$ which implies that $v_{6r+2,i}v_{6r+3,i},v_{6r+4,i}v_{6r+5,i}$ $\in E_i$. Thus, we have shown that $E_i$ contains the path $v_{6r+1,i},v_{6r+2,i},v_{6r+3,i},$ $v_{6r+4,i},v_{6r+5,i},v_{6r+6,i},v_{6r+7,i}$, as required. For the second part of the claim recall that $P_i$ decomposes into an edge from $L_0$ to $L_1$, 10 paths of length 6 inside the gadgets and an edge from $L_{61}$ to $L_{62}$. Moreover, if $P_i$ contains an edge $v_{6r+6,i}v_{6r+7,i}$ for some $r\in[10]$, then this edge is the last edge of one of the 10 paths of length 6. It follows that it is preceded by $1+6r+5$ edges, which is an even number. Equivalence ----------- Assume $c$ is a list edge coloring of $(G,L)$. Define a boolean assignment $f:{{\ensuremath{\rm{vrb}}}}(\varphi)\rightarrow\{T,F\}$ by setting $x_i$ to $T$ iff the first edge of the path $P_i$ from Claim \[claim:variable-path-simple\] is colored by $x_i$. Note that $P_i$ is colored alternately with $x_i$ and $\neg x_i$ and every odd edge on $P_i$ (i.e., preceded by an even number of edges) is colored with a satisfied literal. Now consider an arbitrary clause $C$. Let $r=g(C)$. Let $C=\{\ell_1,\ell_2,\ell_3\}$ and let $x_{i_j}$ be the variable from the literal of $\ell_j$, for $j=1,2,3$. By construction, there are three edges $v_{6r+6,i_j}z_{6r+7,C}$, for $j=1,2,3$ with $L(v_{6r+6,i_j}z_{6r+7,C})=\{\ell_j,c_C, d_C\}$. At most two of these edges are colored with $c_C$ or $d_C$, so there is $j=1,2,3$ such that $c(v_{6r+6,i_j}z_{6r+7,C}) = \ell_j$. In particular, $v_{6r+6,i_j}z_{6r+7,C} \in c^{-1}(\{x_{i_j},\neg x_{i_j}\})$ and hence, by Claim \[claim:variable-path-simple\] we know that $v_{6r+6,i_j}z_{6r+7,C}\in P_{i_j}$. However, by the second part of Claim \[claim:variable-path-simple\] this edge is preceded by an even number of edges on $P_{i_j}$. It follows that $\ell_j$ is satisfied. Assume $\varphi$ is satisfiable and let $f:{{\ensuremath{\rm{vrb}}}}(\varphi)\rightarrow\{T,F\}$ be a satisfying assignment. We define a list edge coloring $c$ of $(G,L)$ as follows. Consider any edge $e\in E(L_0,L_1)$. Then $L(e)=\{x_i,\neg x_i\}$. We color $e$ with $x_i$ when $f(x_i)=T$ and with $\neg x_i$ otherwise. Now consider any edge $e\in E(L_{61},L_{62})$. Again $L(e)=\{x_i,\neg x_i\}$. We color $e$ with $x_i$ when $f(x_i)=F$ and with $\neg x_i$ otherwise. By Invariant \[inv:unique\] incident edges get different colors in the partial coloring described so far. In what follows we describe $c|_{E(G_r)}$ for every $r\in[10]$ separately. Fix $r\in [10]$. Consider an arbitrary clause $C\in{{\ensuremath{\mathcal{C}}}}_r$. Let $C=\{\ell_1,\ell_2,\ell_3\}$ and let $x_{i_j}$ be the variable from the literal of $\ell_j$, for $j=1,2,3$. Since $\varphi$ is satisfied by $f$, at least one literal of $C$ is satisfied by $f$, by symmetry we can assume it is $\ell_1$. Consider the three edge disjoint paths $$\begin{aligned} R_1 &=v_{6r+1,i_1},v_{6r+2,i_1},v_{6r+3,i_1}, v_{6r+4,i_1}, v_{6r+5,i_1},v_{6r+6,i_1},v_{6r+7,i_1}, \\ R_2 & =v_{6r+1,i_2},v_{6r+2,i_2},v_{6r+6,i_2},v_{6r+5,i_2},v_{6r+4,i_2},v_{6r+3,i_2},v_{6r+7,i_2}, \\ R_3 &= v_{6r+1,i_3},v_{6r+2,i_3},v_{6r+3,i_3},v_{6r+4,i_3},v_{6r+6,i_3},v_{6r+5,i_3},v_{6r+7,i_3}. \end{aligned}$$ For each $j=1,2,3$ the path $R_j$ is colored by $x_{i_j}$ and $\neg x_{i_j}$ alternately, beginning with $\neg x_{i_j}$ if $f(x_{i_j})=T$ and with $x_{i_j}$ if $f(x_{i_j})=F$. Note that edges of $R_1$, $R_2$ and $R_3$ are colored by colors from their lists. Indeed, this is obvious for every edge apart from $v_{6r+6,i_1},v_{6r+7,i_1}$, because their lists contain $\{x_{i_j},\neg x_{i_j}\}$. Edge $v_{6r+6,i_1},v_{6r+7,i_1}$ is colored with $x_{i_j}$ if $f(x_{i_j})=T$ and with $\neg x_{i_j}$ if $f(x_{i_j})=F$. It follows that $v_{6r+6,i_1},v_{6r+7,i_1}$ is colored with the literal from $\{x_{i_1},\neg x_{i_1}\}$ which is satisfied by $f$, hence it is colored by $\ell_1$, and $\ell_1\in L(v_{6r+6,i_1},v_{6r+7,i_1})$, as required. Finally, we put $c(v_{6r+2,i_1}v_{6r+6,i_1})=a_{i_1}$, $c(v_{6r+4,i_1}v_{6r+6,i_1})=b_{i_1}$, $c(v_{6r+2,i_2}v_{6r+3,i_2})=a_{i_2}$, $c(v_{6r+4,i_2}v_{6r+6,i_2})=b_{i_2}$, $c(v_{6r+2,i_3}v_{6r+6,i_3})=a_{i_3}$, $c(v_{6r+4,i_3}v_{6r+5,i_3})=b_{i_3}$, $c(v_{6r+6,i_2}v_{6r+7,i_2})=c_C$, $c(v_{6r+6,i_3}v_{6r+7,i_3})=d_C$. Thus we have colored all edges of $G_r$ which have lists containing a variable from $C$. Now consider any variable $x_i$ that does not appear in any clause of ${{\ensuremath{\mathcal{C}}}}_r$. Consider the path $v_{6r+1,i},v_{6r+2,i},\ldots, v_{6r+7,i}$. If $f(x_i)=T$, color the path with the sequence of colors $\neg x_i,x_i,\neg x_i,\ldots, x_i$, and otherwise with the sequence of colors $x_i,\neg x_i,x_i,\ldots, \neg x_i$. Thus we have colored all the edges of $G_r$. It is straightforward to check that for every $r\in[10]$ the subgraph $G_r$ is colored properly. It remains to show that vertices in the layers $L_i$ for $i\equiv 1 \pmod 6$ are not incident to two edges of the same color. Clearly, this cannot happen for colors $a_j$ or $b_j$ for any $j\in[n]$, because they are not present on lists of edges incident to $L_i$ for $i\equiv 1 \pmod 6$. Also, it cannot happen for colors $c_C$ or $d_C$ for any clause $C$, because edges with these colors on their list only join $L_{i-1}$ with $L_{i}$ for $i\equiv 1 \pmod 6$, so two incident edges colored with $c_C$ or $d_C$ cannot belong to different gadgets. Finally, consider colors $\{x_i,\neg x_i\}$ for a fixed $i\in [n]$. The edges with these colors form a path of length 62, starting with $v_{0,i}v_{1,i}$, and continued as follows. The edge $v_{0,i}v_{1,i}$ is followed by 10 paths of length 6. For every $r\in[10]$, the $r$-th path of length 10 begins in $v_{6r+1,i}$ and ends in $v_{6r+7,i}=v_{6(r+1)+1,i}$. Finally, the 62-path ends with edge $v_{61,i}v_{62,i}$. Note that $v_{0,i}v_{1,i}$ is colored with the satisfied literal. Next, for every $r\in[10]$, the first edge of the $r$-th 10-path is colored with the non-satisfied literal and its last edge is colored by the satisfied literal. Finally, $v_{61,i}v_{62,i}$ is colored with the non-satisfied literal. It follows that the 62-path of all edges with colors from $\{x_i,\neg x_i\}$ is colored alternately in $x_i$ and $\neg x_i$, as required. This finishes the proof that $c$ is a list edge coloring of $(G,L)$, and the proof of Lemma \[lem:reduction-simple\]. Proof of Theorem \[th:simple\] ------------------------------ Theorem \[th:simple\] follows immediately from Lemma \[lem:reduction-simple\] and Corollary \[cor:eth-3,4-sat\]. Indeed, if there is an algorithm $A$ which solves [[[[List Edge Coloring in Simple Graphs]{}]{}]{}]{}in time $2^{o(|V(G)|^2)}$, then by Lemma \[lem:reduction-simple\] an $n$-variable instance of [[[[$(3,\!4)$-SAT]{}]{}]{}]{}can be transformed to a $O(\sqrt{n})$-vertex instance of [[[[List Edge Coloring in Simple Graphs]{}]{}]{}]{}in polynomial time and next solved in time $2^{o(n)}$ using $A$, which contradicts ETH by Corollary \[cor:eth-3,4-sat\]. Conclusions and further research ================================ In this work we have shown that [[[[List Edge Coloring in Simple Graphs]{}]{}]{}]{}does not admit an algorithm in time $2^{o(n^2)}$, unless ETH fails. This has consequences for designing algorithms for [[[[Edge Coloring]{}]{}]{}]{}: in order to break the barrier $2^{O(n^2)}$ one has to use methods that exploit symmetries between colors, and in particular do not apply to the list version. On the other hand, one may hope that our reductions can inspire a reduction to [[[[Edge Coloring]{}]{}]{}]{}which would exclude at least a $2^{O(n)}$-time algorithm. However it seems that [[[[Edge Coloring]{}]{}]{}]{}requires a significantly different approach. In our reductions we were able to encode information (namely, the boolean value of a variable in a satisfying assignment) in a [*color*]{} of an edge. In the case of [[[[Edge Coloring]{}]{}]{}]{}this is not possible, because one can recolor any edge $e$ by choosing an arbitrary different color $c'$ and swapping $c'$ and the color $c$ of $e$ on the maximal path/cycle that contains $e$ and has edges colored with $c$ and $c'$ only. [^1]: Institute of Informatics, Faculty of Mathematics, Informatics and Mechanics, University of Warsaw, Poland. [^2]: Work supported by the National Science Centre of Poland, grant number 2015/17/N/ST6/01224 (AS). The work of [Ł]{}. Kowalik is a part of the project TOTAL that has received funding from the European Research Council (ERC) under the European Union’s Horizon 2020 research and innovation programme (grant agreement No 677651).

ArXiv

--- abstract: 'In this **supplementary document**, we first show the shape variations of the adopted PCA model in Sec. \[shape\_pca\]. In Sec. \[full\_jacobian\], the full derivations of the analytical Jacobians of all the residuals defined in the main paper are presented. Lastly in Sec. \[more\_res\], we show more results on the KITTI dataset, which qualitatively demonstrate the ability of our method to recover the 3D poses and shapes of cars in challenging real-world environments. Apart from this document, a video showing how our method works on some selected stereo frames can be found on the project page <https://vision.in.tum.de/research/vslam/direct-shape>.' author: - 'Rui Wang$^{1}$, Nan Yang$^{1}$, Jörg Stückler$^{2}$, Daniel Cremers$^{1}$ [^1][^2]' title: | **Supplementary Material\ DirectShape: Photometric Alignment of Shape Priors\ for Visual Vehicle Pose and Shape Estimation** --- Shape Variations of PCA Model {#shape_pca} ============================= To demonstrate that our PCA model can deform and fit properly to a variety of car shapes, we first fit it to 12 selected vehicles from the CAD samples which we used to extract the PCA model. The shapes together with the color coded signed distance function (SDF) are shown in Fig. \[sdfs\]. An animation showing the different car shapes by modifying the shape coefficients $\mathbf{z}$ can be found on the project page. We further show some real-world examples in Fig. \[fig:shape\_variation\], where the PCA model is optimized using our approach to fit the corresponding cars in the second row. We claim that although the adopted PCA shape embedding is a simple linear model, it works nicely for object categories like cars. [.90]{} {width="25.00000%"} {width="25.00000%"} {width="25.00000%"} {width="25.00000%"} {width="25.00000%"} {width="25.00000%"} {width="25.00000%"} {width="25.00000%"} {width="25.00000%"} {width="25.00000%"} {width="25.00000%"} {width="25.00000%"} [.09]{} {width="80.00000%"} {width="96.00000%"} \[fig:shape\_variation\] Full Derivations of Jacobians {#full_jacobian} ============================= Jacobian of Silhouette Alignment Residual ----------------------------------------- As the relative transformation between the left and right cameras are considered to be fixed in this work, the Jacobians of $r_{silh}^{l}$ and $r_{silh}^{r}$ are the same and we will omit the superscript in the following. As shown in Eq. 2 in the main paper, the silhouette alignment residual of pixel $\mathbf{p}$ is defined as $$r_{silh}(\mathbf{p}) = -\text{log} \big( \underbrace{\pi(\mathbf{\Phi}, \mathbf{p}) p_{fg}(\mathbf{p}) + (1 - \pi(\mathbf{\Phi}, \mathbf{p})) p_{bg}(\mathbf{p})}_{\coloneqq A(\pi)} \big),$$ thus using chain rule its Jacobian with respect to the pose and shape parameters $[\boldsymbol{\xi}_{c}^{o}; \mathbf{z}]$ can be factorized to $$\begin{aligned} \mathbf{J}_{silh} & = \frac{\partial r_{silh}(\mathbf{p})}{\partial [\boldsymbol{\xi}_{c}^{o}; \mathbf{z}]} \\ & = -\frac{\partial log(A(\pi))}{\partial A(\pi)} \frac{\partial A(\pi)}{\partial \pi} \frac{\partial \pi(\mathbf{\Phi}, \mathbf{p})}{\partial \mathbf{\Phi}} \frac{\partial \mathbf{\Phi}}{\partial [\boldsymbol{\xi}_{c}^{o}; \mathbf{z}]},\end{aligned}$$ where $$\begin{aligned} \frac{\partial log(A(\pi))}{\partial A(\pi)} = \frac{1}{A(\pi)}, \label{eq:dlog_dA}\end{aligned}$$ $$\begin{aligned} \frac{\partial A(\pi)}{\partial \pi} = p_{fg}(\mathbf{p}) - p_{bg}(\mathbf{p}). \label{eq:dA_dpi}\end{aligned}$$ Recall that the shape embedding projection function $\pi(\mathbf{\Phi}, \mathbf{p})$ is defined as $$\pi(\mathbf{\Phi}, \mathbf{p}) = 1 - \prod_{\mathbf{X}_{o}} \frac{1}{e^{\mathbf{\Phi}(\mathbf{X}_{o})\zeta} + 1},$$ to make it easier to derive its Jacobian, we convert the multiplications in $\pi(\mathbf{\Phi}, \mathbf{p})$ to summations by reformulating it to $$\pi(\mathbf{\Phi}, \mathbf{p}) = 1 - \text{exp}\bigg(\underbrace{\sum_{\mathbf{X}_{o}} \text{log}\bigg(\overbrace{\frac{1} {e^{\mathbf{\Phi}(\mathbf{X}_{o})\zeta} + 1}}^{\coloneqq C(\mathbf{\Phi})}\bigg)}_{\coloneqq B(\mathbf{\Phi})} \bigg).$$ Therefore, $$\begin{aligned} \frac{\partial \pi(\mathbf{\Phi}, \mathbf{p})}{\partial \mathbf{\Phi}} = -\text{exp}(B(\mathbf{\Phi})) \sum_{\mathbf{X}_{o}} \frac{1}{C(\mathbf{\Phi})} \frac{\partial C(\mathbf{\Phi})}{\partial \mathbf{\Phi}}, \label{eq:dpi_dPhi}\end{aligned}$$ where $$\begin{aligned} \frac{\partial C(\mathbf{\Phi})}{\partial \mathbf{\Phi}} & = \frac{\partial (\frac{1}{e^{\mathbf{\Phi}(\mathbf{X}_{o})\zeta} + 1})}{\partial \mathbf{\Phi}} \\ & = (-1) \frac{e^{\mathbf{\Phi}(\mathbf{X}_{o})\zeta} }{(e^{\mathbf{\Phi}(\mathbf{X}_{o})\zeta} + 1)^2}\zeta \\ & = -\frac{\zeta e^{\mathbf{\Phi}(\mathbf{X}_{o})\zeta} }{(e^{\mathbf{\Phi}(\mathbf{X}_{o})\zeta} + 1)^2}. \label{eq:dC_dPhi}\end{aligned}$$ The remaining part to derive is $\partial \mathbf{\Phi}(\mathbf{X}_{o}) / \partial [\boldsymbol{\xi}_{c}^{o}; \mathbf{z}]$. As $\mathbf{\Phi}(\mathbf{X}_{o}) = \mathbf{V(\mathbf{X}_{o})z} + \mathbf{\Phi}_{mean} = \sum_{k=1}^{K}\mathbf{v}_{k}(\mathbf{X}_{o})z_k + \mathbf{\Phi}_{mean}$, we have $$\label{eq:dPhi_dz} \frac{\partial \mathbf{\Phi}(\mathbf{X}_{o})}{\partial \mathbf{z}} = [\mathbf{v}_{1}(\mathbf{X}_{o}), \mathbf{v}_{2}(\mathbf{X}_{o}),...,\mathbf{v}_{K}(\mathbf{X}_{o})].$$ To derive $\partial \mathbf{\Phi}(\mathbf{X}_{o}) / \partial \boldsymbol{\xi}_{c}^{o}$, we first compute the coordinate for $\mathbf{X}_{o}$ in the camera coordinate system as $\mathbf{X}_{c}$, so we have $\mathbf{X}_{o} = \text{exp}\left(\hat{\boldsymbol{\xi}_{c}^{o}}\right)\mathbf{X}_{c}$, where $\text{exp}(\hat{\cdot})$ is the exponential map that maps the twist coordinate to $\text{SE(3)}$. The remaining part of the Jacobian is then computed as $$\label{eq:dPhi_dxi} \frac{\partial \mathbf{\Phi}(\mathbf{X}_{o})}{\partial \boldsymbol{\xi}_{c}^{o}} = \nabla \mathbf{\Phi} \bigg \rvert_{\mathbf{X}_{o}} \frac{\partial \mathbf{X}_{o}}{\partial \boldsymbol{\xi}_{c}^{o}},$$ $$\label{eq:dx_dxi} \frac{\partial \mathbf{X}_{o}}{\partial \boldsymbol{\xi}_{c}^{o}} = \frac{\partial \text{exp}\left(\hat{\boldsymbol{\xi}_{c}^{o}}\right)}{\partial \boldsymbol{\xi}_{c}^{o}}\bigg \rvert_{\boldsymbol{\xi}_{c}^{o}} \mathbf{X}_{c} = \frac{\partial \text{exp}(\hat{\delta \boldsymbol{\xi}})}{\partial (\delta \boldsymbol{\xi})} \bigg \rvert_{\mathbf{0}} \text{exp}(\hat{\boldsymbol{\xi}_{c}^{o}}) \mathbf{X}_{c},$$ where $\nabla \mathbf{\Phi}$ is the spatial gradient of $\mathbf{\Phi}$, $\delta \boldsymbol{\xi}$ is a small increment in $\mathfrak{se}(3)$ and is applied with the exponential map to the left hand side of the pose estimate. The closed form solution for $\partial \text{exp}(\hat{\delta \boldsymbol{\xi}}) / \partial (\delta \boldsymbol{\xi})$ near $\delta \boldsymbol{\xi} = \mathbf{0}$ can be obtained using the infinitesimal generators of $\text{SE(3)}$ (please refer Eq. \[eq:generator1\] and \[eq:generator2\]). Depending on the derivations of the specific derivatives above, the full Jacobian of the silhouette alignment residual can be computed by combining Eq. \[eq:dlog\_dA\], \[eq:dA\_dpi\], \[eq:dpi\_dPhi\], \[eq:dC\_dPhi\] and Eq. \[eq:dPhi\_dz\], \[eq:dPhi\_dxi\], \[eq:dx\_dxi\]. Jacobian of the Photometric Consistency Residual ------------------------------------------------ As defined in the main paper, the photometric consistency residual of pixel $\mathbf{p}$ is $$r_{photo}(\mathbf{p}) = \mathbf{I}_{r}\big(\underbrace{\Pi_{c}(\mathbf{R}_{l}^{r} \Pi_{c}^{-1}(\mathbf{p}, d_{\mathbf{p}}) + \mathbf{t}_{l}^{r})}_{\coloneqq warp(\mathbf{p}, d_{\mathbf{p}})} \big) - \mathbf{I}_{l}\big(\mathbf{p}\big),$$ where the pose and shape parameters $[\boldsymbol{\xi}_{c}^{o}; \mathbf{z}]$ only appear in $d_{\mathbf{p}}$. Using chain rule the Jacobian with respect to the pose and shape parameters can be factorized to $$\begin{aligned} \mathbf{J}_{photo} & = \frac{\partial r_{photo}(\mathbf{p})}{\partial [\boldsymbol{\xi}_{c}^{o}; \mathbf{z}]} \\ & = \nabla \mathbf{I}_r(warp(\mathbf{p}, d_{\mathbf{p}})) \frac{\partial warp(\mathbf{p}, d_{\mathbf{p}})}{\partial d_{\mathbf{p}}} \frac{\partial d_{\mathbf{p}}}{\partial [\boldsymbol{\xi}_{c}^{o}; \mathbf{z}]},\end{aligned}$$ where $warp(\mathbf{p}, d_{\mathbf{p}}) = \Pi_{c}(\mathbf{R}_{l}^{r} \Pi_{c}^{-1}(\mathbf{p}, d_{\mathbf{p}}) + \mathbf{t}_{l}^{r})$ is the pixel warping function from the left image to the right image, $\nabla \mathbf{I}_r(warp(\mathbf{p}, d_{\mathbf{p}}))$ is the image gradient of the right image at the warped pixel location $warp(\mathbf{p}, d_{\mathbf{p}})$. In the following we derive $\partial warp(\mathbf{p}, d_{\mathbf{p}}) / \partial d_{\mathbf{p}}$ and $\partial d_{\mathbf{p}} / \partial [\boldsymbol{\xi}_{c}^{o}; \mathbf{z}]$ successively. Denoting the 3D coordinates of $\mathbf{p}$ in the left and the right camera coordinate systems as $\mathbf{X}_l$ and $\mathbf{X}_r$, we have $$\begin{aligned} warp(\mathbf{p}, d_{\mathbf{p}}) & = \Pi_{c}(\underbrace{\mathbf{R}_{l}^{r} \overbrace{\Pi_{c}^{-1}(\mathbf{p}, d_{\mathbf{p}})}^{\mathbf{X}_l} + \mathbf{t}_{l}^{r}}_{\mathbf{X}_r}),\end{aligned}$$ $$\begin{aligned} \mathbf{X}_l & = d_{\mathbf{p}}\mathbf{K}^{-1}[\mathbf{p}(u), \mathbf{p}(v), 1]^{\top},\end{aligned}$$ $$\begin{aligned} \mathbf{X}_r & = \mathbf{R}_{l}^{r}\mathbf{X}_l + \mathbf{t}_{l}^{r},\\ & = d_{\mathbf{p}}\underbrace{\mathbf{R}_{l}^{r}\mathbf{K}^{-1}[\mathbf{p}(u), \mathbf{p}(v), 1]^{\top}}_{\coloneqq \mathbf{v} = [\mathbf{v}(x), \mathbf{v}(y), \mathbf{v}(z)]^{\top}} + \mathbf{t}_{l}^{r}, \\ & = d_{\mathbf{p}}\mathbf{v} + \mathbf{t}_{l}^{r}, \end{aligned}$$ $$\begin{aligned} \Pi_{c}(\mathbf{X}_r) & = \begin{bmatrix} f_u & 0 & c_u \\ 0 & f_v & c_v \end{bmatrix} \begin{bmatrix} \frac{\mathbf{X}_r(x)}{\mathbf{X}_r(z)} \\ \frac{\mathbf{X}_r(y)}{\mathbf{X}_r(z)} \\ 1 \end{bmatrix} \\ & = \begin{bmatrix} f_u\frac{\mathbf{X}_r(x)}{\mathbf{X}_r(z)} + c_u \\ f_v\frac{\mathbf{X}_r(y)}{\mathbf{X}_r(z)} + c_v, \end{bmatrix},\end{aligned}$$ where $\mathbf{K} = [f_u, 0, c_u; 0, f_v, c_v; 0, 0, 1]$ is the camera intrinsic matrix. $\partial warp(\mathbf{p}, d_{\mathbf{p}}) / \partial d_{\mathbf{p}}$ therefore can be computed as $$\begin{aligned} \frac{\partial warp(\mathbf{p}, d_{\mathbf{p}})}{\partial d_{\mathbf{p}}} & = \begin{bmatrix} f_u \frac{\partial \frac{\mathbf{X}_r(x)}{\mathbf{X}_r(z)}}{d_{\mathbf{p}}} \\ f_v \frac{\partial \frac{\mathbf{X}_r(y)}{\mathbf{X}_r(z)}}{d_{\mathbf{p}}} \end{bmatrix} \\ & = \begin{bmatrix} f_u \frac{\frac{\partial \mathbf{X}_r(x)}{\partial d_{\mathbf{p}}} \mathbf{X}_r(z) - \mathbf{X}_r(x)\frac{\partial \mathbf{X}_r(z)}{\partial d_{\mathbf{p}}}}{\mathbf{X}_r^2(z)} \\ f_v \frac{\frac{\partial \mathbf{X}_r(y)}{\partial d_{\mathbf{p}}} \mathbf{X}_r(z) - \mathbf{X}_r(y)\frac{\partial \mathbf{X}_r(z)}{\partial d_{\mathbf{p}}}}{\mathbf{X}_r^2(z)} \end{bmatrix} \\ & = \begin{bmatrix} f_u \frac{\mathbf{v}(x)\mathbf{X}_r(z) - \mathbf{X}_r(x)\mathbf{v}(z)}{\mathbf{X}_r^2(z)} \\ f_v \frac{\mathbf{v}(y)\mathbf{X}_r(z) - \mathbf{X}_r(y)\mathbf{v}(z)}{\mathbf{X}_r^2(z)} \\ \end{bmatrix}.\end{aligned}$$ To compute $\partial d_{\mathbf{p}} / \partial [\boldsymbol{\xi}_{c}^{o}; \mathbf{z}]$, we first compute the 3D coordinate of the intersecting point of the ray and the zero-level surface based on $d_{\mathbf{p}}$ obtained by ray-casting, then transform it from the camera coordinate system to the object coordinate system and denote it as $\mathbf{X}_o^d$. The Jacobian with respect to the shape encoding vector is then computed as $$\frac{\partial d_{\mathbf{p}}}{\partial \mathbf{z}} = \frac{\partial d_{\mathbf{p}}}{\partial \mathbf{\Phi}} \bigg \rvert_{\mathbf{\Phi}(\mathbf{X}_o^d)} \frac{\partial \mathbf{\Phi}}{\partial \mathbf{z}} \bigg \rvert_{\mathbf{z}},$$ where $\partial \mathbf{\Phi} / \partial \mathbf{z}$ can be computed similarly as in Eq. \[eq:dPhi\_dz\], the derivation of $\partial d_{\mathbf{p}} / \partial \mathbf{\Phi}$ is illustrated in Fig. \[fig:tsdf\_derivative\]. At the intersecting point $\mathbf{X}^{d}_{o}$, the change of the depth along the ray $\partial d$ is approximately proportional to the change of the SDF value $\delta \mathbf{\Phi}$ by a factor of $1/\text{cos}(\theta)$ where $\theta$ is the angle between the ray and the surface normal. Taking the sign into account we have $$\label{eq:dd_dPhi} \frac{\partial d_{\mathbf{p}}}{\partial \mathbf{\Phi}}\bigg \rvert_{\mathbf{\Phi}(\mathbf{X}_o^d)} = -\frac{1}{\text{cos}(\theta)}.$$ The Jacobian with respect to $\boldsymbol \xi^{o}_{c}$ can be factorized to $$\frac{\partial d_{\mathbf{p}}}{\partial \boldsymbol{\xi}^{o}_{c}} = \frac{\partial d_{\mathbf{p}}}{\partial \mathbf{\Phi}} \bigg \rvert_{\mathbf{\Phi}(\mathbf{X}_o^d)} \nabla \mathbf{\Phi} \bigg \rvert_{\mathbf{X}_{o}^{d}} \frac{\partial \mathbf{X}_{o}^{d}}{\partial \boldsymbol{\xi}_{c}^{o}} \bigg \rvert_{\boldsymbol{\xi}_{c}^{o}},$$ which can be computed according to Eq. \[eq:dd\_dPhi\] and \[eq:dx\_dxi\]. Jacobian of Prior Residuals --------------------------- Based on the energy terms defined in the Eq. 7-9 in the main paper, we define the residuals of the priors on the shape and pose parameters as $$\begin{aligned} r_{shape}^{i} & = \frac{z_i}{\sigma_i}, \quad i = 1, 2,..., K \label{eq:shape}\\ r_{trans} & = \mathbf{t}_o^c(y) - g(\mathbf{t}_o^c(x, z))(y), \\ r_{rot} & = 1 - (\mathbf{R}_{o}^{c}[0, -1, 0]^{\top})^{\top} \mathbf{n}_{g}.\end{aligned}$$ {width="25.00000%"} \[fig:tsdf\_derivative\] ### Jacobian of Shape Prior Residuals. Based on Eq. \[eq:shape\] we have $$\begin{aligned} \frac{\partial r_{shape}^{i}}{\partial \boldsymbol{\xi}_{c}^{o}} & = \mathbf{0},\\ \frac{\partial r_{shape}^{i}}{\partial \mathbf{z}} & = [0,...,0, \frac{1}{\sigma_i}, 0, ..., 0].\end{aligned}$$ ### Jacobian of Translation Prior Residuals. Denoting the equation for the ground plane as $\mathbf{n}_g(x)x + \mathbf{n}_g(y)y + \mathbf{n}_g(z)z + d = 0$ with $\mathbf{n}_g$ the plane normal vector and $d$ a constant, the height of the ground plane at $\mathbf{t}_o^c(x, z)$ is $$\begin{aligned} g(\mathbf{t}_o^c(x, z))(y) = -\frac{\mathbf{n}_g(x)\mathbf{t}_o^c(x) + \mathbf{n}_g(z)\mathbf{t}_o^c(z) + d}{\mathbf{n}_g(y)},\end{aligned}$$ thus $$\begin{aligned} r_{trans} = \mathbf{t}_o^c(y) + \frac{\mathbf{n}_g(x)\mathbf{t}_o^c(x) + \mathbf{n}_g(z)\mathbf{t}_o^c(z) + d}{\mathbf{n}_g(y)}.\end{aligned}$$ Its Jacobian with respect to $\boldsymbol{\xi}_{c}^{o}$ then can be computed as $$\begin{aligned} \frac{\partial r_{trans}}{\partial \boldsymbol{\xi}_{c}^{o}} & = \frac{\partial r_{trans}}{\partial \mathbf{t}_o^c}\frac{\partial \mathbf{t}_o^c}{\partial \boldsymbol{\xi}_{c}^{o}} \\ & = [\frac{\mathbf{n}_g(x)}{\mathbf{n}_g(y)}, 1, \frac{\mathbf{n}_g(z)}{\mathbf{n}_g(y)}]\frac{\partial \mathbf{t}_o^c}{\partial \boldsymbol{\xi}_{c}^{o}},\end{aligned}$$ where the last term can be computed as $$\begin{aligned} \frac{\partial \mathbf{t}_o^c}{\partial \boldsymbol{\xi}_{c}^{o}} & = \frac{\partial \mathbf{T}_o^c(0\mathbin{:}2, 3)}{\partial \boldsymbol{\xi}_{c}^{o}} \\ & = \frac{\partial {\mathbf{T}_c^o}^{-1}(0\mathbin{:}2, 3)}{\partial \boldsymbol{\xi}_{c}^{o}} \\ & = \frac{\partial \big((exp(\hat{\delta \boldsymbol{\xi}})\mathbf{T}_c^o)^{-1}(0\mathbin{:}2, 3)\big)}{\partial (\delta \boldsymbol{\xi})} \\ & = \frac{\partial \big((\mathbf{T}_o^c exp(-\hat{\delta \boldsymbol{\xi}}))(0\mathbin{:}2, 3)\big)}{\partial (\delta \boldsymbol{\xi})} \\ & = \big(\mathbf{T}_o^c (-\frac{\partial exp(\hat{\delta \boldsymbol{\xi}})}{\partial (\delta \boldsymbol{\xi})}) \big)(0\mathbin{:}2, 3) \label{eq:generator1}\\ & = -[(\mathbf{T}_o^c \mathbf{G}_0)(0\mathbin{:}2, 3), (\mathbf{T}_o^c \mathbf{G}_1)(0\mathbin{:}2, 3),..., (\mathbf{T}_o^c \mathbf{G}_5)(0\mathbin{:}2, 3)],\label{eq:generator2}\end{aligned}$$ where we use $(0\mathbin{:}2, 3)$ to denote the operation of getting the translation part from the corresponding matrix; $\mathbf{G}_0,...,\mathbf{G}_5$ are the infinitesimal generators of SE(3). Assuming the first three elements in the twist coordinate correspond to the translation part and the last three correspond to the rotation part, the infinitesimal generators are defined as $$\begin{aligned} G_0 = \begin{bmatrix} \phantom{-}0 & \phantom{-}0 & \phantom{-}0 & \phantom{-}1\\ \phantom{-}0 & \phantom{-}0 & \phantom{-}0 & \phantom{-}0\\ \phantom{-}0 & \phantom{-}0 & \phantom{-}0 & \phantom{-}0\\ \phantom{-}0 & \phantom{-}0 & \phantom{-}0 & \phantom{-}0 \end{bmatrix},\\ G_1 = \begin{bmatrix} \phantom{-}0 & \phantom{-}0 & \phantom{-}0 & \phantom{-}0\\ \phantom{-}0 & \phantom{-}0 & \phantom{-}0 & \phantom{-}1\\ \phantom{-}0 & \phantom{-}0 & \phantom{-}0 & \phantom{-}0\\ \phantom{-}0 & \phantom{-}0 & \phantom{-}0 & \phantom{-}0 \end{bmatrix},\\ G_2 = \begin{bmatrix} \phantom{-}0 & \phantom{-}0 & \phantom{-}0 & \phantom{-}0\\ \phantom{-}0 & \phantom{-}0 & \phantom{-}0 & \phantom{-}0\\ \phantom{-}0 & \phantom{-}0 & \phantom{-}0 & \phantom{-}1\\ \phantom{-}0 & \phantom{-}0 & \phantom{-}0 & \phantom{-}0 \end{bmatrix},\\ G_3 = \begin{bmatrix} \phantom{-}0 & \phantom{-}0 & \phantom{-}0 & \phantom{-}0\\ \phantom{-}0 & \phantom{-}0 & -1 & \phantom{-}0\\ \phantom{-}0 & \phantom{-}1 & \phantom{-}0 & \phantom{-}0\\ \phantom{-}0 & \phantom{-}0 & \phantom{-}0 & \phantom{-}0 \end{bmatrix},\\ G_4 = \begin{bmatrix} \phantom{-}0 & \phantom{-}0 & \phantom{-}1 & \phantom{-}0\\ \phantom{-}0 & \phantom{-}0 & \phantom{-}0 & \phantom{-}0\\ -1 & \phantom{-}0 & \phantom{-}0 & \phantom{-}0\\ \phantom{-}0 & \phantom{-}0 & \phantom{-}0 & \phantom{-}0 \end{bmatrix},\\ G_5 = \begin{bmatrix} \phantom{-}0 & -1 & \phantom{-}0 & \phantom{-}0\\ \phantom{-}1 & \phantom{-}0 & \phantom{-}0 & \phantom{-}0\\ \phantom{-}0 & \phantom{-}0 & \phantom{-}0 & \phantom{-}0\\ \phantom{-}0 & \phantom{-}0 & \phantom{-}0 & \phantom{-}0 \end{bmatrix}. \end{aligned}$$ Lastly, the Jacobian with respect to $\mathbf{z}$ is $$\begin{aligned} \frac{\partial r_{trans}}{\partial \mathbf{z}} = \mathbf{0}.\end{aligned}$$ ### Jacobian of Rotation Prior Residuals. The rotation prior residual can be reformulated to $$\begin{aligned} r_{rot} & = 1 - (\mathbf{R}_{o}^{c}[0, -1, 0]^{\top})^{\top} \mathbf{n}_{g} \\ & = 1 - [0, -1, 0] {\mathbf{R}_o^c}^{\top}\mathbf{n}_g \\ & = 1 + [0, 1, 0] \mathbf{R}_c^o\mathbf{n}_g \\ & = 1 + \mathbf{r}_2\mathbf{n}_g,\end{aligned}$$ where $\mathbf{r}_1$ is the second row of $\mathbf{R}_c^o$. Therefore, the Jacobian with respect to $\boldsymbol{\xi}_{c}^{o}$ is $$\begin{aligned} \frac{\partial r_{rot}}{\partial \boldsymbol{\xi}_{c}^{o}} & = \frac{\partial \mathbf{r}_1}{\partial \boldsymbol{\xi}_{c}^{o}}\mathbf{n}_g \\ & = \frac{\partial \mathbf{T}_c^o(1,0\mathbin{:}2)}{\partial \boldsymbol{\xi}_{c}^{o}}\mathbf{n}_g \\ & = (\frac{\partial exp(\hat{\delta \boldsymbol{\xi}})}{\partial (\delta \boldsymbol{\xi})} \mathbf{T}_c^o)(1,0\mathbin{:}2) \mathbf{n}_g \\ & = [(\mathbf{G}_0\mathbf{T}_c^o)(1,0\mathbin{:}2),...,(\mathbf{G}_5\mathbf{T}_c^o) (1,0\mathbin{:}2)]\mathbf{n}_g,\end{aligned}$$ where $(1,0\mathbin{:}2)$ denotes the operation of getting the part corresponding to the second row of the rotation matrix. Lastly, we have $$\begin{aligned} \frac{\partial r_{rot}}{\partial \mathbf{z}} = \mathbf{0}.\end{aligned}$$ More Qualitative Results {#more_res} ======================== In Fig. \[fig1\] we qualitatively show the refinements on 3D pose and shape delivered by our method. The results on each stereo image pair are shown in each two-row block. In the first row we show the initial pose and shape estimates and our results projected onto the left image. In the second row, the initial pose (3DOP) and the estimated pose by our method are shown in the first two images, together with the ground truth 3D point cloud. In the following three images, the 3D point cloud estimated by ELAS and by our method, as well as the ground truth are shown, respectively. As shown in the results, dense stereo matching results become extremely noisy on the strong non-lambertian reflective car surfaces. Our results avoid using such results for recovering the 3D poses and shapes of cars, instead it works directly on images by performing joint silhouette and photometric alignment. While it drastically improves the 3D shape reconstruction, it can also effectively recover the 3D poses of the objects. {width="46.50000%"} {width="46.50000%"} {width="18.60000%"} {width="18.60000%"} {width="18.60000%"} {width="18.60000%"} {width="18.60000%"} {width="46.50000%"} {width="46.50000%"} {width="18.60000%"} {width="18.60000%"} {width="18.60000%"} {width="18.60000%"} {width="18.60000%"} {width="46.50000%"} {width="46.50000%"} {width="18.60000%"} {width="18.60000%"} {width="18.60000%"} {width="18.60000%"} {width="18.60000%"} {width="46.50000%"} {width="46.50000%"} {width="18.60000%"} {width="18.60000%"} {width="18.60000%"} {width="18.60000%"} {width="18.60000%"} More qualitative results are listed in Fig. \[fig:eval\_pose\_quality1\] and \[fig:eval\_pose\_quality2\], where the input images are shown on the left and our results overlaid to the input images are shown on the right. {width="44.00000%"} {width="44.00000%"} {width="44.00000%"} {width="44.00000%"} {width="44.00000%"} {width="44.00000%"} {width="44.00000%"} {width="44.00000%"} {width="44.00000%"} {width="44.00000%"} {width="44.00000%"} {width="44.00000%"} {width="44.00000%"} {width="44.00000%"} {width="44.00000%"} {width="44.00000%"} {width="44.00000%"} {width="44.00000%"} \[fig:eval\_pose\_quality1\] {width="44.00000%"} {width="44.00000%"} {width="44.00000%"} {width="44.00000%"} {width="44.00000%"} {width="44.00000%"} {width="44.00000%"} {width="44.00000%"} {width="44.00000%"} {width="44.00000%"} {width="44.00000%"} {width="44.00000%"} {width="44.00000%"} {width="44.00000%"} {width="44.00000%"} {width="44.00000%"} {width="44.00000%"} {width="44.00000%"} \[fig:eval\_pose\_quality2\] [^1]: $^{1}$R. Wang, N. Yang and D. Cremers are with the Department of Computer Science, Technical University of Munich, Garching bei München, 85748, Germany and Artisense Corporation, 350 Cambridge Avenue 250, Palo Alto, CA 94306, USA. [{wangr, yangn, cremers}@in.tum.de]{} [^2]: $^{2}$J. Stückler is with Max Planck Institute for Intelligent Systems Tübingen, Tübingen, 72076, Germany.\ [joerg.stueckler@tuebingen.mpg.de]{}

ArXiv

--- author: - 'S.-W. Wu[^1], A. Bik, Th. Henning' - 'A. Pasquali' - 'W. Brandner' - 'A. Stolte' bibliography: - 'my\_bib.bib' - 'arjanW49\_bib.bib' date: 'Received: 7 May 2014; accepted: 16 July 2014' title: 'The Discovery of a Very Massive Star in W49[^2] ' --- [Very massive stars (M &gt; 100 [M$_{\odot}$]{}) are very rare objects, but have a strong influence on their environment. The formation of this kind of objects is of prime importance in star formation, but observationally still poorly constrained.]{} [We report on the identification of a very massive star in the central cluster of the star-forming region W49.]{} [We investigate near-infrared $K$-band spectroscopic observations of W49 from VLT/ISAAC together with $JHK$ images obtained with NTT/SOFI and LBT/LUCI. We derive a spectral type of W49nr1, the brightest star in the dense core of the central cluster of W49.]{} [On the basis of its $K$-band spectrum, W49nr1 is classified as an O2-3.5If\* star with a $K$-band absolute magnitude of -6.27$\pm$0.10 mag. The effective temperature and bolometric correction are estimated from stars of similar spectral type. After comparison to the Geneva evolutionary models, we find an initial mass between 100 [M$_{\odot}$]{} and 180 [M$_{\odot}$]{}. Varying the extinction law results in a larger initial mass range of 90 - 250 [M$_{\odot}$]{}. ]{} Introduction ============ Even though very massive stars (M &gt; 100 [M$_{\odot}$]{}) are very rare, they have a strong influence on their environment via powerful winds and ionizing radiation, injecting large quantities of momentum and energy into the surrounding interstellar medium. Their fast evolution and the steep slope of the initial mass function (IMF) imply that one has to study the most massive star-forming regions to identify them. The formation mechanisms of very massive stars are by no means fully understood [@Krumholz14VMS]. For a long time, it was put in serious doubts whether these very massive stars could actually form at all. Observational evidence was presented suggesting an upper mass limit of 150 [M$_{\odot}$]{} [@Figer05]. However, recently @Crowther2010aa claimed the existence of very massive stars up to 300 [M$_{\odot}$]{}, especially in and around young massive clusters, such as NGC 3603, the Arches cluster and R136 in the Large Magellanic Cloud. In this letter we present the discovery of a very massive star in one of the most luminous Galactic regions: W49 (Fig \[W49image\]). With dozens of OB-type stars in its core, W49 is one of the most important Galactic sites for studying the formation and evolution of massive stars in the local universe [@Alves:2003aa; @Homeier05]. Given its location in the plane of the Milky Way and a distance of 11.11 $^{+0.79}_{-0.69}$ kpc [@Zhang13], W49 is optically obscured by intervening interstellar dust. This makes an optical identification and spectral classification of the stellar content close to impossible, leaving the near-infrared window (primarily $K$-band) for spectral classification of the highly obscured stars. Here, we report on the spectroscopic identification of a very massive star in W49, which we, hereafter, refer to as W49nr1. We first present our near-infrared observations (imaging and spectroscopy) of W49 (Sect. 2). The spectral features as well as the classification of W49nr1 are described in Sect. 3, where we also derive its stellar parameters like effective temperature ($T_{\mbox{eff}}$), initial mass and age. Finally, we briefly discuss the implications of our results in Sect. 4 and end with conclusions in Sect. 5. Observation and data reduction ============================== A medium-resolution (R=10,000) $K$-band spectrum of W49nr1 was obtained with ISAAC mounted on Antu (UT1) of ESO’s Very Large Telescope (VLT), Paranal, Chile. $J$- and $H$-band images were obtained with SOFI at the New Technology Telescope (NTT), La Silla, Chile and a $K$-band image was acquired with LUCI mounted on the Large Binocular Telescope (LBT), Mount Graham, Arizona. Observations ------------ SOFI $J$- and $H$-band imaging observations of W49 were performed on 2001, June 7 (PI: J. Alves) providing a $5' \times 5'$ field of view with $0.''288~pixel^{-1}$. The data were taken with a DIT (detector integration time) of 6 s and NDIT (number of integrations) of 5 per saved frame. The number of exposures for $J$ and $H$-band are 20 and 15 respectively, which lead to a total exposure time of 600 s ($J$-band) and 450 s ($H$-band). The spatial resolution is $ \sim 0.5-0.7''$. The LUCI $K$-band data were taken on 2009, September 29 with the N3.75 camera, providing a $5' \times 5'$ field of view with $0.''12~pixel^{-1}$. The spatial resolution of the $K$ image is $ \sim 0.6-0.7''$. The observations were taken with a DIT of 2 s and NDIT of 10. Forty-two frames were observed, resulting in a total exposure time of 840 s. Sky frames were taken at an offset positions centred at $\alpha (2000) = 19^h 08^m 35.8^s , \delta (2000) = +08^{\circ} 50' 52.7''$. ![$JHK$ three color image of the central area of W49. The massive star W49nr1 is indicated with a white arrow.[]{data-label="W49image"}](W49nr1){width="\hsize"} The most luminous star in the central cluster of W49, W49nr1 (Table \[tbl-1\]), was observed with ISAAC in the $K$-band on 2004, August 6 (PI: J. Alves), with 3 exposures each with a DIT of 300s. The wavelength range covered by the spectrum is 2.08 $\mu$m to 2.20 $\mu$m. The sky frame and science frames were taken with an object-sky-object pattern, and the nodding offset between the two science frames was set to 20. HR 6572, an A0V star, was used as the standard star to correct for the telluric features from the atmosphere. It was observed about one hour before the science frames, in the same wavelength range as the science observations and with an integration time of 5 s. ![Normalised $K$-band spectrum of W49nr1 with the emission lines annotated.[]{data-label="spectrum"}](spectrumone.pdf){width="\hsize"} Data reduction -------------- ### Imaging The $J$- and $H$-band images were reduced using the ESO SOFI pipeline v1.5.2. and the $K$ data with standard IRAF routines [see also @Pasquali:2011aa; @Bik14]. The images were dark and flat field corrected. For the $JH$ data, a sky frame was created from the science frames by rejecting the brightest pixels while combining the frames in pixel coordinates. A sky frame for the $K$-band data was created by combining images taken at the offset position and rejecting the 3 lowest and 6 highest values. Photometry on the $JHK$ images was performed with `DAOPHOT` [@Stetson:1987aa] under IRAF. Stars are detected with *daofind* with a threshold of 3$\sigma$ above the background. Aperture photometry was performed with *phot* in a radius of  (1-2) $\times$ the FWHM of the PSF. For each filter a reference PSF model was constructed by combining the PSF of at least 20 objects. PSF-fitting photometry was performed with *allstar*, using the PSF model to fit all objects identified with a $3\sigma$ confidence level over the local background. The $K$ image of W49 has severe nebular contamination strongly affecting the photometry of the point sources. To reduce the effect of the nebulosity in the $K$ image, we first removed the stars by means of PSF fitting. The residual frame, with all the stars subtracted, was then smoothed with a kernel of 12 pixels, resulting in a frame containing only the smooth nebular emission. This smoothed frame is subtracted from the original frame. After that, we performed PSF photometry on the nebular subtracted image, resulting in a more accurate photometry. W49nr1 is located in the center of a compact cluster and its photometry is affected by crowding from the neighbouring stars. To quantify the effect of the crowding, we performed aperture photometry at the position of W49nr1 on the residuals in the psf-subtracted frame. This gives an error of 15.8, 6.4 and 8.6 % for $J$, $H$ and $K$ respectively. Finally, we cross-matched the obtained catalogs for each filter to identify the sources detected in more than one band. We calibrated the SOFI and LUCI photometry with 2MASS [@Skrutskie:2006aa]. The final calibration resulted in errors in the zero points of 0.0063, 0.0071 and 0.0055 mag for $J$, $H$ and $K$-band respectively. We did not find a color dependence of the derived zero points. The final errors of the $JHK$ photometry, as listed in Table \[tbl-1\], are a combination of the photometry uncertainty, errors in the zero points and the errors due to crowding. -------------------------------- ------------------------------------------------ $\alpha$(J2000) (h m s) 19:10:17.43 $\delta$(J2000) ($^\circ$   ) +9:06:20.93 $J$ (mag) 16.57$\pm0.18$ $H$(mag) 13.47$\pm0.12$ $K$ (mag) 11.93$\pm0.10$ EW(Br$\gamma$) (Å) 8.2 $\pm$ 1.7 EW() (Å) 2.4 $\pm$ 0.7 EW() (Å) 2.3 $\pm$ 1.0 EW() (Å) 2.6 $\pm$ 0.9 Spectral type O2-3.5If\* $T_{\mbox{eff}}$ (K) 40,000 – 50,000 BC (mag) -5.2 – -4.55 $A_{\mbox{K}}$ (mag) 2.9/2.6 - 3.5 Initial mass ([M$_{\odot}$]{}) 100 – 180/90 – 250 Luminosity (L/L$_{\sun}$) 1.7 - 3.1$\times 10^6$/ 1.2 - 4.9$\times 10^6$ -------------------------------- ------------------------------------------------ : Observed and derived properties of W49nr1[]{data-label="tbl-1"} ### Spectroscopy The ISAAC observations of W49nr1 were reduced using standard IRAF routines. The wavelength calibration was performed using the Xe and Ar arc frames. After the flat fielding and wavelength calibration, the sky was removed by subtracting the frames taken at the A and B nodding position. The spectra were extracted using *doslit* and the different exposures are combined to one final spectrum. To remove the narrow $Br\gamma$ emission from the diffuse nebular emission surrounding the cluster, the background was estimated using a Legendre function, sampling a region close to the star, and subtracted from the spectrum. The spectrum of the telluric standard star was reduced in the same way as the spectrum of W49nr1. Before correcting the spectrum of W49nr1 with the standard star, the Br$\gamma$ line of the standard star was removed by fitting a Lorentzian profile. The resulting atmospheric transmission spectrum was used to correct the science spectrum for telluric absorption using the IRAF task *telluric*. The signal-to-noise ratio (SNR) of the final spectrum is $\sim$90. Results ======= Spectral classification of W49nr1 --------------------------------- The final, normalized $K$-band spectrum of W49nr1 is shown in Fig. \[spectrum\]. The spectrum is dominated by broad emission lines of Br$\gamma$ (2.16 [$\mu$m]{}), (2.189 [$\mu$m]{}), (2.116 [$\mu$m]{}) and (2.10 [$\mu$m]{}). The narrow emission component of $Br\gamma$ is a residual of the nebular subtraction. The and lines are indicative of an early spectral type [@Hanson05]. The broad emission profiles imply an origin in the stellar wind. These properties suggest similarities with the spectral classes O2-3.5If\*, O2-3.5If\*/WN5-7 (“slash” stars) and WN5-7 stars [@Crowther:2011aa]. The sum of the equivalent widths (EWs) of Br$\gamma$ and can be used as a discriminator between these classes. For the WN5-7 stars, the summed EWs are expected to be above 70 Å, while O2-3.5If\* stars have a total EW between 2 and 20 Å, with the “slash” stars lying in between. The total EW of both lines for W49nr1 (Table \[tbl-1\]) is (10.6 $\pm$ 1.8 Å), resulting in a classification of W49nr1 as O2-3.5If\*. Hertzsprung-Russell diagram --------------------------- Based on the classification of W49nr1 as an O2-3.5If\* star, we estimated $T_{\mbox{eff}}$ between 40,000 and 50,000 K and the bolometric correction (BC$_{K}$) between -5.2 and -4.55 mag adopting the derived values for an O2f\*, O3I and an O4I star as representative for this class [@Crowther:2011aa]. From our $HK$ photometry the absolute $K$-band magnitude was derived to be -6.27 $\pm$ 0.1 mag by assuming the distance of 11.11 kpc [@Zhang13], applying the extinction law of @Indebetouw05 and adopting the intrinsic color of $(H-K)=-0.1$ mag for O3I stars from @Martins:2006aa. After applying the BC$_{K}$, the bolometric magnitude of W49nr1 was derived to be between -11.47 and -10.82 mag, and the corresponding bolometric luminosity between $1.7\times 10^6$ and $3.1\times 10^6$ L$_{\sun}$. We plotted the likely parameter space of W49nr1 in the Hertzsprung Russel diagram (HRD) as shown in Fig. \[HRD\]. As $T_{\mbox{eff}}$ and BC$_{K}$ are correlated, the likely location of W49nr1 is a diagonal ellipse. The possible locations of W49nr1 in the HRD was estimated by calculating the luminosity for the three spectral types in this class (O2If\*, O3I and O4I), using their $T_{\mbox{eff}}$ and corresponding BC$_{K}$. To estimate the initial mass and age of W49nr1, the likely parameter space in the HRD was compared with the Geneva theoretical stellar evolution models [@Ekstrom:2012aa; @Yusof:2013aa], using models with and without stellar rotation. From the evolutionary tracks, the initial mass of W49nr1 was estimated to be in the range between $\sim$110 and $\sim$180 [M$_{\odot}$]{} for models without rotation and between $\sim$100 and $\sim$170 [M$_{\odot}$]{} for models with rotation (Fig. \[HRD\], left panel, vertically hashed area). While the initial mass estimate for W49nr1 is insensitive to rotation, the isochrones for the models with and without rotation for the same age are very different (Fig. \[HRD\], right panel). The position of W49nr1 suggests an upper age limit of $\sim$2 Myrs after comparison with the “non-rotating” isochrones, however, considering the models with rotation, an age between 2 and 3 Myrs is more likely. As the extinction towards W49nr1 is high ($A_{\mbox{K}} = 2.9$ mag) the choice of the extinction law can have a large effect on the derived luminosity and therefore on its initial mass and age. To select the best fitting extinction laws, we applied a similar analysis to the color-color diagram of W49 as @Bik12 and found that the slopes of the @Cardelli89 and @Roman07 laws were not consistent with the observations (Wu et al, in prep). The extintion law of @Indebetouw05 was the best fitting law, but also the slopes of @Fitzpatrick99, @Nishiyama09 and @Rieke85 are consistent with the observed colors. Taking into account all the 4 extinction laws, the estimated initial mass range widens to 90 - 250 [M$_{\odot}$]{}(see Fig. \[HRD\]). ![HRD with the possible location of W49nr1 marked as a vertically hashed area and horizontally hashed, taking into account different extinction laws. The three black filled circles stand for an O2If\*, an O3I and an O4I star respectively. *Left panel:* The Geneva evolution tracks [@Ekstrom:2012aa; @Yusof:2013aa] without stellar rotation (solid line) and with rotation (dashed line) for different masses are over plotted. *Right panel:* The main sequence isochrones with different ages, again without stellar rotation (solid line) and with rotation (dashed line).[]{data-label="HRD"}](HRDtrack.pdf){width="\hsize"} ![HRD with the possible location of W49nr1 marked as a vertically hashed area and horizontally hashed, taking into account different extinction laws. The three black filled circles stand for an O2If\*, an O3I and an O4I star respectively. *Left panel:* The Geneva evolution tracks [@Ekstrom:2012aa; @Yusof:2013aa] without stellar rotation (solid line) and with rotation (dashed line) for different masses are over plotted. *Right panel:* The main sequence isochrones with different ages, again without stellar rotation (solid line) and with rotation (dashed line).[]{data-label="HRD"}](HRDiso.pdf){width="\hsize"} Discussion and future prospectives ================================== In this letter we report the discovery of a very massive star in the center of the main cluster in W49. In the following we discuss the uncertainties in the derivation of the stellar parameters and the implications for the properties of the central cluster in W49. We end with a suggestion for further characterization of this object. Stellar paramaters ------------------ Our classification of W49nr1 depends on the empirical relation between the spectral type and the equivalent width of the emission lines as well as the calibration of $K$-band bolometric corrections for early-O stars based upon atmosphere models derived by @Crowther:2011aa. Due to the very few objects used in this study, it is hard to predict the uncertainty of this classification and a larger number of stars is needed to make this calibration more reliable. The evolution of the very massive stars is mostly governed by their stellar wind and mass-loss rate. These input parameters for stellar evolution models add uncertainties to the estimated initial mass and age. As a comparison to the Geneva models we use the relation between the luminosity and the maximum stellar mass for homogeneous hydrogen burners [@Grafener11], resulting in a present-day mass estimate of 110 - 175 [M$_{\odot}$]{} (and 95 - 250 [M$_{\odot}$]{} for taking into account all 4 extinction laws as discussed in Sect. 3.2). The stellar rotation only plays an important part in estimating the age of W49nr1 from the HRD, as the rotational models predict longer time scales for the different evolutionary phases of the massive stars. The $K$-band spectrum is fully dominated by emission lines originating in the stellar wind, hence no estimate of the rotation can be made. High resolution spectroscopy of possible absorption lines to derive its rotation is key to understand the evolutionary status of this extreme star as well as the cluster. By monitoring the radial velocity of the emission lines, multi-epoch spectroscopy could probe for a possible binary nature. Cluster properties ------------------ W49nr1 is located in the center of the compact central cluster in W49 (Fig. \[W49image\]), and thus supports the theoretical expectation of rapid dynamical mass segregation (e.g. @Allison:2009aa). @Homeier05 estimate the mass of this cluster as $10^4$ [M$_{\odot}$]{}. This suggests that W49nr1 is located in an environment quite similar to other very massive stars located in [@Crowther2010aa]. It adds to the growing number of stars with initial masses at or above 150 [M$_{\odot}$]{}, suggesting the absence of a strict upper mass limit for massive stars as also suggested by numerical simulations [@Kuiper:2010aa; @Kuiper:2011aa]. Comparing the cluster mass and the derived stellar mass for W49nr1 to theoretical relations between cluster mass and mass of the most massive star [@Weidner10] shows that the presence of such a massive star is consistent with a normal initial mass function. A full study of the upper end of the IMF requires a spectroscopy classification of many more massive stars (Wu et al, in prep). Conclusions =========== In this letter we present $JHK_s$ imaging and $K$-band spectroscopy observations of W49nr1, the brightest star in the central cluster of W49. According to classification criteria based on the equivalent widths of Br$\gamma$ and given by @Crowther:2011aa, W49nr1 is classified as an O2-3.5If\* star. We estimate the effective temperature to be between 40,000 and 50,000 K and the bolometric luminosity between $1.7\times 10^6$ and $3.1\times 10^6$ L$_{\sun}$. Comparison with the Geneva stellar evolutionary tracks suggests an initial mass range of 100 - 180 [M$_{\odot}$]{} in the case of a single star, relatively independent of rotational velocity. We study the effect of variations in the extinction law on the stellar parameters, resulting in a large initial mass range of 90 - 250 [M$_{\odot}$]{}. Estimates of the present day mass delivers similar values. The age depends severely on rotational velocity and can only be constrained to less than 3 Myrs. The next step will be a full spectroscopic modelling of the near-infrared spectrum of W49nr1 resulting in stricter constraints on the effective temperature and luminosity. Spectral modelling will allow us to identify possible absorption lines at other wavelengths, suitable for measuring the rotational velocity. We thank the anonymous referee for helpful suggestions which have improved the paper significantly. We acknowledge Fabrice Martins and Adrianne Liermann for extensive discussions on the interpretation of the K-band spectrum. A.B. acknowledges the hospitality of the Aspen Center for Physics, which is supported by the National Science Foundation Grant No. PHY-1066293. The LBT is an international collaboration among institutions in Germany, Italy, and the United States. LBT Corporation partners are LBT Beteiligungsgesellschaft, Germany, representing the Max Planck Society, the Astrophysical Institute Potsdam, and Heidelberg University; Istituto Nazionale di Astrofisica, Italy; The University of Arizona on behalf of the Arizona University system; The Ohio State University, and The Research Corporation, on behalf of The University of Notre Dame, University of Minnesota and University of Virginia [^1]: International Max Planck Research School for Astronomy and Cosmic Physics at the University of Heidelberg (IMPRS-HD) [^2]: Based on observations made with ESO Telescopes at the La Silla Paranal Observatory under programme IDs 67.C-0514 and 073.D-0837) and on data acquired using the Large Binocular Telescope (LBT).

ArXiv

--- abstract: '=-1 The entropic brain hypothesis holds that the key facts concerning psychedelics are partially explained in terms of increased entropy of the brain’s functional connectivity. Ayahuasca is a psychedelic beverage of Amazonian indigenous origin with legal status in Brazil in religious and scientific settings. In this context, we use tools and concepts from the theory of complex networks to analyze resting state fMRI data of the brains of human subjects under two distinct conditions: (i) under ordinary waking state and (ii) in an altered state of consciousness induced by ingestion of Ayahuasca. We report an increase in the Shannon entropy of the degree distribution of the networks subsequent to Ayahuasca ingestion. We also find increased local and decreased global network integration. Our results are broadly consistent with the entropic brain hypothesis. Finally, we discuss our findings in the context of descriptions of “mind-expansion” frequently seen in self-reports of users of psychedelic drugs.' author: - 'A. Viol' - 'Fernanda Palhano-Fontes' - Heloisa Onias - 'Draulio B. de Araujo' - 'G. M. Viswanathan' bibliography: - 'entropy\_aya10.bib' title: | Shannon entropy of brain functional complex networks\ under the influence of the psychedelic Ayahuasca --- =-1 Relatively little is known about how exactly psychedelics act on human functional brain networks. During the last few years, new neuroimaging techniques, such as functional magnetic resonance imaging (fMRI) [@buxton2009introduction; @Heeger2002], have allowed noninvasive investigation of global brain activity in a variety of conditions, e.g., under anaesthesia, sleep, coma, and in altered states of consciousness induced by psychedelic drugs [@Schrouff2011; @schoter2012; @Noirhomme2010; @Andrade2011; @draulio2012; @carhatharris2012; @Carhart-Harris2016; @fernanda2014]. Recently, Carhart-Harris [*et al.*]{} proposed a hypothesis known as the [*entropic brain*]{}, which holds that the stylized facts concerning altered states of consciousness induced by psychedelics can be partially explained in terms of higher entropy of the brain’s functional connectivity [@Carhart-Harris2014]. Although the entropy of the brain has never been directly measured, the entropic brain hypothesis is empirically supported by several recent studies. For example, Sarasso [*et al.*]{} have reported complex spatiotemporal cortical activation pattern during anesthesia with ketamine, which can induce vivid experiences (“ketamine dreams”) [@sarasso-xxx]. Similarly, Petri [*et al.*]{} found that after administration of the psychedelic psilocybin, the brain’s functional patterns undergo a dramatic change characterized by the appearance of many transient low-stability structures [@petri-xxx]. Perhaps the most convincing evidence supporting the hypothesis thus far has come from the study undertaken by Tagliazucchi [*et al.*]{} [@tagliazicchi2014], who reported a larger repertoire of brain dynamical states during the psychedelic experience with psilocybin. They inferred an increase in the entropy of the functional connectivity in several regions of the brain, by studying the temporal evolution (i.e., dynamics) of the connectivity graphs. Here we directly measure increases in entropy associated with the functional connectivity of the whole brain under the influence of a psychedelic. Specifically, we analyze fMRI functional connectivity of human subjects before and after they ingest the psychoactive brew Ayahuasca and report an increase in the Shannon entropy. This is the first time that the entropy of the functional networks of the human brain has been directly measured in altered states of mind on a global scale, i.e. considering the entire brain. =‘ Ayahuasca is a beverage of Amazonian indigenous origin and has legal status in Brazil in religious and scientific settings [@labate2014prohibition]. It contains the powerful psychedelic $N,N$- dimethyltryptamine (DMT), together with harmala alkaloids that are known to be monoamine oxidase inhibitors (MAOIs). The beverage is typically obtained by decoction of two plants from the Amazonian flora: the bush [*Psychotria viridis*]{}, that contains DMT, and the liana [*Banisteriopsis caapi*]{}, that contains MAOIs [@mckenna2004]. DMT is usually rapidly metabolized by monoamine oxidase (MAO), but the presence of MAOI allows DMT to cross the blood-brain barrier and to exert its effects [@shanon2002antipodes; @huxley2004doors; @hofmann1983lsd; @Hollister1962235; @grof1980lsd; @Griffiths2006]. Similar to LSD, mescaline and psilocybin [@shanon2002antipodes; @huxley2004doors; @hofmann1983lsd; @Hollister1962235; @grof1980lsd; @Griffiths2006], Ayahuasca can cause profound changes of perception and cognition, with users reporting increase of awareness, flexible thoughts, insights, disintegration of the self, and attentiveness [@shanon2002antipodes; @riba2001]. There is growing interest in Ayahuasca, partially due to recent findings showing that it may be effective in treating mental disorders, such as depression [@Sanches2016] and behavioral addiction [@OSORIO2015; @labate2013therapeutic; @Nunes2016]. Similar therapeutic potential has also been pointed out for other psychedelics [@krebs2012lysergic; @grof1980lsd; @johnson2014pilot; @albaugh1974peyote; @frederking1955intoxicant; @Carhart-Harris2016]. =-1 For analysis we use tools and concepts from the field of complex networks, a brief history of which follows. The application of graph theory to phase transitions and complex systems led to significant progress in understanding a variety of cooperative phenomena over a period of several decades. In the 1960s, the books by Harary, especially [*Graph Theory and Theoretical Physics*]{} [@harary1967graph], introduced readers to powerful mathematical techniques. The chapter by Kastelyn, still considered to be a classic, showed that difficult combinatorial problems of exact enumeration could be attacked via graph theory, including the exact solution of the two-dimensional Ising model (e.g., see Feynman [@feynman1998statistical]). In the 1980s, certain families of neural network models were shown to be equivalent to Ising systems, e.g., the Hopfield network [@hopfield1982] is a content-addressable memory which is isomorphic to a generalized Ising model [@amit1992modeling]. Beginning in the 1990s, new approaches to networks, giving emphasis to concepts such as the node degree distribution, clustering, assortativity, small-worldliness and network efficiencies, led eventually to what has become [the new field of complex networks [@Barabasi2002; @newman2010networks]. These new tools and concepts]{} [@Albert02; @Bornholdt2003; @caldarelli2013scale] have found successful application in the study of diverse phenomena, such as air transportation networks [@Verma2014], terrorist networks [@terror], gene regulatory networks [@Magtanong2011a], and functional brain networks [@sporns2016networks; @Bullmore2009; @mckenna1994; @rubinov2010; @meunier2010]. We approach the human brain from this perspective of [complex networks]{} [@braincomplexsystem2; @braincomplexsystem3]. Ten healthy volunteers were submitted to two distinct scanning sessions: (i) before and (ii) 40 minutes after Ayahuasca intake, when the subjective effects are noticeable. In both cases, participants were instructed to close their eyes and remain awake and at rest, without performing any task. We performed a standard preprocessing on all samples of the fMRI data (see Methods for details concerning data acquisition and preprocessing). Data analysis consists of two main steps. In the first step, we use fMRI data to generate complex networks to represent the actual functional brain connectivity patterns. In the second step, we use the networks generated in step 1 as inputs and calculate network characteristics as output, using techniques from the theory of complex networks. The Methods section describes both steps in detail. More information about most of the methods used here can be found in refs. [@onias2014; @schoter2012; @liuyong2008]. Figure \[braindegree\] shows the networks generated from one subject before and after Ayahuasca intake, for one specific choice of mean node degree. The spheres represent nodes, with sphere size proportional to the degree of the node. The lower plots show histograms of node degrees. The main result that we report here is an increase in the Shannon entropy of the degree distribution for the functional brain networks subsequent to Ayahuasca ingestion. We also find that the geodesic distance increases during the effects of Ayahuasca, i.e. qualitatively the network becomes “larger.” More generally, we also find that these functional brain networks become less connected globally but more connected locally. The key technical innovation is the measurement of the Shannon entropy of the degree distribution of the complex networks that represent the functional connectivity of the human brain. This novel use of the Shannon entropy allows the brain to be studied from the perspective of information theory in a manner previously unexploited. Moreover, the Shannon entropy is also very closely related to the Boltzmann-Gibbs entropy used in statistical mechanics. Hence, our approach to studying the brain experimentally is grounded in two strong theoretical traditions: graph theory and complex networks on the one hand, and information theory and statistical physics on the other. [ Our study also represents a significant advance for the following additional reasons: (i) our results unveil how Ayahuasca (and likely most other tryptamine psychedelics) alter brain function, both locally and globally; (ii) it is the first time that this specific approach has been applied to characterize functional brain networks in altered states of consciousness; (iii) our study of Ayahuasca covers all brain regions; and (vi) the method we have developed can be immediately applied to study a variety of other phenomena (e.g., the effects of medication for mental health disorders). ]{} Results {#sec-results .unnumbered} ======= Increase of the Shannon entropy of the degree distributions {#increase-of-the-shannon-entropy-of-the-degree-distributions .unnumbered} ----------------------------------------------------------- We find evidence of significant changes in the functional brain networks of subjects before and after ingestion of Ayahuasca. Figure \[moment\] shows 2nd as well as 4th central moments of the degree distributions for each subject. The individual values are calculated separately for each network. We find an increase of variance for all subjects after Ayahuasca intake and a decrease of kurtosis for almost all of them (6 subjects). These findings indicate that the degree distributions become less peaked and wider. This behavior is suggestive of an increase of the Shannon entropy for the degree distributions after Ayahuasca ingestion. Figure \[entropy\] shows the average Shannon entropy of the degree distributions as a function of mean degree, considering networks from all subjects, before and after Ayahuasca intake. A fair comparison of the “before” and “after” networks is possible by considering the entropy of networks of identical mean degree. We find an increase in the entropy of the degree distributions after Ayahuasca ingestion. In order to better evaluate the consistency of this result, we also calculate the average Shannon entropy subject-by-subject, before and after Ayahuasca (Figure \[boxentropy\]). We find significant increased entropy for all individual subjects. Iso-entropic randomized networks {#iso-entropic-randomized-networks .unnumbered} -------------------------------- The degree distribution does not completely define a network, however it can have great influence over other network properties. One can quantify this influence by comparing any given network $G$ to other networks chosen randomly from the ensemble of networks that have exactly the same degree distribution. We refer to such networks as “randomized networks.” By definition, all such randomized networks have the same entropy as the original network $G$, i.e. they are iso-entropic to $G$. An efficient way of generating such randomized networks is the Maslov algorithm [@Maslov2002] (see Methods). Whereas entropy is conserved by the Maslov algorithm, the clustering coefficient, geodesic distances and efficiencies are not. By comparing these non-conserved quantities before and after randomization, we can distinguish effects that are due solely to changes in the degree distribution from those that are sensitive to how links are more specifically arranged. We generate a set of 30 iso-entropic randomized networks for each original network, for all subjects both before and after Ayahuasca ingestion. Comparison of the original networks with the randomized networks yields important information concerning to what degree the changes in quantities such as geodesic distance, clustering coefficients, and global and local efficiencies can be accounted for by the changes in the degree distributions (see results described below). Decrease of global integration {#decrease-of-global-integration .unnumbered} ------------------------------ Figure \[distance\] shows an increase of mean geodesic distance and a decrease of global efficiency after Ayahuasca ingestion. To determine how much of the change in geodesic distance is due to the change in the degree distribution, we also calculated the geodesic distance and global efficiency for the iso-entropic randomized networks. Note how the values for those networks are quite different compared to the non-randomized networks. We conclude that the change in degree distribution cannot explain the entire change in geodesic distance. The inset in the middle panels shows the change in the normalized mean geodesic distance and global efficiency, which we define as the ratio $D/D_{\mbox{\tiny rand}}$ and similarly for the global efficiency (see [@Maslov2002; @schoter2012]) . We see, indeed, that these ratios are not close to zero. If the change in degree distribution could account for all the change in geodesic distance and efficiency, then the change in these ratios would be close to zero. Significant changes are also observed at the individual level and are again consistent for all subjects (Figure \[distance\] (e) and \[distance\] (f)). Increase of local integration {#increase-of-local-integration .unnumbered} ----------------------------- Figure \[clustering\] shows an increase of clustering coefficients and local efficiency after Ayahuasca ingestion. In contrast to the behavior of the metrics discussed above, almost identical changes are seen for iso-entropic networks. This result indicates that the variation in degree distribution can account for most of the change in clustering and local efficiency. The insets in the middle panel show the change in the normalized clustering and local efficiency, which we define as the ratio $C/C_{\mbox{\tiny rand}}$ and similarly for the local efficiency (see [@Maslov2002; @schoter2012]). We see, indeed, that these ratios are close to zero. Discussion {#discussion .unnumbered} ========== Our results reveal some remarkable findings, the most important of which is that the entropy increases after Ayahuasca ingestion. The following also increase: geodesic distance, clustering coefficient and local efficiency. However, the global efficiency decreases. Overall, we find an increase of local integration and a decrease of global integration in the functional brain networks. We interpret these findings in the context of some well understood prototypical classes of networks. Regular lattices have fixed coordination number, hence all nodes have the same degree and the Shannon entropy of the degree distribution is thus zero. In contrast, the entropy is high in networks with broad distributions of degree. In the context of the Watts-Strogatz model [@WS1998], clustering and geodesic distance both decrease when highly regular networks are transformed into small-world networks by randomly re-assigning the links. [Whereas clustering and geodesic distances decrease with increasing randomness in such models, we find the opposite behavior for Ayahuasca, i.e., randomness as measured by the Shannon entropy of the node degree distribution increases in parallel with clustering and geodesic distances.]{} Hence, our findings cannot be reduced to simple explanations of greater or lesser randomness. Locally, there is an increase in integration (as measured by network efficiency), but globally there is a decrease in integration. Indeed the increase of geodesic distance and decrease of global efficiency after Ayahuasca intake signify that the functional brain networks are less globally integrated. One possible interpretation of these findings is that the increase of local robustness and the decrease of global integration reflect a variation in modular structure of the network. Recent studies have reported the presence of modularity in functional brain networks on several scales [@ferrarini2009; @meunier2010; @Nicolini2016]. Modular networks are characterized by the existence of reasonably well-defined subnetworks in which internal connections are denser than connections between distinct subnetworks [@meunier2010]. However, traditional algorithms [@Newman2004; @Guimera2005a; @Blondel2008] were not able to detect variation on modular structure features between our sets of networks. Our results are broadly consistent with the entropic brain hypothesis, hence we discuss the latter in the context of our findings. The hypothesis maintains that the mental state induced by psychedelics, which the original authors term “primary-state,” presents relatively elevated entropy in some features of brain organization, compared to the ordinary waking state (termed “secondary”) [@Carhart-Harris2014]. Although it may be somewhat counter-intuitive that the psychedelic state is considered primary while ordinary consciousness is secondary, their hypothesis is inherently plausible considering that a wider spectrum of experiences is possible with psychedelics than in ordinary consciousness. In this sense, ordinary consciousness can be thought of as a restriction or constrained special case of a more primary consciousness. The hypothesized lower entropy of ordinary consciousness relative to primary consciousness is attributed to this reduction of freedom. In fact, the idea that ordinary consciousness is not primary was previously put forth by Alan Watts to describe what later became widely known as [*mindfulness*]{} [@livro_do_watts] Indeed, it is possible to interpret the effects of Ayahuasca, and other psychedelics, as being due to the temporary removal of the some of the restrictions that are necessary for sustaining ordinary (adult trained) consciousness. Without these restrictions, the mind reverts to the more flexible state, in which self-referential narratives and thoughts about the past or the future are no longer experienced as identical to the reality that they are assumed to represent [@livro_do_watts]. Relatively few studies have investigated entropy in brain functional networks, hence additional comments are in order. Tagliazucchi [*et al.*]{} [@tagliazicchi2014] showed that psilocybin (psychedelic present in some species of mushrooms) may be responsible for increases of a different entropy measure in functional connectivity of the 4 regions of Default Mode Network (DMN), a relevant functional network related to resting state. Recently, Yao [*et al.*]{} [@Yao2013a] correlated entropy increases in the human brain with age. This study also supports the view that entropy is correlated to [brain function (and perhaps also its development).]{} Moreover, in agreement with our results, a study by Schroter [*et al.*]{} [@schoter2012] similarly suggests that functional network topology may have a central role in consciousness quality. They investigated the effects on the human brain of the anesthetic propofol, which can induce loss of consciousness [@Sarasso2015]. They reported a decrease of the clustering coefficient, which is strongly influenced by degree distribution (however, geodesic distance remained unchanged). We briefly comment on the limitations of our method: (i) the reduced number of subjects and the fact that all of them were experienced with Ayahuasca do not allow population inferences and do not elucidate whether the effects observed here were only due the acute administration or if previous experience also played a significant role; (ii) expectancy and suggestion were not controlled, as placebo was not used; (iii) networks were built upon a number of critical choices, such as the atlas used to partition the brain, the method used to build the correlation matrix, and the cutoff thresholds for generating the adjacency matrices from correlation matrices [@Smith20121257; @langer2013], which may affect the final results; (iv) the chosen range of correlation values automatically limits the networks’ behavior to a small-world network. Despite this limitation, it is important to highlight that several studies have consistently demonstrated that brain networks exhibit a small-world behavior [@bassett2006]. Finally, we speculate about whether or not our finding of larger mean geodesic distances may have any relation to self-reports of “mind-expansion” by users of psychedelics. Could there be a direct relation between entropy increases and the higher creativity reported by users of psychedelics? Such questions merit further investigation. In conclusion, our results are broadly consistent with the hypothesis that psychedelics increase the entropy in brain functions. By calculating the Shannon entropy of the degree distribution of complex networks generated from fMRI data, we have taken a new low-computational-cost approach to investigating brain function under the influence of psychedelics. Methods {#sec-methods .unnumbered} ======= Data acquisition and preprocessing {#sec-data .unnumbered} ---------------------------------- The fMRI images were obtained in a 1.5 T scanner (Siemens, Magneton Vision), using an EPI-BOLD like sequence comprising 150 volumes, with the following parameters: TR=1700 ms; TE=66 ms; FOV=220 mm; matrix 64$\times$64; voxel dimensions of 1.72mm$\times$1.72mm$ \times $1.72 mm. It also was acquired whole brain high resolution T1-weighted images (156 contiguous sagittal slices) using a multiplanar reconstructed gradient-echo sequence, with the following parameters: TR=9.7 ms; TE=44 ms; flip angle 12$^{\circ}$; matrix 256$ \times $256; FOV= 256 mm, voxel size$ = 1mm \times 1mm \times 1 mm$. The images were obtained from 10 healthy right-handed adult volunteers (mean age 31.3, from 24 to 47 years), all who were experienced users of Ayahuasca with at least 5 years use (twice a month) and at least 8 years of formal education. The experimental procedure was approved by the Ethics and Research Committee of the University of São Paulo at Ribeirão Preto (process number 14672/2006). Written informed consent was obtained from all volunteers, who belonged to the Santo Daime religious organization. Volunteers were not under medication for at least 3 months prior to the scanning session and were abstinent from caffeine, nicotine and alcohol prior to the acquisition. They had no history of neurological or psychiatric disorders, as assessed by DSM-IV structured interview [@american2000diagnostic]. Subjects ingested 120-200 mL (2.2 mL/kg of body weight) of Ayahuasca known to contain 0.8 mg/mL of DMT and $0.21$ mg/mL of harmine. Harmaline was not detected via the chromatography analysis, at the threshold of 0.02 mg/mL [@draulio2012]. preprocessing steps were conducted in FSL (http://www.ndcn.ox.ac.uk/divisions/fmrib) and include: slice-timing correction, head motion correction and spatial smoothing (Gaussian kernel, FWHM = 5 mm). One volunteer was excluded from analysis due to excessive head movement (more than 3mm in some direction), leaving 9 participants (5 women) to our analysis. All images were spatially normalized to the Montreal Neurologic Institute (MNI152) [@Brett2002a] standard space, using a linear transformation. We also evaluated 9 regressors of non-interest using a General Linear Model (GLM): 6 regressors to movement correction, 1 to white matter signal, 1 to cerebrospinal fluid and 1 to global signal. Each volunteer was submitted to fMRI scanning under two distinct conditions: (i) before and (ii) 40 minutes subsequent to Ayahuasca intake. In both cases, volunteers were in an awake resting state: they were requested to stay lying with eyes closed, without performing any task. One volunteer sample was excluded from analysis due to excessive head movement, leaving 9 participants (5 women) to our analysis. Complex network metrics {#complex-network-metrics .unnumbered} ----------------------- For a detailed overview of complex network theory, we refer readers to refs. [@Bornholdt2003; @Barabasi2002; @rubinov2010]. Each element of a network is known as a node (or vertex), and the relation between a pair of nodes is represented by a connecting link (or edge). Links can have weights associated with them and can be directed or undirected (or, equivalently bi-directional). Nodes connected by a single link are known as nearest neighbors [@newman2010networks]. Non-weighted undirected networks, i.e. those with symmetric and unweighted links are isomorphic to a binary symmetric matrix known as the adjacency matrix. When a pair of nodes $i$ and $j$ are neighbors, the adjacency matrix element is $a_{i,j}=1$ and $a_{i,j}=0$ otherwise. Standard quantities of interest that help to characterize the topology and complexity of networks [@onias2014; @rubinov2010] include node degree, geodesic distance, clustering coefficient, and local and global network efficiencies. Definitions: \(i) The degree $k_j$ of a node $j$ is the number of links that it has with other nodes. The degree distribution of a network is the normalized histogram of degrees over all nodes. \(ii) A geodesic path between two nodes is the shortest path from one to the other, assuming such a path exists. The geodesic distance $d_{i,j}$ between nodes $i$ and $j$ is the number of links in the geodesic path. If there is no such path, the geodesic distance is defined as infinite. Given a network $G$ with $N$ nodes, the mean geodesic distance is given by $$D(G)=\frac{1}{N(N-1)}\sum_{i\neq j}{d_{i,j}} ~. \label{distance} $$ \(iii) The clustering coefficient quantifies the density of triads of linked nodes, e.g., the fraction of the neighbors of a node that are themselves neighbors. The clustering coefficient is defined by $$ C(G)=\frac{1}{N}\sum_{i \neq j \neq h } \frac{2}{k_i(k_i-1)} ~ a_{i,j}a_{j,h}a_{h,i} ~, \label{cluster}$$ where $k_i$ is the degree of node $i$ and $a$ is the adjacency matrix element. \(iv) The efficiency, typically defined as the reciprocal of the harmonic mean of geodesic distances, quantifies the influence of the topology on flux of information through the network. Efficiency can be global as well as local. We define global efficiency as $$E_{\mbox{\tiny g}}(G)=\frac{1}{N(N-1)}\sum_{i \neq j \in G} \frac{1}{d_{i,j}} ~,$$ and local efficiency as $$E_{\mbox{\tiny l}} (G)=\frac{1}{N}\sum_{i \in G} \bigg{(}\frac{1}{n_i(n_i-1)} \sum_{j \neq h \in g_i} \frac{1}{d_{h,j}} \bigg{)} ~,$$ where $g_i$ are the subnetworks formed by neighbors of node $i$ and $n_i$ is the number of nodes of this subnetwork [@latora2001]. In addition to these standard network properties, we also use the Shannon entropy [@shannon1949] to quantify disorder or uncertainty. Specifically, we calculate the Shannon entropy functional of the distribution of node degrees. Let $P$ be the normalized probability distribution for node degree $k$, i.e. $\sum_{k} P(k)=1$. We define the Shannon entropy $S[P]$ of the degree distribution $P(k)$ for a network with $N$ nodes by: $$S[P]= - \sum_{k} P(k)\log {P(k)} ~.$$ Often the logarithm of base 2 is used [@cover2006elements] (e.g., in computer science), but we use the natural logarithm instead, so the entropy values shown are in natural information units rather than in bits. Maslov algorithm for generating randomized networks {#maslov-algorithm-for-generating-randomized-networks .unnumbered} --------------------------------------------------- Given $G$, one can select two non-overlapping pairs $(i,j)$ and $(m,n)$ of linked nodes, then unlink them, and cross-link the pairs according to $(i,m)$ and $(j,n)$. If this process is repeated many times, the links become randomized, but the degree of each node remains the same [@Maslov2002]. Hence the entropy of the degree distribution is also a conserved quantity. Calculation of correlation matrix for brain regions {#calculation-of-correlation-matrix-for-brain-regions .unnumbered} --------------------------------------------------- We segmented the brain images into 110 brain regions according to the Harvard-Oxford cortical and subcortical structural atlas (threshold of $> 25\%$, using FMRIB Software Library, www.fmrib.ox.ac.uk/fsl). Six regions had to be excluded from further analysis, as they were not sampled for all subjects, due to technical limitations during image acquisition. For each of the 104 regions, an averaged fMRI time series was computed from all voxels (a voxel is a 3D image block, analogous to the 2D pixel). within that region using Marsbar (SPM toolbox). To reduce confounders, we applied a band-pass filter ($\approx 0.03-0.07$ Hz) using the maximum overlap wavelet transform (MODWT) with a Daubechies wavelet to divide the signal into 4 scales of different frequency bands. In keeping with the literature [@schoter2012; @liuyong2008], that point that resting state typically leads to low frequency ($\approx 0.01$ to $0.1$ Hz) [@Fransson2005], we choose scale 3. We then calculated the Pearson correlation between these wavelet coefficients from all possible pairs, thus obtaining a 104$\times$104 correlation matrix to represent each sample. Only correlations with p $<$ 0.05 were considered. Construction of complex networks from fMRI images {#construction-of-complex-networks-from-fmri-images .unnumbered} ------------------------------------------------- A correlation matrix uniquely defines a weighted network. Nonetheless, we are interested in generating non-weighted networks. Hence, we need a function that maps correlation matrices to adjacency matrices. We use a thresholding function for this purpose. Given a correlation matrix, we obtain the adjacency matrix by applying a threshold to the absolute value of the elements of the correlation matrix. Specifically, if the absolute value of the correlation matrix element $|c_{i,j}|$ is larger than a defined threshold $\eta$, then a link is assumed and the adjacency matrix element is taken to be 1 (i.e., $a_{i,j}=1$), while otherwise there is no link ($a_{i,j}=0$). In order to obtain better statistics, we choose not a single value of $\eta$ but a range of values instead. Then we analyze the behavior of the network properties over this range. Using this approach, we create a number of networks for each fMRI sample, all with the same number of nodes (104 nodes). For each of these networks, we choose $\eta$ such that the density of links is the same before and after Ayahuasca intake. We choose a range for the mean network degree to ensure the networks were fully connected but also sparse (to avoid random network behavior). For this purpose, we adopt the following criteria: the network must have lower global efficiency and greater local efficiency than its randomized version. These criteria also ensure small-world behavior of the networks [@Achard2007] (according to the definition of Watts and Strogatz [@WS1998]). In order to obtain the same threshold range for all subjects, it is necessary to exclude two of them from the analysis, since there is no threshold range common between them and the other subjects. Data from a second subject was also excluded due to excessive head movement. Following the criteria described above, the threshold range is set to $0.28 \leq \eta \leq 0.37$. We generate networks with mean degree in the range $24 \leq \langle k\rangle \leq 39$. [We evaluate measures in degree increments of $\Delta \langle k \rangle=1$, thus obtaining 16 different values of mean degree]{}. In summary, we have 7 human subjects suitable for both conditions (before and after ingestion). The resulting sets of networks allow 16 different comparisons (i.e. of differing mean degrees) for each subject before and after Ayahuasca ingestion. We calculate the topological measurements (using the Brain Connectivity Toolbox for Matlab [@rubinov2010]). Statistical testing {#statistical-testing .unnumbered} ------------------- Comparisons between the two conditions (i.e., before and after Ayahuasca) are obtained from paired-sample Student’s $t$-tests. The $p$-values shown in some of the figures are as follows: values $p<0.05$ in bold and $p<0.005$ indicated by asterisks (\*). The implicitly assumed null hypothesis is that the difference of the paired values are normally distributed with zero mean. Acknowledgements {#acknowledgements .unnumbered} ================ We thank Santo Daime members for volunteering and for providing the Ayahuasca. We thank Sidarta Ribeiro for discussions, José C. Cressoni, Marco A. A. da Silva, and Carlos Viol for feedback and CAPES and CNPq for funding. AV thanks UFV and Science without Borders (CAPES Grant No. 88881.030375/2013-01) for funding and Guillermo Cecchi and Irina Rish for their hospitality and discussions during her year at IBM. Author contributions {#author-contributions .unnumbered} ==================== D.B.A. recruited the volunteers for data acquisition and conceived the study. A.V., F.P.-F. and H.O performed fMRI data preprocessing, complex network construction and evaluated standard network features. A.V. and G.M.V. performed complex network analysis and statistical analysis. All authors contributed equally to the final overall design of the study. Competing financial interests {#competing-financial-interests .unnumbered} ============================= The authors declare no competing financial interests. \(a) Before After   \(b)   ![[**Illustrative example of functional brain networks.**]{} (a) 3 views of a complex network generated from brain fMRI data of one of the subjects, before (left) and after (right) Ayahuasca ingestion (mean node degree $ \langle k \rangle=30$). The spheres represent nodes and sphere size is proportional to the node degree. (b) histograms of the node degrees, corresponding to the networks shown in (a). After Ayahuasca intake, the distribution is wider, indicating a higher entropy. In (a) we have used the BrainNet Viewer (http://www.nitrc.org/projects/bnv) for visualization. []{data-label="braindegree"}](figuras/frequency_degree_before_subj8.png "fig:") ![[**Illustrative example of functional brain networks.**]{} (a) 3 views of a complex network generated from brain fMRI data of one of the subjects, before (left) and after (right) Ayahuasca ingestion (mean node degree $ \langle k \rangle=30$). The spheres represent nodes and sphere size is proportional to the node degree. (b) histograms of the node degrees, corresponding to the networks shown in (a). After Ayahuasca intake, the distribution is wider, indicating a higher entropy. In (a) we have used the BrainNet Viewer (http://www.nitrc.org/projects/bnv) for visualization. []{data-label="braindegree"}](figuras/frequency_degree_after_subj8.png "fig:") (a)![[**Variance and kurtosis of the degree distribution.**]{} Mean $\pm$ 1 standard deviation calculated over all 16 networks of the degree variance (a) and kurtosis (b), shown for each subject (blue $\triangledown$) and after (green $\triangle$) Ayahuasca ingestion. The individual values for the degree variance and kurtosis are calculated separately for each network. We find higher variance and (mostly) lower kurtosis after Ayahuasca, hence the node distributions change shape and become less “peaked.” Such behavior is again consistent with (if not suggestive of) a higher Shannon entropy after Ayahuasca.[]{data-label="moment"}](figuras/variance5.png "fig:")\ (b)![[**Variance and kurtosis of the degree distribution.**]{} Mean $\pm$ 1 standard deviation calculated over all 16 networks of the degree variance (a) and kurtosis (b), shown for each subject (blue $\triangledown$) and after (green $\triangle$) Ayahuasca ingestion. The individual values for the degree variance and kurtosis are calculated separately for each network. We find higher variance and (mostly) lower kurtosis after Ayahuasca, hence the node distributions change shape and become less “peaked.” Such behavior is again consistent with (if not suggestive of) a higher Shannon entropy after Ayahuasca.[]{data-label="moment"}](figuras/kurtosis5.png "fig:")   ![[**Entropy grows after Ayahuasca ingestion.**]{} Mean $\pm$ 1 standard deviation of the Shannon entropy of the distribution of node degrees, calculated over all 7 subjects, as a function of mean degree $k$, before (blue $\triangledown$) and after (green $\triangle$) Ayahuasca intake. The bottom row lists $p$-values for Student’s paired $t$-test, with values $p<0.005$ indicated by asterisks (\*). We thus see evidence against the null hypothesis of no change in entropy. Indeed, we find a significant increase in the entropy of the degree distributions after Ayahuasca ingestion. This entropy increase is the main result that we report.[]{data-label="entropy"}](figuras/entropymean_new5.png) \(a)   ![[**Entropy growth per subject.**]{} (a) Boxplot of the entropy distribution and before (B) and after (A) Ayahuasca ingestion and (b) boxplot of entropy increase, for all 7 subjects. Note the significant increase in entropy after Ayahuasca ingestion. There are 16 values of entropy per subject, as discussed in the text. The bars show minimum and maximum values and the box shows the 2nd and 3th quartiles, with the median shown dividing the box (in red). The asterisks (\*) in the bottom rows in both plots indicate $p$-values $p<0.005$ for Student’s paired $t$-test in (a) and $t$-test for zero mean in (b). Subject-by-subject, we thus find strong evidence against the null hypothesis of no entropy change.[]{data-label="boxentropy"}](figuras/entropyboxsubj85.png) (b)  ![[**Entropy growth per subject.**]{} (a) Boxplot of the entropy distribution and before (B) and after (A) Ayahuasca ingestion and (b) boxplot of entropy increase, for all 7 subjects. Note the significant increase in entropy after Ayahuasca ingestion. There are 16 values of entropy per subject, as discussed in the text. The bars show minimum and maximum values and the box shows the 2nd and 3th quartiles, with the median shown dividing the box (in red). The asterisks (\*) in the bottom rows in both plots indicate $p$-values $p<0.005$ for Student’s paired $t$-test in (a) and $t$-test for zero mean in (b). Subject-by-subject, we thus find strong evidence against the null hypothesis of no entropy change.[]{data-label="boxentropy"}](figuras/entropybox_diff5.png) \(a) (b)   \(c) (d)   \(e) (f)   Figure 5. (See next page for caption.) \(a) (b)   \(c) (d)  \(e) (f)   Figure 6. (See next page for caption.)

ArXiv

--- abstract: 'ISIS (In-situ Storage Imaging Sensor) is a novel CMOS sensor with multiple charge storage capability developed for the ILC vertex detector by the Linear Collider Flavour Identification (LCFI) collaboration. This paper reports test results for ISIS2, the second generation of ISIS sensors implemented in a 0.18 micron CMOS process. The local charge storage and charge transfer were unambiguously demonstrated.' author: - | Yiming Li$^1$, Chris Damerell$^2$, Rui Gao$^1$, Rhorry Gauld$^1$, Jaya John John$^1$,\ Peter Murray$^2$, Andrei Nomerotski$^1$, Konstantin Stefanov$^3$, Steve Thomas$^2$,\ Helena Wilding$^1$, Zhige Zhang$^2$\ 1- Sub-department of Particle Physics\ University of Oxford, Oxford OX1 3RH - UK\ 2- Rutherford Appleton Laboratory - UK\ 3- Sentec Ltd - UK title: ' ISIS2: Pixel Sensor with Local Charge Storage for ILC Vertex Detector' --- Introduction ============ In order to achieve the precision targeted at the ILC, the vertex detector must meet challenging requirements. The resolution of $\sim 3 \mu m $ is a straightforward one to enable the high precision studies of processes requiring b-tagging such as the measurement of top quack forward backward assymmetry, Higgs self-coupling etc [@SiDLOI]. In the $e^+e^-$ machine huge amount ($10^5\sim10^6$ per collision) of $e^+e^-$ pairs will be present as background due to beamstrahlung. Figure \[Fig:ILCtimeslicing\] shows the ILC beam structure. It is essential for the vertex detector to keep the occupancy below $1\%$. Therefore time slicing is necessary. In addition, the material budget will be $0.1 X_0\%$ per layer, which largely limits the cooling option to air cooling. [r]{}[0.5]{} {width="0.45\columnwidth"} Two different approaches exist to tackle the low occupancy issue. The first is fast-readout during the collisions, at least 20 frames in $\sim 1 ms$ time. This has been pursued by Column Parallel CCDs [@CPCbib1; @CPCbib2; @CPCbib3], MAPS [@MAPSbib1; @MAPSbib2] and DEPFET [@DEPFETbib1; @DEPFETbib2] sensor types just to name a few. The ISIS sensor pursues another approach, which is to store the charge at the pixel level during the bunch train and to read out during the quiet time. This reduces the required peak power and avoids power cycling, which is needed for the first approach. ISIS principle -------------- The design concept of ISIS sensor can be illustrated in the cross section of a pixel as in Figure \[Fig:BC-reset-transistor\]. During the collisions the charge generated in the $\sim 20 \mu m$ epitaxial layer is collected under the photogate. Then the raw charge is transferred down to the pixel-level storage cells, equivalent to a short CCD column. The signal charge will be shifted to the register every $50 \mu s$ as the time slicing required. The charge-voltage conversion and readout of full array will happen in the $~200 ms$ quiet time between the bunch trains. The CCD register is protected from the epitaxial layer by the deep $p^+$ implant. ![ISIS design concept.[]{data-label="Fig:BC-reset-transistor"}](figures/BC-reset-transistor.eps){width="90.00000%"} History ------- Fast framing CCD optical cameras based on ISIS principle has been under development since around ten years ago [@GojiEtoh], with the maximum frame rate of $\sim 100 Megaframes/s$ achieved. The ISIS sensor as a particle detector for ILC vertexing was pioneered by Linear Collider Flavour Identification (LCFI) collaboration in 2003. [r]{}[0.4]{} {width="0.35\columnwidth"} The first generation device ISIS1 was produced to prove the feasibility of local charge storage on a relatively large CCD pitch. A example of ISIS1 pixels is shown in Figure \[Fig:ISIS1\]. The ISIS1 sensors was manufactured by e2v technologies [@e2v] with $\sim 2 \mu m $ CCD process. There are 5 storage cells for each pixel, each with area of $160 \times 40 \mu m^2$. They were successfully tested with X-ray and testbeams [@ISIS1bib1; @ISIS1bib2; @ISIS1bib3]. ISIS2, the second generation of ISIS sensor was designed and manufactured on a smaller pitch and with more storage cells. The ISIS2 sensors were received in 2008 and the main results of their testing will be summarized in the following sections. Section \[sec:design\] will explain the design of ISIS2 in detail. Section \[sec:test-result\] will report the testing results on test structure and main arrays. ISIS2 Design {#sec:design} ============ The ISIS2 sensors were manufactured in a $\sim 0.18 \mu m$ CMOS process by Jazz Semiconductor [@Jazz]. The small feature size of the foundry enables 20 storage registers every pixel, each with the area of $3 \times 5 \mu m^2$ (comparing to the $20 \times 40 \mu m^2$ storage cell on ISIS1). ![ISIS2 sensor wired-bonded in a ceramic package(left) and ISIS2 floor plan(right).[]{data-label="Fig:package-floorplan"}](figures/ISIS2-from-package-to-floorplan.eps){width="90.00000%"} Figure \[Fig:package-floorplan\] shows a photo of a packaged ISIS2 sensor and its floor plan. The total size of the sensor is $5 \times 5 mm^2$. The sensor has 32 columns and 128 rows, divided into the upper and lower halves, with buried-channel and surface-channel reset transistor respectively. The columns are equally divided into four sections featured by different deep $p^+$ well variations. The readout from 8 columns in each sections are multiplexed, and the rows are controlled by a rolling shutter. The ISIS2 imaging pixel size is $40 \times 20 \mu m^2$ while the pixel layout size is $80 \times 10 \mu m^2$ as in Figure \[Fig:pixel-layout\]. This miniaturization is a major progress compared to ISIS1 thanks to the small feature size provided by the foundry. Each pixel of ISIS2 contains a 3-phase CCD with 20 storage cells, a reset transistor, source follower and a row select transistor. There is a charge injection input next to the photogate and this controlled charge injection is useful for testing. All clock and bias signals are shared between all pixels except the Summing Gate (SG) and Row Select (RSEL) signals, which can be controlled by a separate transfer gate at row level. A diagram of the pixel structure described here is shown in Figure \[Fig:pixel-diagram\]. ![ISIS2 pixel layout. An $80 \times 10 \mu m^2$ pixel and an imaging pixel of the size $40 \times 20 \mu m^2$ are in shadow.[]{data-label="Fig:pixel-layout"}](figures/pixel-layout.eps){width="60.00000%"} ![Diagram of an ISIS2 pixel.[]{data-label="Fig:pixel-diagram"}](figures/pixel-diagram.eps){width="90.00000%"} [r]{}[0.6]{} {width="0.57\columnwidth"} The separation of the buried channel and the epitaxial layer with the deep $p^+$ implant is another improvement after ISIS1 only possible in a modern CMOS process. Figure \[Fig:pwell-xsec\] shows the part of the cross-section of an ISIS2 pixel under the photogate. The opening of the deep $p^+$ implant is shown (not to scale) through which the deposited charge is collected. In order to investigate the deep $p^+$ implant, four sections of the ISIS2 chip include following variations: (1)without deep $p^+$; (2)with deep $p^+$ but without the opening under the photogate; (3) with deep $p^+$ and the opening; (4) with deep $p^+$ and a larger opening compared to (3). The fabrication process enables a range of additional ISIS2 variations which can be helpful in studying the performance optimization. These variations are listed below: - [Deep $p^+$ well (as mentioned above);]{} - [Reset transistors implemented in surface channel and buried channel;]{} - [Different CCD gate and gap width;]{} - [Change in dopant concentration up to 20%]{} ISIS2 Test Results {#sec:test-result} ================== Test Structure -------------- A simple test structure (as shown in Figure \[Fig:test-structure\]) was included on the same ISIS wafer. It is similar to a pixel in the main array except for the absence of the storage CCD cells. This structure allows us to establish the optimum operation point before studying the full array. Over half of the test structure area is taken by the $4 \times 5 \mu m^2$ photogate. The small size of the output node will result in its small capacitance, hence a large charge to voltage conversion factor and very low noise. On the other hand the small size together with the tapered shape from the output gate to the output node could complicate charge transfer between the two and also means that the edge effects and 3D fringing fields can become important. ![ISIS2 test structure.[]{data-label="Fig:test-structure"}](figures/test-structure.eps){width="60.00000%"} It turned out that the fringing effect is important because the voltage on the output gate (OG) is pulled up by the constant 5 V voltage on the neighbouring output gate. It was found that the OG bias should set to below 0 V otherwise the charge will leak from the summing gate directly into the output node. The OG bias value for a typical CCD could be above 1 V. Figure \[Fig:fringe-effect\] shows an example of such behaviour [@Vertex09ISIS]. The sensor was illuminated with LED light and then the photogate was lowered while the summing gate was kept high. The charge collected should be temporarily held within summing gate if the OG functions properly. The line labeled “SG” shows the charge kept under summing gate and “PG” shows the charge that leaked through OG to the output node. It is clear that for a low OG voltage the charge can be transferred as expected while at high OG the transfer is severely hampered by fringe field and happens prematurely. ![Charge that can be held under summing gate(SG) and that leaks to output node(PG) at different OG voltage. [@Vertex09ISIS][]{data-label="Fig:fringe-effect"}](figures/fringe-effect.eps){width="60.00000%"} [r]{}[0.5]{} {width="0.45\columnwidth"} Another unexpected feature discovered in ISIS2 is the highly resistive polysilicon gate. The gates were left unintentionally undoped which caused the high resistivity. The immediate result is that the time constant to charge up the gate to a certain voltage is substantially longer than expected, and a charge transfer between two neighbouring gates typically takes $\sim ms$. The time constant also varies with the temperature. From Figure \[Fig:slow-readout\] which shows the voltage output it is clear that the response at lower temperature is much slower. Though the time of operation was largely limited by the charge transfer speed, it was also found that the charge survives up to a few seconds in the sensor which enables us to investigate charge transfers in the CCD structure and study the transfer efficiency. The absolute calibration of the test structure is done using a standard $^{55}Fe$ radiative source. A minimum noise of $6 e^-$ allows the $K_{\alpha}$ and $K_{\beta}$ peaks to be resolved, which correspond to $1620 e^-$ and $1780 e^-$ generated in the epitaxial layer, as shown in Figure \[Fig:Fe55-calibration\]. It can be derived that the response of the output node is $24 \mu V/e^-$ as designed. ![Amplitude spectrum of $^{55}Fe$ source. The left peak is the pedestal and the two labelled signal peaks on the right are caused by the $^{55}Fe$ X-rays. These are the direct hits on output node at $31^{\circ}C$.[]{data-label="Fig:Fe55-calibration"}](figures/Fe55-calibration.eps){width="80.00000%"} With the X-ray source it is possible to estimate the charge transfer efficiency (CTE) on the test-structure. The output from two cases are compared: (1) direct X-ray hits on the output node and (2) hits collected on the photogate and transferred through summing gate to the output node. Separation of the $K_{\alpha}$ peak and pedestal peak in both cases give the charge on the output node. The ratio of the charge in case (2) over case (1) is the CTE estimated as in Table \[Tab:CTE-test-structure\]. This measurement was repeated at room temperature and low temperature, and the CTE estimation agrees well. This relatively low CTE is probably caused by the tapered shape of transition from the summing gate to the output node. Temperature $-10^{\circ}C$ $31^{\circ}C$ ------------- ---------------- --------------- CTE 94.2% 94.5% : CTE on test-structure.[]{data-label="Tab:CTE-test-structure"} In addition to the X-ray source, the charge transfer was also studied with charge injection, dark current integration and LED pulses. One example using dark current is given in Figure \[Fig:test-structure-linearity\] which shows the dependence of the output voltage on the photogate integration time. From this measurement the summing gate capacity can be estimated to $5000 \sim 10000 e^-$, depending on the summing gate bias. ![Dependence of the output voltage on the photogate integration time.[]{data-label="Fig:test-structure-linearity"}](figures/test-structure-linearity.eps){width="60.00000%"} Main Array ---------- An example of successful charge transfer in the ISIS2 main array is shown in Figure \[Fig:first-ISIS2-main-array\]. The oscilloscope traces both with and without charge injection are shown. The difference emphasized with the circle is due to the output of the injected charge after being transferred down the 20 storage cells. The linear accumulation of leakage current on the twenty CCD registers is also clearly visible. ![The oscilloscope trace of a successful charge transfer in the ISIS2 main array.[]{data-label="Fig:first-ISIS2-main-array"}](figures/first-ISIS2-main-array.eps){width="70.00000%"} The slow readout speed mentioned earlier means that it takes a long time (up to seconds) to read a full pixel, during which time the dark current will cause a considerable shot noise and more worryingly, could saturate the well capacity. A lot of efforts therefore have been directed to minimize the readout time. The most straightforward way of reducing the dark current will be to reduce temperature, but this will lead to higher resistivity of the polysilicon gate, hence even longer readout time. The operation temperature between $-5^{\circ}C$ and room temperature was found to be optimal. After optimization the transfer time for the CCD gates was set to 85% of the summing gate time. At room temperature a readout rate of about 10 Hz was achieved. In order to measure the CTE for the 20 storage cells with $^{55}Fe$ X-ray source, a challenge is to disentangle the hits on the photogate and those on the storage cells. There are two different approaches on this issue. Since three-phase CCD registers are used, the charge can be transferred in both directions. The first approach uses charge manipulation and compares the two cases as illustrated in Figure \[Fig:charge-reversal\]. In case (2) the charge from photogate is first moved forward, then backward for $n$ cells, then forward again. In case (1) charge always moves forward, but will be paused in the middle simply to keep both sequences at the same length. Therefore there will be $2n$ more transfers in case (2) than in (1). Assume the outputs measured are $S_1$ and $S_2$, then $S_2/S_1 = CTE^{2n} \simeq 1 - 2n\times CTI $, where $CTI = 1 - CTE$ is Charge Transfer Inefficiency. This measurement gives $CTE \gtrsim 99\%$. The uncertainty of this measurement is quite large, mainly due to the temperature fluctuation during the measurements. ![CTE measurement method using reversing of charge in the CCD register.[]{data-label="Fig:charge-reversal"}](figures/charge-reversal.eps){width="80.00000%"} Another independent approach to measure the CTE is to define the time window for X-ray illumination. Two measurements based on this idea were made at different setups. One of them uses a moving X-ray source and the other use a fast mechanical shutter between the source and the sensor to control the timing of the X-ray illumination. If the initial charge package is $S_0$, after $N$ transfers the charge left will be $S_N = S_0 \times ( 1- CTI)^N \sim S_0(1- N\times CTI)$. So the CTE is calculated from the measured output signal dependence on the number of transfers as in Figure \[Fig:CTE-slopes\]. The measurement with a moving source gives a CTE of 99.3% [@Vertex09ISIS] and that with a shutter measures 98.4%. It should be noted that these are measured under different conditions (temperature, frequency) and using different ISIS2 variations, so they are not directly comparable. Still these numbers are consistent within uncertainties and show a reasonably high transfer efficiency. ![Position of $^{55}Fe$ signal peak versus number of transfers. The left plot was obtained with a moving $^{55}Fe$ source [@Vertex09ISIS] and the right with a mechanical shutter between the source and the sensor.[]{data-label="Fig:CTE-slopes"}](figures/CTE-slopes.eps){width="80.00000%"} As mentioned in Section \[sec:design\], all rows are read out by a rolling shutter and each eight columns are multiplexed. Figure \[Fig:multiplexed-output\] shows an example of the multiplexed voltage output. ![The voltage output of 8 columns multiplexed.[]{data-label="Fig:multiplexed-output"}](figures/column-multiplex.eps){width="0.55\columnwidth"} Conclusion ========== The tests of ISIS2 sensors showed successful charge storage and transfer abilities for a short CCD register implemented in a CMOS process. A few unexpected features such as the slow charge transfer caused by the resistive polysilicon gates were well understood and should be easy to correct in future iterations. A large area ISIS sensor with more compact and optimized pixel geometry, and readout architecture with data serialization is a viable technology for the ILC vertexing. [99]{} H. Aihara [*et al.*]{}, [*SiD Letter of Intent*]{}, arXiv:0911.0006v1\[physics.ins-det\] (2009). C.J.S. Damerell, Nucl. Instr. Meth. [**A568**]{} 240 (2006). A. Nomerotski, Nucl. Instr. Meth. [**A598**]{} 33 (2009). K. Stefanov, Nucl. Instr. Meth. [**A569**]{} 48 (2006). R. Turchetta [*et al.*]{}, Nucl. Instr. Meth. [**A458**]{} 677 (2001). G. Deptuch [*et al.*]{}, IEEE Trans. Nucl. Sci. [**51**]{} 2313 (2004). J. Kemmer, G. Lutz, Nucl. Instr. Meth. [**A253**]{} 365 (1984). P. Fischer [*et al.*]{}, Nucl. Instr. Meth. [**A582**]{} 843 (2007). Dao Vu Truong Son [*et al.*]{}, Sensors [**10**]{} 16-35 (2010). http://www.e2v.com. J.J. Velthius [*et al.*]{}, Nucl. Instr. Meth. [**A599**]{} 161 (2009). D. Cussans [*et al.*]{}, Nucl. Instr. Meth. [**A604**]{} 393 (2009). Z. Zhang [*et al.*]{}, Nucl. Instr. Meth. [**A607**]{} 538 (2009). http://www.towerjazz.com/. Z. Zhang [*et al.*]{}, [*In-situ Storage Image Sensor for a Vertex Detector at the ILC*]{}, Proceeding of Vertex 2009, Veluwe, Netherlands (2009). To be published by PoS.

ArXiv

--- abstract: 'The standard theory of relativity is based on the hypothesis of locality. The locality principle assumes that an object is affected only by its immediate surroundings and not by variables in the past. It follows that in standard relativity theory even wave properties are measured instantaneously. This contradicts the Bohr-Rosenfeld principle, according to which fields cannot be determined instantaneously. Nonlocal special relativity resolves the problem by taking past history into account. The current status of nonlocal electrodynamics is discussed and a new consequence of nonlocality, namely, a certain additional amplitude shift due to nonlocality in the spin-rotation coupling is presented.' author: - Bahram Mashhoon bibliography: - 'sample.bib' title: 'Nonlocal Special Relativity: Amplitude Shift in Spin-Rotation Coupling' --- [address=[Department of Physics and Astronomy, University of Missouri, Columbia, MO 65211, USA]{} ]{} Introduction ============ The fundamental quantum laws of microphysics have all been formulated with respect to ideal inertial observers. Such observers do not exist in nature; indeed, all realistic observers are more or less accelerated. Nevertheless, the basic significance of the hypothetical inertial observers for theoretical physics must be noted. These are the fundamental observers—namely, the free observers that remain at rest and thus characterize the global inertial frames in Minkowski spacetime [@Ein; @Min]. Their importance in the historical development of physics has been emphasized by I. B. Cohen [@Coh]: “I believe it fair to say that it was the freedom to consider problems either in a purely mathematical way or in a “philosophical” (or physical) way that enabled Newton to express the first law and to develop a complete inertial physics. After all, physics as a science may be developed in a mathematical way but it always must rest on experience—and experience never shows us pure inertial motion. Even in the limited examples of linear inertia discussed by Galileo, there was always some air friction and the motion ceased almost at once, as when a projectile strikes the ground. In the whole range of physics explored by Galileo there is no example of a physical object that has even a component of pure inertial motion for more than a very short time. It was perhaps for this reason that Galileo never framed a general law of inertia. He was too much a physicist.” To extend the laws of physics to accelerated observers, a physical connection must be set up between accelerated and ideal inertial observers in Minkowski spacetime. The standard theory of relativity is based on the following pointwise relation: the accelerated observer is equivalent—at each event along its world line—to an otherwise identical momentarily comoving inertial observer. However, such a *hypothesis of locality* is a relic of Newtonian mechanics of point particles and must be generalized for fields to a *nonlocal* ansatz, which then leads naturally to a nonlocal special relativity theory [@Mas]. From this acceleration-induced nonlocality, one would then expect a nonlocal general relativity theory as well, since inertia and gravitation are intimately linked in accordance with the principle of equivalence of inertial and gravitational masses. Recently, it has been possible to develop a nonlocal generalization of Einstein’s theory of gravitation via the teleparallel equivalent of general relativity. The resulting nonlocal theory implies that gravity is nonlocal even in the Newtonian regime. The nonlocally modified Newtonian gravitation appears to provide a natural explanation for the dark matter problem in cosmology; in fact, *nonlocal gravity simulates dark matter*. Further consequences of this theory are under active investigation at the present time [@He1; @He2; @Blo; @Bah; @Car]. It is important to emphasize that in the nonlocal theory, the fields are always local; however, they satisfy nonlocal integro-differential equations. Nonlocal aspects of the gravitational interaction have also been explored by Novello and his collaborators [@No1; @Nov; @YNO; @Luc; @Duq]. Novello’s approach has involved an antisymmetric tensor of the third rank that was first introduced by Fierz in 1939 [@No1; @Nov]. The spin-two field theory based on the Fierz tensor has its origin in the teleparallel approach to general relativity [@YNO]. In this way, Novello *et al.* have developed a parallel framework for a nonlocal theory of gravity [@Duq]. *It is a great pleasure for me to dedicate this paper to Mário Novello on the occasion of his seventieth birthday*. The plan of this paper is as follows. The essential steps in the argument for nonlocality are reviewed in the following section. The difficulties associated with nonlocal electrodynamics are then described. Observational results will be essential in the determination of the appropriate nonlocal kernel in this case. As an example, a specific consequence of nonlocality for *electromagnetic fields*, namely, the contribution of nonlocality to the amplitude shift in the spin-rotation coupling is treated in detail. The final section contains a brief discussion of our results. Bohr-Rosenfeld Principle and Nonlocality ======================================== Consider Maxwell’s electrodynamics in an inertial frame of reference in Minkowski spacetime. The electric and magnetic fields, $\mathbf{E}(t,\mathbf{x})$ and $\mathbf{B}(t,\mathbf{x})$, respectively, that satisfy Maxwell’s equations are assumed to be fields measured by the fundamental observers at rest in the inertial frame. In 1933, Bohr and Rosenfeld pointed out that in fact only *spacetime averages* of these fields have immediate physical significance. That is, $\mathbf{E}(t,\mathbf{x})$ and $\mathbf{B}(t,\mathbf{x})$ occur in Maxwell’s equations as idealizations [@Boh; @Ros]. To illustrate this point, Bohr and Rosenfeld considered a simple situation involving the measurement of the electric field using a macrophysical object of volume $V$ and typical spatial dimension $L$, $V \sim L^3$, with *uniform* volume charge density $\rho$. When placed in the external electric field $\mathbf{E}(t,\mathbf{x})$, the object moves with respect to the fundamental observers according to the Lorentz force law, namely, $$\label{1} \frac{d\mathbf{P}}{dt}=\rho \int_V\mathbf{E}(t,\mathbf{x})~d^3x\,.$$ Suppose that the motion of the object is monitored over an interval of time $T=t'' - t'$ and the momentum $\mathbf{P}$ is measured to be $\mathbf{P'}$ and $\mathbf{P''}$ at the initial and final instants, $t'$ and $t''$, respectively, of the experiment. Then, $$\label{2} \mathbf{P''}-\mathbf{P'}=\rho \int_{t'}^{t''}\int_V\mathbf{E}(t,\mathbf{x})~d^3x~dt=\rho \langle \mathbf{E} \rangle TV \,,$$ where the measured electric field is $$\label{3} \langle \mathbf{E} \rangle=\frac{1}{\Delta} \int_{\Delta} \mathbf{E}(x)~ d^4x$$ with $\Delta=TV$ and $x^{\mu}= (ct, \mathbf{x})$. Henceforth we use units such that $c=1$, unless specified otherwise; moreover, the signature of the spacetime metric is $+2$ in our convention. It is assumed here that the times needed by the fundamental observers for momentum measurements are $\ll T$ and the corresponding displacements caused by these measurements are $\ll L$. *The gist of the Bohr-Rosenfeld argument is that fields cannot be measured instantaneously.* While this argument appears to be relatively innocuous for classical field measurements via ideal inertial observers, it leads to nonlocal special relativity for accelerated observers in Minkowski spacetime as a direct consequence of the existence of invariant acceleration scales. We recall that an accelerated observer is generally endowed with intrinsic acceleration lengths such as $c^2/g(\tau)$ and $c/\Omega(\tau)$, where $\tau$ is the proper time of the observer and $g$ is the magnitude of its translational acceleration, while $\Omega$ is the angular speed of rotation of its spatial frame with respect to a nonrotating (i.e., Fermi-Walker transported) frame. Let $\lambda^{\mu}{}_{(\alpha)}$ be the observer’s orthornormal tetrad frame along its world line, then in general $$\label{4} \frac{d\lambda^{\mu}{}_{(\alpha)}}{d\tau}=\Phi_{(\alpha)}{}^{(\beta)}(\tau)~\lambda^{\mu}{}_{(\beta)}\,,$$ where $\Phi_{(\alpha)(\beta)} = - \Phi_{(\beta)(\alpha)}$ is the spacetime-invariant antisymmetric acceleration tensor of the observer. Here, in analogy with electrodynamics, the translational and rotational accelerations of the observer constitute the spacetime-invariant “electric” and “magnetic” components of the acceleration tensor $\Phi_{(\alpha)(\beta)}$. Let $\psi(x)$ be a basic radiation field as determined by the fundamental inertial observers and let $\Psi(x)$ be the field measured by the accelerated observer. According to the locality principle of the standard relativity theory, $$\label{5} \Psi (\tau)= \hat{\psi}(\tau) , \qquad \hat{\psi} = \Lambda \psi\,,$$ where $\hat{\psi}$ is the field measured instantaneously by the infinite set of hypothetical momentarily comoving inertial observers whose straight world lines are tangent to the world line of the accelerated observer. Moreover, at each instant $\tau$, $\hat{\psi}$ is related to $\psi$ by an element $\Lambda$ of a matrix representation of the Lorentz group. To satisfy the Bohr-Rosenfeld principle, however, we must generalize Eq. . The most general *linear* relationship between $\Psi$ and $\hat{\psi}$ that is consistent with causality is $$\label{6} \Psi(\tau)= \hat{\psi}(\tau) + u(\tau - \tau_{0})\int_{ \tau_{0}}^{\tau}K(\tau,\tau') \hat{\psi}(\tau')d\tau' \,.$$ Here, $u(t)$ is the unit step function such that $u(t) = 0$ for $t<0$ and $u(t) = 1$ for $t>0$ and $\tau_{0}$ is the initial instant of proper time at which the observer is accelerated. As in the case of the Bohr-Rosenfeld Eq. , Eq.  is manifestly covariant under the inhomogeneous Lorentz group of spacetime transformations. It remains to determine kernel $K$, which must be proportional to the acceleration of the observer. This is the main problem of nonlocal special relativity. The basic nonlocal ansatz  is a Volterra integral equation of the second kind. Therefore, it has the fundamental property that the relationship between $\Psi(\tau)$ and $\psi(\tau)$ is unique in the space of functions of physical interest in accordance with the Volterra-Tricomi theorem [@Vol; @Tri]. For a *pure radiation field*, we impose the requirement that a *constant* $\Psi$ should uniquely correspond to a *constant* $\psi$; then, a variable $\psi$ will always lead to a variable $\Psi$. In this way, *no observer can ever stay completely at rest with a pure radiation field*. Following this line of thought, we finally arrive at the kernel [@BaM; @BMa; @Chi; @ChM; @Hel] $$\label{7} K(\tau,\tau')=k(\tau')=-\frac{d\Lambda(\tau')}{d\tau'}\Lambda^{-1}(\tau')\,.$$ With this kernel, our nonlocal ansatz  takes the form $$\label{8} \Psi(\tau)= \hat{\psi}(\tau_{0}) + \int_{ \tau_{0}}^{\tau}\Lambda(\tau')\frac{d\psi(\tau')}{d\tau'}d\tau' \,$$ for $\tau \ge \tau_{0}$. It turns out that the nonlocal contribution in our ansatz  is negligible when the intrinsic scale of the phenomenon under observation is sufficiently small in comparison with the scale of variation of the state of the observer. This happens to be the case for most Earth-based experiments, since $c^{2}/|\mathbf{g}_{\oplus}|\approx 1$ light year and $c/|\boldsymbol{\Omega}_{\oplus}|\approx 28$ astronomical units. The implications of the nonlocal theory have been worked out in detail—see, for instance, Ref. [@BMA] for the nonlocal Dirac equation. Once acceleration is turned off, its memory in general persists in the form of an additive constant field. In nonlocal special relativity, local fields satisfy integro-differential field equations that contain the memory of past acceleration. For fundamental scalar or pseudoscalar radiation fields, $\Lambda=1$ and hence $K=0$; indeed, nonlocal special relativity predicts that such radiation fields do not exist in nature. This prediction is consistent with observation. The nonlocal kernel given by Eq.  has been derived for a *pure radiation field*. Its application to the electromagnetic field encounters difficulties, however, as it is not clear how one must treat nonlocal electrostatics and magnetostatics in this framework. This problem is the subject of the next section. Nonlocal Electrodynamics ======================== Consider an electromagnetic field $F_{\mu\nu}(x)$ and the corresponding gauge potential $A_{\mu}(x)$, $F_{\mu\nu}=\partial_{\mu}A_{\nu}-\partial_{\nu}A_{\mu}$, in the global inertial frame in Minkowski spacetime. Along the world line of the accelerated observer, the fields measured by the momentarily comoving inertial observers are $$\label{9} A_{(\alpha)}(\tau)=A_{\mu}\lambda^{\mu}{}_{(\alpha)}, \qquad F_{(\alpha)(\beta)}(\tau)=F_{\mu\nu}\lambda^{\mu}{}_{(\alpha)}\lambda^{\nu}{}_{(\beta)}\,.$$ Let $\mathcal{A}_{(\alpha)}(\tau)$ and $\mathcal{F}_{(\alpha)(\beta)}(\tau)$ be the fields determined by the accelerated observer; then, in accordance with our nonlocal ansatz, $$\begin{aligned} \label{10} \mathcal{A}_{(\alpha)}(\tau)=A_{(\alpha)}(\tau)+u(\tau-\tau_{0})\int_{\tau_{0}}^{\tau}K_{(\alpha)}{}^{(\beta)}(\tau,\tau')A_{(\beta)}(\tau')d\tau'\,\end{aligned}$$ and $$\begin{aligned} \label{11} \mathcal{F}_{(\alpha)(\beta)}(\tau)=F_{(\alpha)(\beta)}(\tau)+u(\tau-\tau_{0})\int_{\tau_{0}}^{\tau}K_{(\alpha)(\beta)}{}^{(\gamma)(\delta)}(\tau,\tau') F_{(\gamma)(\delta)}(\tau')d\tau'\,.\end{aligned}$$ If the kernel in Eq.  is chosen in accordance with Eq. , namely, $$\begin{aligned} \label{12} K_{(\alpha)}{}^{(\beta)}(\tau,\tau')=-~\Phi_{(\alpha)}{}^{(\beta)}(\tau')\,,\end{aligned}$$ then a constant $A_{\mu}(x)$ will be determined to be constant by all accelerated observers. This circumstance poses no difficulty, as the electromagnetic field vanishes altogether in this case. However, the situation is quite different if the kernel in Eq.  is chosen in accordance with Eq. ; then, constant electromagnetic fields in the laboratory will always be measured to be constant by any accelerated observer. This conclusion appears to contradict the results of Kennard’s experiment [@Ken; @Peg]. The issue has been discussed in detail in [@MaN]; in this case, the kernel must be determined from $\Phi_{(\alpha)(\beta)}$, the Minkowski metric tensor and the Levi-Civita tensor (with $\epsilon_{0123}=1$ in our convention). A simple choice for the kernel in Eq.  turns out to be a linear combination of the kernel given by Eq. , namely, $$\label{13} \kappa_{(\alpha)(\beta)}{}^{(\gamma)(\delta)}=-2\Phi_{\lbrack(\alpha)}{}^{\lbrack(\gamma)}~\delta_{(\beta)\rbrack}{}^{(\delta)\rbrack}$$ and its dual $$\label{14} \kappa^*_{(\alpha)(\beta)}{}^{(\gamma)(\delta)}=\frac{1}{2}\epsilon_{(\alpha)(\beta)}{}^{(\rho)(\sigma)}~\kappa_{(\rho)(\sigma)}{}^{(\gamma)(\delta)}\,.$$ That is, we tentatively assume that in vacuum $$\label{15} K_{(\alpha)(\beta)}{}^{(\gamma)(\delta)}(\tau,\tau^{'})= p~\kappa_{(\alpha)(\beta)}{}^{(\gamma)(\delta)}(\tau')+ q~\kappa^*_{(\alpha)(\beta)}{}^{(\gamma)(\delta)}(\tau')\,,$$ where $p$ and $q$ are real constant coefficients to be determined from experiment [@MaN] . Some of the properties of kernel  have been discussed in [@MaN; @Mas]. In particular, the right dual of kernel  is equal to its left dual, while its mixed duals vanish; that is, kernel  has a *unique* dual and this property is a direct consequence of the relation $\Phi_{(\alpha)(\beta)}=-\Phi_{(\beta)(\alpha)}$. It is interesting to define $$\label{15a} \chi_{(\alpha)(\beta)}{}^{(\gamma)(\delta)}=-2\Phi^*_{\lbrack(\alpha)}{}^{\lbrack(\gamma)}~\delta_{(\beta)\rbrack}{}^{(\delta)\rbrack},$$ where $\Phi^*_{(\alpha)(\beta)}$ is the dual of the acceleration tensor, $$\label{15b} \Phi^*_{(\alpha)(\beta)}=\frac{1}{2}\epsilon_{(\alpha)(\beta)}{}^{(\gamma)(\delta)}~\Phi_{(\gamma)(\delta)}\,.$$ This tensor is antisymmetric; therefore, kernel  has a unique dual as well, and it is then straightforward to show, using a generalized Kronecker delta [@Hel], that $$\label{15c} \chi^*_{(\alpha)(\beta)}{}^{(\gamma)(\delta)}=-\kappa_{(\alpha)(\beta)}{}^{(\gamma)(\delta)}\,.$$ Taking the dual of this equation, we find $$\label{15d} \chi_{(\alpha)(\beta)}{}^{(\gamma)(\delta)}=\kappa^*_{(\alpha)(\beta)}{}^{(\gamma)(\delta)}\,.$$ Inspection of Eqs.  and  reveals that under the double duality operation, kernels $\kappa_{(\alpha)(\beta)}{}^{(\gamma)(\delta)}$ and $\kappa^*_{(\alpha)(\beta)}{}^{(\gamma)(\delta)}$—and hence the nonlocal electromagnetic kernel —are simply multiplied by $-1$; for instance, $$\label{15e} \frac{1}{2}\epsilon_{(\alpha)(\beta)}{}^{(\rho)(\sigma)}~\kappa^*_{(\rho)(\sigma)}{}^{(\gamma)(\delta)}=-\kappa_{(\alpha)(\beta)}{}^{(\gamma)(\delta)}\,.$$ Available experimental results regarding electrodynamics of accelerated media are rather meager and possibly unreliable; in any case, the effect of nonlocality is expected to be very small and hence rather difficult to detect. One may therefore try to find indirect evidence for nonlocality. For instance, in the correspondence limit of large quantum numbers, electrons in atoms can be regarded as following on average classical accelerated paths and one can study their behavior under the influence of incident electromagnetic radiation. In particular, it turns out that the impulse approximation of quantum theory corresponds to the hypothesis of locality. A detailed investigation reveals that in the correspondence limit, quantum results are in better qualitative agreement with the predictions of the nonlocal theory than with the standard relativity theory based on the locality principle [@BMN]. Some of the observational consequences of kernel  have been worked out in [@MaN; @Mas]. It is argued there that we should assume $p\ge0$ and $q\ne0$; moreover, we expect that $|q|\ll 1$, since $q$ is associated with violations of parity and time reversal invariance. A further implication of this kernel involves the amplitude shift in spin-rotation coupling, to which we now turn. Spin-Rotation Coupling and Nonlocality ====================================== The nonlocal aspects of spin-rotation coupling in electrodynamics have been the subject of recent studies [@BM1; @BM2]. On the other hand, starting from classical electrodynamics and kernel , we are interested here in the measurement of the electromagnetic field when a circularly polarized plane monochromatic wave of frequency $\omega$ is normally incident on an observer that rotates uniformly with frequency $\Omega > 0$ about the direction of incidence of the radiation. Fourier analysis of $\mathcal{F}_{(\alpha)(\beta)}(\tau)$ implies in this case that the frequency measured by the rotating observer is $\gamma~(\omega\mp \Omega)$, where $\gamma$ is the Lorentz factor of the rotating observer and the upper (lower) sign refers to an incident positive (negative) helicity wave. The only exception occurs in the resonance case of incident positive helicity wave when $\omega=\Omega$, in which case the complex amplitude of the measured field varies with time as $1-i(p+iq)\gamma~\Omega\tau$; this feature is hence a direct consequence of nonlocality [@MaN]. The complex amplitude of the incident circularly polarized radiation, as measured by the rotating observer, also depends upon the helicity of the radiation as a direct consequence of nonlocality; indeed, this dependence away from resonance is given by [@MaN] $$\begin{aligned} \label{16} 1+(\pm p+iq)\frac{\Omega}{\omega\mp \Omega}\,.\end{aligned}$$ The general phenomenon of spin-rotation coupling, which is due to the inertia of intrinsic spin, has helped elucidate the energy shift that is observed when a spinning particle passes through a *rotating* spin flipper [@Neu; @Kai]. For electromagnetic radiation, the *frequency shift* that occurs when circularly polarized radiation of frequency $\omega$ passes through a rotating device (with rotation frequency $\Omega \ll \omega$) that flips the helicity of the radiation has been known observationally for a long time and has been simply explained via the photon picture [@All; @GAr; @Gar; @Sim; @Bag; @Pip; @Nie]. Let $N$ photons pass through the nonrelativistic device and undergo helicity flip; hence, the magnitude of the angular momentum of the rotating device $\mathcal{J}$ must change by $\delta \mathcal{J}=2N\hbar$, as a consequence of the law of conservation of angular momentum. Since the device is rotating with frequency $\Omega$, its energy $\mathcal{E}$ must change by $\delta \mathcal{E} = \Omega ~ \delta \mathcal{J}$ or $2N\hbar \Omega$. It then follows from energy conservation that each photon must suffer an energy shift equal to $2 \hbar \Omega$ and hence a corresponding frequency shift equal to $2\Omega$. In connection with the phenomenon of frequency shift, it is important to remark that *nonlocality* brings about a corresponding extra *amplitude shift* as well. This is described in the next section by means of nonlocal classical electrodynamics using kernel . Amplitude Shift =============== Imagine, for instance, a simple situation involving the passage of light wave of definite helicity propagating along the $z$ axis through a rotating half-wave plate (HWP) as in Figure 1. In the *background global inertial frame*, the incident positive-helicity wave has initial frequency $\omega_{i}$ and constant amplitude $\alpha_{i}$, while the corresponding outgoing negative-helicity wave has final frequency $\omega_{f}$ and constant amplitude $\alpha_{f}$. We assume here that $\gamma \approx 1$, $\omega_i \gg \Omega$ and terms proportional to $(\Omega/\omega_i)^2$ are negligible, so that we can—among other things—approximate $\omega\mp \Omega$ in Eq.  by $\omega$. At the *rotating outer boundaries of the HWP*, imagine observers at rest in the rotating frame of the HWP; according to these observers, the measured frequencies and amplitudes are then given as in the previous section by (cf. Figure 1) $$\begin{aligned} \label{17} \omega'_{1}\approx\omega_{i}-\Omega\,, \qquad \alpha'_{1}\approx \alpha_{i}~\Big \lbrack1+(p+iq)\frac{\Omega}{\omega_{i}}\Big \rbrack\end{aligned}$$ and $$\begin{aligned} \label{18} \omega'_{2}\approx\omega_{f}+\Omega\,, \qquad \alpha'_{2}\approx \alpha_{f}~\Big \lbrack1+(-p+iq)\frac{\Omega}{\omega_{f}}\Big \rbrack\,.\end{aligned}$$ Here we have neglected time dilation; indeed, all terms beyond the linear order in $\Omega$ are neglected in this analysis. The rotation of the HWP is assumed to be uniform; therefore, the wave frequency remains constant in the rotating frame, namely, $\omega'_{1}=\omega'_{2}$. Moreover, $\alpha'_{2}=\mathcal{T}\alpha'_{1}$, where $\mathcal{T}$ is the net transmission amplitude through the uniformly rotating HWP as measured *in the rotating frame*. Thus Eqs.  and  imply that $$\begin{aligned} \label{19} \omega_{f}\approx \omega_{i}-2\Omega\,,\end{aligned}$$ which is the expected *frequency shift*, and $$\begin{aligned} \label{20} \alpha_{f}\Big \lbrack1+(-p+iq)\frac{\Omega}{\omega_{f}}\Big \rbrack \approx \mathcal{T}\alpha_{i}\Big \lbrack1+(p+iq)\frac{\Omega}{\omega_{i}}\Big \rbrack\,.\end{aligned}$$ The response of the *rotating* HWP to the passage of radiation could in general be complicated [@Pip; @Bia; @Hau]. On the other hand, a static HWP can be simply treated in the standard classical manner [@Bor]. For the calculation of $\mathcal{T}$, we adopt an elementary approach that should be adequate for a sufficiently narrow incident beam along the axis of rotation of a slowly rotating HWP, namely, in the rest frame of the HWP, the passage of radiation of frequency $\omega'=\omega'_{1}=\omega'_{2}$ through the HWP is treated just as in the absence of rotation [@Bor]. In the horizontal $(x,y)$ plane of the HWP, the slow axis (with principal index of refraction $n_s$) is perpendicular to the fast axis (with principal index of refraction $n_f$) such that $\delta n= n_s-n_f>0$ is sufficiently small. The thickness $D$ of the HWP along the vertical $z$ direction is then connected to its birefringence $\delta n$ via the characteristic relation $\omega' \delta n~D/c= \pi +2\pi m$, where $m$ is a positive integer. In the background inertial frame, the circular polarization states of the radiation can be expressed in terms of unit vectors as $(\hat {\mathbf{x}}\pm i\hat{\mathbf{y}})/\sqrt{2}$ [@MaN]. In the rotating frame, which is the rest frame of the HWP, the corresponding unit vectors are $\approx (\hat {\mathbf{x'}}\pm i\hat{\mathbf{y'}})/\sqrt{2}$ for $\gamma \approx 1$ [@MaN]. Furthermore, we assume that the rotating $x' (y')$ axis coincides with the slow (fast) axis. Using the formulas given in [@Bor], the transmission amplitudes for light that is *linearly polarized* along the slow and fast axes can then be calculated and we find that $\mathcal{T}_s \approx -\mathcal{T}_f$; hence, for *circularly polarized* radiation $$\begin{aligned} \label{21} \mathcal{T}\approx \mathcal{T}_s=\frac{e^{i\zeta_s}}{1-if(n_s)e^{i\zeta_s}\sin \zeta_s}\,,\end{aligned}$$ where $$\begin{aligned} \label{22} \zeta_s=\frac{1}{c}\omega' n_s~D, \qquad f(n) = \frac{(n-1)^2}{2n}\,.\end{aligned}$$ Here, the refractive index of the slow axis $n_s$ is assumed to be slightly larger than the refractive index of the fast axis $n_f$ in such a way that $f(n_s)\approx f(n_f)$. More precisely, calculating $\delta f=f(n_s)-f(n_f)$ using Taylor expansion and assuming that $\delta f\ll f$, we find that $\delta n$ must satisfy the condition $$\begin{aligned} \label{23} \frac{\delta n}{n_f}\ll\frac{n_f-1}{n_f+1}\,.\end{aligned}$$ Moreover, we note that $\exp (i\zeta)\sin \zeta$ in the denominator of Eq.  is invariant under $\zeta \mapsto \zeta \pm (\pi +2\pi m)$. It follows from Eqs.  and  that the emerging amplitude is enhanced due to nonlocality, namely, $\alpha_{f} \approx \mathcal{T}~\Sigma~\alpha_{i}$, where $$\begin{aligned} \label{24} \Sigma(\Omega)\approx 1+p\frac{2\Omega}{\omega_{i}}\,.\end{aligned}$$ That is, $q$ drops out at the linear order in $\Omega$ and $\Sigma(\Omega)-1$ is the relative *amplitude shift due to nonlocality* given approximately by $2p~\Omega / \omega_{i}$. Assuming that $p>0$ is, say, of the order of unity, the amplitude upshift is expected to be very small compared to unity. For an incident negative-helicity wave, $\Omega\mapsto -\Omega$ in Eqs.  and  and so there would be an upshift in frequency by $2\Omega$ and a corresponding downshift in amplitude, so that the emerging amplitude would be diminished due to nonlocality by the factor $\Sigma(-\Omega)\approx 1-2p~\Omega/\omega_{i}$. The frequency shift given by Eq.  is a general consequence of spin-rotation coupling; indeed, as mentioned in the previous section, such a shift is experimentally well known and has been the subject of a number of investigations—see, for instance, [@All; @GAr; @Gar; @Sim; @Bag; @Pip; @Nie; @Cou; @Bas; @Bl1; @Bl2; @Bl3] and the references cited therein. The situation is different, however, for the amplitude shift, which has not yet been detected. In fact, nonlocality implies that for $p>0$ the corresponding relative shift in amplitude is positive (negative) when the helicity of the incident wave is in the same (opposite) sense as the rotation of the HWP. The amplitude shift due to nonlocality occurs in addition to the amplitude shift that comes about as a direct result of spin-rotation coupling and is revealed through the dependence of the transmission amplitude $\mathcal{T}$ upon $\omega' \approx \omega_i -\Omega$ via $\zeta$; indeed, this dependence can be clearly seen in the transmission coefficient, $$\begin{aligned} \label{25} |\mathcal{T}|^2\approx \frac{1}{1+f(f+2)\sin^2 \zeta}\,,\end{aligned}$$ where $f(f+2)=(n^2-1)^2/(4n^2)$, $\zeta \approx \omega_i~n~D/c-\Omega~n~D/c$ and $n$ can be either $n_s$ or $n_f$. For visible light with $\omega_{i}/(2\pi)\approx 5\times 10^{14}$ Hz and a HWP rotating uniformly with frequency $\Omega/(2\pi)\approx 25$ Hz, we have $2\Omega/\omega_{i}\approx 10^{-13}$, which implies a rather small relative shift in amplitude, too small perhaps to be detectable at present. This is consistent with the relatively low level of amplitude sensitivity of current observational data; in fact, experiments of this type have not reported any similar amplitude shift—see [@All; @GAr; @Gar; @Sim; @Bag; @Pip; @Nie; @Cou; @Bas; @Bl1; @Bl2; @Bl3] and the references cited therein. Perhaps the amplitude shift due to nonlocality would be easier to detect with microwaves or radio waves. Discussion ========== The main conceptual steps that lead to a nonlocal theory of relativity have been outlined in this paper. The problems associated with nonlocal electrodynamics have to do with the determination of the corresponding nonlocal kernel. In this regard, future experimental results concerning electrodynamics of accelerated systems will be decisive. A new consequence of nonlocal electrodynamics, namely, an extra amplitude shift proportional to $p$ in spin-rotation coupling is pointed out. Estimates suggest that this direct result of nonlocality is negligibly small in most situations. The possibility of detecting this novel effect is briefly discussed. This paper is based in part on lectures delivered at the XIV Brazilian School of Cosmology and Gravitation (August 30 - September 11, 2010). Thanks are due to Mário Novello and the organizing committee for their kind invitation and excellent hospitality. I am grateful to Friedrich Hehl for helpful discussions on all aspects of nonlocal electrodynamics. [99]{} A. Einstein, *The Meaning of Relativity*, Princeton University Press, Princeton, NJ, 1955. H. Minkowski, in *The Principle of Relativity*, by H. A. Lorentz, A. Einstein, H. Minkowski and H. Weyl, Dover, New York, 1952. I.  B. Cohen, *The Birth of a New Physics*, Doubleday Anchor Books, Garden City, NY, 1960, pp. 164–165. B. Mashhoon, *Ann. Phys. (Berlin)* [**17**]{}, 705 (2008) \[arXiv:0805.2926 \[gr-qc\]\]. F. W. Hehl and B. Mashhoon, *Phys. Lett. B* **673**, 279 (2009) \[arXiv:0812.1059 \[gr-qc\]\]. F. W. Hehl and B. Mashhoon, *Phys. Rev. D* **79**, 064028 (2009) \[arXiv:0902.0560 \[gr-qc\]\]. H.-J. Blome, C. Chicone, F. W. Hehl and B. Mashhoon, *Phys. Rev. D* **81**, 065020 (2010) \[arXiv:1002.1425 \[gr-qc\]\]. B. Mashhoon, “Nonlocal Gravity,” in *Cosmology and Gravitation, Proc. XIV Brazilian School of Cosmology and Gravitation (Rio de Janeiro, 2010)*, edited by M. Novello and S. E. Perez Bergliaffa, Cambridge Scientific Publishers, UK, 2011, pp. 1–9 \[arXiv:1101.3752 \[gr-qc\]\]. C. Chicone and B. Mashhoon, *J. Math. Phys.* [**53**]{}, 042501 (2012) \[arXiv:1111.4702 \[gr-qc\]\]. M. Novello, arXiv: gr-qc/0212026. M. Novello and R. P. Neves, *Class. Quantum Grav.* [**19**]{}, 5335 (2002). Yu. N. Obukhov and J. G. Pereira, *Phys. Rev. D* [**67**]{}, 044008 (2003). Luciane R. de Freitas, M. Novello and N. Pinto-Neto, *J. Math. Phys.* [**35**]{}, 734 (1994). M. Novello and S. L. S. Duque, *Int. J. Mod. Phys. D* [**4**]{}, 79 (1995). N. Bohr and L. Rosenfeld, *K. Dan. Vidensk. Selsk. Mat. Fys. Medd.* **12**, No. 8 (1933);\ translated in *Quantum Theory and Measurement*, edited by J. A. Wheeler and W. H. Zurek, Princeton University Press, Princeton, NJ, 1983. N. Bohr and L. Rosenfeld, *Phys. Rev.* **78**, 794 (1950). V. Volterra, *Theory of Functionals and of Integral and Integro-Differential Equations*, Dover, New York, 1959. F. G. Tricomi, [*Integral Equations*]{}, Interscience, New York, 1957. B. Mashhoon, *Phys. Rev. A* **47**, 4498 (1993). B. Mashhoon, “Nonlocal Electrodynamics,” in *Cosmology and Gravitation*, *Proc. VII Brazilian School of Cosmology and Gravitation (Rio de Janeiro, August, 1993)*, edited by M. Novello, Editions Frontières, Gif-sur-Yvette, 1994, pp. 245–295. <http://www.cbpf.br/~cosmogra/Escolas/ind_class_field.html> C. Chicone and B. Mashhoon, *Ann. Phys. (Berlin)* **11**, 309 (2002). C. Chicone and B. Mashhoon, *Phys. Lett. A* **298**, 229 (2002). F. W. Hehl and Yu. N. Obukhov, [*Foundations of Classical Electrodynamics: Charge, Flux, and Metric*]{}, Birkhäuser, Boston, MA, 2003. B. Mashhoon, *Phys. Rev. A* **75**, 042112 (2007) \[arXiv: hep-th/0611319\]. E. H. Kennard, [*Phil. Mag.*]{} [**33**]{}, 179 (1917). G. B. Pegram, [*Phys. Rev.*]{} [**10**]{}, 591 (1917). B. Mashhoon, *Phys. Lett. A* **366**, 545 (2007) \[arXiv: hep-th/0702074\]. B. Mashhoon, *Phys. Rev. A* **72**, 052105 (2005) \[arXiv: hep-th/0503205\]. B. Mashhoon, *Phys. Rev. A* **79**, 062111 (2009) \[arXiv:0903.1315 \[gr-qc\]\]. B. Mashhoon, *Ann. Phys. (Berlin)* [**523**]{}, 226 (2011) \[arXiv:1006.4150 \[gr-qc\]\]. B. Mashhoon, R. Neutze, M. Hannam and G. E. Stedman, *Phys. Lett. A* **249**, 161 (1998). B. Mashhoon and H. Kaiser, *Physica B* **385-386**, 1381 (2006) \[arXiv: quant-ph/0508182\]. P. J. Allen, *Am. J. Phys.* **34**, 1185 (1966). B. A. Garetz and S. Arnold, *Opt. Commun.* **31**, 1 (1979). B. A. Garetz, *J. Opt. Soc. Am.* **71**, 609 (1981). R. Simon, H. J. Kimble and E. C. G. Sudarshan, *Phys. Rev. Lett.* **61**, 19 (1988). V. Bagini *et al*., *Eur. J. Phys.* **15**, 71 (1994). A. B. Pippard, *Eur. J. Phys.* **15**, 79 (1994). G. Nienhuis, *Opt. Commun.* **132**, 8 (1996). J. Courtial *et al*., *Phys. Rev. Lett.* **81**, 4828 (1998). I. V. Basistiy *et al*., *Opt. Lett.* **28**, 1185 (2003). K. Y. Bliokh, Y. Gorodetski, V. Kleiner and E. Hasman, *Phys. Rev. Lett.* **101**, 030404 (2008). K. Y. Bliokh *et al*., *Nature Photonics* **2**, 748 (2008). K. Y. Bliokh, *J. Opt. A: Pure Appl. Opt.* **11**, 094009 (2009). I. Bialynicki-Birula and Z. Bialynicka-Birula, *Phys. Rev. Lett.* **78**, 2539 (1997). J. C. Hauck and B. Mashhoon, *Ann. Phys. (Berlin)* **12**, 275 (2003). M. Born and E. Wolf, *Principles of Optics*, Pergamon Press, Oxford, 1975, pp. 61–66.

ArXiv

--- abstract: | Interaction networks, consisting of agents linked by their interactions, are ubiquitous accross many disciplines of modern science. Many methods of analysis of interaction networks have been proposed, mainly concentrating on node degree distribution or aiming to discover clusters of agents that are very strongly connected between themselves. These methods are principally based on graph-theory or machine learning. We present a mathematically simple formalism for modelling context-specific information propagation in interaction networks based on random walks. The context is provided by selection of sources and destinations of information and by use of potential functions that direct the flow towards the destinations. We also use the concept of dissipation to model the aging of information as it diffuses from its source. Using examples from yeast protein-protein interaction networks and some of the histone acetyltransferases involved in control of transcription, we demonstrate the utility of the concepts and the mathematical constructs introduced in this paper. bibliography: - 'networks.bib' --- [**Information Flow in Interaction Networks**]{} [Aleksandar Stojmirović and Yi-Kuo Yu[^1]]{} .2in National Center for Biotechnology Information\ National Library of Medicine\ National Institutes of Health\ Bethesda, MD 20894\ United States Introduction ============ Interaction networks are abundant and have recently gained significant publicity in many diverse modern disciplines such as electronics [@CJS01], sociology [@WF94; @Newman04] and epidemiology [@BBPV05]. In its simplest form, an interaction network consists of a collection of entities (or agents), where two agents are linked if they interact in some way. For example, in an acquaintance network [@ASBS00], the agents represent persons and two persons are linked together if they know each other while the Woldwide Web network consists of web pages with links between pages [@BKFP00]. Mathematically, networks correspond exactly to graphs (or multigraphs), with agents as vertices and links as edges, which can be weighted and/or directed depending on the exact application being modeled. The key to analysis of interaction networks is the assumption of information transitivity: information can flow through or can be exchanged via paths of interactions. Biology in post-genomic era also contains numerous examples of molecular networks [@Galitski04]. Metabolic networks have been modeled by representing metabolites as nodes and chemical reactions as links: two metabolites are linked if they participate in the same reaction [@MZ03]. Genetic networks have genes as nodes with two genes being linked if they interact through directed transcriptional regulation [@GBBK02]. Protein-protein interaction networks have proteins as nodes, with the links representing physical interactions (binding) between proteins [@PHJ04]. Large scale high-throughput studies in model organisms such as *Saccharomyces cerevisiae* (baker’s yeast) [@ICOYHS01; @UGCM00], *Drosophilla melanogaster* (fruit-fly) [@GBBC03], *Caenorhabditis elegans* (roundworm) [@LABG04] and humans [@SWLH05; @RVHH05], provided extensive datasets of protein-protein interactions, stored in publicly-available databases such as the Database of Interacting Proteins (DIP) [@XSDHKE02; @SMSPBE04]. Unfortunately, there is very little consistency between the protein-protein interaction data coming from different high-throughput experiments [@SSM03] and significant effort has been expended in devising ways to discover false positives and false negatives [@SSRS06]. This problem is not restricted to protein-protein interactions: microarray data also contains non-negligible inconsistencies [@MM04] . Numerous approaches have been proposed for analysis of biological and, in particular, protein-protein interaction networks [@AS06]. However, due to space restrictions, we will refer to just a few. Most algorithms aim to discover ‘functional modules’ [@HHLM99], representing well connected clusters of nodes with the same or similar function, by using clustering techniques from graph theory and/or machine learning [@SPAD02; @SM03; @RG03; @PEO04; @NJAC05; @XHDZ05; @CSW06; @CY06; @HCZR06]. Very frequently, these techniques make use of additional experimental data which is not present in the network structure itself. For example, methods for discovery of complexes from protein-protein interaction networks often refer to the data from dataset from different species [@KSKS03; @SIKS05; @SSKK05], microarray expression studies [@SPAD02; @CY06], or human-curated functional classifications [@NJAC05; @CSW06]. Our approach to analyzing interaction networks is very different, relying solely on the network structure. We model diffusion of information through the network by discrete-time random walks moving from the nodes representing the sources of information to their destinations. The choice of sources and destinations provides the *context of analysis* with the nodes most affected by information flow being called *Information Transduction Modules*. We use two modes of diffusion, dual to each other, which we call absorbing and emitting, with our absorbing mode directly corresponding to deeply investigated absorbing Markov chains [@KS76]. Random walks and corresponding Markov chains are one of the subjects of spectral graph theory [@Chung97] but we do not use eigenspace decomposition in our work, instead relying on a basic matrix algebra approach similar to that of @KS76. The algorithm *Functional Flow* by @NJAC05, also modeling diffusion of information from sources, is closest to our emitting model. However, to delineate a certain biological context, we additionally direct the flow from sources to selected destinations using potential functions and allow the information content to dissipate (evaporate) from the network at each time step, thus modeling natural ‘aging’ of information. Our models allow investigation of several types of biological questions from protein-protein interaction networks. Many proteins perform their function in cooperation with other proteins through, often large, protein complexes. Thus, to elucidate the function of a given protein, it is useful to know the most likely members of complexes it may belong to and their relations to each other. Additionally, if two proteins are known to have similar function, what, if any, are the proteins they share in their respective complexes? To help answer such questions, we employ our absorbing diffusion mode. The answers to the above questions can provide the general interaction environment of one or more proteins. It is also very instructive to identify specific modules mediating interactions between distant (in network terms) proteins. Our emitting diffusion mode can be used to find possible candidates for members of such modules. Furthermore, analysis of interaction modules obtained from considering different proteins in the same biological context may lead to discovery of fundamental units of information transduction. To achieve this we developed the concept of information interference. More concrete definitions will be presented in the body of the text. This paper is organized as follows. Section 2 outlines the theory behind our models of information diffusion in networks. For better readability, all the theorems and proofs, using mainly the basic concepts and results from the matrix algebra are given in Appendix (the reader may wish to consult the standard linear algebra textbooks such as [@HK71] or [@BR97] for background). Section 3 introduces the methods of analysis of results obtained using the concepts of Section 2, while Section 4 presents concrete examples centered around yeast histone acetyltransferases. We finish with discussion and conclusion in Section 5. Theory ====== Preliminaries ------------- We represent an interaction network as a weighted directed graph $\Gamma=(V,E,w)$ where $V$ is a finite set of vertices of size $n$, $E\subseteq V\times V$ is a set of edges and $w$ is a non-negative real-valued function on $V\times V$ that is positive on $E$, giving the weight of each edge (the weight of non-existing edge is defined to be $0$). Assuming an ordering of vertices in $V$, we represent a real-valued function on $V$ as a state (column) vector ${\mathbf{{\boldsymbol{\varphi}}}}\in{\mathbb{R}}^n$ and the connectivity of $\Gamma$ by the *weight* matrix ${\mathbf{W}}$ where $W_{ij}=w(i,j)$ (the weight of an edge from $i$ to $j$). If $\Gamma$ is an unweighted undirected graph, ${\mathbf{W}}$ is the adjacency matrix of $\Gamma$ where $$W_{ij} = \begin{cases} 2 &\text{if $i=j$ and $(i,i)\in E$},\\ 1 &\text{if $i\neq j$ and $(i,j)\in E$}, \\ 0 &\text{if $(i,j)\not\in E$.}\end{cases}$$ Throughout this paper, we will not make distinction between a vertex $v\in V$ and its corresponding state given by a particular ordering of vertices. Let ${\mathbf{P}}$ denote the $n\times n$ *transition* matrix of $\Gamma$ where $$P_{ij} = \frac{W_{ij}}{\sum_k W_{ik}},$$ that is, ${\mathbf{P}}$ is the weight matrix of $\Gamma$ normalized by row. The matrix ${\mathbf{P}}$ can be used to model random walks on $\Gamma$: for any pair of vertices $i$ and $j$, $P_{ij}$ gives the probability of the random walk moving from vertex $i$ to vertex $j$ in one time step, which is proportional to the weight $W_{ij}$. Since the matrix ${\mathbf{P}}$ is stochastic (all rows sum to unity), it can also be interpreted as the transition matrix for Markov chain on the set $V$. In the following sections we will model information diffusion as a random walk on $\Gamma$ with particular starting and terminating points. Constrained diffusion {#subsec:transition} --------------------- In this section we select certain vertices as sources or sinks of information and solve for the number of times a vertex is visited. Let $S$ denote the set of selected vertices, let $T=V\setminus S$ and let $m={\left\vertT\right\vert}$. Assuming that the first $n-m$ states correspond to vertices in $S$, we write the matrix ${\mathbf{P}}$ in the canonical form: $$\label{eqn:Pcannonical} {\mathbf{P}}=\left[ \begin{array}{cc}{\mathbf{P}}_{SS} & {\mathbf{P}}_{ST}\\ {\mathbf{P}}_{TS} & {\mathbf{P}}_{TT}\end{array}\right].$$ Here ${\mathbf{P}}_{AB}$ denotes a matrix giving probabilities of moving from $A$ to $B$ where $A,B$ stand for either $S$ or $T$. The states (vertices) belonging to the set $T$ are called *transient*. ### Absorption in sinks Suppose now that the set $S$ represents the set of *sinks* of information: any information reaching a sink vertex is absorbed and cannot not leave it. Let ${\mathbf{F}}(t)$ denote an $m \times (n-m)$ matrix such that $F_{ij}(t)$ is the probability that the information originating at $i\in T$ is absorbed at $j\in S$ in $t$ or fewer steps. Since information can only be absorbed once in any state $s\in S$, it follows that the information reaching $j$ avoided all other sinks. For the same reason, $F_{ij}(t)$ can be interpreted as the expected number of visits to the state $j$ of a random walk starting at $i$ for all times up to $t$. Absorption at $j$ after not more than $t$ steps can be achieved in two ways: either the content reached vertex $j$ in the first step, with probability $P_{ij}$ or it moved to some transient vertex $k$ in the first step and was absorbed by $j$ from there in at most $t-1$ steps, with probability $P_{ik}F_{kj}(t-1)$. Therefore, we have for all $t=1,2,\ldots$, $$\label{eqn:sink0} F_{ij}(t+1) = P_{ij} + \sum_{k\in T} P_{ik}F_{kj}(t),$$ or in the matrix form $$\label{eqn:sink1} {\mathbf{F}}(t+1) = {\mathbf{P}}_{TS} + {\mathbf{P}}_{TT}{\mathbf{F}}(t).$$ We solve for the long-term or equilibrium state, where ${\mathbf{F}}(t+1)={\mathbf{F}}(t)={\mathbf{F}}$. In this case, Equation (\[eqn:sink1\]) becomes $$\label{eqn:sink2} {\mathbf{F}} = {\mathbf{P}}_{TS} + {\mathbf{P}}_{TT}{\mathbf{F}},$$ or $$\label{eqn:sink3} ({\mathbb{I}}-{\mathbf{P}}_{TT}){\mathbf{F}} = {\mathbf{P}}_{TS},$$ where ${\mathbb{I}}$ denotes the identity matrix. If ${\mathbb{I}}-{\mathbf{P}}_{TT}$ is invertible, let ${\mathbf{G}}=({\mathbb{I}}-{\mathbf{P}}_{TT})^{-1}$. Equation (\[eqn:sink3\]) then has a unique solution $$\label{eqn:sink4} {\mathbf{F}} = {\mathbf{G}}{\mathbf{P}}_{TS}.$$ ### Diffusion from sources Now consider the dual problem where $S$ is a set of sources of information. Each source emits a unit of information at each time step and no information can enter any source: we assume any information entering a source vanishes. Let ${\mathbf{H}}(t)$ denote an $(n-m) \times m$ matrix such that $H_{ij}(t)$ is the total expected number of times the transient vertex $j$ is visited by a random walk emitted from source $i$ for the time up to $t$. The information emitted from $i$ can arrive at $j$ at time $t$ in two different ways: either the content was emitted from $i$ at time $t$ and reached $j$ directly, or it was emitted at an earlier time step, was located at some transient vertex at time $t-1$ and moved from there to $j$ at time $t$. The former option contributes $P_{ij}$ while the latter contributes $H_{ik}(t-1)P_{kj}$ for all $k\in T$ towards $H_{ij}$. Therefore, we have for all $t=1,2,\ldots$, $$\label{eqn:source0} H_{ij}(t+1) = P_{ij} + \sum_{k\in T} H_{ik}(t)P_{kj},$$ or in the matrix form $$\label{eqn:source1} {\mathbf{H}}(t+1) = {\mathbf{P}}_{ST} + {\mathbf{H}}(t){\mathbf{P}}_{TT}.$$ Similarly to the previous case, we are interested in the steady state, representing the total expected number of visits, where ${\mathbf{H}}(t+1)={\mathbf{H}}(t)={\mathbf{H}}$. In this case, Equation (\[eqn:source1\]) becomes $$\label{eqn:source2} {\mathbf{H}} = {\mathbf{P}}_{ST} + {\mathbf{H}}{\mathbf{P}}_{TT},$$ or $$\label{eqn:source3} {\mathbf{H}}({\mathbb{I}}-{\mathbf{P}}_{TT}) = {\mathbf{P}}_{ST}.$$ If ${\mathbb{I}}-{\mathbf{P}}_{TT}$ is invertible, Equation (\[eqn:source3\]) has a unique solution $$\label{eqn:source4} {\mathbf{H}} = {\mathbf{P}}_{ST}{\mathbf{G}}.$$ ### Existence and interpretation of solutions {#subsubsec:Green} It can immediately be observed that existence of solutions to Equation (\[eqn:source3\]) and Equation (\[eqn:sink3\]) are equivalent: they both depend on the existence of the inverse of ${\mathbb{I}}-{\mathbf{P}}_{TT}$. Specifically, they are special cases of the discrete Laplace equation on $T$ with the Dirichlet boundary condition on $S$ [@Chung97; @CY00]. Given a square matrix ${\mathbf{M}}$, the matrix ${\mathbb{I}}- {\mathbf{M}}$ is often called the *discrete Laplace operator* of ${\mathbf{M}}$. Let $\Delta={\mathbb{I}}- {\mathbf{P}}_{TT}$ ($\Delta$ is the discrete Laplace operator of ${\mathbf{P}}$ restricted to $T$). Equation (\[eqn:sink3\]) can then be written as $$\label{eqn:laplace1} \Delta{\mathbf{F}} = {\mathbf{P}}_{TS}.$$ Denote by ${\mathbf{e}}_k$ the $k$-th standard basis (column) vector of length $n-m$ where $({\mathbf{e}}_k)_j=\delta_{kj}$ ($\delta$ here is the Kronecker’s delta). Let ${\mathbf{f}}_k={\mathbf{F}}{\mathbf{e}}_k$ denote the $k$-th column of ${\mathbf{F}}$ and let ${\mathbf{p}}_k={\mathbf{P}}_{TS}{\mathbf{e}}_k$. Then, solving Equation (\[eqn:laplace1\]) is equivalent to solving the discrete Laplace equation $$\label{eqn:laplace2} \Delta{\mathbf{f}}_k = {\mathbf{p}}_k$$ for all $k\in S$. The standard basis vectors ${\mathbf{e}}_k$ provide exactly the *Dirichlet boundary conditions* on the set $S$ (the set $S$ can be assumed to be a boundary of $T$). It is also easy to see that Equation (\[eqn:source3\]) can be written as $$\label{eqn:laplace3} {\mathbf{H}}\Delta = {\mathbf{P}}_{ST}.$$ Hence, the solution to (\[eqn:laplace3\]) is obtained by solving the discrete Laplace equation in terms of the discrete Laplace operator of the transpose of ${\mathbf{P}}$. The *Green’s function* is defined to be the inverse of the Laplacian. In our case the inverse of $\Delta$ is exactly the matrix ${\mathbf{G}}=({\mathbb{I}}-{\mathbf{P}}_{TT})^{-1}$ and hence the existence of solutions to Equations (\[eqn:source3\]) and (\[eqn:sink3\]) is equivalent to existence of the Green’s functions to the corresponding Laplacian. In the absorbing Markov chain theory [@KS76], the matrix ${\mathbf{G}}$ is known as the *Fundamental matrix* of the corresponding absorbing Markov chain. The entry $G_{ij}$ represents the mean number of times the random walk reaches vertex $j\in T$ having started in state $i\in T$. We now present some elementary sufficient conditions for existence of the Green’s functions of the discrete Laplacians of the graphs. The full proofs are given in Appendix \[app:Green\]. For the development of the discrete Green’s functions (for undirected graphs) in terms of the eigenvalues and eigenfunctions of the Laplacian, we refer the reader to the paper by @CY00. \[prop:specrad0\] Suppose that $\Gamma$ is a weighted directed graph such that for every $p\in T$ there exists $s\in S$ such that there exists a directed path from $p$ to $s$. Then, the matrix ${\mathbb{I}}-{\mathbf{P}}_{TT}$ is invertible and $$\label{eq:invassum0} ({\mathbb{I}}-{\mathbf{P}}_{TT})^{-1} = \sum_{k=0}^\infty ({\mathbf{P}}_{TT})^k.$$ Proposition \[prop:specrad0\] thus guarantees existence of the Green’s functions if every transient vertex can be connected to a source or sink via a directed path. If the underlying graph is undirected, this condition can be rephrased as follows: every connected component of $V$ contains at least one vertex from $S$. In the context of information diffusion, the connectivity condition implies that all information entering the transient set at any specific time must eventually leave it, either by absorption into $S$ when $S$ is a set sinks, or by dissipation when $S$ represents the set of sources. We will further discuss the concept of dissipation in \[sec:dissipation\]. Assuming the Green’s function exists, the entries of the matrices ${\mathbf{F}}$ and ${\mathbf{H}}$ can be interpreted in several different ways. Fundamentally, both $F_{ij}$ and $H_{ij}$ represent the total expected number of times the vertex $j$ is visited by the information originating at the vertex $i$ while avoiding all members of the boundary set $S$ (the proofs are given in Appendix \[app:interpr1\]). It is also clear, by Equation (\[eq:invassum0\]), that ${\mathbf{F}}$ and ${\mathbf{H}}$ are both non-negative matrices and that ${\mathbf{F}}=\lim_{t\to\infty} {\mathbf{F}}(t)$ and ${\mathbf{H}}=\lim_{t\to\infty} {\mathbf{H}}(t)$. In addition, the rows of ${\mathbf{F}}$ all sum to $1$ (Lemma \[lemma:Fstoch\] in Appendix \[app:interpr2\]) and thus $F_{ij}$ is the overall probability an information originating from transient vertex $i$ is absorbed at the sink $j$ while avoiding all other sinks. If we assume that a random walk deposits a fixed amount of information content each time it visits a node, we can interpret $H_{ij}$ is the overall amount of information content originating from the source $i$ deposited at the transient vertex $j$. If $\Gamma$ is an undirected graph with symmetric weight matrix ${\mathbf{W}}$ and $S$ contains a single source, the value of $H_{ij}$ is directly proportional to the degree of the transient vertex $j$ (Appendix \[app:interpr2\]). Hence, in this case, the total average number of times of visits for each transient node is proportional to its degree. This is no longer true if ${\mathbf{W}}$ is not symmetric. Furthermore, we can interpret $F_{ij}$ as the sum of probabilities of paths originating at the vertex $i\in T$ and terminating at the vertex $j\in S$ that avoid all other nodes in the set $S$, and $H_{ij}$ as the sum of probabilities of paths originating at the vertex $i\in S$ and terminating at the vertex $j\in T$, also avoiding all other nodes in the set $S$. Each such path has a finite but unbounded length. However, unlike $F_{ij}$, $H_{ij}$ does not represent a probability because the events of the information being located at $j$ at the times $t$ and $t'$ are not mutually exclusive (a random walk can be at $j$ at time $t$ and revisit it at time $t'$). For $F_{ij}$, the absorbing events at different times are mutually exclusive. Information dissipation {#sec:dissipation} ----------------------- It was mentioned previously that the requirement that every transient node is connected to a node in the set $S$ is effectively equivalent to the property that all information content entering the transient set leaves it at the nodes in $S$. In the present section we extend our model to allow the information to dissipate not only at those nodes but also at the transient nodes. Let ${\boldsymbol{\alpha}}$ and ${\boldsymbol{\beta}}$ be vectors of length $n$ such that for all $i\in V$, $\alpha_i>0$ and $\beta_i>0$. We form the matrix ${\mathbf{\tilde{P}}}$ with entries $$\tilde{P}_{ij} = \alpha_i\beta_jP_{ij},$$ and use the new matrix to compute the matrices ${\mathbf{\tilde{F}}}$ and ${\mathbf{\tilde{H}}}$ by replacing the matrix ${\mathbf{P}}$ in the previous section with $\tilde{P}$ so that. $$\label{eq:dissipate1} {\mathbf{\tilde{F}}} = {\mathbf{\tilde{G}}}{\mathbf{\tilde{P}}}_{TS}.$$ and $$\label{eq:dissipate2} {\mathbf{\tilde{H}}} = {\mathbf{\tilde{P}}}_{ST}{\mathbf{\tilde{G}}}.$$ where ${\mathbf{\tilde{G}}}=({\mathbb{I}}-{\mathbf{\tilde{P}}}_{TT})^{-1}$, provided ${\mathbb{I}}-{\mathbf{\tilde{P}}}_{TT}$ is invertible. The entry $\alpha_i$ gives the proportion of the signal leaving the vertex $i$ that is retained (we call the value of $1-\alpha_i$ the *outgoing dissipation coefficient* of the node $i$) while the entry $\beta_j$ gives the proportion of the signal entering the vertex $j$ that is retained (the value $1-\beta_j$ is called the *incoming dissipation coefficient* of the node $j$). The case where $\alpha_i=\beta_i=1$ for all $i\in V$ gives back the original matrix ${\mathbf{P}}$. Note that our definition allows entries of ${\boldsymbol{\alpha}}$ and ${\boldsymbol{\beta}}$ that are greater than $1$, corresponding to negative dissipation coefficients. Such coefficients lead to amplification of the signal. However, in order for the Green’s function ${\mathbf{\tilde{G}}}$ to exist, any amplification should be balanced by dissipation. We now establish a sufficient condition for existence of ${\mathbf{\tilde{G}}}$. The proof, as well as a discussion of its generalization, is given in Appendix \[app:dissipation\]. \[prop:dissipation\] Let $\alpha_*=\max\{\alpha_i:i\in V\}$ and $\beta_*=\max\{\beta_i:i\in V\}$ and suppose $\alpha_*\beta_*<1$. Then, the matrix ${\mathbb{I}}-{\mathbf{\tilde{P}}}_{TT}$ is invertible and $$\label{eq:invassum1} ({\mathbb{I}}-{\mathbf{\tilde{P}}}_{TT})^{-1} = \sum_{k=0}^\infty ({\mathbf{\tilde{P}}}_{TT})^k.$$ Proposition \[prop:dissipation\] makes no assumptions on the connectivity of the graph: the equilibrium solutions exist regardless of the graph topology. The reason for the removal of the connectivity conditions is that a unit of information originating anywhere in the network has a nonzero probability of being dissipated at each time step and therefore will disappear in the long term, with a portion possibly reaching a sink in the absorbing model. The vectors of coefficients $\alpha$ and $\beta$ provide us with the ability to consider different rates of dissipation at different vertices. We demonstrate the utility of the extended model in examples involving protein-protein interaction networks (Section \[sec:examples\]), where we use vertex specific dissipation to construct ‘evaporating nodes’ that dissipate most of the information coming in but allow unrestricted outward flow. A possible further generalization of this model is for the entries of the vectors ${\boldsymbol{\alpha}}$ and ${\boldsymbol{\beta}}$ to be functions of the state variable of the dynamical system instead of constants. The dynamical system in this case would become non-linear, allowing us to model amplification or dissipation of the information depending on the time specific state of the system. Potentials ---------- Our models so far, including the dissipation modifications described above, model ‘free diffusion’ of information through the network: the likelihood for the signal to move from vertex $i$ to vertex $j$ is proportional to the relative weight of the edge $(i,j)$ among all edges emanating from $i$ (dissipation only affects the total amount transmitted). In order to direct the flow of information towards or away from selected nodes, we adjust the weights of edges of our network graph $\Gamma$ using *potentials*, real-valued monotone functions defined on the nodes that depend on the distances from selected points. Let $\rho$ denote the path-metric on the weighted directed connected graph $\Gamma=(V,E,w)$, where for all $i,j\in V$, $\rho(i,j)$ denotes the sum of the reciprocals of the weights of the edges forming the shortest directed path from $i$ to $j$. Suppose $R$ is a subset of $T$ such that for each $k\in R$ there exists a monotone potential function $\theta_k:{\mathbb{R}}\to{\mathbb{R}}$. For each vertex $j\in V$ define the *total potential* at $j$, denoted $\Theta(j)$ by $$\Theta(j) = \sum_{k\in R} \theta_k(\rho(j,k)).$$ Let $\hat{\Gamma}$ denote the new weighted directed graph $(V,E,\hat{w})$ where $$\label{eq:pseudosinkpt} \hat{W}_{ij} = W_{ij}\exp\left(- \Theta(j)\right).$$ The form of Equation (\[eq:pseudosinkpt\]) ensures that the signal preferentially diffuses from each vertex towards the vertices adjacent to it that have lower potential relative to other adjacent vertices. A vertex $i\in V$ is called a *destination* if $\Theta$ has a minimum at $i$. There can be multiple destinations in a network. The natural candidates for destinations are the members of the set $S$ since all information entering them does not leave them. Some transient states, with the weights of their outgoing edges adjusted to partially accumulate the signal, are also good candidates for destinations. Let $K$ be a subset of $T$ and let $0\leq\gamma\leq 1$. From the already modified graph $\hat{\Gamma}$, we form the graph $\Gamma'$ represented by the weight matrix ${\mathbf{W}}'$ where $$\label{eq:Rgenmat1} W'_{ij} = \begin{cases} \hat{W}_{ij} & \text{if $i\not\in K$,}\\ \gamma \hat{W}_{ij} & \text{if $i\in K$ and $i\neq j$,}\\ \hat{W}_{ij} + (1-\gamma)\sum_{k\neq i} \hat{W}_{ik} & \text{if $j\in K$ and $i = j$}.\end{cases}$$ The effect of this modification is to turn each vertex $i\in K$, called a *pseudosink*, into a partial sink: some proportion of the weights of edges emanating out of $i$ is transferred to the edge pointing back to $i$. The parameter $\gamma$, representing the proportion of information allowed to leave each pseudosink while the remainder is accumulated, is called the *pseudosink leakage coefficient*. The value $\gamma=1$ implies no change in edge weights. The value $\gamma=0$ is a special case because no directed path exists between pseudosinks and source nodes in the resulting graph $\Gamma'$ and Proposition \[prop:specrad0\] does not apply. In this case, there are two possibilities leading to the existence of the Green’s function: either set the outgoing dissipation coefficient of the pseudosinks to something less than $1$, or treat the pseudosinks as parts of the boundary set $S$, as a ‘non-emitting source’ defined in \[subsec:emittingmodel\] below. Note that, while dissipation is applied to the transition matrix ${\mathbf{P}}$, potentials and pseudosinks are applied to the weight matrix ${\mathbf{W}}$ prior to normalization. Since applications of potentials and pseudosinks do not commute, potentials are applied before pseudosinks, although pseudosinks can be potential centers (members of the set $R$). Theoretical Methods for Analysis ================================ In the previous section we introduced the basic concepts related to our models of diffusion of information through networks as well as some modifications to the underlying graph and the transition matrix that lead to biologically realistic models. After all modifications are applied, we obtain the matrices ${\mathbf{\tilde{F}}}$ and ${\mathbf{\tilde{H}}}$, the Green’s functions arising where $S$ represents sinks and sources, respectively. Here we turn to the practical interpretation of these results, which depend on the boundary conditions imposed on the vertices in $S$. Absorbing model --------------- In the case where $S$ represents sinks of information (the *absorbing model*), the entries of the matrix ${\mathbf{\tilde{F}}}$ have a clear probabilistic interpretation: $\tilde{F}_{ij}$ is the probability that information starting at transient vertex $i$ reaches the sink $j$ while avoiding all other sinks, taking into account the dissipation as well as the new weights induced by the potentials. Generally, each sink $j$ exerts a ‘region of influence’, including the transient points with large $\tilde{F}_{ij}$. Depending on the distributions of sinks within the network, some transient node may have a $\tilde{F}_{ij}$ small for all $j$: information emerging from these points is more likely to dissipate than to reach any of the sinks. If $S'\subset S$ is a selection of sink nodes, then $\sum_{j\in S'} F_{ij}$ gives the total probability of information reaching the set $S'$ from the vertex $i$, avoiding all other nodes in $S$. In this context, we call the nodes in $S'$ *explicit sinks* (since we investigate the probabilities of reaching them) and the remaining nodes in $S$ *implicit sinks*, the points that serve as sinks of information but are not considered. Furthermore, if the sinks are treated as general boundary points, with boundary values not restricted to $0$ and $1$, the entries of ${\mathbf{\tilde{F}}}$ can be interpreted as temperatures [@ZBY07]. Emitting model {#subsec:emittingmodel} -------------- Where $S$ represents sources (the *emitting model*), the entries of ${\mathbf{\tilde{H}}}$ can be interpreted as visiting times or as information contents: $\tilde{H}_{ij}$ is the total information content emitted from the source $i$ deposited at the transient vertex $j$. Information is dissipated at all sources and the value of $\tilde{H}_{ij}$ is dependent on transient dissipation coefficients ${\boldsymbol{\alpha}}$ and ${\boldsymbol{\beta}}$ and the potentials. For biological applications, we will consider the case where at least one pseudosink is present in addition to one or several sources, with the potential directing the flow towards the pseudosinks. The distribution of entries of the $i$-th row of ${\mathbf{\tilde{H}}}$ will then describe the *information transduction module* (ITM) involved in transfer of information from $i$ to the pseudosinks, with the nodes with largest entries being most significant. Let ${\boldsymbol{\xi}}$ denote the vector of length ${\left\vertS\right\vert}$ such that for all $i\in S$, $\xi_i\geq 0$. We call $\xi_i$ the *source strength* of the source $i$, representing the amount of information emitted from $i$ at each time step. In this context, we call $i\in S$ an *emitting source* if $\xi_i>0$ and a *non-emitting source* if $\xi_i=0$. Non-emitting sources are essentially information ‘black holes’, dissipating any information coming in and not emitting any. ### Total content For any $i\in S$, let ${\boldsymbol{\epsilon}}_i$ denote the standard $i$-th row basis vector of length $n-m$, where $({\boldsymbol{\epsilon}}_i)_j=\delta_{ij}$. For $x>0$ define the vector ${\boldsymbol{\phi}}_i$ by $${\boldsymbol{\phi}}_i = \xi_i{\boldsymbol{\epsilon}}{\mathbf{\tilde{H}}},$$ that is, ${\boldsymbol{\phi}}_i$ denotes the $i$-th row of ${\mathbf{\tilde{H}}}$ multiplied by $\xi_i$. Its entries give the amount of information content originating from the source $i$ of strength $\xi_i$ deposited at transient vertices. The value of ${\left\Vert{\boldsymbol{\phi}}_i\right\Vert}_1$ is then the total amount of content originating at source $i$ deposited at the transient states. In our examples in the following sections we choose the source strengths ${\boldsymbol{\xi}}$ so that ${\left\Vert{\boldsymbol{\phi}}_i\right\Vert}_1$ is the same for all $i\in S$ (we call the resulting vectors $\phi_i$ normalized content vectors). The *joint information content* vector, denoted ${\boldsymbol{\tau}}$, is defined by $${\boldsymbol{\tau}}= \sum_{i\in S} {\boldsymbol{\phi}}_i.$$ The vector ${\boldsymbol{\tau}}$ implicitly depends on the matrix ${\mathbf{\tilde{H}}}$ and the source strength vector ${\boldsymbol{\xi}}$: we have ${\boldsymbol{\tau}}={\boldsymbol{\xi}}{\mathbf{\tilde{H}}}$. ### Participation ratio Let ${\mathbf{x}}\in{\mathbb{R}}^n$ be any vector and recall that for any $0\leq p<\infty$, the $\ell_p$-norm of ${\mathbf{x}}$, denoted ${\left\Vert{\mathbf{x}}\right\Vert}_p$, is given by ${\left\Vert{\mathbf{x}}\right\Vert}_p = \left(\sum_k {\left\vertx_k\right\vert}^p \right)^{1/p}$. Define the *participation ratio* of ${\mathbf{x}}$, denoted $\pi({\mathbf{x}})$ by $$\pi({\mathbf{x}}) = \frac{{\left\Vert{\mathbf{x}}\right\Vert}_1^2}{{\left\Vert{\mathbf{x}}\right\Vert}_2^2} = \frac{\left(\sum_{k} {\left\vertx_k\right\vert} \right)^2}{\sum_k x_k ^2}.$$ Participation ratio is well known under a slightly different definition in the physics literature [@Thouless74]. It gives the number of components of ${\mathbf{x}}$ whose magnitude is ‘significant’. Clearly, $\pi$ is independent of the scale of ${\mathbf{x}}$: we have for any $\lambda>0$, $\pi(\lambda{\mathbf{x}})=\pi({\mathbf{x}})$. We illustrate the usage by examples. Let ${\mathbf{x}}=[1,1,1,1,1]$. Then, $\pi({\mathbf{x}})=\frac{5^2}{5}=5$. All components are equally significant and this is reflected in the participation ratio. Now consider ${\mathbf{x}}=[1,1,0,0,0]$. We have, $\pi({\mathbf{x}})=\frac{2^2}{2}=2$. Only the first two components are non-zero and are of equal magnitude. Finally, let ${\mathbf{x}}=\left[1,\frac12,\frac14,\frac18,\frac{1}{16}\right]$. We obtain $\pi({\mathbf{x}})\approx 2.8181$. Here all five components are non-zero but their magnitudes differ significantly. The participation ratio here implies that the first two components and to a large extent the third are significant while the remaining two are much smaller. In our biological examples, we use $\pi({\boldsymbol{\tau}})$ to choose the number of the transient vertices with largest total mass to display as a ‘significant’ subgraph, together with all sources and pseudosinks. ### Interference Given the vector of source strengths ${\boldsymbol{\xi}}$, the entry of $\tau_j$ can be interpreted as providing the total amount of information deposited at the vertex $j$. It is also possible to investigate the interaction of the signals from different sources using the concept of destructive interference. For any vector ${\mathbf{x}}\in{\mathbb{R}}^n$, let $\mu$ denote an *interference function* such that $0\leq \mu({\mathbf{x}})\leq {\left\Vert{\mathbf{x}}\right\Vert}_1$. When applied to a vector containing information content from different sources, interference function is interpreted as removing some of the information present due to the interaction of the various information types and returning the remaining information content. Interference functions can take various forms depending on the nature of the types of information in each application. Suppose ${\mathbf{x}}$ consists of two components representing information types that are assumed to completely cancel out each other. In this case, the interference function takes the form $\mu({\mathbf{x}}) = {\left\vertx_1-x_2\right\vert}$. When ${\mathbf{x}}$ has more than two components, there are may possible ways to generalize the above example. We distinguish two general modes of interference: exclusive and partial. Exclusive interference mode represents the case where simultaneous presence of all types of information is necessary for destructive interference. For example, if each information type carries the same weight, the interference function is: $$\label{eq:interf1} \mu({\mathbf{x}}) = \sum_{k} \left(x_k - \nu\right),$$ where $\displaystyle\nu = \min_{k} x_k$. We call the partial interference the case where presence of all types of information is not necessary. It can be modeled in many ways depending on the desired interpretation. For example, if there are three sources, we can use complex numbers to set $\mu$ so that $$\mu({\mathbf{x}}) = {\left\vert\sum_{k=1}^3 x_k\exp\left(\frac{\iota k\pi}{3} \right)\right\vert},$$ where $\iota$ denotes the imaginary unit. In this case, some content is lost when any two types of signal are present but all three must be present for complete annihilation. Given the interference function $\mu$, define the *interference strength function* $\psi:{\mathbb{R}}^n\to{\mathbb{R}}\cup\{\infty\}$ by $$\label{eq:interfstr1} \psi({\mathbf{x}}) = \begin{cases} {\left\Vert{\mathbf{x}}\right\Vert}_1 \log\left(\frac{{\left\Vert{\mathbf{x}}\right\Vert}_1}{\mu({\mathbf{x}})}\right) & \text{if ${\left\Vert{\mathbf{x}}\right\Vert}_1>0$,}\\ 0 & \text{if ${\left\Vert{\mathbf{x}}\right\Vert}_1=0$.} \end{cases}$$ By the definition of $\mu$ Since $0\leq \mu({\mathbf{x}})\leq {\left\Vert{\mathbf{x}}\right\Vert}_1$, it follows that $\psi$ takes non-negative values (including $+\infty$). The value of $\psi$ is infinite if $\mu({\mathbf{x}})=0$ (perfect interference) and finite otherwise. For an $m\times n$ matrix ${\mathbf{X}}$ define the vector ${\boldsymbol{\sigma}}({\mathbf{X}})$ of length $n$ having the components $$\label{eq:interfstr2} \sigma_i({\mathbf{X}}) = \psi({\mathbf{X}}{\mathbf{e}}_i)$$ (recall that ${\mathbf{e}}_i$ is the standard column basis vector and hence ${\mathbf{X}}{\mathbf{e}}_i$ represents the $i$-th column of ${\mathbf{X}}$). We will call ${\boldsymbol{\sigma}}$ the *interference strength vector*. For our applications, the entries of the matrix ${\mathbf{X}}$ above are interpreted as information contents over some graph: $X_{ij}$ is the the content of type $i$ at the vertex $j$. For each node $j$, the $\ell_1$-norm in Equation (\[eq:interfstr1\]) can be interpreted in this context as the total information content at $j$ and the value of $\mu$ applied to the $j$-th column of ${\mathbf{X}}$ as the information content remaining after interference. Hence, interference strength of each node measures how much information content was lost by interference, adjusted by the node’s joint information content. The matrix ${\mathbf{\tilde{H}}}$ is therefore a natural input to $\psi$ and ${\boldsymbol{\sigma}}$, however other derived matrices can be used such as ${\mathbf{\tilde{H}}}$ adjusted for source strength by multiplying each row by its corresponding source strength $\xi_i$. Furthermore, rows of ${\mathbf{X}}$ can come from different ${\mathbf{\tilde{H}}}$ matrices, using different potentials or dissipation coefficients, as long as the underlying vertex set is the same. The general purpose of interference strength is to measure the amount of interaction or overlap between different ITMs. Biological Examples {#sec:methods} =================== \[sec:examples\] The theory and methods outlined in previous sections can be applied to any interaction network. This section will present some examples using biological networks, more specifically, yeast protein-protein interaction networks. Since the interaction data obtained using many high-throughput methods is generally inconsistent [@SSM03], we use the core yeast dataset from DIP, version ScereCR20060402, consisting of 2554 proteins and 5952 interactions for all our examples. The core dataset, obtained using the methods of @DSXE02, contains only the most reliable interactions from the DIP dataset of all yeast protein-protein interactions. Our examples are restricted to investigation of information transduction modules related to yeast histone acetyltransferases (HATs). Histones are nuclear proteins that are major components of eukaryotic chromatin [@Wolffe92]: eukaryotic DNA is organized as a repeating array of nucleosomes consisting of 146 bp of DNA wound around a histone octamer consisting of two of each of histone proteins H2A (Hta1, Hta2 in yeast), H2B (Htb1, Htb2 in yeast), H3 (Hht1, Hht2 in yeast) and H4 (Hhf1, Hhf2 in yeast). It has been repeatedly demonstrated that transcription is strongly influenced by the chromatin structure and DNA-histone interactions in particular. The regions of DNA that interact with histones are generally unavailable for transcription and transcriptional activation and deactivation are connected with chromatin alterations [@Wolffe01]. Histone acetyltransferases are enzymes that acetylate histones, leading to weakening of the nucleosome structure and making the DNA involved accessible to transcription factors [@Struhl98; @WK98]. *Saccharomyces cerevisiae* contains several HATs from two major classes with a variety of biological functions and substrate specificities [@SB00]. The proteins Hat1, Gcn5, Elp3, Spt10 and Hpa2 belong to the GNAT superfamily [@NL97], while Esa1, Sas2 and Sas3 belong to the MYST family [@BSABBCCDDFHMVWH96; @SEGSPZCLA98]. The proteins TAF1 (TATA-binding protein associated factor), a subunit of the TFIID complex, and Nut1 (Med5), a subunit of the mediator complex [@BY05], have also been associated with histone acetyltransferase activity [@MYKBBOWWBKNA96; @LBGMK00]. Unfortunately, the core dataset does not contain the relevant data for all known HATs. The HATs Hpa2 and Spt10 are not present in the core while HAT1 has interactions only with Hat2 and its substrate Hhf2. We chose to primarily concentrate on HATs Gcn5, Esa1 and Elp3 because they are well researched and the interaction data is abundant. They are all involved in transcriptional activation, unlike Sas2, which promotes silencing [@OSMBYSW01]. Gcn5 is the best characterized of all HATs, preferentially acetylating histone H3 [@SS99]. It forms the catalytic subunit of the ADA and SAGA transcriptional activation complexes [@GDCRBCOOAWBW97]. In addition to Gcn5, the SAGA complex also contains the proteins Tra1, TAF5, TAF6, TAF9, TAF10, TAF12, Hfi1 (Ada1), Ada2, Ngg1 (Ada3), Spt3, Spt7, Spt8 and Spt20 (Ada5) [@TT05]. The ADA complex contains a subset of proteins from the SAGA complex, namely Gcn5, Hfi1, Ada2, Ngg1 and Spt20, plus the adaptor protein Ahc1 [@ESSHYBW99]. The TAF proteins in SAGA also belong to the TFIID complex, which overall consists of 15 subunits including a TATA-binding protein and 14 TAFs [@SW00]. Esa1 is the catalytic subunit of the NuA4 histone acetyltransferase complex essential for growth in yeast [@SEGSPZCLA98; @AUSCGBPWC99] that catalyses acetlyaltion of the histone H4. It has been established that the NuA4 complex, containing, in addition to Esa1, the proteins Tra1, Epl1 Yng2, Eaf1, Eaf2, Eaf3, Eaf5, Eaf6, Act1, Arp4 and Yaf9, is recruited by a variety of transcriptional complexes as a transcriptional coactivator and is involved in DNA repair [@DC04]. Elp3 is a part of the six component elongator complex , which is associated with RNA polymerase II during transcript elongation [@WOBFEOLATS99]. The elongator complex also includes the proteins Iki3 (Elp1), Elp2–4, Iki1 (Elp5) and Elp6 [@KG01]. This section contains four examples of the application of our models, depicted in Figures \[fig:fig1\]–\[fig:emitting2b\]. Subsection \[subsec:absorbing\] describes possible complexes associated with the HATs Gcn5, Esa1 and Elp3, taken individually and in competition, that can be inferred from the protein-protein interaction network using the absorbing model. Subsection \[subsec:emitting\] investigates possible physical interaction interfaces between the MADS box protein Mcm1 [@SS95] and the HATs Esa1 and Gcn5. In this case, the emitting model is employed to discover the pathways through which Mcm1 can recruit the above HATs and whether they are recruited through the same interface. Before presenting our results we describe the model parameters and computational techniques used. Parameters and computation -------------------------- ### Dissipation For all our examples, we set $\alpha_i=1$ for every node $i$ in our interaction network so that the outgoing flow from any node is not dissipated. Modeling the incoming dissipation the coefficients $\beta_i$ can take two values: one for ‘ordinary’ and one for *evaporating* vertices. In our examples that use the absorbing model (\[subsec:absorbing\]), $\beta_i$ is set to $0.70$ for ordinary nodes and $0.01$ for evaporating nodes while the examples using the emitting model (\[subsec:emitting\]) set $0.87$ for ordinary nodes and $0.01$ for evaporating nodes. The evaporating nodes consisted of cytoskeleton proteins Act1, Myo1, Myo2, Myo3, Myo4, Myo5, Smy1, Smy2, Sla1, Arc40, Arp2, Rvs167, Tpm1, Tpm2, Aip1 and Las17 and histones (Hta1, Hta2, Htb1, Htb2, Hht1, Hhf2, Htz1, Hho1). The coefficients for the ordinary nodes were chosen using the following reasoning. For the emitting model we considered the dissipation rate that would allow the random walk emitted from the source to reach an ‘average’ node along the shortest path to it with the probability slightly less than $0.5$, say $0.49$. We found that the average length of the shortest path between two points in the yeast core dataset is $5.23$ and hence our coefficient is $0.49^{(1/5.23)}=0.872$, which is rounded to $0.87$. A different coefficient was needed for the absorbing examples because we were interested in only the immediate complexes containing our selected HATs: the coefficient $\beta_i=0.87$ would lead to most of the members of the RNA polymerase II holoenzyme to be retrieved as members of the resulting ITM. We chose to consider the shortest paths of length $2$, rather than of the average length $5.23$. Using the same calculation as above, we obtain $0.49^{(1/2)}=0.7$. The reason for having evaporating nodes with larger dissipation rate is that both the cytoskeleton proteins and the histones form extended structures in the cell and the nucleus, respectively. In our physical interaction network, we assume that information can flow from one protein to another through an intermediate node if all three nodes are brought close together in space and time. Information is not likely to flow through proteins that are parts of extended structures because proteins with completely different biological function may bind them at different locations and at different times. Therefore, allowing significant information flow through such nodes would yield biologically implausible results. However, depending on the exact context of the investigation, such nodes may have an important role to play and removing them completely from the interaction networks or assigning them to the boundary set $S$ would not be appropriate. Hence, we set a very high incoming dissipation rate at evaporating nodes while allowing the information to originate from them. In terms of our models, this approach means that the evaporating nodes will have very small visiting times in the emitting models and hence will not be components of any ITM. On the other hand, depending on the exact network topology, they may be part of ITMs obtained by the emitting model. Note that other proteins that bind their interacting partners in a non space and time specific manner can be chosen as additional evaporating nodes; we chose histones and cytoskeleton proteins due to their direct relevance to our selected examples. ### Potentials All our examples use attracting potentials centered at each pseudosink or sink. The potential function, heuristic in nature, is the same in every example has the the form $$\label{eq:pseudosinkpt2} \theta_k(x) = \begin{cases} a_1x & \text{if $0<x\leq b$,}\\ a_1x + a_2(x-b)^2 & \text{if $x>b$,} \end{cases}$$ where $a_1=0.8181$, $a_2= 0.05$, $b=2$ and $k$ is any pseudosink or a sink. The potential function shown above is long-range, affecting the whole graph, with a linear portion for short ranges $0\leq x\leq 2$ and quadratic for distances larger than $2$. We do not expect to see qualitative changes in the results if the form of the potential function is modified as long as it has the effect of attracting information towards the destination. The sources (in the case of emitting models) and evaporating points were excluded from the graph prior to calculating distances (their distances from the centers were set to an arbitrary large number) in order to exclude the paths passing through them from consideration. The reason for excluding the paths passing through sources was that, by construction, the information never enters a source from a transient vertex, while the evaporating points were excluded because most of the signal entering them is dissipated. ### Numerical implementation The code for computation of the results was implemented in the Python programming language, using the NumPy and SciPy packages [@JEP01]. In particular, the computation of the matrices ${\mathbf{\tilde{F}}}$ and ${\mathbf{\tilde{G}}}$ (Equations (\[eq:dissipate1\]–\[eq:dissipate2\])) was performed by the embedded FORTRAN code from the UMFPACK [@Davis04] solver of sparse systems of linear equations, using the Automatically Tuned Linear Algebra Software (ATLAS) [@WP05] implementation of Basic Linear Algebra Subprograms (BLAS). The graphical representations of the subgraphs of interest were produced by the *neato* program from the Graphviz graph visualization suite [@GN00]. HAT complexes: absorbing examples {#subsec:absorbing} --------------------------------- --------- --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- **(a)** ![ITMs obtained by running the absorbing model with Esa1(a), Gcn5(b) and Elp3(c) as a sink. The shades of grey at the nodes represent the probability of the information originating at the corresponding protein being absorbed at the sink, the darker nodes indicating higher probability.[]{data-label="fig:fig1"}](graph40.eps "fig:") **(b)** ![ITMs obtained by running the absorbing model with Esa1(a), Gcn5(b) and Elp3(c) as a sink. The shades of grey at the nodes represent the probability of the information originating at the corresponding protein being absorbed at the sink, the darker nodes indicating higher probability.[]{data-label="fig:fig1"}](graph41.eps "fig:") **(c)** ![ITMs obtained by running the absorbing model with Esa1(a), Gcn5(b) and Elp3(c) as a sink. The shades of grey at the nodes represent the probability of the information originating at the corresponding protein being absorbed at the sink, the darker nodes indicating higher probability.[]{data-label="fig:fig1"}](graph42.eps "fig:") --------- --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Figure \[fig:fig1\] shows the three subgraphs of the yeast core interaction graph consisting of the top scoring nodes according to the absorbing model with Esa1, Gcn5 and Elp3 as single sinks, respectively. The information orginating at the proteins shown has more than $0.07$ probability of being absorbed by the sink (under the influence of the potential centered at the sink) as opposed to being dissipated. Hence, the subgraphs show the proteins that are likely to be in the same complex with the HATs chosen as sinks. Figure \[fig:fig1\](a), with Esa1 as the sink, shows all the proteins from the NuA4 complex that are available in the core dataset as highly significant. Some of the proteins from ADA and SAGA complexes can also be seen because Tra1 belongs to these complexes as well as to NuA4. The four types of histones forming the histone octamer can also be seen interacting with Arp4. The proteins Vps51–54 on the right of Figure \[fig:fig1\](a) belong to the Vps Fifty-three thethering (VFT) complex, involved in vesicle assembly [@RWSSK03]. The proteins Tlg1 and Ypt6 are interacting partners of the VFT complex [@RWSSK03]. The relation between VFT and NuA4 is not established as these two complexes are localized in different cellular compartments: NuA4 in the nucleus and VFT in golgi-vacuole transport vesicles. The relationship observed in Figure \[fig:fig1\](a) results exclusively from the Yng2–Vps51 interaction, which was orginally observed in a yeast-two-hybrid screen by @ITMOCNYKS00 [@ICOYHS01]. Based on the above information, it appears that VFT and NuA4 complexes do not interact *in vivo*. Note that the histones as well as actin, although selected as evaporating points, can be seen in the figure because the outgoing flow from evaporating nodes is allowed. In a similar fashion, Figure \[fig:fig1\](b), with Gcn5 as the sink, shows the members of SAGA, ADA and TFIID transcriptional activator complexes as well as many other transcription factors, mostly members of subcomplexes of the RNA polymerase II holoenzyme. Also worth mentioning is Cti6, which bridges the Cyc8-Tup1 corepressor and the SAGA coactivator to overcome repression of the GAL1 gene [@PPKTT02]. The Cyc8 protein is also shown while Tup1 is not, most likely because it is involved in many other interactions away from Gcn5, bringing down its relative significance. Figure \[fig:fig1\](c), with Elp3 as the sink, clearly outlines the elongator complex, as well as some members of the core RNA polymerase II complex (Rbp2–5, Rbp7, Rpc10, Rpo26) [@MY98]. Figure \[fig:fig2\] shows the top scoring nodes according to the absorbing model with Esa1, Gcn5 and Elp3 as simultaneous sinks with attracting potentials. In this case, the information originating at the depicted nodes has more than $0.05$ total probability of being absorbed by any of the sinks as opposed to being dissipated. Fewer nodes can be seen in this figure as compared to Figure \[fig:fig1\] because the three attracting potentials are now involved that may cancel each other out. It can be seen that the elongator complex centered around Elp3 is not connected to the subgraph around Esa1 and Gcn5. Although all of the NuA4, SAGA, ADA and elongator complexes belong to the RNA polymerase II holoenzyme, they do so at different times. The NuA4, ADA and SAGA complexes have a role in initiation of transcription while the elongator complex is involved in transcript elongation [@Martinez02]. The green (mixture of cyan and yellow) color of Tra1 is indicative of the fact that it is a subunit of both Esa1-containing NuA4 complex and the Gcn5-containing SAGA complex. ![ITM obtained by running the absorbing model with Esa1, Gcn5 and Elp3 as simultaneous sinks. The strength of each of cyan, yellow and magenta color component of the node shows the square root of the probability of absorption at Esa1, Gcn5 and Elp3, respectively.[]{data-label="fig:fig2"}](graph60.eps) Transcription factor interaction interfaces; emitting examples {#subsec:emitting} -------------------------------------------------------------- Mcm1 is a yeast transcription factor essential for cell viability. It controls many cellular functions including cell cycle transition [@ASWN95], mating [@MBGSAV02] and arginine metabolism [@MD93], through interactions with different cofactors. It has been determined that Mcm1 acts both as an activator and a repressor of transcription [@BHS92; @MD93] and here we explore the possible ways it can interact with the NuA4 and SAGA HAT complexes. --------- ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- **(a)** ![ITMs resulting from the emitting model with Mcm1 as a source and Esa1 as a pseudosink using the original yeast core dataset (a) and the modified dataset additionally including the edges Tra1–Gal4 and Tra1–Gcn4 (b). The proteins containing the largest amounts of deposited information are shown, with the information content indicated by shading (darkest nodes contain the most information).[]{data-label="fig:emitting1"}](graph10.eps "fig:") **(b)** ![ITMs resulting from the emitting model with Mcm1 as a source and Esa1 as a pseudosink using the original yeast core dataset (a) and the modified dataset additionally including the edges Tra1–Gal4 and Tra1–Gcn4 (b). The proteins containing the largest amounts of deposited information are shown, with the information content indicated by shading (darkest nodes contain the most information).[]{data-label="fig:emitting1"}](graph14.eps "fig:") --------- ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Figure \[fig:emitting1\](a) shows the subgraph consisting of the $22$ proteins with the largest deposited information content obtained by running our emitting model with Mcm1 as a source and Esa1 as a pseudosink. The number of proteins to display ($20$ plus the source and the pseudosink) was chosen because the participation ratio for the information content vector (excluding the source and the pseudosink) was $20.33$. The ITM shown in Figure \[fig:emitting1\](a) gives the likely pathways of physical interaction from Mcm1 to Esa1, according to the yeast core interaction dataset. It can be immediately observed that Esa1 is reached solely through Tra1, which is known to be the general interaction domain of both NuA4 and SAGA HAT complexes [@AUSCGBPWC99; @GSPY98]. Directly associated with Mcm1 are the proteins Arg80–Arg82, belonging to the ArgR complex involved in regulation of arginine metabolism [@DM91]. The majority of the ITM is dominated by the members of the SRB mediator subcomplex of the RNA polymerase II holoenzyme (Srb2, Srb4, Srb7) [@BY05] and the TFIID, SAGA and ADA complexes. Also prominent are transcriptional activators Gal4 and Gcn4 [@Hinnebusch05; @TJS06]. The subgraph image suggests two possible interaction pathways: the main (based on the intensities of deposited information) through Srb4 and members of SAGA/ADA complex and the alternative through Ume6–TAF10–Spt7. Ume6 is a DNA binding protein that acts as a transcriptional repressor by recruiting histone deacetylases, which have the catalytic activity opposite to the HATs [@KABR03]. While simultaneous existence of activating and repressing pathways is biologically plausible, we do not anticipate both pathways to be in action at the same time. On the other hand, interaction of Mcm1 with the NuA4 through any of the above pathways *in vivo* is doubtful because both pathways lead through the interacting partners of Tra1 in the SAGA complex that are not associated with it in the NuA4 complex [@DC04; @TT05]. Note that the direct physical interaction of the ArgR/Mcm1 complex and the SAGA complex was hypothesized by @RGB02 in relation to regulation of arginine metabolism. Nevertheless, it is likely that the yeast core dataset does not contain all the interactions of Tra1 and that the interactions not in the dataset may provide us with the plausible explanation. @BHSA01 have indicated that HAT complexes are recruited through Tra1 by Gal4 and Gcn4 transcriptional activators. To investigate if adding the implied edges would significantly change the resulting ITM we added the Gcn4–Tra1 and Gal4–Tra1 links to the core dataset and rerun the emitting model with all other parameters unchanged. The resulting ITM, with participation ratio of $21.66$, is shown in Figure \[fig:emitting1\](b). We observe few changes: the proteins Ssn3, Srb5, Srb6 and Gal11, belonging to the mediator complex, replaced Cti6 and Srb7, thus placing more emphasis to the mediator complex. In this example, our emitting model appears to be quite robust to changes in the pseudosink leakage parameter $\gamma$. Using the original core dataset, in addition to the original run with $\gamma=0.3$, we ran our model with $\gamma=0$, $\gamma=0.5$ and $\gamma=1$, obtaining participation ratios of $19.43$, $20.34$ and $20.75$ and very little change in constitution of the ITMs. For example, when $\gamma=1$, the new ITM contains the NuA4 proteins Arp4 and Yng2 in the place of Cti6 and Srb7. Hence, larger pseudosink leakage coefficient allows exploration of the nodes surrounding the pseudosinks without affecting the remainder of the ITM in a major way. Such exploration is very desirable for protein-protein interaction networks because it reveals more of the complexes around pseudosinks, thus giving some of the characteristics of the absorbing model to the emitting model. Note that many of the interacting partners of the sources are found in the ITM solely due to proximity of the source. --------- --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- **(a)** ![ITM resulting from the emitting model with Esa1 and Gcn5 as sources and Mcm1 as a pseudosink: (a) information content, (b) interference strength.[]{data-label="fig:emitting2a"}](graph30.eps "fig:") **(b)** ![ITM resulting from the emitting model with Esa1 and Gcn5 as sources and Mcm1 as a pseudosink: (a) information content, (b) interference strength.[]{data-label="fig:emitting2a"}](graph31.eps "fig:") --------- --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- To explore the extend the HATs Esa1 and Gcn5 share their interaction interface with Mcm1 we set Esa1 and Gcn5 as sources and Mcm1 as a pseudosink destination. Figure \[fig:emitting2a\] shows the ITM based on the total information content (participation ratio $24.62$, 28 nodes shown), with the nodes shaded according to total content and interference strength. The proteins shown as nodes in Figure \[fig:emitting2a\] have appeared in one of the previous figures, mostly forming parts of NuA4, SAGA/ADA, TFIID and mediator complexes. The nodes with the largest total content are Tra1, Ada2, Ngg1 and Srb4 and the latter three are also the nodes with by far the largest interference strength. This fact does not surprise us because although Tra1 is a member of both NuA4 and SAGA complexes, information flowing from Gcn5 to Mcm1 largely avoids it. The paths used by the information emitted from Esa1 and Gcn5 separately can best be seen in a color figure (Figure \[fig:emitting2b\](a)) where the information content from Esa1 and Gcn5 is shown as cyan and yellow, respectively. The nodes colored strongly cyan contain mostly information from Esa1 while those colored yellow contain mostly the information from Gcn5. The nodes colored green contain information from both sources. In this way it can be observed that members of NuA4 contain the information solely from Esa1, some SAGA proteins contain the information solely from Gcn5, while Ada2, Ngg1 and Srb4 contain a significant amount of information from both sources. --------- ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- **(a)** ![Information content of members of the ITM arising from the emitting model with Esa1 and Gcn5 as sources and Mcm1 as a pseudosink: (a) using the yeast core dataset; (b) using the modified dataset additionally including the edges Tra1–Gal4 and Tra1–Gcn4. The strength of the cyan and yellow color component of the node corresponds to the information content originating from Esa1 and Gcn5, respectively.[]{data-label="fig:emitting2b"}](graph32.eps "fig:") **(b)** ![Information content of members of the ITM arising from the emitting model with Esa1 and Gcn5 as sources and Mcm1 as a pseudosink: (a) using the yeast core dataset; (b) using the modified dataset additionally including the edges Tra1–Gal4 and Tra1–Gcn4. The strength of the cyan and yellow color component of the node corresponds to the information content originating from Esa1 and Gcn5, respectively.[]{data-label="fig:emitting2b"}](graph33.eps "fig:") --------- ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Using additional links based on Brown *et al.* (Figure \[fig:emitting2b\](b)) produces effects similar to Figure \[fig:emitting1\](b): the common interface through the mediator complex is emphasized at the expense of the paths through the SAGA complex. For example, note the difference in color of Spt7, Gcn4 and Gal4 between Figure \[fig:emitting2b\](a) and Figure \[fig:emitting2b\](b). The common interface through the mediator complex appears biologically more plausible than directly through members of the SAGA complex but we are as yet unable to find direct evidence in the literature confirming either possibility. Discussion and conclusion ========================= The proposed information diffusion models appear to capture some of the essential features of the yeast protein-protein interaction network in our examples. Our absorbing model performed well in identifying complexes related to sinks while the emitting model with pseudosinks is able to illuminate the possible interaction interfaces between sources and pseudosinks. Application of the concept of destructive interference in this context provides a way to assess the degree of overlap of different ITMs. The salient feature of our models is a novel use of attraction potentials and dissipation. While the entries of the Green’s function can be interpreted in graph-theoretic terms as sums of weights of paths from a source to a transient vertex (for the emitting model) or from a transient vertex to a sink (for the absorbing model), the potentials, together with the choice of boundary, provide a unique context for information diffusion in the network. The weights of the edges and hence the nature of the underlying graphs are changed every time a different potential is applied, thus bringing forward different aspects of the network. The potential function used for our examples was heuristic in nature and we hope that our work would generate interest in developing theoretical foundations for directed information propagation through networks. Dissipation coefficients provide a natural and extremely flexible way of controlling the spread of information content through the network. While @GN02 proposed a similar formulation for penalizing longer paths connecting two nodes in a network, they did so in the context of hierarchical clustering and using a single dissipation rate. Node specific dissipation rates are important because they allow construction of ‘evaporating nodes’ and possible integration of additional information to our model. Having the dissipation rates dependent on the environment of the node may lead to a more sophisticated model of information transduction. When modelling physical cellular protein networks, the main limitation of our approach is that the the publicly available representations of protein-protein interaction networks contain a limited amount of information. Each interaction is shown as either occurring or not occurring, without reference to the dynamics, time-scale, or specificity of binding. Furthermore, the spatial location of the interactions on the protein molecules is not available, so that it cannot be determined if a protein known to belong to two separate complexes, such as Tra1 in our examples, can belong to both at the same time and therefore transmit information between them. Therefore, our model of protein cellular networks is only metaphorical at this stage. However, our diffusion paradigm can be adapted to account for additional information about proteins, such as their concentrations, cellular compartment localizations, post-translational modifications or rate constants for binding interactions, as it becomes available. One way to do that is to associate each protein to a vector instead of a scalar value and to construct an evolution operator that reflects the nature of the additional information. In such circusmstances, the dynamics of information flow could be as revealing as the steady state we use at this stage. The quality of the interaction dataset also has a strong influence to the outcomes of our models. Addition or deletion of edges may make the results more realistic, as in our emitting examples, but also may completely alter the ITM produced, if a particular edge provides a shortcut towards the destination. Hence, in order to obtain the results useful in field of application, it is imperative to use datasets of interactions that precisely reflect the network being investigated. In the case of yeast protein-protein interactions, @CKZG07 were recently able to derive a significantly more reliable collection of interactions, primarily based on two large-scale studies of protein complexes by tandem affinity purification of complexes followed by mass spectroscopic identification of individual proteins [@GAGK06; @KCYZ06]. It is interesting that the same transcriptional complexes encountered in our examples are prominent in the unified physical interactome map presented by @CKZG07. The problem of ‘shortcuts’ through the network was also observed by @SPAD02, who completely eliminated certain nodes in their effort to model signal transduction pathways using the yeast protein-protein interactions. Our evaporating nodes, with a very large incoming dissipation rate, have a similar role with an added advantage that they can be visible as parts of complexes observed using the absorbing model. The list of evaporating nodes used by us is not exhaustive and it would be necessary to add further classes of proteins to it for large-scale investigations of the yeast protein interactome using our methods. In this paper, we introduced a flexible mathematical framework for analysis of interaction networks and indicated its utility by examples. We believe that the ability to select a particular context for information propagation by setting various model parameters will be extremely useful for addressing questions involving interaction networks in biology and many other disciplines. Acknowledgments =============== We thank Drs John Wootton and David Landsman for encouragement and comments. This work was supported by the Intramural Research Program of the National Library of Medicine at National Institutes of Health. Existence of Green’s Functions {#app:Green} ============================== In this appendix we provide the elementary proofs of the results about existence of the Green’s functions stated in the main text. As before, $\Gamma=(V,E,w)$ denotes a weighted directed graph with $N$ vertices, with the weight matrix ${\mathbf{W}}$ and transition matrix ${\mathbf{P}}$. We also have $T\subset V$ and $S=V\setminus T$. Recall that for every matrix ${\mathbf{M}}$, the induced $\ell_\infty$ norm of ${\mathbf{M}}$, written ${\left\Vert{\mathbf{M}}\right\Vert}_\infty$, is defined by $${\left\Vert{\mathbf{M}}\right\Vert}_\infty=\sup_{{\mathbf{x}}\in{\mathbb{R}}^n}\frac{{\left\Vert{\mathbf{M}}{\mathbf{x}}\right\Vert}_\infty}{{\left\Vert{\mathbf{x}}\right\Vert}_\infty},$$ where ${\left\Vert{\mathbf{x}}\right\Vert}_\infty=\max_i {\left\vertx_i\right\vert}$. One can easily show that $$\label{eqn:l1norm} {\left\Vert{\mathbf{M}}\right\Vert}_\infty = \max_i \sum_j {\left\vertM_{ij}\right\vert}.$$ Also recall that the spectral radius of a square matrix ${\mathbf{M}}$ is defined to be the largest absolute value of its eigenvalues. It is well known that that for every eigenvalue $\lambda$ of ${\mathbf{M}}$ and any $k=1,2,\ldots$, $$\label{eq:specrad} {\left\vert\lambda\right\vert} \leq {\left\Vert{\mathbf{M}}^k\right\Vert}_\infty^{1/k}.$$ \[lemma:mat\] Let ${\mathbf{M}}$ be a square matrix with the spectral radius strictly less than $1$. Then, (i) ${\mathbf{M}}^k\to {\mathbf{0}}$ as $k\to\infty$, (ii) The matrix ${\mathbb{I}}-{\mathbf{M}}$ is invertible and $({\mathbb{I}}-{\mathbf{M}})^{-1} = \sum_{k=0}^\infty {\mathbf{M}}^k$. By the Jordan matrix decomposition, we can write ${\mathbf{M}}={\mathbf{V}}\boldsymbol{\Lambda}{\mathbf{V}}^{-1}$ for some matrix ${\mathbf{V}}$, where $\boldsymbol{\Lambda}$ is a block-diagonal matrix of the form $$\boldsymbol{\Lambda}=\left[\begin{array}{cccc} {\mathbf{B}}_1 & {\mathbf{0}} & \cdots & {\mathbf{0}}\\ {\mathbf{0}} & {\mathbf{B}}_2 & \cdots & {\mathbf{0}}\\ \vdots & \vdots & \ddots & \vdots\\ {\mathbf{0}} & {\mathbf{0}} & \cdots & {\mathbf{B}}_N \end{array}\right],$$ with each of the sub-blocks ${\mathbf{B}}_j$, $1\leq j\leq N$, is of the form ${\mathbf{B}}_j =\lambda_j{\mathbb{I}}+ {\mathbf{C}}_j$ where $${\mathbf{C}}_j= \left[\begin{array}{ccccc} 0 & 1 & 0 & \cdots & 0\\ 0 & 0 & 1 & \cdots & 0\\ \vdots & \vdots & \ddots & \ddots & \vdots\\ 0 & 0 & 0 & \cdots & 1 \\ 0 & 0 & 0 & \cdots & 0 \end{array}\right]$$ and $\lambda_1, \ldots \lambda_N$ are eigenvalues of ${\mathbf{M}}$. Hence, ${\mathbf{M}}^k={\mathbf{V}}\boldsymbol{\Lambda}^k{\mathbf{V}}^{-1}$ and $$\boldsymbol{\Lambda}^k=\left[\begin{array}{cccc} {\mathbf{B}}_1^k & {\mathbf{0}} & \cdots & {\mathbf{0}}\\ {\mathbf{0}} & {\mathbf{B}}_2^k & \cdots & {\mathbf{0}}\\ \vdots & \vdots & \ddots & \vdots\\ {\mathbf{0}} & {\mathbf{0}} & \cdots & {\mathbf{B}}_N^k \end{array}\right].$$ For each eigenvalue $\lambda_j$ and each block ${\mathbf{B}}_j$, we can write $${\mathbf{B}}_j^k = (\lambda_j{\mathbb{I}}+{\mathbf{C}}_j)^k = \sum_{p=0}^k \binom{k}{p} \lambda_j^{k-p} {\mathbf{C}}_j^p.$$ It can easily be shown that for each $j$, ${\mathbf{C}}_j$ is a nilpotent matrix, that is, if ${\mathbf{C}}_j$ is an $m\times m$ matrix, then ${\mathbf{C}}^m={\mathbf{0}}$. Therefore, for $k\geq m-1$, $${\mathbf{B}}_j^k = \lambda_j^{k-m+1} \left( \sum_{p=0}^{m-1} \binom{k}{p} \lambda_j^{m-p-1} {\mathbf{C}}_j^p \right).$$ Observe that the above expression in parenthesis gives an (upper triangular) matrix whose entries are $m-1$-th degree polynomials in $k$ and hence, that the whole expression for ${\mathbf{B}}_j^k$ is dominated by $\lambda_j^{k-m+1}$. Since, by the spectral radius assumption, ${\left\vert\lambda_j\right\vert}<1$ for each $i$, it follows that for each $j$, ${\mathbf{B}}_j^k\to{\mathbf{0}}$ as $k\to\infty$ and hence $\boldsymbol{\Lambda}^k\to{\mathbf{0}}$ as $k\to\infty$ by the block structure. This proves the first statement. For the second statement suppose that ${\mathbb{I}}-{\mathbf{M}}$ is singular. Then ${\mathbb{I}}-{\mathbf{M}}$ has $0$ as an eigenvalue and hence $\lambda=1$ is an eigenvalue of ${\mathbf{M}}$, contradicting our assumption about the spectral radius of ${\mathbf{M}}$. Therefore, ${\mathbb{I}}-{\mathbf{M}}$ is invertible. Furthermore, it can easily be obtained using the block diagonal structure of $\boldsymbol{\Lambda}$ and the ratio test that the sum $\sum_{k=0}^\infty {\mathbf{M}}^k$ converges, Hence, $$({\mathbb{I}}-{\mathbf{M}})\sum_{k=0}^\infty {\mathbf{M}}^k = \sum_{k=0}^\infty {\mathbf{M}}^k - \sum_{k=0}^\infty {\mathbf{M}}^{k+1} = {\mathbb{I}}+ \sum_{k=1}^\infty {\mathbf{M}}^k - \sum_{k=1}^\infty {\mathbf{M}}^k = {\mathbb{I}}.$$ Since the matrix ${\mathbf{P}}$ is stochastic, we have ${\left\Vert{\mathbf{P}}\right\Vert}_\infty=1$ and hence the spectral radius of ${\mathbf{P}}$ is bounded by $1$. Since ${\mathbf{P}}_{TT}$ is a submatrix of ${\mathbf{P}}$, we have ${\left\Vert{\mathbf{P}}_{TT}\right\Vert}_\infty\leq 1$ and its spectral radius is also bounded by $1$. To prove Proposition \[prop:specrad0\] (denoted Proposition \[prop:specrad\] below) we will show that the spectral radius of ${\mathbf{P}}_{TT}$ is strictly smaller than $1$ if there is some vertex in $S$ that can be reached from any transient node via a directed path. Before presenting the main proof, we require several lemmas. \[lemma:sprad1\] Let ${\mathbf{B}}$ and ${\mathbf{C}}$ be $n\times n$ matrices with non-negative entries such that ${\left\Vert{\mathbf{B}}\right\Vert}_\infty\leq 1$ and ${\left\Vert{\mathbf{C}}\right\Vert}_\infty\leq 1$ and let ${\mathbf{D}}={\mathbf{C}}{\mathbf{B}}$. Suppose there exists $1\leq p\leq n$ such that $0 < \sum_j B_{pj}< 1$. Then, for every $1\leq i\leq n$ such that $C_{ip} > 0$, $$\sum_j D_{ij} < 1.$$ Let $K=\{k: C_{ik}>0 \}$. Then $p\in K$ and $$\begin{aligned} \sum_j D_{ij} & = \sum_j \sum_k C_{ik}B_{kj}\\ & = \sum_{k\in K} C_{ik} \sum_j B_{kj}\\ & \leq \sum_{k\in K\setminus\{p\}} C_{ik} {\left\Vert{\mathbf{B}}\right\Vert}_\infty + C_{ip} \sum_j B_{pj}\\ & < \sum_{k\in K\setminus\{p\}} C_{ik}+ C_{ip}\\ & \leq 1.\end{aligned}$$ \[prop:specrad3\] Let $\Gamma$ be a weighted directed graph with weight matrix ${\mathbf{W}}$. Let $i$ and $j$ be distinct nodes of $\Gamma$ connected by a directed path from $i$ to $j$ of length $n\geq 1$. Then $W^n_{ij}>0$. We use induction. If $i$ and $j$ are connected with a path of length $1$, then there exists an edge $(i,j)\in E$ and hence $W_{ij}>0$. Assume that $W^m_{ij}>0$ if $i$ and $j$ are connected by a directed path from $i$ to $j$ of length $m$. Suppose $i$ and $j$ are connected by a path of length $m+1$. Then there exists a vertex $k$ such that $i$ and $k$ are connected by a directed path from $i$ to $k$ of length $m$ and there exists a directed edge $(k,j)$. Hence, by our assumption $W^m_{ik}>0$ and $W_{kj}>0$. Therefore, $$W^{m+1}_{ij} = \sum_{k'\in V} W^{m}_{ik'}W_{k'j}\geq W^m_{ik}W_{kj} > 0.$$ \[lemma:sprad4\] Let ${\mathbf{M}}={\mathbf{P}}_{TT}$, let $i\in T$ and suppose there exists $s\in S$ such that there exists a directed path from $i$ to $s$ of length $m$. Then for all $n\geq m$, $$\sum_{k\in T} M^{n}_{ik} < 1.$$ Let $i\in T$ and let $s\in S$ be a vertex such that there exists a directed path from $i$ to $s$ of length $m$. Let $J$ be the set of vertices in $T$ directly adjacent to a vertex in $S$. Then, by our assumption, for every $i\in T$ there exists $j\in J$ such that there exists a directed path from $i$ to $j$ of length $m-1$. Since the matrix ${\mathbf{P}}_{TT}$ can be treated as the weight matrix for the subgraph of $\Gamma$ restricted to vertices in $T$, it follows by Lemma \[prop:specrad3\] that $M^{m-1}_{ij}>0$. Since every point in $J$ is adjacent to a point in $S$, it also follows that $\sum_{k\in T} M_{jk}< 1$. Clearly, ${\left\Vert{\mathbf{M}}\right\Vert}_\infty\leq 1$ and hence ${\left\Vert{\mathbf{M}}^{m-1}\right\Vert}_\infty\leq 1$. Applying Lemma \[lemma:sprad1\] to the matrices ${\mathbf{M}}$ and ${\mathbf{M}}^{m-1}$ we obtain that for every $i\in T$, $\sum_{k\in T} M^{m}_{ik} < 1$. Let $t\geq m$ and assume $\sum_{k\in T} M^{t}_{ik} < 1$. We have $$\sum_{k\in T} M^{t+1}_{ik} = \sum_{k\in T} \sum_{k'\in T} M^t_{ik'}M_{k'k} = \sum_{k'\in T} M^t_{ik'} \sum_{k\in T} M_{k'k} \leq \sum_{k'\in T} M^t_{ik'} < 1$$ and our result follows by induction. \[prop:specrad\] Suppose that for every $p\in T$ there exists $s\in S$ such that there exists a directed path from $p$ to $s$. Then, the matrix ${\mathbb{I}}-{\mathbf{P}}_{TT}$ is invertible and $$\label{eq:invassum} ({\mathbb{I}}-{\mathbf{P}}_{TT})^{-1} = \sum_{k=0}^\infty ({\mathbf{P}}_{TT})^k.$$ Let ${\mathbf{M}}={\mathbf{P}}_{TT}$. Observe that our assumption implies that for every $i\in T$ there exists $s\in S$ such that there exists a directed path from $i$ to $s$ of length at most $N$. By Lemma \[lemma:sprad4\], we have for every $i\in T$, $\sum_{k\in T} M^{N}_{ik} < 1$. Hence, ${\left\Vert{\mathbf{M}}^{N}\right\Vert}_\infty< 1$ and therefore the spectral radius of ${\mathbf{M}}={\mathbf{P}}_{TT}$ is strictly smaller than $1$. Our result follows by Lemma \[lemma:mat\]. Information dissipation {#app:dissipation} ----------------------- \[prop:dissipation1\] Let ${\boldsymbol{\alpha}}$ and ${\boldsymbol{\beta}}$ be vectors of length $N$ such that for all $i\in V$, $\alpha_i>0$ and $\beta_i>0$. Define the $N \times N$ matrix ${\mathbf{\tilde{P}}}$ with entries $$\tilde{P}_{ij} = \alpha_i\beta_j P_{ij},$$ Let $\alpha_*=\max\{\alpha_i:i\in V\}$ and $\beta_*=\max\{\beta_i:i\in V\}$ and suppose $\alpha_*\beta_*<1$. Then, the matrix ${\mathbb{I}}-{\mathbf{\tilde{P}}}_{TT}$ is invertible and $$\label{eq:invassum2} ({\mathbb{I}}-{\mathbf{\tilde{P}}}_{TT})^{-1} = \sum_{k=0}^\infty ({\mathbf{\tilde{P}}}_{TT})^k.$$ Let ${\mathbf{M}}={\mathbf{\tilde{P}}}_{TT}$ and let $i\in T$. Then, $$\begin{aligned} \sum_{j\in T} M_{ij} & = \sum_{j\in T} \alpha_i\beta_jP_{ij} \leq \alpha_*\beta_* \sum_{j\in T} P_{ij}< 1.\end{aligned}$$ Hence, ${\left\Vert{\mathbf{M}}\right\Vert}_\infty<1$ and thus the spectral radius of ${\mathbf{\tilde{P}}}_{TT}$ is strictly smaller than $1$. Our result then follows by Lemma \[lemma:mat\]. More generally, it is possible to interpret dissipation in the light of Proposition \[prop:specrad\] by constructing a new graph $\tilde{\Gamma}$ with the vertex set $\tilde{V}=V\cup\{v\}$, where $v$ denotes an additional vertex. The weight matrix of $\tilde{\Gamma}$, denoted ${\mathbf{\tilde{W}}}$, has entries $$\tilde{W}_{ij} = \begin{cases} \alpha_i\beta_j P_{ij} & \text{if $i\in V$ and $j\in V$,}\\ 1 - \sum_{k\in V} \alpha_i\beta_k P_{ik} & \text{if $i\in V$ and $j=v$,}\\ 0 & \text{if $i=v$.} \end{cases}$$ Clearly, a random walk on $\tilde{\Gamma}$ is equivalent to a random walk on $\Gamma$ with dissipation: the dissipated information is directed towards the additional vertex $v$ and then disappears. If we place $v$ in the boundary set $\tilde{S}$, by Proposition \[prop:specrad\], the necessary condition for existence of the Green’s function $({\mathbb{I}}-{\mathbf{\tilde{P}}}_{TT})^{-1}$ is that from every transient node $i$ there exists a directed path to either a node $s\in S$ or a node $j\in T$ such that $\sum_{k\in V} \alpha_j\beta_k P_{jk}< 1$ (such node $j$ is adjacent to $v$ in the graph $\tilde{\Gamma}$. Proposition \[prop:dissipation1\] then just represents the special case where every transient vertex is adjacent to $v$ in $\tilde{\Gamma}$. Interpretations of the matrices ${\mathbf{F}}$ and ${\mathbf{H}}$ {#app:interpr} ================================================================== ${\mathbf{F}}$ and ${\mathbf{H}}$ as matrices of expected visiting times {#app:interpr1} ------------------------------------------------------------------------ We will show that both $F_{ij}$ and $H_{ij}$ can be interpreted as the expected number of times a random walk originating at the vertex $i$ visits the vertex $j$, while avoiding all vertices in the boundary set $S$. Note that in the case of the matrix ${\mathbf{F}}$, we have $i\in T$ and $j\in S$ while for the matrix ${\mathbf{H}}$, $i\in S$ and $j\in T$. We will use ${\mathbb{E}}$ to denote the expectation operator. Suppose the boundary set $S$ represents sinks and let $Z_{ij}$ be a random variable denoting the total number of times a random walk starting at $i\in T$ is absorbed at $j\in S$. Then, $${\mathbb{E}}(Z_{ij}) = F_{ij}.$$ Let $Y_{ij}(t)$ be the random variable taking the value $1$ if the random walk originating at $i\in T$ is absorbed at $j\in S$ at time $t$, with probability $\sum_{k\in T} P_{ik}^{t-1} P_{kj}$, and taking the value $0$ otherwise. We have $Z_{ij}=\sum_{t=1}^\infty Y_{ij}(t)$ and ${\mathbb{E}}(Y_{ij}(t)) = \sum_{k\in T} P_{ik}^{t-1} P_{kj}$. Thus, $$\begin{aligned} {\mathbb{E}}(Z_{ij}) & = {\mathbb{E}}\left(\sum_{t=1}^\infty Y_{ij}(t)\right)\\ & = \sum_{t=1}^\infty {\mathbb{E}}(Y_{ij}(t))\\ & = \sum_{t=1}^\infty \sum_{k\in T} P_{ik}^{t-1} P_{kj}\\ & = \sum_{k\in T} \sum_{t=0}^\infty P_{ik}^{t}P_{kj}\\ & = \sum_{k\in T} G_{ik} P_{kj}\\ & = F_{ij}. \qedhere\end{aligned}$$ Suppose the boundary set $S$ represents sources and let $Z_{ij}$ be a random variable denoting the total number of times a random walk starting at $i\in S$ visits the node $j\in T$. Then, $${\mathbb{E}}(Z_{ij}) = H_{ij}.$$ In the same fashion as above, let $Y_{ij}(t)$ be the random variable taking the value $1$ if the random walk originating at $i\in S$ is at $j\in T$ at time $t$, with probability $\sum_{k\in T} P_{ik} P^{t-1}_{kj}$, and taking the value $0$ otherwise. We have $Z_{ij}=\sum_{t=1}^\infty Y_{ij}(t)$ and ${\mathbb{E}}(Y_{ij}(t)) = \sum_{k\in T} P_{ik} P_{kj}^{t-1}$. Thus, $$\begin{aligned} {\mathbb{E}}(Z_{ij}) & = {\mathbb{E}}\left(\sum_{t=1}^\infty Y_{ij}(t)\right)\\ & = \sum_{t=1}^\infty {\mathbb{E}}(Y_{ij}(t))\\ & = \sum_{t=1}^\infty \sum_{k\in T} P_{ik} P^{t-1}_{kj}\\ & = \sum_{k\in T} \sum_{t=0}^\infty P_{ik}P^{t}_{kj}\\ & = \sum_{k\in T} P_{ik}G_{kj}\\ & = H_{ij}. \qedhere\end{aligned}$$ Invariants of ${\mathbf{F}}$ and ${\mathbf{H}}$ {#app:interpr2} ----------------------------------------------- Let ${\mathbf{1}}\in{\mathbb{R}}^n$ denote the vector whose entries are all $1$’s. Since all rows of ${\mathbf{P}}$ sum to unity, it follows that ${\mathbf{P}}{\mathbf{1}} = {\mathbf{1}}$ and hence ${\mathbf{1}}$ is a right eigenvector of ${\mathbf{P}}$ for the eigenvalue $\lambda=1$. Define ${\mathbf{d}}$ as a vector of length $n$ having entries $d_i = \sum_j W_{ij}$. If $\Gamma$ is unweighted graph, $d_i$ gives the degree of the node $i$. Assuming ${\mathbf{W}}$ is symmetric, $$\sum_k P_{kj}d_k = \sum_k W_{kj} = \sum_k W_{jk} = d_j$$ and therefore ${\mathbf{d}}$ is a left eigenvector of ${\mathbf{P}}$ corresponding to the eigenvalue $\lambda=1$. This leads to the following result. \[lemma:Fstoch\] Suppose that the matrix ${\mathbb{I}}-{\mathbf{P}}_{TT}$ is invertible. Let ${\mathbf{u}}$ and ${\mathbf{v}}$ be the left and right eigenvector of the matrix ${\mathbf{P}}$ corresponding to the eigenvalue $\lambda=1$, respectively. Write ${\mathbf{u}} = [{\mathbf{u}}_S\ {\mathbf{u}}_T ]$ and ${\mathbf{v}}=\left[\begin{array}{c}{\mathbf{v}}_S\\ {\mathbf{v}}_T \end{array}\right]$. Then, $$\label{eqn:leigen0} {\mathbf{u}}_T = {\mathbf{u}}_S{\mathbf{H}},$$ and $$\label{eqn:reigen0} {\mathbf{v}}_T = {\mathbf{F}}{\mathbf{v}}_S.$$ Using the canonical form of the matrix ${\mathbf{P}}$ (Equation (\[eqn:Pcannonical\])) and the fact that ${\mathbf{u}}$ and ${\mathbf{v}}$ are left and right eigenvectors of ${\mathbf{P}}$ respectively, we obtain $$\label{eqn:leigen1} {\mathbf{u}}_T = {\mathbf{u}}_S{\mathbf{P}}_{ST} + {\mathbf{u}}_T{\mathbf{P}}_{TT},$$ and $$\label{eqn:reigen1} {\mathbf{v}}_T = {\mathbf{P}}_{TS}{\mathbf{v}}_S + {\mathbf{P}}_{TT}{\mathbf{v}}_T.$$ Rearranging Equations (\[eqn:leigen1\]) and (\[eqn:reigen1\]) leads to $$\label{eqn:leigen2} {\mathbf{u}}_T({\mathbb{I}}-{\mathbf{P}}_{TT}) = {\mathbf{u}}_S{\mathbf{P}}_{ST},$$ and $$\label{eqn:reigen2} ({\mathbb{I}}-{\mathbf{P}}_{TT}){\mathbf{v}}_T = {\mathbf{P}}_{TS}{\mathbf{v}}_S.$$ Our result then follows as the consequence of invertibility of ${\mathbb{I}}- {\mathbf{P}}_{TT}$. Since ${\mathbf{1}}$ is a right eigenvector of ${\mathbf{P}}$, it follows from (\[eqn:reigen0\]) that for all $i$, $\sum_{j\in S} F_{ij} = 1$. Furthemore, recall that if $\Gamma$ is an undirected graph, ${\mathbf{W}}$ is symmetric and ${\mathbf{d}}$ is a left eigenvector of ${\mathbf{P}}$ for $\lambda=1$. Assuming the matrix ${\mathbf{H}}$ exists, we obtain from Lemma \[lemma:Fstoch\] that, if $S$ contains a single point, the matrix ${\mathbf{H}}$ is a row vector, which is a multiple of ${\mathbf{d}}_T$. [^1]: to whom correspondence should be addressed

ArXiv

--- abstract: 'We analyze in detail the penumbral structure found in a recent radiative MHD simulation. Near $\tau=1$, the simulation produces penumbral fine structure consistent with the observationally inferred interlocking comb structure. Fast outflows exceeding $8\,\mbox{km\,s}^{-1}$ are present along almost horizontal stretches of the magnetic field; in the outer half of the penumbra, we see opposite polarity flux indicating flux returning beneath the surface. The bulk of the penumbral brightness is maintained by small-scale motions turning over on scales shorter than the length of a typical penumbral filament. The resulting vertical rms velocity at $\tau=1$ is about half of that found in the quiet Sun. Radial outflows in the sunspot penumbra have two components. In the uppermost few $100$ km, fast outflows are driven primarily through the horizontal component of the Lorentz force, which is confined to narrow boundary layers beneath $\tau=1$, while the contribution from horizontal pressure gradients is reduced in comparison to granulation as a consequence of anisotropy. The resulting Evershed flow reaches its peak velocity near $\tau=1$ and falls off rapidly with height. Outflows present in deeper layers result primarily from a preferred ring-like alignment of convection cells surrounding the sunspot. These flows reach amplitudes of about $50\%$ of the convective rms velocity rather independent of depth. A preference for the outflow results from a combination of Lorentz force and pressure driving. While the Evershed flow dominates by velocity amplitude, most of the mass flux is present in deeper layers and likely related to a large-scale moat flow.' author: - 'M. Rempel' bibliography: - 'natbib/apj-jour.bib' - 'natbib/papref\_m.bib' title: 'Penumbral fine structure and driving mechanisms of large-scale flows in simulated sunspots' --- Introduction ============ Since the discovery of the Evershed effect about a century ago [@Evershed:1909], the origin of large-scale outflows in sunspot penumbrae has been a central element in observational and theoretical studies of sunspots. Over the past decades, advancements in ground- and space-based observing capabilities have revealed the stunning fine structure of sunspot penumbrae that is manifest in the intensity, magnetic field and velocity structure (see, for example, recent reviews by @Solanki:2003 [@Thomas:Weiss:2004; @Thomas:Weiss:2008] and high resolution observations by @Scharmer:etal:2002 [@Langhans:etal:2005; @Rimmele:Marino:2006; @Ichimoto:etal:2007; @Ichimoto:etal:2007:sc; @Langhans:etal:2007; @Scharmer:etal:2007; @Rimmele:2008; @Franz:Schlichenmaier:2009; @BellotRubio:etal:2010]). All quantities show in the penumbra a primarily radial filamentary structure. Strong horizontal outflows take place in regions with almost horizontal field, embedded in a background of more vertical field – which has been referred to as “uncombed penumbra” [@Solanki:Montavon:1993] or “interlocking-comb” structure [@Thomas:Weiss:1992]. The connection between the Evershed flow and the intensity structure is less clear. While earlier work pointed toward a flow preferentially in the dark component, more recently @Schlichenmaier:etal:2005 and @Ichimoto:etal:2007 showed that Evershed flow and intensity variations show a positive correlation in the inner and negative correlation in the outer penumbra. Another controversial aspect is the depth profile of the Evershed flow. While @Rimmele:1995 and @Stanchfield:etal:1997 found the Evershed flow in elevated flow channels, more recent work by @Schlichenmaier:etal:2004 [@BellotRubio:etal:2006] and @Borrero:etal:2008 points toward a flow primarily in the deep photosphere that declines with height. A variety of simplified models have been proposed to explain the penumbral fine structure [e.g. @Danielson:1961; @Meyer:Schmidt:1968; @Galloway:1975; @Thomas:1988; @Degenhardt:1989; @Degenhardt:1991; @Grosser:1991; @Wentzel:1992; @Thomas:Montesinos:1993; @Montesinos:Thomas:1997; @Schlichenmaier:etal:1998a; @Schlichenmaier:etal:1998b; @Spruit:Scharmer:2006; @Scharmer:Spruit:2006]. Studies of idealized magnetoconvection in inclined magnetic field [see, e.g., @Hurlburt:etal:1996; @Hurlburt:etal:2000] revealed traveling wave-like convection modes, which produce at the surface a combination of horizontal flow velocities and pattern motions that have been associated with flow properties observed in penumbrae. Recently substantial progress was made in “realistic” numerical simulations that include of the effects of partial ionization and radiative transfer. These models were first applied to sections of sunspots [@Schuessler:Voegler:2006; @Heinemann:etal:2007; @Rempel:etal:2009; @Kitiashvili:etal:2009] and later to full sunspots [@Rempel:etal:Science; @Rempel:2010:IAU]. Not all of the simplified models listed above contain a self-consistent description of the Evershed effect, however, such a flow could be added as an additional degree of freedom to most of them. The models that include physical processes responsible for driving large-scale outflows are based on either stationary or dynamic flux tube models. In the case of stationary flux tube models [@Meyer:Schmidt:1968; @Thomas:1988; @Degenhardt:1989; @Degenhardt:1991; @Thomas:Montesinos:1993; @Montesinos:Thomas:1997], a pressure difference is imposed at the footpoints of the flux tube, which leads to siphon flows. In the dynamic flux tube model of @Schlichenmaier:etal:1998a [@Schlichenmaier:etal:1998b], fast outflows result from a combination of hot plasma rising at the inner footpoint and additional pressure driving that results from radiative cooling in the photosphere. In the more recent radiative MHD simulations, the penumbral fine structure is a byproduct of anisotropic overturning convection and the Evershed flow has been interpreted as the convective flow component in the direction of the magnetic field [@Scharmer:etal:2008]. A more detailed analysis of [@Rempel:etal:Science] concluded that convection in penumbra and quiet Sun differ primarily in terms of anisotropy of the velocity field, while typical convective rms velocities and length scales of energy and mass transport are comparable. The convective nature of the penumbra is manifest in all radiative 3D simulations to date but it is still debated in the context of observational constraints. While there is evidence for overturning convection in some investigations [@Zakharov:etal:2008; @Rimmele:2008; @Bharti:etal:2010], many others primarily identify the upflow component of the Evershed flow in the inner and downflows in the outer penumbra with little evidence for overturning convection [@BellotRubio:etal:2005; @Ichimoto:etal:2007; @Franz:Schlichenmaier:2009; @BellotRubio:etal:2010]. In this paper, we present a detailed analysis of the model presented in [@Rempel:etal:Science] with special focus on the physical origin of large-scale outflows. After a brief description of the numerical model in Sect. \[sect:model\], we focus first on the photospheric appearance of the penumbra in Sect \[sect:phot\]. In Sect. \[sect:forcings\] - \[sect:simple-model\] we present a detailed analysis of the subsurface structure responsible for the driving of large-scale outflows. Sect. \[sect:filaments\] analyzes the field geometry and connectivity in penumbral flow channels and compares our results with findings from previous models, in particular models based on the flux tube picture. Sect. \[sect:deep-flow\] analyzes deeper reaching outflow components beneath the penumbra that are not directly associated with the Evershed flow. The results are summarized and discussed in Sect. \[sect:concl\]. Numerical model {#sect:model} =============== Our investigation is based on the simulation of a pair of opposite polarity sunspots described in @Rempel:etal:Science. The simulation in a $98\times 49$ Mm wide and $6.1$ Mm deep box was designed to study the formation and structure of penumbrae under a variety of different field strength and inclination angles. To this end the simulation was initialized with a pair of opposite polarity sunspots each having a flux of $1.6\cdot 10^{22}$ Mx each, but different field strengths of about $3$ and $4$ kG, respectively. In order to focus on details of sunspot fine structure a rather high grid resolution of $3072\times 1536 \times 384$ ($32$ km horizontal and $16$ km vertical) was used at the expense that this simulation could cover only a rather short time span. While the original presentation in @Rempel:etal:Science was based on a run of $1.5$ h in high resolution ($3.5$ h total) we have progressed the simulation in the meantime to $4$ h in high resolution ($6$ h total). During the extension of the simulation several aspects of the penumbral structure evolved. During the later stages of the simulation filaments became more radially aligned and the mean intensity profile shows in the inner penumbra a constant value of about $0.7 I_{\odot}$ with a more steep drop toward the umbra. The overall properties of the Evershed flow did not show a significant variation in the time frame covered by this simulation; we base our detailed analysis of the physical origin of the flow pattern on the last hour of this simulation run (starting about $5$ hours after the initialization). We also emphasize for clarification that this simulation uses gray radiative transfer. When we refer in the following discussion to intensity, we mean the bolometric intensity, if we refer to optical depth, we mean the optical depth computed with the Rosseland mean opacity. Photospheric appearance {#sect:phot} ======================= Azimuthal averages in photosphere --------------------------------- As described in detail in @Rempel:etal:Science, the simulation domain contains a pair of opposite polarity sunspots, with the most extended penumbrae found in between the opposite polarity spots along the horizontal x-direction. The most coherent penumbra is found in the sunspot with the initially stronger field strength of about $4$ kG [see the spot on the right in Fig. 1 of @Rempel:etal:Science]. We focus our detailed analysis on the latter, for which an intensity image is presented in Fig. \[fig:int\_region\]. We are here in particular interested in the extended penumbra on the left side of the spot for which we highlighted the sub-domain used for azimuthal averages in subsequent figures. The dashed lines indicate $R=10$ Mm and $R=20$ Mm from the center of the spot. Fig. \[fig:global\_prop\] summarizes properties at the $\tau=1$ level, averaged azimuthally over the $90$ degree wedge shown in Fig. \[fig:int\_region\] and about $1$ hour in time. The intensity normalized by the quiet Sun brightness $I_{\odot}$ (panel a) shows a sharp increase from umbra toward penumbra from about 0.15 to 0.7 $I_{\odot}$. In the inner penumbra, the intensity stays constant on a plateau with about $0.7 I_{\odot}$ and increases then almost linear toward the edge of the penumbra where it reaches $0.95 I_{\odot}$ (due to the nearby opposite polarity spot in our simulation setup the intensity does not reach $I_{\odot}$). The plateau-like intensity profile formed during later stages of this simulation and was not present in the results reported earlier by @Rempel:etal:Science. The radial outflow (panel b) starts at about $R=10$ Mm, reaches its peak of about $4\,\mbox{km\,s}^{-1}$ near $R=15$ Mm and drops off toward the outer edge of the penumbra. $R=10$ Mm corresponds to the position at which the average field inclination (displayed in panel d) angle exceeds $45$ degrees, which was already found by @Rempel:etal:Science as the critical value for the onset of large-scale outflows. The position of the peak velocity coincides with the position of maximum inclination (about 70 degrees) in the middle of the penumbra. The inclination is defined here as $\arcsin(B_R/\vert B\vert)$. Due to the strong variation of inclination angle with azimuth, it makes a difference whether we compute the inclination locally and average in azimuth and time later or whether we base the computation on the averaged magnetic field presented in panel c). We show in panel d) both, the average of the inclination (solid) and the inclination of the average field (dashed). The vertical rms velocity (panel b, dashed) increases steadily throughout the penumbra from a few $100\,\mbox{m\,s}^{-1}$ at $R=10$ Mm to about $2\,\mbox{km\,s}^{-1}$ at the outer edge, which is the value corresponding to quiet Sun granulation. A value of about $1\,\mbox{km\,s}^{-1}$ is required near the inner edge of the penumbra ($R=12$ Mm) to maintain the penumbral brightness of $0.7 I_{\odot}$. We find in this simulation an approximate relationship of the form $I \propto \sqrt{v_{z\,rms}(\tau=1)}$. Filamentation in photosphere ---------------------------- Fig. \[fig:fil\_tau1\] displays the filamentary fine structure seen at the $\tau=1$ level in the penumbra. A filamentary structure is present in all quantities shown, however the strongest evidence is seen in intensity (panel a), vertical magnetic field (panel c), inclination (panel d) and radial flow velocity (panel e). Penumbral filaments show a strong reduction of the vertical magnetic field strength, while the horizontal (radial) field component is moderately enhanced (panel b). The combination of the two leads to the strong variation of the inclination angle in the penumbra. Strong radial outflows (panel e, red color indicates outflows) are seen in the almost horizontal flow channels, toward the outer end of flow channels the inclination angle exceeds $90\deg$, indicating field returning back into the convection zone. Radial outflows with more than $10\,\mbox{km\,s}^{-1}$ outflow velocity are indicated by solid contours. Most of the very fast outflows are found in the inner half of the penumbra, a few of them are associated with fast downflows in the outer penumbra. The vertical velocity (panel f, blue colors indicate upflows) shows strong up and downflows everywhere in the penumbra, strong radially aligned upflows are preferentially found in the center of penumbral filaments. Solid contours indicate regions with more than $5\,\mbox{km\,s}^{-1}$ downflow velocity. They are primarily found near the outer edge of the penumbra. We find downflow speeds of up to $15\,\mbox{km\,s}^{-1}$ near $\tau=1$. Fast downflows in opposite polarity regions have been observed by @Westendorp:etal:2001 [@DelToro:etal:2001]. To clarify the relation between radial flow velocity, intensity and magnetic field strength in a statistical sense we present correlation coefficients in Fig. \[fig:corr\]. Panel a) displays the correlation between intensity and radial velocity, panel c) the correlation between field strength and radial velocity. All correlations are computed based on the fluctuations of these quantities about their azimuthal mean. Intensity is correlated with outflows in the inner penumbra, but weakly anti-correlated further outward. The radial outflow is found in regions with reduced field strength in the inner, but stronger field in the outer penumbra. Similar correlations were found by @Ichimoto:etal:2007 (see Fig. 3 therein) as well as @Schlichenmaier:etal:2005. In the panels on the right (b and d) we present additional correlations, which allow us to make a closer connection to the magnetoconvective structure of the penumbra. The radial dependence of the $I-v_R$ correlation is due to a decorrelation between vertical and radial velocity in the penumbra (panel b, blue) combined with a decorrelation of vertical velocity and intensity (panel b, red). While the latter remains positive, a sign change is present in the former. The physical reason for the decorrelation between vertical and radial velocity is evident from the magnetoconvection pattern shown in Fig. \[fig:fil\_tau1\]. In the inner penumbra filaments are very narrow and the central upflow covers most of the filament, leading to large positive correlation between the brighter upflow and radial outflow. In the outer penumbra the patches of outflowing material become broader and several downflow lanes can be found within these patches, resulting in a reduction of the correlation. Toward the outer edge of the penumbra stronger downflow patches are present, turning the correlation weakly negative. Note that the $I-v_z$ correlation stays low outside $R=20$ Mm due to the proximity of an opposite polarity spot in our simulation setup. Fig. \[fig:corr\] panel d) presents additional correlations between radial velocity and vertical magnetic field (blue) as well as radial magnetic field strength (red). While the former stays negative throughout the penumbra, the correlation with the radial field strength is positive. In the innermost penumbra the strong reduction of $\vert B_z\vert$ dominates the picture and leads to an anti-correlation between $v_R$ and $\vert B\vert$, further out the contribution from the increased $\vert B_R\vert$ in the flow channels dominates and leads to a positive $v_R - \vert B\vert$ correlation. The increase of the inclination angle found in the flow channels is a consequence of a strong reduction of $B_z$ to almost zero, while $B_R$ is moderately enhanced in strength. This asymmetry is due to the fact that $B_R$ benefits from a strong positive contribution of the induction term $(\vec{B}\cdot\nabla)\vec{v}$ due to the Evershed flow, while the corresponding term is negative for $B_z$ due to the upward decreasing vertical velocity near $\tau=1$ (see Sect. \[sect:simple-model\] for more detail). @Ichimoto:etal:2007 found also a negative $v_R - \vert B\vert$ correlation in the inner penumbra with a trend of overall decreasing anti-correlation further out, however, a sign change was not observed. The latter was proposed by @Tritschler:etal:2007 and @Ichimoto:etal:2008 based on observations of the net circular polarization (NCP) in the outer penumbra at different viewing angles. The outflow velocity we find in the simulation is not stationary, we see flow variability that ranges from periodic fluctuations on timescales of $5-10$ minutes in the inner penumbra to quasi-periodic variations over a wider range of timescales starting from $10-20$ minutes in the center and outer penumbra. A flow variability in the $15-40$ minute range was also reported in the simulation of @Kitiashvili:etal:2009 and associated with Evershed clouds [@Shine:etal:1994; @Rimmele:1994; @Cabrera:etal:2007]. It is conceivable that the periodic variations we find near the inner tip of filaments have a relation to twisting motions observed by @Ichimoto:etal:2007:sc and @Bharti:etal:2010. We focus in this paper on the maintenance of the stationary flow component and base our analysis primarily on time and volume averages over sections of the penumbra. The non stationary flow component will be analyzed in a separate publication. Mass and energy fluxes ---------------------- Fig. \[fig:fill\_vy\] panel a) displays filling factors of radial and vertical motions. While the upflow filling factor remains almost constant around $0.4$ to $0.5$ from inner umbra toward the outer penumbra, the filling factor of outflows exceeds $0.8$ in the center of the penumbra. Panel b) shows the vertical rms velocity (black), together with the mean velocity of upflow (blue) and downflow regions (red). The green line presents the mean vertical velocity averaged over regions with radial outflows (flow channels). The latter shows a weak average upflow of about $250\,\mbox{m\,s}^{-1}$ in the inner penumbra and a downflow reaching velocities of more than $500\,\mbox{m\,s}^{-1}$ toward the outer edge of the penumbra. The contributions from small- and large-scale flow components to mass and energy flux in the penumbra are presented in Fig. \[fig:m\_e\_flux\]. In order to properly compare up- and downflow components we perform the analysis here on a constant height surface that is located about $350$ km beneath $\tau=1$ in the quiet Sun (about half a Wilson depression downward). We decompose here the mass flux into positive and negative as well as azimuthal average components. Their contributions as function of radius are presented in panel a), where we show $\langle m_z^+ \rangle$ in blue and $\langle m_z^- \rangle$ in red as well as $\langle m_z \rangle$ in green (the latter is the sum of the former two). Here, $\langle\ldots\rangle$ denotes the azimuthal average and $m_z^{\pm}=(m_z\pm\vert m_z\vert)/2$. While in the innermost penumbra up to about $50\%$ of the mass flux is present in the azimuthal average component, this fraction drops steadily toward the center penumbra. Integrated over the region $R<15$ Mm about $1/3$ of the total upward flowing mass is found in the azimuthal component, while the major fraction ($2/3$) is still overturning laterally. The mass flux in the penumbra is balanced within $R<20.9$ Mm. Integrated over this region the unsigned mass flux in the mean component constitutes about $13\%$ of the total unsigned vertical mass flux in the penumbra. Evaluating the relative contributions from large-scale and small-scale convective motions to the total convective energy flux in the penumbra requires an appropriate decomposition of the vertical mass flux. A separation just into azimuthal mean and the respective fluctuation would not be sufficient since the latter assumes that the large-scale flow is axisymmetric and equally considers filaments with higher temperature and the region in between with lower temperature in the enthalpy flux. The consequence would be an underestimation of the overall contribution from the large-scale flow (in the region $R<20.9$ the net contribution would be $-0.07 L_{\odot}$). Instead we construct the vertical mass flux of the laterally overturning flow component $m_z^S$ as follows: in regions with positive $\langle m_z \rangle$, we reduce the amplitude of upflows such that they are in a mass flux balance with downflows, in regions with negative $\langle m_z \rangle$, we reduce the amplitude of downflows such that they are in balance with the upward directed mass flux. The large-scale mass flux is then given by $m_z^L=m_z-m_z^S$. Unlike the decomposition into azimuthal mean and corresponding fluctuation, this procedure does not change the position of upflow and downflow regions for $m_z^S$ and $m_z^L$ compared to $m_z$. With $H=(e_{\rm int}+p)/\varrho+v^2/2$ we can now compute the energy flux components $F_z^{S/L}=\langle m_z^{S/L} H \rangle$, which are displayed in Fig. \[fig:m\_e\_flux\] panel b). While $F_z^S$ matches the intensity profile very well in the outer penumbra, there is a clear deficit present in the inner penumbra. The deviations in the umbra are due to the fact that our horizontal slice is located above the $\tau=1$ in the umbra and therefore the convective energy flux is zero. The large scale energy flux $F_z^L$ has an amplitude of about $+L_{\odot}$ in the inner penumbra and $-L_{\odot}$ in the outer penumbra. The relevant quantity here is the net contribution after carefully balancing the upflow in the inner and downflow in the outer penumbra. Integrating $F_z^L$ over the region $R<20.9$ Mm (in which the large-scale mass flux is balanced) leads to a net contribution of $12\%$ to the total convective energy flux. The latter is very consistent with the relative mass flux contribution of $13\%$ we found before. In an integral sense the large-scale flow contributes only a small fraction of the energy radiated away in a sunspot penumbra, but locally the contribution can be larger. If we use the difference between $F_z^S/L_{\odot}$ (blue) and $I/I_{\odot}$ (green) in Fig. \[fig:m\_e\_flux\] as a rough estimate for the missing energy flux we identify a contribution of up to $50\%$ in the inner penumbra. Note that we avoided in the above discussion the association between Evershed flow and the large-scale flow component since there is no clear definition of what the former encompasses. If we associate the Evershed flow only with the horizontal flow pattern that corresponds to upflows in the inner and downflows in the outer penumbra we would conclude that this flow pattern plays only a minor role in the penumbral energy transport. This definition would essentially correspond to the “sources” and “sinks” of the Evershed flow that have been identified by @Rimmele:Marino:2006 and @Ichimoto:etal:2007. Also note that the contribution from large-scale flows increases with depth as all intrinsic scales of convection increase with depth. Properly quantifying their contribution in deeper layers requires numerical simulations over longer timescales (since convective timescales increase with depth), which is beyond the scope of the current investigation. Subsurface flow structure and underlying driving forces {#sect:forcings} ======================================================= Flow structure beneath penumbra ------------------------------- Fig. \[fig:vr\_normalized\] presents the subsurface outflow structure as function of depth and radial distance from the center of the spot. The depth is measured relative to the average height of the $\tau=1$ level in the quiet sun. We will use the same height scale in all of the following figures except Fig. \[fig:filament\_avr\], where we use the average $\tau=1$ level in the penumbra as reference. Flow velocities are normalized by the rms velocity found outside the sunspot at the corresponding height level (see also Fig. \[fig:vrms\]). We have chosen this normalization in order to compare flow fields found in the penumbra to convective flows found in almost undisturbed convection. We refer to this velocity reference in the following as $v_{\rm rms}^0$. While the outflow velocity stays around $0.4-0.5\,v_{\rm rms}^0$ in the deeper layers, the near surface layers stand out with flow speeds exceeding $v_{\rm rms}^0$. The two different scaling regimes of the outflow velocity found in the near surface layers (uppermost $500$ km) and the deeper part of the domain indicate already different physical driving mechanism at work, which we will analyze further in the following discussion. We exclude here the lower most $2$ Mm of our domain which are partially influenced by the bottom boundary condition. The solid black lines indicate the average $\tau=1$ and $\tau=0.01$ levels. The radial outflow velocity peaks close to $\tau=1$ and falls off rapidly with height. An outflow is present to about $\tau=0.01$ in the inner and $\tau=10^{-3}$ in the outer penumbra. The azimuthally averaged mass flux changes sign between $\tau=10^{-3}$ and $\tau=10^{-4}$, since it puts more weight on the region above the more dense filament channels with fast outflows. Overall the simulation indicates that radial outflows in the penumbra are expected to be found in the deep photosphere, which is consistent with recent spectropolarimetric inversions [@Schlichenmaier:etal:2004; @BellotRubio:etal:2006; @Borrero:etal:2008], but not earlier work by @Rimmele:1995 and @Stanchfield:etal:1997 where elevated flow channels were inferred. Whether the inflow above $\tau=10^{-3}$ could be related to the inverse Evershed flow observed in the Chromosphere [@Dialetis:etal:1985] is currently an open question, even though also @Borrero:etal:2008 found observational evidence for an inflow near temperature minimum. As described in @Rempel:etal:2009 we switch for reasons of numerical stability to an isothermal equation of state in regions with $\beta=p_{\rm gas}/p_{\rm mag}< 10^{-3}$ and limit the Lorentz force such that the Alfv[é]{}n velocity does not exceed $60\,\mbox{km\,s}^{-1}$ to prevent stringent time step constraints. The latter two could possibly influence this flow pattern near the top boundary while the influence on flows in the photosphere is rather weak. Despite substantial velocities of a few $\mbox{km\,s}^{-1}$ the associated mass and momentum flux is negligible compared to photospheric flows due to the sharp drop in density. The vertical dotted lines indicate 3 regions we refer to in the following analysis. We consider the region in between $R=10$ and $R=12$ Mm as inner penumbra, the region in between $R=12$ and $R=18$ Mm as center penumbra and $R=18$ and $R=20$ as outer penumbra. We have chosen the boundary for the inner penumbra based on the intensity profile (Fig. \[fig:global\_prop\]) that reaches a value typical for a penumbra of $0.7 I_{\odot}$ at $R=12$ Mm. Since our outer penumbra might not be fully representative for conditions in a “typical” outer penumbra due to the presence of a nearby opposite polarity spot with an Evershed flow in the opposite direction, we separated out regions with $R>18$ Mm. $R=18$ Mm is also the distance where most of the dominant filaments of the center penumbra end (see Fig. \[fig:int\_region\] and \[fig:fil\_tau1\]). Fig. \[fig:vrms\] compares rms velocities in the plage region surrounding the sunspot and the center penumbra. The top panels present the absolute rms velocities, the bottom panels relative to $v_{\rm rms}^0$. Blue indicates the vertical rms velocity, green and red the horizontal components (green is along the filaments in the case of the penumbra). In the plage region (more or less undisturbed convection) about half of the kinetic energy is found in vertical motions, the other half equally distributed among the horizontal components. This scaling holds very well over the 3 orders of magnitude in pressure stratification shown here (panel c). In the penumbra the vertical and horizontal rms velocity perpendicular to the filaments show a similar scaling and relative strength, but overall their amplitude is reduced to about $40\%$ of the respective values in the plage region. The rms velocity along the filaments is strongly increased with respect to the vertical rms, indicating a strong degree of anisotropy. The rms velocity in the direction of filaments is proportional to $v_{\rm rms}^0$ in more than $1$ Mm depth and shows a steep increase toward the photosphere. Overall the kinetic energy is reduced in the deeper layers, but doubled in the near surface layers compared to the plage region. The apparent excess of kinetic energy found in the Evershed flow compared to the plage region is due to a vertical redistribution of kinetic energy combined with anisotropy of the flow. Underlying driving forces ------------------------- In order to investigate the physical processes that lead to the driving of large-scale outflows around sunspots we analyze the energy conversion terms in the kinetic energy equation. Starting from the momentum equation we derive the following energy balance (we drop the time derivatives since we are interested in time averages): $$\begin{split} \underbrace{\vec{v}\cdot(\varrho\vec{g}-\nabla p)}_{\rm pressure/buoyancy} &+\underbrace{\vec{v}\cdot(\vec{j}\times\vec{B})}_{\rm Lorentz}\\ &\underbrace{-\varrho\vec{v}\cdot[(\vec{v}\cdot\nabla)\vec{v}]}_{\rm Acceleration} +\underbrace{\vec{v}\cdot F_{visc}}_{\rm Viscosity}=0\;. \end{split}$$ Under the assumption of stationarity the acceleration term is identical to the negative divergence of the kinetic energy flux, $\varrho\vec{v} v^2/2$. A negative acceleration term implies positive divergence, i.e. the volume element is a source of kinetic energy. In Fig. \[fig:energetics\_cmp\] we compare the different contributions to the energy equation for the plage region (panel a) and penumbra (panel b). On a qualitative level there is a large degree of similarity: Pressure/buoyancy forces are the main driver, close to the surface most of that energy input is used up by acceleration forces, the remainder is balanced in about equal parts by work against viscous and Lorentz forces. In the penumbra the total amount of energy input by pressure/buoyancy forces in the uppermost $1.5$ Mm shown here is reduced to about $40\%$ and more concentrated toward the photosphere. The reduction in energy input is consistent with the overall reduced kinetic energy integrated over this depth range. It is also notable that the work against the Lorentz force is not substantially different from the plage region (relative to the respective pressure driving) despite the quite different field strength and field structure. To investigate the driving of flows in plage and penumbra further we split now terms into the contributions from different grid directions as well as flow directions, i.e. we consider the following $18$ terms: $$\begin{aligned} P_i^{\pm}&=&\langle\,v_i^{\pm}[\varrho g_i -(\nabla p)_i]\,\rangle \label{eq:driving_components1}\\ L_i^{\pm}&=&\langle\,v_i^{\pm}(\vec{j}\times\vec{B})_i\,\rangle \label{eq:driving_components2}\\ A_i^{\pm}&=&-\langle\,\varrho v_i^{\pm}[(\vec{v}\cdot\nabla) \vec{v}]_i \,\rangle\label{eq:driving_components3}\end{aligned}$$ Here, $i$ indicates either the Cartesian directions $x,y,z$ in the case of the plage region or the cylindrical components $R,\Phi,z$ in the case of the sunspot penumbra. With $v_i^{\pm}=(v_i\pm\vert v_i\vert)/2$ we denote negative and positive velocity components. Note that we compute all forces on the Cartesian grid and use the transformation to cylindrical coordinates only to separate the directions along and perpendicular to filaments in our nearly axisymmetric penumbra fragment (i.e. we compute terms like $v_r F_r$ and $v_{\Phi} F_{\Phi}$ instead of $v_x F_x$ and $v_y F_y$ with $\vec{F}$ being any one of the forces). The explicit expression for the viscous force is rather complicated due to the non-linearity of the underlying artificial viscosity scheme. In the following discussion we do not explicitly compute the viscous terms, but indicate their approximate magnitude by using the quantity $V_i^{\pm}=-(P_i^{\pm}+L_i^{\pm}+A_i^{\pm})$. We confirmed a close relationship for a few snapshots, for which we restarted the code and extracted all numerical dissipation terms. Formally the energy conversion terms are power densities (work per volume and time). For the sake of making the text more readable we will refer to them in the following discussion very often as “work done by/against ... forces” instead of “work done by/against ... forces per volume and time”. Since the former is simply the latter multiplied by a unit volume element and time interval it has no further impact on the physical meaning of these terms. Fig. \[fig:energetics\_plage\] shows the energy conversion terms for vertical motions (panel a) and horizontal motions (panel b) in the plage region. Note that we only show one horizontal direction due to isotropy. In the vertical direction pressure/buoyancy driving is in balance with work done against acceleration forces. Most of the pressure/buoyancy driving takes place in downflows due to their overdense material that cannot be supported by the pressure gradient. Pressure/buoyancy driving in upflows is much weaker since they are very close to a hydrostatic balance. Close to $z=0$ the sign of pressure/buoyancy driving is changing in upflows as a consequence of the overshoot layer in the upper photosphere. Magnetic and viscous forces play only a minor role in the vertical direction. Horizontal flows are primarily driven by pressure forces. Most of the energy is absorbed by magnetic and viscous forces, only a small amount is balanced by horizontal acceleration in the uppermost $500$ km of the convection zone. In comparison to the plage region, the center penumbra shows distinct differences (Fig. \[fig:energetics\_center\_p\]). Almost all pressure/buoyancy driving takes place in upflow regions: the presence of strong magnetic field causes in the near surface layers to a steepening of the pressure gradient, leading to almost hydrostatic balance in downflows and excess pressure driving in the upflows. This excess pressure driving in upflows is in balance with work against the Lorentz force while in contrast to the plage region vertical acceleration of fluid does not play an important role at all. Most of the energy extracted by Lorentz forces in the vertical direction is deposited into outward acceleration of fluid along filaments. In that sense the Evershed flow is driven by vertical pressure forces in upflows that are deflected into the horizontal direction through the Lorentz-force. This “deflection” is most efficient very close to the surface: integrated from $1$ Mm ($500$, $250$ km) depth to the top boundary about $44\%$ ($64\%$, $97\%$) of the pressure driving in the vertical direction ends up as Lorentz force driving in the radial direction. This way kinetic energy that is normally deposited into the vertical direction (plage region) gets transferred directly into the horizontal direction and accounts for the high anisotropy seen in the penumbra. Also the role of horizontal pressure driving differs substantially from the plage region, which can be seen in Fig. \[fig:energetics\_center\_p2\]a), where we show the energy conversion terms for the radial direction on a different scale. Pressure driving is dominant below $300$ km depth, but the overall magnitude remains smaller than in the plage region at comparable depth. While pressure driving shows a preference for outflows in more than $300$ km depth, it prefers inflows further up. Pressure forces are the primary cause for the deep flow component with velocities of about $0.5\,v_{\rm rms}^0$ we identified in Fig. \[fig:vr\_normalized\], but their overall role for the near surface flow is limited: integrated from $1$ Mm ($500$, $250$ km) depth to the top boundary the contribution from $P^+_R$ relative to $L^+_R$ is $100\%$ ($38\%$, $9\%$). If we consider only the components of the driving that break the symmetry between in- and outflows, $P^+_R-P^-_R$ and $L^+_R-L^-_R$, the corresponding values are $13\%$ ($1\%$, $-9\%$). Note that most of the contribution to $P^+_R-P^-_R$ comes from the region $R>17$ Mm, further inward $P^+_R-P^-_R$ is close to zero (see also Fig. \[fig:energetics\_contour\]). However, pressure forces remain the dominant driver for flows perpendicular ($\Phi$-direction) to filaments (Fig. \[fig:energetics\_center\_p2\]b). Here, pressure driving is offset by work against the Lorentz force, while both acceleration and viscous terms do not contribute substantially. The different role of pressure driving compared to the plage region is primarily a consequence of anisotropy in terms of radially elongated convection cells in the penumbra: radial pressure gradients are reduced, while lateral pressure gradients are enhanced compared to isotropic granulation. In addition the steepening of the vertical pressure gradient (that leads to the shift of pressure driving from down to upflows) results in an overall reduction of pressure close to the photosphere in comparison to the ambient stratification. The clear association between Lorentz force driving and the near surface flow pattern is evident from Fig. \[fig:energetics\_contour\]. Here, we present the quantities $A^+_R-A^-_R$, $L^+_R-L^-_R$ and $P^+_R-P^-_R$ as function of radius and depth. Strong negative values of $A^+_R-A^-_R$ indicate outward acceleration of fluid. These regions are confined to a narrow layer near $\tau=1$. Here, the Lorentz force is the primary driver, pressure terms have weakly negative contributions (they favor inflows). The peak of Lorentz force driving and acceleration is found in between $R=12$ and $R=15$ Mm. In deeper layers pressure terms are in approximate balance with Lorentz force terms, resulting in only minor acceleration work despite their overall amplitude. Toward the outermost edge the Lorentz force is overcompensating outward directed pressure driving resulting in deceleration of radial outflows (positive values of $A^+_R-A^-_R$). For comparison with Fig. \[fig:energetics\_center\_p\] we present in Figs. \[fig:energetics\_inner\_p\] and \[fig:energetics\_outer\_p\] the same quantities for the inner penumbra (from $R=10$ to $R=12$ Mm) and outer penumbra (from $R=18$ to $R=20$ Mm). In the inner penumbra we see a forcing pattern that is very similar to that we found for the center penumbra, in particular with respect to the near surface layers where the Evershed flow is driven. Differences are present in deeper layers; here, the Lorentz force is also dominant in driving outflows and is actually driving these outflows against horizontal pressure forces. In the outer penumbra, we see a fundamentally different situation (which is in part a consequence of the nearby opposite polarity spot and the resulting strongly magnetized downflow lane in between). Here, outward directed pressure forces dominate the picture entirely, however, they do not lead to a strong outward acceleration of fluid. To a large degree they are opposed by the horizontal Lorentz force and the energy is transferred to the vertical direction, where the Lorentz force becomes the major driver for downflows. The latter is due to the fact that magnetic field in the outer penumbra turns back downward. We see in the uppermost layers only a weak signal from the horizontal Lorentz force driving outflows – this is expected since we are in the region where the Evershed flow declines quickly to zero. We will further discuss the deeper reaching flow component in Sect. \[sect:deep-flow\]. Magnetic filament substructure responsible for driving the Evershed flow {#sect:simple-model} ======================================================================== Simplified momentum balance --------------------------- After describing the driving forces behind the Evershed flow in detail in the previous section, we present here a simple model of the underlying thermal, magnetic and velocity structure and reduce the overall picture to the most relevant terms in the equations. In order to carve out the typical structure of the regions responsible for driving the Evershed flow we select regions which have both upflows and outflows. In Fig. \[fig:filament\_avr\]a) we present the mean magnetic field, flow, and thermal structure as function of depth obtained by averaging over all such regions horizontally between $R=12$ and $R=18$ Mm. While the vertical magnetic field (black, dashed) is constant at about $400$ G, the radial component (black, solid) increases monotonically from about $900$ G to $1.8$ kG at the $\tau=1$ level (vertical dotted line) and drops again in higher layers to about $1.6$ kG. The steep increase just below $\tau=1$ is essential for the Lorentz force component driving the outflow as we will describe below. The vertical velocity (blue, dashed) increases monotonically to about $1\,\mbox{km\,s}^{-1}$ just $100$ km beneath $\tau=1$, followed by a sharp decline to a few $100\,\mbox{m\,s}^{-1}$ above $\tau=1$. The radial flow velocity peaks right at $\tau=1$, the maximum amplitude is about $5.5\,\mbox{km\,s}^{-1}$. The solid red line shows the mean temperature profile with the corresponding scale on the right. Panel b) presents volume averages of pressure driving in upflows (black) work against Lorentz force in upflows (red) and work by Lorentz force in outflows (blue) and work against acceleration forces (green) for the same region. The dotted black line shows contributions from horizontal pressure gradients multiplied by a factor of $10$. Averaged over the region shown (from -$0.4$ to $0.2$ Mm) they contribute about $10\%$ to the total acceleration work. Work by the horizontal Lorentz force (blue) is $90\%$ of the work by vertical pressure driving (black). The solid lines are based on all terms in the equations (see Eqs. (\[eq:driving\_components1\]) to (\[eq:driving\_components3\]) ), the dashed lines are an approximation for Lorentz force and acceleration terms based only on volume averages of the following expressions: $$\begin{aligned} L_z&=&-v_z\frac{1}{4\pi}B_R\frac{\partial B_R}{\partial z}\label{eq:wlz} \\ L_R&=&v_R\frac{1}{4\pi}B_z\frac{\partial B_R}{\partial z}\label{eq:wlr} \\ A_R&=&-\varrho v_R v_z \frac{\partial v_R}{\partial z}\label{wq:war} \;.\end{aligned}$$ The excellent agreement allows us to understand the driving mechanism behind the Evershed effect by considering a reduced set of equations. Note that we could go even one step further with these simplifications, by expressing all terms through the mean quantities $\bar{B}_R(z)$, $\bar{B}_z(z)$, $\bar{v}_R(z)$, $\bar{v}_z(z)$, $\bar{p}(z)$, and $\bar{\varrho}(z)$. Despite the fact that we are dealing with nonlinear terms of spatially highly inhomogeneous quantities, the agreement remains excellent except for the acceleration term that falls short by a factor of 2, i.e. quantities such as $\overline{\varrho v_R v_z \partial_z v_R}$ and $\bar{\varrho} \bar{v}_R \bar{v}_z \partial_z \bar{v}_R$ agree in general on a qualitative level for the region we selected to perform the averages. In the vertical direction we have essentially a magneto-hydrostatic balance involving the terms: $$\overline{\frac{\partial}{\partial z}\left(p+\frac{B_R^2}{8\pi}\right)} \approx-\overline{\varrho g}\;.$$ This is evident from the opposing contributions of the terms $-\overline{v_z\left(\partial_z p+\varrho g\right)}$ and $-\overline{v_z B_R\partial_z B_R/(4\pi)}$ in Fig. \[fig:filament\_avr\]b) (black and dashed red curve). The energy extracted by the Lorentz force in the vertical direction leads to a strong acceleration of an outflow in the radial direction. Here, we have a balance between the Lorentz force and acceleration terms: $$\overline{\frac{1}{4\pi}B_z\frac{\partial B_R}{\partial z}}\approx \overline{\varrho v_z\frac{\partial v_R}{\partial z}}\;.$$ The acceleration force results from the upward transport of plasma in a region with an upward increasing Evershed flow velocity. The work by vertical and radial Lorentz force components is in approximate balance, i.e. $$\overline{v_z\frac{1}{4\pi}B_R\frac{\partial B_R}{\partial z}}\approx \overline{v_R\frac{1}{4\pi} B_z\frac{\partial B_R}{\partial z}}\;,$$ leading to a simple relation between vertical and radial flow velocities of the form $$\overline{v_z B_R}\approx \overline{v_R B_z}\label{eq:vrvz-ratio}\;.$$ The latter is the relation one would expect from a simple “deflection” of vertical flows by an inclined magnetic field. Note that most of the acceleration of the outflow takes place about $100$ km beneath the $\tau=1$ level, while the outflow peaks right at $\tau=1$. The latter is a consequence of the strong vertical upflow advecting accelerated fluid a few $100$ km further upward. This upward advected fluid also overpowers the inward directed Lorentz force found right above $\tau=1$ due to the sign change in $\partial_z B_R$. Induction equation ------------------ Since the large positive value of $\partial_z B_R$ right below $\tau=1$ plays an essential role in the acceleration process, we analyze now how this magnetic field structure is maintained in the presence of strong vertical and radial flows. To this end we evaluate the different contributions in the induction equation: $$\frac{\partial \vec{B}}{\partial t}= \underbrace{-(\vec{v}\cdot\nabla)\vec{B}}_{\rm Advection} +\underbrace{(\vec{B}\cdot\nabla)\vec{v}}_{\rm Stretching} \underbrace{-\vec{B}(\nabla\cdot\vec{v})}_{\rm Divergence}\;.$$ The bottom panels of Fig. \[fig:filament\_avr\] present these contributions (black: advection, red: stretching, blue: divergence) for the maintenance of the radial magnetic field structure in panel c) and vertical magnetic field structure in panel d). As before solid lines show the full expressions, dashed-line approximations are described in the text below. For the radial field component (panel c) the dominant source is the stretching term in the induction equation. The major contribution to this term comes from the vertical shear profile of the Evershed flow, leading to an induction term $\overline{B_z\partial_z v_R}$ (red dashed line). The remainder is due to horizontal stretching from terms like $\overline{B_R\partial_R v_R}$. The peak of the stretching term (including all contributions) is located about $100$ km beneath $\tau=1$, where we also find the peak of the Lorentz force driving. This is not exactly where we find the peak of $\bar{B}_R$, since there is an additional strong contribution from the vertical advection that pushes strong radial field upward ($-\overline{v_z\partial_z B_R}$, black dashed line). The remainder is offset by the negative contribution from the diverging convective motions (solid blue line). In the case of the vertical field (panel d), the role of the contributions from advection and stretching are opposite. Here ,advection is the primary mechanism that maintains the field. The positive sign originates primarily from horizontal advection terms (black dashed) with a dominant contribution from $-\overline{v_R\partial_R B_z}$ due to the on average outward decreasing vertical field strength, but there are also positive contributions from vertical advection $-\overline{v_z\partial_z B_z}$. The dominant negative contribution to the stretching term is due to $\overline{B_z\partial_z v_z}$ (red dashed line), which peaks close to $\tau=1$ where $\partial_z \bar{v}_z$ is strongly negative. The remainder is offset again by the negative contribution from the diverging convective motions (solid blue line). In both panels the green curve indicates the negative sum of these three terms, i.e. the amplitude of additional contributions from artificial numerical diffusivity. For both radial and vertical magnetic field the contributions from stretching, advection and divergence are in balance at the level of a few $\%$. This indicates that the magnetic structure within the penumbral filaments in this simulations is not strongly affected on average by the unavoidable artificial magnetic diffusivity of the numerical scheme. Also the fact that our simulation contains almost field free umbral dots on scales even smaller than filaments sets strong constraints on the role artificial diffusivity plays. Filament cross section ---------------------- From Fig. \[fig:filament\_avr\] we deduce a vertical extent of about $200$ km for the region in which most of the energy conversion takes place. This value is obtained through an average over all areas between $R=12$ and $R=18$ km that have upflows and outflows. Since the typical height variation of the $\tau=1$ level in the penumbra is about $200$ km, this indicates that locally within individual filaments the flow is driven in an even narrower boundary layer right beneath the $\tau=1$ level. We illustrate this in Fig. \[fig:fil\_cross\_sect\], which shows magnetic field and velocity together with inclination and energy conversion by horizontal Lorentz force on a vertical cut through three developed and one just forming penumbral filament in the inner penumbra. It is evident that there is a narrow boundary layer forming along the $\tau=1$ surface that is characterized by increased $B_R$ and reduced $B_z$, resulting in a strong increase of inclination. Lorentz force driving is concentrated to an equally thin layer just beneath $\tau=1$, the resulting outflow is broader and extends above $\tau=1$. The latter is a consequence of the presence of overturning motions that transport and distribute accelerated fluid above and along the $\tau=1$ surface. This redistribution does not require additional acceleration work, since the associated kinetic energy flux is close to divergence free (the term for acceleration work is identical to the negative divergence of the kinetic energy flux under the assumption of stationarity). In Fig. \[fig:fil\_cross\_sect\], we highlighted a cross section in the inner penumbra, further out the filamentary structure is less prominent. Nevertheless, we also see there a concentration of the energy conversion terms to very thin sheets beneath $\tau=1$, while fast outflows are found mostly between $\tau=1$ and $\tau=0.01$. In a statistical (average) sense the differences between inner and center penumbra are small (compare Figs. \[fig:energetics\_center\_p\] and \[fig:energetics\_inner\_p\]). Also note that the concentration of energy conversion by Lorentz forces to thin sheets is typical for magnetoconvection in a more general sense; however, the preferred location and quasi-steady maintenance of these regions near $\tau=1$ is restricted to the penumbra. The mechanism responsible for the latter is explained in Sect. \[sect:boundary\_layer\]. Formation of thin boundary layer {#sect:boundary_layer} -------------------------------- In this section we illustrate the crucial role of the vertical advection terms by discussing a simplified model that captures the essential terms on a qualitative level to within a factor of two. We consider only those terms in the radial momentum and induction equations that have been identified as the dominant contributors in the previous discussion: $$\begin{aligned} \frac{\partial v_R}{\partial t}+v_z\frac{\partial v_R}{\partial z}&=& \frac{B_z}{4\pi\varrho}\frac{\partial B_R}{\partial z} \label{eq:eq-m}\\ \frac{\partial B_R}{\partial t}+v_z\frac{\partial B_R}{\partial z}&=& B_z\frac{\partial v_R}{\partial z}\label{eq:eq-i}\;.\end{aligned}$$ Neglecting the advection terms, Eqs. (\[eq:eq-m\]) and (\[eq:eq-i\]) lead to wave solutions of the form (assuming that $B_z$ is nearly constant, which is at least true for the average shown in Fig. \[fig:filament\_avr\]): $$\left[\frac{\partial^2}{\partial t^2}-\frac{B_z^2}{4\pi\varrho} \frac{\partial^2}{\partial z^2}\right](v_R,B_R)=0\;.$$ In this case the profiles of $v_R$ and $B_R$ could not be maintained in place and would spread out with the Alfv[é]{}n velocity corresponding to the vertical magnetic field component, $\vert B_z\vert /\sqrt{4\pi\varrho}$; see, for example, @Vasil:Brummell:2009 for a discussion of the dynamics of magnetic shear layers. Since the Alfv[é]{}n velocity is of the order of $2\,\mbox{km\,s}^{-1}$ (using a mean value of $B_z=400$ G and $\varrho=3\cdot 10^{-7} {\rm g}/{\rm cm}^3$ near the $\tau=1$ level), the rather narrow Evershed flow profile would broaden substantially within a few $100$ secs of time. On the other hand, the inclusion of the advection terms allows for a stationary solution provided that $v_z>\vert B_z\vert/\sqrt{4\pi\varrho}$. Since near $\tau=1$ the vertical field strength drops and upflows can reach locally up to $3\,\mbox{km\,s}^{-1}$, this condition can be met within the thin boundary layer in which most of the driving is taking place. With a sufficiently strong advection term the upflow counteracts the downward traveling Alfv[é]{}n wave, while the upward traveling wave is bound by the photosphere and transition to a low $\beta$ regime. The quasi-steady maintenance of the shear layer despite the back-reaction of Lorentz forces seems to be at odds with Lenz’s rule, however, we have to keep in mind that there is a steady flow of energy through the system ultimately driven by overturning convective motions. Indeed, the energy conversion by the vertical advection term in the induction equation, $B_R/(4\pi)\, v_z\partial_z B_R$, is identical to the energy extracted by the Lorentz force from convective motions in the vertical direction (Eq. \[eq:wlz\]). To summarize, the most important feature responsible for driving the Evershed flow is a strong increase of $B_R$ just beneath the $\tau=1$ level (combined with the presence of a vertical background field of a few $100$ G). The steep gradient of $B_R$ is primarily maintained by the vertical shear profile of the Evershed flow in combination with upward advection due to the strong upflow in the center of the filament. The vertical advection terms in the induction as well as momentum equation are essential for a quasi-stationary configuration with lifetimes far beyond the overturning timescale of convection. They ensure that the peak of $v_R$ and $B_R$ is maintained above the region with the strongest Lorentz force driving as well as induction due to shear, which are proportional to $\partial_z B_R$ and $\partial_z v_R$, respectively. The vertical confinement of the shear layer is guaranteed by the fact that locally the upflow velocity can exceed $\vert B_z\vert/\sqrt{4\pi\varrho}$. Since the outflow velocity is linked to the upflow velocity by the approximate relation $v_R B_z\approx v_z B_R$ the expectation is that the resulting outflow reaches a velocity of about $v_R=\vert B_R\vert/\sqrt{4\pi\varrho}$. With $B_R=1\ldots 2$ kG and $\varrho=3\cdot 10^{-7}{\rm g}{\rm cm}^{-3}$ the resulting velocities should be $v_R=5\ldots 10\,\mbox{km\,s}^{-1}$, which is about the range we find for the velocity within flow channels. Fig. \[fig:v\_alf\] displays outflow and upflow velocities at $\tau=1$ relative to $\vert B_R\vert/\sqrt{4\pi\varrho}$ and $\vert B_z\vert/\sqrt{4\pi\varrho}$. Outflows are close to Alfv[é]{}nic throughout the penumbra (typically $v_R\approx 0.8\,\vert B_R\vert/\sqrt{4\pi\varrho}$), upflows are weakly super-Alfv[é]{}nic. The fact that there is a clear threshold for the onset of this driving mechanism ($v_z>\vert B_z\vert/\sqrt{4\pi\varrho}$ near $\tau=1$) is a possible explanation for a more or less well defined inner edge of the penumbra, or related, a critical field inclination that needs to be exceeded (about $45$ degrees in this simulation). Going further inward toward the umbra the vertical field becomes stronger while vertical velocities are reduced, which makes it harder to pass this threshold locally. Even if it would be passed the resulting radial flow velocities would be less due to the smaller value of $B_R$. Observable consequences ----------------------- Unfortunately the “feature” responsible for driving the Evershed flow remains well hidden beneath the $\tau=1$ level. Even worse, if we compute the Lorentz force from the “visible” part of the magnetic field structure, the Lorentz force is inward directed due to the sign change of $\partial_z B_R$. The amplitude of the visible inward component above $\tau=1$ is about a factor of $5-10$ smaller than the strong outward directed component beneath $\tau=1$, which is responsible for the outward acceleration. The observable parts of the magnetic and velocity field are an increase of $v_R$ and $B_R$ toward $\tau=1$ and a strong vertical gradient of $v_z$ around $\tau=1$. The latter is only visible very deep in the photosphere. The combination of this three factors should lead to positive values of ${\rm d}|B|/{\rm d}\tau$ and ${\rm d}|v_{\rm los}|/{\rm d}\tau$ for a variety of observation angles and therefore contribute to the net circular polarization observed in sunspot penumbrae. Field line structure of filaments {#sect:filaments} ================================= Many simplified models of penumbral filaments and the origin of the Evershed flow such as @Meyer:Schmidt:1968 [@Thomas:1988; @Degenhardt:1989; @Degenhardt:1991; @Thomas:Montesinos:1993; @Montesinos:Thomas:1997; @Schlichenmaier:etal:1998a; @Schlichenmaier:etal:1998b] are based on the thin flux tube approximation. It is not clear from first principles whether penumbral filaments (flow channels) can be identified with flux tubes in a meaningful way. The latter assumes the existence of a well defined flux surface that encloses a flow channel and clearly separates fluid “inside” the channel from fluid “outside” and assumes further that there are well defined “footpoints” when intersected with a horizontal plane somewhere beneath the photosphere. The convective structure of the penumbra as presented in Sect. \[sect:phot\] puts already some limits on the usefulness of the flux tube concept, since overturning convective motions are mostly orthogonal to the flows assumed in the flux tube picture (except footpoints). These convective motions lead to a continuous mass, momentum and energy exchange while fluid is moving outward along penumbral flow channels. Integrated over the penumbra this mass exchange is substantial, since we find only about $13\%$ of the total unsigned vertical mass flux in the large scale flow component. In the following paragraphs we will discuss filaments on the basis of their field line structure and connectivity to allow for a better comparison with models that are based on the flux tube approximation. Field Line connectivity ----------------------- In Fig. \[fig:field\_lines\] we present the magnetic field line connectivity as well as radial outflow velocity. The filaments are representative for different radial positions in the simulated penumbra. The field line analysis presented here was performed using the VAPOR software package developed at NCAR [@Clyne:Rast:2005; @Clyne:etal:2007] (www.vapor.ucar.edu). The field lines are computed from a $15$ minute average, which is about a characteristic wave crossing time along filaments. We have chosen the latter since in particular stationary flux tube models make only sense on timescales beyond that, but the following conclusions are not affected by the averaging in a fundamental way. Filament 1 corresponds to a peripheral umbral dot that almost transitions to a penumbral filament, filament 2 is representative for the inner, 3 and 4 for the center and 5 and 6 for the outer penumbra. Seed points for filament 1 were chosen from a magnetogram at the umbral $\tau=1$ level (regions with reduced field strength indicating peripheral umbral dot). Seed points for filaments 2-5 were chosen based on outflow velocities in the indicated vertical cross sections with more than $5\,\mbox{km\,s}^{-1}$ (most of them are actually around $8\,\mbox{km\,s}^{-1}$). The seed points for filament 6 were chosen from the indicated horizontal plane based on regions with more than $5\,\mbox{km\,s}^{-1}$ downflow speed. The color coding of the field lines indicates the radial outflow velocity (the colors red, yellow, green, and blue correspond to velocities of $<0$, $2$, $4$, and $>8\,\mbox{km\,s}^{-1}$, respectively). Going radially outward from filament 1 to filament 6 we see the following physical picture emerging. Near the umbra-penumbra boundary upflow plumes (similar to those forming umbral dots in the center of the umbra) push mass upward along inclined field lines. The mass loading results in a small bend of the field line and the upflow is guided outward leading to outflow velocities of about a $\mbox{km\,s}^{-1}$. Due to the strong almost vertical field the flow is not powerful enough to bend over field lines completely. This is consistent with a moderate enhancement of the field inclination in peripheral umbral dots by some $10-20\deg$ that has been inferred from spectropolarimetric observations [@Socas-Navarro:etal:2004; @Riethmueller:etal:2008; @Sobotka:Jurcak:2009; @Ortiz:etal:2010]. Going further outward (filament 2) field lines are bent over sufficiently to become horizontal for a distance of a few Mm. In the horizontal stretch we find outflows exceeding $8\,\mbox{km\,s}^{-1}$ (blue color), nevertheless, the field lines remain connected to the top boundary. Near the outer edge of this filament, we see the formation of small dips right before the field lines connect back to the top boundary. The latter is a consequence of the mass flux decoupling on average from the field through the formation of U-loops and reconnection with deeper field lines that extend further out. In addition, downflows present along the edge of filaments can transport field lines together with the mass flowing along them downward beneath the photosphere. Moving to the center penumbra (filaments 3 and 4) the horizontal stretch with fast outflows is extended and field lines start to bend over and return beneath the photosphere (filament 4). Here, most of the mass unloading from field lines happens through either the above mentioned U-loop formation or through field lines that bend over temporarily. Going further outward (filament 5) the latter scenario happens most of the time resulting in a filament that follows the $\tau=1$ level and returns back beneath the photosphere. Note that all of these filaments (2-5) have fast $>8$ km outflows along their almost horizontal stretch in the photosphere regardless of their connectivity further out. Filament 6 shows an example of field connectivity and flows in proximity of one of the fast downflow patches in the outer penumbra (all of the selected field lines have more than $5\,\mbox{km\,s}^{-1}$ downflow velocity near $\tau=1$). Compared to the previous filament most field lines host only outflows in the $2-4$ km range with some faster flows present in the ultimate proximity of the footpoint. Most field lines reach toward the higher photosphere and turn back beneath the photosphere within $3$ Mm. They do not show the extended horizontal stretches that host fast outflows in filaments 2-5. Underlying physical picture --------------------------- Overall we do not see compelling evidence that the field line connectivity is linked to the presence of strong horizontal outflows. Filaments with more than $8\,\mbox{km\,s}^{-1}$ outflow speed can have any connectivity: in the inner penumbra field lines typically connect to the top boundary, in the outer penumbra field lines bend over and return beneath the photosphere. Looking at the smooth transition in the filament structure throughout the penumbra (filaments 2-5) strongly suggests that a similar process is responsible for driving outflows in all of them. Following up on the discussion in Sect. \[sect:forcings\] and \[sect:simple-model\] outflows result from pressure driven upflows that load field lines with mass and bend them over. A very similar situation was already described by @Scharmer:etal:2008 based on the “slab” simulation of @Heinemann:etal:2007. The energetic signature of this process is a balance between pressure and Lorentz forces in upflows and Lorentz and acceleration forces in the radial direction. Horizontal pressure gradients do not enter the picture on average since the upflow cells are elongated in the radial direction, which strongly decreases the role of the radial pressure gradient. The elongation of filaments does not impact the Lorentz force since there only vertical field gradients matter (see Sect. \[sect:simple-model\]). The region in which outflows are driven is confined to a narrow boundary layer just beneath $\tau=1$ as we discussed in Sect. \[sect:simple-model\]. An interesting property of this driving mechanism is that no substantial acceleration work is present in horizontal stretches of the magnetic field, where we find most of the fast horizontal flows: the Lorentz force has no horizontal component there and horizontal pressure gradients do not contribute much. This is however no contradiction, since all of the fluid that appears at or above $\tau=1$ has to pass through the narrow boundary layer with concentrated driving forces (see Fig. \[fig:fil\_cross\_sect\]). The continuous vertical mass exchange along the flow channel maintains an almost steady flow despite the fact that no substantial driving forces exist above $\tau=1$. The driving of outflows depends primarily on conditions in the upflow cell in the inner penumbra, the field line connectivity toward the outer penumbra is secondary and established as a consequence of the outflow (similarly also umbral dots form initially on field lines that might connect to a region several $100$ Mm away, the field line connectivity changes as part of the process until overturning convection is possible). The average field line connectivity found in the penumbra depends on the radial position (combination of ambient field strength and inclination angle). Mass unloading happens primarily through U-loop formation and reconnection in the inner penumbra since flows are not strong enough to completely bend over field lines. In addition, entire field lines can be submerged by laterally overturning convective motions. In the center and outer penumbra, continuous bending of the field lines increases the length of the almost horizontal part near $\tau=1$ until mass can be unloaded at the outer edge of the penumbra. This process happens periodically in the center and permanently over the life time of the filament in the outer penumbra. The fact that the flow speed in the flow channels does not show an increase when field lines bend over completely is a strong indication that the conditions of the outer footpoint are of secondary importance to the process. We cannot rule out additional contributions from unavoidable numerical diffusivity allowing plasma to move across field lines; however, our analysis in Sect. \[sect:simple-model\] did not reveal a very large contribution on average compared to the other terms in the induction equation. The above interpretation shares some similarity with the “fallen flux tube” concept of @Wentzel:1992 combined with the convective driving of outflows described here and previously by @Scharmer:etal:2008. Implications for simplified models ---------------------------------- We address here only models that include driving processes for the Evershed flow, a more general discussion is presented in Sect. \[sect:concl\]. The picture presented above shows substantial differences to stationary siphon flow models that have been proposed to explain the Evershed flow. Those models assume that processes related to turbulent pumping near the outer edge of the penumbra [@Montesinos:Thomas:1997; @Brummell:etal:2008] hold field lines down and establish the field line connectivity that allows then for siphon flows due to pressure differences between the inner and outer footpoint. We see stronger evidence in our simulation for a flow that is driven from within the penumbra regardless of the initial field line connectivity, even though siphon-like flow channels can result from this process in the outer penumbra (see, e.g., filament 5 in Fig. \[fig:field\_lines\]). Since we find fast outflows along horizontal stretches of field lines regardless of their field line connectivity we conjecture that siphon-like flow channels in the outer penumbra are more a consequence of the fast outflow than its cause. This does not rule out additional contributions from processes as described by @Montesinos:Thomas:1997 in the outer penumbra. The filament 6 we highlighted in Fig. \[fig:field\_lines\] is a potential example for a siphon flow related to turbulent pumping near the edge of the penumbra. The footpoint in the outer penumbra is caused by a strong downdraft leading to an U-loop of field lines (see the upward returning flux in the left corner of the indicated sub domain). The outer footpoint of filament 6 has as a consequence fast downflows and low pressure (on average more than $10^5 {\rm dyne}{\rm cm}^{-2}$ lower than the regions most of the field lines connect to further inward). We see a bundle of arch like field lines extending a few $100$ km above $\tau=1$ and having outflow velocities mostly in the $2-4\,\mbox{km\,s}^{-1}$ range (a few faster flows are found in the proximity of the footpoint), which is consistent with the predictions of most stationary siphon models. However, the flow velocities fall short of those present in filaments 2-5, where fast outflows are confined to almost horizontal stretches of the field lines in the deep photosphere over lengths of several Mm. When there are higher reaching arches present such as in filament 3 and 4, flow speeds are declining toward the highpoint, contrary to predictions from stationary siphon flow models. The concentration of driving forces to a very narrow boundary layer beneath $\tau=1$ is not compatible with the acceleration of fluid along several Mm wide arches of field lines extending mostly above $\tau=1$. Fast outflows in the deep photosphere are a natural outcome of the moving flux tube model [@Schlichenmaier:etal:1998a; @Schlichenmaier:etal:1998b] in which processes similar to convective overshoot limit the vertical rise of plasma. The fast outflows found there have been attributed in part to a hot upflow near the inner footpoint that is magnetically deflected outward and in part to horizontal pressure gradients resulting from radiative cooling. The former has some similarity to the magnetic “deflection” we see in our magnetoconvection simulation leading to a process limited to a very narrow height range near $\tau=1$, but we do not see significant contributions from horizontal pressure gradients in the radial direction. The primary difference is that in the simplified flux tube picture there is only one inner footpoint present for the filament, while the driving process seen in our numerical simulation is approximately translation invariant along the filament, i.e. the entire filament is essentially a “footpoint” in which this process accelerates fluid. The translation invariance also implies small contributions from horizontal pressure driving in the radial direction. Deep flow component =================== \[sect:deep-flow\] In the previous section we focused on the driving of flows in the uppermost few $100$ km of the penumbra. While these flows can be primarily explained through strongly enhanced Lorentz force driving, there is no strong contribution present in more than $400$ km depth (cf. Fig. \[fig:energetics\_center\_p2\]). The physical origin of this deeper reaching outflow becomes evident from Fig. \[fig:vr\_vz\_tau\]. At the $\tau=1$ surface horizontal outflows stand out compared to typical granular flows at that height level, while vertical motions in the penumbra are subdued. Furthermore, strong outflows are found preferentially along the x-direction where the nearby opposite polarity spots impose a more horizontal magnetic field. In about $3.4$ Mm depth vertical and horizontal flows do not stand out in terms of amplitude, but rather in terms of the overall flow structure. In the periphery of the sunspot convection cells are arranged in a ring-like pattern in contrast to the random arrangement further away. This preferred arrangement leads to the appearance of large-scale mean flows (outflows away from the sunspot), which have amplitudes comparable to typical horizontal convective flows at the same depth, i.e. the flow amplitude should be a certain fraction of the rms velocity rather independent of depth as indicated in Fig. \[fig:vr\_normalized\]. While the appearance of an approximately axisymmetric mean flow is a consequence of geometric arrangement, the preferred outflow direction requires additional ingredients. A pure arrangement of convection cells in a ring-like fashion should lead to a pair of convection rolls, generating an inflow close to the spot and an outflow further out. A preference for the outflow can be a consequence of the presence of strong magnetic field in the center, which inhibits motions converging toward the spot but has less influence on the diverging motions further out. In addition large-scale pressure gradients can lead to the preference of outflows. We see in this simulation a combination of both. According to Fig. \[fig:energetics\_center\_p2\]a) radial flows in the center penumbra are driven primarily by pressure forces in more than $300$ km depth. We see a preference for pressure driving of outflows, only a weak asymmetry is introduced by the Lorentz force, which opposes inflows more strongly than outflows. Note that most of the pressure driving originates from the region in between $R=16$ and $R=18$ Mm, where the deep flow component gains speed. While the outflow close to $\tau=1$ clearly dominates in terms of flow velocity, the deep reaching flow component transports significantly more mass than the shallow Evershed flow. In Fig. \[fig:mass\_flux\_moat\], we compare vertical profiles of $\langle v_R\rangle$ and $\langle v_R \varrho\rangle$ at the position $R=18$ Mm in the outer penumbra. At a depth of $3$ Mm the mass flux is about $10$ times larger than the the mass flux of the Evershed flow in the photosphere. It is very likely that the deep flow component is related to large-scale moat flows observed around sunspots in the photosphere as well as deeper layers through helioseismology (see, e.g., @Gizon:Birch:2005:lrsp for a recent review and references therein). To clearly establish this relationship we need however a numerical simulation covering a substantially longer time span as well as depth range, which is beyond this investigation. Conclusions {#sect:concl} =========== We presented a detailed analysis of the recent numerical sunspot simulation by @Rempel:etal:Science. Our investigation focused on properties of penumbral fine structure near the $\tau=1$ level as well as the physical mechanisms behind the driving of large-scale outflows in sunspots at photospheric levels and beneath. Our main conclusions can be summarized as follows: 1. We find penumbral fine structure near $\tau=1$, which is compatible with the observationally inferred picture of an interlocking comb structure with fast $>8\,\mbox{km\,s}^{-1}$ Evershed flows along almost horizontal stretches of magnetic field. 2. Correlations between radial flow velocity, intensity, and field strength at the $\tau=1$ level show a good qualitative agreement with recent observations and are a direct consequence of convective energy transport in a sunspot penumbra. 3. The net contribution from large-scale flows to the mass and convective energy flux in the penumbra is found to be about $12-13\%$. Local contributions in the inner penumbra can reach $50\%$. Maintaining the penumbral brightness of $0.7 L_{\odot}$ requires about $1\,\mbox{km\,s}^{-1}$ vertical rms velocity at the $\tau=1$ level. 4. We find in the sunspot penumbra two flow components, which we separate according to their scaling with respect to the convective rms velocity outside the penumbra ($v_{\rm rms}^0$). While the deep component (more than $500$ km beneath the photosphere) has an almost constant ratio to $v_{\rm rms}^0$ of about $50\%$ rather independent of depth, the shallow component (uppermost $500$ km) shows an increase toward the surface steeper than $v_{\rm rms}^0$. While the latter is essentially the Evershed flow, the former is likely related to a deep reaching moat flow component. 5. The near surface flow component is almost exclusively driven through the horizontal component of the Lorentz force along filaments. The energy for maintaining this flow is provided by vertical pressure forces in upflow regions, which are effectively deflected horizontally by an inclined magnetic field. 6. The Evershed flow is driven in a thin boundary layer beneath $\tau=1$. Essential ingredient is a strong vertical increase of $B_R$ beneath $\tau=1$ combined with a moderate vertical average field strength of a few $100$ G. The Evershed flow is strongly magnetized and reaches a peak velocity of about $\vert B_R\vert/\sqrt{4\pi\varrho}$ at $\tau=1$. 7. Upward advection of momentum and magnetic field by overturning convective motions in the penumbra is crucial for maintaining the conditions under which a quasi-stationary Evershed flow can be driven. 8. The deep reaching flow component results from a preferred geometric alignment of convection cells in the periphery of the sunspot. Asymmetries in pressure and Lorentz forces lead to a dominance of the outflow component. The flow amplitude is about $50\%$ of the convective rms velocity, rather independent of depth. 9. Flow channels cannot be easily identified with magnetic flux tubes. The field line connectivity is changing between inner and outer penumbra and we see no compelling evidence that the field line connectivity plays a major role in determining the Evershed flow speed; on the contrary, field line connectivity is established primarily as a consequence of the outflow. A variety of simplified as well as magnetoconvective models for the penumbra have been discussed to explain the Evershed effect. The majority of the simplified models that have been used to describe the acceleration of horizontal flows in the penumbra is based on stationary or dynamic flux tube models. We see strong limitations for the applicability of the thin flux tube approximation in the context of penumbral flow channels. The continuous mass exchange along flow channels due to overturning convective motions is not part of the flux tube picture (it is essentially orthogonal to the assumptions), but found to be substantial in the presented numerical simulation. In addition the changing field line connectivity along the flow channels from inner to outer penumbra does not allow for an easy identification with a flux tube, at least not for the whole length of it. It is in general difficult to find meaningful compact footpoints that are representative of the field in inner and outer penumbra at the same time. We see limitations for the applicability of stationary flux tube models such as @Meyer:Schmidt:1968 [@Thomas:1988; @Degenhardt:1989; @Degenhardt:1991; @Thomas:Montesinos:1993; @Montesinos:Thomas:1997] as explanation for the flows in our simulation. Both, field line connectivity and the causality between outflows and field line connectivity point toward processes in which the conditions in the outer footpoint (if it exists at all) are of minor influence on the flow pattern and outflows are mostly driven by convective motions within the penumbra. In addition fast outflows are found preferentially in the deep photosphere along almost horizontal stretches of field lines regardless of their connectivity, which is different from the situation described in most stationary siphon models to date. This does not rule out additional contributions in the outer penumbra from processes similar to those described in @Montesinos:Thomas:1997, where turbulent pumping by convective motions at the periphery of the penumbra plays a crucial role in establishing the field geometry. Filament 6 in Fig. \[fig:field\_lines\] is one possible example for such a configuration. Note that our simulation might not be fully representative of a “typical” outer penumbra due to the setup with a nearby opposite polarity spot, even though, the combination of strong horizontal field in between both sunspots with a strong downflow lane due to the converging Evershed flows from both sides has a tendency to enhance submergence of field by convective motions. The moving flux tube model of @Schlichenmaier:etal:1998a [@Schlichenmaier:etal:1998b] naturally produces outflows located in the deep photosphere and similarly to the situation in our simulation most driving is focused on the inner footpoint. The acceleration of outflows is attributed to a combination of “deflection” of hot upflows and horizontal pressure gradients resulting from radiative cooling. In contrast to moving flux tube models we see in our simulation strong limits on the overall role horizontal pressure driving plays for the acceleration of plasma in the radial direction, while the magnetic “deflection” of hot upflows is not restricted to the inner footpoint but found everywhere throughout the penumbra. Recently @Thomas:2010 and also Priest (2010, private communication) have suggested that flows in the penumbra could be described in terms of “dynamical” siphon models, in which pressure gradients are produced by MHD processes such as magnetoconvection and these drive time-dependent flows along the magnetic field, which then reacts by Lorentz forces to the presence of the flow: in this view, field line connectivity plays only a secondary role and is more a consequence than a cause. While describing part of the picture, however, one also needs to understand the fluid motions that are responsible for the filamentation and most of the energy transport. Furthermore, we showed that the magnetic driving of outflows originates from narrow boundary layers that form beneath $\tau=1$ in regions where convective upflows are present. These regions arise as a consequence of magnetoconvection and are not captured by flux tube models that do not include a filament sub-structure and overturning convection. @Galloway:1975 presented a phenomenological model based on the “roll-convection” picture introduced by @Danielson:1961, which explains the Evershed flow as a consequence of unbalanced horizontal Lorentz force components: while the Lorentz force balances the gas pressure deficit of the sunspot on average, the filamentation of the penumbra implies a strong azimuthal variation and therefore a violation of this balance locally that is responsible for driving the Evershed flow. It is argued further that the Evershed flow is located in dark filaments, where downward directed motions concentrate magnetic field and lead in return to an above average Lorentz force. While our magnetoconvection simulation certainly produces a Lorentz force that is strongly varying in the azimuthal direction, our results disagree with @Galloway:1975 in detail as we find that most of the Evershed flow is driven in upflow regions. Several other penumbra models focus on the fine structure of the penumbra, without necessarily addressing the driving of large-scale outflows. One of these models which has gotten a lot of attention recently is the “gappy” penumbra model by @Spruit:Scharmer:2006 [@Scharmer:Spruit:2006]. In its original form this model proposed essentially field free upflow plumes embedded in an inclined background field. While this structure certainly captures the essence of the convective picture seen in the numerical simulations presented here, there has been a lot of discussion of how field free these “gaps” really are and to which extent the field strength seen in present simulations is affected by numerical diffusion [see, e.g., discussion in @Nordlund:Scharmer:2010]. The physical explanation of the Evershed flow presented here implies the presence of strong Lorentz forces in the uppermost few $100$ km beneath the $\tau=1$ level in the penumbra, which requires the presence of strong $1-2$ kG magnetic field. In that sense our results are incompatible with models that predict field free gaps at photospheric levels! Furthermore we do not see evidence that the field structure is heavily influenced by numerical dissipation as the analysis presented in the bottom panels of Fig. \[fig:filament\_avr\] reveals. Also it remains very controversial whether the wealth of spectropolarimetric observations could be explained by an essentially unmagnetized Evershed flow [see, e.g., @Thomas:2010]. Nevertheless, the results presented here indicate an almost Salomonian solution to this discussion: Strong field is confined to a narrow boundary layer just beneath $\tau=1$, while further down the field strength is substantially reduced compared to the ambient plasma (but still of the order of $1$ kG). This scenario was also brought forward as a possible solution for this problem in a recent review by @Scharmer:2009. Recently a variety of different magnetoconvection simulations with radiative transfer such as @Heinemann:etal:2007 [@Rempel:etal:2009; @Kitiashvili:etal:2009] have been used to model the penumbra. The models by @Heinemann:etal:2007 and @Rempel:etal:2009 focused primarily on the transition from umbra toward inner penumbra, which corresponds roughly to the innermost edge of the region we analyzed in this investigation. The energy conversion terms for that region are displayed in Fig. \[fig:energetics\_inner\_p\] and do not show (except for the overall amplitude) a fundamental difference to Fig. \[fig:energetics\_center\_p\]. From this we conclude that the driving mechanisms for horizontal outflows in @Heinemann:etal:2007 and @Rempel:etal:2009 is essentially the same as discussed here. The overall picture we described in Sect. \[sect:filaments\] is similar to @Scharmer:etal:2008. @Kitiashvili:etal:2009 studied in an idealized setup the influence of field strength and inclination on large-scale flows and found a strong dependence of outflow speeds on the average inclination and field strength, which is consistent with the mechanism explained here. They also reported on temporal variations of the Evershed flow speed on timescales of $15$ to $40$ minutes. Our analysis essentially reinforces conclusions of @Scharmer:etal:2008 [@Rempel:etal:Science] that the penumbra is anisotropic magnetoconvection and that the Evershed flow can be understood as convective flow component in the direction of the magnetic field. The similarity between plage region and penumbra with respect to the different terms in the kinetic energy equation (Fig. \[fig:energetics\_cmp\]) is quite astonishing, a comparison between the depth profiles of rms velocities points toward anisotropy as the main difference (see @Rempel:etal:Science, supporting online material). Nevertheless, there are also notable differences which clearly differentiate the Evershed flow from horizontal flows in typical convection. While the latter is entirely pressure driven, the Evershed flow is almost completely Lorentz force driven. Only flows that turn over laterally in penumbral filaments remain pressure driven. In addition pressure/buoyancy driving takes place primarily in upflow regions, in contrast to field free convection that is driven by top heavy downflow regions. An interesting new aspect pointed out in this paper is the confinement of the underlying driving mechanism to very narrow boundary layers that exist just beneath $\tau=1$. For clarification we want point out that the driving of the Evershed flow is achieved through the radial component of the Lorentz force while the total work done by Lorentz forces remains negative, i.e. the net effect is a sink for kinetic energy. The overall underlying energy source is convective instability, which enters the kinetic energy balance through pressure/buoyancy driving. The Lorentz force facilitates the energy exchange between the pressure driving in the vertical direction and the horizontal Evershed flow acceleration. A necessary condition for the latter is the shift of pressure/buoyancy driving from downflow to upflow regions in the penumbra we described above. In addition to the mechanism leading to the fast Evershed flow in the upper most few $100$ km of a sunspot penumbra we have also identified a mechanism leading to the formation of a larger scale outflow in deeper layers. In contrast to the Evershed flow the deeper flow scales proportional to the convective rms velocity (outside the sunspot), the dominant radial outflow reaches typically amplitudes $\sim 0.5\,v_{\rm rms}$. The main reason for this flow cell is a preferred circular alignment of convection cells surrounding the sunspot. As a consequence the azimuthal average over this ring-like pattern of convection cells does not vanish and leads to mean flow speeds scaling proportional to $v_{\rm rms}$. A preference for the outflow results from a combination of pressure and Lorentz forces. While this flow does not stand out in terms of flow velocity as the Evershed flow, the radial mass flux is substantially larger, which make a connection with the large-scale moat flows observed around sunspots likely. In this simulation we do not see evidence for a converging collar flow that was found previously in 2D axisymmetric simulations [@Hurlburt:Rucklidge:2000; @Botha:etal:2006; @Botha:etal:2008]. In a future publication we will investigate the subsurface structure of this flow component in deeper domains and evolution over timescales longer than those covered by the numerical simulation presented here. The deep flow component described in Sect. \[sect:deep-flow\] should be in principle observable through local helioseismology. The clear prediction is here an outflow of plasma with an amplitude of about $50\%$ of the convective rms velocity reaching downward several Mm beneath the penumbra. This result is in contradiction with some recent helioseismic inversions such as @Zhao:etal:2001 [@Zhao:etal:2010], which point toward an inflow in a depth range from $1.5$ to $5$ Mm. One the other hand @Gizon:etal:2009 reported on an outflow over the uppermost $5$ Mm. While most of the processes responsible for driving the Evershed flow are located beneath the $\tau=1$ level, there are nevertheless several aspects of the magnetoconvective penumbra model presented here that can be constrained through observations. As presented in Sect. \[sect:phot\] most of the energy is transported in the penumbra by laterally overturning convective motions. We find a very tight relationship between intensity and vertical rms velocity of the form $I \propto \sqrt{v_{z\,rms}(\tau=1)}$. From this follows that the vertical rms velocity at $\tau=1$ in the penumbra with $I\approx 0.7 I_{\odot}$ should be about half of the value found in the quiet sun, i.e. about 1 $\mbox{km\,s}^{-1}$. This value is consistent with the recent findings of @Franz:Schlichenmaier:2009, who computed from [*Hinode*]{} observations velocity distributions functions for both quiet Sun and penumbra (see Fig.3 in their paper). The half width at half maximum of the vertical velocity distribution function for the penumbra is about $500\,\mbox{m\,s}^{-1}$, while the same analysis results in $1\,\mbox{km\,s}^{-1}$ for the quiet Sun, i.e. the latter falls short by about a factor of $2$ compared to the value we find for the quiet Sun at $\tau=1$ (due to a combination of limited observational resolution as well as the sharp decline of $v_z$ above $\tau=1$). If we assume that the same shortfall applies also to the penumbra, the vertical velocity structure reported in @Franz:Schlichenmaier:2009 is at least in a statistical sense fully consistent with the amount of overturning convection we see in the numerical simulation presented here. Other consequences of the magnetoconvective model are the sign changes in the $I - v_R$ and $B - v_R$ correlations presented in Fig. \[fig:corr\]. They are consistent with the analysis of [*Hinode*]{} data presented by @Ichimoto:etal:2007 (see Fig. 3 therein). A positive $B - v_R$ correlation in the outer penumbra was suggested by @Tritschler:etal:2007 and @Ichimoto:etal:2008 based on observations of the net circular polarization (NCP) at different viewing angles. As explained in Sect. \[sect:simple-model\] the observable consequences of the Evershed flow driving mechanism are a moderate increase of $B_R$ and a steep increase of $v_R$ toward $\tau=1$. $v_z$ shows a very steep gradient in the deep photosphere. The peak velocity of the Evershed flow in the deep photosphere should be around $\vert B_R\vert/\sqrt{4\pi\varrho}$. The fact that the simulated Evershed flow is a deep photospheric flow is a direct consequence of its convective origin. On the observational side the depth dependence of the Evershed flow is debated. While the investigations by @Rimmele:1995 and @Stanchfield:etal:1997 point toward flows in elevated flow channels, recent work by @Schlichenmaier:etal:2004, @BellotRubio:etal:2006, and @Borrero:etal:2008 is in support of a flow in the deep photosphere declining with height. Another point heavily debated is the presence or absence of overturning convection in the penumbra. Support for overturning convection is found by @Ichimoto:etal:2007:sc [@Zakharov:etal:2008; @Rimmele:2008; @Bharti:etal:2010], while @BellotRubio:etal:2005 [@Ichimoto:etal:2007; @Franz:Schlichenmaier:2009; @BellotRubio:etal:2010] see primarily support for Evershed flow related upflows in the inner and downflows in the outer penumbra – a flow pattern that accounts in our model only for a small fraction of the unsigned mass and energy flux integrated over the penumbra. It appears that overcoming the discrepancy between the presence of overturning convection in MHD simulations and the lack of evidence in many high resolution observations is one of the biggest challenges both numerical models and observations will face in the future. It is unlikely that the absence of overturning convection is the solution to this discrepancy; a brightness of $0.7 I_{\odot}$ or more requires overturning mass flux at a level not much less than granulation. An other possible solution could be related to thin boundary layers, which are indicated but not well resolved in the simulation presented here. If quantities such as flow velocities and magnetic field change dramatically over short distances, moving a $\tau$-surface by a distance comparable to our grid spacing can make a dramatic differences for the visibility of such feature. This clearly indicates that the simulation presented here can only be considered as a first step in that direction. A convergence study of the properties highlighted in this investigation covering the resolution range from $96\times 32$ to $16\times 12$ km resolution (horizontal $\times$ vertical) is in progress. A preliminary analysis shows that most magnetoconvective properties of the penumbra are robust (qualitative agreement over the whole range investigated, quantitative agreement from $48\times 24$ km resolution upward), while the photospheric appearance of sunspot fine structure improves substantially with resolution. The currently highest resolution case is presented in @Rempel:2010:IAU. The National Center for Atmospheric Research is sponsored by the National Science Foundation. Computing resources were provided by NCAR’s Computational and Information Systems Laboratory (CISL). M. Rempel is grateful to Manfred Sch[ü]{}ssler, Alfred De Wijn, Michael Kn[ö]{}lker, Keith MacGregor, and Eric Priest for helpful comments on the manuscript. M. Rempel also thanks the referee of this paper, J.H. Thomas, for providing feedback that substantially improved the presentation.

ArXiv

--- author: - 'W.E.A. Lorenz[^1]' - 'R.O. Kuzian' - 'S.-L. Drechsler' - 'W.-D. Stein' - 'N. Wizent' - 'G. Behr' - 'J. Málek' - 'U. Nitzsche' - 'H. Rosner' - 'A. Hiess' - 'W. Schmidt' - 'R. Klingeler' - 'M. Loewenhaupt' - 'B. Büchner' bibliography: - '99.bib' title: 'Highly Dispersive Spin Excitations in the Chain Cuprate [Li$_2$CuO$_2$]{}' --- [**1. INTRODUCTION**]{} ======================= Li$_2$CuO$_2$ is the first [@Hoppe1970] and the most frequently studied compound of the growing class of edge-shared spin-chain cuprates [@Matsuda1996; @Matsuda2001; @MasudaPRB2005; @Enderle2005; @Drechsler2007PRL; @Drechsler2007JP]. Owing to its structural simplicity with ideally planar CuO$_2$ chains (see Fig. 1) it has been considered as a model quasi-one-dimensional (1D) frustrated quantum spin system. In almost all edge-shared cuprate chain compounds the spins are expected to be coupled along the chains via nearest neighbor (NN) ferromagnetic (FM) and next-nearest neighbor (NNN) antiferromagnetic (AFM) exchange interactions, $J_1$ and $J_2$, respectively. Due to the induced frustration FM and spiral in-chain correlations are competing. While in the 1D model the ground state is governed by the ratio $\alpha=-J_2/J_1$, the actual 3D magnetic order sensitively depends on the strength of the inter-chain couplings and anisotropy. In [Li$_2$CuO$_2$]{}, below ${T_\mathrm{N}}\approx 9 {\,\mathrm{K}}$ [@Sapina1990; @Chung2003] a long-range collinear commensurate AFM inter-chain with FM in-chain (CC-AFM-FM) magnetic ordering evolves. However, a proper understanding, necessary for ![(Color online) Left: The crystallographic structure of [Li$_2$CuO$_2$]{} comprises two AFM coupled CuO$_2$ spin-chains per unit cell running along the $b$-axis (orange [ $\bullet$]{} – Cu$^{2+}$, red [ $\bullet$]{} – O$^{2-}$, bright blue [ $\bullet$]{} – Li$^{+}$). The unit cell is indicated by the outer black cuboid. Right: the main intra- and inter-chain exchange paths, $J_1$, $J_2$, and $\tilde{J}_2$ marked by blue arcs and dashed lines, respectively. Notice the frustration introduced by an AFM inter-chain coupling for any non-FM in-chain ordering. []{data-label="fig::Struct"}](Fig1alorenz.eps "fig:"){width="13.00000%"} ![(Color online) Left: The crystallographic structure of [Li$_2$CuO$_2$]{} comprises two AFM coupled CuO$_2$ spin-chains per unit cell running along the $b$-axis (orange [ $\bullet$]{} – Cu$^{2+}$, red [ $\bullet$]{} – O$^{2-}$, bright blue [ $\bullet$]{} – Li$^{+}$). The unit cell is indicated by the outer black cuboid. Right: the main intra- and inter-chain exchange paths, $J_1$, $J_2$, and $\tilde{J}_2$ marked by blue arcs and dashed lines, respectively. Notice the frustration introduced by an AFM inter-chain coupling for any non-FM in-chain ordering. []{data-label="fig::Struct"}](Fig1blorenz.eps "fig:"){width="35.00000%"} a critical evaluation of theoretical studies, especially of electronic/magnetic structure calculations [@Mizuno1998; @Weht1998; @deGraaf2002; @DrechslerJMMM2007; @Xiang2007], is still missing. {width="\textwidth"} {width="\textwidth"} In particular, this concerns a precisely enough knowledge of the main exchange interactions. The knowledge of realistic values is helpful also for the understanding of related “frustrated ferromagnets” [@Dmitriev2008; @Plekhanov2009] such as Ca$_2$Y$_2$Cu$_5$O$_{10}$ [@Matsuda2005], La$_6$Ca$_8$Cu$_{24}$O$_{41}$ [@Matsuda1996] with FM in-chain ordering and LiVCuO$_4$, LiCu$_2$O$_2$ with helimagnetism and multiferroicity, all being of considerable current interest. A previous inelastic neutron scattering (INS) study aimed to determine the exchange integrals was not conclusive [@Boehm1998]. It revealed an anomalous low-lying branch of hardly dispersive and overdamped spin excitations in chain direction. Its linear spin-wave (LSW) analysis results in unrealistically small in-chain exchange integrals. The missing, but expected, dispersive quasi-1D spin chain excitation remained as a challenging puzzle for the community [@Weht1998; @Mizuno1998; @Mizuno1999]. In Sec. 3 we present new INS data which unambiguously show the presence of a strongly dispersive in-chain spin mode. The main exchange integrals are derived applying the LSW-theory [@Oguchi1960] and in Sec. 4 we compare our results with those of related chain cuprates as well as with predictions of band structure and cluster calculations. A criticism of improper <span style="font-variant:small-caps;">Curie-Weiss</span> analysis of spin susceptibility data is provided and consequences for the direct FM Cu-O exchange parameter $K_{pd}$ entering extended [<span style="font-variant:small-caps;">Hubbard</span>]{} models are discussed. [**2. EXPERIMENTAL**]{} ======================= A single crystal of $^7$[Li$_2$CuO$_2$]{} was grown for INS experiments by the travelling solvent floating zone technique under high pressure [@Behr2008]. In order to avoid vaporization of Li$_2$O during growth a 4:1 Ar:O$_2$ atmosphere at $50\ab{\,\ab{bar}}$ was chosen. A fast growth rate of $10\,\ab{mm/h}$ inhibits growth of impurity phases. Isotope enriched $^7$Li was employed to avoid the significant neutron absorption coefficient of $^6$Li. The sample was characterized by X-ray powder diffraction, polarized light microscopy, magnetization and specific heat measurements. By X-ray powder diffraction no impurity phase was found. The macroscopic magnetization and specific heat data of the sample agree with literature data, i.e. AFM order is found below ${T_\mathrm{N}}= 9.2{\,\mathrm{K}}$ and a weak FM component evolves below $T_2 \approx 3{\,\mathrm{K}}$ [@Ortega1998; @Staub2000; @Chung2003]. INS experiments were performed with thermal and cold neutrons at the three-axis-spectrometers IN8 and IN12 at the ILL, Grenoble, France. Four single crystals with a total mass of 3.8g were mounted together in the $(0$ $K$ $L)$ scattering plane with a resulting sample mosaicity of $3^\circ$. For both instruments, focusing PG(002) monochromator and analyzer have been utilized. The measurements at IN8 were taken with fixed final momentum $k_\ab{f}=2.662{\,\mathrm{\AA}}^{-1}$ with PG-filter on $k_\ab{f}$. IN12 was configured with $k_\ab{f}=1.5{\,\mathrm{\AA}}^{-1}$ and Be-filter on $k_\ab{f}$. Most scans have been done in the CC-AFM-FM phase at $T=4.1{\,\mathrm{K}}$ well above another not yet well understood magnetic phase below  . Anyhow, the observed changes of the INS spectra in this low-$T$ phase (not shown here) are weak. [**3. RESULTS AND DATA ANALYSIS**]{} ==================================== The results of our INS studies are summarized in [Fig.\[fig::INS\]]{}. Representative spectra of constant energy scans as taken at IN12 and IN8 for moment transfer along the chains ($b^*$) are displayed in [Fig.\[fig::INS\]]{}(a). The main result is the observation of a highly dispersive excitation which is strong at the magnetic zone center and significantly weakens at higher energies. With the chosen experimental setup the magnetic branch could be traced up to energy transfers of $25{\,\mathrm{meV}}$. The measured data points along $(0$ $K$ $1)$ taken at IN12 with cold neutrons are summarized in the color map [Fig.\[fig::INS\]]{}(b). Note, that the reflections are periodic with the magnetic unit cell and their strongly reduced intensity above ${T_\mathrm{N}}$ observed up to energy transfers of $15{\,\mathrm{meV}}$ does confirm their magnetic nature (see Fig. 2 (c)). At the magnetic zone center $(0$ $0$ $1)$, a gap of $\Delta = 1.36{\,\mathrm{meV}}$ is observed. The excitations for momentum transfer along $(0$ $0$ $1+L)$ are only weakly dispersive in agreement with the results of Ref.  [@Boehm1998]. Respective constant ${\mathbf}{q}$-scans are shown in [Fig.\[fig::INS\]]{}(c). Note, that the mosaicity of the sample broadens the excitations along $L$ which is less pronounced for moment transfer along $K$ due to the longer $b^*$-axis. As shown in [Fig.\[fig::INS\]]{}(b) we observe further inelastic features for moment transfer along the chain. In addition, the data in [Fig.\[fig::INS\]]{}(b) also exhibit weak and presumably incommensurate (IC) magnetic scattering below the magnon gap energy. The origin of these low-energy excitations is not yet clear and will be addressed in future studies. Furthermore, there is a continuous feature appearing at double the energy of the anisotropy gap which is possibly attributed to two-magnon scattering. We also mention that we have observed low-lying and strongly broadened excitations along $b^*$ similarly as in Ref. [@Boehm1998]. However, the intensity of these excitations was roughly two orders of magnitude weaker than that reported ibidem. According to ESR measurements [@Ohta1993] the exchange interactions show an uniaxial anisotropy with the easy-axis directed along the crystallographic $a$-axis. We describe the corresponding Cu momenta by the spin-[<span style="font-variant:small-caps;">Hamilton</span>]{}ian $$\hat{H}=\frac{1}{2}\sum_{\mathbf{m},\mathbf{r}} \left[J_{\mathbf{r}}^{z}\hat{S}_{\mathbf{m}}^{z}\hat{S}_{\mathbf{m} +\mathbf{r}}^{z}+J_{\mathbf{r}}^{xy} \hat{S}_{\mathbf{m}}^{+}\hat{S}_{\mathbf{m}+\mathbf{r}}^{-} \right]\label{eq:H}$$ where $\mathbf{m}$ enumerates the sites in the magnetic (Cu) lattice, the vector $\mathbf{r}$ connects sites with an exchange coupling $J_{\bf r}$ [@remarkboehm]. The $z$-axis is taken along the easy-axis, i.e. the $a$-axis. Within the LSW-theory [@Oguchi1960], the dispersion-law reads $$\begin{aligned} \omega_{\mathbf{q}} & = & \sqrt{\left(J_{\mathbf{q}}^{xy}-J_{\mathbf{0}}^{xy}+\tilde{J}_{\mathbf{0}}^{xy} -D\right)^{2}-\left(\tilde{J}_{\mathbf{q}}^{xy}\right)^{2}},\label{eq:wq}\end{aligned}$$ where $J_{\mathbf{q}}\equiv(1/2)\sum_{\mathbf{r}}J_{\mathbf{r}}\exp\left(\imath\mathbf{qr}\right)$ is the Fourier transform of the in-chain exchange integrals, and analogously for the inter-chain integrals $\tilde{J}_{\mathbf{q}}$. The exchange anisotropy $D\equiv J_{0}^{z}-J_{0}^{xy}-\tilde{J}_{0}^{z}+\tilde{J}^{xy}_0$ causes the abovementioned spin gap $\Delta =\omega_{\mathbf{0}}$ in our case (see Fig. 3). Their relation reads $$\Delta=\sqrt{D\left(D-2\tilde{J}_{0}^{xy}\right)},\quad \mbox {or} \quad D=\tilde{J}_{\mathbf{0}}^{xy}-\sqrt{\left(\tilde{J}_{\mathbf{0}}^{xy}\right)^{2}+\Delta^{2}} \ . \label{eq:D}$$ In the summations over $\mathbf{r}$ we retain only the leading terms (see Fig. 1). According to LSDA+$U$ based magnetic structure calculations they are given by the following in-chain integrals: $J_{1},\: J_{2},\: J_{3}$, (corresponding to $\mathbf{r}=\mathbf{b},\:2\mathbf{b},\:3\mathbf{b}$ respectively) and inter-chain integrals: $\tilde{J}_{111},\:\tilde{J}_{131}$, (corresponding to $\mathbf{r}_{111} =\left(\mathbf{a}+\mathbf{b}+\mathbf{c}\right)/2$, $\mathbf{r}_{131}=\left(\mathbf{a}+3\mathbf{b}+\mathbf{c}\right)/2$). For $q_aa$ or $q_cc=\pi$ the inter-chain dispersion caused by $\tilde{J}_{\mathbf{q}}^{xy}$ vanishes and Eqn. (\[eq:wq\]) simplifies: $$\omega_{\mathbf{q}} = J_{\mathbf{q}}^{xy}-J_{\mathbf{0}}^{xy}+\tilde{J}_{\mathbf{0}}^{xy} -D=J_{\mathbf{q}}^{xy}-J_{\mathbf{0}}^{xy}+\Delta_1 ,$$ i.e. the single-chain dispersion can be read off [*directly*]{} by subtracting the effective gap $\Delta_1$, only. Our INS data are well fitted by <span style="font-variant:small-caps;">Gauss</span>ian distributions. The maxima of the main branch of the spectrum were analyzed within LSW-theory [Eqn.(\[eq:wq\])]{}. The inspection of Fig. 3 reveals a strong in-chain dispersion and a much weaker one in the perpendicular $c$ and $a$ (not shown here) directions. The results of the fit are given in [Table\[tab::ExchangeParameters\]]{}. The total width $W=\omega_{\pi/b}-\Delta= 2\mid J_1+J_3+J_5+\cdots-0.5\left( \tilde{J}^{xy}_0-D \right) \mid -\Delta \approx 2\mid J_1\mid $, i.e. the in-chain dispersion yields a direct measure of the NN coupling since $W$ is [*un*]{}affected by $J_2, J_4, J_6 \cdots$. Supposing a monotonous behaviour of $\omega_q$ up to the zone boundary at $Q=0.5$ (corresponding to $q=\pi/b$), from the measured part of the full $\omega_q$-curve one obtains already a rigorous lower bound for $\mid J_1\mid \stackrel{>}{\sim}$ 150 K. Our analysis shows that the in-chain NN interaction $J_1$ is strongly FM, but frustrated by an AFM NNN-coupling $J_2$ which affects the shape of $\omega_q$. In comparison, the AFM inter-chain coupling is weak, clearly demonstrating the magnetically quasi-1D character of the compound. For the in-chain coupling we find $\alpha = -J_2/J_1 = 0.33$, [*unambigously above*]{} the critical ratio $\alpha_{\ab{crit}}=1/4$ for an isotropic <span style="font-variant:small-caps;">Heisenberg</span>-chain [@Bursill1995]. Note that the dispersion near the zone center behaves like $\omega(q)\propto q^4$ to be discussed below. We confirm also theoretical predictions [@deGraaf2002; @Xiang2007] (see also Tab. 1 for our results) that the main inter-chain coupling is indeed the NNN coupling along $(\frac{a}{2},\frac{3b}{2},\frac{c}{2})$. The NN inter-chain exchange has a negligible effect on the dispersion in the (0 $K$ $L$)-plane. Although further couplings can not be accessed from our fits, the main $J$ values given here do not change much, if the INS data are analyzed in more complex models with additional exchange paths, especially $J_3$ and $\tilde{J}_1$. When taken into account, both remain small ($\sim 4$ K and 1 K, respectively) in full accord with L(S)DA+$U$ (see Tab. 1). [**4. DISCUSSION**]{} ===================== The actual CC-AFM-FM ordering seemingly contradicts the IC “spiral” phase expected for a frustration ratio of $\alpha \approx 0.33$ in a 1D-approach (in the sense of the wave vector $q_0\neq 0,\pi/b$ where the magnetic structure factor $S(q)$ becomes maximal [@Bursill1995]). Hence, the obtained relatively small but frustrated AFM inter-chain coupling may hinder the spiral formation. Thus, the 3D critical point compared with $\alpha_c^{1D}$ is upshifted: $$\alpha_c^{\rm \tiny 3D,iso}=\alpha_c^{1D}\left(1+\beta_1 +9\beta_2 +25\beta_3 + \cdots \right) \ , \label{interchain}$$ where $\beta_n= -\tilde{J}_n/J_1$. Eqn. (\[interchain\]) has been derived in the isotropic (iso) case [@Drechsler2005]. Ignoring all other very weak inter-chain couplings $\tilde{J}_n$ we arrive with our results $\tilde{J}_2=9.04$ K and $J_1=-228$ K at $\alpha^{\rm \tiny 3D, iso}_c=0.339$. Anisotropies (aniso), as found here, further stabilize the CC-AFM-FM state. From Eqns. (2,3) we estimate finally $\alpha_c^{\rm \tiny 3D,aniso}\approx 0.39$. The combined effect of AFM inter-chain coupling and easy-axis anisotropy is also responsible for the anomalous $q^4$-dependence of the spin excitations mentioned above. In the limit $q_bb \rightarrow 0$ we expand Eqn. (2) and obtain $$\omega(q)\approx\Delta +A_{\Gamma}(q_bb)^2+B_{\Gamma}(q_bb)^4,$$ ------------------------------------------- -------------- ------------------ ------------------------ ---------- ---------------- ---------------- --------------- $J_1$ $\alpha$ $J_2=-\alpha\cdot J_1$ $J_3$ $\tilde{J}_2$ D $\tilde{J_1}$ INS / present work $-228\pm 5$ $0.332\pm 0.005$ $76\pm 2$ (3.8) $9.04\pm 0.05$ $-3.29\pm 0.2$ (1) 3$d$O2$p$ / present work[@remarkmapping]: $-218$ $0.30$ $66$ $-0.4$ $-$ $-$ $-$ 3$d$O2$p$ [@Malek2008]: $-143$ $0.23$ $33$ $-1$ $-$ $-$ $-$ 3$d$O2$p$ [@Mizuno1998]: $-103$ $0.47$ $49$ $-2$ $-$ $-$ $-$ two-chain phenomenol. [@Mizuno1999]: $-100$ $0.40$ $40$ $-$ 16 $-$ 16 LSDA+$U$, $U=$ 6 eV [@Drechsler2009]: $-216 \pm 2$ 0.31 $66\pm 2$ $5\pm 2$ $13\pm 2$ $-$ 0$\pm 2$ GGA +$U$, $U=$ 6 eV[@Xiang2007]: $-171$ 0.60 98 $-$ 18 $-$ 0.23 RFPLO, LAPW+SO[@Mertz2005]: $-$ $-$ $-$ $-$ $-$ $-15.6 $ $-$ ------------------------------------------- -------------- ------------------ ------------------------ ---------- ---------------- ---------------- --------------- $A_{\Gamma}$ and the quadratic dispersion vanish exactly at $$\alpha^q_0=\frac{1}{4}\left(1+\frac{9\beta_2}{\delta}\right) = 0.33098, \quad \delta =1-\frac{D}{4\tilde{J}_2}.$$ Accidentally [Li$_2$CuO$_2$]{} is very close to this point and its dispersion near the zone center is [*quasi-quartic*]{}. But in the presence of a spin gap $\Delta$ caused by the anisotropy $D$ this vanishing of the quadratic dispersion doesn’t yet signal an instability of the CC-AFM-FM state. Near the Z-point $\left(0,0,\pi/c \right)$ the quadratic coefficient $A_Z$ is already essentially negative (see Fig. 3 (b, right panel)). Since along the line $Z-R(0,\pi/b,\pi/c)$ the inter-chain dispersion vanishes one can easily read off the 1D Fourier components of the exchange interactions $J^{xy}_{\mathbf{q}}$ (see Eqn. (4)). A similar rare situation occurs in the 2D frustrated CsCuCl$_4$ system in a high magnetic field above its saturation limit [@coldea2002]. The clearly visible minima at $q_{b,0}b=\cos^{-1}\left(1/4\alpha \right)\approx\pm 0.72=\pm 0.11$ (r.l.u.) correspond to the two equivalent propagation vectors of a low-lying spiral excitation [@Kuzian2007] with a pitch angle of about 41.2$^{\circ}$ above a CC-AFM-FM ground state observed to the best of our knowledge for the first time. Next, we briefly compare our results with those obtained so far by INS-studies for Ca$_2$Y$_2$Cu$_5$O$_{10}$ (CYCO) [@Matsuda2005] with a similar FM in-chain ordering and a frustrating AFM inter-chain interaction. There the reported $J_1$ read -80 K and -93 K for fits where $J_2=0$ and $J_2= 4.6$ K, respectively. However, such tiny values of $J_2$ are unlikely [@Mizuno1998; @Mizuno1999; @Malek2008]. For a standard Cu-O hybridization a much larger value is expected [@Kuzian2009] in accord with the observed sizable part of the total magnetic moment (22 % ) residing at O [@Matsuda2002]. Re-fitting their INS data yields $J_1$,$J_2$ values of the same order as we found for [Li$_2$CuO$_2$]{} (LCO) ($J_1^{\tiny \rm CYCO}\sim J_1^{\tiny \rm LCO}$ and $J_2^{\tiny \rm CYCO}\sim 0.5J_{2}^{\tiny \rm LCO}$). A detailed comparison of both systems will be given elsewhere [@Kuzian2009]. With respect to their large $J_1$ values the question may arise why they have been not recognised so far in analyzing thermodynamic properties? In this context a critical evaluation of the reported AFM “<span style="font-variant:small-caps;">Curie-Weiss</span>” (CW) temperatures $\Theta_{\tiny \rm CW}^{\tiny \rm CYCO}\approx -15$ K [@Yamaguchi1999] or small FM values: 5-10 K [@Kudo2005] and $\Theta_{\tiny \rm CW}^{\tiny \rm LCO}\approx -40$ [@Sapina1990; @Boehm1998] or -8 K [@Ebisu1998] is very instructive. All these data have been derived from the linear fits of inverse susceptibility plots below 300–400 K (of the type denoted as “pseudo”-CW-lines in Fig. 5). But here we estimate $$\Theta_{\tiny \rm CW}\approx \frac{1}{2}\left[\mid J_1\mid-J_2- \frac{z_{int-ch}}{2}\left(\tilde{J}_1+\tilde{J}_{2} \right) \right] > +54 \ \mbox{K},$$ where the inter-chain coordination number $z_{int-ch}=4,8$ for CYCO and LCO, respectively. Thus, we arrive at about $ 60$ K and 58$\pm 4$ K, respectively. The inspection of [Fig.\[fig::InvChi\]]{} clearly shows that very high $T$ [@remarkCW], [*far above*]{} any available data would be required to extract $\Theta_{\tiny \rm CW}$ from $1/\chi(T)$-data. To satisfy the incorrect $\Theta_{\tiny \rm CW}$-values strongly underestimated $J_1$-values have been adopted in Refs. [@Yamaguchi1999; @Kudo2005; @Ebisu1998; @Boehm1998; @Mizuno1998]. Even the 2$^{nd}$ order of high $T$-series expansion (HTS) approaches the experimental curve only above 400 K. Hence, even more any attempt to detect even a <span style="font-variant:small-caps;">Curie</span>-law [@Mizuno1998] near 300 K must fail. Our $J_1$-values for LCO and CYCO from INS-data provide support for the large value we found in Li$_2$ZrCuO$_4$ from thermodynamic properties [@Drechsler2007PRL] but puzzle the tiny value reported for LiVCuO$_4$ [@Enderle2005] with a similar Cu-O-Cu bond angle. Finally, we turn to a microscopic analysis. In Tab. 1 and Fig. 3 we compare our INS derived exchange integrals with theoretical results. First, we list the in-chain couplings $J_n$ as obtained from the mapping of a five-band extended Hubbard $pd$ model (on open chain Cu$_n$O$_{2n+2}$-clusters $n=5,6$) [@remarkmapping] on a corresponding $J_1$-$J_2$-$J_3$-[<span style="font-variant:small-caps;">Heisenberg</span>]{} model (see Fig. 1). But here, to reproduce the main experimental exchange integrals, a refinement has been performed most importantly by considering a larger direct FM Cu-O exchange $K_{pd}=81$ meV compared with 50 meV adopted in Ref.  [@Mizuno1999]. We note that practically only $J_1$ is significantly affected by $K_{pd}$. Thereby $\mid J_1 \mid \propto K_{pd}$ holds approximately. Notice that the contribution of $K_{pd}$ is much more important for the large negative (FM) value of $J_1$ than that of the intra-atomic FM Hund’s rule coupling on O. Since the available spectroscopic data at 300 K depend only weakly on $K_{pd}$ not much is known on its magnitude. In the past $K_{pd}$ has been used mostly as a fitting parameter for thermodynamic properties ranging from 50 to 110 meV for CuGeO$_3$ [@Mizuno1998; @Braden1996]. The INS data reported here provide a unique way to restrict its value phenomenologically and opens a door for systematic studies of this very important interaction and well-founded comparisons with other edge-shared CuO$_2$ chain compounds. Secondly, in the LSDA+$U$ there is practically only one adjustable parameter $U_d-J_H$, where $U_d$ denotes the Coulomb onsite repulsion (between 6 and 10 eV) and $J_H$ denotes the intra-atomic exchange ($ \approx 1$ eV) both on Cu-sites. Comparing the total energy of various ordered magnetic states, a set of in-chain and inter-chain integrals can be derived [@Xiang2007]. As a result one arrives again at very close numbers to our INS derived set [@Drechsler2009]. Noteworthy, both ED a well as the LSDA+$U$ provides a justification to neglect any long-range exchange beyond the third NN. The latter also explains why there is only one important inter-chain exchange integral $\tilde{J}_2$ ($\beta_2$). The excellent agreement between the INS-data analyzed in the simple LSW-theory and the theoretical results/predictions suggests that in the present case the effect of quantum fluctuations as well as of spin-phonon interaction seems to be rather weak. The former point is also supported by the relatively large value of the magnetic moment $m\approx 0.96\mu_{\tiny \rm B}$ [@Sapina1990; @Chung2003] in the ordered state below $T_N$ and a consequence of the fact that the FM state is an eigenstate of the 1D spin-model in contrast to the <span style="font-variant:small-caps;">Néel</span> state. Concerning the value of the anisotropy, there is no good agreement between contemporary DFT calculations [@Mertz2005] and much smaller values obtained in various experiments [@Ohta1993; @Boehm1998] including our data (see Tab. 1). **5. SUMMARY** ============== The main results of our INS study are (i) the relatively large dispersion of spin excitations in the CuO$_2$ chains of [Li$_2$CuO$_2$]{}  due to the large value of the FM NN in-chain coupling $J_1$ and (ii) the observation of a low-energy spiral excitation over a commensurate collinear <span style="font-variant:small-caps;">Néel</span> ground state in the vicinity of the 3D critical point above the corresponding 1D point. The obtained main exchange integrals can be approximately reproduced adopting an enhanced value for the direct FM exchange $K_{pd}$ between Cu 3$d$ and O 2$p$ states within an extended five-band [<span style="font-variant:small-caps;">Hubbard</span>]{}-model. Further support for the empirical exchange integrals comes from L(S)DA+$U$ calculations, if a moderate value of $U$ somewhat smaller than the $U_d$ in exact diagonalization for the extended [<span style="font-variant:small-caps;">Hubbard</span>]{}-model is employed. The achieved detailed knowledge of the main exchange couplings derived from the INS-data provides a good starting point for an improved general theoretical description of other CuO$_2$-chain systems and to adress a microscopic theory of their exchange anisotropy. We thank the DFG \[grants KL1824/2 (BB, RK & WEAL), DR269/3-1 (S-LD & JM), & the E.-Noether-progr. (HR)\], the progr.PICS \[contr.CNRS 4767, NASU 243 (ROK)\], and ASCR(AVOZ10100520) (JM) for financial support as well as M. Boehm, M. Matsuda, A. Boris, H. Eschrig, V.Ya. Krivnov, D. Dmitriev, S. Nishimoto, E. Plekhanov and J. Richter for valuable discussions. [99]{} url\#1[`#1`]{} and references therein. *et al.* [*et al.*]{} . . The $J_n$ given in Tab. 1 in the 4$^{th}$ row we have estimated in the same way as proposed in the present work and Ref. [@Malek2008]. They slightly differ from those for a dimer and an artificial trimer (without the central Cu site) given by the authors. . The $J$-values shown in Tab. 1 are linear interpolations between the values given for $U_{eff}=U-J_{d}=4$ and 6 eV by the authors since in our calculation the standard value $J_{d}=1$ eV for the Hund’s rule coupling has been used. [*et al.*]{} [*et al.*]{} The relatively large dispersion in view of $J_1=100$ K, only, (shown in Fig. 4) results from an adopted significant inter-chain coupling (see Tab. 1) $\tilde{J}_1=\tilde{J}_2=16$ K. Their sum strongly exceeds our empirical INS or L(S)DA+$U$- derived effective value by a factor of 3.5. . *et al*., Our notation coincides with that used in Ref. [@Matsuda1996] but differs from that of Ref. [@Boehm1998]: we denote the interaction of a pair of spins $\hat{S}_{\mathbf{m}}^{\alpha }$ and $\hat{S}_{\mathbf{m}+\mathbf{R}}^{\alpha }$ as $J_{\mathbf{R}}^{\alpha }$. We use positive(negative) signs for AFM(FM) exchange, respectively. [*et al.*]{} [*et al.*]{} $ T > 1000-2000{\,\mathrm{K}}$, where the 2$^{nd}$ order of the HTS approaches the <span style="font-variant:small-caps;">Curie-Weiss</span> line. Our parameters were adapted to describe optical, EELS and O 1$s$ XAS-spectral data [@Malek2008; @Neudert1999] at 300 K. Careful measurements in the whole $T$-range down to 4 K as reported here for the INS would be very helpful for their further refinement, especially concerning the intersite Coulomb interactions $V_{pd}$, $V_{pp}$, and $V_{dd}$. But here and in Ref.[@Malek2008] $V_{pd}$ etc. have been ignored for the sake of simplicity. A detailed discussion will be given elsewhere. [*et al.*]{} . [*et al.*]{} [*et al.*]{} . \[Cbib\] [^1]: E-mail:

ArXiv

--- abstract: 'We study the distribution of the largest eigenvalue in the “Pfaffian” classical ensembles of random matrix theory, namely in the Gaussian orthogonal (GOE) and Gaussian symplectic (GSE) ensembles, using semi-classical skew-orthogonal polynomials, in analogue to the approach of Nadal and Majumdar (NM) for the Gaussian unitary ensemble (GUE). Generalizing the techniques of Adler, Forrester, Nagao and van Moerbeke, and using “overlapping Pfaffian” identities due to Knuth, we explicitly construct these semi-classical skew-orthogonal polynomials in terms of the semi-classical orthogonal polynomials studied by NM in the case of the GUE. With these polynomials we obtain expressions for the cumulative distribution functions of the largest eigenvalue in the GOE and the GSE. Further, by performing asymptotic analysis of these skew-orthogonal polynomials in the limit of large matrix size, we obtain an alternative derivation of the Tracy-Widom distributions for GOE and GSE. This asymptotic analysis relies on a certain Pfaffian identity, the proof of which employs the characterization of Pfaffians in terms of perfect matchings and link diagrams.' author: - Anthony Mays$^1$ - Anita Ponsaing$^2$ - Grégory Schehr$^3$ - bibliography: - 'MPS2020.bib' title: 'Tracy-Widom distributions for the Gaussian orthogonal and symplectic ensembles revisited: a skew-orthogonal polynomials approach' --- Introduction ============ Since their discovery more than 25 years ago, the Tracy-Widom (TW) distributions [@forrester1993spectrum; @tracy1994level; @TracWido1996] have become cornerstones of extreme value statistics of strongly correlated variables [@majumdar2020extreme]. While they were initially found as the limiting distributions describing the typical fluctuations of the largest eigenvalues of large random matrices belonging to the classical Gaussian ensembles of random matrix theory (RMT), namely the Gaussian orthogonal, unitary and symplectic ensembles (respectively denoted as GOE, GUE and GSE), they have since found a large number of applications (for a review see [@majumdar2007course]). Indeed, TW distributions have emerged in a variety of problems at the interface between statistical mechanics and mathematics, including the longest increasing subsequence of random permutations [@baik1999distribution], directed polymers [@baik1999distribution; @baik2000limiting; @baik2018pfaffian] and related growth models [@prahofer2000universal; @majumdar2004anisotropic; @imamura2004fluctuations], in the Kardar-Parisi-Zhang (KPZ) universality class in (1 + 1) dimensions as well as for the continuum (1+1)-dimensional KPZ equation [@sasamoto2010one; @calabrese2010free; @dotsenko2010bethe; @amir2011probability; @le2012kpz; @gueudre2012directed; @barraquand2020half], sequence alignment problems [@majumdar2005exact], height fluctuations of non-intersecting Brownian motions over a fixed time interval [@forrester2011non; @liechty2012nonintersecting; @nguyen2017non], height fluctuations of non-intersecting interfaces in the presence of a long-range interaction induced by a substrate [@nadal2009nonintersecting], or more recently in the context of trapped fermions [@dean2015finite; @dean2016noninteracting; @stephan2019free; @dean2019noninteracting], as well as in finance [@biroli2007top]. Remarkably, the TW distributions have been recently observed in experiments on nematic liquid crystals [@takeuchi2010universal; @takeuchi2011growing] (for the GOE and GUE) as well as in experiments involving coupled fibre lasers (for the GOE), and in dissipative self-assembled systems [@makey2020universality] (for the GUE). In the pioneering works on the largest eigenvalue in the classical ensembles of RMT [@forrester1993spectrum; @tracy1994level; @TracWido1996], the authors used the powerful tools of determinantal (for GUE) or Pfaffian (for GOE and GSE) point processes. This naturally led to the expression of these distributions in terms of a Fredholm determinant (for GUE) or a Fredholm Pfaffian (for GOE and GSE). Using rather involved “operator theoretic" techniques [@tracy1994level; @TracWido1996], it was further shown how to relate these Fredholm determinants and Pfaffians to sets of partial differential equations. In the limit of large matrix size $N$, this eventually led to a fairly explicit expression of these distributions for GOE, GUE and GSE in terms of a special solution of a Painlevé II equation (the so called Hastings-McLeod solution, see also below). More recently, an alternative derivation of the TW distribution for the GUE was proposed by Nadal and Majumdar in Ref. [@NadaMaju2011] using (semi-classical) orthogonal polynomials. The idea of the method is rather simple and also quite instructive since one sees how the Painlevé II equation emerges from the asymptotic analysis of the three-term recurrence relation satisfied by these orthogonal polynomials, which are some deformations of the standard Hermite polynomials, in the limit of large matrix size $N$. Furthermore, this approach was further extended in Ref. [@PerrSche2014] to compute the distribution of the first gap (between the first two eigenvalues), and more generally the statistics of near extreme eigenvalues in the GUE, which could be expressed in a rather compact form in terms of Painlevé transcendents, from which very precise asymptotics could be derived (see also [@witte2013joint] for yet another derivation of the statistics of the first gap in the GUE). It would thus be very useful to obtain such an alternative derivation of the TW distributions in the other classical ensembles, namely the GOE and the GSE. This would be particularly interesting in the case of GOE, since this would provide a very efficient method to compute the statistics of near-extreme eigenvalues for this ensemble, which is directly relevant to describe static [@monthus2013typical] and dynamical [@fyodorov2015large] properties of a well known mean-field spin-glass model, namely the spherical Sherrington-Kirkpatrick model. Up to now, the statistics of near-extreme eigenvalues in these ensembles have only been studied numerically [@perret2015density]. The goal of this paper is precisely to extend the method of Ref. [@NadaMaju2011] and provide an alternative derivation of the TW distributions in the GOE and the GSE, by developing an approach based on (semi-classical) skew-orthogonal polynomials. This is a first important step towards a precise and useful description of the statistics of near-extreme eigenvalues, e.g. the first gap between the two largest eigenvalues, in terms of Painlevé transcendents in these ensembles [@mays2020prep]. Summary of main results ======================= In the following we consider Gaussian random matrices ${\mathbf{M}}= [m_{ij}]$ belonging to the aforementioned classical ensembles of random matrices with real symmetric (GOE), complex Hermitian (GUE) or real quaternionic self-dual (GSE) entries respectively [@Mehta2004; @Forrester2010] (see also Appendix \[sec:classical\]), characterised by a Dyson index $\beta = 1, 2$ and $4$ respectively. In these three cases, the probability measure associated to the matrix ensemble is given by[^1] $$\begin{aligned} \label{e:MatPDFs} \Pr({\mathbf{M}}) \propto e^{-\beta ({\mathrm{Tr}\,}{\mathbf{M}}^2)/2} \;.\end{aligned}$$ In what follows we denote by G$\beta$E these ensembles with $\beta=1$ for the GOE, $\beta=2$ for the GUE and $\beta=4$ for the GSE. By performing a change of variables from the matrix entries $m_{ij}$ to the eigenvalues and eigenvectors of ${{\mathbf{M}}}$, one obtains the joint probability density function (JPDF) of the (real) eigenvalues $\lambda_1, \lambda_2, \cdots, \lambda_N$ in the G$\beta$E ensembles as (see [@Mehta2004; @Forrester2010]) $$\begin{aligned} \label{e:evJPDFs} {\mathcal{P}}_{\beta} (\lambda_1, \dots, \lambda_N) &=\frac{1} {Z_{\beta,N}}\; \prod_{j=1}^N e^{- \beta \lambda_j^2 /2 } \prod_{j<k} |\lambda_k -\lambda_j|^{\beta},\end{aligned}$$ where $Z_{\beta, N}$ is a normalization constant such that $$\begin{aligned} \int_{-\infty}^{\infty} d\lambda_1 \cdots \int_{-\infty}^{\infty} d\lambda_N {\mathcal{P}}_{\beta} (\lambda_1, \dots, \lambda_N) =1 \;\end{aligned}$$ and is given explicitly by $$\begin{aligned} \label{e:ZbetaN} Z_{\beta, N} = \beta^{-\frac{N}{2} - \frac{N \beta}{4} (N-1)} (2\pi)^{\frac{N}{2}} \prod_{j=0}^{N-1} \frac{\Gamma \left( 1+(j+1) \frac{\beta}{2} \right)} {\Gamma \left( 1+\frac{\beta}{2} \right)} \;,\end{aligned}$$ where $\Gamma(z)$ is the gamma function. We will compute the cumulative distribution function (CDF) of the largest eigenvalue, i.e. $F_{\beta, N} (y) \equiv \Pr (\lambda_{\max}^{(\beta)} <y)$, or equivalently, the probability that all eigenvalues are less than some upper bound $y$ $$\begin{aligned} \label{d:CDF} F_{\beta, N} (y) \equiv \Pr (\lambda_{\max}^{(\beta)} <y) = N! \int_{-\infty}^y d\lambda_{1} \int_{\lambda_1}^y d \lambda_2 \cdots \int_{\lambda_{N-1}}^y d\lambda_N \mathcal{P}_{\beta} (\lambda_{1}, \dots,\lambda_N) \;,\end{aligned}$$ where the factorial $N!$ comes from the fact that in Eq. (\[d:CDF\]), the eigenvalues are ordered, i.e. $\lambda_1 < \lambda_2 < \dots < \lambda_N \leq y$. (Note that this ordering is not required here, however it will be convenient later to work with ordered eigenvalues and therefore we impose the ordering from the beginning.) It is well known that the JPDF in Eq. (\[e:evJPDFs\]) can be interpreted as the Boltzmann weight of a one-dimensional gas of $N$ charged particles where $\lambda_i$ denotes the position of the $i$-th particle and $\beta$ the inverse temperature [@dyson1962statistical]. These particles interact via a repulsive logarithmic interaction while they are subjected to an external quadratic potential: this is the so-called log-gas. Hence the CDF $F_{\beta, N} (y)$ in Eq. (\[d:CDF\]) is the partition function of this log-gas in the presence of a hard wall at position $y$ [@majumdar2014top] — such partition functions are called “restricted partition functions” in the following. To compute $F_{\beta, N} (y)$, it is useful to introduce sets of *orthogonal* and *skew-orthogonal* polynomials. Specifically, we define the $y$-dependent inner (or scalar) product for $\beta=2$ $$\begin{aligned} \label{d:IP2} (f,g)_{2}^y = \int_{-\infty}^y e^{-\lambda^2} f(\lambda) g(\lambda) d\lambda \;,\end{aligned}$$ and the skew-inner products for $\beta=4$ $$\begin{aligned} \nonumber \langle f , g \rangle_{4}^y &= \frac{1}{2} \int_{- \infty}^{y} dx \; e^{-2 x^2} \left[ f(x) g'(x)- g(x) f'(x) \right]\\ \label{d:IP4} &= \frac{1}{2} \int_{- \infty}^{y} dx \; e^{-x^2} \left[ f(x) \frac{d} {dx} \left( e^{-x^2} g(x)\right) - g(x) \frac{d} {dx} \left( e^{-x^2} f(x) \right) \right],\end{aligned}$$ and for $\beta=1$ $$\begin{aligned} \nonumber \langle f , g \rangle_{1}^y &= \frac1{2} \int_{-\infty}^{y} dx \; e^{-x^2/2} f(x) \int_{-\infty}^y d z \; e^{-z^2/2} g (z) {\mathrm{sgn}}(z- x)\\ \label{d:IP1} &= \frac1{2} \int_{-\infty}^{y} dx \; e^{-x^2/2} f(x) \int_{x}^{y} d z \; e^{-z^2/2} g (z) - \frac1{2} \int_{-\infty}^{y} dx \; e^{-x^2/2} f(x) \int_{-\infty}^{x} d z \; e^{-z^2/2} g (z).\end{aligned}$$ Then for each ensemble, we seek a set of (monic) polynomials $\{ p_j (x,y) \}$ for $\beta = 2$, $\{Q_j (x, y) \}$ for $\beta = 4$, and $\{ R_j (x, y) \}$ for $\beta = 1$ (by increasing order of complexity, as we will see) with the orthogonality/skew-orthogonality properties $$\begin{aligned} \label{d:2orthog} (p_j, p_k)_2^y = h_j(y) \delta_{j,k},\end{aligned}$$ $$\begin{aligned} \nonumber \langle Q_{2j} , Q_{2k} \rangle_{4}^y&= \langle Q_{2j+1} , Q_{2k+1} \rangle_{4}^y= 0 \;, \\ \label{d:4sorthog} \langle Q_{2j} , Q_{2k+1} \rangle_{4}^y&= -\langle Q_{2k+1} , Q_{2j} \rangle_{4}^y = q_j(y) \delta_{j,k} \;,\end{aligned}$$ and $$\begin{aligned} \nonumber \langle R_{2j} , R_{2k} \rangle_{1}^y&= \langle R_{2j+1} , R_{2k+1} \rangle_{1}^y= 0 \;, \\ \label{d:1sorthog} \langle R_{2j} , R_{2k+1} \rangle_{1}^y&=-\langle R_{2k+1} , R_{2j} \rangle_{1}^y = r_j (y) \delta_{j,k} \;,\end{aligned}$$ where, to be explicit, the respective normalizations are $$\begin{aligned} \label{e:ynorm2} h_j(y) &:= (p_j, p_j)_2^y\\ \label{e:ynorm4} q_j(y) &:=\langle Q_{2j} , Q_{2j+1} \rangle_{4}^y\\ \label{e:ynorm1} r_j(y) &:= \langle R_{2j} , R_{2j+1} \rangle_{1}^y \;.\end{aligned}$$ (Note that the orthogonal and skew-orthogonal polynomials depend on the parameter $y$, although we will often suppress the explicit notation of that dependence for brevity.) In fact, as for the case of the GUE [@NadaMaju2011], the CDF $F_{\beta,N}(y)$ can be expressed only in terms of the norms $h_j(y)$, $q_j(y)$ and $r_j(y)$ for $\beta = 2, 4$ and $1$ respectively. For $\beta=2$, it was indeed shown in [@NadaMaju2011] that $$\begin{aligned} \label{e:F2N}F_{2, N} (y) &= \prod_{j=0}^{N-1} \frac{h_j(y)} {h_j (\infty)} = \frac{2^{N(N-1)/2}}{\pi^{N/2}} \prod_{j=0}^{N-1} \frac{h_j(y)} {j!} \;.\end{aligned}$$ In the present paper we show that for $\beta =1$ (and where $N$ is restricted to be even for simplicity), with the polynomials $R_j$ from , we have $$\begin{aligned} \label{e:F1N} F_{1,N} (y) = \prod_{j=0}^{N/2 -1} \frac{r_j (y)} {r_j (\infty)} = \frac{2^{\frac{N}{2} \left( \frac{N}{2} -1 \right)}} {\pi^{N/4}} \prod_{j=0}^{N/2 -1} \frac{1} {(2j)!} {\mathrm{Pf}\,}{\mathbf{V}}_{N-1},\end{aligned}$$ where the matrix ${\mathbf{V}}_m$, whose explicit expression is given in below, contains the $\beta=2$ polynomials $p_{j} (x,y)$ and their normalizations $h_j(y)$. For $\beta=4$ we require a slightly modified (by a simple rescaling) skew-inner product with associated modified skew-orthogonal polynomials $\tilde{Q}_{j}$ and normalizations $\tilde{q}_j$ \[see Eqs. (\[e:Rescaleqj\]) and (\[e:RescaleQj\])\], which gives us $$\begin{aligned} \label{e:F4N} F_{4, N} (y) = \prod_{j=0}^{N-1} 2^{ -2j -\frac{1}{2}} \frac{\tilde{q}_j (\sqrt{2} y)} {q_j (\infty)} = \frac{2^{N^2}}{\pi^{N/2}} \prod_{j=0}^{N-1} \frac{1}{(2j +1)!} {\mathrm{Pf}\,}{\mathbf{W}}_{2N-1}\Big|_{y \mapsto \sqrt{2} y},\end{aligned}$$ where the matrix ${\mathbf{W}}_m$ is given in , and again contains the $\beta=2$ polynomials $p_{j} (x,y)$ and their normalizations $h_j(y)$. In Fig. \[f:CDFs\] we present a comparison between a numerical evaluation of these formulae and and a direct numerical computation of these CDFs by sampling GOE and GSE random matrices, showing very good agreement. We emphasize that the expressions on the right hand side of Eqs. and depend only on the $\beta=2$ orthogonal polynomials, and do not depend on the skew-orthogonal polynomials at all. In the case of the GUE ($\beta = 2$), the orthogonal polynomials $p_k$ for the inner product in have already been studied, first in [@NadaMaju2011] and later in [@PerrSche2014]. Here we call these polynomials the *Nadal–Majumdar (NM) polynomials*. Interestingly, these NM polynomials naturally arise also in the study of the so called [*level curvature distribution*]{} at the soft edge of random Hermitian matrices [@fyodorov2011level]. Although they do not have a known closed formula, they satisfy the three-term recurrence relation \[since they are orthogonal with respect to the inner product in (\[d:IP2\])\] $$\begin{aligned} \label{e:NM1} \lambda p_k (\lambda, y) &= p_{k+1} (\lambda, y) + {\check{\mathsf{S}}}_k (y) p_k (\lambda, y) + {\check{\mathsf{R}}}_k (y) p_{k-1} (\lambda, y)\\ \label{e:NM2} {\check{\mathsf{R}}}_{k}(y) &= \frac{h_k (y)} {h_{k-1}(y)}\\ \label{e:NM3} {\check{\mathsf{S}}}_k(y) & \neq 0,\end{aligned}$$ where the last expression follows because the domain of integration in the inner product is not symmetric. (Note that we have used the “check” and sans serif font to distinguish ${\check{\mathsf{R}}}_k$ from the $\beta=1$ polynomials $R_j$. We use the same style for ${\check{\mathsf{S}}}_k$ for consistency.) In the limit $y\to \infty$ the NM polynomials become the (monic, “physicist’s”) Hermite polynomials, i.e. [@NadaMaju2011] $$\begin{aligned} \label{e:limHerms} p_j(\lambda, y) = \frac{1}{2^j} H_j (\lambda) + O \left( e^{-y^2} \right),\end{aligned}$$ where the Hermite polynomials of index $j$, $H_j(x)$, are orthogonal with respect to the weight function $e^{-x^2}$, and the division by $2^j$ is here to ensure monicity. In fact, in this limit the inner product (\[d:IP2\]) and skew-inner products – all reduce to their classical Gaussian counterparts, with norms [@Mehta2004; @Forrester2010] $$\begin{aligned} \label{e:InfinityNorms} h_j (\infty)= \frac{\pi^{1/2}} {2^j} \Gamma (j+1), \quad q_j(\infty) = \frac{\pi^{1/2}} {2^{4j+\frac{3}{2}}} \Gamma (2j+2), \quad r_j(\infty) = \frac{\pi^{1/2}} {2^{2j}} \Gamma (2j+1) \;.\end{aligned}$$ The corresponding classical skew-orthogonal polynomials are known, and recalled in Appendices \[a:limpolys4\] and \[a:limpolys1\]. However, for finite $y$, there are no known statements analogous to – for $\beta=1$ and $4$ polynomials. Yet, as a first approach, we can iteratively use the skew-inner products and to construct these polynomials. An important property is that these polynomials are not unique, since skew-inner products are invariant under the polynomial transformation $$\begin{aligned} \label{e:oddsymm} \eta_{2j+1} \mapsto \eta_{2j+1} + c \; \eta_{2j}\end{aligned}$$ where $c$ is any constant (and $\eta_k = Q_k$ or $R_k$), and therefore a set of skew-orthogonal polynomials is unique only up to this symmetry in the odd degree polynomials. By specifying the constant we employ this iterative process to construct the skew-orthogonal polynomials defined in Eqs. (\[d:4sorthog\]) and (\[d:1sorthog\]) in Appendix \[s:iterative\]. However this method is not convenient for the asymptotic analysis of the quantities in . Instead, in [@NadaMaju2011; @PerrSche2014], it was shown that the recurrence relations – can be exploited to obtain the asymptotic behaviors of the norms $h_j(y)$ and the polynomials $p_j (\lambda,y)$ themselves in the limit of large $N$ and large $y$. Here, we extend the approach developed in [@AdlevanMoer2002; @AdleForrNagavanMoer2000] to obtain explicit expressions for the sets of semi-classical skew-orthogonal polynomials $\{ Q_j\}$ and $\{ R_j\}$ in the basis of the orthogonal polynomials $\{ p_j\}$ (the NM polynomials). This is the content of Proposition \[p:alpha4\] (for the GSE) and Proposition \[p:alpha1\] (for the GOE). Interestingly, the proofs of these results are achieved by using results on [*overlapping Pfaffians*]{}, studied by Knuth [@Knuth1996]. This is the first main technical contribution of this work. As a byproduct of our analysis, we also recover the classical skew-orthogonal polynomials as the $y\to \infty$ limit of our results here (see Appendices \[a:beta4\] and \[a:beta1\]). We will then use this explicit construction, together with the asymptotic analysis of the polynomials $p_j(x, y)$, to compute the large $N$ asymptotic limit of $F_{1,N}$ and $F_{4,N}$. Indeed for the case of the GSE, we show that[^2] $$\begin{aligned} \label{e:TW4_intro} \lim_{N \to \infty} F_{4,N} \left( y = \sqrt{2N} + \frac{s}{2^{7/6} N^{1/6}} \right) = \exp{\left(-\frac{1}{2} \int_s^\infty (x-s) q^2(x) dx \right)} \cosh{\left(\frac{1}{2} \int_s^\infty q(x)dx \right)},\end{aligned}$$ where $q(x)$ is the Hastings-McLeod solution of the Painlevé II equation, i.e. $$\begin{aligned} \label{e:PII_intro} q''(x) = x q(x) + 2 q^3(x) \;, {\rm with} \;\; q(x) \underset{x \to \infty}{\sim} {\rm Ai}(x) \;,\end{aligned}$$ and ${\rm Ai}(x)$ is the standard Airy function. On the other hand, for the case of the GOE, we show that $$\begin{aligned} \label{e:TW1_intro} \lim_{N \to \infty} F_{1, N} \left( y = \sqrt{2N} + \frac{s}{\sqrt{2} N^{1/6}} \right) &= \exp\left( -\frac{1}{2} \int_s^{\infty} (x-s) q(x)^2 dx \right) \exp\left( - \frac{1}{2} \int_s^{\infty} q(x) dx \right),\end{aligned}$$ with, again, $q(x)$ given in Eq. (\[e:PII\_intro\]). We thus recover the known expressions of the TW distributions for GSE and GOE [@TracWido1996], by using here a completely different method. This is the second main achievement of the present paper. The key result used to obtain the TW distributions is an identity proved in Proposition \[lem:PfW\] \[see Eq. (\[e:expansion:Pf\])\] that allows us to obtain explicit expressions of the Pfaffians entering the expressions in Eqs. (\[e:F1N\]) and (\[e:F4N\]), which are then conveniently amenable to an asymptotic analysis in the limit of large $N$. The proof of this identity (\[e:expansion:Pf\]) relies on the expression of a Pfaffian as a sum over perfect matchings recalled in of the Appendices — this representation is used extensively throughout the present paper. The paper is organized as follows. In Section \[sec:PartFns\] we use the polynomials $Q_j$ and $R_j$, defined in and respectively, to find Pfaffian expressions for restricted partition functions such as the CDFs $F_{1,N}(y)$ and $F_{4,N}(y)$ using standard techniques. In Section \[sec:SOPs\] we construct explicitly these skew-orthogonal polynomials in terms of the NM polynomials $p_j (x,y)$ and their normalizations $h_j (y)$, finding in particular Pfaffian expressions for the coefficients and the normalizations $q_j(y)$ and $r_j(y)$. In Sections \[s:F4asympt\] and \[s:F1asympt\] we present the asymptotic analysis of $F_{4,N}(y)$ and $F_{1,N}(y)$ respectively, leading to the expressions given in Eqs. (\[e:TW4\_intro\]) and (\[e:TW1\_intro\]). Finally, Section \[s:conclusion\] contains our conclusions and perspectives. Several technical details about the results presented in this paper have been left to the Appendices. [![The histograms correspond to a numerical evaluation of the CDF of the largest eigenvalue sampled from $50,000$ matrices in the $\beta=1$ ensemble for $N=8$ (left panel) and in the $\beta=4$ ensemble for $N=4$ (right panel). The solid red line represents the exact result given, in the left panel, by Eq. (\[e:F1N\]) and, in the right panel, by Eq. (\[e:F4N\]).[]{data-label="f:CDFs"}](F18_cont.pdf "fig:"){width="40.00000%"}]{} Restricted partition functions and generalizations {#sec:PartFns} ================================================== In this section, we show how to compute restricted partition functions such as the CDFs $F_{\beta,N}(y)$ in Eq. (\[d:CDF\]). We actually consider slightly more general quantities defined as the following averages over the eigenvalue JPDFs for the GOE ($\beta=1$), GUE ($\beta=2$) and the GSE ($\beta =4$): $$\begin{aligned} \nonumber \hat{Z}_{\beta, N} [a, y] &= \left\langle \prod_{j=1}^N a(\lambda_j) \right\rangle_{\mathcal{P}_{\beta}}^y\\ \label{e:GenPF}&= \frac{1} {Z_{\beta,N}} \int_{-\infty}^y d\lambda_1 \cdots \int_{-\infty}^y d\lambda_N \prod_{j=1}^N a(\lambda_j) e^{- \beta \lambda_j /2 } \prod_{j<k} |\lambda_k -\lambda_j|^{\beta}.\end{aligned}$$ Each of the $\hat{Z}_{\beta, N} [a, y]$ will be put into determinant/Pfaffian form — the construction of the associated matrices will depend on its own set of monic polynomials. While these polynomials are in principle arbitrary, it is convenient to specify them to be the respective orthogonal/skew-orthogonal polynomials. If we think of the integral in as an average over a truncated version of the density , i.e. $$\begin{aligned} \label{eq:janossy1} {\mathcal{P}}_{\beta} (\lambda_1, \dots, \lambda_N; y) := {\mathcal{P}}_{\beta} (\lambda_1, \dots, \lambda_N) \chi_{(-\infty, y)} (\lambda_1, \dots, \lambda_N) \;,\end{aligned}$$ where $\chi_{A} ({{\mathbf x}}) =1$ if ${{\mathbf x}}\in A^N$ and zero otherwise, then we are in the realm of Janossy densities [@Janossy1950] (see [@Soshnikov2004] for a clear introduction to the topic and references). In [@BoroSosh2003] the authors discussed “determinantal” Janossy densities (where the particle JPDF and $n$-point correlation functions can be written in terms of a determinant) and found the matrix kernel for the determinant. In [@Soshnikov2003] these results were extended to “Pfaffian” Janossy densities, that is, the author found the matrix kernel for Janossy JPDFs and $n$-point correlation functions that are expressed as Pfaffians. Our eigenvalue JPDFs have this determinantal ($\beta=2$) or Pfaffian ($\beta=1,4$) structure, and so the $n$-point correlations will also have determinantal/Pfaffian structure. We will explicitly construct these correlation functions in a future work, and use them to calculate gap probabilities and the density of states near the largest eigenvalue [@mays2020prep]. Here, however, we restrict ourselves to the calculation of the averages , which gives us the CDF of the largest eigenvalue via $$\begin{aligned} \label{e:FZ} F_{\beta, N} (y) = \hat{Z}_{\beta, N} [1,y] \;.\end{aligned}$$ Below we treat the case $\beta = 2$, $\beta = 4$ and $\beta = 1$, again by increasing order of complexity. $\beta=2$ --------- Although this is not needed for the $\beta=1,4$ cases, for completeness we also include the $\beta=2$ result, which can be obtained using the Vandermonde determinant identity (the procedure is a slight modification to that in [@Forrester2010 §5.2.1]) $$\begin{aligned} \label{e:GenPF2} \hat{Z}_{2,N} [a, y] = \frac{N!}{Z_{2 ,N}} \det \left[ \gamma_{j,k}^{(2)} [a, y] \right]_{j,k= 0, \dots, N-1},\end{aligned}$$ where $Z_{2, N}$ is given in and $$\begin{aligned} \label{e:GenPF2b} \gamma_{j,k}^{(2)} [a, y] := \int_{-\infty}^y a(\lambda) e^{-x^2} p_j (\lambda, y) p_k (\lambda, y) d\lambda \;.\end{aligned}$$ The polynomials $p_j$ in are the NM polynomials, i.e. the monic polynomials of degree $j$ that are orthogonal with respect to the inner product . A consequence of this (in the limit $y\to \infty$) is the known result [@Mehta2004; @Forrester2010] $$\begin{aligned} \label{e:Z2norms} Z_{2,N} = N! \prod_{j=0}^{N-1} h_j (\infty),\end{aligned}$$ where $h_j (\infty)$ is given in , which agrees with . With $a(x)=1$ the integral $\gamma_{j,k}^{(2)}$ becomes the inner product , so with the orthogonal polynomials $p_j$ we use to obtain the known result . $\beta=4$ --------- The average for $\beta=4$ is $$\begin{aligned} \label{e:GenPF4} \hat{Z}_{4,N} [a, y] = \frac{N! 2^N}{Z_{4, N}}\; {\mathrm{Pf}\,}\left[ \gamma^{(4)}_{j,k}[a, y] \right]_{j,k= 0, \dots, 2N-1},\end{aligned}$$ where $$\begin{aligned} \gamma^{(4)}_{j,k} [a ,y] := \frac{1}{2} \int_{- \infty}^{y} d\lambda \; a(\lambda) e^{-\lambda^2} \left[ Q_j(\lambda, y) \frac{d} {d\lambda} \left( e^{-\lambda^2} Q_k(\lambda, y)\right) - Q_k(\lambda, y) \frac{d} {d\lambda} \left( e^{-\lambda^2} Q_j (\lambda, y) \right) \right]\end{aligned}$$ and the $Q_j$ are monic polynomials of degree $j$ that are skew-orthogonal with respect to the skew-inner product . Using the theory of Section \[sec:SOPs\] below, we can make a quick check of by noting that when $a(x) = 1$ the matrix in is of the form , and so from the Pfaffian is given by the product $q_0 (y) q_1 (y) \cdots q_{N-1} (y)$. Then, in the limit $y\to \infty$, we recover the result analogous to [@Mehta2004; @Forrester2010] $$\begin{aligned} \hat{Z}_{4,N}[1,y]\Big|_{y\to \infty} =1 {\qquad \Rightarrow \qquad}Z_{4, N} = N! 2^N \prod_{j=0}^{N-1} q_j (\infty),\end{aligned}$$ where $q_j (\infty)$ is given in , and this agrees with . [*Proof*:]{}This result is obtained using the same techniques as applied in [@TracWido1998; @Mehta2004; @Forrester2010], but with a truncated domain of integration, and a correspondingly different set of polynomials. To keep this paper self-contained, we will go through the details. We start with the identity [@Mehta2004] $$\begin{aligned} \label{e:beta4Vdm} \prod_{1\leq j< k\leq N} (\lambda_j - \lambda_k)^4 = \det \left[ \begin{array}{c} \lambda_j^{k-1}\\ (k-1) \lambda_j^{k-2} \end{array}\right]_{j=1, \dots, N \atop k= 1, \dots, 2N},\end{aligned}$$ and note that each even row is the derivative of the odd row immediately above it. Then in this matrix, for each column, by adding linear combinations of the columns to the left of that column (starting from the left-most column) we can create arbitrary monic polynomials, while preserving the derivative relationship between the even and odd rows. So for our purpose, we choose the polynomials to be the $Q_j$, which are skew-orthogonal with respect to the skew-inner product , giving $$\begin{aligned} \nonumber \hat{Z}_{4, N} [a, y] &= \frac{1}{Z_{4, N}} \int_{-\infty}^y d\lambda_1 \cdots \int_{-\infty}^y d\lambda_N \prod_{j=1}^N a(\lambda_j) e^{- 2 \lambda_j } \det \left[ \begin{array}{cc} Q_{2k-2} (\lambda_j) & Q_{2k-1} (\lambda_j)\\ Q_{2k-2}' (\lambda_j) & Q_{2k-1}' (\lambda_j) \end{array}\right]_{j,k= 1, \dots, N}\\ \label{e:beta4Vdm2}&= \frac{1}{Z_{4, N}} \sum_{P\in S_{2N}} \varepsilon(P) \prod_{j=1}^N \int_{-\infty}^y d\lambda\; a(\lambda) e^{-2\lambda^2} Q_{P(2j-1) -1} (\lambda) Q_{P(2j) -1}' (\lambda),\end{aligned}$$ where the second line follows from Laplace expansion of the determinant, and we apply the integrals to each matched pair of $Q$ and $Q'$. (Note that we suppress the dependence on $y$ for brevity.) For each pair of indices on the $Q$ and $Q'$ in , we then match up each permutation with the corresponding permutation where that index pair is interchanged, hence picking up a $(-1)$, giving $$\begin{aligned} \nonumber &\hat{Z}_{4, N} [a, y] =\frac{1}{Z_{4, N}} \times\\ &\sum_{P\in S_{2N} \atop P(2j)> P(2j-1)} \varepsilon(P) \prod_{j=1}^N \int_{-\infty}^y d\lambda\; a(\lambda) e^{-2\lambda^2} \Big( Q_{P(2j-1) -1} (\lambda) Q_{P(2j) -1}' (\lambda) - Q_{P(2j) -1} (\lambda) Q_{P(2j-1) -1}' (\lambda) \Big),\end{aligned}$$ where we need to restrict the sum to just those permutations obeying the rule $P(2j)> P(2j-1)$ for all $j$. Introducing a factor of $\frac{1}{2}$ for each integral (incurring a pre-factor of $2^N$), then using the definition of the Pfaffian recalled in we obtain $$\begin{aligned} \hat{Z}_{4, N} [a, y] =\frac{2^N N!}{Z_{4, N}} {\mathrm{Pf}\,}\left[ \frac{1}{2} \int_{-\infty}^y d\lambda\; a(\lambda) e^{-2\lambda^2} \Big( Q_{j} (\lambda) Q_{k}' (\lambda) - Q_{k} (\lambda) Q_{j}' (\lambda) \Big) \right]_{j,k=0, \dots, 2N-1}.\end{aligned}$$ The equality between the first and second lines in gives the result in . $\Box$ While , with $a(x)=1$, gives us the CDF $F_{4,N}$, we will need the explicit forms of the polynomials $\{ Q_j \}$ before we can obtain the expression in . This will be achieved below in Section \[sec:SOPs\]. $\beta=1$, with $N$ even ------------------------ Recall that we have restricted $N$ to be even in this work. The parity of $N$ plays an important role since for the $\beta=1$ case we have the difficulty of the absolute value of the Vandermonde determinant in . To deal with it, we apply the method of integration over alternate variables, which was introduced by de Bruijn [@deBruijn1955] and applied to integrals similar to by Mehta [@Mehta2004]. However, this method is dependent on the parity of $N$, which can be seen when one pairs up the rows in below — when $N$ is odd there would be one unpaired row, which needs to be specially dealt with. For simplicity, we will only work with the $N$ even case here, and the techniques for dealing with the $N$ odd case are contained in [@Mehta2004; @Forrester2010; @ForrMays2009; @Mays2011thesis]. With $N$ even the average for $\beta=1$ is $$\begin{aligned} \label{e:GenPF1} \hat{Z}_{1,N} [a ,y] = \frac{2^{N/2} N!}{Z_{1,N}} {\mathrm{Pf}\,}\left[ \gamma^{(1)}_{j,k}[a ,y] \right]_{j,k= 0, \dots, N-1},\end{aligned}$$ where $$\begin{aligned} \label{d:gamma1} \gamma^{(1)}_{j,k} [a ,y] := \frac1{2} \int_{-\infty}^{y} dx \; a(x) e^{-x^2/2} R_j (x, y) \int_{-\infty}^{y} d z \; a(z) e^{-z^2/2} R_k (z, y) \, {\mathrm{sgn}}(z-x)\end{aligned}$$ and the $R_j$ are monic polynomials of degree $j$ that are skew-orthogonal with respect to the skew-inner product . As with $\beta=4$ above, we can recover the known result [@Mehta2004; @Forrester2010] with $a(x)=1, y\to \infty$ $$\begin{aligned} \hat{Z}_{1,N} [1,y]\Big|_{y\to \infty} = 1 {\qquad \Rightarrow \qquad}Z_{1, N} = N! 2^{N/2} \prod_{j=0}^{N/2 -1} r_j (\infty),\end{aligned}$$ where $r_j (\infty)$ is given in . [*Proof*:]{}As in the case of $\beta=4$ above, the techniques used here are found in [@TracWido1998; @Mehta2004; @Forrester2010] but we will delve into some of the details using the truncated integral for completeness. We start by ordering the eigenvalues $-\infty < \lambda_1 <\cdot\cdot\cdot < \lambda_N < y$ (incurring a factor of $N!$) in so that we can remove the absolute value from the product of differences. Then we use the Vandermonde determinant expression (suppressing the polynomial dependence on $y$) [$$\begin{aligned} \nonumber &\hat{Z}_{1,N} [a ,y]= \frac{N!}{Z_{1, N}} \int_{-\infty}^{y}d\lambda_N \int_{-\infty}^{\lambda_N}d\lambda_{N-1} \cdot\cdot\cdot \int_{-\infty}^{\lambda_2} d\lambda_1 \prod_{j=1}^N e^{-\lambda_j^2/2}\; a(\lambda_j)\prod_{1\leq j < k \leq N}(\lambda_k -\lambda_j)\\ \nonumber &=\frac{N!}{Z_{1, N}} \int_{-\infty}^{y}d\lambda_N \int_{-\infty}^{\lambda_N} d\lambda_{N-1} \cdot\cdot\cdot \int_{-\infty}^{\lambda_2} d\lambda_1 \det \left[ e^{-\lambda_j^2/2} a(\lambda_j) \lambda_j^{k-1} \right]_{j,k=1,...,N}\\ &=\frac{N!}{Z_{1, N}} \int_{-\infty}^{y} d\lambda_N \int_{-\infty}^{\lambda_N} d\lambda_{N-1} \cdot\cdot\cdot \int_{-\infty}^{\lambda_2} d\lambda_1 \det \left[ e^{-\lambda_j^2/2} a(\lambda_j) R_{k-1}(\lambda_j) \right]_{j,k=1,...,N},\end{aligned}$$]{}where the third equality follows from elementary column operations. This is the same procedure that was applied to in the $\beta=4$ case above, and it allows us to obtain any set of monic polynomials in the columns; for our purposes we specify the polynomials to be the $\{R_j\}$, which are skew-orthogonal with respect to the skew-inner product . Now we wish to apply the method of integration over alternate variables (mentioned above), and to prepare for that we change the order of the integrals, with even integrals on the left and odd integrals on the right [$$\begin{aligned} \nonumber &\hat{Z}_{1,N} [a ,y]=\frac{N!}{Z_{1, N}}\\ &\times \int_{-\infty}^{y}d\lambda_N \int_{-\infty}^{\lambda_N} d\lambda_{N-2} \cdot\cdot\cdot \int_{-\infty}^{\lambda_4} d\lambda_2 \int_{\lambda_{N-2}}^{\lambda_N} d\lambda_{N-1} \cdots \int_{\lambda_{2}}^{\lambda_{4}} d\lambda_{3} \int_{-\infty}^{\lambda_2} d\lambda_1 \det \left[ e^{-\lambda_j^2/2} a(\lambda_j) R_{k-1} (\lambda_j) \right]_{j,k=1,...,N}.\end{aligned}$$]{}The purpose of this manipulation is that now in each odd integral (i.e. over the variables $\lambda_{2n-1}$) the only dependence of the corresponding variable is in the $(2n-1)$st row of the determinant, so the odd integrals can be applied to their respective rows: [$$\begin{aligned} \label{e:integAVs} \hat{Z}_{1,N} [a ,y]&=\frac{N!}{Z_{1, N}} \int_{-\infty}^{y} d\lambda_N \int_{-\infty}^{\lambda_N}d\lambda_{N-2} \cdot\cdot\cdot \int_{-\infty}^{\lambda_4} d\lambda_2 \det \left[ \begin{array}{c} \int_{-\infty }^{\lambda_{2j}} e^{-\lambda^2/2} a(\lambda) R_{k-1}(\lambda) d\lambda\\ e^{-\lambda_{2j}^2/2} a(\lambda_{2j}) R_{k-1}(\lambda_{2j}) \end{array}\right]_{j=1,...,N/2 \atop k=1,...,N},\end{aligned}$$ ]{}where we have added the first row to the third row, and the first and third rows to the fifth row, and so on, so all the integrals have lower terminal $-\infty$. (This sequence of steps is the *method of integration over alternate variables*.) We see that the determinant in is now symmetric in the variables $\lambda_2, \lambda_4,..., \lambda_N$, and so we can remove the ordering $\lambda_2 < \lambda_4 < ... < \lambda_N$ at the cost of dividing by $(N/2)!$. Taking the Laplace expansion of the determinant we find $$\begin{aligned} \hat{Z}_{1,N} [a ,y]= \frac{1}{Z_{1,N}} \frac{N!} {(N/2)!} \sum_{P\in S_N} \varepsilon(P) \prod_{j=1}^{N/2} \mu_{P(2j -1), P(2j)},\end{aligned}$$ where $$\begin{aligned} \label{def:mu_GOE} \mu_{j,k}:=\int_{-\infty}^{y} dx\: e^{-x^2/2}\: a(x) \:R_{k-1}(x) \int_{-\infty}^x dz\: e^{-z^2/2}\: a(z)\: R_{j-1}(z),\end{aligned}$$ and $\varepsilon(P)$ is the sign of the permutation $P$. By defining $$\begin{aligned} \gamma_{j,k}^{(1)} :=\frac{1}{2}(\mu_{j,k} -\mu_{k,j}),\end{aligned}$$ incurring a factor of $2^{N/2}$, then we can restrict the sum to terms with $P(2j)> P(2j -1)$ for all $j$, giving $$\begin{aligned} \hat{Z}_{1,N} [a ,y]=\frac{1}{Z_{1,N}} 2^{N/2} \frac{N!} {(N/2)!} \sum_{P\in S_N \atop P(2j) > P(2j-1)} \varepsilon(P) \prod_{j=1}^{N/2} \gamma_{P(2j-1), P(2j)}^{(1)}.\end{aligned}$$ Now using we have the result in – \[where we cancel the factor of $(N/2)!$ to account for summing over distinct terms only\]. $\Box$ As for $\beta=4$ above, we will need to first find the skew-orthogonal polynomials $\{ R_j \}$ before we can use to obtain the expression for the CDF $F_{1,N}$ in . This is precisely the aim of the next section. Explicit construction of the skew-orthogonal polynomials {#sec:SOPs} ======================================================== The averages and in Section \[sec:PartFns\] above contain integrals over the respective skew-orthogonal polynomials $Q_j$ and $R_j$. The major advantage of using these polynomials can be seen if we first consider the case of $\beta=2$, from the expression : we see that when $a(x)= 1$ the matrix in the determinant becomes $\left[ (p_j, p_k)_2^y \right]_{j,k=0, \dots, N-1}$, and so the determinant will be simply calculated if the polynomials $p_j$ are orthogonal with respect to the inner product since the resulting matrix is diagonal. Indeed, this was the approach taken in [@NadaMaju2011; @PerrSche2014], where the orthogonal polynomials are the NM polynomials, which obey the relations –. We will use the same approach for the $\beta=4$ and $\beta=1$ cases; that is we will construct the polynomials $Q_j$ and $R_j$ such that the matrices in and are of *skew-diagonal form*[^3] $$\begin{aligned} \label{d:sdiag} {\mathbf{S}}= \begin{bmatrix} 0&s_1& 0 &0 & \cdots& 0& 0\\ -s_1& 0& 0& 0 & \cdots& 0& 0\\ 0& 0& 0& s_2& \cdots& 0& 0\\ 0& 0& -s_2 & 0 & & \vdots& \vdots\\ \vdots& \vdots& \vdots& &\ddots\\ 0& 0& 0& \cdots &&0& s_N\\ 0& 0& 0& \cdots &&-s_N& 0 \end{bmatrix}.\end{aligned}$$ The only non-zero elements of ${\mathbf{S}}$ are in $2\times 2$ blocks $\begin{bmatrix} 0& s_j\\ -s_j &0 \end{bmatrix}$ on the diagonal, and we then have the simple result $$\begin{aligned} \label{e:Pfsdiag} {\mathrm{Pf}\,}{\mathbf{S}}= \prod_{j=1}^N s_j.\end{aligned}$$ In other words, we are looking for two sets of monic polynomials $\{Q_j\}, \{ R_j\}$ that satisfy the conditions in and respectively. Such polynomials are called *skew-orthogonal polynomials*. Recall that these polynomials are only unique up to the transformation , where $\eta_j= Q_j$ ($\beta=4$) and $\eta_j=R_j$ ($\beta=1$). As discussed in Introduction we can, in principle, construct the polynomials iteratively using the conditions and , but this technique does not yield expressions that are amenable to asymptotic analysis. Nor is there a known closed form or recursive expression for these polynomials. Rather, we apply the method of [@AdlevanMoer2002; @AdleForrNagavanMoer2000] where we aim to express our skew-orthogonal polynomials in the basis of the NM polynomials $\{ p_j \}$ from the analogous $\beta=2$ problem $$\begin{aligned} \label{e:SOPvsOP} \eta_j= p_j + \alpha_{j, j-1} p_{j-1} + \dots \alpha_{j, 0} p_0, \qquad \mbox{($\eta_j= Q_j$ for $\beta=4$, $\eta_j=R_j$ for $\beta=1$)}.\end{aligned}$$ If we can find the coefficients $\alpha_{j,k}$ in then we can use the properties of the polynomials $\{ p_j \}$ to obtain asymptotic results for the problems that we consider here. Note that implies that we have freedom in the choice of the $\alpha_{2j+1, 2j}$ (that is, the second term in the odd-degree polynomials) in equation , and we will typically choose $\alpha_{2j+1, 2j} =0$. \[We will see that this choice is quite natural once we have the general formula for the coefficients, see .\] We find that the coefficients and the polynomial normalizations $q_j(y)$ and $r_j(y)$ are given as ratios of Pfaffians. To contrast this with the classical Gaussian case (in the limit $y\to \infty$) we have included the skew-orthogonal polynomials for $\beta=4$ in Appendix \[a:limpolys4\] and for $\beta=1$ in Appendix \[a:limpolys1\]. The key to the method of [@AdleForrNagavanMoer2000] is an operator $A$, which acts thusly $$\begin{aligned} \label{d:Aop} A f [x] = e^{x^2/2} \frac{d}{dx} \left( e^{-x^2/2} f(x) \right).\end{aligned}$$ We will also need the inverse operator $$\begin{aligned} \nonumber A^{-1} f[x] &= \frac{e^{x^2/2}} {2} \int_{-\infty}^{\infty} {\mathrm{sgn}}(x-z) e^{-z^2/2} f(z) dz\\ \label{d:Ainvop} &= \frac{e^{x^2/2}} {2} \left( \int_{-\infty}^{x} e^{-z^2/2} f(z) dz - \int_{x}^{\infty} e^{-z^2/2} f(z) dz \right) \;.\end{aligned}$$ That this is the inverse can be checked by explicitly calculating $A A^{-1} f[x]$ and $A^{-1}A f[x]$, and using the identity $\frac{d}{dx} {\mathrm{sgn}}(x-z) = 2 \delta (x-z)$ (where care is taken to use the *distributional derivative*). The use of these operators will allow us to find relations between the $\beta=2$ inner product and the $\beta=4,1$ skew-inner products and , and then to find the sought relations between the polynomials themselves. Before proceeding, we point out for the interested reader that the original motivation for developing the technique in [@AdlevanMoer2002] was to relate the $\tau$-function solutions of the Toda lattice equations and the so-called “Pfaff $\tau$-function” solutions of the related Pfaff lattice. The Toda $\tau$-functions are matrix integrals that have determinantal expressions, but they also define polynomials that diagonalize the related matrix of inner products, which is essentially the matrix in . One can then analogously define the Pfaff lattice, which has solutions given by Pfaff $\tau$-functions, which can be expressed in terms of Pfaffians (instead of determinants). Further, the polynomials defined by these Pfaff $\tau$-functions skew-diagonalize matrices of skew-inner products like and . The conversion between the Toda lattice equations and the Pfaff lattice equations is essentially the expression of the new (Pfaff) polynomials in the basis of the original (Toda) polynomials, and applying the matrix operations in and . We refer the reader to Ref. [@AdlevanMoer2002] (and references therein) for more details. $\beta=4$ {#s:SOPs4} --------- Recall that the goal is to find the coefficients $\alpha_{j,k}$ in so that we can express the $Q_j$ in the basis of the $p_j$, which are the $\beta=2$ orthogonal polynomials \[orthogonal with respect to the inner product \], and we will use the operator $A$ from . This operator will allow us to develop both the $\beta=1$ and $\beta=4$ cases in the same framework, however we will need to define a slightly modified $\beta=4$ skew inner product, the skew-orthogonal polynomials of which are related to the $Q_j$ by a simple rescaling. The modified $\beta=4$ skew inner product is defined as $$\begin{aligned} \nonumber {\savebox{\@brx}{\(\m@th{\langle}\)} \mathopen{\copy\@brx\kern-0.5\wd\@brx\usebox{\@brx}}}f , g {\savebox{\@brx}{\(\m@th{\rangle}\)} \mathclose{\copy\@brx\kern-0.5\wd\@brx\usebox{\@brx}}}_{4}^y &:= \frac{1}{2} \int_{- \infty}^{y} dx \; e^{-x^2} \left[ f(x) g'(x)- g(x) f'(x) \right]\\ \label{d:modIP4}&= \frac{1}{2} \int_{- \infty}^{y} dx \; e^{-x^2/2} \left[ f(x) \frac{d} {dx} \left( e^{-x^2/2} g(x)\right) - g(x) \frac{d} {dx} \left( e^{-x^2/2} f(x) \right) \right],\end{aligned}$$ which is the same as , except that we have replaced $e^{-x^2} \mapsto e^{-x^2/2}$. We also define the associated monic skew-orthogonal polynomials $\{\tilde{Q}_j \}_{j=0,1, \dots}$ and normalizations $\{\tilde{q}_j \}_{j=0,1, \dots}$: $$\begin{aligned} \label{d:4modsorthoga} {\savebox{\@brx}{\(\m@th{\langle}\)} \mathopen{\copy\@brx\kern-0.5\wd\@brx\usebox{\@brx}}}\tilde{Q}_{2j} , \tilde{Q}_{2k} {\savebox{\@brx}{\(\m@th{\rangle}\)} \mathclose{\copy\@brx\kern-0.5\wd\@brx\usebox{\@brx}}}_{4}^y&= {\savebox{\@brx}{\(\m@th{\langle}\)} \mathopen{\copy\@brx\kern-0.5\wd\@brx\usebox{\@brx}}}\tilde{Q}_{2j+1} , \tilde{Q}_{2k+1} {\savebox{\@brx}{\(\m@th{\rangle}\)} \mathclose{\copy\@brx\kern-0.5\wd\@brx\usebox{\@brx}}}_{4}^y= 0,\\ \label{d:4modsorthogb} {\savebox{\@brx}{\(\m@th{\langle}\)} \mathopen{\copy\@brx\kern-0.5\wd\@brx\usebox{\@brx}}}\tilde{Q}_{2j} , \tilde{Q}_{2k+1} {\savebox{\@brx}{\(\m@th{\rangle}\)} \mathclose{\copy\@brx\kern-0.5\wd\@brx\usebox{\@brx}}}_{4}^y&= -{\savebox{\@brx}{\(\m@th{\langle}\)} \mathopen{\copy\@brx\kern-0.5\wd\@brx\usebox{\@brx}}}\tilde{Q}_{2k+1} , \tilde{Q}_{2j} {\savebox{\@brx}{\(\m@th{\rangle}\)} \mathclose{\copy\@brx\kern-0.5\wd\@brx\usebox{\@brx}}}_{4}^y = \tilde{q}_j(y) \delta_{j,k}.\end{aligned}$$ Note that the use of the tilde $\tilde{~}$ here and elsewhere in this paper (which matches the notation in [@Forrester2010]) denotes that the quantity is related to this modified $\beta=4$ skew-inner product , rather than the standard skew-inner product . By performing a change of variables we have the relations $$\begin{aligned} \label{e:Rescaleqj} q_j (y) &= 2^{ -2j -\frac{1}{2}} \tilde{q}_j \left( \sqrt{2} y \right),\\ \label{e:RescaleQj} Q_k (x, y) &= 2^{-k /2} \tilde{Q}_k \left( \sqrt{2} x, \sqrt{2} y \right),\end{aligned}$$ and so we can recover the polynomials that we are searching for. (Note that the factor of $2^{-k/2}$ ensures that the polynomials remain monic.) We can check these relations by generating the first few polynomials ${\tilde{Q}}_j$, as done for the $Q_j$ in Appendix \[s:iterative\], $$\begin{aligned} \tilde{Q}_0 (\lambda, y)& = 1, \qquad \tilde{Q}_1 (\lambda, y) = \lambda, \qquad \tilde{Q}_2 (\lambda, y)= \lambda^2 +\tilde{b} \lambda + \frac{1 -y \tilde{b}} {2},\\ \tilde{Q}_3 (\lambda, y)&= \lambda^3 - 3 \frac{1- y \tilde{b}} {2} \lambda -\tilde{b} (1+ y^2)\end{aligned}$$ \[where we used for $\tilde{Q}_3 (\lambda)$\] and the corresponding normalizations $$\begin{aligned} \tilde{q}_0 (y)&:= {\savebox{\@brx}{\(\m@th{\langle}\)} \mathopen{\copy\@brx\kern-0.5\wd\@brx\usebox{\@brx}}}\tilde{Q}_0 , \tilde{Q}_1 {\savebox{\@brx}{\(\m@th{\rangle}\)} \mathclose{\copy\@brx\kern-0.5\wd\@brx\usebox{\@brx}}}_{4}^y = \frac{\sqrt{\pi}}{4 } \;{\mathrm{erfc}}(-y) = \frac{e^{-y^2}} {2 \tilde{b}},\\ \tilde{q}_1 (y)&:= {\savebox{\@brx}{\(\m@th{\langle}\)} \mathopen{\copy\@brx\kern-0.5\wd\@brx\usebox{\@brx}}}\tilde{Q}_2 , \tilde{Q}_3 {\savebox{\@brx}{\(\m@th{\rangle}\)} \mathclose{\copy\@brx\kern-0.5\wd\@brx\usebox{\@brx}}}_{4}^y = \frac{1} {8} \left( 3 \sqrt{\pi}\, {\mathrm{erfc}}(- y) - e^{-y^2} y (9 +2 y^2) - e^{-y^2} ( 4 +y^2) \tilde{b} \right),\end{aligned}$$ with the parameter $\tilde b$ given by $$\begin{aligned} \tilde{b}= \frac{2 e^{-y^2}} {\sqrt{\pi} (1+{\mathrm{erf}}( y))}= \frac{2 e^{-y^2}} {\sqrt{\pi} \, {\mathrm{erfc}}(- y)}.\end{aligned}$$ To use this modified skew-inner product we will introduce the operator $A$ from into the $\beta=2$ inner product and we have the properties (by integrating by parts) $$\begin{aligned} \label{e:fAg} (f, Ag)_{2}^y &= - (g, A f)_{2}^y + \Omega (f, g; y),\\ \label{e:fAf} (f, Af)_{2}^y &= \frac{\Omega (f, f; y)} {2},\end{aligned}$$ where $$\begin{aligned} \label{d:Omega} \Omega (f, g; y) &:= \Big[ e^{-x^2} f(x) g(x) \Big]_{-\infty}^y = \lim_{x\to y} \Big( e^{-x^2} f(x) g(x) \Big) - \lim_{x\to -\infty} \Big( e^{-x^2} f(x) g(x) \Big).\end{aligned}$$ Then we can write $$\begin{aligned} \nonumber {\savebox{\@brx}{\(\m@th{\langle}\)} \mathopen{\copy\@brx\kern-0.5\wd\@brx\usebox{\@brx}}}f , g {\savebox{\@brx}{\(\m@th{\rangle}\)} \mathclose{\copy\@brx\kern-0.5\wd\@brx\usebox{\@brx}}}_{4}^y &= \frac{1}{2} \Big( (f, Ag)_{2}^y - (g, Af)_{2}^y \Big)\\ \label{e:IP4vsIP2} &= (f, Ag)_{2}^y - \frac{\Omega (f, g; y)} {2}.\end{aligned}$$ So we are searching for coefficients $\tilde{\alpha}_{j,k}$ $$\begin{aligned} \label{e:4SOPvsOP} \tilde{Q}_{j} = p_j + \tilde{\alpha}_{j,j-1} \; p_{j-1} + \dots + \tilde{\alpha}_{j,1} \; p_1 + \tilde{\alpha}_{j,0}\end{aligned}$$ such that the relations – hold, and we will use – to recast the problem in terms of the $\beta=2$ inner product and associated polynomials. Note that the coefficients ${\tilde{\alpha}}_{j,k}$ depend on $y$. Here we only present the results, with the detailed derivations in Appendix \[a:beta4\]. Define the skew-symmetric matrix $$\begin{aligned} \nonumber {\mathbf{W}}_m &= \left[ \begin{array}{cccccc} 0 & w_{0,1} & w_{0,2} & w_{0,3} & \cdots& w_{0,m}\\ -w_{0,1} & 0 & w_{1,2} & w_{1,3} & \cdots& w_{1,m}\\ -w_{0,2} & -w_{1,2} & 0 & w_{2,3} & \cdots& w_{2,m}\\ -w_{0,3} & -w_{1,3} & -w_{2,3} & 0 & & \\ \vdots & \vdots & \vdots & & \ddots & \\ -w_{0,m} & -w_{1,m} & -w_{2,m} & & & 0 \end{array} \right]\\ \label{d:Wmat}&= \left[ \begin{array}{cccccc} 0 & h_{1} + \frac{\Omega_{0,1}}{2} & \frac{\Omega_{0,2}}{2} & \frac{\Omega_{0,3}}{2} & \cdots & \frac{\Omega_{0,m}}{2}\\ -h_{1} - \frac{\Omega_{0,1}}{2} & 0 & h_2 + \frac{\Omega_{1,2}}{2} & \frac{\Omega_{1,3}}{2} & \cdots & \frac{\Omega_{1,m}}{2}\\ -\frac{\Omega_{0,2}}{2} & -h_2 - \frac{\Omega_{1,2}}{2} & 0 & h_3 + \frac{\Omega_{2,3}}{2} & \cdots & \frac{\Omega_{2,m}}{2}\\ -\frac{\Omega_{0,3}}{2} & -\frac{\Omega_{1,3}}{2} & -h_3 - \frac{\Omega_{2,3}}{2} & 0 && \\ \vdots & \vdots & \vdots & & \ddots & \\ -\frac{\Omega_{0,m}}{2} & -\frac{\Omega_{1,m}}{2} & -\frac{\Omega_{2,m}}{2} & & & 0 \end{array} \right],\end{aligned}$$ where $$\begin{aligned} w_{j,k}:= \delta_{j+1,k} h_k + \frac{1}{2}\Omega_{j,k}, \qquad \Omega_{j,k} := \Omega (p_j, p_k; y) = e^{-y^2} p_j (y, y) p_k (y, y).\end{aligned}$$ Note that it will turn out that ${\mathbf{W}}_m= \left[ {\savebox{\@brx}{\(\m@th{\langle}\)} \mathopen{\copy\@brx\kern-0.5\wd\@brx\usebox{\@brx}}}p_j, p_k {\savebox{\@brx}{\(\m@th{\rangle}\)} \mathclose{\copy\@brx\kern-0.5\wd\@brx\usebox{\@brx}}}_4^y \right]_{j,k=0, \dots, m}$, and so we can use the fact that ${\tilde{Q}}_j$ is a linear combination of the $p_j$’s to obtain information about the coefficients ${\tilde{\alpha}}_{j,k}$ in . Then we use a result of Knuth on overlapping Pfaffians [@Knuth1996 (5.0)–(5.1)] (see also Appendix \[a:beta4\] for more details) to obtain the following. \[p:alpha4\] Assuming $$\begin{aligned} \label{e:alpha04} \tilde{\alpha}_{j,j-1} (y) &=0, \qquad \mbox{$j$ odd},\end{aligned}$$ then for $j\geq 2$ $$\begin{aligned} \label{e:alphas4} \tilde{\alpha}_{j,k} (y)&= \left\{ \begin{array}{ll} -\frac{{\displaystyle{\mathrm{Pf}\,}{\mathbf{W}}_{j-1}^{(k\mapsto j)}}} {{\displaystyle{\mathrm{Pf}\,}{\mathbf{W}}_{j-1}}}, &\quad \mbox{$j$ even, $k\leq j-1$},\\ \\ -\frac{{\displaystyle{\mathrm{Pf}\,}{\mathbf{W}}_{j-2}^{(k\mapsto j)}}} {{\displaystyle{\mathrm{Pf}\,}{\mathbf{W}}_{j-2}}}, &\quad \mbox{$j$ odd, $k\leq j-2$}, \end{array}\right.\end{aligned}$$ where ${\mathbf{W}}_{m}^{(\eta \mapsto \nu)}$ is the matrix ${\mathbf{W}}_{m}$ from with all occurrences of the index $\eta$ replaced by the index $\nu$, and $$\begin{aligned} \label{e:alphajj} \tilde{\alpha}_{j,j} (y) &=1, \qquad \forall j.\end{aligned}$$ The normalizations are $$\begin{aligned} \label{e:qs} \tilde{q}_j (y)= \frac{{\mathrm{Pf}\,}{\mathbf{W}}_{2j+1}} {{\mathrm{Pf}\,}{\mathbf{W}}_{2j-1}},\end{aligned}$$ with the convention $$\begin{aligned} \label{e:PfWm0} {\mathrm{Pf}\,}{\mathbf{W}}_{-1} = 1.\end{aligned}$$ The proofs of these results are contained in Appendix \[a:beta4\]. With Proposition \[p:alpha4\], we can obtain the skew-orthogonal polynomials for the skew-inner product via where the $p_j$ are the NM polynomials, and the normalizations are obtained from via $$\begin{aligned} \label{e:Pfm2} q_j (y) = 2^{-2j- \frac{1}{2}} \frac{{\mathrm{Pf}\,}{\mathbf{W}}_{2j+1}} {{\mathrm{Pf}\,}{\mathbf{W}}_{2j-1}} \bigg|_{y\mapsto \sqrt{2} y}.\end{aligned}$$ From we know that the CDF of the largest eigenvalue is expressed in terms of the average , with the function $a(x) =1$. Also, with $a (x) =1$ we have $\gamma_{j,k}^{(4)}(1) = \langle Q_j, Q_k \rangle_{4}^y$. So using the skew-orthogonal polynomials , the relations tell us that the matrix in $\hat{Z}_{4,N} [1, y]$ is of the form , and so its Pfaffian is given by with $s_j = q_j$. Thus, substitution of the normalization into , with $a(x)=1$, yields the result in . The expression for $\tilde{\alpha}_{j,k}$ can be seen to recover the classical Gaussian case (with $y\to \infty$), since in this limit the polynomials $p_j$ are the Hermite polynomials \[from \] and also that $\Omega_{j,k}=0$, so the matrix is then the same as that in [@Forrester2010 Prop 6.2.1]. The derivation of the $\alpha_{j,k}$ in then proceeds identically. $\beta=1$, $N$ even ------------------- As above, we want to express the skew-orthogonal polynomials $\{ R_j \}$ from in terms of the polynomials $\{ p_j \}$ from . So we look for coefficients $\alpha_{j,k}$ such that $$\begin{aligned} \label{e:Rjalpha1} R_j = p_j + \alpha_{j, j-1} p_{j-1} + \dots + \alpha_{j, 1} p_1 + p_0,\end{aligned}$$ and again these coefficients will depend on $y$. To make further progress, we use the operator $A^{-1}$ from . First we note from that $$\begin{aligned} \nonumber (f, A^{-1}g) = (A A^{-1}f, A^{-1}g) &= -(A^{-1}f, g) + \Omega(A^{-1}f, A^{-1} g; y)\\ \label{e:AinvSymm} &= -(g, A^{-1}f) - \Phi(f, g) - \Phi (g,f),\end{aligned}$$ where $$\begin{aligned} \label{d:Phifg} \Phi(f, g)&:= \frac{1}{2} \int_{-\infty}^{y} e^{-z^2/2} f(z) dz \int_{y}^{\infty} e^{-z^2/2} g(z) dz.\end{aligned}$$ Now we can re-write the skew-inner product as $$\begin{aligned} \nonumber \langle f, g \rangle_{1}^y &= - \frac{1}{2} \Big( (f, A^{-1}g )_{2}^y - (g, A^{-1}f )_{2}^y + \Phi(f,g) - \Phi (g,f) \Big)\\ \label{e:IP1vsIP2} &= -(f, A^{-1}g)_{2}^y - \Phi(f, g).\end{aligned}$$ From here we follow the same procedure as for $\beta=4$, but replacing the matrix ${\mathbf{W}}_m$ in with the more complicated matrix $$\begin{aligned} \nonumber {\mathbf{V}}_m &= \left[ \begin{array}{cccccc} 0 & v_{0,1} & v_{0,2} & v_{0,3} & \cdots& v_{0,m}\\ -v_{0,1} & 0 & v_{1,2} & v_{1,3} & \cdots& v_{1,m}\\ -v_{0,2} & -v_{1,2} & 0 & v_{2,3} & \cdots& v_{2,m}\\ -v_{0,3} & -v_{1,3} & -v_{2,3} & 0 & & \\ \vdots & \vdots & \vdots & & \ddots & \\ -v_{0,m} & -v_{1,m} & -v_{2,m} & & & 0 \end{array} \right]\\ \label{d:Vm} &= \begin{bmatrix} 0 & h_0 - X_{0,1} -\Phi_{0,1} & X_{2,0} + \Phi_{2,0} & X_{3,0} + \Phi_{3,0}& \dots \\ -h_0 +X_{0,1} +\Phi_{0,1} & 0 & h_1 - X_{1,2} -\Phi_{1,2} & X_{3,1} + \Phi_{3,1}& \dots \\ -X_{2,0} - \Phi_{2,0} & -h_1 +X_{1,2} +\Phi_{1,2} & 0 & h_2 - X_{2,3} -\Phi_{2,3}& \dots \\ -X_{3,0} - \Phi_{3,0} & -X_{3,1}- \Phi_{3,1} & -h_2 +X_{2,3} +\Phi_{2,3} & 0\\ \vdots & \vdots & \vdots \end{bmatrix},\end{aligned}$$ where $$\begin{aligned} \label{d:Phijk} \Phi_{j,k} &:= \Phi(p_j, p_k) = \frac{1}{2} \int_{-\infty}^{y} e^{-z^2/2} p_j (z, y) dz \int_{y}^{\infty} e^{-z^2/2} p_k (z, y) dz,\\ \label{d:Xjk} X_{j,k} &:= \frac{1}{2} \left( \int_{-\infty}^y p_j (x, y) e^{-x^2/2}\; {\mathrm{erf}}\left( \frac{x}{\sqrt{2}} \right) dx \right) \int_{-\infty}^{\infty} e^{-z^2/2} p_k (z, y) dz.\end{aligned}$$ Note that we have the equality $$\begin{aligned} {\mathbf{V}}_m= \Big[ \langle p_j, p_k \rangle_{1}^y \Big]_{j,k=0, \dots, m}= -\Big[ (p_j, A^{-1} p_k)_{2}^y + \Phi(p_j, p_k) \Big]_{j,k=0, \dots, m}.\end{aligned}$$ We now give expressions for the coefficients $\alpha_{j,k} (y)$ and normalizations $r_j(y)$ in terms of the matrix ${\mathbf{V}}_m$. (We discuss the construction of the matrix in Appendix \[a:beta1\].) \[p:alpha1\] Assuming $$\begin{aligned} \label{e:alpha01} \alpha_{j,j-1} (y) &=0, \qquad \mbox{$j$ odd},\end{aligned}$$ then for $j\geq 2$ $$\begin{aligned} \label{e:alphas1b} \alpha_{j,k} (y)&= \left\{ \begin{array}{ll} -\frac{{\displaystyle{\mathrm{Pf}\,}{\mathbf{V}}_{j-1}^{(k\mapsto j)}}} {{\displaystyle{\mathrm{Pf}\,}{\mathbf{V}}_{j-1}}}, &\quad \mbox{$j$ even, $k\leq j-1$},\\ \\ -\frac{{\displaystyle{\mathrm{Pf}\,}{\mathbf{V}}_{j-2}^{(k\mapsto j)}}} {{\displaystyle{\mathrm{Pf}\,}{\mathbf{V}}_{j-2}}}, &\quad \mbox{$j$ odd, $k\leq j-2$}, \end{array}\right.\end{aligned}$$ where ${\mathbf{V}}_{m}^{(\eta \mapsto \nu)}$ is the matrix ${\mathbf{V}}_{m}$ with all occurrences of the index $\eta$ replaced by the index $\nu$, and $$\begin{aligned} \label{e:alphajj1} \alpha_{j,j} (y) &=1, \qquad \forall j.\end{aligned}$$ The normalizations are $$\begin{aligned} \label{e:rs} r_j (y)= \frac{{\mathrm{Pf}\,}{\mathbf{V}}_{2j+1}} {{\mathrm{Pf}\,}{\mathbf{V}}_{2j-1}}\end{aligned}$$ with the convention $$\begin{aligned} \label{e:PfVm0} {\mathrm{Pf}\,}{\mathbf{V}}_{-1} = 1.\end{aligned}$$ Note that the coefficients $\alpha_{j,k}$ depend on $y$. We give the proof of Proposition \[p:alpha1\] in Appendix \[a:beta1\]. The coherence of the $\alpha_{j,k}$ in Proposition \[p:alpha1\] with the $y\to \infty$ classical Gaussian result in is not as straightforward as in the $\beta=4$ case above, and we go through the details in Appendix \[a:limpolys1\]. The extra complications are because the technique of [@AdleForrNagavanMoer2000] did not use an exact analogue of our matrix ${\mathbf{V}}_m$ in ; they instead used some shrewd linear algebra to express the matrix $[(p_j, A^{-1} p_k)]$ in terms of the matrix $[(p_j, A p_k)]$ and some other matrices containing the polynomial normalizations $h_j$. This approach worked as it relied on inverting matrices that are (almost) diagonal, however the analogous step in our case (with finite $y$) involves inverting a full $N \times N$ matrix, and so it is infeasible here. At any rate, setting $y = \infty$, we see from that the matrix $\Phi (p_j, p_k) =0$, and from that the polynomials $p_j$ become the Hermite polynomials leading us to the simple expression for the elements of $X_{j,k}$. Using these facts we recover the classical case in , where $\alpha_{2j+1, 2j-1} = -j$ and is zero otherwise — see Appendix \[a:limpolys1\] for the details. As with the $\beta=4$ case above, by use of the polynomials $R_j$ the matrix in $\hat{Z}_{1,N} [1,y]$ of has the skew-diagonal structure in and so its Pfaffian is given by with $s_j= r_j$. Substitution of into gives the expression for the CDF of the largest eigenvalue in . Skew-orthogonal polynomials for more general weight functions ------------------------------------------------------------- As mentioned after Eq. , the density functions in this paper are of the form , which are a type of Janossy density, so a natural question is to ask if our methods can be applied more generally. We see from and that the key step involved in calculating the polynomial coefficients in Proposition \[p:alpha4\] (for the GSE) and Proposition \[p:alpha1\] (for the GOE) is writing the corresponding skew-inner product in terms of the GUE inner product. The quantity separating the procedure here from the classical case in [@AdleForrNagavanMoer2000] is $\Omega$ in , which \[via \] also determines the quantity $\Phi$ in . This $\Omega$ function is particular to the Gaussian weight and the eigenvalue domain $(-\infty, y)$, however, from following the matrix manipulations in Appendices \[a:beta4\] and \[a:beta1\], we can conclude that our method will work for Janossy densities over more general domains and for the other classical weight functions. Indeed, let $$\begin{aligned} w_{\beta} (x):= \left\{ \begin{array}{cc} e^{-\beta x^2/2},& \mbox{(Gaussian)},\\ x^{a \beta/2 } e^{-\beta x/2},& \mbox{(Laguerre)},\\ (1-x)^{a \beta/2} (1+x)^{b \beta/2},& \mbox{(Jacobi)},\\ ( 1+ x^2)^{-\beta (N-1)/2 +1},& \mbox{(Cauchy)}, \end{array}\right.\end{aligned}$$ and define the inner product $$\begin{aligned} (f,g)_2^{Y} := \int_Y w_2 (x) f(x) g(x) dx\end{aligned}$$ and the skew-inner products $$\begin{aligned} \label{f:GenIP4} \langle f, g \rangle_4^Y&:= \frac{1}{2} \int_Y w_4 (x) \left[ f(x) g'(x)- g(x) f'(x) \right] dx,\\ \label{f:GenIP1} \langle f , g \rangle_{1}^Y &:= \frac{1}{2} \int_Y w_1 (x) f(x) \int_Y w_1 (z) g (z) {\mathrm{sgn}}(z- x)\, dz\, dx,\end{aligned}$$ where $Y\subset U_w$, with $U_w$ the maximal domain for the weight function $w_\beta$. Then we can define a new $\Omega$ and apply the procedures in Appendices \[a:beta4\] and \[a:beta1\] to obtain the skew-orthogonal polynomials in terms of the orthogonal polynomials. (Of course, explicitly calculating the $\beta=1$ and $\beta=4$ polynomials using this method relies on knowing the orthogonal polynomials for the corresponding $\beta=2$ problem; a non-trivial hurdle.) Note that in the case of Hermitian matrix models (i.e. with $\beta = 2$), such orthogonal polynomials have been studied in the context of the counting statistics of eigenvalues in these ensembles (see e.g. [@cao2014continuous; @witte2012variance]). For the purposes of illustration assume $Y:= (y_1, y_2) \subset \mathbb{R}$, then we replace and by $$\begin{aligned} \Omega (f, g; Y) &:= \lim_{x\to y_2} \Big( w_2(x) f(x) g(x) \Big) - \lim_{x\to y_1} \Big( w_2(x) f(x) g(x) \Big)\\ \Phi (f,g; Y)&:= \frac{1}{2} \int_{y_1}^{y_2} w_1(z) f(z) dz \left( \int_{y_2}^{\infty} w_1(z) g(z) dz - \int_{-\infty}^{y_1} w_1(z) g(z) dz \right),\end{aligned}$$ and then Propositions \[p:alpha4\] and \[p:alpha1\] hold, with the matrices ${\mathbf{W}}_m$ and ${\mathbf{V}}_m$ modified accordingly. Asymptotic analysis of the CDF of the largest eigenvalue for $F_{4,N}$ for large $N$ {#s:F4asympt} ==================================================================================== In this section we show that our formula for the CDF $F_{4,N} (y)$ is amenable to an asymptotic analysis, in the large $N$ limit, which allows us to obtain an alternative derivation of the Tracy-Widom formula for $\beta = 4$ [@TracWido1996]. Indeed, we will show below that, from the expression in (\[e:F4N\]), we can obtain $$\begin{aligned} \label{e:TW4} \lim_{N \to \infty} F_{4,N} \left( y = \sqrt{2N} + \frac{s}{2^{7/6}}N^{-1/6} \right) = \exp{\left(-\frac{1}{2} \int_s^\infty (x-s) q^2(x) dx \right)} \cosh{\left(\frac{1}{2} \int_s^\infty q(x)dx \right)},\end{aligned}$$ where $q(x)$ is the Hastings-McLeod solution of the Painlevé II equation as in . To show this result (\[e:TW4\]), starting from our expression in (\[e:F4N\]), we will first provide an explicit expression for ${\mathrm{Pf}\,}{\mathbf{W}}_{2N-1}$, where the matrix ${\mathbf{W}}_{m}$ is defined in (\[d:Wmat\]). It is convenient first to define $$\begin{aligned} \label{d:M1} M_{i_1, i_2}(y) = \frac{1}{2}\dfrac{\prod_{m={i_1}}^{i_2-1} h_{2m+2} (y)}{\prod_{m=i_1}^{i_2} h_{2m+1} (y)} p_{2i_1}(y, y) p_{2i_2+1}(y, y) e^{-y^2} \;, \;\;\;\; i_2 \geq i_1 \;,\end{aligned}$$ where we recall that the $p_k$’s are the NM orthogonal polynomials (\[e:NM1\])–(\[e:NM3\]) while the $h_k$’s are their corresponding norms (\[e:ynorm2\]). (We will often suppress the explicit dependence on $y$ for concision.) In view of the asymptotic analysis, it is useful to rewrite as $$\begin{aligned} \label{d:M2} M_{i_1, i_2}(y) = \frac{1}{2} \frac{1}{\sqrt{{\check{\mathsf{R}}}_{2i_2+1} (y)}} \prod_{m=i_1}^{m=i_2-1} \sqrt{\frac{{\check{\mathsf{R}}}_{2m+2} (y)}{{\check{\mathsf{R}}}_{2m+1} (y)}} \, \psi_{2i_1}(y, y) \psi_{2i_2+1}(y, y) \;, \;\;\;\; i_2 \geq i_1 \;\end{aligned}$$ in terms of ${\check{\mathsf{R}}}_m = h_m/h_{m-1}$ from and the so-called “wave functions” $\psi_k(x, y)$ given by $$\begin{aligned} \label{d:psi} \psi_k(x, y) = \frac{p_k(x, y)}{\sqrt{h_k (y)}} e^{-\frac{x^2}{2}} \;. \end{aligned}$$ With these definitions we find the following convenient expression for the Pfaffian in . \[lem:PfW\] With $M_{j,k}$ defined in and ${\mathbf{W}}_m$ from we have (suppressing the explicit dependence on $y$) $$\begin{aligned} \label{e:expansion:Pf} {\mathrm{Pf}\,}{\mathbf{W}}_{2N-1}(y) &= \left(\prod_{j=0}^{N-1} h_{2j+1} \right) \left(1 + \sum_{p=1}^N \; \sum_{{\mathcal{I}}_{2p}} M_{i_1, i_2} M_{i_3,i_4} \cdots M_{i_{2p-1}, i_{2p}} \right) \;,\end{aligned}$$ where we have used the notation $$\begin{aligned} \label{d:I2p} {\mathcal{I}}_{2p}: 0\leq i_1 \leq i_2 < i_3 \leq i_4 < i_5 \leq i_6 < \dots <i_{2p-1}\leq i_{2p} \leq N-1\end{aligned}$$ for the terminals on the second sum. (Note that the indices in the sum obey both strict and non-strict inequalities in the sequence $i_{2j-1}\leq i_{2j} < i_{2j+1} \leq i_{2j+2}$.) [*Proof*:]{}First we define $$\begin{aligned} \label{e:sigmaj} \sigma_{j} := \frac{p_{j} (y, y)}{\sqrt{2}} e^{-y^2/2}\end{aligned}$$ so that $$\begin{aligned} \label{d:welts} w_{j,k} = h_k \delta_{j+1,k} + \sigma_j \sigma_k.\end{aligned}$$ Here we use the expression for the Pfaffian in , where the sum is over all perfect matchings on $2N$ sites $\{ 0, 1, \dots, 2N-1\}$ and so $$\begin{aligned} \label{e:PfWPMs} {\mathrm{Pf}\,}{\mathbf{W}}_{2N-1}(y) = \sum_{\mu\in M_{2N}} \varepsilon (\mu)\; w_{i_1, j_1} w_{i_2, j_2}\cdot\cdot\cdot w_{i_N, j_{N}},\end{aligned}$$ where we recall that the perfect matchings $\mu\in M_{2N}$ are represented by link diagrams as in Figure \[f:PMeg\]. We see from that the summand will include a factor of $h_{i+1} + \sigma_i \sigma_{i+1}$ if and only if the link diagram of the perfect matching includes a “little link” from site $i$ to $i+1$ (see Figure \[f:LLs\]), and otherwise every factor is of the form $\sigma_i \sigma_{j}$. We denote by $M_{2N; i_j, \dots, i_k}$ the set of perfect matchings of $2N$ sites with “little links” $(i_j, i_j+1), \dots, (i_k, i_k +1)$ and no others. Then becomes $$\begin{aligned} \nonumber {\mathrm{Pf}\,}{\mathbf{W}}_{2N-1}(y) &= \sum_{\mu\in M_{2N}} \varepsilon (\mu)\; \Big( \sigma_{0} \dots \sigma_{2N-1} \Big)\\ \nonumber &+ \sum_{i_1=0}^{2N-2} \sum_{\mu\in M_{2N; i_1}} \varepsilon (\mu)\; \Big( \sigma_{0} \dots \sigma_{i_1 -1} \Big) h_{i_1+1} \Big( \sigma_{i_1+2} \dots \sigma_{2N-1} \Big) \\ \nonumber &+ \sum_{i_1=0}^{2N-2} \sum_{i_2=i_1+2}^{2N-2} \sum_{\mu\in M_{2N; i_1, i_2}} \varepsilon (\mu)\; \Big( \sigma_{0} \dots \sigma_{i_1 -1} \Big) h_{i_1+1} \Big( \sigma_{i_1 +2} \dots \sigma_{i_2 -1}\Big) h_{i_2+1} \Big( \sigma_{i_2+2} \dots \sigma_{2N-1} \Big) \\ \nonumber &\vdots\\ \label{e:PfW1} &+ \sum_{i_1=0}^{2N-2} \sum_{i_2=i_1+2}^{2N-2} \dots \sum_{i_N=i_{N-1}+2}^{2N-2} \sum_{\mu\in M_{2N; i_1, \dots i_{N-1}}} \varepsilon (\mu)\; h_{i_1+1} \dots h_{i_{N-1}+1},\end{aligned}$$ where we see that each $h_{j+1}$ replaces a pair $\sigma_j \sigma_{j+1}$ in the summand. In all summands are now independent of the matching $\mu$, except for the factor of $\varepsilon(\mu)$, and so we factor these out and want to show that $$\begin{aligned} \label{e:SignSum} \sum_{\mu\in M_{2N; I}} \varepsilon (\mu) = 1\end{aligned}$$ for any set of indices $I$. From we have that there is an even number of perfect matchings, excluding the identity perfect matching $\{ (0,1), (2,3), \dots, (2N-2, 2N-1) \}$, which has “little links” at all sites. We can pair these non-identities in such a way that for each perfect matching with a sign of $(+1)$ there is a partner with sign $(-1)$, and so the sum in will have contribution of zero from these terms, leaving just the identity matching. We do this pairing according to the following algorithm. Any link diagram of a non-identity matching will have at least one non-identity link \[i.e. a link not of the form ($2j$, $2j + 1$)\], and will match one of the two forms in Figure \[f:LinkPairing1\], where $2j$ is the left site of the left-most non-identity link (and thus $2j+1$ is by necessity also part of a non-identity link). Every non-identity perfect matching $\mu$ of the form in Figure \[f:LinkPairing1\] (a) can be paired with a $\hat{\mu}$ of the form in Figure \[f:LinkPairing1\] (b), where the link patterns are identical except at the four sites $\{2j, 2j+1, 2s, 2t+1\}$. The extra crossing in $\hat{\mu}$ then implies that $\varepsilon( \mu) =-\varepsilon( \hat{\mu})$, so $$\begin{aligned} \sum_{\mu\in M_{2N;I}} \varepsilon(\mu) = \varepsilon(\text{identity}) = 1.\end{aligned}$$ This leaves us with $$\begin{aligned} \nonumber {\mathrm{Pf}\,}{\mathbf{W}}_{2N-1}(y) &= \Big( \sigma_{0} \dots \sigma_{2N-1} \Big)+ \sum_{i_1=0}^{2N-2} \; \Big( \sigma_{0} \dots \sigma_{i_1 -1} \Big) h_{i_1+1} \Big( \sigma_{i_1+2} \dots \sigma_{2N-1} \Big) \\ \nonumber &+ \sum_{i_1=0}^{2N-2} \sum_{i_2=i_1+2}^{2N-2} \; \Big( \sigma_{0} \dots \sigma_{i_1 -1} \Big) h_{i_1+1} \Big( \sigma_{i_1 +2} \dots \sigma_{i_2 -1}\Big) h_{i_2+1} \Big( \sigma_{i_2+2} \dots \sigma_{2N-1} \Big) \\ \nonumber &\vdots\\ \label{e:PfW2} &+ h_{1} h_{3} \dots h_{2N -1},\end{aligned}$$ where we have replaced the bottom line in by the product over the odd indexed $h_j$, which is the only term in that sum (since there is only one way to replace all pairs of $\sigma_j \sigma_{j+1}$). We now just need to match up the expression in with products of $M_{i_1, i_2}$ from — first we note $$\begin{aligned} \label{d:Msigma} M_{i_1, i_2}(y) = \sigma_{2i_1}(y) \frac{h_{2i_1+2}}{h_{2i_1 +1}} \frac{h_{2i_1+4}} {h_{2i_1 +3}} \cdots \frac{h_{2i_2}} {h_{2i_2 -1}} \frac{\sigma_{2i_2+1}(y)} {h_{2i_2 +1}} \;, \;\;\;\; i_2 \geq i_1 \;.\end{aligned}$$ In each term of , we start from the left with $\sigma_0$ and pair up each even $\sigma_{2j}$ with the nearest odd $\sigma_{2k+1}$ to its right ($k\geq j$). This will be easiest to see if we start with an example of one of the terms in , such as the term $$\begin{aligned} \label{e:PfTermEg} \Big( \sigma_{0} \dots \sigma_{i_1 -1} \Big) h_{i_1+1} \Big( \sigma_{i_1 +2} \dots \sigma_{i_2 -1}\Big) h_{i_2+1} \Big( \sigma_{i_2+2} \dots \sigma_{2N-1} \Big) = \sigma_0 \sigma_1 h_3 \sigma_4 \sigma_5 \sigma_6 h_8 \sigma_9 \sigma_{10} \sigma_{11}\end{aligned}$$ with $N=6, i_1=2, i_2=7$, where the corresponding link diagram is drawn in Figure \[f:LinkDiagEg1\]. In the diagram, we have included the labels of the $h_j$ which are present ($h_3$ and $h_8$) and, for convenience, the odd $h_j$ which are missing using a “hat” ($\hat{h}_1, \hat{h}_5, \hat{h}_7, \hat{h}_9$ and $\hat{h}_{11}$). (-0.5,0)–(13.5,0); at (0,0) [$0$]{}; at (0.5,1) [$\hat{h}_1$]{}; at (1,0) [$1$]{}; at (2,0) [$2$]{}; at (2.5,1) [$h_3$]{}; at (3,0) [$3$]{}; at (4,0) [$4$]{}; at (4.5,1) [$\hat{h}_5$]{}; at (5,0) [$5$]{}; at (6,0) [$6$]{}; at (6.5,1) [$\hat{h}_7$]{}; at (7,0) [$7$]{}; at (7.5,1) [$h_8$]{}; at (8,0) [$8$]{}; at (8.5,1) [$\hat{h}_9$]{}; at (9,0) [$9$]{}; at (10,0) [$10$]{}; at (10.5,1) [$\hat{h}_{11}$]{}; at (11,0) [$11$]{}; (0,0) to\[out=60,in=180\] (0.5,0.33) to\[out=0,in=120\] (1,0); (2,0) to\[out=60,in=180\] (2.5,0.33) to\[out=0,in=120\] (3,0); (4,0) to\[out=60,in=180\] (4.5,0.33) to\[out=0,in=120\] (5,0); (6,0) to\[out=60,in=180\] (7.5,0.66) to\[out=0,in=120\] (9,0); (7,0) to\[out=60,in=180\] (7.5,0.33) to\[out=0,in=120\] (8,0); (10,0) to\[out=60,in=180\] (10.5,0.33) to\[out=0,in=120\] (11,0); Dividing through by $h_1 h_3 \cdots h_{11}$ we obtain $$\begin{aligned} \label{e:sigmaMeg} \frac{\sigma_0 \sigma_1} {h_1} \frac{\sigma_4 \sigma_5} {h_5} \frac{\sigma_6 h_8 \sigma_9}{h_7 h_9} \frac{\sigma_{10} \sigma_{11}} {h_{11}}= M_{0,0} M_{2,2} M_{3,4} M_{5,5}\end{aligned}$$ using . We see that all the possible combinations of $M_{i_1,i_2} M_{i_3,i_4} M_{i_5,i_6} M_{i_7,i_8}$, with $0\leq i_1\leq i_2< i_3\leq i_4< i_5\leq i_6< i_7\leq i_8 \leq 5$, will appear in the second line of — each of the odd $h_{2j+1}$ that appear are cancelled on division by $h_1 h_3 \cdots h_{11}$ and all the even $h_{2k}$ are subsumed into the $M_{i,j}$ containing the surrounding $\sigma$’s. In general, for any $M_{i_1, i_2}$ we will have a factor $\frac{\sigma_{2i_1} \sigma_{2i_2 +1}} {h_{2i_2 +1}}$ for each neighbouring even–odd pair of $\sigma$’s and the $h$ corresponding to the right hand edge, and this pair will be accompanied by a factor of $\frac{h_{2j}} {h_{2j-1}}$ for each missing pair $\sigma_{2j-1} \sigma_{2j}$ in the interval $i_1 < j < i_2$, which is the expression in . Then, to obtain , we rewrite as a sum over the number of $\sigma$ pairs in each term and divide through by $h_1 h_3 \cdots h_{2N-1}$. Finally, we rewrite the products of $\sigma$’s and $h$’s as in and the indices obey the rule in . $\Box$ We recall that the $h_k$’s as well as the $M_{i_1, i_2}$ depend explicitly on $y$. The expression for the Pfaffian in (\[e:expansion:Pf\]) is quite convenient to analyse the large $N$ limit of $F_{4,N}(y)$ which thus reads \[see Eq. (\[e:F4N\])\] $$\begin{aligned} \label{e:F4N2} F_{4,N}(y) =\frac{2^{N^2}}{\pi^{N/2}} \prod_{j=0}^{N-1} \frac{1}{(2j +1)!} \left(\prod_{j=0}^{N-1} h_{2j+1} \right) \left(1 + \sum_{p=1}^N \; \sum_{{\mathcal{I}}_{2p}} M_{i_1, i_2} M_{i_3,i_4} \cdots M_{i_{2p-1}, i_{2p}} \right) \Bigg|_{y \mapsto \sqrt{2} y} \;.\end{aligned}$$ Let us first check from this formula (\[e:F4N2\]) that $\lim_{y \to \infty}F_{4,N}(y) = 1$. From (\[d:M1\]), and the knowledge from that when $y \to \infty$, the norms $h_j$’s converge to the norms of the Hermite polynomial of degree $j$, i.e. $$\begin{aligned} \label{e:norm2lim} h_j (\infty) = \lim_{y \to \infty} h_{j}(y) = \sqrt{\pi} \frac{j!}{2^j},\end{aligned}$$ it is rather clear that $$\begin{aligned} \label{e:norm1} \lim_{y \to \infty} \left(1 + \sum_{p=1}^N \; \sum_{{\mathcal{I}}_{2p}} M_{i_1, i_2} M_{i_3,i_4} \cdots M_{i_{2p-1}, i_{2p}} \right)\ = 1\;. \end{aligned}$$ We can also use to obtain $$\begin{aligned} \label{e:norm3} \lim_{y \to \infty} \prod_{j=0}^{N-1} h_{2j+1}(y) = {\pi}^{N/2} \frac{\prod_{j=0}^{N-1} (2j+1)!}{2^{\sum_{j=0}^{N-1}(2j+1)}} = \pi^{N/2} \frac{\prod_{j=0}^{N-1} (2j+1)!}{2^{N^2}} \;,\end{aligned}$$ which implies, by combining (\[e:F4N2\]), (\[e:norm1\]) and (\[e:norm3\]), that $$\begin{aligned} \label{e:norm4} \lim_{y \to \infty} F_{4,N}(y) = 1 \;,\end{aligned}$$ as it should. We now proceed to obtain the scaled limit . Let us start by analyzing the first factors of $F_{4,N}(y)$ in (\[e:F4N2\]) and define $$\begin{aligned} \label{d:ZN} {\cal Z}_N = \frac{2^{N^2}}{\pi^{N/2}} \prod_{j=0}^{N-1} \frac{1}{(2j +1)!} \left(\prod_{j=0}^{N-1} h_{2j+1} \right) \;.\end{aligned}$$ It is easy to check that $$\begin{aligned} \label{e:ZN1} \frac{{\cal Z}_{N-1} {\cal Z}_{N+1}}{{\cal Z}_N^2} = \frac{1}{N(N+1/2)} \frac{h_{2N+1}}{h_{2N-1}} = \frac{{\check{\mathsf{R}}}_{2N+1} {\check{\mathsf{R}}}_{2N}}{N(N+1/2)} \;.\end{aligned}$$ Let us assume the asymptotic scaling behavior $$\begin{aligned} \label{e:scaling} \ln {\cal Z}_N(y) \mathop{\rightarrow}\limits_{N\to \infty} f(2^{7/6} N^{1/6} (y-\sqrt{2N})) \;,\end{aligned}$$ with some function $f$, independent of $N$, yet to be determined. Assuming this scaling behavior (\[e:scaling\]), and setting $y = \sqrt{2N} + (s/2^{7/6}) N^{-1/6}$, the left hand side of Eq. (\[e:ZN1\]) becomes $$\begin{aligned} \label{e:ZN2} \ln {\cal Z}_{N-1} + \ln {\cal Z}_{N+1} - 2 \ln {\cal Z}_{N} = 2^{4/3} f''(s) N^{-2/3} + {o}(N^{-2/3}) \;.\end{aligned}$$ Let us now analyse the right hand side of (\[e:ZN1\]) in the large $N$ limit, where, from [@NadaMaju2011], we have the asymptotic behavior $$\begin{aligned} \label{e:RNasympt1} {\check{\mathsf{R}}}_N\left(\sqrt{2N} + \frac{x}{\sqrt{2}} N^{-1/6}\right) = \frac{N}{2} \left(1 - N^{-2/3} q^2(x) + {o}(N^{-{2/3}}) \right) \;,\end{aligned}$$ where $q(s)$ is defined in . This implies, setting again $y = \sqrt{2N} + (s/2^{7/6}) N^{-1/6}$, that $$\begin{aligned} \label{e:RNasympt2} {\check{\mathsf{R}}}_{2N}(\sqrt{2}\, y) = N \left(1 - (2N)^{-2/3} q^2(s) + o(N^{-2/3}) \right) \;.\end{aligned}$$ Hence the logarithm of the right hand side of Eq. (\[e:ZN1\]) reads $$\begin{aligned} \label{e:RNasympt3} \ln\left( \frac{{\check{\mathsf{R}}}_{2N+1}(\sqrt{2}\, y) {\check{\mathsf{R}}}_{2N}(\sqrt{2}\, y)}{N(N+1/2)} \right) = - 2^{1/3} N^{-2/3} q^2(s) + o(N^{-2/3}) \;.\end{aligned}$$ Taking the logarithm of the relation in (\[e:ZN1\]) and equating the leading terms, of order ${\cal O}(N^{-2/3})$ on both sides, one finds $$\begin{aligned} \label{e:RNasympt4} f''(s) = -\frac{1}{2} q^2(s) \;.\end{aligned}$$ Integrating twice this relation (\[e:RNasympt4\]), using that $\lim_{s \to \infty} f'(s) = 0$ \[since the probability density function $F'_{4,N}(y) \to 0$ as $y \to \infty$\] as well as $\lim_{s \to \infty} f(s) = 0$ \[since $F_{4,N}(y) \to 1$ as $y \to \infty$, see Eq. (\[e:norm4\])\], one obtains $$\begin{aligned} \label{e:RNasympt5} f(s) = -\frac{1}{2} \int_s^{\infty} (x- s) q^2(x) dx \;.\end{aligned}$$ Therefore, recalling (\[e:scaling\]) one obtains $$\begin{aligned} \label{e:RNasympt6} \lim_{N\to \infty} {\cal Z}_N \left(y = \sqrt{2N} + (s/2^{7/6})N^{-1/6}\right) \Big|_{y\mapsto \sqrt{2} y}= \exp{\left[-\frac{1}{2} \int_s^{\infty} (x- s) q^2(x) dx\right] } \;,\end{aligned}$$ which gives the first factor of the Tracy-Widom distribution for $\beta=4$ \[see Eq. (\[e:TW4\])\]. We now analyse the large $N$ behavior of the second factor in the expression of the Pfaffian in Eq. (\[e:expansion:Pf\]). For this purpose, we will take advantage of the analysis performed in [@PerrSche2014]. In fact, one can show that, in the large $N$ limit, the multiple sums in Eq. (\[e:expansion:Pf\]) are dominated by the region where $i_1, i_2, \cdots, i_{2p}$ are close to $N$. For later convenience, we reverse the order of the indices in the product of $M_{i_1, i_2}$ by looking for $M_{N- k_1, N- k_2}$, then from the results obtained in [@PerrSche2014] for the asymptotic forms of the “wave functions” in (\[d:psi\]) $$\begin{aligned} \label{e:LargePsi} \psi_N (y, y) \mathop{\sim}\limits_{N\to \infty} 2^{1/4} N^{-1/12} q \left( \sqrt{2} N^{1/6} (y- \sqrt{2N} ) \right)\end{aligned}$$ and using $y\mapsto \sqrt{2} y = 2\sqrt{N} + \frac{s}{2^{2/3} N^{1/6}}$ we have $$\begin{aligned} \label{e:largePsi2} \psi_{2(N- k)} (y, y) &\sim 2^{1/6} N^{-1/12} q\left( s+ 2^{2/3} \frac{k}{N^{1/3}} \right).\end{aligned}$$ Using for the pre-factors in one gets $$\begin{aligned} \label{e:Masympt2a} M_{N-k_1, N-k_2}\left(y = 2\sqrt{N} + \frac{s}{2^{2/3} N^{1/6}} \right) \sim \frac{1}{(2N)^{2/3}} q\left(s + 2^{2/3} \frac{k_1}{N^{1/3}}\right) q\left(s + 2^{2/3} \frac{k_2}{N^{1/3}}\right) \;,\end{aligned}$$ which we will be the useful form in the following. Indeed, performing first the change of variables $i_j = N - k_j$ in the second factor of Eq. (\[e:F4N2\]) and then using (\[e:Masympt2a\]) one finds, at leading order for large $N$, setting again $y = \sqrt{2N} + (s/2^{7/6}) N^{-1/6}$, $$\begin{aligned} \label{e:Masympt3} \sum_{{\mathcal{I}}_{2p}} M_{i_1, i_2} M_{i_3,i_4} \cdots M_{i_{2p-1}, i_{2p}} \Big|_{y\mapsto \sqrt{2} y} &\sim \frac{1}{(2N)^{\frac{2p}{3}}}\sum_{{\mathcal{K}}_{2p}} q\left(s + 2^{\frac{2}{3}} \frac{k_1}{N^{\frac{1}{3}}} \right) \cdots q\left(s + 2^{\frac{2}{3}} \frac{k_{2p}} {N^{\frac{1}{3}}} \right) \;,\end{aligned}$$ where, similar to , we denote $$\begin{aligned} \label{e:K2p} {\mathcal{K}}_{2p}: N \geq k_1 \geq k_2 > k_3 \geq k_4 > k_5 \geq k_6> \dots > k_{2p -1} \geq k_{2p} \geq 1\; .\end{aligned}$$ In the limit $N \to \infty$ the discrete sums over the $k_j$’s become integrals. Performing the change of variables $v_j = 2^{2/3} k_j/N^{1/3}$ one finds $$\begin{aligned} \label{e:Masympt4a} &&\;\;\; \sum_{{\mathcal{I}}_{2p} } M_{i_1, i_2} M_{i_3,i_4} \cdots M_{i_{2p-1}, i_{2p}} \nonumber \\ &\sim& \frac{1}{2^{2p}} \int_0^{\infty} dv_{2p}\int_{v_{2p}}^{\infty} dv_{2p-1} \cdots \int_{v_2}^{\infty} dv_1 \, q(s+v_1) \cdots q(s+v_{2p-1}) q(s+v_{2p}) \;.\end{aligned}$$ Since the integrand in (\[e:Masympt4a\]) is completely symmetric under the permutation of the variables $v_i$’s, the nested integral can actually simply be written as $$\begin{aligned} \label{e:Masympt5} &&\;\;\; \sum_{{\mathcal{I}}_{2p} } M_{i_1, i_2} M_{i_3,i_4} \cdots M_{i_{2p-1}, i_{2p}} \sim \frac{1}{(2p)!} \left(\frac{1}{2} \int_s^\infty dx \, q(x)\right)^{2p} \;.\end{aligned}$$ Finally, summing over $p$ in Eq. (\[e:F4N2\]), one obtains $$\begin{aligned} \label{e:Masympt6a} \;\;\; \lim_{N \to \infty} \left[1 + \sum_{p=1}^N \; \sum_{{\mathcal{I}}_{2p} } M_{i_1, i_2} M_{i_3,i_4} \cdots M_{i_{2p-1}, i_{2p}}\right] &=& \sum_{p=0}^\infty \frac{1}{(2p)!} \left(\frac{1}{2} \int_s^\infty dx \, q(x)\right)^{2p} \nonumber \\ &=& \cosh\left(\frac{1}{2} \int_s^\infty dx \, q(x) \right) \;.\end{aligned}$$ Combining Eqs. (\[e:F4N2\]), (\[d:ZN\]), (\[e:RNasympt6\]) and (\[e:Masympt6a\]), one obtains the desired expression given in (\[e:TW4\]) for the $\beta=4$ Tracy-Widom distribution. Asymptotic analysis of the CDF of the largest eigenvalue for $F_{1,N}$ for large $N$ {#s:F1asympt} ==================================================================================== We now show that starting with , we can obtain the limiting formula for $\beta=1$ [@TracWido1996] $$\begin{aligned} \label{e:TW1} \lim_{N \to \infty} F_{1, N} \left( y = \sqrt{2N} + \frac{s}{\sqrt{2} N^{1/6}} \right) &= \exp\left( -\frac{1}{2} \int_s^{\infty} (x-s) q(x)^2 dx \right) \exp\left( - \frac{1}{2} \int_s^{\infty} q(x) dx \right),\end{aligned}$$ where we proceed in much the same way as in Section \[s:F4asympt\] above for $\beta=4$. From the definitions in and we have [$$\begin{aligned} \nonumber X_{j,k} + \Phi_{j,k} &= \frac{1}{2} \left( \int_{-\infty}^y e^{-\frac{x^2}{2}} p_j (x, y) \; {\mathrm{erf}}\left( \frac{x}{\sqrt{2}} \right) dx \right) \int_{-\infty}^{\infty} e^{-\frac{x^2}{2}} p_k (x, y) dx + \frac{1}{2} \int_{-\infty}^{y} e^{-\frac{x^2}{2}} p_j (x, y) dx \int_{y}^{\infty} e^{-\frac{x^2}{2}} p_k (x, y) dx\\ \label{e:Ljk1} &= \frac{1}{2} \left( \int_{-\infty}^y e^{-\frac{x^2}{2}} p_j (x, y) \; {\mathrm{erfc}}\left( - \frac{x}{\sqrt{2}} \right) dx \right) P_k (\infty, y) - \frac{1}{2} P_{j} (y, y) P_{k} (y, y),\end{aligned}$$]{} where we have introduced the notation $$\begin{aligned} P_j(x, y) := \int_{-\infty}^x e^{-z^2/2} p_{j} (z, y) dz, \qquad P_j(\infty, y) := \int_{-\infty}^{\infty} e^{-z^2/2} p_{j} (z, y) dz.\end{aligned}$$ For later use we also similarly define $$\begin{aligned} \label{d:CapPsi} \Psi_j(\infty, y) := \int_{-\infty}^{\infty} \psi_j (z, y) dz. $$ Using the identities and to perform the integrals in gives $$\begin{aligned} \label{e:Pjxy} P_j(x, y)&= \frac{1}{2} {\mathrm{erfc}}\left( - \frac{x}{\sqrt{2}} \right) P_j (\infty, y) - e^{-x^2/2}p_{j-1} (x, y) + e^{-x^2/2} \mathrm{LoP}_{j-2}\\ \nonumber \int_{-\infty}^x e^{-\frac{z^2}{2}} p_j (z, y) \; {\mathrm{erfc}}\left( - \frac{z}{\sqrt{2}} \right) dz &= - e^{-\frac{x^2}{2}} {\mathrm{erfc}}\left( - \frac{x}{\sqrt{2}} \right) p_{j-1} (x, y) + \frac{1}{4} {\mathrm{erfc}}\left( - \frac{x}{\sqrt{2}} \right)^2 P_j (\infty, y)\\ \label{e:erfcinteg} &+ e^{-\frac{x^2}{2}} \mathrm{LoP}_{j-2} + e^{-x^2} \mathrm{LoP}_{j-2}\end{aligned}$$ where the notation $\mathrm{LoP}_{j-2}$ denotes “lower-order polynomials” up to degree $j-2$, that is, some combination of $p_0 (x, y), p_1 (x,y), \dots, p_{j-2} (x,y)$. Noting that ${\mathrm{erfc}}(x) \in (0,2)$ (so it is bounded) and recalling the $O \left( e^{-y^2} \right)$ corrections in , we substitute and into to give (at leading order for large $y$) $$\begin{aligned} \label{e:Guess2} X_{j,k} + \Phi_{j,k} &\mathop{\sim}\limits_{y \to \infty} \frac{1}{2} e^{-\frac{y^2}{2}} \Big( p_{k-1} (y, y) P_j (\infty, y)- p_{j-1} (y, y) P_k (\infty, y) \Big).\end{aligned}$$ Keeping just the leading order polynomial (meaning that we use only the larger of $j$ or $k$), we substitute into to obtain $$\begin{aligned} \label{e:Vm1} {\mathbf{V}}_m &\sim \begin{bmatrix} 0 & h_0 -\frac{e^{-y^2/2}}{2} P_{0}\, p_{0} & -\frac{e^{-y^2/2}}{2} P_{0}\, p_{1} & -\frac{e^{-y^2/2}}{2} P_{0}\, p_{2} & \dots \\ -h_0 +\frac{e^{-y^2/2}}{2} P_{0}\, p_{0} & 0 & h_1 -\frac{e^{-y^2/2}}{2} P_{1}\, p_{1} & -\frac{e^{-y^2/2}}{2} P_{1} \,p_{2} & \dots \\ \frac{e^{-y^2/2}}{2} P_{0}\, p_{1} & -h_1 +\frac{e^{-y^2/2}}{2} P_{1}\, p_{1} & 0 & h_2 -\frac{e^{-y^2/2}}{2} P_{2}\, p_{2} & \dots \\ \frac{e^{-y^2/2}}{2} P_{0}\, p_{2} & \frac{e^{-y^2/2}}{2} P_{1}\, p_{2} & -h_2 +\frac{e^{-y^2/2}}{2} P_{2}\, p_{2} & 0\\ \vdots & \vdots & \vdots & \end{bmatrix},\end{aligned}$$ where $P_0 = P_0 (\infty, y)$ and we have suppressed all the function arguments to save space. Looking at the matrix in , we see that it has identical structure to if we make the following replacements $$\begin{aligned} N& \mapsto \frac{N}{2}, &h_{k}(y) &\mapsto h_{j-1} (y), &p_{j}(y,y) &\mapsto -P_{j} (\infty, y), &p_{k}(y,y) &\mapsto p_{k-1} (y,y),\end{aligned}$$ where we use $j$ to denote the row index and $k$ for the column index. This allows us to use the Pfaffian identity in to conclude that for large $y$ (recalling that $N$ is even) $$\begin{aligned} \label{e:F1lim1} F_{1,N}(y) &\sim \frac{2^{\frac{N}{2} \left( \frac{N}{2} -1 \right)}} {\pi^{N/4}} \prod_{j=0}^{N/2 -1} \frac{1} {(2j)!} \left(\prod_{j=0}^{\frac{N}{2} -1} h_{2j} (y) \right) \left(1 + \sum_{p=1}^{N/2} \; \sum_{{\mathcal{I}}_{2p}} T_{i_1, i_2} T_{i_3,i_4} \cdots T_{i_{2p-1}, i_{2p}} \right),\\ &{\mathcal{I}}_{2p}: 0\leq i_1 \leq i_2 < i_3 \leq i_4 < i_5 \leq i_6 < \dots <i_{2p-1}\leq i_{2p} \leq \frac{N}{2}-1\end{aligned}$$ with $$\begin{aligned} \nonumber T_{i_1, i_2}(y) &= -\frac{1}{2}\dfrac{\prod_{m={i_1}}^{i_2-1} h_{2m+1} (y)}{\prod_{m=i_1}^{i_2} h_{2m} (y)} P_{2i_1}(\infty, y) p_{2i_2}(y, y) e^{-y^2/2} \;, \;\;\;\; i_2 \geq i_1,\\ \label{e:TPfelt} &= -\frac{1}{2} \prod_{m=i_1+1}^{m=i_2} \sqrt{\frac{{\check{\mathsf{R}}}_{2m-1} (y)}{{\check{\mathsf{R}}}_{2m} (y)}} \Psi_{2i_1}(\infty, y) \psi_{2i_2}(y,y) \;, \;\;\;\; i_2 \geq i_1,\end{aligned}$$ where $\Psi_j (\infty, y)$ is from and ${\check{\mathsf{R}}}_k(y) = h_k (y) /h_{k-1}(y)$ is from . In terms of the proof of Proposition \[lem:PfW\], only superficial modifications are needed, with the main change here being that a “little link” from site $j$ to site $j+1$ (as in Figure \[f:LLs\]) now corresponds to a factor of $h_j \delta_{j-1,j} - \frac{P_j (\infty, y) p_j (y,y)}{2} e^{-y^2/2}$. Denoting the prefactor in by $$\begin{aligned} {\mathcal{Z}}_N:= \frac{2^{\frac{N}{2} \left( \frac{N}{2} -1 \right)}} {\pi^{N/4}} \prod_{j=0}^{N/2 -1} \frac{1} {(2j)!} \left(\prod_{j=0}^{\frac{N}{2} -1} h_{2j} (y) \right)\end{aligned}$$ then $$\begin{aligned} \frac{{\mathcal{Z}}_{N-2} {\mathcal{Z}}_{N+2}} {{\mathcal{Z}}_N^2} = \frac{4}{N(N-1)} \frac{h_N (y)}{h_{N-2} (y)} = \frac{4 {\check{\mathsf{R}}}_N {\check{\mathsf{R}}}_{N-1}}{N(N-1)} \end{aligned}$$ and so using $$\begin{aligned} \label{e:log1} \ln \left( \frac{4 {\check{\mathsf{R}}}_N {\check{\mathsf{R}}}_{N-1}}{N(N-1)} \right) \sim - 2 \frac{q(s)^2} {N^{2/3}}+ o \left( N^{-2/3} \right).\end{aligned}$$ Now, analogously to we assume $$\begin{aligned} \label{e:beta1scaling} \ln {\cal Z}_N(y) \mathop{\rightarrow}\limits_{N\to \infty} f(\sqrt{2} N^{1/6} (y-\sqrt{2N})) \;,\end{aligned}$$ and then with $y= \sqrt{2N} + \frac{s}{\sqrt{2} N^{1/6}}$ we have $$\begin{aligned} \label{e:log2} \ln {\cal Z}_{N-2} + \ln {\cal Z}_{N+2} - 2 \ln {\cal Z}_{N} = \frac{4}{N^{2/3}} f''(s) + {o} \left( N^{-2/3} \right) \;.\end{aligned}$$ Equating and we have (to leading order) $$\begin{aligned} f''(s) = - \frac{1}{2} q^2(s)\end{aligned}$$ identically with the $\beta=4$ case in . Therefore, we have $$\begin{aligned} \lim_{N\to \infty} {\cal Z}_N \left(y = \sqrt{2N} + \frac{s}{\sqrt{2} N^{1/6}} \right)= \exp{\left[-\frac{1}{2} \int_s^{\infty} (x- s) q^2(x) dx\right] } \;,\end{aligned}$$ which is the first factor in . For the right-most factor in we use the known asymptotic behaviour , with $y= \sqrt{2N} + \frac{s}{\sqrt{2} N^{1/6}}$, to find $$\begin{aligned} \psi_{N- 2k} (y, y) \sim 2^{1/4} N^{-1/12} q\left( s+ 2 \frac{k}{N^{1/3}} \right).\end{aligned}$$ For $\Psi_{N -2k} (\infty, y)$ in , we recall that this is an integral over the entire domain, so the integral will be dominated by the behaviour of the integrand in the bulk regions. From and we have the large $y$ behaviour $$\begin{aligned} \psi_{N} (x, y) &\mathop{\sim}\limits_{y\to \infty} \frac{H_N (x)} {\pi^{1/4} 2^{N/2} \sqrt{\Gamma (N+1)}} e^{-x^2/2}\end{aligned}$$ and using the recursion relations [@AbraSteg1972 Chapter 22] $$\begin{aligned} \label{e:Hermids} H_{j+1} (x) = 2x H_j - H'_j (x), \qquad H'_j(x) = 2j H_{j-1} (x)\end{aligned}$$ we can show $$\begin{aligned} \label{e:Hermint} \int_{-\infty}^{\infty} e^{-\frac{z^2}{2}} H_j (z) dz &= \left\{ \begin{array}{cl} 2^{j+ 1/2} \Gamma \left( \frac{j+1}{2} \right),& \mbox{$j$ even},\\ 0,& \mbox{$j$ odd}, \end{array}\right.\end{aligned}$$ where the second line follows since Hermite polynomials of odd degree are odd functions. So we have (recalling that $N$ is even) $$\begin{aligned} \label{e:PsiLim} \Psi_{N- 2k} (\infty, y) &\mathop{\sim}\limits_{y\to \infty} \frac{2^{3/4}} {N^{1/4}},\end{aligned}$$ and with we obtain $$\begin{aligned} T_{\frac{N}{2} -k_1, \frac{N}{2} -k_2} \left( y= \sqrt{2N} + \frac{s}{\sqrt{2} N^{1/6}} \right) \sim - \frac{1}{N^{1/3}} q\left( s+ 2 \frac{k_2}{N^{1/3}} \right).\end{aligned}$$ Then we find ourselves at the analogue of $$\begin{aligned} \sum_{{\mathcal{I}}_{2p}} T_{i_1, i_2} T_{i_3,i_4} \cdots T_{i_{2p-1}, i_{2p}} &\sim \frac{(-1)^p}{N^{\frac{p}{3}}}\sum_{{\mathcal{K}}_{p}} q\left(s + 2 \frac{k_2}{N^{\frac{1}{3}}} \right) q\left(s + 2 \frac{k_4}{N^{\frac{1}{3}}} \right) \cdots q\left(s + 2 \frac{k_{2p}} {N^{\frac{1}{3}}} \right) \;,\end{aligned}$$ where we denote $$\begin{aligned} \label{e:Kp} {\mathcal{K}}_{p}: \frac{N}{2} \geq k_2 > k_4 > k_6> \dots > k_{2p -2} > k_{2p} \geq 1\; .\end{aligned}$$ \[Note that, in contrast to $\mathcal{K}_{2p}$ in , ${\mathcal{K}}_{p}$ contains only the even indices. In the $\beta=4$ case, both even and odd indices contributed factors of $q$, as can be seen in . But here the odd indices are attached to the integrals $\Psi_{N-2k_{\mathrm{odd}}}$, and only contribute factors of $N$ and $2$ as per .\] Changing variables $v_j = 2k_j /N^{1/3}$ gives $$\begin{aligned} \nonumber &\sum_{{\mathcal{I}}_{2p}} T_{i_1, i_2} T_{i_3,i_4} \cdots T_{i_{2p-1}, i_{2p}} \\ \nonumber & \sim \frac{(-1)^p}{2^{p}} \int_0^{\infty} q(s+v_{2p}) dv_{2p }\int_{v_{2p}}^{\infty} q(s+v_{2p-2}) dv_{2p-2} \cdots \int_{v_6}^{\infty} q(s+v_4) dv_4 \int_{v_4}^{\infty} q(s+v_2) dv_2 \\ &= \frac{(-1)^p}{p!} \left(\frac{1}{2} \int_s^\infty dx \, q(x)\right)^{p},\end{aligned}$$ where, in the final line, we have removed the ordering from the integration variables since the integrand is symmetric in the $v_j$’s. Summing over $p$ and taking the limit we have $$\begin{aligned} \nonumber \lim_{N \to \infty} \left[1 + \sum_{p=1}^{N/2} \sum_{{\mathcal{I}}_{2p} } T_{i_1, i_2} T_{i_3,i_4} \cdots T_{i_{2p-1}, i_{2p}}\right] &= \sum_{p=0}^\infty \frac{1}{p!} \left(-\frac{1}{2} \int_s^\infty dx \, q(x)\right)^{p}\\ &=\exp\left(-\frac{1}{2} \int_s^\infty dx \, q(x) \right),\end{aligned}$$ which is the second factor in . Lastly, we note that the Pfaffian identity that we used here for $\beta=1$ and for $\beta=4$ above will hold more generally, for all anti-symmetric matrices of the form $$\begin{aligned} {\mathbf{M}}= {\mathbf{T}}+ {\mathbf{B}},\end{aligned}$$ where ${\mathbf{T}}$ has upper triangular elements $\delta_{j+1,k} t_{j,k}$ and ${\mathbf{B}}= [b_{j,k}]_{j,k=1, \dots, 2N}$ has upper-triangular entries $b_{j,k}= f_j g_k$ for some functions $f$ and $g$. In which case, $$\begin{aligned} {\mathrm{Pf}\,}{\mathbf{M}}&= \left(\prod_{j=0}^{N-1} t_{2j+1} \right) \left(1 + \sum_{p=1}^N \; \sum_{{\mathcal{I}}_{2p}} L_{i_1, i_2} L_{i_3,i_4} \cdots L_{i_{2p-1}, i_{2p}} \right) \;,\end{aligned}$$ where $$\begin{aligned} L_{i_1, i_2} = \dfrac{\prod_{m={i_1}}^{i_2-1} t_{2m+2}}{\prod_{m=i_1}^{i_2} t_{2m+1}} f_{2i_1}\, g_{2i_2+1}, \;\;\;\; i_2 \geq i_1 \;,\end{aligned}$$ with the summation indices ${\mathcal{I}}_{2p}$ defined in . Conclusions and perspectives {#s:conclusion} ============================ In this paper, we have revisited the computation of the cumulative distribution function of the largest eigenvalue in the classical ensembles of RMT, namely the GOE and the GSE, using the techniques of skew-orthogonal polynomials, thus extending the approach of Nadal and Majumdar [@NadaMaju2011] developed for the GUE. By adapting the method of Refs. [@AdleForrNagavanMoer2000; @AdlevanMoer2002], we have constructed explicitly these (semi-classical) skew-orthogonal polynomials in terms of the so-called “Nadal-Majumdar” orthogonal polynomials introduced in the case of the GUE. This construction involves some non-trivial Pfaffians, which we have related to “overlapping Pfaffians”, studied originally by Knuth [@Knuth1996]. We were then able to carry out the asymptotic analysis of these skew-orthogonal polynomials and of their norms to obtain the well known Tracy-Widom distributions, using a method which is quite different from the original one [@TracWido1996], and also different from the more recent one obtained via the so-called stochastic Airy operator [@bloemendal2013limits]. This relied on a certain Pfaffian identity, the most general statement of which is given at the end of Section \[s:F1asympt\]. As discussed in Section \[sec:PartFns\], it is known that “Pfaffian” Janossy densities (of which our $\beta=1$ and $\beta=4$ densities are examples) have $n$-point correlation functions given by Pfaffians. These correlation functions can be calculated via standard techniques (see [@Mehta2004; @Forrester2010]) — these calculations will be presented in a follow-up work [@mays2020prep]. By using the skew-orthogonal polynomials constructed in the present work, this will allow us to analyze the density of states near the largest eigenvalue and the statistics of the gap between the two largest eigenvalues in the GSE and the GOE. These quantities are particularly interesting in the challenging case of GOE since they naturally enter into the computation of physical observables in the spherical Sherrington-Kirkpatrick model of mean-field spin glasses [@fyodorov2015large]. A.M. would like to thank Michael Wheeler, Peter Forrester and Shi-Hao Li for helpful discussions. A.M. and A.P. are supported by the Australian Research Council (ARC) Centre of Excellence for Mathematical and Statistical Frontiers (ACEMS), ARC Grant No. CE140100049. A.M. thanks LPTMS for their hospitality during a visit supported by CNRS. Reminder on the classical ensembles of RMT: GOE, GUE and GSE {#sec:classical} ============================================================ For self-consistency, we recall here the definition of the classical ensembles of RMT studied in this paper: - [The *Gaussian Orthogonal Ensemble* (GOE) is the set of $N\times N$ real symmetric matrices $$\begin{aligned} \label{d:GOE} {\mathbf{M}}= \frac{{\mathbf{Y}}+ {\mathbf{Y}}^T}{2}, \end{aligned}$$ where ${\mathbf{Y}}$ contains standard normally distributed elements $y_{j,k} \sim \mathcal{N}[0,1]$ resulting in the matrix PDF proportional to $e^{- ({\mathrm{Tr}\,}{\mathbf{M}}^2)/2}$ which is invariant under orthogonal conjugation ${\mathbf{M}}\mapsto {\mathbf{O}}^T {\mathbf{M}}{\mathbf{O}}$.]{} - [The *Gaussian Unitary Ensemble* (GUE) is the set of complex Hermitian matrices $$\begin{aligned} {\mathbf{M}}= \frac{{\mathbf{Y}}+ {\mathbf{Y}}^{\dagger}}{2} \end{aligned}$$ with real independent Gaussian components $y_{j,k} \sim \mathcal{N}[0,\frac{1}{\sqrt{2}}] + i \mathcal{N}[0,\frac{1}{\sqrt{2}}]$ giving a matrix PDF proportional to $e^{- {\mathrm{Tr}\,}{\mathbf{M}}^2}$ which is invariant under unitary conjugation ${\mathbf{M}}\mapsto {\mathbf{U}}^{\dagger} {\mathbf{M}}{\mathbf{U}}$.]{} - [The *Gaussian Symplectic Ensemble* (GSE) is defined similarly for normally distributed quaternionic entries. We provide in Appendix \[app:Quats\] some definitions related to quaternions, however this will not be required for understanding the current work, as we use the equivalent $2\times 2$ representation of quaternions $$\begin{aligned} \label{d:quat}\begin{bmatrix} a_1+ ib_1 &a_2 +i b_2\\ -a_2+ ib_2 &a_1 -i b_1 \end{bmatrix} \qquad (a_1, a_2, b_1, b_2 \in \mathbb{R}). \end{aligned}$$ The ensemble is then the set of $2N\times 2N$ matrices, $$\begin{aligned} \label{d:GSE} {\mathbf{M}}= \frac{{\mathbf{Y}}+ {\mathbf{Y}}^{\dagger}} {2}, \end{aligned}$$ where each $2\times 2$ block of ${\mathbf{Y}}$ is of the form with each independent real component normally distributed $a_1, b_1, a_2, b_2 \sim \mathcal{N}[0, \frac{1}{2} ]$. The matrix PDF is then proportional to $e^{- {\mathrm{Tr}\,}{\mathbf{M}}^2}$, which is invariant under symplectic conjugation, that is conjugation by a unitary matrix ${\mathbf{M}}\mapsto {\mathbf{U}}^{\dagger} {\mathbf{M}}{\mathbf{U}}$, with the restriction that $$\begin{aligned} {\mathbf{U}}{\mathbf{Z}}_{N} {\mathbf{U}}^T = \pm {\mathbf{Z}}_{N} \;, \end{aligned}$$ where $$\begin{aligned} \label{d:Zmat} {\mathbf{Z}}_N:= \begin{bmatrix} 0& 1\\ -1& 0 \end{bmatrix} \otimes {\mathbf{I}}_N = \begin{bmatrix} 0& 1& 0 & 0&0& &\\ -1& 0& 0 & 0&0& &\dots\\ 0&0&0&1& 0 &&\\ 0&0&-1&0& 0& &\\ &\vdots &&&&&\ddots \end{bmatrix},\end{aligned}$$ and ${\mathbf{I}}_{N}$ is the $N\times N$ identity matrix. ]{} Quaternions {#app:Quats} =========== Here we provide a brief overview of some definitions related to quaternions. A *quaternion* is typically written in the form $$\begin{aligned} q= q_0 + iq_1 +j q_2 +kq_3, \qquad q_0, q_1, q_2, q_3 \in \mathbb{R},\end{aligned}$$ where $i, j$ and $k$ are the quaternionic generalization of the imaginary unit and obey Hamilton’s famous bridge equation $$\begin{aligned} i^2 = j^2 = k^2 = ijk = -1,\end{aligned}$$ which defines their algebraic behaviour. (Note that we restrict the coefficients $q_j$ to be real — these are called *real quaternions* by other authors [@Mehta2004; @Forrester2010], to contrast with the more general case where the coefficients are complex. We have no need of the more general case in this work.) A more convenient representation of the same algebra is given by mapping the quaternions to the $2\times 2$ complex matrices $$\begin{aligned} \label{d:quat2x2} q= \begin{bmatrix} a && b\\ -\bar{b} && \bar{a} \end{bmatrix},\end{aligned}$$ where $a:= q_0 + iq_1 \in \mathbb{C}$ and $b= q_2+ i q_3 \in \mathbb{C}$. This representation is equivalent to a linear combination of the Pauli spin matrices (see, for example, [@Forrester2010]). The analogue of complex conjugation for quaternions is $$\begin{aligned} q^*= q_0 - iq_1 -j q_2 -kq_3 \qquad \longleftrightarrow \qquad q^{\dagger} = \begin{bmatrix} \bar{a} & & -b\\ \bar{b} & & a \end{bmatrix},\end{aligned}$$ where we see that $q^*$ is the same as the Hermitian conjugate of the $2\times 2$ matrix representation. A matrix of quaternionic entries is said to be *self-dual* if $$\begin{aligned} {\mathbf{Q}}:= [q_{j,k}] = [q^*_{k,j}] =: {\mathbf{Q}}^*,\end{aligned}$$ or equivalently, if the matrix of $2\times 2$ quaternionic blocks is Hermitian. The Gaussian Symplectic Ensemble in is then equivalently defined as the set of $2N \times 2N$ Hermitian matrices $[q_{j,k}]$ with entries $$\begin{aligned} q_{j,j} = \begin{bmatrix} x_{j,j}& 0\\ 0 & x_{j,j} \end{bmatrix}, \qquad q_{j,k} = \begin{bmatrix} z_{j,k}& w_{j,k}\\ -\bar{w}_{j,k} & \bar{z}_{j,k} \end{bmatrix}, \qquad (k>j)\end{aligned}$$ with $$\begin{aligned} x_{j,j} \mathop{\sim}\limits_d \mathcal{N}\left[ 0, \frac{1}{2} \right] \in \mathbb{R} \quad \mbox{and} \quad z_{j,k}, w_{j,k} \mathop{\sim}\limits_d \mathcal{N}\left[ 0, \frac{1}{2\sqrt{2}} \right] + i \mathcal{N}\left[ 0, \frac{1}{2\sqrt{2}} \right] \in \mathbb{C}.\end{aligned}$$ In we write the matrix PDF for $\beta=1,2$ and $4$, however for $\beta=4$ this requires the use of the *quaternion trace*, which for a quaternionic matrix $\hat{{\mathbf{Q}}}_{N\times N}$, is $$\begin{aligned} \label{e:qTr} \mathrm{qTr}\; \hat{{\mathbf{Q}}}_{N\times N} = \sum_{j=1}^N (q_0)_{j,j} = \frac{1}{2} {\mathrm{Tr}\,}{\mathbf{Q}}_{2N\times 2N},\end{aligned}$$ where in the second equality this is the usual matrix trace and ${\mathbf{Q}}_{2N\times 2N}$ is the equivalent matrix with entries given by the $2\times 2$ matrices . A related concept is the *quaternion determinant*, which is defined for self-dual quaternion matrices by $$\begin{aligned} \label{def:qdet} {\mathrm{qdet}}\; \hat{{\mathbf{Q}}}_{N\times N} :=\sum_{P\in S_N}(-1)^{N-|c(P)|}\prod_{(ab\cdots s)\in c(P)} \; (q_{ab}q_{bc}\cdots q_{sa})_{0},\end{aligned}$$ where $c(P)$ is the set of cycles of the permutation $P$, and the subscript ${(\dots)}_{0}$ denotes that one takes the scalar part $q_{0}$ of the resulting quaternion. As with the quaternion trace in , there is a relationship between the quaternion determinant and the usual determinant, given by $$\begin{aligned} \label{e:qdet} \left( {\mathrm{qdet}}\; \hat{{\mathbf{Q}}}_{N\times N} \right)^2 = \det {\mathbf{Q}}_{2N\times 2N},\end{aligned}$$ where again the matrix on the right is the equivalent complex matrix made of the $2\times 2$ blocks . Pfaffians {#a:Pfaffs} ========= Pfaffians are very closely related to quaternion determinants , however they do not require any of the quaternionic technicalities, so we prefer to use Pfaffians in this work. A brief historical survey on the topic is provided in [@Knuth1996 §6] \[Pfaffian\] Let ${\mathbf{M}}=[m_{j, k}]_{j,k =1,... ,2N}$, where $m_{j, k}=-m_{k, j}$, so that ${\mathbf{M}}$ is an anti-symmetric matrix of even size. Then the *Pfaffian* of ${\mathbf{M}}$ is defined by $$\begin{aligned} \nonumber \mathrm{Pf}\; {\mathbf{M}}&=\sum^*_{P\in S_{2N} \atop P(2j)>P(2j -1)} \varepsilon (P) m_{P(1),P(2)} m_{P(3),P(4)}\cdot\cdot\cdot m_{P(2N-1), P(2N)}\\ \nonumber &=\frac{1}{N!} \sum_{P\in S_{2N} \atop P(2j) >P(2j -1)} \varepsilon (P) m_{P(1),P(2)} m_{P(3),P(4)}\cdot\cdot\cdot m_{P(2N-1), P(2N)}\\ \label{def:Pf} &=\frac{1}{2^N N!}\sum_{P\in S_{2N}} \varepsilon (P) m_{P(1), P(2)} m_{P(3), P(4)}\cdot\cdot\cdot m_{P(2N-1),P(2N)},\end{aligned}$$ where $S_{2N}$ is the group of permutations of $2N$ letters and $\varepsilon (P)$ is the signature of the permutation $P$. The \* above the first sum indicates that the sum is over distinct terms only (that is, all permutations of the pairs of indices are regarded as identical). Note that in the second equality of (\[def:Pf\]) the factors of $2$ are associated with the restriction $P(2j)>P(2j -1)$ while the factorial is associated with counting only distinct terms \[$N!$ is the number of ways of arranging the $N$ pairs of indices $ P(2l-1), P(2l)$\]. Pfaffians can be calculated via a version of Laplace expansion, however the Pfaffian minors ${\mathbf{M}}^{(j,k)}$ that one needs to calculate are obtained by blocking out both the $j$th and $k$th row and the $j$th and $k$th column. The definition of a Pfaffian is very close to that of a determinant, and for the matrix ${\mathbf{M}}$ (antisymmetric of size $2N \times 2N$), they are related by $$\begin{aligned} \label{e:PfDet2a} ({\mathrm{Pf}\,}{\mathbf{M}})^2 = \det {\mathbf{M}}.\end{aligned}$$ The clear similarity between and highlights the equivalent nature of quaternion determinants and Pfaffians; they are specifically connected via the matrix ${\mathbf{Z}}_N$ in , where we note that $$\begin{aligned} \label{e:detZ} {\mathrm{Pf}\,}({\mathbf{Z}}_{N}) = \det ({\mathbf{Z}}_{N}) = 1.\end{aligned}$$ With ${\mathbf{Z}}_N$ we have $$\begin{aligned} \label{e:Pfqdet1} {\mathrm{Pf}\,}( {\mathbf{M}}) &= {\mathrm{qdet}}({\mathbf{M}}{\mathbf{Z}}_{N} ) = {\mathrm{qdet}}( {\mathbf{Z}}_{N} {\mathbf{M}}),\\ \label{e:Pfqdet2} {\mathrm{Pf}\,}( {\mathbf{M}}) &= (-1)^N {\mathrm{qdet}}({\mathbf{M}}{\mathbf{Z}}_{N}^T ) = (-1)^N {\mathrm{qdet}}( {\mathbf{Z}}_{N}^T {\mathbf{M}}).\end{aligned}$$ We will also have need of the identity [@deBruijn1955] $$\begin{aligned} \label{e:PfBAB} {\mathrm{Pf}\,}({\mathbf{B}}{\mathbf{M}}{\mathbf{B}}^T)= \det ({\mathbf{B}}) {\mathrm{Pf}\,}({\mathbf{M}}),\end{aligned}$$ where ${\mathbf{B}}$ is a general $2N \times 2N$ matrix. Pfaffians and elementary row/column operations ---------------------------------------------- Given the similarity between Pfaffians and determinants, it is not surprising that Pfaffians have similar behaviour to determinants, particularly for elementary row and column operations. Recall the determinant identity $$\begin{aligned} \label{e:scaledet} \alpha \det \begin{bmatrix} {{\mathbf m}}_1& {{\mathbf m}}_2 & \dots & {{\mathbf m}}_N \end{bmatrix} = \det \begin{bmatrix} {{\mathbf m}}_1& \dots & {{\mathbf m}}_{j-1} & \alpha {{\mathbf m}}_j & {{\mathbf m}}_{j+1} & {{\mathbf m}}_N \end{bmatrix},\end{aligned}$$ for a general $N\times N$ matrix, where $j$ is any integer from $1$ up to $N$. That is, the determinant can be scaled by scaling any column (or row) of the matrix. There is an equivalent identity for Pfaffians, however when we scale a column/row we also scale its corresponding row/column by the same factor. Explicitly, with ${\mathbf{M}}$ as above, we have $$\begin{aligned} \label{e:scalePf} \alpha \; {\mathrm{Pf}\,}{\mathbf{M}}&= {\mathrm{Pf}\,}\begin{bmatrix}[ccc|c|ccc] {} &&& \alpha m_{1,j} & {} \\ {} & * && \vdots & &*& \\ {} &&& \alpha m_{j-1,j} & {} \\ \hline -\alpha m_{1,j} & \dots & -\alpha m_{j-1,j} & 0& \alpha m_{j, j+1} & \dots & \alpha m_{j, 2N}\\ \hline &&&- \alpha m_{j, j+1}&&&\\ &* && \vdots&& * &\\ &&&- \alpha m_{j, 2N}&&& \end{bmatrix},\end{aligned}$$ where the “$*$” represents that the remaining matrix elements are unchanged. Note this row and column scaling preserves the anti-symmetry of the matrix. If we take $\alpha \mapsto \sqrt{\alpha}$ then it can be seen that this scaling is consistent with and . Similarly, we have an analogue of the identity $$\begin{aligned} \det \begin{bmatrix} {{\mathbf m}}_1& {{\mathbf m}}_2 & \dots & {{\mathbf m}}_N \end{bmatrix} = \det \begin{bmatrix} {{\mathbf m}}_1& \dots & {{\mathbf m}}_{j-1} & {{\mathbf m}}_j +\alpha {{\mathbf m}}_k & {{\mathbf m}}_{j+1} & {{\mathbf m}}_N \end{bmatrix}\end{aligned}$$ where the determinant is unchanged by adding to any column/row a scalar multiple of any other column/row. The Pfaffian analogue is obtained by adding a multiple of a column (or row) to another column (or row), and adding the same multiple of the same row (or column) to the matching row (or column), $$\begin{aligned} \label{e:Pfrowadd} {\mathrm{Pf}\,}{\mathbf{M}}= {\mathrm{Pf}\,}\widehat{{\mathbf{M}}}_{j,k; \alpha},\end{aligned}$$ where the matrix $\widehat{{\mathbf{M}}}_{j,k; \alpha}$ is identical to ${\mathbf{M}}$ except for column and row $j$, which equal $$\begin{aligned} Col(j)\mapsto Col(j) + \alpha Col(k), \qquad Row(j)\mapsto Row(j) + \alpha Row(k)\end{aligned}$$ Note that anti-symmetry is preserved, and we again see that this is consistent with . Lastly, we have $$\begin{aligned} \det \begin{bmatrix} {{\mathbf m}}_1& {{\mathbf m}}_2 & \dots & {{\mathbf m}}_N \end{bmatrix} = - \det \begin{bmatrix} {{\mathbf m}}_1& \dots & {{\mathbf m}}_{j-1} & {{\mathbf m}}_k & {{\mathbf m}}_{j+1} \dots & {{\mathbf m}}_{k-1} & {{\mathbf m}}_j & {{\mathbf m}}_{k+1} & \dots & {{\mathbf m}}_N, \end{bmatrix}\end{aligned}$$ where determinants pick up a factor of $(-1)$ for each column/row swap. The analogous result for Pfaffians is more complicated $$\begin{aligned} \label{e:Pfrowswap} {\mathrm{Pf}\,}{\mathbf{M}}= - {\mathrm{Pf}\,}\widehat{{\mathbf{M}}}_{j \leftrightarrow k},\end{aligned}$$ where the matrix $\widehat{{\mathbf{M}}}_{j \leftrightarrow k}$ is identical to ${\mathbf{M}}$ except that $$\begin{aligned} Col(j) \leftrightarrow Col(k) \quad \mbox{and} \quad Row(j) \leftrightarrow Row(k),\end{aligned}$$ where the swaps happen in succession. This again preserves anti-symmetry and is consistent with . Pfaffians and perfect matchings {#a:PfsPMs} ------------------------------- In order to prove Proposition \[lem:PfW\] we will use an expression equivalent to in terms of perfect matchings and link patterns. Expressions for Pfaffians in terms of perfect matchings have been known for a long time, and they are discussed in many places — we refer to [@Knuth1996; @Rote2001]. A *perfect matching* $\mu$ is a set of links between $2N$ sites, where each site is connected to exactly one other site. Diagrammatically, this is expressed as a *link diagram*, and most easily seen via an example: let $$\begin{aligned} \label{e:PMeg2} \mu= \{(2,3), (5,1), (4, 6)\}\end{aligned}$$ and the link diagram is given in Figure \[f:PMeg\]. The sign $\varepsilon(\mu)$ of the perfect matching is given by$(-1)^{\# \chi}$, where $\# \chi$ is the number of crossings in the link pattern — for the example in we have $\varepsilon(\mu) = (-1)^1$. We denote the set of all perfect matchings on $2N$ sites by $M_{2N}$, and the number of perfect matchings is $$\begin{aligned} \label{e:NumPMs} |M_{2N}| = (2N -1)!! = (2N -1) \cdot (2N-3) \cdots (3) \cdot (1),\end{aligned}$$ since there are $2N-1$ sites for the first site to pair with, then $2N-3$ sites for the second site to pair with, etc. (-0.5,0)–(5.5,0); at (0,0) [$1$]{}; at (1,0) [$2$]{}; at (2,0) [$3$]{}; at (3,0) [$4$]{}; at (4,0) [$5$]{}; at (5,0) [$6$]{}; (0,0) to\[out=60,in=180\] (2,0.75) to\[out=0,in=120\] (4,0); (1,0) to\[out=60,in=180\] (1.5,0.33) to\[out=0,in=120\] (2,0); (3,0) to\[out=60,in=180\] (4,0.5) to\[out=0,in=120\] (5,0); (Note that usually a perfect matching is defined as a set of edges on a graph such that every vertex is included exactly once. However this characterization will not be useful for us, and for a complete graph it is equivalent to the definition we use in terms of link patterns.) The connection to Pfaffians comes from the fact that there is a bijection from $M_{2N}$ to a subset of $S_{2N}$, the set of permutations of $\{1, \dots, 2N \}$. The bijection is found by taking a perfect matching $\mu$ and ordering the components of each pair such that $\mu= \{ (\mu_{j,L}, \mu_{j,R}) \}_{j= 1, \dots, N}$, where $\mu_{j,L}$ and $\mu_{j,R}$ are respectively the left and right terminals of link $j$. (In graph parlance, this creates a *directed link pattern*, where all links point from, say, left to right.) Then we institute an ordering between the pairs according to some scheme (say, that $\mu_{j,L}< \mu_{j+1 ,L}$), which results in a unique representative ordered set of pairs for each perfect matching. Then, by removing the pairing, we obtain a unique $s\in S_{2N}$. For the example in we find $$\begin{aligned} \label{e:PMeg3} M_{6} \ni \{(2,3), (5,1), (4, 6)\} = \{ (1, 5), (2,3), (4, 6)\} \mapsto (1,5,2,3,4,6) \in S_6.\end{aligned}$$ The reverse mapping $S_{2N} \supset \hat{S}_{2N} \to M_{2N}$ is clear: $s\in \hat{S}_{2N}$ is a permutation of $1, \dots, 2N$ such that $s(2j-1)< s(2j)$ and $s(2j-1)< s(2k-1)$ for $j<k$. In order for this mapping to make sense, we need $\varepsilon(\mu)= \varepsilon(s)$, that is the number of crossings in the perfect matching $\mu$ must be the same as the sign of the permutation $s$, given by $(-1)^\tau$ where $\tau$ is the number of transpositions required to return $s$ to the identity permutation. This can be shown by first noting that the identity permutation gives a link pattern with no crossings, and then that a crossing can always be removed by a single transposition, while a link pattern with no crossings can be transformed to the identity by an even number of transpositions. The conditions defining $\hat{S}_{2N}$ are the same restrictions on $S_{2N}$ as those implied by the first line of , and so we have the following equivalent expression for the Pfaffian $$\begin{aligned} \label{e:PfPerfMatch} {\mathrm{Pf}\,}{\mathbf{M}}= \sum_{\mu\in M_{2N}} \varepsilon (\mu)\; m_{i_1, j_1} m_{i_2, j_2}\cdot\cdot\cdot m_{i_N, j_{N}},\end{aligned}$$ where $M_{2N}$ is the set of all perfect matchings $\mu= \{ (i_1, j_1), \dots, (i_N, j_N) \}$ on $2N$ sites, and $\varepsilon (\mu)$ is the sign of the perfect matching, or equivalently, the sign of the corresponding permutation. Iterative construction of the first few skew-orthogonal polynomials {#s:iterative} =================================================================== In this Appendix, we iteratively construct the first few skew-orthogonal polynomials defined in Eqs. (\[d:4sorthog\]) and (\[d:1sorthog\]). First for $\beta=4$, by monicity, we must have $Q_0 (\lambda) = 1$ and by we can assume that $Q_1 (\lambda) = \lambda$, then we use the skew-inner product relations to iteratively solve for the higher degree polynomials, so the first four skew-orthogonal polynomials are $$\begin{aligned} Q_0 (\lambda, y)& = 1, \qquad Q_1 (\lambda, y) = \lambda, \qquad Q_2 (\lambda, y)= \lambda^2 +b \lambda + \frac{1 -2yb} {4},\\ Q_3 (\lambda, y)&= \lambda^3 - 3 \frac{1- 2y b} {4} \lambda -b \frac{1+2 y^2} {2}\end{aligned}$$ \[where we used for $Q_3 (\lambda, y)$\] with normalizations $$\begin{aligned} q_0 (y)&:= \langle Q_0 , Q_1 \rangle_{4}^y = \frac{\sqrt{\pi}}{4 \sqrt{2}} \;{\mathrm{erfc}}(-\sqrt{2} y) = \frac{e^{-2y^2}} {4 b},\\ q_1 (y)&:= \langle Q_2 , Q_3 \rangle_{4}^y = \frac{1} {64} \left( 3 \sqrt{2 \pi} {\mathrm{erfc}}(- \sqrt{2} y) -2 e^{-2 y^2} y (9 +4 y^2) - 4 e^{-2 y^2} ( 2 +y^2) b \right),\end{aligned}$$ where $$\begin{aligned} b= \frac{\sqrt{2} e^{-2y^2}} {\sqrt{\pi} (1+{\mathrm{erf}}(\sqrt{2} y))}= \frac{\sqrt{2} e^{-2y^2}} {\sqrt{\pi} {\mathrm{erfc}}(- \sqrt{2} y)}.\end{aligned}$$ For $\beta=1$, again by monicity and we have, $R_0(\lambda) =1, R_1(\lambda) = \lambda$ and using the relations we can obtain the first four polynomials $$\begin{aligned} R_0(\lambda, y)& =1, \qquad R_1(\lambda) = \lambda,\\ R_2 (\lambda, y)&= \lambda^2 + \lambda \frac{c}{\sqrt{\pi}} \left(2 e^{-y^2 /2} + \sqrt{2 \pi} y \: {\mathrm{erfc}}(-y/ \sqrt{2}) \right) + c e^{y^2/2} {\mathrm{erfc}}(-y) -1,\\ R_3 (\lambda, y)&= \lambda^3 +\lambda c \left( \frac{2 ye^{-y^2/2}} {\sqrt{\pi}} - \frac{2}{c} - e^{y^2/2} \: {\mathrm{erfc}}(-y) + y^2 \sqrt{2 }\: {\mathrm{erfc}}(-y/ \sqrt{2} ) \right) - \frac{2 c e^{-y^2/2}} {\sqrt{\pi}}\end{aligned}$$ with $$\begin{aligned} c= \left( 2e^{y^2/2} {\mathrm{erfc}}(-y) - \sqrt{2} {\mathrm{erfc}}(-y/ \sqrt{2}) \right)^{-1}\end{aligned}$$ and $$\begin{aligned} r_0(y) &=\frac{\sqrt{\pi}} {2} \left( {\mathrm{erfc}}(-y) - \frac{e^{-y^2/2}} {\sqrt{2}} {\mathrm{erfc}}(- y/ \sqrt{2}) \right) = \frac{\sqrt{\pi} e^{-y^2/2}}{4 c}\\ \nonumber r_1 (y) &= \frac{\sqrt{\pi }}{8} \text{erfc}(-y) -\frac{y e^{-y^2}}{4} - c \left( \frac{e^{-\frac{3y^2}{2}}}{\sqrt{\pi }} + \frac{y^2 \sqrt{\pi}}{2 \sqrt{2}} {\mathrm{erfc}}(-y) {\mathrm{erfc}}\left(-\frac{y}{\sqrt{2}}\right)\right.\\ &\left. +\frac{y e^{-\frac{y^2}{2}}}{2} {\mathrm{erfc}}(-y) +\frac{y e^{-y^2} {\mathrm{erfc}}\left(-\frac{y}{\sqrt{2}}\right)}{\sqrt{2}}- \frac{\sqrt{\pi} e^{\frac{y^2}{2}} }{4} {\mathrm{erfc}}(-y)^2 \right).\end{aligned}$$ Skew-orthogonal polynomials for $\beta=4$ {#a:beta4} ========================================= For the ease of the reader, we try to use the same notation as in [@Forrester2010 §6.2 & §6.4], where the case $y= \infty$ is discussed in detail. Also note that all the quantities in this section depend on $y$, however we will suppress the explicit notation of such, to save space. The goal is to write the $\beta=4$ skew-orthogonal polynomials $\{ Q_j \}$, defined by and , in terms of the polynomials orthogonal with respect to the inner product , the NM polynomials $p_j$ which obey the relations –. However, as discussed in Section \[s:SOPs4\] we will instead use the modified skew-inner product , and look for polynomials $\{\tilde{Q}_j \}$ that obey the relations and , up to the invariance . Since the orthogonal polynomials form a complete set we can find coefficients $\tilde{\alpha}_{j,k}$ such that $$\begin{aligned} \label{e:pQ} \tilde{Q}_j = p_j + \tilde{\alpha}_{j,j-1}p_{j-1} + \dots + \tilde{\alpha}_{j,1}p_1 + \tilde{\alpha}_{j,0} p_0, \qquad \tilde{\alpha}_{j,k}\in \mathbb{C}.\end{aligned}$$ Recall that the tilde $\tilde{~}$ means that the quantity is associated with this modified skew-inner product. From monicity and we have $$\begin{aligned} \label{e:alpharules} \tilde{\alpha}_{j,j}=1, \qquad \tilde{\alpha}_{2j+1, 2j}=0.\end{aligned}$$ We can write in the matrix form $$\begin{aligned} {\mathbf{\tilde{Q}}}= {\mathbf{\tilde{X}}}{{\mathbf p}}\end{aligned}$$ where $$\begin{aligned} \label{d:tQp} {\mathbf{\tilde{Q}}}= \left[ \begin{array}{c} \tilde{Q}_0 \\ \tilde{Q}_1\\ \vdots \end{array} \right], \qquad {{\mathbf p}}= \left[ \begin{array}{c} p_0 \\ p_1\\ \vdots \end{array} \right]\end{aligned}$$ $$\begin{aligned} \label{d:Xmat} {\mathbf{\tilde{X}}}= \left[ \begin{array}{cccccc} 1 & 0 & 0 & 0 & 0 & \cdots\\ 0 & 1 & 0 & 0 & 0 & \cdots\\ \tilde{\alpha}_{2, 0} & \tilde{\alpha}_{2, 1} & 1 & 0 & 0 & \cdots\\ \tilde{\alpha}_{3, 0} & \tilde{\alpha}_{3, 1} & 0 & 1 & 0 & \cdots\\ \tilde{\alpha}_{4, 0} & \tilde{\alpha}_{4, 1} & \tilde{\alpha}_{4, 2} & \tilde{\alpha}_{4, 3} & 1 & \\ \vdots & \vdots & \vdots & \vdots & & \ddots \end{array} \right].\end{aligned}$$ For the calculation, we will find it more convenient to work with the equation $$\begin{aligned} \label{e:xinvtQ} {{\mathbf p}}= {\mathbf{\tilde{X}}}^{-1} {\mathbf{\tilde{Q}}}.\end{aligned}$$ Since the skew-orthogonal polynomials will also form a complete set, we know that ${\mathbf{\tilde{X}}}$ is invertible and we denote $$\begin{aligned} \label{d:Xinvmat} {\mathbf{\tilde{X}}}^{-1}= \left[ \begin{array}{cccccc} 1 & 0 & 0 & 0 & 0 & \cdots\\ \tilde{\beta}_{2,0} & 1 & 0 & 0 & 0 & \cdots\\ \tilde{\beta}_{2, 0} & \tilde{\beta}_{2, 1} & 1 & 0 & 0 & \cdots\\ \beta_{3, 0} & \beta_{3, 1} & \beta_{3, 2} & 1 & 0 & \cdots\\ \tilde{\beta}_{4, 0} & \tilde{\beta}_{4, 1} & \tilde{\beta}_{4, 2} & \tilde{\beta}_{4, 3} & 1 & \\ \vdots & \vdots & \vdots & \vdots & & \ddots \end{array} \right]= \left[ \begin{array}{cccccc} 1 & 0 & 0 & 0 & 0 & \cdots\\ 0 & 1 & 0 & 0 & 0 & \cdots\\ \tilde{\beta}_{2, 0} & \tilde{\beta}_{2, 1} & 1 & 0 & 0 & \cdots\\ \beta_{3, 0} & \beta_{3, 1} & 0 & 1 & 0 & \cdots\\ \tilde{\beta}_{4, 0} & \tilde{\beta}_{4, 1} & \tilde{\beta}_{4, 2} & \tilde{\beta}_{4, 3} & 1 & \\ \vdots & \vdots & \vdots & \vdots & & \ddots \end{array} \right]\end{aligned}$$ where we have used the assumptions analogous to $$\begin{aligned} \label{e:betarules} {\tilde{\beta}}_{j,j}=1, \qquad {\tilde{\beta}}_{2j+1, 2j}=0.\end{aligned}$$ So instead of looking for the coefficients in we will solve for the coefficients ${\tilde{\beta}}_{j,k}$ in $$\begin{aligned} \label{e:Qp} p_j = {\tilde{Q}}_j + {\tilde{\beta}}_{j,j-1} {\tilde{Q}}_{j-1} + \dots + {\tilde{\beta}}_{j,1} {\tilde{Q}}_1 + {\tilde{\beta}}_{j,0} {\tilde{Q}}_0, \qquad {\tilde{\beta}}_{j,k}\in \mathbb{C},\end{aligned}$$ and then hope to invert the relations to recover the ${\tilde{\alpha}}_{j,k}$. We also define the matrix of inner products $$\begin{aligned} {\mathbf{\tilde{q}}}= \Big[ {\savebox{\@brx}{\(\m@th{\langle}\)} \mathopen{\copy\@brx\kern-0.5\wd\@brx\usebox{\@brx}}}{\tilde{Q}}_j, {\tilde{Q}}_k {\savebox{\@brx}{\(\m@th{\rangle}\)} \mathclose{\copy\@brx\kern-0.5\wd\@brx\usebox{\@brx}}}_{4}^y \Big]= \left[ \begin{array}{cccccc} 0 & {\tilde{q}}_0 & 0 & 0 & 0 & \cdots\\ -{\tilde{q}}_0 & 0 & 0 & 0 & 0 & \cdots\\ 0 & 0 & 0 & {\tilde{q}}_1 & 0 & \cdots\\ 0 & 0 & -{\tilde{q}}_1 & 0 & 0 & \cdots\\ 0 & 0 & 0 & 0 & 0 & \\ \vdots & \vdots & \vdots & \vdots & & \ddots \end{array} \right].\end{aligned}$$ Using we can write the modified $\beta=4$ skew-inner product in terms of the $\beta=2$ inner product, with the inclusion of the operator $A$ defined in . To make use of this we first note that if $f_k$ is any monic polynomial of degree $k$ then we have $$\begin{aligned} Af_k[x] &= -\left( xf_k (x) - f_k' (x) \right) = -\left( p_{k+1} (x) + \sum_{j=0}^{k-1} c_j p_{j} (x) \right),\end{aligned}$$ where we have decomposed $xf_k (x) - f_k' (x)$ into a sum over the (monic) orthogonal polynomials $p_j$, with coefficients $c_j$. Combining this fact with , and the normalization of the $p_j$ from we have the matrix $$\begin{aligned} \nonumber {\mathbf{A}}&:= \left[ (p_j, A p_k )_{2}^y \right]_{j, k =0, ..., N -1}\\ &= \left[ \begin{array}{ccccccc} \frac{\Omega_{0, 0}} {2} & h_1+ \Omega_{0, 1} & \Omega_{0, 2} & \Omega_{0, 3} & \Omega_{0, 4} & \Omega_{0, 5} &\cdots\\ -h_1 & \frac{\Omega_{1, 1}} {2} & h_2+ \Omega_{1, 2} & \Omega_{1, 3} & \Omega_{1, 4} & \Omega_{1, 5} &\cdots\\ 0 & -h_2 & \frac{\Omega_{2, 2}} {2} & h_3+ \Omega_{2, 3} & \Omega_{2, 4} & \Omega_{2, 5} &\cdots\\ 0 & 0 & -h_3 & \frac{\Omega_{3, 3}} {2} & h_4+ \Omega_{3, 4} & \Omega_{3, 5} &\cdots\\ 0 & 0 & 0 & -h_4 & \frac{\Omega_{4, 4}} {2} & h_5+ \Omega_{4, 5} &\\ \vdots & \vdots & \vdots & \vdots & \ddots & \ddots & \ddots \end{array}\right].\end{aligned}$$ So now we can write $$\begin{aligned} \nonumber \mathbf{\tilde{q}}&:= \left[ {\savebox{\@brx}{\(\m@th{\langle}\)} \mathopen{\copy\@brx\kern-0.5\wd\@brx\usebox{\@brx}}}\tilde{Q}_j, \tilde{Q}_k {\savebox{\@brx}{\(\m@th{\rangle}\)} \mathclose{\copy\@brx\kern-0.5\wd\@brx\usebox{\@brx}}}_{4}^y \right]= {\savebox{\@brx}{\(\m@th{\Big\langle}\)} \mathopen{\copy\@brx\kern-0.5\wd\@brx\usebox{\@brx}}}{{\mathbf{\tilde{Q}}}{\mathbf{\tilde{Q}}}^T}{\savebox{\@brx}{\(\m@th{\Big\rangle}\)} \mathclose{\copy\@brx\kern-0.5\wd\@brx\usebox{\@brx}}}_{4}^y = {\savebox{\@brx}{\(\m@th{\Big\langle}\)} \mathopen{\copy\@brx\kern-0.5\wd\@brx\usebox{\@brx}}}{{\mathbf{\tilde{X}}}{{\mathbf p}}{{\mathbf p}}^T {\mathbf{\tilde{X}}}^T}{\savebox{\@brx}{\(\m@th{\Big\rangle}\)} \mathclose{\copy\@brx\kern-0.5\wd\@brx\usebox{\@brx}}}_{4}^y\\ \nonumber & = {\mathbf{\tilde{X}}}{\savebox{\@brx}{\(\m@th{\Big\langle}\)} \mathopen{\copy\@brx\kern-0.5\wd\@brx\usebox{\@brx}}}{{{\mathbf p}}{{\mathbf p}}^T}{\savebox{\@brx}{\(\m@th{\Big\rangle}\)} \mathclose{\copy\@brx\kern-0.5\wd\@brx\usebox{\@brx}}}_{4}^y {\mathbf{\tilde{X}}}^T = {\mathbf{\tilde{X}}}\Big( {\mathbf{A}}- \frac{1}{2} \left[\Omega_{j,k} \right] \Big) {\mathbf{\tilde{X}}}^T\\ \label{e:xwxt}&= {\mathbf{\tilde{X}}}{\mathbf{W}}{\mathbf{\tilde{X}}}^T,\end{aligned}$$ where ${\mathbf{W}}= {\mathbf{A}}- \frac{1}{2} \left[\Omega_{j,k} \right]$ is the anti-symmetric matrix in . (Note that for a matrix ${\mathbf{M}}$ the notation ${\savebox{\@brx}{\(\m@th{\langle}\)} \mathopen{\copy\@brx\kern-0.5\wd\@brx\usebox{\@brx}}}{\mathbf{M}}{\savebox{\@brx}{\(\m@th{\rangle}\)} \mathclose{\copy\@brx\kern-0.5\wd\@brx\usebox{\@brx}}}_4^y$ implies that the average is applied elementwise to the matrix.) Rearranging $$\begin{aligned} \label{e:xinvq} {\mathbf{\tilde{X}}}^{-1} {\mathbf{\tilde{q}}}({\mathbf{\tilde{X}}}^{-1} )^T = {\mathbf{W}},\end{aligned}$$ and expanding out the left hand side we get $$\begin{aligned} \nonumber \Big[ {\mathbf{\tilde{X}}}^{-1} {\mathbf{\tilde{q}}}({\mathbf{\tilde{X}}}^{-1})^T \Big]_{j,k} &= \sum_{m = 0, 1, \dots, j \atop n = 0, 1, \dots, k} {\tilde{\beta}}_{j,m} \tilde{{{\mathbf q}}}_{m,n} {\tilde{\beta}}_{k,n}\\ \nonumber &=\sum_{m\, \mathrm{even}} {\tilde{\beta}}_{j, m} \tilde{{{\mathbf q}}}_{m,m+1} {\tilde{\beta}}_{k,m+1}+ \sum_{m\, \mathrm{odd}} {\tilde{\beta}}_{j, m} \tilde{{{\mathbf q}}}_{m,m-1} {\tilde{\beta}}_{k,m-1}\\ &=\sum_{m\, \mathrm{even}} {\tilde{\beta}}_{j, m} {\tilde{q}}_{m/2} {\tilde{\beta}}_{k,m+1} -\sum_{m\, \mathrm{odd}} {\tilde{\beta}}_{j, m} {\tilde{q}}_{(m-1)/2} {\tilde{\beta}}_{k,m-1},\end{aligned}$$ noting that this is a finite sum since all $\beta_{\mu, \nu}$ are zero when $\nu> \mu$. So we have the set of equations $$\begin{aligned} \label{e:betaeqns} 0&= \sum_{m \, \mathrm{even}} {\tilde{q}}_{m/2} \left( {\tilde{\beta}}_{j,m} {\tilde{\beta}}_{k, m+1}- {\tilde{\beta}}_{j,m+1} {\tilde{\beta}}_{k, m} \right) - w_{j,k}\end{aligned}$$ and we are now in a position to solve for the normalizations ${\tilde{q}}_j$ and the coefficients ${\tilde{\beta}}_{j,k}$. Expressions for ${\tilde{q}}_j$ ------------------------------- Let the matrices in be of size $2n \times 2n$. Then, taking the Pfaffian we get $$\begin{aligned} {\mathrm{Pf}\,}{\mathbf{W}}= {\mathrm{Pf}\,}({\mathbf{\tilde{X}}}^{-1} {\mathbf{\tilde{q}}}({\mathbf{\tilde{X}}}^T)^{-1})= \det ({\mathbf{\tilde{X}}}^{-1}) {\mathrm{Pf}\,}{\mathbf{\tilde{q}}}= {\mathrm{Pf}\,}{\mathbf{\tilde{q}}}\;,\end{aligned}$$ where we used the Pfaffian identity for the second equality, and the fact that ${\mathbf{\tilde{X}}}^{-1}$ is a triangular matrix with $1$s on the diagonal for the third equality. Because ${\mathbf{\tilde{q}}}$ is a skew-diagonal matrix, as in , we have $$\begin{aligned} \label{e:prodqj} {\mathrm{Pf}\,}{\mathbf{\tilde{q}}}= \prod_{j=0}^{n-1} {\tilde{q}}_j = {\mathrm{Pf}\,}{\mathbf{W}}_{2n-1}.\end{aligned}$$ Beginning with $n=1$ and iterating, we obtain , with the convention . Expressions for ${\tilde{\beta}}_{j,k}$ --------------------------------------- Let $k$ be even, then the last term in the sum of is $-{\tilde{\beta}}_{j,k+1} {\tilde{q}}_{k/2}$ (when $m= k$), and so solving for this ${\tilde{\beta}}$ we obtain $$\begin{aligned} \label{e:betake} {\tilde{\beta}}_{j,k+1} = \frac{1}{{\tilde{q}}_{k/2}} \left[ \sum_{m=0, \atop m\, \mathrm{even}}^{k-2} {\tilde{q}}_{m/2} \left( {\tilde{\beta}}_{j,m} {\tilde{\beta}}_{k, m+1} - {\tilde{\beta}}_{j,m+1} {\tilde{\beta}}_{k, m}\right) -w_{j,k} \right], \qquad \mbox{$k$ even}\end{aligned}$$ For $k$ odd the last term (when $m=k-1$) is ${\tilde{q}}_{(k-1)/2}( {\tilde{\beta}}_{j, k-1} - {\tilde{\beta}}_{j,k} {\tilde{\beta}}_{k, k-1})$, but recall from that (when $k$ is odd) we have set ${\tilde{\beta}}_{k, k-1} =0$ \[using \], so we obtain $$\begin{aligned} \label{e:betako} {\tilde{\beta}}_{j, k-1} = -\frac{1}{{\tilde{q}}_{(k-1)/2}} \left[ \sum_{m=0, \atop m\, \mathrm{even}}^{k-3} {\tilde{q}}_{m/2} \left( {\tilde{\beta}}_{j,m} {\tilde{\beta}}_{k, m+1} - {\tilde{\beta}}_{j,m+1} {\tilde{\beta}}_{k, m}\right) -w_{j,k} \right], \qquad \mbox{$k$ odd}.\end{aligned}$$ From these two expressions we see that each ${\tilde{\beta}}_{j,2k}$ and ${\tilde{\beta}}_{j, 2k+1}$ only depends on the ${\tilde{\beta}}$s in the same row, and in columns $0,1,\dots 2k-1$. This allows us to inductively solve for the ${\tilde{\beta}}$: first we solve for ${\tilde{\beta}}_{j,0}, {\tilde{\beta}}_{j,1}$, then ${\tilde{\beta}}_{j,2}, {\tilde{\beta}}_{j,3}$, etc. It is this decoupling of the ${\tilde{\beta}}$ equations that are the reason for working with ${\mathbf{\tilde{X}}}^{-1}$ instead of ${\mathbf{\tilde{X}}}$. \[p:betas4\] $$\begin{aligned} \label{e:betas} {\tilde{\beta}}_{j,k}&= \left\{ \begin{array}{cl} \frac{{\displaystyle{\mathrm{Pf}\,}{\mathbf{W}}_{k+1}^{(k\mapsto j)}}} {{\displaystyle{\mathrm{Pf}\,}{\mathbf{W}}_{k+1}}},&\quad \mbox{$k$ even},\\ \frac{{\displaystyle{\mathrm{Pf}\,}{\mathbf{W}}_{k}^{(k\mapsto j)}}} {{\displaystyle{\mathrm{Pf}\,}{\mathbf{W}}_{k}}},&\quad \mbox{$k$ odd}, \end{array} \right.\end{aligned}$$ where ${\mathbf{W}}_{\mu}^{(\eta \mapsto \nu)}$ is the matrix ${\mathbf{W}}_{\mu}$ from with all occurrences of the index $\eta$ replaced by the index $\nu$. [*Proof*:]{}As mentioned above, we will employ an inductive proof. We need both even and odd base cases. Expanding out with $k=0$ we have $$\begin{aligned} &0= -{\tilde{\beta}}_{j,1} {\tilde{q}}_0 {\tilde{\beta}}_{0,0} - w_{j,0}\\ &\Rightarrow {\tilde{\beta}}_{j,1}= -\frac{w_{j,0}} {{\tilde{q}}_0}= \frac{w_{0,j}} {w_{0,1}} = \frac{{\displaystyle{\mathrm{Pf}\,}{\mathbf{W}}_{1}^{(1\mapsto j)}}} {{\displaystyle{\mathrm{Pf}\,}{\mathbf{W}}_{1}}}.\end{aligned}$$ Similarly, \[recalling that ${\tilde{\beta}}_{1,0}=0$ from \] with $k=1$, we get $$\begin{aligned} &0= {\tilde{\beta}}_{j,0}{\tilde{q}}_0 {\tilde{\beta}}_{1,1}- {\tilde{\beta}}_{j,1} {\tilde{q}}_0 {\tilde{\beta}}_{1,0} - w_{j,1}\\ &\Rightarrow {\tilde{\beta}}_{j,0}= \frac{w_{j,1}} {{\tilde{q}}_0} = \frac{{\displaystyle{\mathrm{Pf}\,}{\mathbf{W}}_{1}^{(0\mapsto j)}}} {{\displaystyle{\mathrm{Pf}\,}{\mathbf{W}}_{1}}}.\end{aligned}$$ Now we move to the inductive step. For convenience, here we restrict to $k$ even. Assume that we have for all ${\tilde{\beta}}_{j,0}, {\tilde{\beta}}_{j,1}, \dots, {\tilde{\beta}}_{j,k-1}$ and we substitute and into to get $$\begin{aligned} \label{e:beta1} {\tilde{\beta}}_{j,k+1} = \frac{{\mathrm{Pf}\,}{\mathbf{W}}_{k-1}} {{\mathrm{Pf}\,}{\mathbf{W}}_{k+1}} \left[ \sum_{m=0, \atop m\, \mathrm{even}}^{k-2} \left( \frac{{\displaystyle{\mathrm{Pf}\,}{\mathbf{W}}_{m+1}^{(m\mapsto j)}}} {{\mathrm{Pf}\,}{\mathbf{W}}_{m-1}} \frac{{\displaystyle{\mathrm{Pf}\,}{\mathbf{W}}_{m+1}^{(m+1 \mapsto k)}}} {{\displaystyle{\mathrm{Pf}\,}{\mathbf{W}}_{m+1}}} - \frac{{\displaystyle{\mathrm{Pf}\,}{\mathbf{W}}_{m+1}^{(m+1 \mapsto j)}}} {{\displaystyle{\mathrm{Pf}\,}{\mathbf{W}}_{m+1}}} \frac{{\displaystyle{\mathrm{Pf}\,}{\mathbf{W}}_{m+1}^{(m\mapsto k)}}} {{\mathrm{Pf}\,}{\mathbf{W}}_{m-1}} \right) -w_{j,k} \right].\end{aligned}$$ Using [@Knuth1996 (1.1)] we obtain $$\begin{aligned} \nonumber &{\mathrm{Pf}\,}{\mathbf{W}}_{m+1}^{(m\mapsto j)} {\mathrm{Pf}\,}{\mathbf{W}}_{m+1}^{(m+1\mapsto k)}- {\mathrm{Pf}\,}{\mathbf{W}}_{m+1}^{(m+1\mapsto j)} {\mathrm{Pf}\,}{\mathbf{W}}_{m+1}^{(m\mapsto k)}\\ \label{e:KnuthPf1} &= {\mathrm{Pf}\,}{\mathbf{W}}_{m-1} {\mathrm{Pf}\,}{\mathbf{W}}_{m+3}^{(m+2\mapsto k, m+3\mapsto j)}- {\mathrm{Pf}\,}{\mathbf{W}}_{m+1}{\mathrm{Pf}\,}{\mathbf{W}}_{m+1}^{(m\mapsto k,m+1\mapsto j)}.\end{aligned}$$ The notation in [@Knuth1996] is quite different to that used here, so we briefly outline how follows from [@Knuth1996 (1.1)], which we quote here, rearranged for convenience $$\begin{aligned} \label{e:KnuthPf2} - f[\alpha x z] f[\alpha w y] + f[\alpha w z] f[\alpha x y] &= f[\alpha] f[\alpha w x y z]- f[\alpha w x] f[\alpha y z]\end{aligned}$$ where $w,x,y,z \in \mathbb{Z}$ are matrix indices and $\alpha \in \mathbb{Z}^p$ is an ordered set of indices. For index sets $\alpha_1\in \mathbb{Z}^p, \alpha_2 \in \mathbb{Z}^q$ the product $\alpha_1 \alpha_2 \in \mathbb{Z}^{p+q}$ is the concatenation of the index sets. The function $f[\alpha]$ is then the Pfaffian of the matrix $\big[ f[jk] \big]$ with index set $\alpha$, i.e. $$\begin{aligned} \label{e:KnuthPf3} f[\alpha] = {\mathrm{Pf}\,}\Big[ f[jk] \Big]_{j,k \in \alpha}\end{aligned}$$ defined recursively, where for a pair of indices $f[jk]$ is the matrix element, and $$\begin{aligned} \label{e:KnuthPf4} f[jk]=-f[kj]\end{aligned}$$ since Pfaffian matrices are anti-symmetric. So then to match with we take $$\begin{aligned} \alpha=\{ 0, 1, \dots, m-1\}, \quad w=\{m\}, \quad x= \{m+1\}, \quad y=\{k\}, \quad z= \{j\},\end{aligned}$$ and apply to rearrange the indices as needed. Substituting into we obtain $$\begin{aligned} {\tilde{\beta}}_{j,k+1} &= \frac{{\mathrm{Pf}\,}{\mathbf{W}}_{k-1}} {{\mathrm{Pf}\,}{\mathbf{W}}_{k+1}} \left[ \sum_{m=0, \atop m\, \mathrm{even}}^{k-2} \frac{{\displaystyle{\mathrm{Pf}\,}{\mathbf{W}}_{m+3}^{(m+2 \mapsto k, m+3\mapsto j)}}} {{\displaystyle{\mathrm{Pf}\,}{\mathbf{W}}_{m+1}}} - \frac{{\displaystyle{\mathrm{Pf}\,}{\mathbf{W}}_{m+1}^{(m\mapsto k, m+1\mapsto j)}}} {{\mathrm{Pf}\,}{\mathbf{W}}_{m-1}} -w_{j,k} \right],\end{aligned}$$ which is a telescoping sum, leaving $$\begin{aligned} {\tilde{\beta}}_{j,k+1}&= \frac{{\mathrm{Pf}\,}{\mathbf{W}}_{k-1}} {{\mathrm{Pf}\,}{\mathbf{W}}_{k+1}} \left[ \frac{{\displaystyle{\mathrm{Pf}\,}{\mathbf{W}}_{k+ 1}^{(k+1 \mapsto j)}}} {{\displaystyle{\mathrm{Pf}\,}{\mathbf{W}}_{k-1}}} - \frac{{\displaystyle{\mathrm{Pf}\,}{\mathbf{W}}_{1}^{(0\mapsto k, 1\mapsto j)}}} {{\mathrm{Pf}\,}{\mathbf{W}}_{-1}} -w_{j,k} \right]\\ &= \frac{{\displaystyle{\mathrm{Pf}\,}{\mathbf{W}}_{k+ 1}^{(k+1 \mapsto j)}}} {{\displaystyle{\mathrm{Pf}\,}{\mathbf{W}}_{k+1}}}\end{aligned}$$ since ${\displaystyle{\mathrm{Pf}\,}{\mathbf{W}}_{1}^{(0\mapsto k, 1\mapsto j)}} = w_{k,j}= -w_{j,k}$, and we also used the convention . For the odd case, one proceeds from in a similar fashion. $\Box$ Note from that $$\begin{aligned} {\tilde{\beta}}_{2n+1, 2n} = \frac{{\displaystyle{\mathrm{Pf}\,}{\mathbf{W}}_{2n+1}^{(2n\mapsto 2n+1)}}} {{\displaystyle{\mathrm{Pf}\,}{\mathbf{W}}_{2n+1}}},\end{aligned}$$ and since $w_{n,n}=0$ for all $n$, then the Pfaffian in the numerator has two identical columns (the right-most) and two identical rows (the bottom-most), which implies $$\begin{aligned} \label{e:beta0} {\displaystyle{\mathrm{Pf}\,}{\mathbf{W}}_{2n+1}^{(2n\mapsto 2n+1)}}=0 {\qquad \Rightarrow \qquad}{\tilde{\beta}}_{2n+1, 2n} =0.\end{aligned}$$ Also, we clearly have $$\begin{aligned} {\tilde{\beta}}_{j,j}&= \left\{ \begin{array}{cl} \frac{{\displaystyle{\mathrm{Pf}\,}{\mathbf{W}}_{j+1}^{(j\mapsto j)}}} {{\displaystyle{\mathrm{Pf}\,}{\mathbf{W}}_{j+1}}},&\quad \mbox{$j$ even},\\ \frac{{\displaystyle{\mathrm{Pf}\,}{\mathbf{W}}_{j}^{(j\mapsto j)}}} {{\displaystyle{\mathrm{Pf}\,}{\mathbf{W}}_{j}}},&\quad \mbox{$j$ odd}, \end{array}\right\} \quad= 1\end{aligned}$$ so we recover . Expressions for ${\tilde{\alpha}}_{j,k}$ in Proposition \[p:alpha4\] {#a:Proofbeta4} -------------------------------------------------------------------- From the matrix product $$\begin{aligned} {\mathbf{\tilde{X}}}^{-1} {\mathbf{\tilde{X}}}={\mathbf{I}}\end{aligned}$$ we have $$\begin{aligned} \label{e:alphas3} {\tilde{\alpha}}_{j,k}= - \sum_{m=k}^{j-1} {\tilde{\beta}}_{j,m} {\tilde{\alpha}}_{m,k} \end{aligned}$$ for $j>k$. Using this and the expressions for the ${\tilde{\beta}}_{j,k}$ in we can find expressions for the ${\tilde{\alpha}}_{j,k}$. *Proof of Proposition \[p:alpha4\]*: From we have $$\begin{aligned} \nonumber {\tilde{Q}}_j&= p_j - \sum_{k=0}^{j-1} {\tilde{\beta}}_{j, k} {\tilde{Q}}_k\\ \label{e:tqpj} &=p_j - {\tilde{\beta}}_{j,j-1} {\tilde{Q}}_{j-1} - \sum_{k=0}^{j-2} {\tilde{\beta}}_{j, k} {\tilde{Q}}_k,\end{aligned}$$ so ${\tilde{\alpha}}_{j,j} = {\tilde{\beta}}_{j,j}=1$. Then, with , this also implies $$\begin{aligned} {\tilde{\beta}}_{j,j-1} {\tilde{Q}}_{j-1} = {\tilde{\beta}}_{j,j-1} \left( p_{j-1} + {\tilde{\alpha}}_{j-1, j-2} p_{j-2}+ \dots \right),\end{aligned}$$ and thus $$\begin{aligned} {\tilde{Q}}_j&= p_j -{\tilde{\beta}}_{j,j-1} p_{j-1} - \mbox{lower degree polynomials}.\end{aligned}$$ So we have that $$\begin{aligned} \label{e:talphatbeta} {\tilde{\alpha}}_{j,j-1}= -{\tilde{\beta}}_{j, j-1},\end{aligned}$$ which is equal to zero when $j$ is odd by . Now we have consistency with both and . For we will use an inductive proof similar to that used in Proposition \[p:betas4\]. We see from that each ${\tilde{\alpha}}_{j,k}$ only depends on the ${\tilde{\beta}}$’s (which are known) and the ${\tilde{\alpha}}$’s above it in the same column of the matrix ${\mathbf{\tilde{X}}}$ \[in \]. From we have $$\begin{aligned} \label{e:alphaPf1} {\tilde{\alpha}}_{j,j-1} = - {\tilde{\beta}}_{j, j-1}= \left\{ \begin{array}{cl} - \frac{{\displaystyle{\mathrm{Pf}\,}{\mathbf{W}}_{j-1}^{(j-1\mapsto j)}}} {{\displaystyle{\mathrm{Pf}\,}{\mathbf{W}}_{j-1}}}, &\quad \mbox{$j$ even},\\ 0, &\quad \mbox{$j$ odd}, \end{array}\right.\end{aligned}$$ and from $$\begin{aligned} \label{e:alphaPf2} {\tilde{\alpha}}_{j,j-2} = -{\tilde{\beta}}_{j,j-2}{\tilde{\alpha}}_{j-2, j-2} - {\tilde{\beta}}_{j, j-1} {\tilde{\alpha}}_{j-1, j-2}= -{\tilde{\beta}}_{j,j-2} = \left\{ \begin{array}{ll} - \frac{{\displaystyle{\mathrm{Pf}\,}{\mathbf{W}}_{j-1}^{(j-1\mapsto j)}}} {{\displaystyle{\mathrm{Pf}\,}{\mathbf{W}}_{j-1}}}, &\quad \mbox{$j$ even},\\ - \frac{{\displaystyle{\mathrm{Pf}\,}{\mathbf{W}}_{j-2}^{(j-2\mapsto j)}}} {{\displaystyle{\mathrm{Pf}\,}{\mathbf{W}}_{j-2}}}, &\quad \mbox{$j$ odd}, \end{array}\right.\end{aligned}$$ since one of ${\tilde{\beta}}_{j, j-1}$ or ${\tilde{\alpha}}_{j-1, j-2}$ must be zero by or . The equations and give us expressions for all ${\tilde{\alpha}}$s on the first and second lower diagonals of ${\mathbf{\tilde{X}}}$. So for any column $k$, there is a row $j$ for which all the ${\tilde{\alpha}}_{j- m, k}$ above it are known, so we have our base cases. Now for the inductive step, we expand to obtain $$\begin{aligned} \label{e:alphas1} {\tilde{\alpha}}_{j,k}&= \left\{ \begin{array}{ll} -{\tilde{\beta}}_{j,k} -{\displaystyle \sum_{m=k+2}^{j-1} {\tilde{\beta}}_{j,m} {\tilde{\alpha}}_{m,k}}, & \qquad \mbox{$j$ even, $k\leq j-1$, $k$ even},\\ -{\tilde{\beta}}_{j,k} -{\displaystyle \sum_{m=k+1}^{j-1} {\tilde{\beta}}_{j,m} {\tilde{\alpha}}_{m,k}}, & \qquad \mbox{$j$ even, $k\leq j-1$, $k$ odd},\\ -{\tilde{\beta}}_{j,k} -{\displaystyle \sum_{m=k+2}^{j-2} {\tilde{\beta}}_{j,m} {\tilde{\alpha}}_{m,k}}, & \qquad \mbox{$j$ odd, $k\leq j-1$, $k$ even},\\ -{\tilde{\beta}}_{j,k} -{\displaystyle \sum_{m=k+1}^{j-2} {\tilde{\beta}}_{j,m} {\tilde{\alpha}}_{m,k}}, & \qquad \mbox{$j$ odd, $k\leq j-1$, $k$ odd}.\\ \end{array}\right.\end{aligned}$$ We assume that ${\tilde{\alpha}}_{m,k}$ is given by for all $m\leq j-1$ ($j$ even) or $m\leq j-2$ ($j$ odd), while all ${\tilde{\beta}}$s are given by . Taking $j,k$ both even (the other cases follow similarly), we substitute these known ${\tilde{\alpha}}$’s and ${\tilde{\beta}}$’s into the first row of to give $$\begin{aligned} \nonumber {\tilde{\alpha}}_{j,k} &= - \frac{{\displaystyle{\mathrm{Pf}\,}{\mathbf{W}}_{k+1}^{(k\mapsto j)}}} {{\displaystyle{\mathrm{Pf}\,}{\mathbf{W}}_{k+1}}} + \sum_{m=k+2 \atop m\, \mathrm{even}}^{j-2} \frac{{\displaystyle{\mathrm{Pf}\,}{\mathbf{W}}_{m+1}^{(m\mapsto j)}}} {{\displaystyle{\mathrm{Pf}\,}{\mathbf{W}}_{m+1}}} \frac{{\displaystyle{\mathrm{Pf}\,}{\mathbf{W}}_{m-1}^{(k\mapsto m)}}} {{\displaystyle{\mathrm{Pf}\,}{\mathbf{W}}_{m-1}}}+ \sum_{m=k+3 \atop m\, \mathrm{odd}}^{j-1} \frac{{\displaystyle{\mathrm{Pf}\,}{\mathbf{W}}_{m}^{(m\mapsto j)}}} {{\displaystyle{\mathrm{Pf}\,}{\mathbf{W}}_{m}}} \frac{{\displaystyle{\mathrm{Pf}\,}{\mathbf{W}}_{m-2}^{(k\mapsto m)}}} {{\displaystyle{\mathrm{Pf}\,}{\mathbf{W}}_{m-2}}}\\ \label{e:alphas2}&=- \frac{{\displaystyle{\mathrm{Pf}\,}{\mathbf{W}}_{k+1}^{(k\mapsto j)}}} {{\displaystyle{\mathrm{Pf}\,}{\mathbf{W}}_{k+1}}} + \sum_{m=k+2 \atop m\, \mathrm{even}}^{j-2} \frac{{\displaystyle{\mathrm{Pf}\,}{\mathbf{W}}_{m+1}^{(m\mapsto j)}}} {{\displaystyle{\mathrm{Pf}\,}{\mathbf{W}}_{m+1}}} \frac{{\displaystyle{\mathrm{Pf}\,}{\mathbf{W}}_{m-1}^{(k\mapsto m)}}} {{\displaystyle{\mathrm{Pf}\,}{\mathbf{W}}_{m-1}}}+ \frac{{\displaystyle{\mathrm{Pf}\,}{\mathbf{W}}_{m+1}^{(m+1 \mapsto j)}}} {{\displaystyle{\mathrm{Pf}\,}{\mathbf{W}}_{m+1}}} \frac{{\displaystyle{\mathrm{Pf}\,}{\mathbf{W}}_{m-1}^{(k\mapsto m+1)}}} {{\displaystyle{\mathrm{Pf}\,}{\mathbf{W}}_{m-1}}},\end{aligned}$$ keeping in mind that we have the convention that ${\mathrm{Pf}\,}{\mathbf{W}}_{-1} =1$. We now use [@Knuth1996 (5.1)] (again quoted here and rearranged for convenience) $$\begin{aligned} \nonumber f[\alpha xuw] f[\alpha vyz] -f[\alpha xuv] f[\alpha wyz] &= -f[\alpha u v w] f[\alpha x y z] + f[\alpha uyz] f[\alpha xvw]\\ \label{e:Knuth2} & +f[\alpha z] f[ \alpha u v w x y] - f[\alpha y] f[\alpha uvwxz]\end{aligned}$$ with $$\begin{aligned} \alpha= \{0, 1, \dots, k-1, k+1, \dots, m-1 \}, \quad &x= \{ j\}, \quad u= \{ k\}, \quad v= \{ m\}, \quad w= \{ m+1\}\\ &y=z=\emptyset \qquad \mbox{(the empty set)}.\end{aligned}$$ Rearranging indices according to , the equality gives $$\begin{aligned} &{\mathrm{Pf}\,}{\mathbf{W}}_{m+1}^{(m\mapsto j)} {\mathrm{Pf}\,}{\mathbf{W}}_{m-1}^{(k\mapsto m)}+ {\mathrm{Pf}\,}{\mathbf{W}}_{m+1}^{(m+1\mapsto j)} {\mathrm{Pf}\,}{\mathbf{W}}_{m-1}^{(k\mapsto m+1)}\\ &= {\mathrm{Pf}\,}{\mathbf{W}}_{m+1} {\mathrm{Pf}\,}{\mathbf{W}}_{m-1}^{(k\mapsto j)}- {\mathrm{Pf}\,}{\mathbf{W}}_{m-1}{\mathrm{Pf}\,}{\mathbf{W}}_{m+1}^{(k\mapsto j)},\end{aligned}$$ and substituting into we get $$\begin{aligned} {\tilde{\alpha}}_{j,k} &=- \frac{{\displaystyle{\mathrm{Pf}\,}{\mathbf{W}}_{k+1}^{(k\mapsto j)}}} {{\displaystyle{\mathrm{Pf}\,}{\mathbf{W}}_{k+1}}} + \sum_{m=k+2 \atop m\, \mathrm{even}}^{j-2} \frac{{\mathrm{Pf}\,}{\mathbf{W}}_{m-1}^{(k\mapsto j)}} {{\displaystyle{\mathrm{Pf}\,}{\mathbf{W}}_{m-1}}} - \frac{{\mathrm{Pf}\,}{\mathbf{W}}_{m+1}^{(k\mapsto j)}} {{\displaystyle{\mathrm{Pf}\,}{\mathbf{W}}_{m+1}}}.\end{aligned}$$ This is a telescoping sum, which reduces to . The other cases in are calculated similarly. $\Box$ $\beta =4$ polynomials in the classical limit {#a:limpolys4} --------------------------------------------- In the classical limit ($y\to \infty$) the skew inner product becomes $$\begin{aligned} \langle f, g \rangle_4:= \frac{1}{2} \int_{- \infty}^{\infty} dx \; e^{-2 x^2} \left[ f(x) g'(x)- g(x) f'(x) \right],\end{aligned}$$ and the associated skew-orthogonal polynomials obeying $$\begin{aligned} \nonumber \langle Q_{2j} , Q_{2k} \rangle_{4}&= \langle Q_{2j+1} , Q_{2k+1} \rangle_{4}= 0\\ \langle Q_{2j} , Q_{2k+1} \rangle_{4}&= -\langle Q_{2k+1} , Q_{2j} \rangle_{4} = q_j \delta_{j,k}\end{aligned}$$ are given by [@NagaWada1991; @AdleForrNagavanMoer2000] $$\begin{aligned} \nonumber Q_{2j+1}(x) &= p_{2j+1} (\sqrt{2} x), &Q_{2j}(x) &= \sum_{t=0}^{j} \left( \prod_{s=t+1}^{j} \frac{h_{2s}}{h_{2s-1}} \right) p_{2t} (\sqrt{2} x)\\ \label{e:LimPolys4} &&&= \sum_{t=0}^{j} \frac{j!}{t!}\; p_{2t} (\sqrt{2} x)\end{aligned}$$ \[up to the invariance \], where the polynomials $$\begin{aligned} \label{e:monHerms} p_{j} (x) = \frac{1}{2^j} H_{j} (x)\end{aligned}$$ are the (monic, “physicist’s”) Hermite polynomials in and $h_j = h_{j} (\infty)$ from . The corresponding normalizations $q_j = q_j(\infty)$ are also from . As mentioned after Proposition \[p:alpha4\], it can be seen that the results of that Proposition reduce to the classical polynomials , since in the limit $y\to \infty$ the matrix $[\Omega_{j,k}]=0$ in , and we then follow exactly the steps in [@AdleForrNagavanMoer2000] to obtain . Skew-orthogonal polynomials for $\beta=1$ {#a:beta1} ========================================= We again suppress the explicit dependence on $y$ to save space, although all the quantities here depend on $y$. We can follow the same steps as for the $\beta=4$ case in Appendix \[a:beta4\] to find the coefficients $\alpha_{j,k}$ in . With ${{\mathbf p}}$ from we first rewrite equation as $$\begin{aligned} {\mathbf{R}}= {\mathbf{X}}{{\mathbf p}}{\qquad \Rightarrow \qquad}{{\mathbf p}}= {\mathbf{X}}^{-1} {\mathbf{R}},\end{aligned}$$ where $$\begin{aligned} {\mathbf{R}}= \left[ \begin{array}{c} R_0 \\ R_1\\ \vdots \end{array} \right],\end{aligned}$$ and ${\mathbf{X}}$ and ${\mathbf{X}}^{-1}$ are the same as in and , but without the tildes. Also define the matrices $$\begin{aligned} {{\mathbf r}}&:= \big[ \langle R_j, R_k \rangle_1^y \big]_{j,k = 0, 1, \dots, N-1},\\ \label{d:Bmat} {\mathbf{B}}&:= [(p_j, A^{-1} p_k)]_{j,k = 0, 1, \dots, N-1},\\ \label{d:bPhi} \boldsymbol{\Phi} &:= [\Phi_{j,k}]_{j,k = 0, 1, \dots, N-1},\end{aligned}$$ where ${{\mathbf r}}$ is of skew-diagonal form . Then $$\begin{aligned} \nonumber {{\mathbf r}}&= \Big[ \langle R_j, R_k \rangle_{1}^{y} \Big]= \left\langle {\mathbf{R}}{\mathbf{R}}^T \right\rangle_{1}^{y}= \left\langle {\mathbf{X}}{{\mathbf p}}{{\mathbf p}}^T {\mathbf{X}}^T \right\rangle_{1}^{y}\\ \nonumber & ={\mathbf{X}}\left\langle {{\mathbf p}}{{\mathbf p}}^T \right\rangle_{1}^{y} {\mathbf{X}}^T\\ \nonumber &= -{\mathbf{X}}\Big( {\mathbf{B}}+ \boldsymbol{\Phi} \Big) {\mathbf{X}}^T\\ \label{e:xvxt}&= {\mathbf{X}}{\mathbf{V}}{\mathbf{X}}^T,\end{aligned}$$ where the anti-symmetric matrix ${\mathbf{V}}$ is defined in — we will discuss the derivation of the specific structure of the elements of ${\mathbf{V}}$ in Appendix \[a:VmElts\] below. (As above, the averages over matrix arguments imply that the average is applied elementwise to the matrix.) We now follow the same steps as in – to get $$\begin{aligned} \label{e:XrXinv} &{\mathbf{X}}^{-1} {{\mathbf r}}\left( {\mathbf{X}}^{-1} \right)^T = {\mathbf{V}}\\ \label{e:betaeqns1} {\qquad \Rightarrow \qquad}& \sum_{m \, \mathrm{even}} r_{m/2} \left( \beta_{j,m} \beta_{k, m+1}- \beta_{j,m+1} \beta_{k, m} \right) - v_{j,k} =0\end{aligned}$$ with ${\mathbf{V}}= {\mathbf{V}}_m= [v_{j,k}]_{j,k=0,\dots, m}$ from . Assuming $m=2n$, taking the Pfaffian of we have $$\begin{aligned} {\mathrm{Pf}\,}{{\mathbf r}}= \prod_{j=0}^{n-1} r_j = {\mathrm{Pf}\,}{\mathbf{V}}_{2n-1}\end{aligned}$$ and we obtain , with the convention . Then, since the equations in are of the same form as , we apply the same reasoning as that in Proposition \[p:betas4\] to obtain solutions for the $\beta_{j,k}$ $$\begin{aligned} \beta_{j,k}&= \left\{ \begin{array}{cl} \frac{{\displaystyle{\mathrm{Pf}\,}{\mathbf{V}}_{k+1}^{(k\mapsto j)}}} {{\displaystyle{\mathrm{Pf}\,}{\mathbf{V}}_{k+1}}},&\quad \mbox{$k$ even},\\ \frac{{\displaystyle{\mathrm{Pf}\,}{\mathbf{V}}_{k}^{(k\mapsto j)}}} {{\displaystyle{\mathrm{Pf}\,}{\mathbf{V}}_{k}}},&\quad \mbox{$k$ odd}, \end{array} \right.\end{aligned}$$ where again, ${\mathbf{V}}_{\mu}^{(\eta \mapsto \nu)}$ is the matrix ${\mathbf{V}}_{\mu}$ with all occurrences of the index $\eta$ replaced by the index $\nu$. Now using the equations $$\begin{aligned} {\mathbf{X}}^{-1} {\mathbf{X}}= {\mathbf{I}}{\qquad \Rightarrow \qquad}\alpha_{j,k} = -\sum_{m=k}^{j-1} \beta_{j,m} \alpha_{m,k},\end{aligned}$$ we follow the same steps as in Appendix \[a:Proofbeta4\] and we establish the remaining statements in Proposition \[p:alpha1\]. Entries of the matrix ${\mathbf{V}}_m$ {#a:VmElts} -------------------------------------- For a general polynomial $$\begin{aligned} p_{j} (x) = c_{j,j} x^j + c_{j, j-1} x^{j-1} + \dots + c_{j,1} x + c_{j,0}\end{aligned}$$ we use the identities (calculated via repeated integration by parts) $$\begin{aligned} \label{e:Ointeg} \int_{a}^{b} e^{-u^2/2} u^{2k+1} du &= (2k)!! \left( \sum_{m=0}^k \frac{e^{-a^2/2} a^{2m} - e^{-b^2/2} b^{2m}}{(2m)!!} \right)\\ \nonumber \int_{a}^{b} e^{-u^2/2} u^{2k} du &= (2k-1)!! \left( \sum_{m=1}^k \frac{e^{-a^2/2} a^{2m-1} - e^{-b^2/2} b^{2m-1}}{(2m-1)!!} \right)\\ \label{e:Einteg}&+ (2k-1)!! \sqrt{\frac{\pi}{2}} \left( {\mathrm{erf}}\left( \frac{b}{\sqrt{2}} \right)- {\mathrm{erf}}\left( \frac{a}{\sqrt{2}} \right) \right)\end{aligned}$$ to obtain $$\begin{aligned} \int_{-\infty}^{\infty} e^{-z^2/2} p_k (z) dz = \sqrt{2 \pi} \sum_{t=0}^{\lfloor k/2 \rfloor} c_{k, 2t} (2t -1)!!\end{aligned}$$ and $$\begin{aligned} A^{-1}p_{k} [z] &= \left( \frac{e^{x^2/2}}{2}\; {\mathrm{erf}}\left( \frac{x}{\sqrt{2}} \right) \int_{-\infty}^{\infty} e^{-z^2/2} p_{k} (z) dz \right) - p_{k-1} (x) - (\mbox{lower order polynomials}).\end{aligned}$$ So then, with ${\mathbf{B}}$ defined in , we have $$\begin{aligned} \nonumber &{\mathbf{B}}= \begin{bmatrix} -\Phi_{0,0} & -h_0 + X_{0,1} & ?? & ??&\\ X_{1,0} & -\Phi_{1,1} & -h_1 + X_{1,2} & ??& \cdots\\ X_{2,0} & X_{2,1} & -\Phi_{2,2} & \\ &\vdots &&\ddots \end{bmatrix}\\ &= \begin{bmatrix} -\Phi_{0,0} & -h_0 + X_{0,1} & ?? & ??&\\ h_0 -X_{0,1} -\Phi(0,1) -\Phi(1,0) & -\Phi_{1,1} & -h_1 + X_{1,2} & ??& \cdots\\ X_{2,0} & h_1 -X_{1,2} -\Phi(1,2) -\Phi(2,1) & -\Phi_{2,2} & &\\ &\vdots&&\ddots \end{bmatrix},\end{aligned}$$ where the $??$ represents currently unknown elements, and the second equality comes from the use of . Adding the matrix $\boldsymbol{\Phi}$ from gives the (negative of the) anti-symmetric matrix ${\mathbf{V}}$ from , allowing us to specify the $??$ as so $$\begin{aligned} \nonumber &{\mathbf{B}}+\boldsymbol{\Phi}= -{\mathbf{V}}=\\ &\begin{bmatrix} 0 & -h_0 + X_{0,1} +\Phi(0,1) & -X_{2,0} - \Phi(2,0) & -X_{3,0} - \Phi(3,0)& \\ h_0 -X_{0,1} -\Phi(0,1) & 0 & -h_1 + X_{1,2} +\Phi(1,2) & -X_{3,1} - \Phi(3,1)& \cdots\\ X_{2,0} + \Phi(2,0) & h_1 -X_{1,2} -\Phi(1,2) & 0 & -h_2 + X_{2,3} +\Phi(2,3)&\\ X_{3,0} + \Phi(3,0) & X_{3,1}+ \Phi(3,1) & h_2 -X_{2,3} -\Phi(2,3) & 0&\\ &\vdots &&&\ddots \end{bmatrix}.\end{aligned}$$ $\beta =1$ polynomials in the classical limit {#a:limpolys1} --------------------------------------------- Similar to Appendix \[a:limpolys4\] above we have the $y \to \infty$ limit of the skew-inner product as $$\begin{aligned} \langle f, g \rangle_1 = \frac1{2} \int_{-\infty}^{\infty} dx \; e^{-x^2/2} f(x) \int_{-\infty}^{\infty} d z \; e^{-z^2/2} g (z) {\mathrm{sgn}}(z- x),\end{aligned}$$ with the associated skew-orthogonal polynomials obeying the equations $$\begin{aligned} \nonumber \langle R_{2j} , R_{2k} \rangle_{1}&= \langle R_{2j+1} , R_{2k+1} \rangle_{1}= 0\\ \langle R_{2j} , R_{2k+1} \rangle_{1}&=-\langle R_{2k+1} , R_{2j} \rangle_{1} = r_j \delta_{j,k}.\end{aligned}$$ These polynomials are given explicitly \[up to the invariance \] by [@NagaWada1991; @AdleForrNagavanMoer2000] $$\begin{aligned} \nonumber R_{2j} (x) &= p_{2j} (x), &R_{2j+1} (x) &= p_{2j+1} (x) - \frac{h_{2j}} {h_{2j-1}} p_{2j-1} (x)\\ \label{e:limHerms1} &&&= p_{2j+1} (x) - j\, p_{2j-1} (x),\end{aligned}$$ where the polynomials $p_j (x)$ are the Hermite polynomials in and $h_j = h_j (\infty)$. The normalizations $r_j = r_j (\infty)$ are from . To check coherence between and we can use integration by parts, the identities and $$\begin{aligned} \frac{d}{dx} {\mathrm{erf}}\left( \frac{x}{\sqrt{2}} \right) &= \sqrt{\frac{\pi}{2}} e^{-x^2/2}\end{aligned}$$ to give us $$\begin{aligned} \int_{-\infty}^{\infty} e^{-x^2/2} H_j (x) {\mathrm{erf}}\left( \frac{x}{\sqrt{2}} \right) dx = \left\{ \begin{array}{cl} 2^{(j+2)/2} (j-1)!! , \quad& \mbox{$j$ odd},\\ 0,& \mbox{$j$ even}. \end{array}\right.\end{aligned}$$ Substitution into yields $$\begin{aligned} \nonumber X_{j,k} \Big|_{y \to \infty} &= \frac{1}{2^{j+k+1}} \left( \int_{-\infty}^{\infty} H_j (x) e^{-x^2/2}\; {\mathrm{erf}}\left( \frac{x}{\sqrt{2}} \right) dx \right) \int_{-\infty}^{\infty} e^{-z^2/2} H_k (z) dz\\ \label{e:XjkLim} &= \left\{ \begin{array}{cl} \Gamma \left( \frac{j+1}{2} \right) \Gamma \left( \frac{k+1}{2} \right) = h_k (\infty) \frac{\Gamma \left( \frac{j+1}{2} \right)}{\Gamma \left( \frac{k+2}{2} \right)}, \quad& \mbox{$j$ odd} \wedge \mbox{$k$ even},\\ 0& \mbox{otherwise}, \end{array}\right.\end{aligned}$$ where we used for the integral over $H_k$. The second line (equalling zero) follows easily from the fact that the error function is an odd function and that $H_j(x)$ is an even or odd function depending on the parity of $j$. We will also make use of the formula $$\begin{aligned} \label{e:hjDupe} h_j (\infty) = \Gamma \left( \frac{j+1}{2} \right) \Gamma \left( \frac{j+2}{2} \right),\end{aligned}$$ which can be shown via Legendre’s duplication formula for gamma functions. In the case that $y= \infty$ then from the function $\Phi_{j,k}=0$ and we also use to find that the matrix ${\mathbf{V}}_m$ in has entries $$\begin{aligned} \label{e:VmLim} {\mathbf{V}}_m= \begin{bmatrix} 0&h_0 &0 &X_{3,0}&0&X_{5,0}&\\ -h_0& 0& 0& 0& 0&0& \\ 0&0&0&h_2 & 0& X_{5,2}& \cdots\\ -X_{3,0}&0 &-h_2& 0& 0&0&\\ 0&0&0&0&0& h_4\\ -X_{5,0}& 0& -X_{5,2}& 0& -h_4& 0\\ &&\vdots &&&&\ddots \end{bmatrix}\end{aligned}$$ meaning $$\begin{aligned} v_{j,k} = \left\{ \begin{array}{cl} h_j, &\quad \mbox{$j$ even $\wedge$ $k=j+1$},\\ 0,&\quad \mbox{$j$ odd $\vee$ $k$ even},\\ X_{k,j}, &\quad \mbox{$j$ even $\wedge$ $k$ odd $\wedge$ $j<k-1$}, \end{array} \right.\end{aligned}$$ with the anti-symmetry condition $$\begin{aligned} \label{e:VmLimAS} v_{j,k}= -v_{k,j}.\end{aligned}$$ So ${\mathbf{V}}_m$ is a sparse chequerboard matrix (as in [@AdlevanMoer2002 Eqn. (6.4)]), and in particular, the second row of ${\mathbf{V}}_m$ has the structure $$\begin{aligned} [-h_0\; 0\; 0 \;\dots \;0 ].\end{aligned}$$ This latter fact tells us that if we perform a Pfaffian Laplace expansion (as discussed in Appendix \[a:Pfaffs\] above) along the first row (with $j=0$), then the Pfaffian minors $M_{1,k}$ have a first row entirely of zeros, except when $k=2$. Since $v_{j,k}=X_{k,j} =0$ for all odd $j$ (with $j>k$), this patterns repeats for all the Pfaffian sub-minors and so $$\begin{aligned} \label{e:PfVm1} {\mathrm{Pf}\,}{\mathbf{V}}_{2j-1} = h_0 h_2 \cdots h_{2j-2},\end{aligned}$$ which gives us the denominator of $\alpha_{j,k}$ in . We can also understand this via the definition in terms of perfect matchings in : the structure of the upper triangle of the matrix in tells us that $v_{j,k}=0$ unless $j$ is even and $k$ is odd, meaning that all even sites in the link diagram connect to the right and all odd sites connect to the left. The only possible diagram satisfying this condition is the identity link pattern in Figure \[f:denomLinks\], which corresponds to the product in . For the numerator of $\alpha_{j,k}$ we have four cases to consider, being the four possibilities given by the parities of $j$ and $k$. : In the $2k$-th column we have the matrix entries $$\begin{aligned} v_{s, 2k} \mapsto v_{s, 2j} = X_{2j, s} =0 \qquad (s< 2k)\end{aligned}$$ while in the $2k$-th row we have $$\begin{aligned} v_{2k, t} \mapsto v_{2j, t} = -X_{2j, t} =0 \qquad (2k<t)\end{aligned}$$ so we have zeros above and to the right of the $(2k, 2k)$ entry (in the same column and row), which gives us $$\begin{aligned} \label{e:PfLim1} {\mathrm{Pf}\,}{\mathbf{V}}_{2j-1}^{(2k\mapsto 2j)} =0,\end{aligned}$$ since at least one of these zero factors must appear in each term of the Pfaffian. : Similar to the above, we have $$\begin{aligned} v_{s, 2k+1} \mapsto v_{s, 2j} = X_{2j, s} =0 \qquad (s< 2k+1)\end{aligned}$$ and $$\begin{aligned} v_{2k+1, t} \mapsto v_{2j, t} = -X_{2j, t} =0 \qquad (2k+1 < t).\end{aligned}$$ So now we have zeros above and to the right of the $(2k+1, 2k+1)$ entry, which gives us $$\begin{aligned} \label{e:PfLim2} {\mathrm{Pf}\,}{\mathbf{V}}_{2j-1}^{(2k+1 \mapsto 2j)} =0.\end{aligned}$$ : Now we have $$\begin{aligned} v_{s, 2k} \mapsto v_{s, 2j+1} = X_{2j+1, s} =0 \qquad (s< 2k \wedge \mbox{$s$ odd})\end{aligned}$$ so we still have every odd row containing only zeros (in the upper triangle). Thus, as in , the only term in the Laplace expansion that could be non-zero is $h_0 h_2 \cdots h_{2j-3}$. However, $$\begin{aligned} h_{2k} = v_{2k, 2k+1} \mapsto v_{2j+1, 2k+1} = -X_{2j+1, 2k+1} =0,\end{aligned}$$ and so $$\begin{aligned} \label{e:PfLim3} {\mathrm{Pf}\,}{\mathbf{V}}_{2j-1}^{(2k \mapsto 2j+1)} =0.\end{aligned}$$ : Using the expressions and we have the identity $$\begin{aligned} \label{e:Xhid} X_{2m+1, 2t} X_{2t+1, 2n} = h_{2t} X_{2m+1, 2n}, \qquad (m> t > n),\end{aligned}$$ which will make use of below. First we recall from that in the upper triangle $v_{j,k} \neq 0$ only when $j$ is even and when $k$ is odd, which implies that all the even sites in the corresponding diagram connect to the right, and all the odd sites connect to the left. However, we will have an exception to this when we make the replacement $2k+1 \mapsto 2j+1$. Specifically, in terms of link diagrams there are two possibilities for the links involving site $2j+1$: either $(2s, 2j+1)$ or $(2j+1, 2t)$ (so $2j+1$ is either the right or left vertex of the link). We note that the other vertex must be even, since any odd-odd or even-even link results in $X_{\mathrm{odd}, \mathrm{odd}} = 0 = X_{\mathrm{even}, \mathrm{even}}$. It is easiest to consider the two cases separately: - [Assume $2j+1$ connects to the left, that is we have a link $(2s, 2j+1)$. Since all other odd sites connect left and all other even sites connect right, this must be the identity link diagram, similar to Figure \[f:denomLinks\], so $s=j$.]{} - [Assume $2j+1$ connects to the right, that is we have a link $(2j+1, 2t)$, then we must have identity links at sites to the left of $2k$ and to the right of $2t+1$, as depicted in Figure \[f:linksii1\]. \[The left-pointing arrow on the edge $(2j+1, 2t)$ indicates that the left vertex is greater than the right vertex, which is the opposite convention to all the other links, and this introduces a negative sign from .\] In this case, we see from the diagram that there are 2 possible connections for $2t-2$, and then another 2 possible connections for $2t-4$, and so on. Thus there are $2^{t-k-1}$ link diagrams corresponding to Figure \[f:linksii1\]]{}. Summing over the possible values of $t=k+1, \dots, j-1$ in (ii), and adding the identity link pattern from (i), we have the number of valid link patterns on $N$ sites $L(N)$ given by $$\begin{aligned} \label{e:NumLinks} L(N)= 1+ \sum_{t=k+1}^{j-1} 2^{t-k-1} = 2^{j-k-1}.\end{aligned}$$ So for $2k+1< 2j-1$ we have an even number of terms in the Pfaffian, and it turns out that they all cancel. To show this, note that the restriction that all odd vertices connect to the left and all even vertices connect to the right (except for $2j+1$ and $2t$) means that a general link diagram must look like that in Figure \[f:GenLinks0\]. That is, big interconnected links, with a large rainbow link $(2j+1, \mathrm{even})$, and interspersed with little links. The big links must interconnect at neighbouring sites, since otherwise we would have two neighbouring vertices pointing in the same direction, violating the even/right–odd/left rule. We can construct every diagram of the type in Figure \[f:GenLinks0\] by application of the equality , by recasting that equation into the link diagram equalities in Figure \[f:linkIdentity2\], for the particular case when $m=j$. In Figure \[f:linkIdentity2\] (a) note the link diagram on the right has a left-pointing arrow (implying that the row index is larger than the column index), and so from we introduce a negative sign on the corresponding matrix entry. In Figure \[f:linkIdentity2\] (b) we have left-pointing arrows on both sides of the equality, but we have an additional sign introduced since the diagrams differ by an odd number of crossings. In Figure \[f:GenLinks2\] we give the example of constructing the link diagram in Figure \[f:GenLinks0\] from the identity diagram by repeated application of equalities in Figure \[f:linkIdentity2\] — starting from the left at the link $(2k, 2j+1)$ we first apply equality (a), and then, moving to the right, we repeatedly apply (b) until we have the final diagram. Each application of the equalities (a) and (b) introduces a negative sign. In the identity diagram there are $j-k-1$ little links to the right of site $2k+1$, so there are ${j-k-1 \choose p}$ link diagrams obtained from $p$ uses of the equalities in Figure \[f:linkIdentity2\], which gives us that $$\begin{aligned} \nonumber {\mathrm{Pf}\,}{\mathbf{V}}_{2j-1}^{(2k+1 \mapsto 2j+1)} &= (h_0 h_2 \cdots h_{2k-2}) v_{2k,2j+1} (h_{2k+2} \cdots h_{2j-2}) \sum_{p=0}^{j-k-1} (-1)^p {j-k-1 \choose p}\\ \label{e:altBinom} &= 0 \qquad \mbox{(for $k< j-1$)},\end{aligned}$$ where $(h_0 h_2 \cdots h_{2k-2}) v_{2k,2j+1} (h_{2k+2} \cdots h_{2j-2})$ is the term from the identity link diagram (i.e. the top diagram in Figure \[f:GenLinks2\]). The second equality follows since the sum of alternating binomial coefficients is equal to zero, which can be seen from the binomial expansion of $(x-y)^{j-k-1}$, with $x, y\to 1$. Thus $\alpha_{2j+1, 2k+1}=0$ when $k<j-1$. From we see the only scenario where we do not have an even number of cancelling link diagrams is when $k = j-1$, and we have only the identity link pattern. In this case, equation becomes $$\begin{aligned} \label{e:PfLim} {\mathrm{Pf}\,}{\mathbf{V}}_{2j-1}^{(2j-1 \mapsto 2j+1)} = h_0 h_2 \cdots h_{2j-4} X_{2j+1, 2j-2}\end{aligned}$$ since $v_{2j-2, 2j-1}\mapsto v_{2j-2, 2j+1}= X_{2j+1, 2j-2}$. Substituting and (with $m=2j-1$) into we have $$\begin{aligned} \alpha_{2j+1, 2j-1} = - \frac{X_{2j+1, 2j-2}} {h_{2j-2}} = - \frac{h_{2j}} {h_{2j-1}} = -\frac{\Gamma(j+1)}{\Gamma(j)} = -j,\end{aligned}$$ where we used for the second equality. Combining this result with , , and we recover . [^1]: Note that, for $\beta = 4$, the ${\mathrm{Tr}\,}$ function needs to be interpreted as a *quaternion trace* \[see Eq. \]. [^2]: Note that the factor $2^{-7/6}$ differs by a factor $2^{-1/6}$ from the result obtained in the original paper [@TracWido1996]. This mistake was actually noticed in [@nadal2011matrices p.47] — see also [@borot2012right]. There, this factor was corrected by matching with known asymptotic results for large (positive and negative) arguments. Here, we obtain this correct factor $2^{-7/6}$ by a direct computation. [^3]: The term *skew-diagonal* is used here in analogy with the term *diagonal*, that is, the (non-trivial) skew-symmetric(${\mathbf{M}}=-{\mathbf{M}}^T$) analogue of a diagonal matrix.

ArXiv

--- abstract: 'We report the development of the achromatic half-wave plate (AHWP) at millimeter wave for cosmic microwave background polarization experiments. We fabricate an AHWP consisting of nine A-cut sapphire plates based on the Pancharatnam recipe to cover a wide frequency range. The modulation efficiency and the phase are measured in a frequency range of 33 to 260 GHz with incident angles up to 10 degrees. We find the measurements at room temperature are in good agreement with the predictions. This is the most broadband demonstration of an AHWP at millimeter wave.' author: - Kunimoto Komatsu - Tomotake Matsumura - Hiroaki Imada - Hirokazu Ishino - Nobuhiko Katayama - Yuki Sakurai bibliography: - 'report.bib' title: 'Demonstration of the broadband half-wave plate using the nine-layer sapphire for the CMB polarization experiment' --- [**\***Kunimoto Komatsu, ]{} INTRODUCTION {#sec:intro} ============ Cosmic inflation is one of the theoretical models that give rise to the initial conditions of the hot big-bang in our universe. The rapid space expansion, immediately after the beginning of the universe, produced quantum fluctuations in space-time. Thus, it generated primordial gravitational waves, which imprinted the B-mode polarization in the cosmic microwave background (CMB). The strength of the gravitational waves is represented by a tensor-to-scalar ratio $r$. The majority of the single-scalar-field slow-roll inflation models predict the value of $r$ to be $> 0.01$. The target accuracy for future CMB polarization measurements is set to be less than 0.001 [@litebird; @cmc-s4]. The required level of the B-mode polarization measurements for the primordial gravitational waves is on the order of a nano-Kelvin corresponding to $r\sim 0.001$; a precise control of systematic effects is needed. A main source of systematic effects in experiments that rely on pair-differencing, is caused by the different characteristics in a mutually orthogonal pair of detectors in the polarization sensitive orientation. The $1/f$ noise in the measurement system and from atmospheric fluctuation is also a source of the contamination in the large angular scales, where the primordial gravitational wave signal is prominent. The measurements with a polarization modulation employing a rotating half-wave plate (HWP) can mitigate those systematics. Another challenge is to separate the CMB polarization with respect to the polarized emission from our own Galaxy. We need to subtract more than 99% of the polarized foreground emission to achieve the target precision of $r$. The standard way to differentiate the foreground emission and the CMB is to make use of the difference in frequency spectra of the sources; the CMB is known to have a perfect blackbody while the foreground emission, such as the synchrotron and dust emissions, have spectra different from the CMB. In order to measure the difference of the spectrum shape, we need an optical system with broad frequency coverage. In the past, the HWP was first implemented to the CMB experiment by MAXIPOL [@maxipol], and has been followed by a number of CMB experiments, including ABS, EBEX, SPIDER, POLARBEAR [@abs; @ebex; @spider; @polarbear]. Upcoming experiments also plan to employ a similar system [@ysakurai_spie; @charles_spie; @so; @swipe]. The HWP is made of a birefringent material plate with an optic axis[@hecht] parallel to the surface. When the thickness of the plate is chosen properly, the phase difference between the ordinary and extraordinary electric waves passing through the plate becomes $\pi$ radians, i.e., a half wavelength. When the HWP is rotated with respect to a linear polarization-sensitive detector (in Figure \[fig:signal\_meas\]), the incident plane of polarization rotates at a rate of twice the HWP rotation angle, and the measured intensity by the detector appears at a rate of four times the HWP angle. The continuously rotating HWPs modulate the signal at four times the rotational frequency of the HWP. As a result, we can reconstruct linear polarization components $Q$ and $U$ from the signal modulated at four times the rotational frequency of a HWP with a single detector. Correspondingly, the requirement to match the detector properties between two detectors is greatly relaxed. The retardance of a waveplate can be written as $\delta=2\pi \frac{\Delta n d}{\lambda}$, where $\Delta n = |n_{\mathrm{e}}-n_{\mathrm{o}}|$ is the difference between the indices $n_{\mathrm{o}}$ and $n_{\mathrm{e}}$ for the ordinary and extraordinary rays, respectively, $d$ is the thickness, and $\lambda$ is the wavelength. The single HWP, which is made of a birefringent material plate, can be used only at the specific wavelength and its harmonics determined by the material and thickness. While the single HWP is generally a single-frequency device, Pancharatnam proposed to stack multiple wave plates to broaden the frequency range [@pancharatnam_1; @pancharatnam_2]. In this paper, we describe our prototype design of the Pancharatnam-based achromatic HWP (AHWP) composed of nine sapphire plates for use in the CMB polarimetry. The design and the experimental demonstration appear in Hanany et al. (2005), Savini et al. (2006), and Pisano et al. (2006) [@hanany; @savini; @pisano]. We designed and constructed the prototype AHWP and evaluated it experimentally in a millimeter wave band, from 33 to 260 GHz, which is the widest demonstrated bandwidth at millimeter-wave. We discussed the results including all the features which we have observed in the measured modulation efficiency. This development is motivated to develop a broadband HWP for the next-generation CMB polarization satellite, LiteBIRD [@litebird; @litebird_new]. The initial design of the observational frequency band of the low-frequency telescope (LFT) was from 34 to 270 GHz, and thus we aim for this range as a development goal. After the design iteration, now LiteBIRD LFT covers from 34 to 161 GHz [@litebird_new]. As a result, we present the results of the development that cover the wider range as compared to the current LiteBIRD LFT frequency coverage. And in this paper, when referring to the frequency range or band of LiteBIRD, it refers to the old one. FORMALISM {#sec:form} ========= We construct the system in Figure \[fig:signal\_meas\], where the power of polarized light after passing through a continuous rotating AHWP is measured by a single polarization sensitive detector. The polarization state is expressed by using Stokes vectors and Mueller matrices as, $$S_{\rm out}=GR(-\omega_{\rm hwp} t)\Gamma_{\rm AHWP} R(\omega_{\rm hwp} t)S_{\rm in}, \label{eq:Sout}$$ $$R(\rho)=\left( \begin{array}{cccc} 1&0&0&0\\ 0&\cos2\rho&-\sin2\rho&0 \\ 0&\sin2\rho&\cos2\rho&0 \\ 0&0&0&1 \end{array} \right), \label{mt:rotation_mueller}$$ $$G=\frac{1}{2}\left( \begin{array}{cccc} 1&1&0&0\\ 1&1&0&0 \\ 0&0&0&0 \\ 0&0&0&0 \end{array} \right), \label{mt:polarizer}$$ $$\Gamma_{\rm AHWP}=\left( \begin{array}{cccc} M_{\rm II}&M_{\rm IQ}&M_{\rm IU}&M_{\rm IV} \\ M_{\rm QI}&M_{\rm QQ}&M_{\rm QU}&M_{\rm QV} \\ M_{\rm UI}&M_{\rm UQ}&M_{\rm UU}&M_{\rm UV} \\ M_{\rm VI}&M_{\rm VQ}&M_{\rm VU}&M_{\rm VV} \end{array} \right), \label{mt:mueller_mt}$$ where $S_{\rm in}=(I_{\rm in}, Q_{\rm in}, U_{\rm in}, V_{\rm in})$ and $S_{\rm out}=(I_{\rm out}, Q_{\rm out}, U_{\rm out}, V_{\rm out})$ are the Stokes vectors of the incident and outgoing radiation. $\Gamma_{\rm AHWP}$ is the Mueller matrix of an AHWP, $R$ is the rotation matrix, $\rho$ is the rotation angle of the AHWP with respect to the detector coordinate, $G$ is Mueller matrix for a polarizer which defines the polarization-sensitive orientation of the detector, $t$ is time, and $\omega_{\rm hwp}$ is the angular frequency of the AHWP rotation. ![\[fig:signal\_meas\] A conceptual sketch of the three-layer AHWP polarimetry. The linearly polarized plane waves propagate from left to right in this figure. An example of the modulated signal resulting from one revolution of the HWP is shown in the most right hand side. The amplitude is related to the polarized intensity and the phase is related to the polarization angle of the incident radiation.](concept_modsignal_meas.pdf){width="\hsize"} The circular polarization of the CMB is negligible. The circular polarization level for the Galactic synchrotron emission is estimated to be about 10 nK at most for 10 GHz frequency [@king2016] and about 4 pK at 100GHz assuming the frequency dependence of a power of -3.5 in the Rayleigh-Jeans units (a factor of 1.29 is different between Rayleigh-Jeans units and blackbody temperature at 100GHz). The circular polarization due to the dust is considered to be even lower [@Montero_Camacho_2018]. In summary, the circular polarization from foreground components is negligible compared to the CMB B-mode polarization of about a few nK. Therefore, we set $V_{\rm in}=0$ in this paper. For the normal incidence, the detected signal, $I_{\rm out}$, can be written as a function of time as: $$\begin{split} I_{\rm out}(t)=&D_{\rm 0I}I_{\rm in}+D_{\rm 0Q}Q_{\rm in}+D_{\rm 0U}U_{\rm in} \\ &+D_{\rm 2I}I_{\rm in}\cos(2\omega_{\rm hwp} t-2\phi_{\rm 0})+D_{\rm 2}\sqrt{Q_{\rm in}^{2}+U_{\rm in}^{2}}\cos(2\omega_{\rm hwp} t-2\phi_{2}) \\ &\qquad \qquad \qquad \qquad \qquad \qquad +D_{\rm 4}\sqrt{Q_{\rm in}^{2}+U_{\rm in}^{2}}\cos(4\omega_{\rm hwp} t-4\phi_{\rm 4}), \\ \label{eq:Iout} \end{split}$$ where, $$\begin{split} D_{\rm 0I}&=\frac{1}{2}M_{\rm II} \\ D_{\rm 0Q}&=\frac{1}{4}(M_{\rm QQ}+M_{\rm UU}) \\ D_{\rm 0U}&=\frac{1}{4}(M_{\rm QU}-M_{\rm UQ}) \\ D_{\rm 2I}&=\frac{1}{2}\sqrt{M_{\rm UI}^{2}+M_{\rm QI}^{2}} \\ \phi_{\rm 0}&=\frac{1}{2}\arctan\frac{M_{\rm UI}}{M_{\rm QI}} \\ D_{\rm 2}&=\frac{1}{2}\sqrt{M_{\rm IQ}^{2}+M_{\rm IU}^{2}} \\ \phi_{\rm 2}&=\frac{1}{2}\arctan\frac{M_{\rm IU}}{M_{\rm IQ}} +\frac{1}{2}\arctan\frac{U_{\rm in}}{Q_{\rm in}} \\ D_{\rm 4}&=\frac{1}{4}\sqrt{(M_{\rm QQ}-M_{\rm UU})^{2}+(M_{\rm QU}+M_{\rm UQ})^{2}}\\ \phi_{\rm 4}&=\frac{1}{4}\arctan\frac{M_{\rm QU}+M_{\rm UQ}}{M_{\rm QQ}-M_{\rm UU}}+\frac{1}{4}\arctan\frac{U_{\rm in}}{Q_{\rm in}}. \\ \end{split}$$ From the demodulation at $4\omega_{\rm hwp}$, we can extract the polarization power of the incident light. In order to demonstrate the performance of the AHWP and to compare it with the model, we define the modulation efficiency, $\epsilon$, as: $$\epsilon=\frac{D_{\rm 4}\sqrt{Q_{\rm in}^{2}+U_{\rm in}^{2}}}{D_{\rm 0I}I_{\rm in}+D_{\rm 0Q}Q_{\rm in}+D_{\rm 0U}U_{\rm in}}.$$ The modulation efficiency is the ratio between the signal power that is modulated at 4 times frequency of the HWP rotational frequency to the detected power. There are two reasons why we defined the modulation efficiency in this way. First, we implicitly assume that the AHWP is going to be rotated continuously and thus, we only pick up the term which is relevant to the 4 times of the rotation frequency already. Secondly, we want to define the efficiency in such that we can make a comparison between the measured degree of polarization and the prediction. This is driven by the fact that it is easy to prepare the fully polarized incident source using a wire grid polarizer. The connection between the lab measured efficiency and the CMB analysis is addressed in T. Matsumura et al. [@tmatsumura]. We also use $\phi_{4}$ as the phase of the modulated signal to compare the calculation with the measured data. For an ideal single HWP, $\Gamma$ becomes Eq. \[mt:birefringent\_wo\_refl\], we recover the following expression [@tmatsumura] $$\epsilon=\frac{\sin^{2}\frac{\delta}{2}\sqrt{Q_{\rm in}^{2}+U_{\rm in}^{2}}}{I_{\rm in}+\cos^{2}\frac{\delta}{2}Q_{\rm in}}.$$ In the case of $(Q_{\rm in}, U_{\rm in})=(0,1)$, the modulation efficiency is simply proportional to $\sin^{2}\frac{\delta}{2}$. In our development, we aim for modulation efficiency greater than 0.98 in a frequency range of 34-270 GHz with an incident angles up to 10 degrees for linearly-polarized incident light. The value of the modulation efficiency 0.98 is the requirement of LiteBIRD LFT telescope. The choice of the maximum incident angle is driven by the size of the field-of-view in LiteBIRD LFT telescope [@kashima]. SAMPLE PREPARATION {#sec:sample_prep} ================== Design optimization ------------------- The AHWP consists of multiple sapphire plates with optic axes relatively rotated among the plates. We use sapphire as the birefringent material for the HWP. Sapphire has superior optical and thermal properties for a HWP: about 10% difference in the refractive indices between the ordinary and the extraordinary rays [@b.r.johnson], the low loss-tangent at millimeter-wave frequency, and the high thermal conductivity, $10^2$-$10^3$ W/K/m, at a temperature of 4 - 10 K [@frank_pobell]. According to M. N. Afsar [@afsar], the variation of the refractive index of sapphire in the frequency range of 60-400 GHz is less than 0.1% for both ordinary and extraordinary rays. In this paper, we assume the refractive indexes are constant at 34-270 GHz. We have conducted the design optimization for the AHWP using a simulation. The Mueller matrix of $N$-layer AHWP is described as $$\Gamma_{\rm AHWP}=\prod_i^N R(-\chi_{i})\Gamma R(\chi_{i}), \label{mt:mueller_ahwp_wo_refl}$$ where $\Gamma$ is the Mueller matrix of a single birefringent plate, $\chi_{i}$ is the orientation of the optic axis with respect to the $x$-axis. We assume that all plates of which the AHWP is composed have the same thickness $d_\mathrm{ c } = \frac{1}{2}\frac{c}{\Delta n \nu_\mathrm{ c } }$, and the same refractive indices, where $\nu_\mathrm{ c } $ is the center of the frequency band and $c$ is the speed of light. In this process, we did not consider any effects of reflection to save computational time. After the completion of the optimization, we considered the performance including the reflection using the calculation referring to T. Essinger-Hileman [@essinger]. Without any reflections, the Mueller matrix of a birefringent material is give in Eq. \[mt:birefringent\_wo\_refl\] with the retardance of $\delta$, $$\Gamma=\left( \begin{array}{cccc} 1&0&0&0\\ 0&1&0&0\\ 0&0&\cos\delta&-\sin\delta\\ 0&0&\sin\delta&\cos\delta \end{array} \right). \label{mt:birefringent_wo_refl}$$ The design optimization is carried out using a brute force method. For the frequency range of 34 - 270 GHz, the center frequency is given as $\nu_c=152$ GHz, and the corresponding thickness is calculated to be 3.14 mm, where we use $n_\mathrm{o}=3.047$ and $n_\mathrm{e}=3.361$ for the refractive indices at low temperature [@b.r.johnson] and we assume lossless sapphire plates. We use the averaged modulation efficiency as the figure-of-merit for this optimization. We generate random numbers for the relative optic axis angle to find an optimal set of the angles offering the high averaged modulation efficiency in the frequency range. We repeated the optimization using various initial starting value to search the wide range of the parameter space in the case of $S_{\rm in}=(1, 0, 1, 0)$. As a result we have concluded that the nine-layer AHWP can cover almost all of the targeted bandwidth. Figure \[fig:design\] shows the modulation efficiency and phase as a function of frequency with the optimized design for a single plate, the three- and nine-layers. Table \[tab:design\] shows the optimized values of the relative angles in nine-layer AHWP. ![\[fig:design\] Calculated modulation efficiency and phase as a function of frequency for the optimized nine-layer AHWP design. For comparison, we also show the calculated result of a single HWP and three-layer AHWP. We calculate for $S_{\rm in}=(1,0,1,0)$.](modeff_phase_design_ne3361_no3047_t314_ain45.pdf){width="\hsize"} number of plates $d_\mathrm{ c } $ \[mm\] $\chi_{i} [^{\circ}]$ ------------------ -------------------------- ---------------------------------------------------- 3 3.14 0, 58, 0 9 3.14 0, 18.5, 37.5, 73.9, 141.5, 73.9, 37.5, 18.5, 22.7 : \[tab:design\] Designed values of the relative angles in the three- and nine-layer AHWP. The thickness of each plate is identical. $\chi_{i}$ is the optic axis angle of the $i$-th plate relative to the first layer. The design of the three-layer AHWP from T. Matsumura at al. [@tmatsumura]. Fabrication ----------- We have fabricated the AHWP based on the optimized parameters. We use a commercially available sapphire sample, with a diameter of 100 mm. Although the optimized thickness is 3.14 mm for a targeted frequency range, we have used sapphire plates with a thickness of 2.53 mm, which was readily available. We believe that this slight difference in thickness is not critical for demonstration purposes. We measure the thickness of the sapphire along the circumference of the disk, and the variation of the thickness within the sample is found to be less than 8 $\mu$m. The surface condition is unpolished. We have stacked the plates without glue at the interface of two plates, and fixed them with an aluminum holder as shown in Figure \[fig:9AHWP\_pic\]. The anti-reflection coating is not applied on any surfaces. To stack the sapphire plates, we use the universal measurement machine (UMM). The UMM consists of a microscope and a rotating table. Each sapphire plate has an orientation flat (OF) at its side that is in perpendicular to the optic axis and can be used as the reference of the optic axis with the accuracy $3^{\circ}$. This accuracy is due to the uncertainty of a dicing capability in a sapphire manufacturing. Sapphire plates are stacked in the aluminum holder while adjusting with respect to OF to the designed orientation of the optic axis of each plate. Figure \[fig:9AHWP\_pic\] shows the assembled nine-layer AHWP. Sapphire plates are fixed in the holder with pressure applied by an aluminum ring. The relative angular uncertainty between the OF of the plates is less than 10 arcmin. The implication of this uncertainty is addresssed in Section 6.1.1. After assembly, we evaluate the presence of air gaps between the layers. We investigate the thickness of the air gap by inserting several thin stainless steel plates with varying thickness ($\geq$50 $\mu$m), between layers of the AHWP. We identify that the air gap between the first and second layer is around 50 $\mu$m, and that air gaps between other layers are less than 50 $\mu$m. We will account for the presence of the gap in the analysis. ![The assembled nine-layer AHWP.[]{data-label="fig:9AHWP_pic"}](9AHWP_sample.pdf){width="0.5\hsize"} EXPERIMENT {#sec:exp} ========== Experimental setup ------------------ Figure \[fig:setup\] shows the setup for measuring the modulation efficiency and the transmittance in the frequency ranges of 33-140 GHz and 150-260 GHz. The millimeter waves are generated by a Continuous Wave (CW) generator and six different active multipliers. The CW generator can generate microwaves up to 20 GHz. Active multipliers up-convert the frequency of the signal from the CW generator. The multiple factors and bandwidths of individual active multipliers are $\times$4 (33-50 GHz), $\times$4 (50-75 GHz), $\times$6 (75-110 GHz), $\times$8 (90-140 GHz), $\times$12 (150-220 GHz) and $\times$24 (210-260 GHz). We pair the active multiplier with a diode detector for the measurements in each band. Two feedhorns for the source (the active multiplier) and for the detector are placed at the foci of the off-axis parabolic mirrors. The millimeter waves emitted from the source horn are collimated by the first mirror. The plane waves propagate through the first attenuator, the first wire grid, the 70 mm diameter aperture, the sample to measure, the second attenuator, and the second wire grid. The plane waves are focused by the second mirror to be fed to the detector horn. The millimeter wave from the source is linearly polarized. The detector used is a single polarization-sensitive detector. We use two free-standing wire grids to define the polarization angle of incident light to the sample. The transmittance of the wire grid for the light having the electric field perpendicular (parallel) to the wire orientation is 0.99 (0.01) in our measurement frequency range. The signal is modulated by an optical chopper with a frequency of 80 Hz to be amplified by a lock-in amplifier. The detector outputs the detected power in voltage. We set the transmission axis of the two wire grids in parallel. All the measurements were performed at room temperature. An optical measurement of AHWP at cryogenic temperatures is also in progress and it is beyond the scope of this paper. ![A sketch of the measurement system. The millimeter waves propagate along the orange lines from the right hand to the left hand side of the figure. The aperture size is approx. 70 mm in diameter.[]{data-label="fig:setup"}](IPMU_system_v4.pdf){width="0.8\hsize"} Measurement of refractive index ------------------------------- To predict the modulation efficiency, we need to know the two refractive indices of an A-cut sapphire plate at the room temperature. We obtain the indices by measuring transmittance, i.e. Fabry-Pérot interference, using the same setup in Figure \[fig:setup\]. When the plane of polarization of the incident wave and the detector-sensitive direction are parallel to the optic axis, transmittance $T$ and the complex refractive index $\tilde{n}$ of an extraordinary (ordinary) ray are simply related by Eq. \[eq:trans\] for the case of the normal incidence [@hecht]. Since the loss tangent of sapphire is small enough to use the approximation, $\tilde{n}$ is expressed using the refractive index $n$ and the loss tangent $\tan\delta$ in Eq. \[eq:c\_index\]. The variable $d$ is the thickness of the sapphire plate and $k_{0}$ is the wave number in the vacuum, $$\begin{aligned} \label{eq:trans} T(\tilde{n})&=&\left| \frac{2\tilde{n}}{2\tilde{n} \cos k_{0}\tilde{n}d + i(\tilde{n}^{2}+1) \sin k_{0}\tilde{n}d} \right|^{2}, \\ \label{eq:c_index} \tilde{n}&\simeq&n \left(1-\frac{i}{2}\tan\delta \right).\end{aligned}$$ We measure the indices of one sapphire plate, and assume the same indices for the rest, which is valid because all the samples are originated from the same batch. We set the sapphire plate to the sample holder in Figure \[fig:setup\]. We then measure output voltages of the lock-in amplifier at 33 to 140 GHz and 150 to 260 GHz in $\sim$ 1 GHz interval. After removing the A-cut sapphire plate, we remeasure output voltages of the lock-in amplifier for the same frequency range with the same frequency step. The source is a coherent source, and thus the measurement system is susceptible to the effect of the standing wave. We mitigate the effect of a standing wave by moving the detector along the optical path by $\lambda/4$ at each frequency. The spectral shape of transmittance for polarization parallel and perpendicular to the optic axis are computed by taking a ratio of the acquired data between the sapphire plate case and the air case. Measurement of modulation efficiency ------------------------------------ The nine-layer AHWP is mounted on a sample holder which can be automatically rotated by a stepping motor. The sample holder continuously rotates around the optical axis in this system with a revolution rate of $\omega_{\rm hwp}=2\pi f_{\rm hwp}$, where $f_{\rm hwp}$ is about 0.02 Hz. The frequency of the electromagnetic source is swept during the rotation. We measure the modulated signal as the output voltage of the lock-in amplifier. The measured frequency range is 33 to 140 GHz and 150 to 260 GHz in $\sim$ 1 GHz interval. The measurement time at each frequency is 60 seconds, during which the AHWP rotates about 360 degrees. The sampling rate of the demodulated signal from the lock-in amplifier is 100 Hz. For each frequency, we fit the acquired data using Eq. \[eq:fit\_TOD\]. $$I(t, \nu) = a_{0}(\nu)+\sum^{8}_{m=1}a_{m}(\nu)\cos{(m \omega_{\rm hwp}t+ m\phi_{m}(\nu))}. \label{eq:fit_TOD}$$ While acquiring data at all frequencies, the AHWP keeps on continuously rotating. The initial value of $\phi_{n}$ is different for each frequency. This offset is recorded and subtracted for each $\phi_{n}$ at given frequyency. The $m=4$ component is the modulation signal of the AHWP. The $m=2$ component appears due to the differential indices which results the frequency dependent transmittance, reflectance, and emissivity between for the ordinary and extraordinary rays of the HWP. The other components are included to capture all the features even though we do not expect the odd $m$ components within the framework of the formalism in this paper. We will address more on this point in the section \[sec:discuss\]. The modulation efficiency and the phase are obtained as $a_{4}/a_{0}$ and $\phi_{4}$, respectively. The initial value of $\phi_{4}$ is determined by the initial rotation angle of the sample holder. We have repeated the measurements for incident angles reletive to the AHWP of 0 and $\pm10^{\circ}$ for $p$- and $s$-polarization, which corresponds to the field-of-view of LiteBIRD and the CMB telescope that observes small angle scales. In many CMB experiments, the intensity of the observation signal is integrated by the detector with a specific bandwidth. Therefore, we introduce the band average modulation efficiency to evaluate the integrated modulation signal. For the evaluation, we use the frequency bands centered at 40, 50, 60, 68, 78, 89, 100, 119, 140, 166, 195, and 235 GHz with the bandwidth of about 30%, that are covered by the LiteBIRD LFT. We normalize $I(t, \nu)$ by $a_{0}$ for each frequency, and integrate this normalized modulated signals as $$\int_{\nu_{i}}^{\nu_{f}} \frac{I(t,\nu)}{a_{0}(\nu)} d\nu = \sum^{\nu_{f}}_{\nu=\nu_{i}}\frac{I(t,\nu)}{a_{0}(\nu)} = A_{0}+\sum^{8}_{m=1}A_{m}\cos{(m \omega_{\rm hwp}t+ m\phi_{m})}, \label{eq:integ_TOD}$$ where $\nu_{i}$ and $\nu_{f}$ are the lower and higher boundary in each frequency band, respectively. We define $A_{4}/A_{0}$ as the band-averaged modulation efficiency. RESULTS {#sec:result} ======= Measurement of refractive index ------------------------------- From the fitting result for the frequency-dependent transmittance of the sapphire plate using Eq. \[eq:trans\] (Figure \[fig:trans\]), we obtain the values of the refractive index and the loss tangent at room temperature for the A-cut sapphire plate, which are summarized in Table \[tab:no\_and\_ne\]. In Figure \[fig:trans\], the difference between the fit and the measured data becomes larger at the lower frequency. We think this is caused by the stability of the source or the effect of the diffraction at the aperture, which prevents the full cancellation of the standing effect. (Figure \[fig:setup\]). ![\[fig:trans\] The transmittance for the ordinary ray (left) and the extraordinary ray (right) for a A-cut sapphire plate. The top plot shows the measurement (the red dots with error-bars) and fitted (the blue solid line) results. The bottom plot shows the residuals of the fitting. ](Trans.pdf){width="\hsize"} --------------------- --------------------------------- --------------------- --------------------------------- -- -- Refractive index Loss tangent ($\times 10^{-4}$) Refractive index Loss tangent ($\times 10^{-4}$) $ 3.059 \pm 0.002 $ $ 0.9 \pm 0.3 $ $ 3.397 \pm 0.003 $ $ 1.6 \pm 0.5$ --------------------- --------------------------------- --------------------- --------------------------------- -- -- : Fitted result to the refractive index and loss tangent for an A-cut sapphire plate at the room temperature. []{data-label="tab:no_and_ne"} Measurement of modulation efficiency ------------------------------------ Figure \[fig:signal\] shows one example for the output voltages of the lock-in amplifier and fitted result using Eq. \[eq:fit\_TOD\] as a function of the rotation angle at 150 GHz. The output voltage is normalized by the DC $m=0$ component. Since the lock-in amplifier can not output a negative voltage, the output signal is inverted when the signal becomes negative due to noise or offset. This causes the amplitude of the modulation signal to be reduced and causes the modulation efficiency to be underestimated. Therefore, the part of the modulated signal close to zero is removed. For all frequencies and incident angles, we confirm that the residual is less than 3% (in RMS) of the DC $m=0$ component. Figure \[fig:overview\] shows the frequency dependence of the modulation efficiency and the phase for each incident angle $\theta$ with $p$- and $s$-polarization. The prediction for the normal incidence takes into account the reflections at plates and does not consider the air gaps between the plates. In the prediction, we use the values of $n_\mathrm{o}$ and $n_\mathrm{e}$ at the room temperature in Table \[tab:no\_and\_ne\]. We can see two features in Figure \[fig:overview\]: the sharp dips that appear at about every 18 GHz, and the fast oscillatory features that fluctuate quickly and with a small amplitude. The dips originate from Fabry-Pérot interference within each plate that composes the AHWP. By contrast, the oscillatory feature is from the reflection at the boundaries between the first/last plate and air. Table \[tab:modeff\] shows the measured band-averaged modulation efficiency and Table \[tab:phasediff\] shows the maximum difference of the phase variation within a bandwidth, which is defined as the LiteBIRD frequency band. About the error source of the modulation efficiency is discussed in section 6. ![The output voltages of the lock-in amplifier as a function of the HWP rotation angle at 150 GHz for the normal incident angle. The output voltage is proportional to the millimeter wave power injected into the detector and normalized by $a_{0}$. The top plot shows the measurement (the blue crosses) and fitted (the orange solid line) results. The bottom plot shows the residuals of fitting.[]{data-label="fig:signal"}](150GHz_p0.pdf){width="0.6\hsize"} ![\[fig:overview\] The modulation efficiency and the phase from 33 to 260 GHz are plotted in $\sim$ 1 GHz interval, where $\theta$ is the incident angle of the millimeter waves for the AHWP. The predictions are plotted in in 0.2 GHz interval. The top side panels show the comparison of the measured data and the prediction for normal incidence. The bottom side panels show the comparison of normal and oblique incidence.](9AHWP_modeff_phase_wogap.pdf){width="\hsize"} -------------- -------------------- --------------------- --------------------- ---------------------- ---------------------- ---------- -- $\theta=0^{\circ}$ $\theta=10^{\circ}$ $\theta=10^{\circ}$ $\theta=-10^{\circ}$ $\theta=-10^{\circ}$ band \[GHz\] bandwidth \[%\] (p-pol.) (s-pol.) (p-pol.) (s-pol.) 40 30 0.902 0.895 0.892 0.897 0.900 50 30 0.961 0.960 0.959 0.961 0.960 60 23 0.971 0.970 0.970 0.970 0.971 68 23 0.969 0.969 0.969 0.969 0.970 78 23 0.976 0.975 0.976 0.976 0.977 89 23 0.981 0.981 0.981 0.982 0.982 100 23 0.985 0.985 0.986 0.985 0.986 119 30 0.984 0.983 0.983 0.983 0.984 140 30 0.984 0.984 0.984 0.984 0.984 166 30 0.983 0.984 0.983 0.984 0.983 195 30 0.979 0.979 0.978 0.979 0.979 235 30 0.959 0.955 0.955 0.954 0.954 -------------- -------------------- --------------------- --------------------- ---------------------- ---------------------- ---------- -- : The measured band-averaged modulation efficiency within the bandwidth for the nine-layer AHWP at each incident angle. []{data-label="tab:modeff"} -------------- -------------------- --------------------- --------------------- ---------------------- ---------------------- ---------------- $\theta=0^{\circ}$ $\theta=10^{\circ}$ $\theta=10^{\circ}$ $\theta=-10^{\circ}$ $\theta=-10^{\circ}$ band \[GHz\] bandwidth \[%\] (p-pol.) (s-pol.) (p-pol.) (s-pol.) 40 30 $4.86^{\circ}$ $5.11^{\circ}$ $5.34^{\circ}$ $5.05^{\circ}$ $4.77^{\circ}$ 50 30 $6.86^{\circ}$ $7.25^{\circ}$ $6.78^{\circ}$ $7.10^{\circ}$ $6.64^{\circ}$ 60 23 $5.20^{\circ}$ $5.41^{\circ}$ $5.08^{\circ}$ $5.56^{\circ}$ $5.01^{\circ}$ 68 23 $3.90^{\circ}$ $4.17^{\circ}$ $3.82^{\circ}$ $3.97^{\circ}$ $3.86^{\circ}$ 78 23 $2.52^{\circ}$ $2.86^{\circ}$ $2.74^{\circ}$ $2.65^{\circ}$ $2.47^{\circ}$ 89 23 $0.82^{\circ}$ $0.79^{\circ}$ $0.68^{\circ}$ $0.69^{\circ}$ $0.74^{\circ}$ 100 23 $0.95^{\circ}$ $0.94^{\circ}$ $0.83^{\circ}$ $0.89^{\circ}$ $1.05^{\circ}$ 119 30 $3.15^{\circ}$ $3.05^{\circ}$ $3.25^{\circ}$ $3.13^{\circ}$ $3.22^{\circ}$ 140 30 $1.83^{\circ}$ $1.83^{\circ}$ $1.97^{\circ}$ $1.86^{\circ}$ $1.82^{\circ}$ 166 30 $1.49^{\circ}$ $1.51^{\circ}$ $1.41^{\circ}$ $1.51^{\circ}$ $1.74^{\circ}$ 195 30 $2.08^{\circ}$ $2.31^{\circ}$ $1.74^{\circ}$ $1.98^{\circ}$ $2.00^{\circ}$ 235 30 $7.12^{\circ}$ $7.65^{\circ}$ $6.63^{\circ}$ $7.37^{\circ}$ $6.84^{\circ}$ -------------- -------------------- --------------------- --------------------- ---------------------- ---------------------- ---------------- : The maximum difference of the phase variation within the bandwidth for the nine-layer AHWP at each incident angle. []{data-label="tab:phasediff"} DISCUSSION {#sec:discuss} ========== Sources of error ---------------- Figure \[fig:overview\] shows that the prediction and the measurement data are in good agreement for both modulation efficiency and phase. Nevertheless, we identify some discrepancies between the measurement results and the predictions. Here, we discuss possible sources of the the discrepancies between the measurement results and the predictions. ### Relative angular uncertainties In the fabrication of the nine-layer stacked AHWP using the UMM, we find the relative angle position error of OF is less than 10 arcmin. The OF can be used as the reference of the optic axis with an accuracy of $3^{\circ}$ (180 arcmin). Therefore the angular position uncertainty of the optic axis of the $i$-th plate is less than $(i-1)\times 190$ arcmin. Here we consider a conservative case; all the plates have angular position shifts of 190 arcmin relative to the former plate in the same direction. Figure \[fig:angle\_err\_effect\] shows the comparison of the modulation efficiency and the phase with and without this angular position shifts. The bottom plot shows the difference between them. In this calculation, we ignore the air gaps and fix all the other parameters to their designed values. From the comparison result, the differences are found to be less than 0.33 (0.06 in RMS) for the modulation efficiency and less than $17^{\circ}$ ($14^{\circ}$ in RMS) for the phase. This $14^{\circ}$ in RMS is obtained in a very conservative way. In the fabrication of the AHWP used for the LiteBIRD observation, we expect to use the sapphire plates produced from the same ingot and define each optic axis in higher accuracy by optical measurement nor X-ray without referring OF. The position determination precision in stacking the plates is expected to be order of 10 arcmin with random variation. In addition to this, we can calibrate the modulation efficiency and the phase in the ground facility with an accuracy sufficient to remove the foreground components with a required level. ![Calculated results of the modulation efficiency and the phase for the nine-layer AHWP with and without the angular position shifts. The bottom plots show the differences from the case with no uncertainty.[]{data-label="fig:angle_err_effect"}](modeff_phase_no3059_ne3397_t2530_angle_err_v2.pdf){width="\hsize"} ### Thickness uncertainty of sapphire plates We estimate the thickness uncertainty in individual sapphire plates to be $\pm$4 $\mu$m from measurements of thickness variation along the circumference of the disk. We therefore calculate the modulation efficiency and the phase by assuming a plate thickness of 2.534 mm and 2.526 mm and compare it with the calculated result of the plate thickness of 2.530 mm. We again ignore the air gaps and set all the other parameters to designed values. From the comparison, the uncertainties of the modulation efficiency and the phase are estimated to be less than 0.05 (0.007 in RMS) and $0.4^{\circ}$ ($0.06^{\circ}$ in RMS), respectively. Note that the finite thickness itself of the AHWP can result a potential systematic effect. In this paper, the total thickness of the AHWP is about 23 mm. When such a large thickness AHWP is employed with a converging or a diverging optical system, the focus position might be affected. In case of LiteBIRD, the AHWP is placed as a first optical element, and thus the incident radiation is in parallel. As a result, this particular issue is not a relevant unless we further look into the higher order effect. This is beyond the AHWP design issue but rather a system wide design issue, which depends on what optical system to use with the AHWP. Therefore, we do not address further in this paper. ![Calculated results of the modulation efficiency and the phase of the nine-layer AHWP for the plate thicknesses of 2.530 mm, 2.534 mm and 2.526 mm. The bottom plot shows the differences of those values from the ones with 2.530 mm.[]{data-label="fig:thickness_err_effect"}](modeff_phase_no3059_ne3397_t2530_thickness_err.pdf){width="\hsize"} ### Refractive indices uncertainty From Table \[tab:no\_and\_ne\], we estimate the difference between the two refractive indices of the A-cut sapphire to be $\Delta n = 0.338 \pm 0.005$. We compare the calculation results that $\Delta n$ set to 0.343, 0.333 and 0.338. Here we ignore the air gaps and set all the other parameters to the designed values. With the comparison, we find the uncertainties in the modulation efficiency and phase to be 0.03 (0.003 in RMS) and $0.4^{\circ}$ ($0.2^{\circ}$ in RMS), respectively. ![Calculated results of the modulation efficiency and the phase of the nine-layer AHWP for the refractive index differences of 0.338, 0.343 and 0.333. The bottom plot shows the difference of them from the ones with 0.338.[]{data-label="fig:index_err_effect"}](modeff_phase_no3059_ne3397_t2530_index_err.pdf){width="\hsize"} ### The effect of air gaps After the assembly of the AHWP, we identify the air gap between the first and second layer to be around 50 $\mu$m, and air gaps between other plates are found to be less than 50 $\mu$m. We compare the modulation efficiency and the phase of the nine-layer AHWP in calculation with and without the 10 $\mu $m and 50 $\mu $m air gaps between all plates. Figure \[fig:airgap\_effect\] shows the comparison results. We find no difference in the modulation efficiency and the phase around 175.3 GHz. This is because the transmittance of all the sapphire plates is close to unity at this frequency and the air gaps do not contribute to the reflections at the boundaries of the plates. On the other hand, Figure \[fig:airgap\_effect\] implies that the dips and oscillatory features seem to depend on the magnitude of the air gap. This is because those two features originate from the reflections in the AHWP and the air gaps affect those reflections. For example, on the high frequency side, the depth of the dips and oscillatory feature monotonically increases according to the thickness of the air gap. On the low frequency side, where the thickness of the air gap is small, the dip depth is decreased, and where the thickness becomes large, it starts to increase. This trend is consistent with the difference between the prediction and the measured data in Figure \[fig:overview\]. The air gaps cause the changes in the modulation efficiency and phase to be 0.9 in maximum (0.1 in RMS) and $14^{\circ}$ at maximum ($2^{\circ}$ in RMS), respectively. We find that the air gaps affect the modulation efficiency and phase around the dips significantly. ![Dependence of the modulation efficiency and the phase on the air gaps. The bottom plot shows the difference of them from the ones without the air gap.[]{data-label="fig:airgap_effect"}](modeff_phase_no3059_ne3397_t2530_gap_effect_v2.pdf){width="\hsize"} ### Summary of sources of error From the consideration in this subsection, we find that the air gaps are the largest source of the change in the AHWP performance around the dips. Figure \[fig:overview2\] shows a comparison of the measured data and the prediction that takes into account the largest source, air gaps. In Figure \[fig:overview2\], the prediction taking the measured air gaps into account reproduces the tendency of the depth of the dip and oscillatory feature better than before. But the residuals of the prediction and the measured data is not decreased because of the frequency shift of the dips caused by the uncertainties of thickness and refractive indices. In the prediction, the air gaps are inserted between each plate as a parallel flat plate having a refractive index of 1. The thickness of the air gap between the first and the second plate is set to 50 $\mu$m and other gaps are set to 8 $\mu$m. The thickness of the air gaps other than those between the first and second plates is obtained from the thickness variation along the circumference of the plates. ![\[fig:overview2\] The modulation efficiency (left panels) and the phase (right panels) from 33 to 260 GHz are plotted in $\sim$ 1 GHz interval. The predictions takes into account the air gaps based on the measurements and are plotted in 0.2 GHz interval.](9AHWP_modeff_phase_diff_w_wo_gap.pdf){width="\hsize"} Amplitude of modulated signal for each mode ------------------------------------------- To obtain the modulation efficiency and the phase, we use Eq. \[eq:fit\_TOD\] for the fitting of the modulated signal. Here, we discuss the modes other than $m = 4$. The $m=2$ mode appears due to the difference of the frequency-dependenct transmittance between the two refractive indices in the HWP caused by the absence of the anti-reflection coating. Figure \[fig:overview\_2f\] shows a comparison of the measured data and the prediction of the $m=2$ mode which takes into account the air gaps. While we think the model of the air gap is not complete, we qualitatively recover the agreement between the prediction and the measured data. ![\[fig:overview\_2f\] The amplitude of $m=2$ mode from 33 to 260 GHz are plotted in $\sim$ 1 GHz interval. The predictions takes into account the air gaps and are plotted in 0.2 GHz interval.](9AHWP_2f_wgap.pdf){width="0.5\hsize"} The origin of other modes, $m=1, 3, 5, 6, 7, 8$, is not physically motivated within the formalism described in Eq. \[eq:Iout\]. The peak amplitude is generally signal-to-noise above 100 for $m=1, 3, 5, 6, 8$ in the range of above 40 GHz. Thus, the identified peaks are not due to the noise. One of the potential contributors to the peaks at $m \neq 2, 4$ is from the imperfection of the assembly. Figure \[fig:overview\_nf\] shows the measured amplitude of the mode at $m=1, 3, 5, 6, 7, 8$. For the mode $m=1, 3, 5$, we identify the general trend of higher amplitude as it extends to the higher frequency. This is generally consistent with the effect of the air gap. The modes at $m=6, 7, 8$, on the other hand, show the different tendency. The modes $m=6, 8$ originate from the refractive index for an extraordinary ray as a function of an incident angle. When an incident wave enters with an incident angle of $\theta$ and an azimuthal angle of $\phi$ with respect to the optic axis, the refractive index $n_\mathrm{e}^\prime (\theta,\phi)$ is given by[@Orfanidis] $$\begin{aligned} n_\mathrm{e}^\prime (\theta,\phi) = n_\mathrm{e} \sqrt{ 1 - \left( \frac{ 1 } { n_\mathrm{ o }^2 } - \frac{ 1 } { n_\mathrm{ e }^2 } \right) n_1^2 \sin^2 \theta \cos^2 \phi },\end{aligned}$$ where $n_1$ is an index of the ambient space. $n_\mathrm{e}^\prime (\theta,\phi)$ includes higher order cosines of $\phi$. It cannot account for the odd order cosines of $\phi$. This is the reason the modes $m=6, 8$ have the different tendency. However, the reason of the $m=7$ mode tendency remains unclear. We also looked at the correlation between $m=1$ and $m=3, 5$ as shown in Figure \[fig:comparison\_1f\_odd\]. If the majority of the effect is due to the air gap and the air gap has a wedge-like shape, we expect the rotational synchronous $m=1$ mode and potentially higher harmonics. Figure \[fig:comparison\_1f\_odd\] shows the positive correlation between the $m=1$ mode and the $m=3, 5$, which supports the idea of the effect of the wedge shape air gap and its harmonics. Candidates for other sources of $m=1, 3, 5$ modes are the rotational speed instability and the vibration and wobble in the rotation, and these are expected to be more pronounced at higher frequencies. The $m\neq4$ modes can be filtered out at the demodulation step. There is a potential leakage from $m=2$ to higher harmonics as a source of the conversion from unpolarized light to polarized light. Such a possibility can be addressed particularly when the incident angle is not normal incident to the HWP. The result is highly dependent on the performance of the AR coating, which is not accounted for in this paper. An example of the study can be found in T. Essinger–Hileman et al. [@essinger_spie] and H. Imada et al. [@imada_isstt2018] Thus, we decide not to explore beyond the identification of the existence of $m\neq4$ mode. ![\[fig:overview\_nf\] The $m=1, 3, 5, 6, 7, 8$ modes from 33 to 260 GHz are plotted in every 0.9 GHz.](compare_nf_even_and_odd_wo_phase.pdf){width="\hsize"} ![\[fig:comparison\_1f\_odd\] The correlation of $a_{1}/a_{0}$ and $a_{3}/a_{0}$, $a_{5}/a_{0}$ from 33 to 260 GHz. The data is every 0.9 GHz.](compare_1f_and_odd.pdf){width="0.6\hsize"} Dependency for incident angle ----------------------------- In Table \[tab:modeff\] we show the band-averaged modulation efficiencies with the incident angle of 10 degrees. The differences of the efficiencies between normal and the 10 degree oblique incidence are less than 0.005 (0.01) for the highest (lowest) bands. For the other bands, the differences are less than 0.001. The bottom side panels of Figure \[fig:overview\] show the modulation efficiency and the phase for the two incident angles, suggesting there is no significant difference between them. In order to understand the incident angle dependence more precisely, we repeat the measurement with a finer frequency step of 0.15 GHz between 230 and 240 GHz, and with incident angle of $\pm 5^{\circ}$. Figure \[fig:230-240\] shows the incident angle dependencies, and we find that there is a frequency shift of the dip. In case of the incident angle of $5^{\circ}$ ($10^{\circ}$), the refraction angle at the first plate of the AHWP is calculated to be about $1.5^{\circ}$ ($3^{\circ}$) for each ray. Since the difference in the refractive indices is small at the boundary between the two sapphire plates, the refraction angles within each plate are similar to the first plate. When $\theta \neq 0$, frequencies of Fabry-Pérot interference spectrum scaled by $(\cos \theta)^{-1}$ compared with the normal incidence ($\theta = 0$). That causes the frequency shift to the higher side. For the refraction angle of $1.5^{\circ}$ ($3^{\circ}$), the frequency shift is computed to be about 0.08 (0.33) GHz. The estimation of the frequency shift is in good agreement to the shift measured in the modulation efficiency and the phase. Therefore, we find the observed frequency shift can be explained by the incident angle dependence. The dependency for incident angle in AHWP is also related to the potential conversion from the incident unpolarized light to the polarized light. This is one of the important effect to be addressed in CMB experiments. H. Imada et al. [@imada_isstt2018] addressed this effect, and the further study to propagate this effect to quantify the impact to the cosmology is in progress. ![Frequency dependence of the modulation efficiency and the phase for 230 to 240 GHz. The data points of the measurement result are plotted for every 0.15 GHz.[]{data-label="fig:230-240"}](modeff_phase_230-240.pdf){width="\hsize"} Further design optimization --------------------------- In spite of the fact that the nine layer AHWP becomes broadband as expected, the band-averaged modulation efficiency is lower than 0.98 in some bands. We find three reasons why the band averaged modulation efficiency is lower than 0.98. The first two reasons are due to the hardware preparation, and the third one arises from the AHWP design. The first reason is the fact that the thickness of the sapphire plates used is slightly thinner than the optimized one (Table \[tab:design\]). This causes an overall frequency shift to higher frequency of the modulation efficiency. The second reason is caused by the large dips and oscillatory features caused by the air gaps, which also decreases the overall modulation efficiency when averaged over the band. Particularly, this influence is large on the higher frequency bands. The third reason is caused by the large phase variation on the higher and lower frequency bands (Table \[tab:phasediff\]). Table \[tab:modeff\_phase\_pre\] shows the prediction of the averaged modulation efficiencies and the maximum phase variation with reflection effects, no air gaps and the optimized thickness. The band averaged modulation efficiencies at the highest and lower frequency band are less than 0.98. This exception is due to the large phase variation in the frequency band. The phase variation decreases in the amplitude of the modulated signal that is integrated in the band and the band averaged modulation efficiency. Therefore, the averaged value becomes smaller, indicating that the phase variation is important. The less phase variation within an observational band is also important to minimize the systematic effect, which results the susceptibility of the polarization angle sensitive orientation of a HWP polarimeter to the incident radiation spectrum and the system spectral shape. Therefore, we study the further design optimization that minimizes the phase variation within the band while the modulation efficiency is kept high enough in preparation. band \[GHz\] bandwidth \[%\] band averaged modulation efficiency $\Delta \phi_{4}$ -------------- ----------------- ------------------------------------- ------------------- 40 30 0.969 $6.26^{\circ}$ 50 30 0.977 $7.49^{\circ}$ 60 23 0.982 $3.87^{\circ}$ 68 23 0.991 $2.07^{\circ}$ 78 23 0.995 $0.55^{\circ}$ 89 23 0.995 $0.85^{\circ}$ 100 23 0.989 $2.41^{\circ}$ 119 30 0.990 $1.87^{\circ}$ 140 30 0.993 $1.98^{\circ}$ 166 30 0.993 $2.01^{\circ}$ 195 30 0.990 $3.08^{\circ}$ 235 30 0.938 $16.46^{\circ}$ : The prediction of the averaged modulation efficiency and the maximum phase variation of the nine-layer AHWP with reflection effects, no air gaps, and optimized thickness. []{data-label="tab:modeff_phase_pre"} CONCLUSIONS {#sec:conclusion} =========== We design and evaluate the prototype of the nine-layer AHWP for use in CMB experiments. We find the measurements in the modulation efficiency and the phase at the room temperature are in good agreement with the predictions. Thus, we demonstrate our nine-layer AHWP to be broadband. However, we find small discrepancies between the measurements and the predictions. The primary contribution around the dips is attributed to the existence of air gaps between the stacked plates. The agreement gets even better when the effect is considered. We measure the incident angle dependence of the modulation efficiency and the phase in the incident angle range comparable to the field of view of LiteBIRD and the CMB telescope that observes small angle scales, about 10 degrees. The dependence of the incident angle is found to be explained by the internal reflections in individual plates. In many CMB experiments, the intensity of the observation signal is integrated by the detector in a specific bandwidth. Therefore, we average the modulation efficiency over the bandwidth to obtain the value usable for the experiments. From the evaluation using the band-averaged modulation efficiency, we find that the smaller the phase variation in the bandwidth, the larger the averaged modulation efficiency. Therefore, the optimization to obtain higher modulation efficiency requires us to have uniform phase values in the bandwidth, which will be presented in future work. This paper based on a SPIE conference proceedings paper [@kkomatsu_spie]. This work is supported by JSPS KAKENHI Grant Number JP17H01125, JP15H05441, JP15H05891, JP17K14272, JP18J20148 and JSPS Core-to-Core Program, A. Advanced Research Networks, and the World Premier International Research Center Initiative (WPI Initiative), MEXT, Japan. We would like to thank Dr. Samantha Stever for editorial suggestions to this paper. **First Author** is a PhD student at Okayama University. He received his BS and MS degrees in physics from Okayama University in 2016 and 2018, respectively. His current research interests the verification of inflation theory using B-mode polarization of CMB created by the primordial gravitational waves. Related to it, he is developing the polarization modulator of LiteBIRD. Biographies and photographs of the other authors are not available.

ArXiv

--- author: - Abdelghani Zeghib title: 'Quelques remarques sur les actions analytiques des réseaux des groupes de Lie de rang supérieur [[^1]]{}' --- amssym.def amssym.tex \[theoreme\][[Corollaire]{}]{} \[theorem\][[Th' eorème algébrique]{}]{}\[theorem\][[Th' eorème géométrique]{}]{} \[proposition\][[Lemme]{}]{} \[theorem\][[Proposition fondamentale]{}]{}\[theorem\][[Definition]{}]{} \[theorem\][[Fait]{}]{}\[theorem\][[Affirmation]{}]{}\[theorem\][[Remarque]{}]{}\[theorem\][[Remarque terminologique]{}]{}\[theorem\][[Remarques]{}]{}\[theorem\][[Question]{}]{}\[theorem\][[Questions]{}]{}\[theorem\][[Exemple]{}]{}\[theorem\][[Conjecture]{}]{}\[theorem\][[Question algébrique]{}]{}\[theorem\][[Projet]{}]{}\[theorem\][[Conjecture géométrique]{}]{}\[theorem\][[Hypothèse]{}]{} [ ]{}\ [**Abridged English version**]{}\ Let $\rho_0$ denote the standard affine action of $\Gamma$, a subgroup of finite index of $SL(n, {\Bbb Z})$, on the torus $T^n$. In order to prove that it is locally rigid, it is a natural idea to show that every $C^1$ nearby action $\rho$ is “linearizable”. From a result on persistance of fixed points of [@sto], and a local linearizability result of [@C-G] (in the analytic case), we get a local linearization of $\rho$ around some fixed point. The main ingredient is then, to introduce a [*Siegel neighborhood*]{} of the fixed point, which is a maximal [*invariant*]{} open set on which the action is linear. We show that the action on the Siegel neighborhood is $C^\omega$-conjugate to the standard action on a punctured torus, i.e. $\rho_0$ restricted to the torus with some rational points removed. Holes may exist, in general, as in the case of Katok-Lewis examples which are constructed by a blowing up process. But in the torus case, a homological consideration leads to the existence of (at least) a periodic point lying in a hole. Its Siegel neighborhood must cut that of our initial fixed point (because we are on a torus), contradicting the maximality of these neighborhoods. Therefore the Siegel neighborhood is the whole torus, that is $\rho$ is conjugate to $\rho_0$. This idea may be adapted to handle the following situations. Let $\Gamma$ be a subgroup of finite index of $SL(n, {\Bbb Z})$, and $\rho_0$ its standard action on $T^n$ ($n \geq 3$). i\) The product of $\rho_0$ by the trivial action of $\Gamma$ on any compact (real) analytic manifold $N$ is $C^1$ locally rigid among $C^\omega$-actions, i.e. it is $C^\omega$-conjugate to every analytic $\Gamma$-action on $T^n \times N$, which is $C^1$-close to it. ii) The orbit $Diff^\omega(T^n).\rho_0$, i.e. the space of conjugates of $\rho_0$ in the space $Rep(\Gamma \to Diff^\omega(T^n))$ of analytic actions of $\Gamma$ on $T^n$, is closed-open for the $C^1$ topology, and closed for the $C^0$ topology. iii\) Any faithful analytic action of $\Gamma$ on $T^n$ preserving a non-atomic measure, and having a fixed point in the support of this measure, is (up to an automorphism) $C^\omega$-conjugate to $\rho_0$. Let $\Gamma$ be a subgroup of finite index of $SL(n+1, {\Bbb Z})$, $n \geq 3$, and $\rho_0$ its standard projective action on $S^n$. The orbit $Diff^\omega(S^n).\rho_0$ in $Rep(\Gamma \to Diff^\omega(S^n))$ is closed-open for the $C^1$ topology and closed for the $C^0$ topology.\ ———————————————————————– Pour montrer que l’action usuelle d’un sous-groupe d’indice fini de $SL(n, {\Bbb Z})$ ($n \geq 3$) sur le tore $T^n$ est localement rigide, une idée naturelle consiste à montrer que toute action proche est “linéarisable”. Dans cette note, nous montrons comment faire marcher cette approche, et aussi la généraliser à d’autres situations. Les preuves elles mêmes, hormis un résultat de linéarisablité locale de [@C-G], sont linéaires, i.e. découlent de propriétés de l’action usuelle. Les détails ainsi que des résultats complémentaires paraîtront ultérieurement. Toutes les actions considérées ici sont [**analytiques**]{} (réelles). #### 1. Linéarisation: du local au global. Soit $M$ une variété analytique réelle et $\Gamma$ un groupe agissant analytiquement sur $M$, via un homomorphisme $\rho: \Gamma \to $ Diff$^\omega(M)$. Supposons que cette action admette un point fixe $x_0$.Nous avons alors une représentation infinitésimale: $r: \Gamma \to GL(T_{x_0}M)$. Supposons que $\rho$ soit analytiquement linéarisable au voisinage de $x_0$, c’est-à-dire qu’elle est conjuguée à $r$, dans un voisinage de $x_0$. Plus précisément, cela signifie qu’il existe $U$, voisinage de $0$ dans $T_{x_0}M$, et $V$ voisinage de $ x_0$ dans $M$, et un difféomorphisme analytique $\phi: U \to V$, tels que, pour tout $\gamma \in \Gamma$, nous avons l’égalité $\phi r(\gamma) = \rho(\gamma) \phi$, dans un certain voisinage de $x_0$ (dépendant de $\gamma$). Pour pouvoir en profiter dynamiquement, nous aurons besoin d’ouverts $U$ et $V$ invariants respectivement par $r$ et $\rho$. ${}$ 1\) Parmi les ouverts [*étoilés*]{} (en 0) de $T_{x_0}M$ dans lesquels $\phi$ se prolonge analytiquement en un difféomorphisme local sur son image, il existe un seul ouvert maximal, noté ${\cal E}$. Il est en particulier invariant par $r$, et le prolongement $\bar{\phi}$ de $\phi$ à ${\cal E}$ vérifie la semi-conjugaison globale: $\bar{\phi} r(\gamma) (u) = \rho(\gamma) \bar{\phi} (u)$, $ \forall \gamma \in \Gamma$ et $\forall u \in {\cal E}$. 2\) Il existe un ouvert maximal parmi les ouverts connexes, contenant 0, $r$-invariants, et sur lesquels $\phi$ se prolonge en un difféomorphisme analytique local semi-conjuguant $r$ et $\rho$. (Dans la suite, nous choisirons un tel ouvert qu’on notera ${\cal M}$, et noterons $\Phi: {\cal M} \to M$, le prolongement de $\phi$). En général, il n’ y a pas une notion consistante de domaine maximal d’extension d’une application analytique; sauf en dimension (réelle) 1, auquel cas, on peut parler d’intervalle maximal de prolongement analytique. Dans notre cas ici, nous considèrons le prolongement maximal de $\phi$ le long des demi-rayons issus de $0 \in T_{x_0}M$. Cela détermine un ensemble étoilé (en 0) sur lequel $\phi$ se prolonge radialement analytiquement. Nous obtenons un ouvert étoilé, en considérant l’intérieur de cet ensemble.Nous nous se restreignons ensuite aux points au voisinage desquels le prolongement de $\phi$ est un difféomorphisme analytique local. C’est notre domaine ${\cal E}$. Tout est naturel, donc, ${\cal E}$ est $r$-invariant, et la semi-conjugaison $\bar{\phi} r(\gamma) = \rho(\gamma) \bar{\phi}$ est satisfaite sur ${\cal E}$. (Plus analytiquement, soit $u \in T_{x_0}M$, si l’application $t \to \phi(tu)$ se prolonge à $[0, T[$, alors, l’application $t \to \phi(t (r(\gamma)))$ se prolonge également dans $[0, T[$ par la formule $\phi(t (r(\gamma))) = \rho (\gamma) (\phi(tu))$...). Enfin, comme il y a un ouvert connexe $r$-invariant satisfaisant la semi-conjugaison ci-dessus, il y en a un (ouvert connexe) maximal (parmi tous les ouverts connexes, non nécessairement étoilés). #### 2. Exemples. En dépit de son “évidence”, la proposition ci-dessus ne semble pas exister dans la littérature, où l’on s’intéresse particulièrement au cas où la représentation linéaire $r$ est constituée soit de dilatations (et contractions), soit de transformations orthogonales. et de plus, en général, $\Gamma = {\Bbb Z}$. (On parle alors de disque, ou domaine, de Siegel...). Il est vrai que c’est dans ces cas que $\Phi: {\cal M} \to M$ jouit d’intéressantes propriétés topologiques. Dans le cas général, $\Phi$ n’est pas nécessairement un revêtement sur son image. On peut par exemple s’amuser à considérer l’exemple de $\Gamma= {\Bbb Z}$ agissant sur le tore $ {T}^2 = S^1 \times S^1$, par un difféomorphisme $f = g \small{\circ} A$, où $A$ est un automorphisme hyperbolique de ${T}^2$ et $g = (g_1, g_2)$, avec $g_1$ et $g_2$ difféomorphismes analytiques de $S^1$ fixant $0$. Pour un choix générique des dérivées $g_1^\prime(0)$ et $g_2^\prime(0)$, la matrice dérivée $D_{(0,0)}f$ n’aura pas de résonances (ce qui entraîne en particulier que $\det(D_{(0,0)}f) \neq \pm 1$), et ainsi d’après le théorème de linéarisation de Sternberg, $f$ est linéarisable au voisinage de $(0,0)$. Il est exceptionnel que $\Phi: {\cal M} \to M$ soit un revêtement sur son image. On peut montrer qu’il est impossible que la connexion plate de ${\cal M}$ ($\subset {\Bbb R}^2$) descende par $\Phi$ (ou en d’autres termes que les identifications de points de ${\cal M}$ ayant une même image par $\Phi$, se fassent à l’aide d’applications affines), et ce à cause du fait que $\det(D_{(0,0)}f) \neq \pm 1$. D’autres exemples s’obtiennent en considérant l’action à gauche d’un réseau $\Gamma$ d’un groupe de Lie $G$ sur le quotient $G / \Gamma$. Cette action fixe le point correspondant à l’élément neutre de $G$, et y est (localement) linéarisable. Ici, $\Phi$ est la restriction de l’application exponentielle ${\cal G} \to G/ \Gamma$, à ${\cal M}$, qui est un ouvert connexe $Ad(\Gamma)$-invariant, et sur lequel l’application exponentielle est un difféomorphisme local. Même dans le cas de $G = SL(n, {\Bbb R})$, $\Phi$ présente assez de pathologies topologiques. #### 3. Un cas rigide. Les réseaux irréductibles des groupes de Lie semisimples, de centre fini, sans facteur compact et de rang réel $\geq 2$, e.g. $SL(n, {\Bbb R})$, $n\geq 3$, sont bien connus par leurs propriétés de rigidité (et super-rigidité). Nous n’en citerons qu’une que nous allons tout de suite exploiter: un résultat de C. Cairns et E. Ghys [@C-G], affirmant, avec les notations ci-dessus (i.e. que $\Gamma$ agit analytiquement en fixant $x_0$), que si $\Gamma$ est un tel réseau, alors son action est linéarisable au voisinage de $x_0$. Il est donc intéressant de comprendre $\Phi$ dans ce cas. L’exemple précèdent de $\Gamma$ agissant sur $G/ \Gamma$, montre les limites d’une rigidité espérée pour $\Phi$. La raison, dans ce cas, réside, probablement dans la “relative pauvreté dynamique” de la représentation linéaire $r$, qui n’est rien d’autre que la représentation $Ad$ de $\Gamma$; par exemple, $Ad(\Gamma)$ agit proprement sur un ouvert (non-vide) de ${\cal G}$. Dans la suite, nous traiterons le cas où $\Gamma$ est un sous-groupe de $SL(n, {\Bbb R})$ agissant sur une variété $M$ de dimension ($n \geq 3$). Nous nous ramènons au cas où $r$ est fidèle (nous utilisons pour cela un théorème de Margulis affirmant que le noyau de $r$ est soit fini, soit d’indice fini). D’après la super-rigidité de Margulis, $r$ est soit la représentation canonique soit sa duale; nous supposerons pour simplifier les notations que c’est la canonique. On peut commencer par essayer de comprendre ${\cal M}$. Son complémentaire ${\cal C}= {\Bbb R}^n - {\cal M}$ est un fermé invariant par l’action de $\Gamma$ sur ${\Bbb R}^n$. De tels ensembles ont été intensivement étudiés dans la littérature; mais tous les résultats les concernant peuvent se déduire, après manipulation algébrique, du Théorème de M. Ratner, résolvant la conjecture de Raghunathan [@Rat]. Pour voir que ce théorème s’applique bien, on identifie ${\Bbb R}^n-\{0\}$ à l’espace homogène $SL(n, {\Bbb R}) /H$, où $H$ est le stabilisateur d’un certain point. Une partie fermée ${\cal C}$ de ${\Bbb R}^n-\{0\}$, $\Gamma$-invariante, s’identifie aussi à une partie fermée ${\cal C}^\prime$ de $SL(n, {\Bbb R})$, $\Gamma$-invariante à gauche et $H$-invariante à droite, i.e. ${\cal C}^\prime= \Gamma. {\cal C}^\prime. H$; et par conséquent, elle s’identifie à une partie fermée de $\Gamma \setminus SL(n, {\Bbb R})$, invariante par $H$ (agissant à droite). Le théorème s’applique car $\Gamma$ est un réseau, et $H$ est engendré par ses éléments unipotents. Le théorème de Ratner affirme que pour $u \in {\Bbb R}^n -\{0\}$, la composante connexe de $u$ dans l’adhérence $\overline{\Gamma. u}$ de son orbite, est de la forme $G.u$, où $G$ est un sous-groupe de Lie connexe, contenant le stabilisateur de $u$ (qui est un conjugué de $H$) et tel que $\Gamma \cap G$ soit un réseau de $G$. On peut en déduire (mais c’est aussi faisable à l’aide d’outils plus élémentaires, dans ce cas précis) que si $\Gamma$ n’est pas, à automorphisme près, un sous-groupe d’indice fini de $SL(n, {\Bbb Z})$, alors, $\Gamma$ agit minimalement sur ${\Bbb R}^n-\{0\}$, et en particulier ${\cal C}= \emptyset$. Si $\Gamma$ est à automorphisme près un sous-groupe d’indice fini de $SL(n, {\Bbb Z})$, alors l’orbite d’un point est, soit dense, soit discrète, auquel cas, ce point est de la forme $\lambda u$, où $u$ est rationnel (et $\lambda \in {\Bbb R}$). En général, nous avons des parties fermées invariantes propres de la forme $\Gamma. (F.u)$ ($= F.(\Gamma.u)$), où $u$ est rationnel et $F$ est un fermé de $[0, \infty [$ ne contenant pas 0 (la notation $F.u$ désigne l’ensemble $\{ tu, t \in F \}$). Tout fermé invariant propre est réunion finie de tels fermés élémentaires. Soit $\Gamma$ un réseau de $SL(n, {\Bbb R}), n>2$, agissant analytiquement en fixant un point $x_0$ sur une variété $M$ de dimension $n$. Alors, ${\cal M} $ est unique, on appellera $\Phi({\cal M})$ l’[**ouvert de Siegel**]{} en $x_0$. Le prolongement maximal $\Phi: {\cal M} \to M$ vérifie: i\) $\Phi$ est un revêtement sur son image. ii\) La connexion affine plate sur ${\cal M}$ descend par $\Phi$ en une connexion plate $\nabla$ sur $\Phi({\cal M})$. iii\) $\nabla$ ne peut pas se prolonger (même localement) en dehors de l’ouvert de Siegel. iv\) $\nabla$ est (localement) unique, au sens que si $\Gamma$ préserve une connexion définie sur un ouvert invariant contenu dans l’ouvert de Siegel, alors, c’est la restriction de $\nabla$.\ De plus, il y a deux situations possibles:\ On y distingue deux cas. 1\) $\Gamma$ n’est pas à automorphisme près un sous-groupe d’indice fini de $SL(n, {\Bbb Z})$, alors ${\cal M} = {\Bbb R}^n$, $\Phi$ est un difféomorphisme (global) sur son image, et $\nabla$ est complète. 2\) $\Gamma$ est à automorphisme près un sous-groupe d’indice fini de $SL(n, {\Bbb Z})$, ${\cal M}\subset {\Bbb R}^n$ est un ouvert $\Gamma$-invariant comme décrit ci-dessus, et $\Phi$ est un difféomorphisme sur son image $\Phi({\cal M})$.\ $\Gamma$ est à automorphisme près un sous-groupe d’indice fini de $SL(n, {\Bbb Z})$, et $\Phi$ n’est pas injective. Dans ce cas, $\Phi$ transite à travers un difféomorphisme $ \Phi^*: (T^n)^* \to \Phi({\cal M})$, où $(T^n)^*$ est le tore $T^n$ privé d’un nombre fini de points rationnels, et $\Phi^*$ respecte les actions et les connexions. La possibilité conservative se produit exactement lorsque l’une ou l’autre des deux conditions suivantes est satisfaite: 1\) $\Gamma$ préserve une mesure finie sans atomes dont le support contient $x_0$. Cette mesure est alors la mesure standard sur le tore troué. 2\) Il existe un élément $\gamma \in \Gamma$ d’ordre infini tel que l’ensemble des points non-errants de $\rho(\gamma)$ contient un voisinage de $x_0$. D’après la description ci-dessus des ensembles fermés invariants par $\Gamma$ agissant sur ${\Bbb R}^n$, nous tirons en particulier que tout ouvert invariant est connexe. En particulier, s’il y a deux ouverts maximaux sur lesquels $\phi$ se prolonge analytiquement, alors leur intersection est connexe, et par suite, $\phi$ se prolonge à leur réunion, donc ${\cal M}$ est unique. Les propriétés de $\Phi$ découlent de la rigidité de l’ensemble: $${\cal R} = \{ (x, y) \in {\cal M} \times {\cal M} \subset {\Bbb R}^n \times {\Bbb R}^n / \Phi(x) = \Phi(y) \}$$ C’est une sous-variété analytique de ${\cal M} \times {\cal M}$, localement fermée dans ${\Bbb R}^n \times {\Bbb R}^n$, invariante par l’action diagonale de $\Gamma$ sur ${\Bbb R}^n \times {\Bbb R}^n$. Pour la décrire, nous considèrons l’action diagonale de $SL(n, {\Bbb R})$ sur ${\Bbb R}^n \times {\Bbb R}^n-\{(0,0)\}$. Cette action admet une orbite ouverte ${\cal O} = \{ (u, v) / {\Bbb R}u \neq {\Bbb R}v \}$, et des orbites dégénérées de la forme $O_\alpha = \{(u, \alpha u), u \in {\Bbb R}^n \}$, $\alpha$ étant un réel. (On suppose $n >2$). Évidemment, ${\cal R}$ qui est une relation d’équivalence, contient la diagonale ${\cal O}_1$. L’égalité ${\cal R} = {\cal O}_1$ signifie que $\Phi$ est injective. On se convainc facilement que si ${\cal R} \neq {\cal O}_1$, alors ${\cal R}$ contient des points de l’orbite ouverte ${\cal O}$. Choisissons un point $(e_1, e_2)$ de ${\cal O}$. Son stabilisateur $H$ (dans $SL(n, {\Bbb R})$) est engendré par ses éléments unipotents; pour $n=3$, $H$ est unipotent. Pour étudier l’adhérence de $\Gamma. (e_1,e_2)$, nous aurons besoin de comprendre les sous-groupes connexes contenant $H$. Notons $P$ le plan ${\Bbb R}e_1 \bigoplus {\Bbb R}e_2$, et $S_P$ le sous-groupe de $SL(n, {\Bbb R})$ préservant $P$; nous avons une projection $\pi: S_P \to GL(P) \simeq GL(2, {\Bbb R})$ ($H$ s’identifie à $\pi^{-1}(1)$). Soit $G$ un groupe de Lie connexe contenant $H$ et différent de $SL(2, {\Bbb R})$, alors l’une des deux possibilités suivantes se présente: i\) $G \subset S_P$; plus exactement $G = \pi^{-1}(L)$, où $L$ est un sous-groupe de $GL(2, {\Bbb R})$. ii\) Il existe $e \in P$, et $G = S_e$, ou $G= S_{ {\Bbb R}e}$, où $S_{\{ \}}$ désigne le stabilisateur. Dans le cas de $\Gamma = SL(n, {\Bbb Z})$, il existe $G$ avec $H \subset G \subset S_P$, et $G \cap \Gamma$, un réseau de $G$, si et seulement si $P$ est un 2-plan rationnel (i.e. engendré par deux vecteurs rationnels). Pour $\Gamma$ quelconque, l’ensemble des 2-plans $P$ avec un groupe $G$ comme précédemment, vérifiant que $G \cap \Gamma$ est un réseau de $G$, est dénombrable. On peut donc se restreindre à l’étude du cas où l’adhérence de l’orbite $\Gamma.(e_1, e_2)$ est déterminée par un groupe $G$ égal à $S_e$ ou $S_{{\Bbb R}e }$. En fait $G= S_e$ car $S_{{\Bbb R}e}$ n’est pas unimodulaire. Donc pour l’action de $\Gamma$ sur ${\Bbb R}^n$, l’orbite de $e$ est discrète; il en découle que $\Gamma$ est à automorphisme près un sous-groupe d’indice fini de $SL(n, {\Bbb Z})$. L’événement, $e$ est colinéaire à $e_1$ ou à $e_2$, correspond à un nombre dénombrable de cas; nous pouvons donc supposer qu’il n’a pas lieu. Maintenant $S_e$ agit transitivement sur ${\Bbb R}^n -{\Bbb R}e$; en particulier tous les points $(x, y)$ d’un voisinage de $(e_1,e_2)$ dans ${\cal R}$, s’écrivent: $(x, y)= (A(e_1), A(e_2))$, avec $A \in S_e$. Comme $e$ est défini à facteur près, et appartient à $P$ ($= {\Bbb R}e_1 \bigoplus {\Bbb R}e_2$), nous pouvons écrire $e_2= \alpha e_1 +e$, pour un certain $\alpha$. Donc $y = A(e_2) = \alpha A(e_1) + A(e) = \alpha x + e$ (car $A \in S_e$). En d’autres termes, un voisinage de $(e_1, e_2)$ dans ${\cal R}$ coïncide avec le graphe de l’homothétie-translation $x \to \alpha x +e$. Notons $f$ cette transformation. Alors, dans un voisinage de $e_1$, nous avons $\Phi \small{\circ} f = \Phi$. Mais comme $f$ est définie partout, l’égalité se prolonge à tout ${\cal M}$, qui doit être invariant par $f$. Il en va de même pour les transformations $h= AfA^{-1}$, qui sont de la forme $x \to \alpha x + A(e)$, où $A$ parcourt $\Gamma$. Sachant que $\Gamma$ est à automorphisme près un sous-groupe d’indice fini de $SL(n, {\Bbb Z})$, nous en déduisons que les parties translationnelles $A(e)$ de ces transformations forment un réseau de ${\Bbb R}^n$, et en particulier, si $\alpha \neq 1$, alors il existe un commutateur de ces transformations qui est une translation non triviale; et par suite, il existe un réseau de translations $g$ vérifiant $\Phi \small{\circ} g = \Phi$. Si $\alpha \neq \pm 1$, il existerait trop de transformations $h$ satisfaisant $\Phi \small{\circ} h= \Phi$, contredisant le fait que $\Phi$ est un difféomorphisme local. Ceci montre que $\Phi$ transite à travers un difféomorphisme $\Phi^*$ comme énoncé. La connexion plate descend à l’ouvert de Siegel $\Phi{\cal M})$ car le groupe de Galois du revêtement $\Phi$ agit affinement sur ${\cal M}$. Le fait qu’elle ne peut pas se prolonger à un ouvert strictement plus grand que l’ouvert de Siegel, provient du fait que cette connexion est déjà “pratiquement complète”. Si les ouverts de Siegel de deux points fixes de $\Gamma$ s’intersectent, alors ils sont identiques. #### 4. Dimension $>n$. La discussion précédente se généralise partiellement lorsque $\Gamma$ agit sur une variété $M$ de dimension $n +p$, telle que la représentation $r$ de $\Gamma$ dans ${\Bbb R}^n \times {\Bbb R}^p$ soit le produit de la représentation canonique dans ${\Bbb R}^n$ par la représentation triviale dans ${\Bbb R}^p$. Par exemple, on peut se ramener essentiellement à cette situation, lorsque $p <n$. Les transformations (partielles) $f$ de ${\Bbb R}^{n+p}$ vérifiant $\Phi \small{\circ} f = \Phi$ sont de la forme: $$f: (u,v) \in {\Bbb R}^n \times U \subset {\Bbb R}^n \times {\Bbb R}^p \to (\alpha(v)u + e(v), g(v))$$ où $\alpha: U \to {\Bbb R}$, $e: U\to {\Bbb R}^n$ et $g: U \to {\Bbb R}^p$, sont des applications analytiques définies sur l’ouvert (de ${\Bbb R}^p$) $U$ et $g$ est un difféomorphisme sur son image. (Évidemment, il peut se passer que $\alpha(v) \equiv 1$, et $e(v) \equiv 0$). Notons que vu la forme de $f$, la composition, à gauche ou à droite, de $f$ avec des éléments de la forme $r(\gamma)$, est définie dans tout le domaine de définition de $f$. Le produit $g = r(\gamma) f^{-1} r(\gamma)^{-1} f$, est un difféomorphisme de ${\Bbb R}^n \times U$ de la forme $g: (u, v) \to (u+ e(v)- r(\gamma) (e(v)), v)$. En particulier, dès qu’il existe un $e(v) \neq 0$, alors, il y ’en aura un réseau, $\Gamma$ est à automorphisme près un sous-groupe d’indice fini de $SL(n, {\Bbb Z})$, et le niveau ${\Bbb R}^n \times \{v\}$ s’envoie par $\Phi$ sur un tore troué. Cette discussion nous fournit en particulier le fait suivant. Soit $\Gamma$ un réseau de $SL(n, {\Bbb R})$ agissant analytiquement sur une variété $M$, en fixant un point $x_0$, avec une représentation infinitésimale, produit de la représentation canonique dans ${\Bbb R}^n$ par une représentation triviale. Supposons que l’action préserve une mesure finie dont le support contient un voisinage de $x_0$. Alors, $\Gamma$ est à automorphisme près un sous-groupe d’indice fini de $SL(n, {\Bbb Z})$, et sur un ouvert invariant contenant $x_0$, l’action est conjuguée à l’action sur $(T^n)^* \times N$, où $(T^n)^*$ est un tore troué sur lequel $\Gamma$ agit de la façon usuelle, et $N$ est une variété de dimension $p= $ dim$M-n$, sur lequel $\Gamma$ agit trivialement. On en déduit en particulier: Soit $\Gamma$ un réseau de $SL(n, {\Bbb R})$, $n>2 $, agissant fidèlement analytiquement sur une variété $M$ de dimension $<2n$, en préservant une mesure finie pleine (i.e.son support est égal à $M$). Si $\Gamma$ n’est pas à automorphisme près un sous-groupe d’indice fini de $SL(n, {\Bbb Z})$, alors, l’action n’a aucun point périodique. [*Dans tous les énoncés précédents, $M$ n’était pas supposée compacte!* ]{} #### 5. Première application: actions sur le tore. Nous esquissons dans ce qui suit une preuve de la rigidité locale de l’action usuelle d’un sous-groupe $\Gamma$ d’indice fini dans $SL(n, {\Bbb Z})$ sur le tore (parmi les actions $C^\omega$). En fait, pour simplifier les notations, nous supposerons que $\Gamma = SL(n, {\Bbb Z})$. Nous supposerons aussi que $n \geq 6$, et ce pour pouvoir “scinder” $\Gamma$ en réseaux de rang supérieur. Le cas général, c’est-à-dire $n\geq 3$, demande un peu plus d’analyse. Soit $\rho$ une action $C^1$ proche de l’action standard $\rho_0$. D’après un théorème de Stowe [@sto], $\rho$ admet un point fixe $x_0$ (proche de 0). On peut se convaincre facilement que l’ouvert de Siegel en $x_0$ est un tore troué, c’est-à-dire qu’on ne peut pas être dans le cas dissipatif du Théorème 1. L’action $\rho$ sera conjuguée à $\rho_0$ si l’on démontre que l’ouvert de Siegel est un vrai tore, i.e. qu’il est non-troué. Pour montrer cela, il suffit de montrer que $\Gamma$, ou même un sous-groupe d’indice fini, possède un point fixe, disons $x_1$, résidant dans un trou (i.e. le complémentaire de l’ouvert de Siegel de $x_0$). En effet, alors, puisqu’on est sur un tore, les deux domaines de Siegel doivent s’intersecter, contredisant le Corollaire 2. On cherchera un point fixe dans un trou, comme intersection de deux sous-variétés, lieux de points fixes de deux sous-groupes engendrant $\Gamma$, dont on sait que le nombre d’intersection total est supérieur à leur nombre d’intersection dans l’ouvert de Siegel en $x_0$. Plus précisément, soit $e_1, \ldots, e_n$ la base canonique de ${\Bbb R}^n$, et notons $E_1$ (resp. $E_2$) les plans engendrés par $e_1, e_2, e_3$ (resp. $e_4, \ldots, e_n$). Ils sont (ponctuellement) fixés par $ \Gamma_1= SL(n-3, {\Bbb Z})$ et $\Gamma_2= SL(3, {\Bbb Z})$ (plongés naturellement dans $SL(n, {\Bbb Z})$). Notons $F_1$ (resp. $F_2$) l’ensemble (analytique) des points fixes de $\rho(\Gamma_1)$ (resp. $\rho(\Gamma_2)$). D’après un théorème de Stowe, lorsque $\rho$ est $C^1$ proche de $\rho_0$, ces ensembles sont des sous variétés (proches des celles qui correspondent à $\rho_0$). (On utilise pour appliquer le théorème de Stowe, le théorème d’annulation cohomologique de Margulis). Évidemment $F_1$ et $F_2$ sont des prolongements dans les trous, de $\Phi(E_1)$ et $\Phi(E_2)$ respectivement. Considérons un élément $\gamma \in SL(n, {\Bbb Z})$ tel que $E_2^\prime = r(\gamma)(E_2)$ intersecte transversalement $E_1 + {\Bbb Z}^n$ en un point de ${\Bbb R}^n-{\cal M}$ (cela existe car ${\Bbb R}^n-{\cal M}$ est constitué de points rationnels). Le nombre d’intersection de $F_2^\prime = \rho(\gamma) (F_2) $ avec $F_1$ est le même que dans le cas de l’action standard. Il est égal au cardinal de $E_2^\prime \cap (E_1 + {\Bbb Z}^n)$ mod. ${\Bbb Z}^n$. Mais le nombre d’intersection de $F_2^\prime$ avec $F_1$ à l’intérieur de l’ouvert de Siegel vaut le nombre d’intersection de $\Phi(E_2^\prime)$ avec $\Phi(E_1)$, qui est égal au cardinal de $(E_2^\prime \cap (E_1 + {\Bbb Z}^n) -{\cal M})$ mod. ${\Bbb Z}^n$. Le choix de $\gamma$ assure que ces deux nombres d’intersection sont différents, et par suite $F_2^\prime$ et $F_1$ s’intersectent en dehors de l’ouvert de Siegel. On remarque maintenant que $E_1$ et $E_2$ sont en fait fixés (ponctuellement) par des groupes plus grands que $\Gamma_2$ et $\Gamma_1$ respectivement; et qui sont des produits semi-directs évidents $\Gamma_2 \ltimes N_2$ et $\Gamma_1 \ltimes N_1$ , où $N_1$ et $N_2$ sont unipotents. Il en résulte que le groupe engendré par $r(\gamma) (\Gamma_2 \ltimes N_2) r(\gamma)^{-1}$ et $\Gamma_1 \ltimes N_1$ fixe un point dans un trou. Mais, à cause de la transversalité entre $E_2^\prime$ et $E_1$, ce groupe est d’indice fini dans $SL(n, {\Bbb Z})$. Les premiers résultats de rigidité locale sont dus à Hurder [@Hur] et Katok-Lewis [@K-L] (dans le cas lisse). La rigidité locale de l’action des sous-groupes d’indice fini de $SL(n, {\Bbb Z})$ sur $T^n$ est parue dans [@KLZ]. L’un des travaux les plus récents sur la question est [@M-Q], où l’on démontre la rigidité locale des “actions algébriques” faiblement hyperboliques. Ici, à l’aide des développements du §4, on peut adapter l’approche ci-dessus pour traiter la rigidité locale dans une situation qui n’est pas faiblement hyperbolique, où $\Gamma$ agit diagonalement sur un produit $T^n \times N$, où $N$ est une variété compacte quelconque sur laquelle $\Gamma$ agit trivialement (Voir [@N-T] pour des résultats proches). Soit $\Gamma$ un sous-groupe d’indice fini de $SL(n, {\Bbb Z})$, $n >2$. Considérons $\rho_0$ l’action produit de l’action usuelle de $\Gamma$ sur $T^n$, par l’action triviale sur une variété compacte $N$. Alors $\rho_0$ est localement rigide sous perturbation $C^1$, parmi les actions analytiques; plus précisément, toute action analytique $C^1$ proche de $\rho_0$, est analytiquement conjuguée à $\rho_0$. Pour les actions sur $T^n$, nous avons le résultat de non-dégénérescense suivant. Soit $\Gamma$ un sous-groupes d’indice fini de $SL(n, {\Bbb Z})$, et $(\rho_i)$ une suite d’actions analytiquesde $\Gamma$ conjuguées à son action usuelle sur $T^n$. Supposons que cette suite converge au sens de la topologie $C^0$ vers une action analytique $\rho$. Alors $\rho$ est analytiquement conjuguée à l’action usuelle. Ce type de résultat est étranger à la théorie hyperbolique, notamment dans le cas classique des difféomorphismes d’Anosov, puisqu’il existe des applications dites DA, dérivées d’Anosov. Le théorème suivant unifie les deux résultats précédents dans le cas de $T^n$, et rend vraisemblable une rigidité globale des actions analytiques dans ce cas (sans hypothèse d’hyperbolicité ou de préservation de volume...). Soit $\Gamma$ un sous-groupes d’indice fini de $SL(n, {\Bbb Z})$, et $Rep(\Gamma, Diff^\omega(T^n))$ l’espace de ses actions analytiques sur $T^n$. L’orbite $Diff^\omega(T^n).\rho_0$, i.e. l’espace des actions conjuguées à l’action standard $\rho_0$, est ouvert dans $Rep(\Gamma, Diff^\omega(T^n)$ au sens de la topologie $C^1$, et fermée au sens de la topologie $C^0$. En particulier, toute action homotope à l’action usuelle, au sens de la topologie $C^1$, à travers des actions $C^\omega$, lui est $C^\omega$ conjuguée. Nous ne savons pas démontrer que l’orbite $Diff^\omega(T^n).\rho_0$ est ouverte dans $Rep(\Gamma, Diff^\omega(T^n))$, au sens de la topologie $C^0$, pour la simple raison que nous ne disposons pas d’un résultat de persistance de point fixe de $\Gamma$ sous perturbation $C^0$. (Une telle persistance paraît vraisemblable dans ce contexte précis). Enfin, nous avons ce résultat global, qui réduit (essentiellement) la rigidité globale à l’existence de points périodiques. Toute action fidèle analytique d’un sous-groupe d’indice fini de $SL(n, {\Bbb Z})$ sur $T^n$ ayant un point fixe et préservant une mesure finie non-atomique dont le support contient le point fixe est analytiquement conjuguée à l’action usuelle. #### 6. Deuxième application: actions sur la sphère. Le groupe de Lie $SL(n+1, {\Bbb R})$ agit projectivement sur la sphère $S^n$, ce qui donne par restriction une action de ses réseaux. Une super-rigidité dans ce contexte consiste à se demander si réciproquement, une action d’un réseau $\Gamma$ de $SL(n+1, {\Bbb R})$ se prolonge en une action de $SL(n+1, {\Bbb R})$ (qui sera par suite nécessairement l’action projective usuelle, à automorphisme près). La rigidité locale est une version locale de cette question. Elle a été démontrée, mais seulement pour les réseaux co-compacts, d’abord par Kanai [@Kan] avec une restriction (technique) sur la dimension, et ensuite dans le cas général par Katok-Spatzier [@K-S] (qui démontrent en fait la rigidité locale des réseaux co-compacts de rang $\geq 2$ agissant sur des bords). Ici, nous considérons des réseaux (non co-compacts) de $SL(n+1, {\Bbb R})$ qui sont des sous-groupes d’indice fini dans $SL(n+1, {\Bbb Z})$ (à automorphisme près). Leur avantage ici, est qu’il contiennent des réseaux de $SL(n, {\Bbb R})$, qui admettent donc des points fixes, auxquels nous pouvons ainsi appliquer Théorème 1. Nous avons le résultat (global) suivant. \[projective.rigid\] Soit $\Gamma$ un sous-groupe d’indice fini de $SL(n+1, {\Bbb Z})$, agissant fidèlement analytiquement sur une variété compacte $M$ de dimension $n \geq 3$, tel que $\Gamma \cap SL(n, {\Bbb Z})$ admet un point fixe. Alors, à automorphisme près, c’est l’action usuelle de $\Gamma$ sur la sphère $S^n$ ou sur l’espace projectif ${\Bbb R}P^n$. Nous supposons pour simplifier les notations que $\Gamma = SL(n+1, {\Bbb Z})$. Ses éléments de la forme $\pmatrix {1 &0 \cr 0& A}, A \in SL(n, {\Bbb R})$ forment un sous-groupe isomorphe à $SL(n, {\Bbb Z})$ que nous notons $\Gamma_0$. Les sous-groupes abéliens unipotents constitués des éléments de la forme $\pmatrix {1 & * \cr 0& 1}$, et $\pmatrix {1 & 0 \cr * & 1}$ sont notés $N^+$ et $N^-$ respectivement. Ils sont isomorphes à ${\Bbb Z}^n$, et donnent lieu à deux produits semi-directs $\Gamma_0 \ltimes N^+$ et $\Gamma_0 \ltimes N^-$ isomorphes au produit semi-direct $SL(n, {\Bbb Z}) \ltimes {\Bbb Z}^n$ (où $SL(n, {\Bbb Z})$ agit sur ${\Bbb Z}^n$ vu comme le réseau canonique de ${\Bbb R}^n$, dans un cas via la représentation canonique de $SL(n, {\Bbb R})$, et dans l’autre cas, via sa représentation duale). Par hypothèse, $\Gamma_0$ fixe un point, disons $x_0$, nous supposons que sa représentation infinitésimale est la canonique. Maintenant, l’idée est de revêtir $M$ d’une structure projective $\Gamma$-invariante, en suivant la recette déterminée par la structure projective usuelle (sur $S^n$ ou ${\Bbb R}P^n$). Nous montrons d’abord, que tout comme dans le cas standard, au moins à indice fini près, $N^+$ fixe $x_0$. L’idée est que le centralisateur dans $\Gamma_0$, d’un élément $\gamma$ de $N^+$, est suffisamment grand, et ne peut avoir que des points fixes isolés, qui sont, dans leur ensemble, invariants par $\gamma$. Ainsi, $\Gamma_0 \ltimes N^+$ admet $x_0$ comme point fixe , et l’action de $\Gamma_0$ y est (localement) linéarisable. Nous montrons alors que $\Gamma_0 \ltimes N^+$ préserve (localement) une structure projective plate au voisinage de $x_0$ (celle déterminée par la connexion affine locale préservée par $\Gamma_0$). Soit ${\cal S}(x_0, \Gamma_0)$ l’ouvert de Siegel de $x_0$ relatif à l’action de $\Gamma_0$. Il correspond nécessairement au cas dissipatif du Théorème 1 (le cas conservatif est exclu, car il y a un groupe plus grand, $\Gamma$, qui agit). Il s’identifie donc à un ouvert de ${\Bbb R}^n$ invariant par l’action de $SL(n, {\Bbb Z})$ ($= \Gamma_0$). Le bord de ${\cal S}(x_0, \Gamma_0)$ dans $M$ peut être très compliqué. La connexion projective s’étend, en particulier, à ${\cal S}(x_0, \Gamma_0)$, mais aussi individuellement aux images $\rho(\gamma) ({\cal S}(x_0, \Gamma_0))$, $\gamma \in N^+$ (car $N^+$ préserve la connexion projective). Nous montrons que la connexion projective plate s’étend au voisinage de la frontière de ${\cal S}(x_0, \Gamma_0)$. En développant la situation (équivariante) dans le substratum de la géométrie projective plate (i.e. $S^n$ ou ${\Bbb R}P^n$), nous identifierons le bord, et par suite l’adhérence de ${\cal S}(x_0, \Gamma_0)$. Il s’agit d’une hémisphère (projective) dont le bord est soit $S^{n-1}$ soit ${\Bbb R}P^{n-1}$. Dans ce dernier cas, l’adhérence de ${\cal S}(x_0, \Gamma_0)$ couvre toute la variété $M$, qui sera donc ${\Bbb R}P^n$. Dans le cas où le bord est $S^{n-1}$, nous nous aidons de $N^-$ pour trouver une deuxième hémisphère complémentaire à ${\cal S}(x_0, \Gamma_0)$, ce qui permet d’identifier $M$ à $S^n$. Il est probable que ce théorème soit vrai sans l’hypothèse de compacité de $M$ (l’énoncé serait alors qu’une telle action n’existe pas dans ce cas). On en déduit en particulier la rigidité locale (parmi les actions analytiques), mais aussi une propriété de non-dégénérescence, comme dans le cas du tore ci-dessus. Soit $\Gamma$ un sous-groupe d’indice fini de $SL(n+1, {\Bbb Z})$, $n\geq 3$. Alors l’orbite par $Diff^\omega(S^n)$ de son action standard sur $S^n$ est ouverte-fermée dans $Rep(\Gamma, Diff^\omega(S^n))$ muni de la topologie $C^1$ et elle est fermée au sens de la topologie $C^0$. [00]{} C. Cairns, É. Ghys, The local linearization problem for smooth ${\rm SL}(n)$-actions. Enseign. Math. (2) 43 (1997) 133–171. S. Hurder, Rigidity for Anosov actions of higher rank lattices. Ann. of Math. (2) 135 (1992) 361–410. M. Kanai, A new approach to the rigidity of discrete group actions. Geom. Funct. Anal. 6 (1996) 943–1056. A. Katok, J. Lewis, Local rigidity for certain groups of toral automorphisms. Israel J. Math. 75 (1991) 203–241. A. Katok, J. Lewis, R. Zimmer, Cocycle superrigidity and rigidity for lattice actions on tori. Topology 35 (1996) 27–38. A. Katok, R. Spatzier, Nonstationary normal forms and rigidity of group actions. Electron. Res. Announc. Amer. Math. Soc. 2 (1996) 124–133 V. Nitica, A. Török, Cohomology of dynamical systems and rigidity of partially hyperbolic actions of higher-rank lattices. Duke Math. J. 79 (1995) 751–810. G. Margulis, N. Qian, Rigidity of weakly hyperbolic actions of higher real rank semisimple Lie groups and thier lattices. Preprint. M. Ratner, Raghunathan’s topological conjecture and distributions of unipotent flows. Duke Math. J. 63 (1991) 235–280. D. Stowe, Stable orbits of differentiable group actions. Trans. Amer. Math. Soc. 277 (1983) 665–684. CNRS, UMPA, École Normale Supérieure de Lyon\ 46, allée d’Italie, 69364 Lyon cedex 07, FRANCE\ Zeghib@umpa.ens-lyon.fr\ http://umpa.ens-lyon.fr/\~zeghib/ [^1]: Version légèrement détaillée d’une note soumise aux C.R.A.S.

ArXiv

--- abstract: | Propagation of gravitational disturbances at the speed of light is one of the key predictions of the General Theory of Relativity. This result is now backed indirectly by the observations of the behaviour of the ephemeris of binary pulsar systems. These new results have increased the interest in the mathematical theory of gravitational waves in the last decades and several mathematical approaches have been developed for a better understanding of the solutions. In this paper we develop a modal series expansion technique in which solutions can be built for plane waves from a seed integrable function. The convergence of these series is proven by the Raabe-Duhamel criteria and we show that these solutions are characterized by a well-defined and finite curvature tensor and also a finite energy content. author: - | L. Acedo[^1]\ Instituto Universitario de Matemática Multidisciplinar,\ Building 8G, $2^{\mathrm{o}}$ Floor, Camino de Vera,\ Universitat Polit$\grave{\mbox{e}}$cnica de Val$\grave{\mbox{e}}$ncia,\ Valencia, Spain\ title: Modal series expansions for plane gravitational waves --- [**Keywords:**]{} Gravitational waves, exact solutions in General Relativity, modal series Introduction {#sec_1} ============ The mathematical theory of gravitational waves has a troubled history. Confusion among curvature and coordinate singularities lead Einstein and Rosen to publish a paper in which they claimed that gravitational radiation does not exist [@EinsteinRosen]. This question was still pondered by Rosen as late as 1979 [@Rosen1979]. Nowadays, the indirect evidence obtained by the meticulous tracking of the stars in the PSR B1913$+$16 system has provided bedrock evidence of the exactitude of the quadrupole formula for the gravitational energy losed by a binary system in the form of gravitational waves [@Sivaram; @Weisberg2010]. The confidence contributed by this experimental evidence has encouraged a lot of research into the mathematical properties of isolated and colliding gravitational waves in the last decades [@Griffiths]. In the case of plane gravitational waves the differential equation corresponds to the solution of the Sturm-Liouville problem: $$\label{SL} p''(u)+V(u)\, p(u)=0\; ,$$ where the apostrophes denote the second-order derivative, $p(u)$ is a metric function and $V(u)$ is the fourth Weyl scalar, $\Psi_4$ [@Chandrasekhar]. Exact solutions in terms of special functions can be obtained for special selections of the fuction $V(u)$. However, this is not possible in general and one must resort to numerical integration or Taylor series in powers of $u$. Although approximate solutions with a certain domain of validitiy can be obtained with these traditional techniques, the main hurdle is the difficulty to prove convergence and finite energy content with numerical calculations. In this context, Bini et al. [@Bini] proposed some constraints that a realistic gravitational wave should verify: (i) The Riemann tensor exists everywhere and it is continuous (ii) Space-time is asymptotically flat for $u=x-ct \rightarrow \pm \infty$ (iii) The metric function is regular (iv) The gravitational wave carries a finite energy content. We will show that the method described in this paper is adequate to obtain a family of solutions verifying these properties. Recently, it has been found that certain classes of non-linear differential equations with quadratic terms can be solved analytically by means of a modal series expansion. This method has been successfully applied to a version of the Kermack-McKendrick model with mass action terms [@AcedoNARWA; @AcedoPhysA] and also to the Lorenz system in the laminar flow regime [@Acedo2013]. This method can be described in general as follows: for any non-linear operator $\mathcal{R}$ acting upon a function $g$ we may define a nonlinear differential equation: $$\label{difeg} \mathcal{R}(g)=0\; .$$ A tentative series solution for this equation is proposed in terms of the functional series $\mathcal{F}=\left\{ f_n \right\}$, $n=0,1,2,\ldots$ in the form: $$\label{genser} g = \sum_{n=0}^\infty \, a_n f_n \; ,$$ where $a_n$ are real or complex numbers. The series in Eq.  (\[genser\]) is called modal under the nonlinear operator $\mathcal{R}$ if $\mathcal{R}(f_m)$ can be written as an algebraic combination of a finite subset of functions in $\mathcal{F}$ for any integer $m$. This allows us to find a recurrence relation for the coefficients $a_n$, $n=0,1,2,\ldots$. The number of free coefficients in Eq. (\[genser\]) will correspond to the highest-order derivative in the operator $\mathcal{R}(f_m)$. In the case of General Relativity the equations are order two in the metric. Sometimes the recurrence relation involve terms with indexes larger than the one we are computing and the calculation cannot be closed. Even if an adequate recurrence is found, we must test the final series solution for convergence because this is not assured from the outline of the heuristic modal method. We will develop the modal series technique for plane gravitational waves to show that exact solutions with finite values of the metric and the curvature exist. These waves form an infinite plane front perpendicular to the the propagation axis, $x$, and the Minkowski’s metric is recovered in the asymptotic limits $x\rightarrow \pm \infty$. The structure of the paper is as follows: in Section \[sect\_2\] we develop the modal formalism for gravitational plane waves. In Section \[sect\_3\] we prove the convergence of these solutions for any coordinates and discuss the definition of the curvature. The stress-energy pseudotensor is explicitly calculated in Section \[sect\_4\]. We also prove here that the energy content is finite. Conclusions, remarks and prospective future work are discussed in Section \[sect\_5\]. Modal transseries for gravitational waves {#sect_2} ========================================= We consider the metric for a plane-wave in the Rosen’s form: $$\label{metric} d s^2=-c^2 dt^2+d x^2+p^2(u)\, d y^2+q^2(u)\, d z^2= 2\, d u\, d v + p^2(u) \, d y^2 + q^2(u) \, d z^2 \; ,$$ where $u=x-c t$ and $v=x+c t$. Einstein’s field equations in vacuum for this metric [@Rindler] are deceptively simple: $$\label{FieldEqs} \displaystyle\frac{\ddot{p}(u)}{p(u)}+\displaystyle\frac{\ddot{q}(u)}{q(u)}=0\; ,$$ where $\ddot{p}(u)$, $\ddot{q}(u)$ denote the second derivative of the metric functions with respect to the wave phase parameter $u$. This is a Petrov type-N metric with a Weyl scalar $\Psi_4=-\ddot{p}(u)/p(u)$ [@Chandrasekhar]. Alternatively, we can use the so-called Brinkmann’s coordinates: $$\label{Brinkmann} d s^2=2 d U d V+H(U) (Y^2-Z^2) d U^2+ d Y^2+d Z^2 \; ,$$ with $H(U)=-\Psi_4=\ddot{p}(u)/p(u)$. The coordinate transformation to arrive at Eq. (\[Brinkmann\]) from Eq. (\[metric\]) is $u=U$, $v=V+(\dot{p} Y^2/p+\dot{q} Z^2/q)/2$, $y=Y/p$ and $z=Z/p$. This implies that all the relevant information is contained in the non-zero Weyl scalar, $\Psi_4$. However, to perform this coordinate transformation a couple of functions verifying Eq. (\[FieldEqs\]), $p(u)$ and $q(u)$, must exist. We are interested in this paper in finding plane waves series solutions, propagating in vacuum without dispersion, according to the following criteria: 1. The functions $p(u)$ and $q(u)$ are everywhere continuous and differentiable up to second order, at least. This is necessary for the definition of the Christoffel’s symbols and Ricci curvature tensor. 2. The limits $\lim_{u \rightarrow \pm \infty} p(u)$ and $\lim_{u \rightarrow \pm \infty} q(u)$ exist. 3. We have $p(u) \neq 0$ and $q(u) \neq 0$ for any $u \in (-\infty,\infty)$, i. e., there are no focus points. 4. The stress-energy pseudotensor is integrable and the energy content of the wave is finite. These are the Bini et al. [@Bini] conditions particularized to this problem. The idea of the series expansion method is to write both $p(u)$ and $q(u)$ in terms of the same function $h(u)$ which serves as a basis for the series. We assume that $p(u)$ is a simple linear combination of $h(u)$ and a constant and that $q(u)$ is another more complicated functional: $$\begin{aligned} \label{pdef} p(u)&=&1+ \epsilon\, h(u) \; , \\ \noalign{\smallskip} \label{qdef} q(u)&=&{\cal F}(u)=\displaystyle\sum_{n=0}^\infty\, a_n h^n(u) \; ,\end{aligned}$$ where $\epsilon$ is a constant. From Eqs. (\[FieldEqs\]), (\[pdef\]) and (\[qdef\]) we have: $$\label{FuncEq} \epsilon\, \ddot{h} \, {\cal F}(h)+\ddot{h}\, (1+\epsilon h)\, {\cal F}'(h)+\dot{h}^2\, (1+\epsilon h)\, {\cal F}''(h)=0\; ,$$ where ${\cal F}'(h)$, ${\cal F}''(h)$ denote the first and second order derivative of ${\cal F}(h)$ with respect its argument, $h$. Now, for a modal series to be possible for this problem the square of the first-order derivative of the base, $\dot{h}^2(u)$, and its second-order derivative, $\ddot{h}(u)$, should be expressed as a simple polynomial in terms of the function $h(u)$ itself. A good candidate for a basis function is the Lorentzian: $$\label{Lorentz} h(u)=\displaystyle\frac{1}{1+u^2}$$ because it has the correct limit for $u \rightarrow \pm \infty$ and also verifies: $$\begin{aligned} \label{hdotL} \dot{h}^2(u)&=&4h^3(1-h) \; , \\ \noalign{\smallskip} \label{hddotL} \ddot{h}(u)&=&2h^2(3-4 h) \; .\end{aligned}$$ With these two relations in Eqs. (\[hdotL\]) and (\[hddotL\]) we can find the following non-linear differential equation for the functional ${\cal F}(h)$: $$\label{FuncEqL} \begin{array}{rcl} & &\epsilon (3 - 4 h){\cal F}(h)+\left(3+(3\epsilon-4)h-4\epsilon h^2\right){\cal F}'(h)\\ \noalign{\smallskip} &+&2 h\left(1+(\epsilon-1) h- \epsilon h^2\right) {\cal F}''(h)=0\; . \end{array}$$ We now consider a series expansion for ${\cal F}$ as given in Eq. (\[qdef\]). By substitution in Eq. (\[FuncEqL\]) and rearranging the terms in such a way that the coefficients of the same powers of $h$ are clearly displayed we have: $$\begin{array}{rcl} & &\displaystyle\sum_{n=0}^\infty \, 3 \epsilon a_n h^n-\displaystyle\sum_{n=1}^\infty \, 4 \epsilon a_{n-1} h^n+\displaystyle\sum_{n=0}^\infty \, 3 (n+1) a_{n+1} h^n +\sum_{n=1}^\infty\, (3 \epsilon-4) n a_n h^n\\ \noalign{\smallskip} &-&\displaystyle\sum_{n=2}^\infty \, 4(n-1)\epsilon a_{n-1}h^n +\displaystyle\sum_{n=1}^\infty\, 2 (n+1)n a_{n+1} h^n+\displaystyle\sum_{n=2}^\infty\, 2 (\epsilon-1) n(n-1) a_n h^n \\ \noalign{\smallskip} &-&\displaystyle\sum_{n=3}^\infty\, 2 \epsilon (n-1)(n-2) a_{n-1} h^n=0\; . \end{array}$$ We can now identify the coefficients of every power of $h(u)$ with zero in order to obtain the recurrence relations for $a_n$, $n=0,1,2,\ldots$. In particular we have: $$\begin{aligned} \label{a1} a_1&=&-\epsilon a_0 \\ \noalign{\smallskip} a_2&=&\displaystyle\frac{3}{5}\epsilon^2 a_0 \\ \noalign{\smallskip} a_3&=&\displaystyle\frac{1}{21}\left[ 8\epsilon a_1-(13 \epsilon-12) a_2 \right]=-\displaystyle\frac{\epsilon^2}{105}\left(4+39 \epsilon\right) a_0\; ,\end{aligned}$$ and for $n \ge 3$ the following recurrence is found: $$\label{recL} a_{n+1}=\displaystyle\frac{2\epsilon(n^2-n+2)}{(n+1)(2n+3)}a_{n-1}- \displaystyle\frac{3\epsilon+(\epsilon-2)n+2(\epsilon-1)n^2}{(n+1)(2n+3)}a_n\; .$$ From this relation it can be checked that the leading term in the expression of $a_n$ is $\epsilon^2$. So, we cannot claim that the modal series in Eq. (\[qdef\]) for $u=0$ ($h(u)=1$) exhibits geometric convergence for $\epsilon < 1$. On the other hand, a numerical computation of the coefficients $a_n$ shows that they decrease with $n$. However, they decrease only algebraically with $n$ as deduced from the double-logarithmic plot in Fig. \[fig1\] and, consequently, $a_{n+1}/a_n \rightarrow 1$ as $n \rightarrow \infty$. This means that convergence of the series for $q(u=0)$ in Eq. (\[qdef\]) cannot be proved by the quotient test. Convergence of the Lorentzian Plane Gravitational Wave {#sect_3} ====================================================== We have found that the modal series expansion for the square-root of the metric of the gravitational wave, $q(u)$, is: $$\label{qseries} q(u)=\displaystyle\sum_{n=0}^\infty \, a_n h^n(u) \; .$$ As the coefficients $\vert a_n \vert$ constitute a decreasing series for sufficiently large $n$ and $h(u) < 1$ for $u > 0$, the series in Eq. (\[qseries\]) is trivially convergent for any nonzero value of $u$. We will show in this section that $q(u=0)$, i. e., the sum of the coefficients $a_n$ is also convergent. To this end, we define a succession $b_n$ as: $$\label{bRaabe} b_n=\left(a_n/a_{n+1}-1\right)n \; .$$ If this succession $b_n$ has a limit for $n\rightarrow \infty$ and $\lim_{n\to\infty} b_n > 1$ then, according to Raabe-Duhamel’s theorem, the series for $a_n$ is convergent [@Arfken]. By dividing both members of the recurrence relation in Eq. (\[recL\]) by $a_n$ we find that: $$\label{testRaabe} \displaystyle\frac{b_n}{n}+1=\displaystyle\frac{R(n)}{Q(n)\left(b_{n-1}/(n-1)+1\right)-P(n)}\; ,$$ where $P(n)$, $Q(n)$ and $R(n)$ are the following second-order polynomials in $n$: $$\begin{aligned} P(n)&=&3\epsilon+(\epsilon-2) n+2(\epsilon-1)n^2 \\ \noalign{\smallskip} Q(n)&=&2\epsilon(n^2-n+2)\\ \noalign{\smallskip} R(n)&=&(n+1)(2 n+3)\; .\end{aligned}$$ For very large values of $n$ we can assume $b_n/n \simeq b_{n-1}/(n-1)$ and Eq. (\[testRaabe\]) becomes a second-order polynomial equation for $b_n/n+1$. We can test numerically that $b_n$ is positive and the relevant root behaves as: $$\label{bsol} b_n=\displaystyle\frac{3}{2}+\displaystyle\frac{1}{(1+\epsilon) n}+{\cal O}\left(\displaystyle\frac{1}{n^2}\right)\; .$$ Now, we have that $\epsilon \ge -1$ because, otherwise, $p(u)$ as given in Eq. (\[pdef\]) becomes zero for a given $u$ and from condition (iii) we want to avoid focus points. In that case, Eq. (\[bsol\]) implies that $\lim_{n\to \infty} b_n = 3/2 >1$ and the series for $q(u)$ as deduced from Eqs. (\[qdef\]) and (\[recL\]), converges. Notice that the parameter $\epsilon$ is not neccessarily small, it should only be larger than $-1$. In Fig. \[fig2\] we have plotted the functions $p(u)$ and $q(u)$ as given by Eqs. (\[pdef\]), (\[qdef\]) and the recurrence relation for the coefficients $a_n$ in Eq. (\[recL\]) for $\epsilon=0.9$ and $a_0=1$. We notice that the passing gravitational wave will provoke and expansion of rods along the $y$ direction and their contraction in the $z$ direction. So, it could be detected by laser interferometry. Nevertheless, the space-time is asymptotically Minkowskian before and after the passage of the wave as requested in condition (ii). It is interesting to check if a similar solution can be found for another choosing of the basis function $h(u)$. Another convenient basis in terms of the hyperbolic cosine is: $$\label{Hyperbolic} h(u)=\displaystyle\frac{1}{1+\cosh(u)}\; ,$$ which verifies the modal recurrences: $$\begin{aligned} \label{hdotH} \dot{h}^2(u)&=&h^2 (1- 2 h) \\ \noalign{\smallskip} \label{hddotH} \ddot{h}(u)&=&h (1-3 h) \; ,\end{aligned}$$ and the functional Eq. (\[FuncEq\]) takes the form: $$\label{FuncEqH} \begin{array}{rcl} & &\epsilon\, (1 - 3 h)\, {\cal F}(h)+\left(1+(\epsilon-3)h-3 \epsilon h^2\right)\,{\cal F}'(h) \\ \noalign{\smallskip} &+& h\,\left(1+(\epsilon-2) h- 2 \epsilon h^2\right)\, {\cal F}''(h)=0\; . \end{array}$$ The non-linear differential equation for ${\cal F}(h)$ can be solved by the modal series expansion as before and we obtain the coefficients: $$\begin{aligned} a_1&=&-\epsilon a_0 \\ \noalign{\smallskip} a_2&=&\displaystyle\frac{3 \epsilon}{4} a_0-\displaystyle\frac{(2\epsilon-3)}{4} a_1=\displaystyle\frac{\epsilon^2}{2} a_0 \\ \noalign{\smallskip} a_3&=&\displaystyle\frac{2 \epsilon}{3} a_1+\displaystyle\frac{5}{9} (2-\epsilon) a_2=-\displaystyle\frac{1}{18}\epsilon^2 (2+5\epsilon) \; ,\end{aligned}$$ and for $n \ge 3$ the following general recurrence relation is found: $$\label{recH} a_{n+1}=\epsilon \displaystyle\frac{(2 n^2-3 n+4)}{(n+1)^2} a_{n-1}+\displaystyle\frac{(2-\epsilon) n^2+n-\epsilon}{(n+1)^2} a_n\; .$$ The convergence of the series for $q(u=0)$ requires the existence of the following sum: $$\label{alphadef} \displaystyle\sum_{n=0}^\infty \, \displaystyle\frac{a_n}{2^n}=\displaystyle\sum_{n=0}^\infty \alpha_n\; .$$ The application of Raabe-Duhamel’s test in this case for the succession $b_n=(\alpha_n/\alpha_{n+1}-1)n$ proceeding in the same way as before yields: $$\label{RaabeH} b_n=\displaystyle\frac{3}{2}+\displaystyle\frac{2+5\epsilon}{4(2+\epsilon)} \displaystyle\frac{1}{n}+{\cal O}\left(\displaystyle\frac{1}{n^2}\right)\; .$$ So, $\lim_{n\to\infty} b_n=3/2$ for any $\epsilon > -2$ and the series in Eq. (\[alphadef\]) and also $q(u)$ as defined in Eq. (\[qdef\]) with the coefficients given by the recurrence in Eq. (\[recH\]) is convergent. For $\epsilon \le -2$ the metric coefficients become null for some values of $u$ and the condition (iii) is not fulfilled. So, we should restrict the solution to values $\epsilon > -2$. This gravitational wave is plotted in Fig. \[fig3\] for $\epsilon=0.9$ and $a_0=1$. Finally, we must discuss the convergence of the first and second-order derivatives of $q(u)$ because they are related with Christoffel symbols and the Ricci tensor. The first derivative is zero for $u=0$ because the derivative of every term in the series of Eq. (\[qdef\]) is proportional to $\dot{h}(u)$. Concerning the second-order derivative the situation is more subtle because it is given by the series: $$\label{qddseries} \ddot{q}(u)=\displaystyle\sum_{n=2}^\infty\, a_n n (n-1) h^{n-2}(u) \dot{h}^2(u)+\displaystyle\sum_{n=1}^\infty\, a_n n h^{n-1}(u) \ddot{h}(u)\; ,$$ For $u\neq 0$ both series in Eq. (\[qddseries\]) converge by the quotient test because $\lim_{n\to\infty} a_{n+1}/a_{n}=1$ and $h(u) < 1$. But for $u=0$ we face a problem with the second term in Eq. (\[qddseries\]) which becomes proportional to $\sum_{n=1}^\infty \, n a_n$ and, consequently, diverges. However the second derivative $\ddot{q}(0)$ can still be defined as the right and left limits of $\ddot{q}(u)$ for $u \to 0$ or, directly, from Eq. (\[FieldEqs\]) as follows: $$\ddot{q}(0)=-q(0)\displaystyle\frac{\ddot{p}(0)}{p(0)}\; .$$ For example, in the case of the Lorentzian solution we obtain $\ddot{q}(u=0)=0.314565$ by computing $q(0)$ with $10^6$ terms in the series of Eq. (\[qdef\]). In comparison we have $\ddot{q}(u=0.1)=0.318772$ directly from the series in Eq. (\[qddseries\]) and with the same number of terms. This proves the consistency of the solution. Energy-stress pseudotensor and the energy content of the wave {#sect_4} ============================================================= Gravitational waves carry a certain amount of energy. This gravitational energy can be calculated from the so-called energy-stress pseudotensor [@Landau]: $$\label{pseudo} \begin{array}{rcl} t^{ik}&=&\displaystyle\frac{c^4}{16 \pi G} \left\{ \left( 2 \Gamma_{lm}^n \Gamma_{np}^p-\Gamma_{lp}^n\Gamma_{mn}^p -\Gamma_{ln}^n\Gamma_{mp}^p \right)\left(g^{il} g^{km}-g^{ik} g^{lm}\right) \right.\\ \noalign{\smallskip} &+&g^{il} g^{mn} \left( \Gamma_{lp}^k \Gamma_{mn}^p+\Gamma_{mn}^k \Gamma_{lp}^p -\Gamma_{np}^k\Gamma_{lm}^p- \Gamma_{lm}^k \Gamma_{np}^p \right)+\left[ i,k\right]\\ \noalign{\smallskip} &+& \left. g^{lm} g^{np} \left( \Gamma_{ln}^i \Gamma_{mp}^k-\Gamma_{lm}^i \Gamma_{np}^k\right) \right\} \; , \end{array}$$ where $[i,k]$ denotes a term coinciding with the previous one save for a permutation of the indexes $i$ and $k$. $G$ is, as usual, the gravitational constant. For the Einstein-Rosen metric in Eq. (\[metric\]) we get the following result: $$\label{tpmetric} t^{00}(u)=t^{01}(u)=-\displaystyle\frac{c^4}{8 \pi G} \left\{\dot{p}^2(u) \, q^2(u)+\dot{q}^2(u)\, p^2(u)+4 \dot{p}(u)\, p(u)\, \dot{q}(u) \, q(u)\right\} \; .$$ We have explicitly calculated the components of the plane wave solutions found in Sec. \[sect\_2\] and the result is plotted in Fig. \[fig4\]. We notice an interesting mathematical feature of these solutions: the energy density of the plane wave may become negative around its center. This is not physically problematic because the energy of a gravitational field is not localized and the only relevant information is the total energy which is positive and finite (For example, $E_{\mbox{total}} = 1.06377\, c^4/(16 \pi G)$ for the Lorentzian wave). One of the advantages of the modal series method discussed in this paper is that the integrability of the stress-energy pseudotensor can be easily proved. The asymptotic behaviour of the components in Eq. (\[tpmetric\]) is given by: $$\label{tpasymp} t^{01}(u) \rightarrow \displaystyle\frac{\epsilon c^4}{4 \pi G} \, \dot{h}^2(u) \; , \; u \rightarrow \pm\infty$$ where we have taken into account Eqs. (\[pdef\]), (\[qdef\]) and (\[a1\]). In the two cases considered in Eqs. (\[Lorentz\]) and (\[Hyperbolic\]) the function $\dot{h}^2(u)$ is integrable in the domain $-\infty < u < \infty$. This is a condition we should impose to any function $h(u)$ used as seed of the modal series method. Conclusions {#sect_5} =========== In this paper we have proposed a modal transseries approach [@Costin] to the systematic analysis of plane wave solutions of Einstein’s Field Equations inspired by the success of this technique in other non-linear problems [@AcedoNARWA; @AcedoPhysA; @Acedo2013]. We have shown that a family of plane waves can be found in vacuum and that these solutions satisfy requirements of physical plausibility: the space-time is Minkowskian very far away from the plane-fronted wave, the metric coefficients are regular and everywhere different from zero, curvature is properly defined and the total gravitational energy content of the wave is finite. These waves travel without changing their shape. However, the shape of these waves is not completely determined by Einstein’s Field Equations as occurs, for example, in the Korteweg-de Vries soliton as a consequence of the interplay between non-linearity and dissipation. It would also be interesting to study the exact spherical symmetric waves and its relation to a binary pulsar or binary black hole source. Moreover, the method developed in this paper could be applied to Non-Abelian waves in Yang-Mills theories which have been analyzed since the seventies of the past century [@Coleman; @Campbell; @Rabinowitch]. Another interesting problem should be the analysis of collisions of plane waves [@Griffiths; @Feinstein] of the type described in this paper by finding a generalization of the modal series with two fronts. Work along these lines will be published in future papers. Acknowledgments {#acknowledgments .unnumbered} =============== The author gratefully acknowledges A. Feinstein for his very useful comments and a critical reading of the manuscript. References {#references .unnumbered} ========== [99]{} A. Einstein and N. Rosen, Journal of the Franklin Institute [**223**]{}, 43-54 (1937). N. Rosen, Gen. Rel. and Grav. [**10**]{}, 351-364 (1979). C. Sivaram, Bull. Astr. Soc. India [**23**]{}, 77-83 (1995). J. M. Weisberg, D. J. Nice and J. H. Taylor, The Astrophysical Journal [**722**]{}, 1030-1034 (2010); arXiv:1011.0718v1. J. B. Griffiths, [*Colliding waves in general relativity*]{} (Oxford: Clarendon Press, 1991). S. Chandrasekhar, [*The mathematical theory of black holes*]{} (Oxford: Clarendon Press, 1983). D. Bini, V. Ferrari and J. Ibañez, Il Nuovo Cimento [**103 B, N. 1**]{} 29-44, 1989. L. Acedo, G. González-Parra and A. J. Arenas, Nonlinear Analysis: Real World Applications [**11**]{} 1819-1825, (2010). L. Acedo, G. González-Parra and A. J. Arenas, Physica A [**389**]{} 1151-1157, (2010). G. González-Parra, L. Acedo and A. J. Arenas, Numerical Algorithms, published online 2013, DOI 10.1007/s11075-013-9776-x. W. Rindler, [*Relativity: Special, General and Cosmological*]{} (New York: Oxford Univ. Press Inc., Second Edition, 2006). G. Arfken, [*Mathematical Methods for Physicists, 3rd. Ed.*]{}, (Academic Press, Orlando, Florida, 1985). L. D. Landau and E. M. Lifshitz, [*The Classical Theory of Fields*]{} (Volume 2: A Course of Theoretical Physics) (New York: Pergamon Press, Third Revised English Edition, 1971) O. Costin, [*Topological construction of transseries and introduction to generalized Borel summability*]{}, in [*Analyzable Functions and Applications*]{} (O. Costin, M. D. Kruskal and A. Macintyre, Eds.), [*Contemp. Math.*]{} [**373**]{} (Providence, RI, USA: Am. Math. Soc., 2005); arXiv: math/0608309v1. S. R. Coleman, Phys. Lett. B [**70**]{} 59-60, (1977). W. B. Campbell and T. A. Morgan, Phys. Lett. B [**84**]{} 87-88, (1979). A. S. Rabinowitch, Int. J. of Adv. Math. Sciences [**1(3)**]{} 109-121, (2013). A. Feinstein and J. Ibañez, Phys. Rev. D [**39(2)**]{} 470-473, (1989). [^1]: E-mail: luiacrod@imm.upv.es

ArXiv

--- abstract: 'It has been shown experimentally that the $^6$Li nucleus shrinks by adding a $\Lambda$ particle while its spectrum is also compressed. We discuss the compatibility of these two effects, contrasting also with a relation between the shrinkage effect and the stability of the spectrum in the $^9_\Lambda$Be hypernucleus. To this end, we employ two-body $d$-$^5_\Lambda$He and $\alpha$-$^5_\Lambda$He cluster models for the $^7_\Lambda$Li and $^9_\Lambda$Be hypernuclei, respectively. We first argue that a Gaussian-like interaction between two clusters leads to a stabilization of the spectrum against an addition of a $\Lambda$ particle, even though the intercluster distance is reduced. In the case of $^7_\Lambda$Li, the spin-orbit interaction between the intercluster motion and the deuteron spin has to be considered also. We show that the shrinkage effect makes the expectation value of the spin-orbit potential larger, lowering the excitation energy for the $^6$Li($3^+)\otimes \Lambda(1/2^+)$ level.' author: - 'K. Hagino and T. Koike' title: ' Relation between shrinkage effect and compression of rotational spectrum in $^7_\Lambda$Li hypernucleus' --- Introduction ============ One of the main interests in physics of hypernuclei is to investigate the impurity effect of $\Lambda$ particle on the structure of atomic nuclei. This includes the change of [*e.g.,*]{} nuclear size[@MBI83], the density distribution[@HKYMR10], deformation properties[@Z80; @ZSSWZ07; @WH08; @SWHS10; @WHK11; @YLHWZM11; @LZZ11; @IKDO11], the neutron drip-line[@VPLR98; @ZPSV08], and fission barrier[@MCH09]. A characteristic feature of $\Lambda$ particle is that it is free from the Pauli principle from nucleons, and thus it can deeply penetrate into the nuclear interior. It has been predicted that the $\Lambda$ particle in the center of a nucleus attracts surrounding nucleons, leading to a shrinkage of nuclear size [@MBI83]. High precision $\gamma$-ray spectroscopy measurements have by now been systematically carried out for the p-shell $\Lambda$ hypernuclei [@HT06]. Such experiments have revealed that the electronic quadrupole transition probability, $B(E2)$, from the first excited state (3$^+$) to the ground state (1$^+$) of $^6$Li is considerably reduced when a $\Lambda$ particle is added [@T01]. By adopting a simple $\alpha$-$d$ and $^5_\Lambda$He-$d$ cluster models for $^6$Li and $^7_\Lambda$Li, respectively, the reduction in $B(E2)$ has been interpreted as a reduction of the intercluster distance by 19%[@T01], in accordance with theoretical calculations [@MBI83; @HKMM99]. Although this interpretation has been well accepted, the spectrum of $^7_\Lambda$Li shows a somewhat puzzling feature. That is, one would naively think that the reduction of the intercluster size leads to a smaller moment of inertia, and thus the rotational excitation energy would increase. This has indeed been observed in the $^{13}_{~\Lambda}$C hypernucleus[@K02] (notice that only the energy of the 3/2$^+$ state has been measured in Ref. [@K02]. However, the theoretical calculation suggests that the partner state of the doublet (5/2$^+$) lies below the 3/2$^+$ state only by at most 0.36 MeV [@HKMYY00]. This corresponds to the spin averaged energy of the doublet of 4.66 MeV, that is compared to the energy of the 2$^+$ state in $^{12}$C, 4.44 MeV). On the contrary, the observed difference in the spin averaged energy between the first ($1/2^+,3/2^+$) and the second ($5/2^+,7/2^+$) doublet levels in $^7_\Lambda$Li is 1.858 MeV [@U06], that is smaller than the excitation energy (2.186 MeV) of the first excited state of $^6$Li (see Fig. 1). For the $^9_\Lambda$Be hypernucleus, the experimental data show that the spin averaged energy of the $^8$Be(2$^+$)$\otimes \Lambda(1/2^+)$ doublet states (3/2$^+$, 5/2$^+$) is almost the same as the energy of the first 2$^+$ state of the $^8$Be nucleus [@A02; @T05], despite a similar shrinkage effect, expected as in the $^7_\Lambda$Li nucleus[@MBI83]. The stability of the rotational energy of $^9_\Lambda$Be was shown to remain the same even when the $\Lambda$ particle is replaced by a charmed baryon $\Lambda_c$ [@BB82]. Apparently, these behaviors of the spectrum of the $^7_\Lambda$Li and $^9_\Lambda$Be cannot be understood in analogy of [*e.g.,*]{} a classical rotor of diatomic molecules. We summarize the experimental spaectra in Fig. 1. ![The experimental low-lying spectra for $^6$Li, $^7_\Lambda$Li, $^8$Be, $^9_\Lambda$Be, $^{12}$C, and $^{13}_{~\Lambda}$C nuclei. The energies are denoted in the units of MeV. The energies for the 1$^+\otimes$1/2$^+$ and 2$^+\otimes$1/2$^+$ levels of the $^7_\Lambda$Li nucleus are obtained by spin-averaging the observed level energies for the 1/2$^+$ (0 MeV) and the 3/2$^+$ (0.069 MeV) states, and the 5/2$^+$ (2.05 MeV) and the 7/2$^+$ (2.52 MeV) states[@U06], respectively. The energy for the 2$^+\otimes$1/2$^+$ level of the $^9_\Lambda$Be is the spin averaged energy between the 5/2$^+$ (3.024 MeV) and 3/2$^+$ (3.067 MeV) states [@T05]. In the $^{13}_{~\Lambda}$C nucleus, only the energy for the 3/2$^+$ state has been measured [@K02], while the energy for the 5/2$^+$ state is expected to be lower than the energy of the 3/2$^+$ state by at most 0.36 MeV [@HKMYY00]. ](fig1) In this paper, we clarify the relation between the two contradictory effects of $\Lambda$ particle on the structure of light atomic nuclei, that is, the shrinkage of the size and the compression/stabilization of the spectrum. An important point to remember is that the $\alpha$ cluster structure is well developed in the ground state of $^6$Li and $^8$Be, while the ground state takes the mean-field-like structure in $^{12}$C and in heavier nuclei. Our aim in this paper is thus to discuss how the spectra of $^7_\Lambda$Li and $^9_\Lambda$Be are compatible with the shrinkage of their size from a viewpoint of a two-body cluster model. The paper is organized as follows. In Sec. II, we first discuss the impurity effect of $\Lambda$ in the $^9_\Lambda$Be nucleus. Using the $\alpha$-$^5_\Lambda$He cluster model together with the semi-classical Bohr-Sommerfeld quantization rule, we investigate how a $\Lambda$ particle influences the rotational spectrum. In Sec. III, we discuss the rotational spectrum of the $^7_\Lambda$Li nucleus. We show that the spin-orbit interaction due to the deuteron spin plays an important role in lowering the excitation energy of the $^6$Li$\otimes\Lambda(1/2^+)$ level. We then summarize the paper in Sec. IV. stabilization of level scheme for $^9_\Lambda$Be ================================================ We first discuss the level scheme of the $^9_\Lambda$Be hypernucleus, and clarify the reasoning why the influence of the $\Lambda$ particle in the level spacing is negligibly small. For this purpose, we use a simple two-body cluster model with $\alpha$-$^5_\Lambda$He configuration. Assuming a local potential, an intercluster potential at an intercluster distance $R$ reads $$V_{\alpha^5_\Lambda {\rm He}}(R)=V^{(N)}_{\alpha\alpha}(R)+V^{(C)}_{\alpha\alpha}(R) +V_{\alpha\Lambda}(R),$$ where $V^{(N)}_{\alpha\alpha}$ and $V^{(C)}_{\alpha\alpha}$ are the nuclear and the Coulomb parts of the potential between the $\alpha$ particle and the core nucleus ($^4$He) of the $^5_\Lambda$He, and $V_{\alpha\Lambda}$ is the potential between the $\Lambda$ particle in $^5_\Lambda$He and the $\alpha$ particle. Assuming that the center of mass of $^5_\Lambda$He is identical to the center of mass of the core nucleus $^4$He, we use the same potential as that given in Ref. [@BFW77] for $V^{(N)}_{\alpha\alpha}$ and $V^{(C)}_{\alpha\alpha}$. That is, $$V^{(N)}_{\alpha\alpha}(R)=-V_0\,e^{-\alpha R^2}, \label{gauss_aa}$$ with $V_0$=122.6225 MeV and $\alpha$=0.22 fm$^{-2}$, and $$V^{(C)}_{\alpha\alpha}(R)=\frac{4e^2}{R}\,{\rm erf}(\beta R^2),$$ with $\beta$=0.75 fm$^{-1}$, where erf$(x)$ is the error function. For $V_{\alpha\Lambda}$, we assume that it is given by the direct part of the double folding potential, that is, $$V_{\alpha\Lambda}(R)= \int d{\mbox{\boldmath $r$}}_Nd{\mbox{\boldmath $r$}}_\Lambda\, \rho_\alpha({\mbox{\boldmath $r$}}_N) \rho_\Lambda({\mbox{\boldmath $r$}}_\Lambda) v_{N\Lambda}({\mbox{\boldmath $R$}}+{\mbox{\boldmath $r$}}_N-{\mbox{\boldmath $r$}}_\Lambda), \label{dfm}$$ where $v_{N\Lambda}$ is a nucleon-$\Lambda$ particle interaction, while $\rho_\alpha$ and $\rho_\Lambda$ are the density distributions for the $\alpha$ particle and the $\Lambda$ particle in $^5_\Lambda$He, respectively. We assume Gaussian density distributions for $\rho_\alpha$ and $\rho_\Lambda$, $$\begin{aligned} \rho_\alpha(r)&=&4(\pi b_\alpha)^{-3/2}e^{-r^2/b_\alpha^2}, \\ \rho_\Lambda(r)&=&(\pi b_\Lambda)^{-3/2}e^{-r^2/b_\Lambda^2}. \end{aligned}$$ Following Ref. [@MBI83], we take $b_\alpha=1.358$ fm, and $b_\Lambda=\sqrt{(4M_N+M_\Lambda)/4M_\Lambda}\,b_\alpha$, where $M_N$ and $M_\Lambda$ are the mass of nucleon and $\Lambda$ particle, respectively. For $v_{N\Lambda}$, we take the central part of the potential given in Ref. [@MBI83], $$v_{N\Lambda}(r)=v_0\,e^{-r^2/b_v^2},$$ with $b_v=1.034$ fm, but we adjust the strength $v_0$ so as to reproduce the energy of the ground state of $^9_\Lambda$Be from the threshold of the $\alpha+^5_\Lambda$He configuration (that is, $-$3.50 MeV). Since $\rho_\alpha$, $\rho_\Lambda$, and $v_{N\Lambda}$ are all given in a Gaussian form, the double folding integral in Eq. (\[dfm\]) can be evaluated analytically as $$V_{\alpha\Lambda}(R)= 4v_0\left(\frac{b_v^2}{b_\Lambda^2+b_\alpha^2+b_v^2}\right)^{3/2} \,\exp\left(-\frac{r^2}{b_\Lambda^2+b_\alpha^2+b_v^2}\right).$$ With the intercluster potential so constructed, we solve the Schrödinger equation for the relative motion for each angular momentum $L$, $$\left(-\frac{\hbar^2}{2\mu_{\alpha^5_\Lambda{\rm He}}}{\mbox{\boldmath $\nabla$}}^2 +V_{\alpha^5_\Lambda{\rm He}}(R)-E\right)\psi_L({\mbox{\boldmath $R$}})=0,$$ where $\mu_{\alpha^5_\Lambda{\rm He}}=4M_N(4M_N+m_\Lambda)/(8M_N+M_\Lambda)$ is the reduced mass for the relative motion between $\alpha$ and $^5_\Lambda$He. The Pauli principle is taken into account by excluding those states which satisfy $2n+L<4$ [@BFW77], where $n$ is the radial node of the wave function. The E2 transition probability can be computed with the intercluster wave functions $\psi_L({\mbox{\boldmath $R$}})$. In the two-body cluster model of $(A_1,Z_1)+(A_2,Z_2)$, the E2 transition operator reads $$\hat{T}_{\rm E2} = e_{\rm E2}R^2Y_{2\mu}(\hat{{\mbox{\boldmath $R$}}}),$$ where $$e_{\rm E2}=\left(\frac{M_1}{M_1+M_2}\right)^2Z_2+\left(\frac{M_2}{M_1+M_2}\right)^2Z_1,$$ is the E2 effective charge, $M_1$ and $M_2$ being the mass of the fragments 1 and 2, respectively.  The calculated spectra for the $^8$Be and $^9_\Lambda$Be are shown in Fig. 2. Since we do not include the spin-dependent part of nucleon-$\Lambda$ interaction, the doublet states (3/2$^+$,5/2$^+$) of $^8$Be(2$^+)\otimes\Lambda(1/2^+)$ are degenerate in energy. In order to obtain the energy and the wave function for resonance states, we use a bound state approximation. That is, we have replaced the intercluster potential for $R$ larger than the barrier radius with a constant so that the wave functions are confined inside the potential barrier[@JRB77; @GK87; @G88]. We have confirmed that the energies obtained with this procedure are close to those energies that give the maximum of the energy derivative of the scattering phase shift. For instance, for the 2$^+$ state of $^8$Be, the energy derivative of the phase shift is maximum at $E$=2.80 MeV, while the energy obtained in the bound state approximation is 2.85 MeV.  From Fig. 2, one sees that the excitation energy slightly increases due to the addition of $\Lambda$ particle. We have confirmed numerically that the potential which we use well reproduces the experimental phase shift for $\alpha$+$\alpha$ scattering[@BFW77]. If the resonance states were properly analysed, this potential would therefore yield the excitation energy of the 2$^+$ state in $^8$Be of 3.04 MeV (see Fig. 1). Then, the energy shift due to the addition of $\Lambda$ particle is estimated to be small indeed, that is, as small as 0.05 MeV. In contrast to the excitation energies, the B(E2) value from the first excited state to the ground state is altered drastically. The large change of $B(E2)$ value is due to the fact that the attraction of the $\Lambda$ particle makes the resonance levels of $^8$Be turn to the bound states in $^9_\Lambda$Be. As a consequence, the wave functions become spatially much more compact, as shown in Fig. 3. The root-mean-square distance between the fragments is 4.90 fm and 3.24 fm for the $L=0$ states of $^8$Be and $^9_\Lambda$Be, respectively. The energy change due to the $\Lambda$ particle shown in Fig. 2, that is, 0.14 MeV, is much smaller than what would have been expected from a classical rotor, $E_L=L(L+1)\hbar^2/2\mu R^2$, which corresponds to the energy shift of 2.64 MeV. Notice that the B(E2) value for the transition from the 2$^+$ to the 0$^+$ states in $^8$Be is consistent with the value obtained in Ref. [@LR86] using a damping factor method to evaluate the integral with the resonance wave functions, although this value is rather large as compared with that obtained in Ref. [@MBI83] using the harmonic oscillator expansion of the wave functions.  We now ask a question why the spectrum of $^9_\Lambda$Be is not influenced much, despite the fact that the wave functions are considerably altered. In order to address this question, the semi-classical approximation may be useful. In Ref. [@R77], Rowley used the Bohr-Sommerfeld quantization rule to show that the spectrum is approximately given by $$E_L-E_{L=0}\sim \frac{\beta\hbar^2}{8\mu}\,L(L+1), \label{rowley}$$ for a Gaussian intercluster potential, $V(R)=V_0\,e^{-\beta R^2}$. We numerically confirm this relation in Appendix A. This formula indicates that the spectrum depends on neither the depth parameter $V_0$ nor the intercluster distance, but only on the range parameter $\beta$ of the potential as well as the reduced mass $\mu$. We show the intercluster potentials for the present problem in Fig. 4 by the solid lines. The potential for the $\alpha+^5_\Lambda$He is deeper than that for $\alpha+\alpha$ by 15.6 MeV, due to an additional attraction caused by the $\Lambda$ particle. These potentials are actually well fitted with the Gaussian form with $V_0=-117.8$ MeV and $\beta=0.222$ fm$^{-2}$ for $^8$Be and with $V_0=-131.2$ MeV and $\beta=0.250$ fm$^{-2}$ for $^9_\Lambda$Be. The quality of the fit can be seen in Fig. 4. The reduced masses, on the other hand, are 1877.8 MeV and 2120.6 MeV for the $\alpha+\alpha$ and the $\alpha+^5_\Lambda$He systems, respectively. Here, we have used the average value of the proton and neutron masses as the nucleon mass. The factor $\beta\hbar^2/8\mu$ in Eq. (\[rowley\]) is thus 0.574 MeV for the $\alpha+\alpha$ system and 0.572 MeV for the the $\alpha+^5_\Lambda$He system. It is remarkable that these values are considerably close to each other. Evidently, a Gaussian-like intercluster potential has a responsibility to make the spectra resemble to each other between $^8$Be and $^9_\Lambda$Be, despite that the absolute value of the energies and thus the radial dependence of the wave functions are considerably changed due to an addition of a $\Lambda$ particle. Level scheme for $^7_\Lambda$Li: role of spin-orbit interaction =============================================================== Let us discuss next the spectrum of the $^7_\Lambda$Li nucleus. If we assume that the nucleus $^6$Li takes the $\alpha$+deuteron ($d$) structure, a big difference of this nucleus from $^8$Be=$\alpha+\alpha$ is that one of the fragments ([*i.e.,*]{} the deuteron) has a finite spin ($S=1$). The potential between $\alpha$ and $d$ then reads [@NTJ83; @NTJK84; @MR85], $$V_{\alpha d}(R)=V_0(R)+V_1(R){\mbox{\boldmath $L$}}\cdot{\mbox{\boldmath $S$}}+ V_2(R)\left[\frac{({\mbox{\boldmath $S$}}\cdot{\mbox{\boldmath $R$}})^2}{R^2}-\frac{1}{3}{\mbox{\boldmath $S$}}^2\right].$$ The spectrum of the ground rotational band of $^6$Li can be understood as follows[@MR85]. This band arises from the states with $2n+L=2$, that is, $L=0$ (1s) and $L=2$ (0d). By combining with the deuteron spin $S=1$, the 1s state yields the total spin and parity of $I^\pi$=1$^+$, while the 0d state yields $I^\pi=1^+,2^+$, and $3^+$ triplet states. These triplet states are degenerate in energy in the central potential $V_0(R)$, but they split by the spin-orbit interaction $V_1(R){\mbox{\boldmath $L$}}\cdot{\mbox{\boldmath $S$}}$ as well as by the tensor interaction $V_2(R)\left[({\mbox{\boldmath $S$}}\cdot\hat{{\mbox{\boldmath $R$}}})^2-{\mbox{\boldmath $S$}}^2/3\right]$. In order to investigate how the spin-orbit intercluster potential affects the spectrum of the $^7_\Lambda$Li nucleus, we closely follow Ref. [@MR85] and use the semi-microscopic cluster model of Buck [*et al.*]{}[@BMR79]. To do so, we assume that the $^7_\Lambda$Li nucleus takes the $d+^5_\Lambda$He structure. In this model, the intercluster potentials $V_0(R), V_1(R)$, and $V_2(R)$ are constructed based on a core+$p$+$n$ three-body model. We regard the $\alpha$ particle and $^5_\Lambda$He as the core nucleus for $^6$Li and $^7_\Lambda$Li, respectively. In the following, for simplicity, we assume that the deuteron is in a pure s-state, and thus the tensor part of the intercluster potential vanishes. This term plays an essential role in reproducing the quadrupole moment of $^6$Li[@NTJ83; @NTJK84; @MR85], but its influence is expected to be small for the spectrum. With this approximation, the central and the spin-orbit potentials read [@MR85] $$\begin{aligned} V_0(R)&=&-2\int dr\, U_0(r,R)\chi(r)^2, \\ V_1(R)&=&-\frac{2M_N}{2(M_N+M_c)} \nonumber \\ &&\times \int dr\,\left[W_0(r,R)+\frac{r}{2R}W_1(r,R)\right]\chi(r)^2, \label{LS}\end{aligned}$$ where $M_c$ is the mass of the core nucleus ([*i.e.,*]{} $^4$He or $^5_\Lambda$He) and $\chi(r)$ is a s-state wave function for relative motion between $p$ and $n$ in the deuteron. $U_0, W_0$ and $W_1$ are multipole components of the nucleon-core interaction, $$V_{cN}=-U({\mbox{\boldmath $R$}}\pm{\mbox{\boldmath $r$}}/2)-W({\mbox{\boldmath $R$}}\pm{\mbox{\boldmath $r$}}/2)\,{\mbox{\boldmath $l$}}\cdot{\mbox{\boldmath $s$}},$$ where ${\mbox{\boldmath $r$}}$ is the relative distance between $p$ and $n$ in the deuteron, ${\mbox{\boldmath $l$}}$ is the relative angular momentum between the core and the nucleon, and ${\mbox{\boldmath $s$}}$ is the nucleon spin. That is, $U_0, W_0$ and $W_1$ are defined as $$\begin{aligned} U({\mbox{\boldmath $R$}}\pm{\mbox{\boldmath $r$}}/2)&=&4\pi\sum_{\lambda}U_\lambda(R,r)Y_\lambda(\pm\hat{{\mbox{\boldmath $r$}}})\cdot Y_\lambda(\hat{{\mbox{\boldmath $R$}}}), \\ W({\mbox{\boldmath $R$}}\pm{\mbox{\boldmath $r$}}/2)&=&4\pi\sum_{\lambda}W_\lambda(R,r)Y_\lambda(\pm\hat{{\mbox{\boldmath $r$}}})\cdot Y_\lambda(\hat{{\mbox{\boldmath $R$}}}).\end{aligned}$$ In the calculations shown below, following Ref. [@MR85], we employ the Hulthen form of the deuteron wave function, that is, $\chi(r)=\sqrt{2\alpha}\,e^{-\alpha r}$ with $\alpha$=0.2316 fm$^{-1}$, and the same potential as in Ref. [@MR85] (with a slight readjustment for the depth parameters) for the nucleon-$\alpha$ potential. The contribution of the $\Lambda$ particle to the $^5_\Lambda$He-nucleon potential is estimated with a single folding potential, $$U_{N\Lambda}(r)= \int d{\mbox{\boldmath $r$}}_\Lambda\, \rho_\Lambda({\mbox{\boldmath $r$}}_\Lambda) v_{N\Lambda}({\mbox{\boldmath $r$}}+{\mbox{\boldmath $r$}}_N-{\mbox{\boldmath $r$}}_\Lambda),$$ where the $\Lambda$ particle density $\rho_\Lambda$ and the nucleon-$\Lambda$ interaction $v_{N\Lambda}$ are given in the previous section. As in Eq. (\[dfm\]), this potential can be computed analytically as $$U_{N\Lambda}(r)= v_0\left(\frac{b_v^2}{b_\Lambda^2+b_v^2}\right)^{3/2} \,\exp\left(-\frac{r^2}{b_\Lambda^2+b_v^2}\right).$$ We adjust the value of $v_0$ in order to reproduce the ground state energy of the $^7_\Lambda$Li nucleus (that is, $-$3.48 MeV from the threshold of $^5_\Lambda$He+$d$). The total central potential for $^5_\Lambda$He-nucleon is given by $U(r)=U_{N\alpha}(r)+U_{N\Lambda}(r)$, where $U_{N\alpha}(r)$ is the central part of the nucleon-$\alpha$ potential. Since we neglect the spin-orbit interaction in the nucleon-$\Lambda$ interaction, the $\Lambda$ particle does not contribute to the spin-orbit part of the intercluster potential, $V_1(R)$, except for the trivial mass factor in Eq. (\[LS\]).    The central part $V_0(R)$ and the spin-orbit part $V_1(R)$ of the intercluster potentials are shown in Fig. 5. For the central part, we also include the Coulomb potential evaluated in the same way as in Ref. [@MR85]. The solid and the dashed lines are the potentials for the $^5_\Lambda$He-$d$ and the $\alpha$-$d$, respectively. The depth of the central part increases by 5.98 MeV due to an addition of a $\Lambda$ particle, while the change of the spin-orbit potential is much smaller. The spectra obtained with these potentials are shown in Fig. 6. In the absence of the spin-orbit potential, $V_1(R)$, an addition of the $\Lambda$ particle slightly increases the energy of the first excited state, from 3.72 MeV in $^6$Li to 3.80 MeV in $^7_\Lambda$Li. This behavior is similar to the change of the spectrum of $^8$Be shown in Fig. 2. The radial wave functions for the ground state and the first excited state obtained without the spin-orbit potential are shown in Figs. 7(a) and 7(b), respectively. As in the case of $^9_\Lambda$Be, the shrinkage effect of the $\Lambda$ particle is substantial [@MBI83; @HKMM99]. Let us now consider the effect of the spin-orbit potential. We fist treat it by the first order perturbation theory, that is, the energy shift for $L=2$ and $I=3$ is approximately given by $$\Delta E_{L=2,I=3}\sim 2\int dR V_1(R)\left[u_{L=2}^{(0)}(R)\right]^2. \label{pert}$$ Here, the factor 2 is the eigenvalue of the operator ${\mbox{\boldmath $L$}}\cdot{\mbox{\boldmath $S$}}$ for $L=2$ and $I=3$, and $u_{L=2}^{(0)}$ is the unperturbed wave function for $L=2$ obtained without the spin-orbit potential. Using the wave functions shown in Fig. 7(b), we obtain $\Delta E_{L=2,I=3}=-1.22$ MeV for $^6$Li and $-1.84$ MeV for $^7_\Lambda$Li. Fig. 7(c) show the integrand of Eq. (\[pert\]). One can clearly see that the overlap of the wave function with the spin-orbit potential shown in the lower panel of Fig. 5 increases significantly for the $^7_\Lambda$Li hypernucleus due to the shrinkage effect, leading to the larger value of the energy shift $\Delta E_{L=2,I=3}$. The spectra obtained by treating the spin-orbit potential exactly, and the corresponding wave functions are shown in Figs. 6 and 7(d), respectively. The effect of the spin-orbit potential is indeed much larger in $^7_\Lambda$Li due to the shrinkage effect, and the energy of the $^6$Li(3$^+)\otimes\Lambda(1/2^+)$ state appears lower than that of the $^6$Li(3$^+$) state. We thus conclude that the spin-orbit potential due to the finite deuteron spin plays an important role in compressing the spectrum of the $^7_\Lambda$Li hypernucleus. We mention that the calculated $B(E2)$ value from the 5/2$^+$ member of the $^6$Li(3$^+)\otimes\Lambda(1/2^+)$ state to the 1/2$^+$ state in the ground state $^6$Li(1$^+)\otimes\Lambda(1/2^+)$ doublet is 3.03 $e^2$fm$^4$, that is in good agreement with the experimental value, 3.6$\pm$0.5$^{+0.5}_{-0.4}$ $e^2$fm$^4$ [@T01]. The calculation also reproduces well the experimental value for the energy of the $^6$Li(3$^+)\otimes\Lambda(1/2^+)$ state, 1.86 MeV (See Fig. 1). Summary ======= We have investigated the structure of $^9_\Lambda$Be and $^7_\Lambda$Li hypernuclei using two-body cluster models of $\alpha+d$ and $^5_\Lambda$He+$d$, respectively. We particularly discussed the relation between the shrinkage effect and the spectra of these hypernuclei. We first showed that the two-body cluster model yields only a small change in the spectra between $^8$Be and $^9_\Lambda$Be even though the intercluster distance is significantly reduced due to an addition of a $\Lambda$ particle. We argued based on the Bohr-Sommerfeld quantization rule that this is caused by the fact that intercluster potentials can be well fitted with a single Gaussian function, for which the spectra are approximately independent of the depth of the potential. A similar effect would be expected also for $^7_\Lambda$Li if the deuteron were a spin-less particle. In reality, a deuteron has a finite spin ($S$=1), and the spin-orbit potential is present in the intercluster potential. We have shown that the shrinkage effect leads to a large overlap between the wave function for the excited state and the spin-orbit potential for the $^7_\Lambda$Li hypernucleus as compared with $^6$Li. As a consequence, the spin-orbit potential acts effectively larger in $^7_\Lambda$Li, eventually leading to the compression of the spectrum. These behaviors of the spectra are peculiarities of a two-body cluster structure. In this sense, $^7_\Lambda$Li and $^9_\Lambda$Be are exceptional cases, as the ground states exhibit a well-developed $\alpha$ cluster structure. For heavier nuclei, the cluster structure appears in excited states while the ground state takes a mean-field-like configuration. In such situations, the shrinkage of radius would push up the spectrum, as has been seen experimentally in $^{13}_{~\Lambda}$C. The deformation degree of freedom also comes into a play there. An interesting case for a future investigation is the hypernucleus $^{19}_{~\Lambda}$F, in which the mean-field structure may be mixed with the $^{17}_{~\Lambda}$O+$d$ cluster structure. A $\gamma$-ray spectroscopy measurement has been planed on this hypernucleus at the J-PARC facility [@T11; @K08]. This will be the first for the sd-shell hypernuclei. The ground and the first excited states of the core $^{18}$F nucleus has a spin and parity of 1$^+$ and 3$^+$, respectively, which are the same for $^{6}$Li. Thus the measurement of the $^{18}$F(3$^+$)$\otimes\Lambda$(1/2$^+$) level will provide a strong stimulus to further studies on the sd-shell hypernuclei and beyond. ![(Color online) (a) The energy difference between the $L=2$ (1d) and the $L=0$ (2s) states for a Gaussian potential $V(R)=V_0\,e^{-\beta R^2}$ for a $\alpha+\alpha$ system as a function of the depth parameter $-V_0$. The solid line is obtained by numerically solving the Schrödinger equation, while the dashed line is the result of the approximate formula given in Eq. (\[rowley\]) based on the Bohr-Sommerfeld quantization rule. The dotted line denotes the energy difference for a classical rotor, evaluated with the root-mean-square (rms) intercluster distance for the 2s state shown in Fig. 8(c). (b) The energy of the $L=0$ (2s) state as a function of the depth parameter $-V_0$ for the Gaussian potential. (c) The rms intercluster distance for the 2s state. ](fig8) We thank H. Tamura for useful discussions and for his continuing encouragement. This work was supported by the Japanese Ministry of Education, Culture, Sports, Science and Technology by Grant-in-Aid for Scientific Research under the program number (C) 22540262. Rotational spectrum of a Gaussian potential =========================================== In this Appendix, we numerically check the performance of the approximate formula given in Eq. (\[rowley\]) for the rotational spectrum for a Gaussian potential, $V(R)=V_0\,e^{-\beta R^2}$. We consider here the $\alpha+\alpha$ system, and use the same geometry for the potential as that given in Eq. (\[gauss\_aa\]), [*i.e.,*]{} $\beta=0.22$ fm$^{-2}$. In this Appendix, we do not include the Coulomb interaction between the two $\alpha$ particles. Fig. 8(a) shows the energy difference between the $L=2$ state and the $L=0$ state for $2n+L=4$. Notice that without the Coulomb interaction the $L=2$ state is bound only for $V_0 < -131.5$ MeV. The solid line shows the numerical result of the Schrödinger equation, while the dashed line is obtained with Eq. (\[rowley\]). One can see that Eq. (\[rowley\]) works well, the deviation from the exact result being about $\pm$ 0.2 MeV for this system. The energy difference indeed depends on the depth parameter only weakly, while the energy and the root mean square (rms) distance between the two $\alpha$ particles for the $L=0$ state varies from $-3.269$ MeV and 3.58 fm at $V_0=-132$ MeV to $-69.538$ MeV and 1.93 fm at $V_0=-300$ MeV, respectively (see Figs. 8(b) and 8(c)). As shown by the dotted line in Fig. 8(a), this is in marked contrast with the energy difference for a classical rotor, $\Delta E = 6\hbar^2/2\mu R^2$, where $\mu=M_\alpha/2$ is the reduced mass for the $\alpha+\alpha$ system and $R$ is the rms intercluster distance. [99]{} T. Motoba, H. Bandō and K. Ikeda, Prog. Theor. Phys. [**80**]{}, 189 (1983). E. Hiyama, M. Kamimura, Y. Yamamoto, T. Motoba, and T.A. Rijken, Prog. Theo. Phys. Suppl. [**185**]{}, 106 (2010). J. Žofka, Czech. J. Phys. [**B30**]{}, 95 (1980). X.R. Zhou, H.-J. Schulze, H. Sagawa, C.X. Wu, and E.G. Zhao, Phys. Rev. C[**76**]{}, 034312 (2007). M.T. Win and K. Hagino, Phys. Rev. C[**78**]{}, 054311 (2008). H.-J. Schulze, M. Thi Win, K. Hagino and H. Sagawa, Prog. Theo. Phys. [**123**]{}, 569 (2010). M.T. Win, K. Hagino, and T. Koike, Phys. Rev. C[**83**]{}, 014301 (2011). J.M. Yao, Z.P. Li, K. Hagino, M.T. Win, Y. Zhang, and J. Meng, Nucl. Phys. [**A868-869**]{}, 12 (2011). B.-N. Lu, E.-G. Zhao, and S.-G. Zhou, Phys. Rev. C[**84**]{}, 014328 (2011). M. Isaka, M. Kimura, A. Dote, and A. Ohnishi, Phys. Rev. C[**83**]{}, 044323 (2011). D. Vretenar, W. Pöschl, G.A. Lalazissis, and P. Ring, Phys. Rev. C[**57**]{}, R1060 (1998). X.-R. Zhou, A. Polls, H.-J. Schulze, and I. Vidaña, Phys. Rev. C[**78**]{}, 054306 (2008). F. Minato, S. Chiba, and K. Hagino, Nucl. Phys. [**A831**]{}, 150 (2009); F. Minato and S. Chiba, Nucl. Phys. [**A856**]{}, 55 (2011). O. Hashimoto and H. Tamura, Prog. Part. Nucl. Phys. [**57**]{}, 564 (2006). K. Tanida [*et al.*]{}, Phys. Rev. Lett. [**86**]{}, 1982 (2001). E. Hiyama, M. Kamimura, K. Miyazaki, and T. Motoba, Phys. Rev. C[**59**]{}, 2351 (1999). H. Kohri [*et al.*]{}, Phys. Rev. C[**65**]{}, 034607 (2002). E. Hiyama, M. Kamimura, T. Motoba, T. Yamada, and Y. Yamamoto, Phys. Rev. Lett. [**85**]{}, 270 (2000). M. Ukai [*et al.*]{}, Phys. Rev. C[**73**]{}, 012501(R) (2006). H. Akikawa [*et al.*]{}, Phys. Rev. Lett. [**88**]{}, 082501 (2002). H. Tamura [*et al.*]{}, Nucl. Phys. [**A754**]{}, 58c (2005). H. Bandō and M. Bando, Phys. Lett. [**109B**]{}, 164 (1982). B. Buck, H. Friedrich, and C. Wheatley, Nucl. Phys. [**A275**]{}, 246 (1977). D.F. Jackson, M. Rhoades-Brown, Nucl. Phys. [**A286**]{}, 354 (1977). S.A. Gurvitz and G. Kalbermann, Phys. Rev. Lett. [**59**]{}, 262 (1988). S.A. Gurvitz, Phys. Rev. A[**38**]{}, 1747 (1988). K. Langanke and C. Rolfs, Phys. Rev. C[**33**]{}, 790 (1986). N. Rowley, Phys. Lett. [**69B**]{}, 25 (1977). H. Nishioka, J.A. Tostevin, and R.C. Johnson, Phys. Lett. [**124B**]{}, 17 (1983). H. Nishioka, J.A. Tostevin, R.C. Johnson, and K.-I. Kubo, Nucl. Phys. [**A415**]{}, 230 (1984). A.C. Merchant and N. Rowley, Phys. Lett. [**150B**]{}, 35 (1985). B. Buck, A.C. Merchant, and N. Rowley, Nucl. Phys. [**A327**]{}, 29 (1979). H. Tamura, [*et. al.*]{}, J-PARC 50GeV PS prposal E13 (2006), http://j-parc/NuclPart/pac-2006/pdf/p13-Tamura.pdf T. Koike, [*Proceedings of Sendai International Symposium Strangeness in Nuclear and Hadronic System (SENDAI08)*]{}, edited by K. Maeda [*et. al.*]{}, 213, (2010)

ArXiv

--- abstract: 'Superconducting loop interrupted by one or three Josephson junctions is considered in many publications as a possible quantum bit, flux qubit, which can be used for creation of quantum computer. But the assumption on superposition of two macroscopically distinct quantum states of superconducting loop contradict to the fundamental law of angular momentum conservation and the universally recognized quantum formalism. Numerous publications devoted to the flux qubit testify to an inadequate interpretation by many authors of paradoxical nature of superposition principle and the subject of quantum description.' author: - 'A.V. Nikulov' title: 'Contradiction between assumption on superposition of flux-qubit states and the law of angular momentum conservation.' --- Introduction {#introduction .unnumbered} ============ Quantum computation and quantum information is one of the most popular themes of the last decades [@book2000; @Nature01]. Many authors propose [@QubitThe01; @QubitThe02] and make [@QubitExp01; @QubitExp02; @QubitExp03; @Ilichev04; @Mooij2003] quantum bits, main element of quantum computer, on base of different two-states quantum systems including superconducting one. The employees of D-Wave Systems Inc. claimed already that they have made the world’s first commercially viable quantum computer [@D-Wave]. Main aim of this paper is to show that the assumptions by numerous authors on macroscopic quantum tunneling [@Tunnel1; @Tunnel2; @Tunnel3] and on superposition of two macroscopically distinct quantum states [@Leggett85; @Leggett02; @Clarke08] of superconducting loop interrupted by Josephson junctions contradict to the fundamental law of angular momentum conservation and the universally recognized quantum formalism. It is important for the problem of practical realisation of the idea of quantum computation since many authors consider such loop as possible quantum bit, flux qubit. Quantum bit is a quantum system with two permitted states, superposition of which is possible. Without superposition of states a quantum system with two permitted states is ordinary but no quantum bit. Therefore the contradiction between the assumption on superposition and fundamental physical laws casts doubt on numerous publication about “flux qubit”. These works may be unavailing. The contradiction between the assumption on “flux qubit” and the law of angular momentum conservation is obvious. It does not mean that this conservation law can be violated. No experimental result obtained for the present can give evidence of superposition of macroscopically distinct quantum states [@QI2005]. Many authors may interpret some experimental results as such evidence because of no enough profound understanding of paradoxical nature of the quantum principle of superposition [@QI2007]. Therefore before to consider the concrete problem of “flux qubit” I will touch “philosophical” problems of quantum foundation and the essence of controversy between creators of quantum theory about the subject of quantum mechanics description. What is subject of quantum mechanics description? ================================================= For centuries science had viewed its aim as the discovery of the real. Scientists believed that they investigate an objective reality as it exists irrespective of any act of observation. But on the atomic level physicists have come into collision with paradoxical phenomena which can not be described up to now as a manifestation of an objective reality. Therefore some creators of quantum theory, Heisenberg, Bohr and others were force to advocate positivism, the point of view according to which the aim of science is investigation no objective reality but only phenomena [@FPP2008]. Other creators of quantum theory, Plank, Einstein, de Broglie, Schrodinger could not agree with this change of science aim. But no realistic description of quantum phenomena could be created. Therefore the quantum mechanics created at the cost of refusal of objective reality description dominates more than eighty years. It is important to understand that quantum mechanics, in contrast to other theories of physics, does not describe a reality. The basic principle of the idea of quantum computation was introduced in 1935 by opponents of the Copenhagen interpretation, Einstein and Schrodinger, who persisted in their opinion that the quantum theory which can describe only phenomena can not be considered as complete. Both Einstein, Podolsky, Rosen [@EPR1935] and Schrodinger where sure that this principle, entanglement or Einstein - Podolsky - Rosen correlation, can not be real because of its contradiction with locality principle. Therefore Einstein, Podolsky, Rosen stated that quantum- mechanical description of physical reality can not be considered complete [@EPR1935] and Schrodinger introduced [@Schrod35] this principle as “entanglement of our knowledge” [@Entangl]. A “philosophical” question: “Could a real equipment be made on base of the principle which can not be describe reality?” forces to consider the essence of entanglement and history of its emergence. Two main paradoxes of quantum phenomena. ---------------------------------------- Two features of quantum phenomena are most paradoxical. The both were introduced into the consideration by Einstein, the principal opponent of the Copenhagen formalism as a complete theory. ### Wave-particle duality Bohr wrote in 1949 [@Bohr1949]: [*“With unfailing intuition Einstein thus was led step by step to the conclusion that any radiation process involves the emission or absorption of individual light quanta or ”photons" with energy and momentum*]{} $$E = h\nu ; \ \ \ p = h\sigma \eqno{(1)}$$ [*respectively, where $h$ is Planck’s constant, while $\nu$ and $\sigma$ are the number of vibrations per unit time and the number of waves per unit length, respectively. Notwithstanding its fertility, the idea of the photon implied a quite unforeseen dilemma, since any simple corpuscular picture of radiation would obviously be irreconcilable with interference effects, which present so essential an aspect of radiative phenomena, and which can be described only in terms of a wave picture. The acuteness of the dilemma is stressed by the fact that the interference effects offer our only means of defining the concepts of frequency and wavelength entering into the very expressions for the energy and momentum of the photon".*]{} ### Indeterminism Further Bohr wrote in [@Bohr1949]: [*“In this situation, there could be no question of attempting a causal analysis of radiative phenomena, but only, by a combined use of the contrasting pictures, to estimate probabilities for the occurrence of the individual radiation processes”.*]{} Before [@Bohr1949] in 1924 [@Bohr1924] Bohr noted that Einstein was first who considered the individual radiation processes as spontaneous, i.e. causeless phenomenon. Superposition of states as a description method of duality and of causeless phenomena. -------------------------------------------------------------------------------------- Advocates of the Copenhagen interpretation believe that the principle of superposition can completely describe paradoxical nature of wave-particle duality and causeless phenomena. ### Double-slit interference experiment. Indeed, it seems that this principle can perfectly describe the duality observed in the double-slit interference experiment. If a particle with a momentum $p$ and an energy $E$ passes the double-slit as a wave $\Psi = A \exp\frac{i(pr-Et)}{\hbar}$ describing an amplitude probability then the probability $$P(y) = |\Psi|^{2} = |\Psi_{1}+\Psi_{2} |^{2} = A_{1}^{2} + A_{2}^{2} + 2A_{1}A_{2}\cos(\frac{pdy}{L\hbar}) \eqno{(2)}$$ to observe the arrival of the particle at a point $y$ of a detecting screen placed on a distance $L$ from two slits separated of a distance $d$ is determined by the superposition $\Psi_{1} + \Psi_{2}$ of possibilities to pass through first $\Psi_{1} = A_{1} \exp\frac{i(pr_{1}-Et)}{\hbar}$ or second $\Psi_{2} = A_{2} \exp\frac{i(pr_{2}-Et)}{\hbar}$ slit. In accordance with this prediction all experiments give the interference patter corresponding to the momentum $p = mv$ of particles, electrons [@electro] with the mass $m \approx 9 \ 10^{-31} \ kg$, neutrons [@neutron] with $m \approx 1.7 \ 10^{-27} \ kg$, atoms [@atom] with $m \approx 3.8 \ 10^{-26} \ kg$ and even massive molecules [@mol2003; @mol2007], for example $C_{30}H_{12}F_{30}N_{2}O_{4}$ with $m \approx 1.7 \ 10^{-24} \ kg$ and a size $a \approx 3.2 \ nm$. The interference patter appears just as a probability when particles pass one by one through the two-slit system [@electro]. ### Probability of what? One may say that the wave-particle duality is observed in the double-slit interference experiment. Electron, for example, in the experiment [@electro] should pass the double-slit as a wave with the de Broglie wavelength $\lambda = 1/\sigma = h/p = h/mv$ in order the interference patter of electrons distribution with a period $\Delta y \approx \lambda L/d $ can emerge at the detecting screen. But each electron is detected as particle at a point of the detecting screen. What is the essence of the de Broglie-Shcrodinger wave function $\Psi$ in this case? According to the orthodox interpretation proposed by Born $|\Psi(r,t)|^{2}$ is a probability density. But probability of what? There is possible a realistic or positivism interpretation. According to the first one $|\Psi(r,t)|^{2}dV$ is a probability that the particle is in a vicinity $dV$ of $r$ at a time $t$. According to positivism point of view such statement has no sense since quantum mechanics can describe only results of observations. The interference observations [@mol2007] of molecules with the size $a = 3.2 \ nm$ exceeding much its de Broglie wavelength $\lambda = h/mv \approx 0.004 \ nm$ corroborate this point of view. It is impossible to localize the molecule with the size $a \approx 3.2 \ nm$ in a volume with a size $\approx 0.1 \ nm$. We must agree with the positivism point of view that the principle of superposition can describe only results of observations and nothing besides. Therefore it is important that we have not the ghost of a chance to observe the quantum interference of a particle larger $\approx 1 \ \mu m$ [@FPP2008]. ### Radioactive decay of atom as classical example of causeless phenomena. Bohr wrote in [@Bohr1949] that “[*in his famous article on radiative equilibrium*]{}” published in 1917 [@Einstein1917] “[*Einstein emphasized the fundamental character of the statistical description in a most suggestive way by drawing attention to the analogy between the assumptions regarding the occurrence of the spontaneous radiative transitions and the well-known laws governing transformations of radioactive substances*]{}”. Further Bohr quotes in [@Bohr1949] an opinion by Einstein about his theory of radiative equilibrium written at the end of the article [@Einstein1917]: “[*The weakness of the theory lies in the fact that, on the one hand, no closer connection with the wave concepts is obtainable and that, on the other hand, it leaves to chance (Zufall) the time and the direction of the elementary processes*]{}”. Thus, radioactive decay of atom may be considered as classical example of causeless phenomenon the time of which is left to chance. By 1928, George Gamow had solved the theory of the alpha decay via quantum tunneling. Following Gamow, as it was made by Einstein in [@Einstein1949], one can describe of uncertain state of radioactive atom with help of a superposition $$\Psi_{atom} = \alpha \Psi_{decay} + \beta \Psi_{no} \eqno{(3)}$$ of decayed $\Psi_{decay}$ and not decayed $\Psi_{no}$ atom. Who or what can a choice make? ------------------------------ According to the positivism point of view of Heisenberg and Bohr the description of the double-slit interference experiment (2) and the radioactive decay (3) with help of the $\Psi$-function is complete. But one can agree with this point of view only if to avoid questions: “How can a particle make its way through two slits at the same time?” and “Who or what can choose result of observation?” Concerning the first question Heisenberg wrote in [@Heisenberg1958] [*“A real difficulty in the understanding of the Copenhagen interpretation arises, however, when one asks the famous question: But what happens ’really’ in an atomic event?”*]{} The creators of the Copenhagen interpretation refused to answer on such question. Concerning the second question there was no agreement between they. Bohr wrote in [@Bohr1949] that at the Solvay meeting 1928 “[*an interesting discussion arose also about how to speak of the appearance of phenomena for which only predictions of statistical character can be made. The question was whether, as to the occurrence of individual effects, we should adopt a terminology proposed by Dirac, that we were concerned with a choice on the part of “nature” or, as suggested by Heisenberg, we should say that we have to do with a choice on the part of the “observer” constructing the measuring instruments and reading their recording. Any such terminology would, however, appear dubious since, on the one hand, it is hardly reasonable to endow nature with volition in the ordinary sense, while, on the other hand, it is certainly not possible for the observer to influence the events which may appear under the conditions he has arranged*]{}”. ### Collapse of wave function. Von Neumann’s projection postulate The orthodox interpretation studied during last eighty years substitutes the answer on the second question with words on collapse of the wave function or a ’quantum jump’ (according Heisenberg [@Heisenberg1958]) at observation. The necessity of the collapse postulated by von Neumann in 1932 [@Neumann1932] reveals the incompleteness of the Copenhagen formalism even according to the positivism point of view. The problems of wave-particle duality and indeterminism were not solved but only taken away outside the theory. The two well known paradoxes, introducing entanglement, have demonstrated clearly this incompleteness. ### Entanglement of two particles states in the EPR paradox demonstrates incompleteness of quantum - mechanical description of physical reality. A. Einstein, B. Podolsky and N. Rosen demonstrated in [@EPR1935] paradoxical nature of the superposition collapse using the law of conservation. In the Bohm’s version [@Bohm1951] of the EPR paradox the spin states of two particles are entangled $$\Psi_{EPR} = \alpha \Psi_{\uparrow }(r_{A})\Psi_{\downarrow }(r_{B}) + \beta \Psi_{\downarrow }(r_{A})\Psi_{\uparrow }(r_{B}) \eqno{(4)}$$ because of the law of angular momentum conservation. Any measurement of spin projection must give opposite results independently of the distance between the particles $r_{A} - r_{B}$ since any other result means violation of this fundamental law. The description of this correlation with help of superposition and its collapse $$\Psi_{EPR} = \Psi_{\uparrow }(r_{A})\Psi_{\downarrow }(r_{B}) \eqno{(5)}$$ implies that a measurement of the particle $A$ can instantly change a state of the particle $B$. This means the observation of real non-locality if superposition (4) is interpreted as description of a reality. Thus, the EPR paradox has prove unambiguously that quantum-mechanical description of physical reality can be considered complete only if non-local interaction is possible in this reality. ### Entanglement of atom and cat states by Schrodinger emphasizes incompleteness of causeless phenomena description. In order to make obvious the incompleteness of causeless phenomena description with help of superposition (3) Schrodinger [@Schrod35] has entangled the states of radioactive atom and a cat with unambiguous cause - effect connection $$\Psi_{cat} = \alpha \Psi_{decay}G_{yes}Fl_{yes}Cat_{dead} + \beta \Psi_{no}G_{no}Fl_{no}Cat_{alive} \eqno{(6)}$$ If the atom decays $\Psi_{decay}$ then the Geiger counter tube $G_{yes}$ discharges and through a relay releases a hammer which shatter a small flask of hydrocyanic acid $Fl_{yes}$. The hydrocyanic acid should kill the cat $Cat_{dead}$. In the opposite case $\Psi_{no}$ the cat should still live $Cat_{alive}$. It is impossible logically to see that the cat is dead $Cat_{dead}$ and alive $Cat_{dead}$ at the same time. When anyone will look on the cat he should see dead $$\Psi_{cat} = \Psi_{decay}G_{yes}Fl_{yes}Cat_{dead} \eqno{(6a)}$$ or alive cat $$\Psi_{cat} = \Psi_{no}G_{no}Fl_{no}Cat_{alive} \eqno{(6b)}$$ The question: “Who or what can choose the cat’s fate?” reveals that even causeless phenomenon must have a cause in its complete description. We must choose between [*nature*]{} as proposed by Dirac or [*the observer*]{} as suggested by Heisenberg. In the first case the description with help of superposition (6) is obviously incomplete. A natural cause because of which the atom could decay is absent the left of $\Psi_{decay}$ and $\Psi_{no}$ in (6). The suggestion of Heisenberg results to the conclusion that no reality can exist without an observer. Can an experimental result be considered as a challenge to macroscopic realism? =============================================================================== Heisenberg upheld just this absence of quantum objective reality [@Heisenberg1958]: [*“In classical physics science started from the belief - or should one say from the illusion? - that we could describe the world or at least parts of the world without any reference to ourselves”*]{}. How can one make a real equipment, which should operate without ourselves, using the quantum description, which has no sense without any reference to ourselves? Heisenberg stated in [@Heisenberg1958] that [*“there is no way of describing what happens between two consecutive observations”*]{} and [*“that the concept of the probability function does not allow a description of what happens between two observations”*]{}. According to this point of view quantum mechanics can not describe the process of quantum computation which should be between observations. Two different “Fathers” of quantum computing -------------------------------------------- Thus, according to the point of view not only opponents, Einstein and Schrodinger, but also the creator of the Copenhagen formalism we have no description of the quantum computation process. Then why could this idea become so popular? The numerous publications about quantum computer result from the ideas of David Deutsch and Richard Feynman [@DiVincenzo]. But it is important to note that Deutsch and Feynman have pointed different ways towards quantum computer. Deutsch invented the idea of the quantum computer in the 1970s as a way to experimentally test the “Many Universes Theory” of quantum physics - the idea that when a particle changes, it changes into all possible forms, across multiple universes [@Father]. This theory is one of the realistic interpretations [@Everett] of quantum mechanics which allows to interpreted most paradoxical quantum phenomena as manifestation of real processes. But this processes should occur across multiple universes [@DeutschFR]. According to Deutsch, [*“quantum superposition is, in Many Universes terms, when an object is doing different things in different universes”*]{} [@Father]. The Many Universes interpretation allows to understand why quantum computer may excel the classical one. It can do [*“a number of computations simultaneously in different universes”*]{} [@Father]. But the idea of many Universes seems mad for most physicists. Therefore most authors follow to Richard Feynman who based the idea of quantum computing on the Copenhagen interpretation. They, as well as Feynman, have an illusion, in spite of opinion of the creators, that the probability function allows a description of what happens between observations. Moreover most modern physicists are sure that quantum mechanics is an universal theory of reality from elementary particles to superconductivity. What is the essence of Bell’s inequality violation in? ------------------------------------------------------ Einstein foresaw possibility of such mass illusion. He wrote to Schrodinger in 1928 [@Lett1928]: [*“The soothing philosophy-or religion?-of Heisenberg-Bohr is so cleverly concocted that it offers the believers a soft resting pillow from which they are not easily chased away”*]{}. Many modern authors are sure that the experimental evidence [@EPRexp] of violation of the Bell’s inequality proves only that Einstein was not right, quantum mechanics is complete theory and we can continue to slip on the soft resting pillow proposed by Heisenberg and Bohr. But some experts understand that the experiments [@EPRexp] rather cast doubt on very existence of physical reality. The violation of the Bell’s inequalities is sole experimental evidence of EPR correlation (entanglement) observation. In order to quantum computer could be a real equipment the entanglement must exist, but not only to be observed. But the entanglement, because of its very nature, contradict to realism, at the least local one and of single Universe. Doubtfulness of numerous publication about superposition and entangled states of superconductor structures. ----------------------------------------------------------------------------------------------------------- The absence of comprehension of these internal conflicts of the idea of quantum computer results to illusion concerning possibility to make quantum bit. Many authors are sure that it is possible not only to make qubits [@Mooij2003; @Leggett02; @Clarke08] but even to entangle their states [@Ilichev04]. Modern physicists have already got accustomed to the principle of superposition in the course of eighty years history of quantum mechanics in its Copenhagen interpretation. Therefore the contradiction of the assumption on superposition of macroscopically distinct quantum states with macroscopic realism [@Leggett85] can not trouble most of they. Many authors interpret thoughtlessly some experimental results obtained at measurements of the superconducting loop interrupted by Josephson junctions as evidence of macroscopic quantum tunneling [@Tunnel1; @Tunnel2; @Tunnel3] and superposition of states [@Mooij2003]. But this interpretation contradicts not only to macroscopic realism but also to fundamental law of angular momentum conservation. ### Superposition of quantum states with macroscopically different angular momentum is quite impossible according to the universally recognized quantum formalism. Superposition and quantum tunneling are assumed between two permitted states $n$ and $n+1$ with equal energy but macroscopically different angular momentum. The angular momentum $M_{p} = (2m_{e}/e)I_{p}S$ is connected with the persistent current circulating in the loop clockwise in the $n$ permitted state and anticlockwise in the $n+1$ one [@Clarke08]. At the values $I_{p} \approx 5 \ 10^{-7} \ A$ and loop area $S \approx 10^{-12} \ m^{2}$ of a typical “flux qubit” [@Mooij2003] the angular momentum equals approximately $M_{p,n} \approx 0.5 \ 10^{5} \ \hbar $ and $M_{p,n+1} \approx -0.5 \ 10^{5} \ \hbar $ in the $n$ and $n+1$ state. At any transition between this states the angular momentum should change on the macroscopic value $M_{p,n} - M_{p,n+1} \approx 10^{5} \ \hbar $. In spite of the obvious contradiction to the law of angular momentum conservation authors of many publications assume that this transition can be causeless, i.e. takes place through superposition of states or quantum tunneling. Such assumption can not be correct according to the universally recognized quantum formalism. ### Possible assumption about an EPR pair of macroscopic systems. It may be that the authors of publications about “flux qubit” assume that superposition and quantum tunneling is possible thanks to a firm coupling with a large solid matrix that absorbs the change in the angular momentum, as it was made in [@Chudnov]. Such fantastic assumption means that states of superconducting condensate are entangled (like in the relation (4)) with a large solid matrix, i.e. the loop, substrate and so forth, of uncertainly large mass. It is impossible to take seriously such fantasy about macroscopic EPR pair. Acknowledgement {#acknowledgement .unnumbered} =============== This work has been supported by a grant “Possible applications of new mesoscopic quantum effects for making of element basis of quantum computer, nanoelectronics and micro-system technic” of the Fundamental Research Program of ITCS department of RAS and the Russian Foundation of Basic Research grant 08-02-99042-r-ofi. [99]{} M.A. Nielsen and I.L. Chuang, *Quantum Computation and Quantum Information* (Cambridge University Press, 2000). E. Hand, Nature [**462**]{}, 376 (2009) D.V. Averin, Sol. State Com. [**105**]{}, 664, (1998). Y. Makhlin, G. Schoen, A. Shnirman, Nature [**398**]{}, 305 (1999) F. Mallet et al., Nature Physics [**5**]{}, 791 (2009) M. Ansmann at al., Nature [**461**]{}, 504 (2009) M. Neeley et al., Science [**325**]{}, 722 (2009). A. Izmalkov et al., M. Phys. Rev. Lett. [**93**]{}, 037003 (2004) I. Chiorescu, Y. Nakamura, C. J. P. M. Harmans, J. E. Mooij, Science [**299**]{}, 1869 (2003). Wim van Dam, Nature Physics [**3**]{}, 220 (2007). R. Rouse, S. Han, and J. E. Lukens, Phys. Rev. Lett. [**75**]{}, 1614 (1995). Shao-Xiong Li et al., Phys. Rev. Lett. [**89**]{}, 098301 (2002). F. Balestro, J. Claudon, J. P. Pekola, and O. Buisson, Phys. Rev. Lett. [**91**]{}, 158301 (2003). A. J. Leggett, A. Garg, Phys. Rev. Lett. [**54**]{}, 857 (1985). A.J. Leggett, Science [**296**]{}, 861 (2002). J. Clarke, F.K. Wilhelm, Nature [**453**]{}, 1031 (2008). V. V. Aristov and A. V. Nikulov, in [*Quantum Informatics 2004*]{}, ed. by Yuri I. Ozhigov; Proceedings of SPIE, vol. 5833, (The International Society for Optical Engineering, Washington, 2005) p. 145; Preprint at http://arxiv.org/abs/cond-mat/0412573. Cited 21 Dec 2004 V.V. Aristov, A.V. Nikulov, in [*Quantum Informatics 2007*]{}, ed. by Yuri I. Ozhigov; Proceedings SPIE Vol. 7023, (The International Society for Optical Engineering, Washington, 2008). A. V. Nikulov, in [*Foundations of Probability and Physics-5*]{}, ed. by Luigi Accardi et al., AIP Conference Proceedings, Vol. 1101 (American Institute of Physics, Melville, New York, 2009) p. 134; Preprint at http://arxiv.org/abs/0812.4118. Cited 22 Dec 2008 A. Einstein, B. Podolsky and Rosen, Phys. Rev. [**47**]{}, 777 (1935). E. Schrodinger, Naturwissenschaften [**23**]{}, 844 (1935); Proc. Cambridge Phil. Soc. [**31**]{}, 555 (1935). C. Brukner, M. Zukowski, A. Zeilinger, The essence of entanglement, http://arxiv.org/abs/quant-ph/0106119. Cited 20 Jun 2001 N. Bohr, in [*"Albert Einstein philosopher - scientist*]{}, ed. by P.A. Schillp. The library of the living philosophers, v. 7 (Evanston, Illinois, 1949) p. 201. N. Bohr N., H Kramers, J. Slater, Phil. Maq. [**47**]{}, 785 (1924). A. Tonomura, J. Endo, T. Matsuda, T. Kawasaki and H. Exawa, Amer. J. Phys. **57**, 117 (1989) R. Gahler and A. Zeilinger, Amer. J. Phys. **59**, 316 (1991). O. Carnal and J. Mlynek, Phys. Rev. Lett. **66**, 2689 (1991); D.W. Keith, C.R. Eckstrom, Q.A. Turchette and D.E. Pritchard, idid **66**, 2693 (1991). L. Hackermuller, S. Uttenthaler, K. Hornberger, E. Reiger, B. Brezger, A. Zeilinger, and M. Arndt, Phys. Rev. Lett. **91**, 090408 (2003). S. Gerlich et al., Nature Physics **3**, 711 (2007). A. Einstein, Phys. Zs. [**18**]{}, 121 (1917) A. Einstein, in [*"Albert Einstein philosopher - scientist*]{}, ed. by P.A. Schillp. The library of the living philosophers, v. 7 (Evanston, Illinois, 1949) p. 665. W. Heisenberg, [*Physics and Philosophy: The Revolution in Modern Science*]{}. (Harper Perennial Modern Classics, 2007) J. von Neumann, [*Mathematical Foundations of Quantum Mechanics*]{}. (Princeton, NJ: Princeton University Press, 1955); [*Mathematishe Grundlagen der Quantem-mechanik*]{}. (Springer, Berlin, 1932). D. Bohm, [*Quantum Theory*]{}. (New York: Prentice-Hall, 1951). D.P. DiVincenzo, Fortschritte der Physik, [**48**]{}, 771 (2000). Q. Norton, The Father of Quantum Computing. http://www.wired.com/science/discoveries/news/2007/02/72734. Cited 15 Feb 2007 H. Everett, Rev. Mod. Phys. [**29**]{}, 454 (1957) . David Deutsch, [*The Fabric of Reality*]{} (The Penguin Press, 1997) A. Einstein, letter to E. Schrodinger, 31 May 1928, reprinted in [*Letters on Wave Mechanics*]{}, ed. M. Klein (New York: Philosophical Library, 1967). A. Aspect, J. Dalibard, G. Roger, Phys. Rev. Lett. [**49**]{}, 1804 (1982); G. Weihs, T. Jennewein, C. Simon, H. Weinfurter, A.Zeilinger, Phys. Rev. Lett. [**81**]{}, 5039 (1998). E. M. Chudnovsky, Phys. Rev. Lett. [**72**]{}, 3433 (1994).

ArXiv

--- abstract: '[In this paper we study the feasibility of inferring the magnetic field from polarized multi line spectra using two methods: *The pseudo line approach* and *The PCA-ZDI approach*.]{} [We use multi-line techniques, meaning that all the lines of a stellar spectrum contribute to obtain a polarization signature. The use of multiple lines dramatically increases the signal to noise ratio of these polarizations signatures. Using one technique, *the pseudo line approach*, we construct the pseudo line as the mean profile of all the individual lines. The other technique, *the PCA-ZDI approach* proposed recently by Semel et al. (2006) for the detection of polarized signals, combines Principle Components Analysis (PCA) and the Zeeman Doppler Imaging technique (ZDI). This new method has a main advantage: the polarized signature is extracted using cross correlations between the stellar spectra and functions containing the polarization properties of each line. These functions are the principal components of a database of synthetic spectra. The synthesis of the spectra of the database are obtained using the radiative transfer equations in LTE. The profiles built with the PCA-ZDI technique are denominated Multi-Zeeman-Signatures.]{} [The construction of the pseudo line as well as the Multi-Zeeman-Signatures is a powerful tool in the study of stellar and solar magnetic fields. The information of the physical parameters that governs the line formation is contained in the final polarized profiles. In particular, using inversion codes, we have shown that the magnetic field vector can be properly inferred with both approaches despite the magnetic field regime.]{}' author: - '[J.C. Ramírez Vélez ]{} , [ M. Semel ]{}, M. Stift , M.J. Martínez González , P. Petit and N. Dunstone' title: Spectropolarimetric multi line analysis of stellar magnetic fields --- Introduction ============ The majority of cool stars, including our Sun, have magnetically confined atmospheres. The study of the magnetic activity is important for understanding: 1) the dynamo effect, responsible for generating solar and stellar magnetic fields, and 2) the different states of stellar evolution and the influence of the magnetic field during these stages of evolution. The development of the Zeeman Doppler Imaging technique, described in a series of five papers, is a benchmark in the stellar spectropolarimetry domain and, consequently, in the study of magnetic stars by observational methods. It opened a new window in the research of magnetic activity in cool stars through the measurement of the circular polarization directly from the spectra. The circular and also the linear polarized signals are detectable through a combination of the Doppler and Zeeman effects [@zdi_1]. Few years after the development of the ZDI technique, Semel & Li (1996) proposed the line addition principle to improve the signal-to-noise ratio and make possible the detection of extremely faint polarization signals in stars. Since then, the addition principle has been used in different and more sophisticated techniques, and it has been particularly successful with the Least Squares Deconvolution (LSD) technique [@lsd], to construct mean intensity and/or polarization profiles. These mean profiles serves in a secondary step as input for codes that produce magnetic stellar surface maps. Observations at different phases of star’s rotational period are required to produce these maps [@zdi_4; @zdi_5; @zdi_codes]. The simplest way to employ the line addition principle is to average the spectral lines. However, this brut line addition technique is not currently used to retrieve the magnetic field vector, not even for solar observations, where the inversion of the Stokes parameters in individual spectral lines is a commonly employed technique. Semel et al. (2009) have recently shown that the signal obtained from the average of multiple spectral lines, the so-called pseudo line, can be employed to estimate the longitudinal magnetic field using the center of gravity method (e.g. Rees et al. 1979). In this work we present, in Sect. 2, more accurate inversion methods applied to the multi line approach by means of the construction and inversion of the pseudo line, and we extend the study to the linear states of polarization since considering the Stokes parameters (Q,U) is necessary to fully determine the orientation of the magnetic field. The second part of this work, Sect. 3, is dedicated to another approach of detection of magnetic fields in stars, where all the lines contained in a given spectral range contribute to the final Stokes profiles. We first present the basis of the so-called PCA-ZDI technique. We then describe the procedure to obtain the final Stokes profiles named Multi-Zeeman-Signatures. We then apply the developed technique to observed spectra of three cool stars. Finally, using synthetic spectra, we will show that the stellar magnetic fields can be correctly retrieved performing direct inversions of the Multi-Zeeman-Signature profiles. The general conclusions are presented in Sect. 4. The pseudo line approach ======================== In this Sect. we concentrate on the line addition technique using the simplest way of combine multiple spectral lines that is averaging them. We have named the *pseudo-line* ($PL$) to the resulting mean signal because it does not come from any physical entity, i.e. it has not associated any intrinsic atomic property like Lande factor, potentials of excitation, etcetera. The Stokes vector of the pseudo line , $\vec{S}_{PL}=(I,Q,U,V)$, is obtained averaging by separate in each Stokes parameter, $$%\vec{S}_{PL}=\ \sum_{i=1}^{n_l} \left( \frac{I_i(X)}{n_l}, \ \ % \frac{Q_i(X)}{n_l}, \ \ \frac{U_i(X)}{n_l}, \ \ \frac{V_i(X)}{n_l} \right), \vec{S}_{PL}=\ \sum_{i=1}^{n_l} \frac{ \left[ I_i(\lambda_R), \ \ Q_i(\lambda_R), \ \ U_i(\lambda_R), \ \ V_i(\lambda_R) \right] } {n_l}, \label{equ:ps}$$ where $\lambda_R$ = $\lambda$ - $\lambda_c$ denotes the reduced wavelength and ${n_l}$ is the number of spectral lines. In table \[tab:tab1\] are listed the individual lines that we have included in this work for the construction of the pseudo line. They have be chosen to be the same lines as in our previous work Semel et al. (2009), hereafter referred as Paper I. Considering the solar case of a single point on the stellar surface, we used the code [diagonal]{} (López Ariste & Semel 1999) to compute the Stokes profiles of the lines listed in table \[tab:tab1\], and consequently, to obtain the profiles of the pseudo line. As mentioned in Paper I, we will assume that there is only one magnetic field vector on the stellar surface responsible of the polarized signals and that the total radiation comes from this magnetic. We do not ignore that more sophisticated scenarios can be considered, but we are not interested in simulate a real stellar spectra but in show that the capabilities of the pseudo line to retrieve the magnetic field vector. In Fig. \[fig:b\_strong\], we show two examples of the pseudo profiles computed using the same given atmospheric model (see Sect. 2.3), but using different magnetic strength field. In the left panel, considering a field strength of 400G, the profiles of all the individual lines are shape similarly, and consequently, the pseudo profiles keep the same shape. In the right panels, considering a field strength of 4kG, the similarity in the shape of the individuals profiles is not preserved. Note for instance the variety of shape in the I and Q profiles. The relatively reduced number of individual lines used to construct the pseudo line is a first step for a future generalization where much more lines can be included. Before to continue with the inspection of the pseudo line we present the procedure that we follow to express the spectral lines in function of the Doppler coordinates. -------- ------------- ------- ------- ----------- -------- -- Line $\lambda_c$ Upper Lower Relative Lande Number (Å) level level intensity factor 1 6141.73 5D2 5P3 1 1.81 2 6232.66 5D1 5P2 2.25 1.99 3 6246.33 5D3 5P3 7 1.58 4 6301.51 5D3 5P2 8.75 1.66 5 6302.50 5D0 5P1 3 2.48 6 6336.83 5D1 5P1 6.75 2.00 7 6400.01 5D4 5P3 27 1.27 8 6408.03 5D2 5P1 5.25 1.01 9 6411.65 5D3 5P2 14 1.18 -------- ------------- ------- ------- ----------- -------- -- : Central wavelength, transition levels, relative intensities and Lande factors of the spectral lines of the multiplet 816 of the neutral iron. The relative intensities are taken from Allen (2000). []{data-label="tab:tab1"} Variable transformation and adequate spectral resolution -------------------------------------------------------- In this Sect. we describe a variable transformation from the wavelength coordinate into the velocity one: $\vec S(\lambda) \rightarrow \vec S(X)$. At present, the typical spectrographes used for stellar spectropolarimetry are cross-dispersed type and cover a range of thousands of Å. We will employ in our analysis a discrete selection of lines, but we present the general procedure as if we were dealing with the complete spectral range. Consider the transformation equation $$X=c \ log(\lambda / \lambda_0) %\ \ \Rightarrow \ \ %\lambda = \lambda_0 exp(X/C) \label{ecu:cv1}$$ where X is expressed in $km/s$, and $\lambda$ in Å. $c$ is the speed of light and $\lambda_0$ the minimum wavelength value in a given spectral interval. Deriving the precedent Eq., $$dX = c \frac{d\lambda }{\lambda},$$ is possible to obtain a constant step in $km/s$ along the spectra when the velocity values are discretized: $dX \rightarrow \Delta X$. We have fixed the constant step in $\Delta X$ to 1$km/s$, such that we can invert the equation (\[ecu:cv1\]) to express $\lambda$ as function of X as: $$\lambda=\lambda_0 exp(X/c) \label{ecu:cv2}$$ where X=\[0,1,2,...\]. Let $\lambda_{obs}$ denote the discrete wavelength values obtained in the spectra after the data reduction. By imposing a constant step in $\Delta X$ one problem arises for certain values because the relation $\lambda (X)$ given by Eq. (\[ecu:cv2\]) will not necessarily coincide with the values of $\lambda_{obs}$. An interpolation is then applied to determine the exact value at the wavelength of interest. Let $\lambda_n$ be one of the inexact values such that $\lambda_{obs,i}$ $<$ $\lambda_n$ $<$ $\lambda_{obs,i+1}$, and let $S(X_n)$ denote any of the Stokes parameters at wavelength $\lambda_n$. The value $S(X_n)$ will then be calculated by: $$S(X_n) = S(\lambda_i) + \frac{S(\lambda_{i+1})-S(\lambda_i)} {\lambda_{i+1}- \lambda_i } (\lambda_n-\lambda_i ) \label{equ:intrapola}$$ With this algorithm, the spectra $S(\lambda)$, observed or synthetic, will be transformed to as function of the new variable $S(X)$. The atmospheric model --------------------- In what follows of this Sect. we explore the use of accurate inversions methods to apply to the pseudo line, i.e. inversions techniques based on the radiative transfer equation (RTE). From experience in the solar data analysis, we know that some correlation between the atmospheric parameters can generate ambiguities in the determination of the magnetic field when the noise level is close to the polarized signal levels, (e.g. del Toro Iniesta & Ruiz Cobos 1996; Bellot Rubio & Collados 2003; Mart' inez González et al. 2006, and references therein). For simplicity, we initially avoid any possible model parametric degeneration by assuming that the atmospheric model is known and only the magnetic field vector is unknown. With this premise, we test the inference of the magnetic field through the inversion of the pseudo profiles under an ideal scenario where all the variables in the RTE are related to the magnetic field vector : The field strength (B), the azimuthal angle ($\gamma_{azi}$) and the inclination field ($\gamma_{incl}$). The atmospheric model we chose to compute the Stokes profiles is as follows: we fixed the gradient of the source function ($\Delta \tau = 10$), the Doppler width to 50 mÅ, the ratio of the line to continuum absorption ($\eta$) is twice the relative strength of the components of the iron multiplet and the damping parameter value is fixed to 0.01. The four Stokes parameters for all the individual Fe lines are computed considering the same atmospheric model and the values of the magnetic field varies randomly in the following ranges : B=\[0,10\] kG, $\gamma_{azi}$=\[0,180\] degrees and $\gamma_{incl}=[0,90]$ degrees. Consider now a given combination of the magnetic field parameters, $Y^n$=($B^n$, $\gamma_{azi}^{\ n}$, $\gamma_{incl}^{\ n}$). From the synthesis of the Fe lines for this combination of parameters, we then obtain the respective pseudo line for this $Y^n$ model. Note that we are in this way *associating* a magnetic model to the pseudo line. The inversion code ------------------ The exercises consist to determine the magnetic field supposed punctual and fixed in a known position in the stellar surface. The inversions in each Fe line and in the pseudo line have been done separately and the goal is to compare the results. We have developed one inversion code per Fe line and one for the pseudo line. The code’s functionality is : First, we construct a representative synthetic database of a large number of profiles (56600). Then, given a Stokes vector to invert, we find the best-fit solution from the database using the same algorithm as described in Ram' irez Vélez et al. (2008). We inverted a set of 500 profiles for each of the lines listed in Table \[tab:tab1\]. To compute the sets of profiles to invert, we followed the same procedure as when constructing the database profiles: we have fixed the atmospheric model and we consider random magnetic field vectors. While inverting the set of Stokes profiles, we applied and compared different strategies to recover the magnetic field vector. Initially, we considered the four Stokes parameters (I,Q,U,V) to simultaneously retrieve the solution of the three magnetic parameters ($B$, $\gamma_{azi}$, $\gamma_{incl}$). We then inverted the same set of profiles using only the I and V Stokes profiles, which is the most common case for stellar observations. Finally, we performed the inversions using only the two linear Stokes parameters (Q and U). The goal of performing different inversion strategies is twofold. On one hand, we compare the degree of accuracy of the inversions using the four Stokes parameters to the most typical case in the data analysis that is when we only dispose of the intensity and circular polarization spectra. On the other hand, this procedure permits to improve the precision of the inversions. Let us call the *discerning criteria* as the use of different Stokes parameters to retrieve the different components of the magnetic field. Given that the inversion code works finding as solution the best-fit of the profiles in the databases, the employment of the *discerning criteria* has the advantage that the databases become largers and thus, the code reduce the error of the inversions. We remark however that the results of the analysis of the pseudo line do not depend in the proposed criteria described below, such that some readers may want to skip the next subsection. ### Database extension To clarify the idea of the extension of the database, let $\vec{P_n}$ denote the Stokes vector for a given set of parameters such that $\vec{P_n} = \vec{P_n} (Y^n) $ and $Y^n$=($B^n$, $\gamma_{azi}^{\ n}$, $\gamma_{incl}^{\ n}$). Let ($B^j$, $\gamma_{incl}^{\ j}$) denote the associated solution values found with the inversion performed using the I and V Stokes parameters. Now, let $\gamma_{azi}^{\ k}$ be the solution value after inversion of the Q and U Stokes parameters of the same $\vec{P_n}$. Since the values of the parameters were retrieved independently, then $j \ne k$ (in more than 90% of the cases). Finally, we use the so-called *discerned criteria* to construct the final solution with a parameters combination $Y^{sol}$=($B^j$, $\gamma_{azi}^{\ k}$, $\gamma_{incl}^{\ j}$). Since $Y^{sol}$ was not originally in the database, we are “extending” the database through the use of the discerning criteria. In Fig. \[fig:disc\], using a set of 500 profiles without any added noise, we show the inversion results following the described criteria. The graphics correspond to the inversions of the Fe I line at 6141 Å, the first line listed in Table \[tab:tab1\]. While using only the I and V Stokes profiles to perform the inversions, first column in Fig. \[fig:disc\], more refined inversions are obtained for $B$ and $\gamma_{incl}$ that when all the Stokes parameters are included in the inversions (third column). This is explained because the number of magnetic models whom parameters $B$ and $\gamma_{incl}$ are close to the original pair of value parameters ($B^{ori}$, $\gamma_{incl}^{ori}$) get increased when the $\gamma_{azi}$ is ignored. In the second column we present the results of the inversions using only the linear Stokes profiles (Q,U). In this case, also an improvement happens for the inversions of the parameter $\gamma_{azi}$ compared to the inversions of the third column when the four Stokes parameters are considered. From the comparison of the inversions, showed in the forth column, we then conclude that the discerning criteria improves the inversion results: In the case of the field strength, the dispersions in the errors diminished considerable and the maximum error value decreased from $\sim$140G (symbols in gray) to $\sim $25G (symbols in black). In the case of the geometrical parameters, the angles $\gamma_{incl}$ and $\gamma_{azi}$, the inversions also improve with the adopted criteria. Note that it is expected that the proposed discerning criteria increase the inversion precision only for those codes that employs a big set of profiles (databases) where the solution is found. Testing the inversions of the pseudo line ----------------------------------------- One advantage of the developed inversion codes is the estimation of the typical errors. Thus, in order to compare quantitatively the inversions retrieved from each line, the results are presented in terms of the absolute value of the difference between the original and the inferred parameters: $$\label{deltab} \vec{B}^{ori} - \vec{B}^{sol}=(| \Delta B |, \ | \Delta \gamma_{azi} |, \ | \Delta \gamma_{incl}|).$$ Additionally, in order to study the effect of the noise in the determination of the magnetic field, we add to the profiles three different noise levels with respective values of 1x10$^{-3}$, 1x10$^{-2}$ and 1x10$^{-1}$ of the continuum level ($I_c$). In Fig. \[fig:histos\] we show the percentage histograms of the inversion’s errors for all the Fe lines (in gray) and for the pseudo line (in black). In the case without any added noise, we can see that all the inverted parameters are well retrieved. The histograms of the errors have a sharp peak at 0 and drop quickly. Note for instance that the field strength is inferred with an error smaller than 10 G with a high probability. The small errors for all the parameters are mainly due to the finiteness of the database, and are identified as the precision errors of the database. In any case, note that the results with the pseudo line are the very same the ones obtained from the inversions of individual spectral lines. We then conclude that the pseudo line encodes the information of the magnetic field vector as the individual spectral lines do!. In the lowest noise case ($10^{-3} I_c$), all parameters are also correctly retrieved, the errors being very close to the precision of the database. However, when increasing the noise levels to $10^{-2}$ and to $10^{-1} I_c$, the error bars become more important, making the error distributions slightly wider. We remark that for these noisy cases the pseudo line is the one that gives the best results, corroborating that the addition of multiple lines increases the signal-to-noise (S/N) ratio. Inversions with an unknown atmospheric model -------------------------------------------- We continue our inspection of the pseudo line relaxing the constraint that the magnetic field is the only free variable. We have thus included some line formation parameters in the atmospheric model, to verify that in a more general case the goodness of the approach remains. Considering as free variables the Doppler width ($V_D$), the absorption line ratio ($\eta_0$), the source function gradient ($\nabla \tau$) and, the magnetic field vector, we have repeated the same exercise as before inverting a set of 500 pseudo profiles. The construction of the pseudo line follows the same principle as before: Given a combination of the atmospheric parameters (excepting $\eta_0$) those values are used to compute all the Fe lines, and subsequently, obtain the pseudo line. In the case of the $\eta_0$ parameter, a factor (denoted $\eta_f$) modifies in the same amount any of the $\eta_0$ values, preserving the relative intensities (i.e. for all Fe lines, $\eta_0$ = $\eta_0 \cdot \eta_f$). In Fig. 4, we show the obtained results. After inversion of the 500 pseudo profiles, we have found that the retrieved values in any of the parameter are correct (with of course and associated error). We then conclude that the atmospheric model and the magnetic field can both be deduced from the inversions of the pseudo line. We argue in favour of the results, that the method used to invert the pseudo line is the very same the one used for the construction. With the present results, we then conclude that the multi line approach based in the direct addition of many spectral lines, i.e. in the construction of the pseudo line, is a very useful tool in the study of solar and stellar magnetic fields. The PCA-ZDI approach ==================== Alternatively to the method presented in the last Sect., instead of make more efforts in this direction considering as many lines as possible in the pseudo line, we prefer to employ a more sophisticated approach recently proposed by Semel et al. (2006), hereafter referred as Paper II. In this Sect. we focus in this second approach named PCA-ZDI. The principal idea of the PCA-ZDI technique is to incorporate the computation of the radiative transfer equations, for instance using an atmospheric model in Local Thermodynamic Equilibrium (LTE), in combination with a powerful statistical tool, the principal components analysis (PCA), in order to retrieve reliable detections of the stellar magnetic fields. This is in fact a way to overcome the employment of the weak field approximation used originally in the formulation of the ZDI technique (Semel 1989) and preserved in the LSD technique [@lsd]. Detection of stellar magnetic fields with PCA-ZDI ------------------------------------------------- As will be shown, the PCA-ZDI technique is a powerful method for detection of magnetic fields in stars. In a summarized description of the developed technique, firstly is required to dispose of a representative database composed of stellar spectra. In fact, despite the domain of application, the construction of the database is a key step when is intended to perform analysis by principal components [@rees_pca_00; @socas_01]. In any case, once defined the database, of what ever number of spectra is constructed this one, it is decomposed in a new basis using the single value algorithm [@golub_96]. The new basis is spanned in terms of eigenvectors which present many interesting properties. For the purposes of this work, we employ the eigenvectors of the new basis as *detectors* to retrieve from the spectra the magnetic *signature* of the Stokes parameters. We have denominated these magnetic signatures as *Multi-Zeeman-Signature* (MZS). The PCA-ZDI technique will thus retrieve a not null MZS profile in the polarized Stokes parameters whenever is present a magnetic field in the stellar atmosphere and the signal to noise ratio allows it. From an observational point of view, the main constrain to retrieve the polarized Stokes parameters in individual spectral lines is the very faint levels expected in the signals Magnetic detectors ------------------ Based in the ideas proposed in the first paper of ZDI [@zdi_1], we have considered that there is only one magnetic element in the stellar surface, at the projected position ($\mu$), and we have calculated the local Stokes profiles at this position. From the results already obtained with synthetic and observed spectra, we advance that the employment of the local profiles does not limit the detection in stars with complex magnetic fields configurations -dipolar, multi-polar and/or multispot-. A more detailed inspection of this statement will be presented in a forthcoming paper. The stellar spectra in the four Stokes parameters have been established with the help of [cossam]{}[^1], described in Stift (2000) and in Wade et al. (2001). [cossam]{} is an LTE line synthesis code in polarised light that calculates the Stokes profiles over arbitrarily large wavelength intervals, both for the sun and for dipolar field geometries in magnetic stars. Direct opacity sampling is carried out over the $\sigma_{-}$, the $\sigma_{+}$, and the $\pi $ components separately of the (anomalous) Zeeman patterns of the individual lines. While covering different positions in the stellar surface and at each one considering different field strengths and orientations is possible to obtain a representative combination of synthetic spectra which will serve to construct the database. The position of the stellar element is specified through the $\mu$=cos($\theta_{LOS}$), where $\theta_{LOS}$ is the angle between the line-of-sight direction and the normal to surface element. $\mu$ varies from \[0.2, 1\] with a step size of 0.2. At each $\mu$ position, we have spanned the magnetic field strength B=\[0,250,...,3000\] G and the inclination angle $\gamma_{incl}$=\[0,15,...,90\] $^{\circ}$. For economy in CPU-time we have fixed the azimuthal angle ($\gamma_{azi}$), but note that this fact does not limit the detection in the linear Stokes parameters since applying rotations of the reference frame to the calculated spectra is possible to pass from the Stokes Q to the Stokes U, or to obtain any desired configuration in the linear parameters, (e.g. Sect. 6.4 in del Toro Iniesta 2003). Consequently is justified to consider as the general case the approach used in the $\gamma_{azi}$ parameter. In this way, the combination of parameters produces a database ($\mathcal M$) with 425 spectra, per Stokes parameter. The spectral range covered from 450 to 750 nm, in steps of 10 mÅ, producing spectra of 300,000 wavelength points in each of the Stokes parameter. In the following, we consider the Stokes spectra as vectors $\vec{S}$ of dimension \[300000\]. Finally, three different atmospheric models have been considered T=3500, 4750 and 5750 $^{\circ}$K. For each model we construct the respective database. In other words, $T$ is fixed to the value $T_0$ in each database. In all cases, we have considered the same atomic solar abundances as in Grevesse & Sauval (1998), leaving out the molecular solar lines. Let $\xi_m$ represent a given combination of the model parameters $$\xi_m = \xi_m(B,\gamma_{incl}, \gamma_{azi}, \mu, T_0)_m, \ \ \ m=[0,1,..,424] \label{equ:param}$$ and let $\vec{S}_m(\xi_m$) denotes the $mth$ combination in any of the Stokes parameters calculated with [cossam]{}. Grouping all the 425 considered spectra models, we obtain an array $\mathcal M$ of dimensions \[300000, 425\] per Stokes parameter. Applying the [*Single Value Decomposition*]{} (SVD) procedure [@golub_96], the new basis with respective set of eigenvectors {$\vec{P}$} = \[$\vec{P}_0, \vec{P}_1,..., \vec{P}_{424}$\] that span the database $\mathcal M$ is found. Consequently, any spectra $\vec{S}_m$ can be expressed as a linear combination of the eigenvectors. Considering the change of variable previously described in Sect. 2.1, the correspondent expression for the spectra $\vec{S}_m(X)$ in terms of the eigenvectors is : $$\vec{S}_{m,i} ({\xi}_m, X) = \sum_{n=0}^{424} \ \alpha_{n,i}^m \ \vec{P}_{n,i}(X) \ ; \ \ \ \ \ \ i=(I,Q,U,V), \label{equ:pca2}$$ where X expressed in km/s has replaced the wavelength value. Note that any Stokes spectra $\vec{S}_{m}$ can be already identified by an unique associated set of coefficients, {$\vec{\alpha}^m$}=\[$\alpha^m_0, \alpha^m_1,...,\alpha^m_{424}$\]. In figure \[fig:exa\_ev\], we show a small portion of the first three eigenvectors of the database constructed for T=4750 $^{\circ}$ K. Space dimension reduction ------------------------- One of the reasons why we employ PCA is because of its capability of data compression and space dimensions reduction. We mean that in Eq. (\[equ:pca2\]) the equality is reached only when $n_{max}$ = 424, but in practice $n_{max}$ could be dramatically reduced such that only the first components are considered. The consequent question is how to determine the number of components really *significants* i.e. how to determine $n_{max}$. To answer this question we have reconstructed all the profiles with 1 component, then with 2, and so on, and we compare the reconstructed spectra with the original ones. The mean percentage contribution per component is plotted in Fig. \[fig:comps\]. From Fig. \[fig:comps\], we can appreciate that rate of contribution of the components to the original spectra decreases faster than an exponential, permitting to truncate the expansion in Eq. (\[equ:pca2\]) to the first components. Given that the contribution for $n=9$ is already inferior to 0.001% in the intensity Stokes parameter and inferior to 0.05% in the polarized ones, we have decided to keep only the first ten terms in Eq. (\[equ:pca2\]). The same number of components will be used when obtaining the Multi-Zeeman-Signatures and when inverting them. Moreover, for any star with the same temperature that the one considered in the database, the same ten eigenvectors will be useful to retrieved the MZS. At this point in the development of the technique we have followed a typical approach with PCA. In the following we employ the eigenvectors as magnetic detectors such that combining them with the ZDI principles, the magnetic signature from the spectra will be extracted. The Multi-Zeeman-Signatures (MZS) --------------------------------- Another of the advantages of the employment of PCA apart from the data compression, is that it permits to retrieve a set of coefficients {$ \vec{\alpha}^m$ } associated univocally to each one of the spectra $\vec{S}_{m}$. We will use then these coefficients to establish a relation to obtain the Multi-Zeeman-Signatures. Given that the eigenvectors have the property of being orthonormals, from Eq. (\[equ:pca2\]) it is then trivial to obtain the set of coefficients for any of the synthetic Stokes spectra as: $$\alpha_{n,i}^m= \vec{S}_{m,i}({\xi}_{m}, X) \cdot \ \vec{P}_{n,i}(X) ; \label{equ:alfas}$$ where n=\[0,1,...,424\] and i=(I,Q,U,V). The relation obtained in Eq. (\[equ:alfas\]) represents the fundamental principle that will be employed to extract the magnetic signature from the observed spectra ($\vec{S}^{obs}$). In order to do it, the last step of the procedure is to consider the rotation velocity of the star and the associated Doppler effect, i.e. combine the MZS profiles with the ZDI principles. It is important to mention that in order to equalize the spectral resolution of the observations with the one of the eigenvectors, we degraded the resolution of the eigenvectors to the spectral resolution of the observations. This last is done through a linear intrapolation similar the one described in Sect. 2.1. We proceed now to apply the same operation expressed in Eq. (\[equ:alfas\]) to the observed spectra $\vec{S}^{obs}$, but when the eigenvectors are Doppler shifted by an amount Y=$\Delta X$ : $$\alpha_{n,i}^{obs} (Y) = \vec{S}_{i}^{obs}(X) \ \cdot \ \vec{P}_{n,i}(X-Y). \label{equ:alfas2}$$ The vectorial product in the right side of in Eq. (\[equ:alfas2\]) that permits to retrieve the $ \alpha $-coefficients at each considered $Y$-value is equivalent to a cross correlation function between the spectra and the magnetic detectors (the eigenvectors). On the other hand, it is convenient to mention that the expression found in Eq. (\[equ:alfas2\]) can be replaced by another one where the eigenvectors {$\vec{P} $} are substituted by the absolute value of the same eigenvectors, denoted by { $\mid \vec{P} \mid $ } : $$\alpha_{n,i}^{obs} (Y) = \vec{S}_{i}^{obs}(X) \ \cdot \ \mid \vec{P}_{n,i} \mid (X-Y) . \label{equ:alfas_abs}$$ Considering the whole range of Doppler velocities, Y=\[$Y_0,Y_1,...,Y_{max}$\], the general expression of the Multi-Zeeman-Signature profiles ($\vec{P_{MZS}}$) is given by: $$(\vec{P_{MZS}})_{n,i}(Y)= \left( \alpha_{n,i}^{obs} (Y_{0}),\ \alpha_{n,i}^{obs} (Y_{1}),\ ...\ ,\ \alpha_{n,i}^{obs} (Y_{max}) \right) . \label{equ:Psp}$$ Note that for each Stokes parameter, (I,Q,U,V), 425 Multi-Zeeman-Signatures are retrieved. Since the eigenvectors are orthonormals, then each one of the 425 Multi-Zeeman-Signatures are in principle independents, making possible to perform independent analysis with each one of them by separate. Magnetic detections ------------------- Once presented the technique, we will now apply it to the real observed spectra. We will employ the eigenvectors obtained for the case of single points in the stellar surface, however this fact does not limit the detection of the global magnetic field. We will first compare the MZS obtained from Eq. 10 to those retrieved following Eq. 11. Let $v_{max}$ and $v_{min}$ denote the maximum and minimum projected rotation velocities of the star $vsini$, such that the star velocities span as $V^{rot}=[v_{min},...,v_{max}]$. If the considered Doppler velocity Y does not coincide with any of the $V^{rot}$ values, the intensity MZS profile recovers the continuum value, and the polarized MZS profile an aleatory sequence of values around zero. Contrary, if Y coincides with one of the $V^{rot}$ values, the magnetic signature in the polarized parameters and the total intensity profile are retrieved. In Fig. \[fig:abs\_iipeg\] we compare the first nine MZS considering the absolute value of the eigenvectors (upper panels) to the first nine MZS when the absolute value is not included (lower panels). The data correspond to the *IIpeg* star observed in August 2007 with the Bernard Lyot telescope, at the Pic du Midi Observatory, using the NARVAL polarimeter. From Fig. \[fig:abs\_iipeg\], we appreciate that when the absolute value is included the nine polarized MSZ are clearly detected with a high S/N ratio. If the absolute value is not considered the level of the first polarized MZS is clearly superior to the noise level but the level of the MZS began to decrease for the rest of eigenvectors. A discussion about how to take advantage of this multiple Multi-Zeeman-Signature detections will be presented in a forthcoming paper. In the following, we prefer to add the nine individual Multi-Zeeman-Signatures in order to obtain a total profile per Stokes parameter. In the last subsection we have showed that the expansion in Eq. (\[equ:pca2\]) can be reduced to the first ten components. We then add the contribution from these components to find the final expression that will be employed to retrieve the Multi-Zeeman-Signature: $$(\vec{P_{MZS}})_{i}(Y)= \sum_{n=0}^{9} \left( \alpha_{n,i}^{obs} (Y_{0}),\ \alpha_{n,i}^{obs} (Y_{1}),\ ...\ ,\ \alpha_{n,i}^{obs} (Y_{max}) \right). \label{equ:Psp}$$ Considering the absolute value of the eigenvectors and following the procedure of the last equation, in Fig. \[fig:detec\], we show two examples of the MZS profiles obtained in intensity and in circular polarization of two cool stars. The binary system HD 155555, left panels, was observed in April 2007 with the Anglo Australian Telescope, at the AAO observatory, using the SEMPOL polarimeter. The solar type star HD 190771, right panels, was observed in August 2007 with the Bernard Lyot telescope, at the Pic du Midi Observatory, using the NARVAL polarimeter. The detection of magnetic fields in this three stars have been also found through the LSD method. In fact, the MZS profiles from Fig. \[fig:detec\] seems similars in shape to those retrieved with the LSD method (Dunstone et al. 2008, Petit et al. 2008). Nevertheless a proper comparison has not been done to distinguish possible differences in the profiles and how important could be those differences when retrieving the magnetic field. In any case, for the moment, the PCA-ZDI technique has been applied with success to retrieve mostly the circular states of polarization in cool stars (Ram' irez Vélez et al. 2006). In fact this is in part due to the absence of representative samples of linear data to analyse, so we leave for a future work studies where the linear states of polarization can be incorporated in the detection and measurement of stellar magnetic fields. Inversions of the Multi-Zeeman-Signatures ----------------------------------------- In this section we intend to show that it is possible to infer the magnetic field through the inversion of the MZS profiles. We will place a simplistic scenario where again we consider local Stokes profiles from different surface elements. We further assume that there is a punctual magnetic field in the surface elements. From now on, the MZS to which we refer will be obtained from the local Stokes profiles through Eq. (13). Given a MZS, the inversion exercise consist in find the position of the surface element and the magnetic field vector. The local Stokes profiles are calculated for a small spectral range \[5400,5500\] Å. The parameters of the Stokes profiles, and thus those of the MZS, varies randomly in the following ranges : the magnetic field strength \[0,10\] kG, the inclination field \[0,90\]$^{\circ}$, the magnetic azimuthal angle \[0,180\]$^{\circ}$ and the projected position of the surface element ($\mu$) \[0,1\]. Since the inversions are performed in the space of the MZS we have produced 6000 MZS that serves as database where we tested the inversion of a set of 600 MZS. As mentioned previously there are two alternatives to produce MZS, namely, whether or not is considered the the absolute value of the eigenvectors (Eqs. 10 and 11). We will test both alternatives and we will show that whether or not is included the absolute value, the inversions are correct. We first present the case where the absolute value of the eigenvectors is employed to retrieve the MZS. In upper panel of Fig. \[fig:examp\] we show an example of a synthetic MZS (in black color) and the respective MZS solution (in gray color). The inversion is correct not only because the similarity between the MZSs but because the values of the parameters solution are close the originals ones (see the title in each panel). Moreover, given that any MZS is associated to a stellar spectrum, in the lower panels of Fig. \[fig:examp\] we corroborate that, as expected, the Stokes profiles associated to the MZS solution fit enough well the originals ones. In Fig. \[fig:examp2\] we show the MZS associated to the same combination of parameters as before but for the case where the absolute value is not considered. The solution parameters in this case are also close to the original ones (upper panels) and the associated solution Stokes profiles gives a good fit to the original Stokes profiles (lower panels). Finally, we present the results of the inversions of the 600 MZS. In Fig. \[fig:inv\_psp\], the upper panels correspond the case when the absolute value of the eigenvectors is included to produce the MZS and the lower panels correspond to the case when is not included. These results show that in both cases, the magnetic field vector is correctly retrieved despite the position in the stellar surface, for all strength fields and orientations. The final conclusion in this section is that the Multi-Zeeman-Signature represents not only an effective tool for the detection of stellar magnetic fields, but also that inversions in the space of the Multi-Zeeman-Signatures can be directly performed. Conclusions =========== In this article, we have focused in the multi line analysis of spectropolarimetric data, with emphasis in the detection and inference of stellar magnetic fields. The use of multiple lines is required to increase the signal to noise ratio in those cases where the polarized signals in individual spectral lines are significantly inferiors to the noise level, remaining hidden for analysis purposes. This is typically the case, for instance, in observations of cool Stars. To deal with this observational constraint, we have presented the development of two new approaches that serve both of them to extract a polarized signal from a mixture of several individual spectral lines. While employing any of the proposed multi-line techniques, namely the pseudo line or the PCA-ZDI approach, we remark three main advantages: (1) the contributions of all the spectral lines to the final polarization signal are included -despite the similarity or not in the individual shape profiles-, (2) it is a valid method for the detection of the circular and the linear states of polarization and (3) it is not limited to weak or to strong magnetic field regimes. In the case of the pseudo line, we have presented an inversion code that builds a database of pseudo lines each one attached to a particular atmospheric model. Initially, considering an ideal scenario where the magnetic field is the only free variable of the atmospheric model, we have found that the results obtained from the inversions of the pseudo line are as goods as those obtained from individual spectral lines. We have also shown that it is expected that in the case of real observed spectra (with a given noise level), the best results in any of the magnetic components are retrieved with the inversions of the pseudo line. Finally, we have showed that the atmospheric model, in addition to the magnetic field, can also be inferred through the inversions of the pseudo line. In the case of the Multi-Zeeman-Signatures, we discussed in detail the development of the technique, and we applied it to real observed data in order to illustrate the type of profile obtained with the PCA-ZDI technique. The circular polarized MZS profiles, like those in Fig. \[fig:detec\] where the polarized signal level is clearly superior to the noise level, represent in a unambiguously way a detection of the magnetic field present in the stellar object. By construction, the Multi-Zeeman-Signatures contain the information of the magnetic field and atmospheric model. To verify this, we have performed inversions in a synthetic set of MZS profiles. The results obtained after the inversions show that the all the considered parameters in the atmospheric model are correctly retrieved. It is pertinent to mention that some refinements have to be done to the technique in order to achieve magnetic fields measurements through the inversion of the MZS profiles. In particular, the Doppler broadening effect due to the rotation of the stars and the case of inversions of continuum fields distributions over the stellar surface will be presented in forthcoming papers. For the moment, we have settled the basis of the approach founding that PCA-ZDI is a robust technique for the analysis of stellar magnetic magnetic fields. The final conclusion of this work is that when the inversion of the magnetic polarized signal are done through the same method used to construct them, the physical parameters that determine the line formation process can be properly recovered, included in particular, but only, the stellar magnetic field. Allen, C.W., 2000, Astrophysical Quantities, New York : Springer Verlag and AIP press. Brown S.F., Donati J.F., Rees D.E. & Semel M., 1991, A&A, 250,463 Bellot Rubio L. R. & Collados M., 2003, A&A, 406, 357 del Toro Iniesta J.C., Introduction to spectropolarimetry, Cambridge University Press, 2003. del Toro Iniesta J.C. & Ruiz Cobos B., 1996, SoPh, 164, 169 Grevesse N. and Sauval A. J., 1998, Space Science Reviews, 85, 161 Donati J.F., & Brown S.F., 1997, A&A, 326, 1135 Donati J.F., Semel M., Carter B.D., Rees D.E. & Collier Cameron A., 1997, MNRAS, 291,658 Dunstone N. J., Hussain G. A. J., Collier Cameron A., Marsden S. C., Jardine M., Stempels H. C., Ram' irez Vélez J. C. & Donati J. -F., 2008, MNRAS, 387, 481 Golub, G. H. & Van Loan, C. F., 1996, *Matrix Computations*, 3rd ed., Johns Hopkins University Press, Baltimore. Hussain G.A.J., Donati J.F, Collier Cameron A. & Barnes J.R, 2000, MNRAS, 318, 961 L[ó]{}pez Ariste A., & Semel M. 1999, , 139, 417 Martínez González M.J., Collados M.J. & Ruiz Cobos B., 2006, A&A, 456, 1159 Petit P., Dintrans B., Solanki SK., Donati J-F, Auriere M., Lignieres F., Morin J., Paletou F., Ram' irez Vélez J.C., Catala C & Fares R., MNRAS, 388, 80 Ram' irez Vélez J.C., López Ariste A. & Semel M., A&A, 487, 731 Rees D.E., López Ariste A., Thatcher J., & Semel M., 2000, A&A, 355, 759 Rees D.E., & Semel M., 1979, A&A, 79, 1 Semel M., Ram' irez Vélez J.C., Mart' inez González M., Asensio Ramos A., Stift M.J., López Ariste A., & Leone F., 2009, A&A, 504, 1003 Semel M., Rees D., Ram' irez Vélez J.C., Stift M. & Leone F., 2006, ASPC, 358, 355 Semel M. & Li J., 1996, SoPh, 164, 417 Semel M., 1989, A&A, 225, 456 Socas Navarro H., L[ó]{}pez Ariste A. & Lites B.W., 2001, ApJ, 553,949 Wade, G. A., Bagnulo, S., Kochukhov, O. P., et al. 2001, A&A, 374, 265 [^1]: Codice per la Sintesi Spettrale nelle Atmosphere Magnetiche

ArXiv

--- abstract: 'We compute the momentum-space probability density of ${}^6$He at leading order in Halo EFT. In this framework, the ${}^6$He nucleus is treated as a three-body problem with a ${}^4$He core (${c}$) and two valence neutrons ($n$). This requires the $nn$ and $n {c}$ t-matrices as well as a ${c}nn$ force as input in the Faddeev equations. Since the $n {c}$ t-matrix corresponds to an energy-dependent potential, we consider the consequent modifications to the standard normalization and orthogonality conditions. We find that these are small for momenta within the domain of validity of Halo EFT. In this regime, the ${}^6$He probability density is regulator independent, provided the cutoff is significantly above the EFT breakdown scale.' author: - Matthias Göbel - 'Hans-Werner Hammer' - Chen Ji - 'Daniel R. Phillips' bibliography: - 'momSpaceProbDensHe.bib' date: 'April 15, 2019' title: 'Momentum-space probability density of ${}^6$He in Halo Effective Field Theory[^1]' --- Introduction {#sec:intro} ============ Probability densities contain important information on the structure of quantum mechanical objects. In chemistry, for example, the atomic orbital model provides a means to visualize the electronic cloud. In nuclear physics, it has long been held that wave functions can be measured in knock-out experiments as long as the quasielastic approximation holds (see, e.g., Refs. [@Ulrych:1998nk; @Yaron:2002nv; @Benhar:2006wy] and references therein). However, in general kinematic situations this is not true: at shorter distance scales strength can be shifted between the initial-state wave function, the current operator, and final-state interactions by means of unitary transformations or a redefinition of field variables [@Haag:1958vt; @Kamefuchi:1961sb; @Chisholm:1961tha; @Furnstahl:2001xq; @More:2017syr]. Here we consider probability densities for halo nuclei. Halo nuclei are characterized by a tightly bound core and a few loosely bound valence nucleons. They show universal properties independent of the details of their structure at short distances [@Jensen:2004zz; @Braaten:2004rn]. Due to this separation of scales in terms of a core momentum scale, $M_\mathrm{core}$, and a halo momentum scale, $M_\mathrm{halo}\ll M_\mathrm{core}$, halo nuclei can be described as effective few-body systems in an expansion in $M_\mathrm{halo}/M_\mathrm{core}$. This expansion is conveniently implemented using the framework of Halo Effective Field Theory (Halo EFT) [@bertulani02; @bedaque03; @hammer17]. Halo nuclei consisting of two valence neutrons and a core are an effective three-body problem in Halo EFT. In this case, the full wave function depends on two Jacobi momenta. As we will discuss below this leads to some complications in the definition of the probability density due to the angular-momentum recoupling that arises in transformations between different sets of Jacobi coordinates. These complications occur for any three-body bound state in which interactions are defined on a partial-wave basis. Here we focus on the momentum-space probability density for ${}^6$He, where two valence neutrons interact with each other and with an $\alpha$-particle core. $^6$He has already been discussed at leading order in Halo EFT [@Rotureau:2012yu; @ji14]. We follow the Halo EFT treatment of $^6$He by Ji [*et al.*]{} [@ji14], which includes the ${}^1S_0$ blue$nn$ interaction and the ${}^2P_{3/2}$ $n{c}$ interaction at leading order using a dibaryon formalism. The ${c}n$ t-matrix then corresponds to an energy-dependent potential. This necessitates modification of the standard normalization and orthogonality conditions, which in turn affects the expression for the momentum-space probability densities. The paper is organized as follows: In Sec. \[sec:6HeHaloEFTreview\] the Halo EFT for [^6^He]{} is reviewed. We write down the Faddeev equations for this system and compute the potentials necessary to reproduce the leading-order ${c}n$ and $nn$ amplitudes in Halo EFT. This is done using separable potentials with two different form-factor choices. The coupling strengths thereby obtained are, in general, energy dependent. The quantum mechanics of energy-dependent potentials is then discussed in Sec. \[sec:energydeppotentials\] at a generic level. As an example the momentum-space probability density a two-body bound state with an energy-dependent contact interaction is worked out. In Sec. \[sec:probdensity\] we derive the expressions for the probability density of [^6^He]{} with and without the modifications due to energy dependence of the potential. We show results for different regulator parameters and different form factors and observe that the low-momentum part of the density is independent of these choices. Furthermore, we show that the modifications of the probability density due to the energy dependence of the potentials are small for momenta in the domain of validity of the EFT. Finally, in Sec. \[sec:future\] we summarize our results and give an outlook. Halo EFT for ${}^6$He {#sec:6HeHaloEFTreview} ===================== From Halo EFT to the ${}^6$He wave function via the Faddeev equations {#subsec:heft} --------------------------------------------------------------------- Our aim is to solve the stationary three-body Schrödinger equation $H\ket{\Psi} = E_3 \ket{\Psi}$ using the Faddeev equations, with the Hamiltonian $H$ derived from the leading-order Halo EFT for [^6^He]{}. In Halo EFT—as in the cluster models that came before it, e.g., Refs. [@Hebach:1967bpg; @Shah:1970wu; @Ghovanlou:1974zza; @Chulkov:1990ac; @Zhukov:1993aw]—[^6^He]{} is a three-body problem with the neutrons $n$ and the core $c$ as degrees of freedom. Within this review we follow [@ji14], where Halo EFT for [^6^He]{} is set up, the Faddeev amplitudes are calculated and the three-body system is renormalized. A general discussion of the Faddeev equations can be found in [@gloeckle83] and a general review in the context of Halo EFT in [@hammer17]. The path from the EFT to our Hamiltonian is that the EFT’s power counting determines the t-matrices that describe two-body scattering to a given accuracy. From there we infer the potential terms $V$ of the Hamiltonian via the Lippmann-Schwinger equation. In fact, for solving the Faddeev equations explicit expressions for the t-matrices are sufficient. Expressions for the potentials are not necessary. Nevertheless, there are two reasons that make explicit formulas for the potentials desirable. The first, somewhat usual, reason is that explicit expressions for the potentials allow us to cross-check whether the found solution really solves the stationary Schrödinger equation $$\left(H_0 + \sum_i V_i+ V_3\right) \ket{\Psi} = E_3 \ket{\Psi}. \label{eq:3BSE}$$ Note that here we follow the usual spectator notation for the three-body problem, as inspired by Faddeev’s treatment [@Faddeev:1960su], and denote the the kinetic energy operator $H_0$, the two-body potential corresponding to spectator $i$ by $V_i$, and the three-body potential by $V_3$. The second reason is that when either the two-body or three-body potential in the Hamiltonian is energy dependent, the standard formula for wave function normalization needs to be modified by adding extra terms related to the potential. We start down the path to solution of Eq. (\[eq:3BSE\]) by reviewing the connection between the two-body t-matrices $t$ and potentials $V$. It is given by the Lippmann-Schwinger equation for the t-matrix $$\label{eq:ls_op} t = V + V G_0 t\,,$$ where $G_0 = {\ensuremath{\left(E - H_0\right)}}^{-1}$ is the free Green’s function. In order to distinguish the different two-body interactions we introduce spectator indices $i \in \{n, {n^\prime}, c\}$. The t-matrix $t_i$ is the one corresponding to $V_i$ via Eq. (\[eq:ls\_op\]). In the case of two-body potentials $V_i$ we are limiting ourselves to separable potentials. They have the advantage of significantly reducing the numerical effort needed to solve the Faddeev equations. In momentum space a separable potential is given by $$\label{eq:def_v_sep} \mel{{\ensuremath{\boldsymbol{p}}}}{V_i(E)}{{\ensuremath{\boldsymbol{{p^{\prime }}}}}} = {\ensuremath{\left(2l_i+1\right)}} P_{l_i}{{\ensuremath{\left({\ensuremath{\boldsymbol{p}}}\cdot {\ensuremath{\boldsymbol{{p^{\prime }}}}}\right)}}} {g_{l_i}{{\ensuremath{\left(p\right)}}}} \lambda_i{{\ensuremath{\left(E\right)}}} {g_{l_i}{{\ensuremath{\left({p^{\prime }}\right)}}}}\,.$$ Since $P_l$ denotes the $l$-th Legendre polynomial Eq. (\[eq:def\_v\_sep\]) encodes an interaction that only happens in the partial wave $l_i$. The strength of the potential is given by $\lambda_i$, which can be energy dependent. The dependence on off-shell momenta completely resides in the so-called form-factors $g{{\ensuremath{\left(p\right)}}}$. These form factors describe the shape of the potentials, their specific forms at high momenta should not affect low-energy observables. In an EFT treatment ${g_{l_i}{{\ensuremath{\left(p\right)}}}}$ can be used to regularize ultraviolet divergences in scattering equations. The separability of the potential implies also the separability of the t-matrix: $$\mel{{\ensuremath{\boldsymbol{p}}}}{t_i(E)}{{\ensuremath{\boldsymbol{{p^{\prime }}}}}} = {\ensuremath{\left(2l_i+1\right)}} P_{l_i}{{\ensuremath{\left({\ensuremath{\boldsymbol{p}}}\cdot {\ensuremath{\boldsymbol{{p^{\prime }}}}}\right)}}} {g_{l_i}{{\ensuremath{\left(p\right)}}}} \tau_i{{\ensuremath{\left(E\right)}}} {g_{l_i}{{\ensuremath{\left({p^{\prime }}\right)}}}}\,. \label{eq:sept}$$ In our EFT treatment the expressions for the t-matrices are given according to a power counting in scattering parameters and/or binding momenta. They are also defined in terms of these, effective-range expansion, parameters. $\tau_i{{\ensuremath{\left(E\right)}}}$ is therefore determined once the power counting and order of the calculation have been specified. The strength parameter $\lambda_i$ follows by partial-wave projecting and reshaping the Lippmann-Schwinger equation (see, e.g., Ref. [@Afnan:1977pi]): $$\label{eq:lb_ls} \lambda_i^{-1}{{\ensuremath{\left(E\right)}}} = \tau_i^{-1}{{\ensuremath{\left(E\right)}}} + 4\pi \int_0^\infty \frac{q^2 {\ensuremath{\left( {g_{l_i}{{\ensuremath{\left(q\right)}}}} \right)}}^2 }{ E - \frac{q^2}{2\mu_{jk}} + \mathrm{i}\epsilon} \dd{q}\,.$$ Now that we have summarized the two-body sector we can proceed to the solution of the three-body problem using the Faddeev equations. In a first step we neglect three-body forces, they will be included later. Our aim is to solve the stationary Schrödinger equation . Since the system is bound, we specify $E_3 = - B_3^{{\ensuremath{\left(0\right)}}}$, with $B_3^{{\ensuremath{\left(0\right)}}}$ the binding energy of the three-body system relative to the ${c}nn$ threshold. Following Faddeev the total state is decomposed into the Faddeev components $$\label{eq:def_fd_wfc} \ket{\psi_i} \coloneqq G_0 V_i \ket{\Psi}\,,$$ which fulfill $$\label{eq:fd_dcmp} \sum_i \ket{\psi_i} = \ket{\Psi}\,.$$ The Schrödinger equation then becomes equivalent to the set of coupled Faddeev equations: $$\label{eq:fd} \ket{\psi_i} = G_0 t_i \sum_{j \neq i} \ket{\psi_j}\,.$$ At this point it is useful to discuss the basis states in terms of which the obtained result $\ket{\Psi}$ shall be represented. In the center-of-mass frame two independent momentum variables are needed to describe the state of the three-particle system. We use standard Jacobi coordinates ${\ensuremath{\boldsymbol{p}}}$ and ${\ensuremath{\boldsymbol{q}}}$, where ${\ensuremath{\boldsymbol{p}}}$ is the relative momentum in the two-body subsystem. The momentum of the third particle relative to the center of mass of the two-body subsystem is given by ${\ensuremath{\boldsymbol{q}}}$. Depending on the choice of the third, spectator, particle three sets of Jacobi coordinates can be defined. The relations between the Jacobi momenta and the single particle momenta $k_i$ are given by $$\begin{aligned} {\ensuremath{\boldsymbol{p}}}_i &\coloneqq \mu_{jk} {\ensuremath{\left(\frac{{\ensuremath{\boldsymbol{k}}}_j}{m_j} - \frac{{\ensuremath{\boldsymbol{k}}}_k}{m_k}\right)}}\,, \\ {\ensuremath{\boldsymbol{q}}}_i &\coloneqq \mu_{i{\ensuremath{\left(jk\right)}}} {\ensuremath{\left(\frac{{\ensuremath{\boldsymbol{k}}}_i}{m_i} - \frac{{\ensuremath{\boldsymbol{k}}}_j + {\ensuremath{\boldsymbol{k}}}_k}{M_{jk}}\right)}}\,,\end{aligned}$$ where $M_{jk} \coloneqq m_j + m_k$, with reduced masses $\mu_{jk} \coloneqq m_j m_k / {\ensuremath{\left(m_j + m_k\right)}}$ and $\mu_{i{\ensuremath{\left(jk\right)}}} \coloneqq m_i M_{jk} / {\ensuremath{\left(m_i + M_{jk}\right)}}$. Different sets of Jacobi momenta with other spectators are given by cyclic permutations of $ijk$. We now discuss the description of angular momenta. The relative orbital angular momentum of the two-body subsystem is given by ${\ensuremath{\boldsymbol{l}}}$, the one of the third particle with respect to the two-body subsystem is denoted by ${\ensuremath{\boldsymbol{\lambda}}}$. The total spin of the two-body subsystem is given by ${\ensuremath{\boldsymbol{s}}}$, the spin of the third particle by ${\ensuremath{\boldsymbol{\sigma}}}$. The total angular momentum of the three-body system is ${\ensuremath{\boldsymbol{J}}}$. In order to do the angular momentum coupling one has the choice between [${\ensuremath{\boldsymbol{j}}}{\ensuremath{\boldsymbol{J}}}$-coupling]{} or [${\ensuremath{\boldsymbol{L}}}{\ensuremath{\boldsymbol{S}}}$-coupling]{}. In [${\ensuremath{\boldsymbol{j}}}{\ensuremath{\boldsymbol{J}}}$-coupling]{} the angular momenta of the subsystem are coupled to ${\ensuremath{\boldsymbol{j}}} \coloneqq {\ensuremath{\boldsymbol{l}}} + {\ensuremath{\boldsymbol{s}}}$. The angular momenta of the third particle make up ${\ensuremath{\boldsymbol{I}}} \coloneqq {\ensuremath{\boldsymbol{\lambda}}} + {\ensuremath{\boldsymbol{\sigma}}}$. Often a description of the states in the ${\ensuremath{\boldsymbol{j}}}{\ensuremath{\boldsymbol{J}}}$-coupling scheme is useful (see [@gloeckle83]): $$\begin{aligned} \label{eq:jJ_coupling} \ket{{\ensuremath{\left(l,s\right)}}j, {\ensuremath{\left(\lambda, \sigma\right)}} I;J,M} \coloneqq \sum_{\mathclap{m_j+m_I=M}} {C{{\ensuremath{\left(j m_j I m_I \,\vline \, J M\right)}}}} \ket{l,s;j, m_j} \ket{\lambda,\sigma;I, m_I \vphantom{l,j}}\,,\end{aligned}$$ where ${C{{\ensuremath{\left(l_1 m_1 l_2 m_2 \,\vline \, L M\right)}}}} \coloneqq \braket{l_1 m_1 l_2 m_2}{(l_1 l_2) L M}$ is the Clebsch-Gordan coefficient. In the case of [^6^He]{} the quantum numbers of the ground state $J^{\pi} = 0^+$. Our assumptions, that the $nn$ interaction happens in the ${}^1S_0$ channel and the $nc$ one acts in the ${}^{2}P_{3/2}$, then fix all quantum numbers in the case of [${\ensuremath{\boldsymbol{j}}}{\ensuremath{\boldsymbol{J}}}$-coupling]{}. This is summarized in the multiindices $$\begin{aligned} \ket{\Omega_c} &\coloneqq \ket{{\ensuremath{\left(0,0\right)}}0, {\ensuremath{\left(0, 0\right)}} 0;0,0}\,, \\ \ket{\Omega_n} &\coloneqq \ket{{\ensuremath{\left(1,{\frac{1}{2}}\right)}}\frac{3}{2}, {\ensuremath{\left(1, {\frac{1}{2}}\right)}} \frac{3}{2};0,0}\,,\end{aligned}$$ where the quantum numbers are written in the same order as in and the multiindex $\ket{\Omega_i}$ is implicitly defined to mean that those quantum numbers refer to the co-ordinate system in which particle $i$ is the spectator. The representations of these indices in [${\ensuremath{\boldsymbol{L}}}{\ensuremath{\boldsymbol{S}}}$-coupling]{} are given in [@ji14]. Now we are equipped to set up the Faddeev equations. We insert the identity $ {\mathbbm{1}}= \sum_\Omega {\int \dd{{p^{}}} {p^{2}}} {\int \dd{{q^{}}} {q^{2}}} \ketbra{p,q;\Omega}$ into Eq. (\[eq:fd\]) in order to obtain a representation of the Faddeev equations. For this purpose the matrix elements of the t-matrix and the Green’s function are necessary: $$\begin{aligned} { \prescript{}{i}{ \mel{p,q;\Omega}{t_i{{\ensuremath{\left(E_3\right)}}}}{{p^{\prime }},{q^{\prime }};\Omega^\prime}_{i}^{} } } &= 4\pi {g_{l_i}{{\ensuremath{\left(p\right)}}}} \tau_i{\left(E_3 - \frac{q^2}{2 \mu_{i{\ensuremath{\left(jk\right)}}}}\right)} {g_{l_i}{{\ensuremath{\left({p^{\prime }}\right)}}}} \label{eq:t_pw_me} \nonumber \\ & {\quad \cross}{\delta_{\Omega, \Omega^\prime}} {\delta_{\Omega, \Omega_i}} \frac{{\delta{{\ensuremath{\left(q-{q^{\prime }}\right)}}}}}{q^2}\,, \\ { \prescript{}{i}{ \mel{p,q;\Omega}{G_0{\ensuremath{\left(E_3\right)}}}{{p^{\prime }},{q^{\prime }};\Omega^\prime}_{i}^{} } } &= {\ensuremath{\left(E_3-\frac{p^2}{2\mu_{jk}}-\frac{q^2}{2\mu_{i{\ensuremath{\left(jk\right)}}}} \right)}}^{-1} \nonumber \\ & {\quad \cross}{\delta_{\Omega, \Omega_i}} \frac{{\delta{{\ensuremath{\left(p-{p^{\prime }}\right)}}}}}{p^2} \frac{{\delta{{\ensuremath{\left(q-{q^{\prime }}\right)}}}}}{q^2}\,,\end{aligned}$$ where $E_3$ is the energy of the three-particle system. In the case of a separable potential it is useful to define new components $\ket{F_i}$, which lead to a simpler set of equations [@Afnan:1977pi]. These are given by $$\label{eq:def_fdt} \ket{\psi_i} \eqqcolon G_0 t_i \ket{F_i}\,.$$ Now, the Faddeev equations become $$\label{eq:fd_f_g} \ket{F_i} = \sum_{j \neq i} G_0 t_j \ket{F_j}\,.$$ By inserting identities and some reshaping one obtains the final version of the Faddeev equations: $$\label{eq:fd_f} F_i{{\ensuremath{\left(q\right)}}} = 4 \pi \sum_{j \neq i} {\int \dd{{q^{\prime }}} {q^{\prime 2}}} X_{ij}{{\ensuremath{\left(q,{q^{\prime }}\right)}}} \tau_j{{\ensuremath{\left({q^{\prime }};E\right)}}} F_j{{\ensuremath{\left({q^{\prime }}\right)}}}\,,$$ with the definitions $$\begin{aligned} F_i{{\ensuremath{\left(q\right)}}} &\coloneqq {\int \dd{{p^{}}} {p^{2}}} {g_{l_i}{{\ensuremath{\left(p\right)}}}} { \prescript{}{i}{ \braket{p,q;\Omega_i}{F_i}_{}^{} } }\,, \\ X_{ij}{{\ensuremath{\left(q,{q^{\prime }}\right)}}} &\coloneqq {\int \dd{{p^{}}} {p^{2}}} {g_{l_i}{{\ensuremath{\left(p\right)}}}} G_0^{{\ensuremath{\left(i\right)}}}{{\ensuremath{\left(p,q;E\right)}}} \nonumber \\ & {\quad \cross}{\int \dd{{p^{\prime }}} {p^{\prime 2}}} { \prescript{}{i}{ \braket{p,q;\Omega_i}{{p^{\prime }},{q^{\prime }};\Omega_j}_{j}^{} } } {g_{l_j}{{\ensuremath{\left({p^{\prime }}\right)}}}} \,.\end{aligned}$$ Eq. (\[eq:fd\_f\]) is a set of homogeneous Fredholm integral equations of the second kind. It can be solved numerically by discretization of the momentum-space integral. Eqs. (\[eq:def\_fdt\]) and (\[eq:fd\_dcmp\]) then determine how the wave function is obtained from the solution for $F_i{{\ensuremath{\left(q\right)}}}$. The two integrals in $X_{ij}{{\ensuremath{\left(q,{q^{\prime }}\right)}}}$ can be reduced to one integration over an angle using the delta functions in the recoupling coefficient ${ \prescript{}{i}{ \braket{p,q;\Omega_i}{{p^{\prime }},{q^{\prime }};\Omega_j}_{j}^{} } }$. In the case of form factors of the type $g_l{{\ensuremath{\left(p\right)}}} = p^l \theta{{\ensuremath{\left(\beta_l -p \right)}}}$ this integral can then be evaluated analytically to yield Legendre functions of the second kind if we neglect the Heaviside step function $\theta$. However, here we carry out this last integral numerically, since this makes it possible to extend the calculation to different form factors at a low effort. By making some adjustments the additional numerical cost can be kept small. This has the additional advantage that no discrepancies due to neglecting the step function are introduced. Checks of the obtained solution are therefore better fulfilled. The formulas actually used for $X_{ij}$ are given in appendix \[subsec:num\_Xij\]. [^6^He]{} contains two neutrons outside the $\alpha$ core, and the state $\ket{\Psi}$ has to be antisymmetrized with respect to their exchange. Demanding $-{\mathcal{P}_{nn}}\ket{\Psi} = \ket{\Psi}$, where ${\mathcal{P}_{nn}}$ is the $nn$ permutation operator, and stating $V_{{n^\prime}} = {\ensuremath{\left(-{\mathcal{P}_{nn}}\right)}} V_n {\ensuremath{\left(-{\mathcal{P}_{nn}}\right)}}$, one finds (using $[{\mathcal{P}_{nn}},V_c]=0$) that $\ket{\psi_{{n^\prime}}} = - {\mathcal{P}_{nn}}\ket{\psi_n}$. Furthermore $\ket{F_{{n^\prime}}} = - {\mathcal{P}_{nn}}\ket{F_n}$ holds. So, in both versions of the Faddeev equations— and — only two out of three equations are linearly independent. Now that the solution procedure has been discussed we specify the t-matrices at leading order in Halo EFT: $$\begin{aligned} \langle {\ensuremath{\boldsymbol{p}}}|t_{c}(E)|{\ensuremath{\boldsymbol{p}}}' \rangle&= \frac{1}{4\pi^2 \mu_{nn}} \frac{1}{\gamma_0 + \mathrm{i} k}\,, \label{eq:t_nn}\\ \langle {\ensuremath{\boldsymbol{p}}}|t_{n}(E)|{\ensuremath{\boldsymbol{p}}}' \rangle&= \frac{3 p {p^{\prime }}}{4\pi^2 \mu_{nc}} \frac{1}{\gamma_1{\ensuremath{\left(k^2-k_R^2\right)}}}\,,\label{eq:t_nc}\end{aligned}$$ where $k = \sqrt{2\mu E}$ is the on-shell momentum of the corresponding system and we have assumed that $p$ and $p^\prime$ are much smaller than the cutoff. For a scattering volume $a_1$ and $p$-wave effective range $r_1$ the momentum of the resonance in the $nc$ interaction is given by $k_R = \sqrt{2/{\ensuremath{\left(a_1 r_1\right)}}}$. Furthermore, we designate $\gamma_1 = -r_1/2$ and $\gamma_0 = 1/a_0$ where $a_0$ is the $s$-wave scattering length. There is then a pole in the $s$-wave at $k={\mathrm{i}}\gamma_0$. The t-matrix $t_{n}$ contains no unitarity piece because of an expansion of the denominator according to the power counting: in this channel the unitarity term is included perturbatively at next-to-leading order and beyond [@bedaque03]. Finally, we discuss briefly how to solve the stationary Schrödinger equation $${\ensuremath{\left[H_0 + \sum_i {\ensuremath{\left(V_i + V_3^{{\ensuremath{\left(i\right)}}}\right)}} \right]}}\ket{\Psi} = E_3 \ket{\Psi}$$ in the presence of the three-body potentials $V_3^{{\ensuremath{\left(i\right)}}}$, which are defined with respect to spectator $i$. This can be done by adjusting the Faddeev equations. As explained in [@gloeckle83], one method is to modify Eq. (\[eq:def\_fd\_wfc\]) to $$\ket{\psi_i} \coloneqq G_0 {\ensuremath{\left(V_i + V_3^{{\ensuremath{\left(i\right)}}} \right)}} \ket{\Psi}\,.$$ The Faddeev equations, as defined in Eq. (\[eq:fd\]), then have to be adjusted. In practice, the equations that are solved to obtain the $\ket{F_i}$, , then change, while the formulas for calculating the $\ket{\psi_i}$ and $\ket{\Psi}$ from the $\ket{F_i}$ are unaltered. We use the three-body force as set up in [@ji14]. See Ref. [@Ryberg:2017tpv] for alternative possibilities. To close this section we explain a complication in the calculation of the three-body wave function. Using the properties of the t-matrices in Eqs. (\[eq:t\_pw\_me\]) and (\[eq:def\_fdt\]) we obtain $$\begin{aligned} \label{eq:fdwc_proj_prop} \ket{\psi_i} &= {\int \dd{{p^{}}} {p^{2}}} {\int \dd{{q^{}}} {q^{2}}} \sum_{\Omega^\prime} { \ket{p,q;\Omega^\prime}_{i}^{} } { \prescript{}{i}{ \braket{p,q;\Omega^\prime}{\psi_i}_{}^{} } } \nonumber \\ &= {\int \dd{{p^{}}} {p^{2}}} {\int \dd{{q^{}}} {q^{2}}} { \ket{p,q;\Omega_i}_{i}^{} } { \prescript{}{i}{ \braket{p,q;\Omega_i}{\psi_i}_{}^{} } } \,.\end{aligned}$$ The Faddeev component $\ket{\psi_i}$ is thus particularly simple in the representation ${ \ket{p,q;\Omega_i}_{i}^{} }$. This simplicity does not extend to the full wave function $\ket{\Psi}$: $ \ket{\Psi} \neq {\int \dd{{p^{}}} {p^{2}}} {\int \dd{{q^{}}} {q^{2}}} { \ket{p,q;\Omega_i}_{i}^{} } { \prescript{}{i}{ \braket{p,q;\Omega_i}{\Psi}_{}^{} } }$ in general because of the presence of the other two Faddeev components. When ${\int \dd{{p^{}}} {p^{2}}} {\int \dd{{q^{}}} {q^{2}}} \sum_{\Omega^\prime} { \ket{p,q;\Omega^\prime}_{i}^{} } { \prescript{}{i}{ \braket{p,q;\Omega^\prime}{\psi_j}_{}^{} } }$ is evaluated a recoupling has to be done between different spectators yielding many non-trivial overlaps. There is thus no single angular momentum state in the used basis containing the complete angular dependence of the state $\ket{\Psi}$. We therefore project on Jacobi momenta plane wave states, and evaluate the probability density for specific Jacobi momenta as the expectation value of the projection operator $P_{{{\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}}} = \ketbra{{{\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}}}{{{\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}}}$; this avoids having to choose specific spin states to project onto. Potentials from amplitudes {#subsec:coupling_strength} -------------------------- As already mentioned, we want to calculate the $\lambda_i$ from the $\tau_i$; we do so by using Eq. (\[eq:lb\_ls\]). The result depends on the chosen form factor. We consider two different functional forms. First: $${g_{l_{}}{{\ensuremath{\left(p\right)}}}} = p^l \theta{{\ensuremath{\left(\beta_l - p\right)}}}\,,$$ where $\beta_l$ denotes the regulation scale and $\theta$ is the Heaviside step function. We call this the [Heaviside form factor]{} here. It is the one used in Ref. [@ji14]. The second is the Yamaguchi form factor $${g_{l_{}}{{\ensuremath{\left(p\right)}}}} = p^l {\ensuremath{\left(1 + \frac{p^2}{\beta_l^2} \right)}}^{-{\ensuremath{\left(l+1\right)}}}\,.$$ The factor of $p^l$ is required to ensure the t-matrix satisfies the Wigner threshold law [@Wigner:1948zz]. The rest of the function ${g_{l_{}}{{\ensuremath{\left(p\right)}}}}$ determines not only the regularization scheme, but also the off-shell behavior of the t-matrix. As long as an appropriate three-body interaction is included, observables should not depend on the off-shell behavior of two-body interactions [@Haag:1958vt; @Polyzou1990]. We can test whether this is really the case in our calculation by examining results for different form factors. In the case of [Heaviside form factor]{}s we employ Eqs.  and in Eqs.  and and obtain for the interaction strength: $$\begin{aligned} \label{eq:std_ff_lambda_c} \lambda_c{{\ensuremath{\left(E\right)}}} &= \frac{1}{4\pi^2\mu_{nn}} \\ & {\quad \cross}{\ensuremath{\left[\frac{1}{a_0} - \sqrt{-k^2} - \frac{2}{\pi} \beta_0 + \frac{2}{\pi} \sqrt{-k^2} \arccot{{\ensuremath{\left(\frac{\sqrt{-k^2}}{\beta_0}\right)}}} \right]}}^{-1}\,, \nonumber \\ \lambda_n{{\ensuremath{\left(E\right)}}} &= \frac{1}{4\pi^2\mu_{nc}} \label{eq:std_ff_lambda_n} \\ & {\quad \cross}{\ensuremath{\left[ \gamma_1 {\ensuremath{\left(k^2 - k_R^2\right)}} - \frac{2 k^2\beta_1}{\pi } - \frac{2\beta_1^3}{3\pi} - \frac{2}{\pi}{\ensuremath{\left(-k^2\right)}}^{3/2} \arccot{{\ensuremath{\left(\frac{\sqrt{-k^2}}{\beta_1}\right)}}} \right]}}^{-1}\,. \nonumber\end{aligned}$$ Note that the expressions depend not only on $k^2$ but also on $\sqrt{-k^2}$. This is necessary in order to obtain potentials from which the t-matrices are reproduced exactly. The form of $\lambda_c(E)$ and $\lambda_n(E)$ which is more convenient for bound states is given in Eqs.  and . For scattering problems we would have to analytically continue via $\sqrt{-k^2} \rightarrow - \mathrm{i} k$. A unitarity term, $\mathrm{i}k$, then becomes visible in the equation for $\lambda_c$, but is canceled by the imaginary part of the analytic continuation of the $\arccot$. However, the equation for $\lambda_n$ does not have a unitarity term outside the $\arccot$. This is due to the fact that we use the $t_{n}$ given in Eq. (\[eq:t\_nc\]), which has no unitarity piece. $\lambda_c$ and $\lambda_n$ are also different because expanding the $\arccot$ in Eqs. (\[eq:std\_ff\_lambda\_c\]) and (\[eq:std\_ff\_lambda\_n\]) in powers of $1/\beta_i$ reveals that, in the limit $\beta_i \to \infty$, the energy dependence of $\lambda_c$ vanishes whereas that of $\lambda_n$ does not. This is why a discussion of energy-dependent potentials is unavoidable in that case, see Sec. \[sec:energydeppotentials\]. For the calculations to be carried out there we need derivatives of the $\lambda_i$ with respect to $E$. They are $$\begin{aligned} \pdv{\lambda_c}{E} &= -8\pi^2 \mu^2_{nn} \lambda_c^2{{\ensuremath{\left(E\right)}}} \nonumber \\ & {\quad \cross}{\ensuremath{\left[ \frac{1}{2\sqrt{-k^2}} - \frac{1}{\pi \sqrt{-k^2}} \arccot{{\ensuremath{\left(\frac{\sqrt{-k^2}}{\beta_0} \right)}}} + \frac{1}{\pi} \frac{\beta_0}{ \beta_0^2 -k^2 } \right]}}\,, \\ \pdv{\lambda_n}{E} &= -8\pi^2 \mu^2_{nc} \lambda_n^2{{\ensuremath{\left(E\right)}}} \nonumber \\ & {\quad \cross}{\ensuremath{\left[ \gamma_1 - \frac{2 \Lambda}{\pi } +\frac{3}{\pi} \sqrt{-k^2} \arccot{{\ensuremath{\left(\frac{\sqrt{-k^2}}{\beta_1}\right)}}} + \frac{1}{\pi} \frac{k^2}{\beta_1 - k^2 /\beta_1} \right]}} \,.\end{aligned}$$ In the case of Yamaguchi form factors the expressions for the strength parameters are: $$\begin{aligned} \lambda_c{{\ensuremath{\left(E\right)}}} &= \frac{1}{4\pi^2\mu_{nn}} {\ensuremath{\left( 1 + 2\frac{k^2}{{\beta_0}^2} + \frac{k^4}{{\beta_0}^4} \right)}} {\ensuremath{\left[ \frac{1}{a_0} - \frac{{\beta_0}}{2} + k^2 \frac{1}{2{\beta_0}} \right]}}^{-1} \eqqcolon \frac{1}{4\pi^2\mu_{nn} } \frac{u_c}{v_c} \,, \\ \lambda_n{{\ensuremath{\left(E\right)}}} &= \frac{1}{4\pi^2\mu_{nc}} {\ensuremath{\left( 1 + \frac{k^2}{{\beta_1}^2} \right)}}^4 \nonumber \\ & {\quad \cross}{\ensuremath{\left[ \gamma_1 {\ensuremath{\left(k^2 -k_R^2\right)}} - \frac{{\beta_1}^3}{16} {\ensuremath{\left( 1 + 9 \frac{k^2}{{\beta_1}^2} + 16 \frac{\sqrt{-k^2}^3}{{\beta_1}^3} - 9 \frac{k^4}{{\beta_1}^4} - \frac{k^6}{{\beta_1}^6} \right)}} \right]}}^{-1} \nonumber \\ &\eqqcolon \frac{1}{4\pi^2\mu_{nc} } \frac{u_n}{v_n}\,,\end{aligned}$$ where $u_c$, $v_c$, $u_n$ and $v_n$ are introduced in order to write the derivatives in a more compact form. These $\lambda_i$ also reproduce and exactly. Their derivatives are: $$\begin{aligned} \pdv{\lambda_c}{E} &= 2 \mu_{nn} \lambda_c {\ensuremath{\left( \frac{ \frac{2}{{\beta_0}^2} + \frac{2k^2}{{\beta_0}^4} }{u_c} - \frac{ \frac{1}{2{\beta_0}} }{v_c} \right)}}\,, \\ \pdv{\lambda_n}{E} &= 2 \mu_{nc} \lambda_n {\ensuremath{\left( \frac{ 4{\ensuremath{\left( 1 + \frac{k^2}{{\beta_1}^2} \right)}}^3 \frac{1}{{\beta_1}^2} }{u_n} - \frac{ - \frac{r_1}{2} - \frac{9}{16}{\beta_1}+\frac{3}{2} \sqrt{-k^2} + \frac{9}{8} \frac{k^2}{{\beta_1}} + \frac{3}{16} \frac{k^4}{{\beta_1}^3} }{v_n} \right)}} \,.\end{aligned}$$ In [@ji14] similar derivations of the strength parameters were done. Tuning the regulator parameters $\beta_0$ and $\beta_1$ of the Yamaguchi form factors as well as the interaction strengths $\lambda_c$ and $\lambda_n$ Ji [*et al.*]{} were able to reproduce to the effective range parameters $a_0$, $r_0$, $a_1$ and $r_1$ without energy-dependent potentials. However, in contrast to what is done here, the inverse t-matrices in [@ji14] still contain pieces $\sim 1/\beta_i$. They therefore do not exactly reproduce the forms and , but only do so up to regulator artifacts associated with higher-order terms in the effective-range expansion. By using energy-dependent potentials we have completely eliminated these artifacts. We have also guaranteed that $r_1$ is reproduced, no matter what value of the regulator parameter $\beta_1$ is employed. The implications of this energy dependence are discussed in the following section. Normalization and orthogonality relations for energy-dependent potentials {#sec:energydeppotentials} ========================================================================= Orthogonality and normalization relations in the presence of energy-dependent potentials ---------------------------------------------------------------------------------------- In this section we consider the Schrödinger Equation for an energy-dependent potential, $V(E)$. We assume that $V$, while energy-dependent, is still Hermitian, i.e., $$\langle \phi_1|V(E)|\phi_2 \rangle=\langle \phi_2|V(E)|\phi_1 \rangle^*\,,$$ for all real energies $E$ and for all states $|\phi_1 \rangle$ and $|\phi_2 \rangle$. We define eigenstates of the problem corresponding to the energy $E$ according to: $$(H_0 + V(E))|\psi_E \rangle=E|\psi_E \rangle\,. \label{eq:SE}$$ with $H_0$ also—as usual—a Hermitian operator. Our discussion of this problem summarizes results from Refs. [@mckellar83; @formanek04]. Consider two eigenstates of $H(E)$ corresponding to different energies, $E_\alpha$ and $E_\beta$ denoted as $|\psi_\alpha \rangle$ and $|\psi_\beta \rangle$, respectively. Since the two states correspond to the solutions of different eigenvalue problems–one for $H(E_\alpha)$ and one for $H(E_\beta)$—they are not orthogonal. By projecting the equation for the eigenstate $|\psi_\alpha \rangle$ onto $|\psi_\beta \rangle$, and vice versa, and taking the difference of the results, we find: $$\langle \psi_\beta|V(E_\alpha) - V(E_\beta)|\psi_\alpha \rangle=(E_\alpha - E_\beta) \langle \psi_\beta|\psi_\alpha \rangle\,,$$ which may be rewritten as: $$\langle \psi_\beta|({\mathbbm{1}}- \Delta V_{\beta \alpha})|\psi_\alpha \rangle=0\,, \label{eq:orthogonality}$$ with $$\Delta V_{\beta \alpha} \coloneqq \frac{V(E_\alpha) - V(E_\beta)}{E_\alpha - E_\beta}\,.$$ Obviously $\Delta V=0$ for an energy-independent potential. But, in a problem with an energy-dependent potential, Eq. (\[eq:orthogonality\]) states that the orthogonality relation is only obtained if, to the usual “1" that is inserted between the two states, we add the contribution from $\Delta V$. Alternatively, we can say that $\langle \psi_\beta|$ is orthogonal to $(1 - \Delta V_{\beta \alpha})|\psi_\alpha \rangle$ and/or $|\psi_\alpha \rangle$ is orthogonal to $\langle \psi_\beta|(1 - \Delta V_{\beta \alpha})$. This is related to the modifications to the standard scalar product for states bound by a pseudopotential discussed by Pricoupenko in Ref. [@Pricoupenko:2006A; @Pricoupenko:2006B]. We may use this interpretation as the starting point for construction of the operator $D^{-1}$, introduced by McKellar and McKay [@mckellar83], that converts the set of states $\langle \psi_\beta|$ into a set of biorthogonal states. I.e., we seek $D$ such that: $$\langle \langle \psi_\beta|=\langle \psi_\beta|D^{-1}$$ with $$\langle \langle \psi_\beta|\psi_\alpha \rangle=\delta_{\beta \alpha}\,.$$ It is then tempting to identify $D^{-1}=1- \Delta V_{\beta \alpha}$. However—as pointed out by Formánek [*et al.*]{} [@formanek04]—this identification is not correct in general, because $\Delta V_{\beta \alpha}$ is an energy-dependent, and hence-state dependent, operator. McKellar and McKay show how to construct $D^{-1}$ as the solution to an integral equation. With the operator $D^{-1}$ in hand we can write a completeness relation: $${\mathbbm{1}}=\sum_\alpha |\psi_\alpha \rangle \langle \langle \psi_\alpha|=\sum_\alpha |\psi_\alpha \rangle \langle \psi_\alpha| D^{-1}\,. \label{eq:id}$$ The presence of $D^{-1}$ here has an important implication: the set of states $\{|\psi_\alpha \rangle\}$, while complete, is not orthornormal, and so [*cannot*]{} be inserted as an identity when deriving quantum-mechanical equations. This is crucial for some forms of scattering-theory equations [@mckellar83], but in the derivations performed in Sec. \[sec:6HeHaloEFTreview\] only complete sets of plane waves were inserted, so it does not affect any of the results presented above. It does, though, affect the normalization condition for the three-body wave function, and hence the definition of the momentum density. From the resolution of the identity operator, (\[eq:id\]), we identify the projector onto the eigenstate of $H(E)$ corresponding to energy $E_\alpha$ as $$P_\alpha=|\psi_\alpha \rangle \langle \langle \psi_\alpha|=|\psi_\alpha \rangle \langle \psi_\alpha| D^{-1}\,. \label{eq:projector}$$ For $P_\alpha$ defined according to Eq. (\[eq:projector\]) to be a standard projector onto a subspace of the Hilbert space of eigenstates of $H(E)$ we need $P_\alpha^2=P_\alpha$. This will only be true if: $$\langle \psi_\alpha| D^{-1}|\psi_\alpha \rangle=1.$$ In the case of a linear-in-energy potential, $V(E)=V_0 + V_1 E$, the argument becomes much simpler, since then $\Delta V_{\beta \alpha}=V_1$ is state independent, and $D^{-1}={\mathbbm{1}}-V_1$ at the operator level. Equation (\[eq:projector\]) then reduces to: $$P_\alpha=|\psi_\alpha \rangle \langle \psi_\alpha| ({\mathbbm{1}}- V_1)\,, \label{eq:Palpha}$$ and the normalization condition is: $$\langle \psi_\alpha|({\mathbbm{1}}- V_1)|\psi_\alpha \rangle=1\,. \label{eq:normconditionlinear}$$ This result permits us to construct the diagonal matrix elements of $D^{-1}$ for an arbitrary $V(E)$. For energies $E \approx E_\alpha$ we can replace $V(E) \approx V(E_\alpha) + V'(E_\alpha) (E-E_\alpha)$. Following the steps of the previous paragraph then gives the normalization of the state $|\psi_\alpha \rangle$ as: $$\langle \psi_\alpha|({\mathbbm{1}}- V'(E_\alpha))|\psi_\alpha \rangle=1\,. \label{eq:normcondition}$$ Formánek [*et al.*]{} arrive at the same conclusion by considering the continuity equation in the presence of an energy-dependent potential [@formanek04]. For yet another derivation, see Appendix \[ap:alternativederivationnorm\]. From now on we refer to the normalization condition $\langle \psi_\alpha|\psi_\alpha \rangle=1$ as the “naive" normalization condition and Eq. (\[eq:normcondition\]) as the “correct" normalization. The projector $P_\alpha$ is not a Hermitian operator. So simply evaluating diagonal matrix elements of $P_\alpha$ using an eigenstate of the momentum operator does not lead to a momentum-space probability density with expected properties, e.g., $\rho_\alpha({\bf p}) \geq 0$. As pointed out by Formánek [*et al.*]{}, if it is to correspond to an observable, the operator $O$ should be Hermitian with respect to matrix elements $\langle \langle \phi|O|\psi \rangle$. This means that $O$ must equal $O^\#$ with the operator $O^\#$ defined by its matrix elements as: $$\langle \langle \psi| O^\# |\phi \rangle \coloneqq \langle \langle \phi|O|\psi \rangle^*\,,$$ or, in operator form: $$O^\# \coloneqq D O^\dagger D^{-1}\,,$$ with $^\dagger$ the usual Hermitian conjugation. (Here we have used, without proof, that $D=D^\dagger$.) The simplest extension of the usual momentum-space projector $|{\bf p} \rangle \langle {\bf p}|$ which obeys this requirement is $$P_{\bf p}=\sqrt{D} |{\bf p} \rangle \langle {\bf p}| \sqrt{D^{-1}}\,. \label{eq:PpD}$$ Note that ${\rm Tr}(P_{\bf p})=1$ for a set of normalized momentum-space eigenstates. Note also that Eq. (\[eq:PpD\]) is only well defined if $D^{-1}$ is a positive semi-definite operator. We now construct the momentum-space probability density by taking the trace of the product of the momentum-space projector (\[eq:PpD\]) and the bound-state projector (\[eq:Palpha\]), which acts as a pure-state density matrix. In general this gives $$\rho_{\alpha}({\bf p})=\langle \psi_\alpha|\sqrt{D^{-1}}|{\bf p} \rangle \langle {\bf p}|\sqrt{D^{-1}}|\psi_\alpha \rangle=|\langle {\bf p}|\sqrt{D^{-1}}|\psi_\alpha \rangle|^2\,, \label{eq:exactpd}$$ which is non-negative, as expected[^2]. Constructing the operator $\sqrt{D^{-1}}$ is, in general, somewhat complicated for the three-body problem. In what follows we will exploit the fact that, for the momenta at which the EFT is valid and for cutoffs $\Lambda$ well above the EFT breakdown scale, $V'(E_\alpha) \ll 1$ (see explicit formulas in Sec. \[sec:6HeHaloEFTreview\]). In that case the formal expression involving $\sqrt{D^{-1}}$ can be replaced by $$\rho_{\alpha}({\bf p})=|\langle {\bf p}|\sqrt{{\mathbbm{1}}-V'(E_\alpha)}|\psi_\alpha \rangle|^2\,. \label{eq:density}$$ This density is positive semi-definite and is also normalized to 1, provided that $|\psi_\alpha \rangle$ is normalized according to Eq. (\[eq:normcondition\]). But in the regime where $V'(E_\alpha) \ll 1$ the density (\[eq:density\]) can be further approximated as: $$\rho_{\alpha}({\bf p})\approx |\langle {\bf p}|\psi_\alpha \rangle|^2 - \Re(\langle {\bf p} |V'(E_\alpha)|\psi_\alpha \rangle \langle \psi_\alpha|{\bf p} \rangle)\,. \label{eq:approxdensity}$$ Note that: 1. If such an evaluation of $\rho_\alpha$ yields a negative number it signifies that we are outside the domain of validity of the expansion used to justify it. 2. The approximation leading to (\[eq:approxdensity\]) preserves the normalization of $\rho_\alpha$, thanks to the normalization condition (\[eq:normcondition\]). The above derivation considered eigenstates of a single momentum operator. In the three-body problem we construct the momentum-space projector for a specific Jacobi representation and get: $$\rho_i({{\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}}) \approx { \prescript{}{i}{ \braket{ {{\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}}}{\Psi}_{}^{} } } \left( { \prescript{}{}{ \braket{ \Psi}{ {{\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}}}_{i}^{} } } - \Re\left[ \sum_j { \prescript{}{}{ \mel{ \Psi \vphantom{\frac{q^2}{2\mu_{i{\ensuremath{\left(jk\right)}}}}} }{ V_j'{{\ensuremath{\left(E_\alpha- \frac{q_j^2}{2\mu_{j{\ensuremath{\left(ki\right)}}}} \right)}}} }{ {{\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}}}_{i}^{} } } \right]\right)\,, \label{eq:Threebodydensity}$$ where the subscript $i$ now indicates the Jacobi momenta ${\bf p}$ and ${\bf q}$ for which we are computing $\rho$, and [*not*]{} the state of interest. We have assumed that the three-body potential, $V_3$, is energy independent, but have not assumed anything about which two-body subsystem(s) the energy-dependent potential is active in. If $j \neq i$ recoupling is necessary in order to compute the matrix element of $V_j$ here. Finally we reiterate that, while the naive probability density contains only the one-body part $|\psi|^2$ (or $|\Psi|^2$), our corrected results, Eqs. (\[eq:approxdensity\]) and (\[eq:Threebodydensity\]), include two-body contributions. Example: two-body bound state with an energy-dependent contact interaction -------------------------------------------------------------------------- The EFT for a two-body system in which both the scattering length and effective range are unnaturally large [@beane00] can be implemented by considering an energy-dependent potential whose matrix elements are: $$\langle {\ensuremath{\boldsymbol{p}}}|V(E)|{\ensuremath{\boldsymbol{p}}}' \rangle=\frac{1}{\Delta_0 + \Delta_2 E}\,,$$ with $\Delta_0$ and $\Delta_2$ parameters that, as we will see below, are related to the bound-state pole position and residue. If $V$ is interpreted as due to an $s$-channel particle exchange then $\Delta_0$ is proportional to the particle’s mass shift, and $\Delta_2$ is related to its kinetic mass [@kaplan97]. Reference [@beane00] showed that the charge density for such a system includes both one- and two-body terms, but is then properly normalized. Here we demonstrate that the normalization condition (\[eq:normcondition\]) has the same property. First we compute the two-body t-matrix for this problem: $t(E)=V(E) + V(E) G_0(E) t(E)$. Since $V(E)$ is independent of momentum $t(E)$ is too, and the Lippmann-Schwinger equation becomes algebraic. The solution is: $$\langle {\ensuremath{\boldsymbol{p}}}|t(E)|{\ensuremath{\boldsymbol{p}}}'\rangle=\frac{1}{\Delta_0 + \Delta_2 E + 8 \pi m_R \Lambda + {\mathrm{i}}4 \pi^2 m_R k}\,,$$ with $m_R$ the two-body reduced mass, $k=\sqrt{2 m_R E}$ the on-shell momentum, and $\Lambda$ a sharp cutoff used to make the integral of $G_0(E)$ finite. We adjust $\Delta_0$ so that the system has one bound state, at $E=-B$. $t$ then takes the form: $$\langle {\ensuremath{\boldsymbol{p}}}|t(E)|{\ensuremath{\boldsymbol{p}}}' \rangle=\frac{1}{4 \pi^2 m_R} \frac{1}{1 - \gamma \rho}\frac{1}{\gamma + {\mathrm{i}}k}\,,$$ with: $$\frac{\Delta_0}{4 \pi^2 m_R}=\gamma - \frac{2\Lambda}{\pi} - \frac{\rho \gamma^2}{2}; \qquad \frac{\Delta_2}{4 \pi^2 m_R}=-\rho m_R \,.$$ where $\rho$ is the effective range for the expansion around the deuteron pole. Note that the second equation has pieces of ${\cal O}(1/\Lambda)$ if we include the regulator in the potential. In the calculation of the coupling strengths in subsection \[subsec:coupling\_strength\] this effect of the $s$-wave regulator (there denoted $\beta_0$) was taken into account. The three-dimensional wave function $\psi$ obtained directly from $t(E)$, without imposing any normalization, is: [@phillips02; @hammer17] $$\langle {\bf p}|\psi \rangle=\frac{1}{\pi} \frac{\sqrt{{\cal Z} \gamma}}{p^2 + \gamma^2}$$ where we have defined ${\cal Z}=1/(1 - \gamma \rho)$ as the wave function renormalization, because $$\int \dd^3p |\langle {\bf p}|\psi \rangle|^2={\cal Z}\,.$$ So the wave function is not normalized according to the naive normalization condition. Now $$\langle {\bf p}'| V'(-B)|{\bf p}\rangle=-\frac{\Delta_2}{(\Delta_0 - \Delta_2 B)^2}= \frac{\rho}{4 \pi^2} \frac{1}{(\gamma - \frac{2 \Lambda}{\pi})^2}$$ Therefore the two-body momentum density in the region of validity of the EFT, Eq. (\[eq:approxdensity\]) is approximately: $$\rho({\bf p}) \approx \frac{{\cal Z} \gamma}{\pi^2}\frac{1}{p^2 + \gamma^2} \left[\frac{1}{p^2 + \gamma^2} - \frac{\rho}{2} \frac{1}{(\gamma - \frac{2 \Lambda}{\pi})}\right]\,, \label{eq:rhop}$$ where we have regulated the divergent integral encountered in the two-body contribution to the probability density using the same sharp cutoff as was employed in the calculation of the t-matrix. (In fact, the same divergent integral that appears everywhere in this analysis, $\int \frac{\dd^3p}{\sqrt{(2 \pi)^3}} \psi({\bf p})=\psi({\bf r}=0)$.) Integrating Eq. (\[eq:rhop\]) over all momenta (and again regulating the second term using a sharp cutoff) we find: $$\int \dd^3p \rho({\bf p})={\cal Z}[1 - \rho \gamma]=1\,.$$ This verifies that the normalization condition (\[eq:normcondition\]) produces a properly normalized probability density for this problem. To close this section we note a few points that are relevant for our application of this formalism to ${}^6$He in the next section - For momenta $p \sim \gamma$ the two-body contribution to $\rho({\bf p})$ is a correction to the one-body piece of ${\cal O}\left(\frac{\rho \gamma^2}{\Lambda}\right)$. For large cutoffs it is thus negligible at low momentum. This justifies the linear approximation used to evaluate $\rho({\bf p})$. - For $p \sim \Lambda$ the two-body contribution is larger than the one-body contribution by a factor $\rho \Lambda$. - In the approximate form of $\rho$ the two-body part of the momentum-space probability density is negative. This is to be expected since ${\cal Z} > 1$. But, since the positive one-body piece dominates at small momentum, it follows that $\rho({\bf p})$ will go through zero at a momentum $p_0 \sim \sqrt{\Lambda/\rho}$. For sufficiently large $\Lambda$ this is well above the breakdown scale of the EFT; the expression we are using for $\rho({\bf p})$ is not reliable for $p \sim p_0$. - The two-body contribution to the norm is proportional to a divergent integral. This, combined with the ${\cal O}(1/\Lambda)$ coefficient, produces a finite effect. Results for the probability density {#sec:probdensity} =================================== Calculation of one-body probability density {#subsec:naive_pbd} ------------------------------------------- The probability density is given by $$\begin{aligned} \label{eq:prob_dens} {\ensuremath{ \rho_{i}{{\ensuremath{\left({{\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}}\right)}}} }} &\coloneqq \mel{\Psi}{{\ensuremath{\left( { \ifthenelse{\equal{}{}}{ { \ket{{{\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}}}_{i}^{} } { \prescript{}{i}{ \bra{{{\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}}} } } } { { \ket{{{\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}}}_{i}^{} } { \prescript{}{i}{ \bra{} } } } } \otimes {\mathbbm{1}^{{{\ensuremath{\left(\mathrm{spin}\right)}}}}}\right)}}}{\Psi} \quad \textrm{or}\\ {\ensuremath{ \tilde{\rho}_{i}{{\ensuremath{\left({{\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}}\right)}}} }} &\coloneqq \mel{\Psi}{{\ensuremath{\left( { \ifthenelse{\equal{}{}}{ { \ket{{{\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}}}_{i}^{} } { \prescript{}{i}{ \bra{{{\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}}} } } } { { \ket{{{\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}}}_{i}^{} } { \prescript{}{i}{ \bra{} } } } } \otimes {\mathbbm{1}^{{{\ensuremath{\left(\mathrm{spin}\right)}}}}}\right)}}{\ensuremath{\left({\mathbbm{1}}- \pdv{V}{E_3}\right)}}}{\Psi}\label{eq:prob_dens_mn}\end{aligned}$$ in the presence of an energy-dependent potential. Note that, in order to define the projection operator, we include an identity in spin space. Hence, the probability density we compute here is averaged over all allowed spin states. $\ket{\Psi}$ is calculated from $\ket{F_i}$ via the equations $\ket{\Psi} = \ket{\psi_n} + \ket{\psi_{n^\prime}} + \ket{\psi_c}$ and $\ket{\psi_i} = G_0 t_i \ket{F_i}$. Since we obtained $F_i{{\ensuremath{\left(q\right)}}} \coloneqq {\int \dd{{p^{}}} {p^{2}}} {g_{l_i}{{\ensuremath{\left(p\right)}}}} { \prescript{}{i}{ \braket{p,q;\Omega_i}{F_i}_{}^{} } }$, it is useful to insert an identity between the t-matrix and the Faddeev amplitude in the partial wave representation. Due to the projection property of the t-matrix it also useful to insert one identity on its left. We then obtain $$\begin{aligned} \label{eq:psi_i_pw} \sum_{\Omega} { \ket{p,q;\Omega}_{i}^{} } { \prescript{}{i}{ \braket{p,q;\Omega}{\psi_i}_{}^{} } } &= { \ket{p,q;\Omega}_{i}^{} } 4\pi {G_0^{{\ensuremath{\left(i\right)}}}{{\ensuremath{\left(p,q\right)}}}} {g_{l_i}{{\ensuremath{\left(p\right)}}}} {\tau_{i}{{\ensuremath{\left(q;-B_3^{{\ensuremath{\left(0\right)}}}\right)}}}} \nonumber \\ &{\quad \cross}{\int \dd{{p^{\prime }}} {p^{\prime 2}}} {g_{l_i}{{\ensuremath{\left({p^{\prime }}\right)}}}} { \prescript{}{i}{ \braket{{p^{\prime }},q;\Omega_i}{F_i}_{}^{} } } \nonumber \\ &= { \ket{p,q;\Omega}_{i}^{} } 4\pi {G_0^{{\ensuremath{\left(i\right)}}}{{\ensuremath{\left(p,q\right)}}}} {g_{l_i}{{\ensuremath{\left(p\right)}}}} {\tau_{i}{{\ensuremath{\left(q;-B_3^{{\ensuremath{\left(0\right)}}}\right)}}}} F_i{{\ensuremath{\left(q\right)}}} \nonumber \\ &= { \ket{p,q;\Omega}_{i}^{} } \psi_i{{\ensuremath{\left(p,q\right)}}} \,,\end{aligned}$$ where the definition ${\tau_{i}{{\ensuremath{\left(q;E_3\right)}}}} \coloneqq \tau_i{{\ensuremath{\left(E_3-q^2/{\ensuremath{\left(2\mu_{i{\ensuremath{\left(jk\right)}}}\right)}}\right)}}}$ is used and the component wave function $\psi_i{{\ensuremath{\left(p,q\right)}}} \coloneqq { \prescript{}{i}{ \braket{p,q;\Omega}{\psi_i}_{}^{} } }$ is introduced. In order to evaluate the expressions for the probability density given in Eq. (\[eq:prob\_dens\]) or in Eq. (\[eq:prob\_dens\_mn\]) using this expression one has to decouple $\ket{\Omega_i}_i$ into the angular and spin part as given in [@ji14]. The emerging overlaps of plane wave states with partial wave states are given by coupled spherical harmonics $$\begin{aligned} {\mathcal{Y}_{l \lambda}^{LM}{{\ensuremath{\left({\ensuremath{\boldsymbol{{p}}}}, {\ensuremath{\boldsymbol{{q}}}}\right)}}}} &\coloneqq \braket{{{\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}}}{p,q;{\ensuremath{\left(l \lambda\right)}} LM}\\ &=\sum_{\mathclap{m_l+m_\lambda=M}} {C{{\ensuremath{\left(l m_l \lambda m_\lambda \,\vline \, L M\right)}}}} {Y_{l,m_l}{{\ensuremath{\left({\ensuremath{\boldsymbol{{p}}}}\right)}}}} {Y_{\lambda,m_\lambda}{{\ensuremath{\left({\ensuremath{\boldsymbol{{q}}}}\right)}}}}\,,\end{aligned}$$ where ${Y_{l,m_l}{{\ensuremath{\left({\ensuremath{\boldsymbol{{p}}}}\right)}}}}$ is the usual spherical harmonic. Using Eq. (\[eq:psi\_i\_pw\]) we obtain[^3] $$\begin{aligned} \label{eq:ov_plw_psi} { \prescript{}{i}{ \braket{{{\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}}}{\Psi}_{}^{} } } &= \sum_{\Omega} { \prescript{}{i}{ \braket{{{\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}}}{p,q;\Omega}_{i}^{} } } { \prescript{}{i}{ \braket{p,q;\Omega}{\Psi}_{}^{} } } \nonumber \\ &= a_i {\sum_{L=0}^1 \sum_{M=-L}^{L} } c_{L,M}{\mathcal{Y}_{11}^{LM}{{\ensuremath{\left({\Upsilon}_i\right)}}}} \ket{L,-M}_n \nonumber \\ &{\quad}- \tilde{a}_i {\sum_{L=0}^1 \sum_{M=-L}^{L} } c_{L,M}{\mathcal{Y}_{11}^{LM}{{\ensuremath{\left(\tilde{{\Upsilon}}_i\right)}}}} {\mathcal{P}_{nn}^\mathrm{(spin)}}\ket{L,-M}_n + \frac{d_i}{4\pi} \ket{0,0}_c\,,\end{aligned}$$ where ${\mathcal{P}_{nn}^\mathrm{(spin)}}$ is the spin space part of the $nn$ commutation operator ${\mathcal{P}_{nn}}$. The coefficients $c_{L,M}$ result from the recoupling from [${\ensuremath{\boldsymbol{j}}}{\ensuremath{\boldsymbol{J}}}$-coupling]{} to [${\ensuremath{\boldsymbol{L}}}{\ensuremath{\boldsymbol{S}}}$-coupling]{} given in [@ji14]. Thus they do not depend on the spectator. They are given by $c_{L,M} \coloneqq {\ensuremath{\left(-1\right)}}^{M} \sqrt{\frac{2^{1-L}}{6L+3}}$ with $L \in \{0,1\}$ and $-L \leq M \leq L$. Note that the result of Eq. (\[eq:ov\_plw\_psi\]) is the sum of scalars times spin states. The other parameters, which depend on the spectator $i$, are given by $$\begin{aligned} \label{eq:pbd_coeff_n_first} a_n &= \psi_n{{\ensuremath{\left(p,q\right)}}}\,, \\ \tilde{a}_n &= \psi_n{{\ensuremath{\left({\kappa_{nnp}^\prime},{\kappa^\prime_{nnq}}\right)}}}\,, \\ d_n &= \psi_c{{\ensuremath{\left({\kappa_{ncp}}, {\kappa_{ncq}}\right)}}}\,, \\ {\Upsilon}_n &= {\ensuremath{\left({\ensuremath{\boldsymbol{{p}}}}, {\ensuremath{\boldsymbol{{q}}}}\right)}}\,, \\ \tilde{{\Upsilon}}_n &= {\ensuremath{\left( {{\ensuremath{\boldsymbol{{\kappa}}}}_{nnp}^{\prime}}, {{\ensuremath{\boldsymbol{{\kappa}}}}_{nnq}^{\prime}}\right)}}\,.\end{aligned}$$ for the $n$ as spectator. The definition of the $\kappa$ are given in appendix \[subsec:jacobi\]. The functions in $a_i$ depend on the norm of the vectors whose unit vectors are contained in ${\Upsilon}_i$, the analogous relation holds for $\tilde{a}_i$ and $\tilde{{\Upsilon}}_i$. Subsequently the expressions for $c$ as spectator are given: $$\begin{aligned} a_c &= \psi_n{{\ensuremath{\left({\kappa_{cnp}}, {\kappa_{cnq}}\right)}}}\,, \\ \tilde{a}_c &= \psi_n{{\ensuremath{\left({\kappa_{cnp}}^\prime, {\kappa_{cnq}}^\prime\right)}}}\,, \\ d_c &= \psi_c{{\ensuremath{\left(p,q\right)}}}\,, \\ {\Upsilon}_c &= {\ensuremath{\left({{\ensuremath{\boldsymbol{{\kappa}}}}_{cnp}}, {{\ensuremath{\boldsymbol{{\kappa}}}}_{cnq}}\right)}}\,, \\ \tilde{{\Upsilon}}_c &= {\ensuremath{\left({{\ensuremath{\boldsymbol{{\kappa}}}}_{cnp}^{\prime}}, {{\ensuremath{\boldsymbol{{\kappa}}}}_{cnq}^{\prime}}\right)}}\,. \label{eq:pbd_coeff_c_last}\end{aligned}$$ Based on Eq. (\[eq:ov\_plw\_psi\]) and the following equations we obtain the expression below for the one-body probability density by using the overlaps of spin states. Most of these overlaps are given in [@ji14]. $$\begin{aligned} \label{eq:ob_pbd} \rho_i{{\ensuremath{\left({\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}\right)}}} &= \mel{\Psi}{{\ensuremath{\left({ \ifthenelse{\equal{}{}}{ { \ket{{{\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}}}_{i}^{} } { \prescript{}{i}{ \bra{{{\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}}} } } } { { \ket{{{\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}}}_{i}^{} } { \prescript{}{i}{ \bra{} } } } } \otimes {\mathbbm{1}^{{{\ensuremath{\left(\mathrm{spin}\right)}}}}}\right)}}}{\Psi} \nonumber \\ &= \bigg[ {\sum_{L=0}^1 \sum_{M=-L}^{L} } c_{L,M}^2 {\ensuremath{\left( a_i^2 \left| {\mathcal{Y}_{11}^{LM}{{\ensuremath{\left({\Upsilon}_i\right)}}}} \right|^2 + \tilde{a}_i^2 \left| {\mathcal{Y}_{11}^{LM}{{\ensuremath{\left(\tilde{{\Upsilon}}_i\right)}}}} \right|^2 \right)}} + {\ensuremath{\left(\frac{d_i}{4\pi}\right)}}^2 \nonumber \\ & {\quad}- 2a_i \tilde{a}_i {\sum_{L=0}^1 \sum_{M=-L}^{L} } c_{L,M}^2 {\ensuremath{\left(-1\right)}}^{1-L} \Re{\ensuremath{\left({\mathcal{Y}_{11}^{LM}{{\ensuremath{\left(\tilde{{\Upsilon}}_i\right)}}}}^* {\mathcal{Y}_{11}^{LM}{{\ensuremath{\left({\Upsilon}_i\right)}}}}\right)}} \nonumber \\ & {\quad}- 2\tilde{a}_i \frac{d_i}{4\pi} c_{0,0} {\mathcal{Y}_{11}^{00}{{\ensuremath{\left(\tilde{{\Upsilon}}_i\right)}}}} -2 a_i \frac{d_i}{4\pi} c_{0,0} {\mathcal{Y}_{11}^{00}{{\ensuremath{\left({\Upsilon}_i\right)}}}} \bigg]\,.\end{aligned}$$ We used that $a_i$, $\tilde{a}_i$ and $d_i$ are real. Because $c_{L,M}^2$ is independent of $M$ it follows that $\rho_i{{\ensuremath{\left({{\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}}\right)}}}$ only depends on $p$, $q$ and $\theta_{pq} \coloneqq \arccos{ {\ensuremath{\left({\ensuremath{\boldsymbol{p}}} \cdot {\ensuremath{\boldsymbol{q}}}/{\ensuremath{\left(pq\right)}}\right)}}} $. The simplification is very helpful, since it reduces the computational effort of angular integrals of the probability density. Additional remarks on it can be found in appendix \[subsec:ang\_simp\]. It also motivates an expansion in $\cos{\theta_{pq}}$: $$\begin{aligned} {\ensuremath{ \rho_{i}^{{\ensuremath{\left(\xi\right)}}}{{\ensuremath{\left(p, q\right)}}} }} &= \int_{0}^\pi \dd{\theta_{pq}} \sin{\theta_{pq}} P_{\xi}{{\ensuremath{\left(\cos{\theta_{pq}}\right)}}} {\ensuremath{ \rho_{i}{{\ensuremath{\left({{\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}}\right)}}} }}\,, \label{eq:rhopw} \\ {\ensuremath{ \tilde{\rho}_{i}^{{\ensuremath{\left(\xi\right)}}}{{\ensuremath{\left(p, q\right)}}} }} &= \int_{0}^\pi \dd{\theta_{pq}} \sin{\theta_{pq}} P_{\xi}{{\ensuremath{\left(\cos{\theta_{pq}}\right)}}} {\ensuremath{ \tilde{\rho}_{i}{{\ensuremath{\left({{\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}}\right)}}} }}\,, \label{eq:rhotpw}\end{aligned}$$ where $P_\xi$ denotes the $\xi$-th Legendre polynomial. Although $\xi$ looks like an angular-momentum quantum number there is, in fact, no wave function component angular momentum—neither $l$ nor $\lambda$ nor $L$—that can be associated with it. Calculation of two-body probability density {#subsec:mod_pbd} ------------------------------------------- We proceed by calculating an approximation to the two-body probability density as given in Eq. (\[eq:prob\_dens\_mn\]), which accounts for the energy dependence of the potentials. For simplicity we call it just the two-body probability density and do not stress its approximate character. In subsection \[subsec:coupling\_strength\] the expressions for the coupling strengths of the potentials $V_c$ and $V_n$ were already derived. As explained in subsection \[subsec:heft\], for $V_{{n^\prime}}$ the equation $$\begin{aligned} V_{{n^\prime}} = {\ensuremath{\left(- {\mathcal{P}_{nn}}\right)}} V_n {\ensuremath{\left(- {\mathcal{P}_{nn}}\right)}}\end{aligned}$$ holds. In following calculations we will use $V_{{n^\prime}}\ket{\Psi} = {\ensuremath{\left(- {\mathcal{P}_{nn}}\right)}} V_n \ket{\Psi}$. Since $V = V_n + V_{{n^\prime}} + V_c + V^{{\ensuremath{\left(3\right)}}}$ and the three-body potential $V^{{\ensuremath{\left(3\right)}}}$ is energy indepedendent, the derivative of the potential with respect to the energy is given by $$\pdv{V}{E_3} = \pdv{V_n}{E_3} + \pdv{V_{{n^\prime}}}{E_3} + \pdv{V_c}{E_3}\,.$$ It is useful to split the calculation in the following way: $$\begin{aligned} \label{eq:prob_dens_mn_dt} &\mel{\Psi}{{\ensuremath{\left( { \ifthenelse{\equal{}{}}{ { \ket{{{\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}}}_{i}^{} } { \prescript{}{i}{ \bra{{{\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}}} } } } { { \ket{{{\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}}}_{i}^{} } { \prescript{}{i}{ \bra{} } } } } \right)}}{\ensuremath{\left({\mathbbm{1}}- \pdv{V}{E_3}\right)}}}{\Psi} = \mel{\Psi}{{\ensuremath{\left( { \ifthenelse{\equal{}{}}{ { \ket{{{\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}}}_{i}^{} } { \prescript{}{i}{ \bra{{{\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}}} } } } { { \ket{{{\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}}}_{i}^{} } { \prescript{}{i}{ \bra{} } } } } \right)}}}{\Psi} \nonumber \\ & {\quad}- \mel{\Psi}{{\ensuremath{\left( { \ifthenelse{\equal{}{}}{ { \ket{{{\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}}}_{i}^{} } { \prescript{}{i}{ \bra{{{\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}}} } } } { { \ket{{{\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}}}_{i}^{} } { \prescript{}{i}{ \bra{} } } } } \right)}}\pdv{V_n}{E_3}}{\Psi} - \mel{\Psi}{{\ensuremath{\left( { \ifthenelse{\equal{}{}}{ { \ket{{{\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}}}_{i}^{} } { \prescript{}{i}{ \bra{{{\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}}} } } } { { \ket{{{\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}}}_{i}^{} } { \prescript{}{i}{ \bra{} } } } } \right)}}\pdv{V_{{n^\prime}}}{E_3}}{\Psi} \nonumber \\ & {\quad}- \mel{\Psi}{{\ensuremath{\left( { \ifthenelse{\equal{}{}}{ { \ket{{{\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}}}_{i}^{} } { \prescript{}{i}{ \bra{{{\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}}} } } } { { \ket{{{\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}}}_{i}^{} } { \prescript{}{i}{ \bra{} } } } } \right)}}\pdv{V_c}{E_3}}{\Psi}\,,\end{aligned}$$ where ${ \ifthenelse{\equal{}{}}{ { \ket{{{\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}}}_{i}^{} } { \prescript{}{i}{ \bra{{{\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}}} } } } { { \ket{{{\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}}}_{i}^{} } { \prescript{}{i}{ \bra{} } } } } \otimes {\mathbbm{1}^{{{\ensuremath{\left(\mathrm{spin}\right)}}}}}$ is abbreviated by ${ \ifthenelse{\equal{}{}}{ { \ket{{{\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}}}_{i}^{} } { \prescript{}{i}{ \bra{{{\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}}} } } } { { \ket{{{\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}}}_{i}^{} } { \prescript{}{i}{ \bra{} } } } }$. The first term is the one-body probability density we already computed. In case of the other terms we can generalize the expression we have to evaluate a bit. We consider the calculation of $$\label{eq:def_O_i_dens} O_i{{\ensuremath{\left({{\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}}\right)}}} \coloneqq \mel{\Psi}{{\ensuremath{\left( { \ifthenelse{\equal{}{}}{ { \ket{{{\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}}}_{i}^{} } { \prescript{}{i}{ \bra{{{\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}}} } } } { { \ket{{{\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}}}_{i}^{} } { \prescript{}{i}{ \bra{} } } } } \otimes {\mathbbm{1}^{{{\ensuremath{\left(\mathrm{spin}\right)}}}}}\right)}}O_i}{\Psi}\,,$$ where $O_i$ can be either a potential or its derivative. Only the properties these have in common are used in further simplifications: $${ \prescript{}{i}{ \mel{p,q;\Omega}{O_i}{{p^{\prime }},{q^{\prime }};\Omega^\prime}_{i}^{} } } \propto {\delta_{\Omega, \Omega^\prime}} {\delta_{\Omega, \Omega_i}} \frac{{\delta{{\ensuremath{\left(q-{q^{\prime }}\right)}}}}}{q {q^{\prime }}} \,.$$ This enables us the reuse the result for the calculation of “potential energy densities"[^4] $\mel{\Psi}{{\ensuremath{\left( { \ifthenelse{\equal{}{}}{ { \ket{{{\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}}}_{i}^{} } { \prescript{}{i}{ \bra{{{\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}}} } } } { { \ket{{{\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}}}_{i}^{} } { \prescript{}{i}{ \bra{} } } } } \right)}}{V_i}}{\Psi}$. Nevertheless, we want to give on this occasion the full matrix elements of the two-body potentials $V_i$ when embedded in a three-body system: $$\begin{aligned} { \prescript{}{i}{ \mel{p,q;\Omega}{V_i{{\ensuremath{\left(E_3\right)}}}}{{p^{\prime }},{q^{\prime }};\Omega^\prime}_{i}^{} } } &= 4\pi {g_{l_i}{{\ensuremath{\left(p\right)}}}} \lambda_i{\left(E_3 - \frac{q^2}{2 \mu_{i{\ensuremath{\left(jk\right)}}}}\right)} {g_{l_i}{{\ensuremath{\left({p^{\prime }}\right)}}}} \nonumber \\ & {\quad \cross}{\delta_{\Omega, \Omega^\prime}} {\delta_{\Omega, \Omega_i}} \frac{{\delta{{\ensuremath{\left(q-{q^{\prime }}\right)}}}}}{q^2}\,.\end{aligned}$$ This expression also directly determines the matrix elements of $V_i^\prime$. Our expressions for the different densities will be expressed using the formula $$\begin{aligned} \begin{aligned} O_i{{\ensuremath{\left({{\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}}\right)}}} &\coloneqq \mel{\Psi}{{\ensuremath{\left( { \ifthenelse{\equal{}{}}{ { \ket{{{\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}}}_{i}^{} } { \prescript{}{i}{ \bra{{{\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}}} } } } { { \ket{{{\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}}}_{i}^{} } { \prescript{}{i}{ \bra{} } } } } \otimes {\mathbbm{1}^{{{\ensuremath{\left(\mathrm{spin}\right)}}}}}\right)}}O_i}{\Psi} \\ &= \bigg[ {\sum_{L=0}^1 \sum_{M=-L}^{L} } c_{L,M}^2 {\ensuremath{\left( a_i a_i^\prime \left| {\mathcal{Y}_{11}^{LM}{{\ensuremath{\left({\Upsilon}_i\right)}}}} \right|^2 + \tilde{a}_i \tilde{a}_i^\prime \left| {\mathcal{Y}_{11}^{LM}{{\ensuremath{\left(\tilde{{\Upsilon}}_i\right)}}}} \right|^2 \right)}} + \frac{d_i d_i^\prime}{{\ensuremath{\left(4\pi\right)}}^2} \\ & {\quad}- {\ensuremath{\left(a_i \tilde{a}_i^\prime + \tilde{a}_i a_i^\prime\right)}} {\sum_{L=0}^1 \sum_{M=-L}^{L} } c_{L,M}^2 {\ensuremath{\left(-1\right)}}^{1-L} \Re{\ensuremath{\left({\mathcal{Y}_{11}^{LM}{{\ensuremath{\left(\tilde{{\Upsilon}}_i\right)}}}}^* {\mathcal{Y}_{11}^{LM}{{\ensuremath{\left({\Upsilon}_i\right)}}}}\right)}} \\ & {\quad}- {\ensuremath{\left(\tilde{a}_i d_i^\prime + d_i \tilde{a}_i^\prime\right)}} \frac{1}{4\pi} c_{0,0} {\mathcal{Y}_{11}^{00}{{\ensuremath{\left(\tilde{{\Upsilon}}_i\right)}}}} -{\ensuremath{\left(a_i d_i^\prime + d_i a_i^\prime\right)}} \frac{1}{4\pi} c_{0,0} {\mathcal{Y}_{11}^{00}{{\ensuremath{\left({\Upsilon}_i\right)}}}} \bigg]\,. \end{aligned}\end{aligned}$$ We used that $a_i$, $a_i^\prime$, $\tilde{a}_i$, $\tilde{a}_i^\prime$, $d_i$ and $d_i^\prime$ are real. For the coefficients $a_i$, $\tilde{a}_i$ and $d_i$ as well as for ${\Upsilon}_i$ and $\tilde{{\Upsilon}}_i$ Eqs. (\[eq:pbd\_coeff\_n\_first\]) to (\[eq:pbd\_coeff\_c\_last\]) hold. The coefficients $a^\prime_i$, $\tilde{a}^\prime_i$ and $d^\prime_i$ involve matrix elements of $O_i$. In order to evaluate ${ \prescript{}{i}{ \mel{p,q;\Omega_i}{O_i}{\psi_j}_{}^{} } }$ in the case $j \neq i$ a recoupling in the partial wave basis has to be made. Due to the properties of $\ket{\psi_j}$ the relation $$\begin{aligned} { \prescript{}{i}{ \mel{p,q;\Omega_i}{O_i}{\psi_j}_{}^{} } } &= {\int \dd{{p^{\prime }}} {p^{\prime 2}}} { \prescript{}{i}{ \mel{p,q;\Omega_i}{O_i}{{p^{\prime }},q;\Omega_i}_{i}^{} } } \nonumber \\ & {\quad}\cross {\int \dd{{\tilde{p}^{\prime }}} {\tilde{p}^{\prime 2}}} {\int \dd{{\tilde{q}^{\prime }}} {\tilde{q}^{\prime 2}}} { \prescript{}{i}{ \braket{{p^{\prime }},q;\Omega_i}{{\tilde{p}^{\prime }},{\tilde{q}^{\prime }};\Omega_j}_{j}^{} } } { \prescript{}{j}{ \braket{{\tilde{p}^{\prime }},{\tilde{q}^{\prime }};\Omega_j}{\psi_j}_{}^{} } }\end{aligned}$$ holds. Accordingly overlaps of partial wave states with different spectators have to be evaluated. Helpful identities are given in \[subsec:ov\_pwb\]. With these the following relations for $O_n$ can be derived: $$\begin{aligned} a_n^\prime &= {\int \dd{{p^{\prime }}} {p^{\prime 2}}} { \prescript{}{n}{ \mel{p,q;\Omega_n}{O_n}{{p^{\prime }},q;\Omega_n}_{n}^{} } } \psi_n{{\ensuremath{\left({p^{\prime }},q\right)}}} \nonumber \\ & {\quad}+ {\int \dd{{p^{\prime }}} {p^{\prime 2}}} { \prescript{}{n}{ \mel{p,q;\Omega_n}{O_n }{{p^{\prime }},q;\Omega_n}_{n}^{} } } \nonumber \\ & {\quad}{\quad}\cross {\int \dd{\Omega_{{\ensuremath{\boldsymbol{{p^{\prime }}}}}}}} {\int \dd{\Omega_{{\ensuremath{\boldsymbol{{\tilde{q}}}}}}}} \sum_{L=0}^1 \sum_{M=-L}^{L} {\ensuremath{\left(-1\right)}}^{L} \frac{2^{1-L}}{6L+3} \nonumber \\ & {\quad}{\quad}\cross {\ensuremath{\left({\mathcal{Y}_{11}^{LM}{{\ensuremath{\left({\ensuremath{\boldsymbol{{p^{\prime }}}}},{\ensuremath{\boldsymbol{{\tilde{q}}}}}\right)}}}}\right)}}^* {\mathcal{Y}_{11}^{LM}{{\ensuremath{\left({{\ensuremath{\boldsymbol{\kappa}}}_{nnp}^{\prime}},{{\ensuremath{\boldsymbol{\kappa}}}_{nnq}^{\prime}}\right)}}}} \psi_n{{\ensuremath{\left({\kappa_{nnp}^\prime},{\kappa^\prime_{nnq}}\right)}}} \nonumber \\ & {\quad}+ \frac{ 1 }{\sqrt{2}} {\int \dd{{p^{\prime }}} {p^{\prime 2}}} { \prescript{}{n}{ \mel{p,q;\Omega_n}{O_n}{{p^{\prime }},q;\Omega_n}_{n}^{} } } \nonumber \\ & {\quad}{\quad}\cross \int_{0}^{\pi} \dd{\theta_{{\ensuremath{\boldsymbol{{p^{\prime }}}}},{\ensuremath{\boldsymbol{{\tilde{q}}}}}}} \sin{\theta_{{\ensuremath{\boldsymbol{{p^{\prime }}}}},{\ensuremath{\boldsymbol{{\tilde{q}}}}}}} \cos{\theta_{{\ensuremath{\boldsymbol{{p^{\prime }}}}},{\ensuremath{\boldsymbol{{\tilde{q}}}}}}} \psi_c{{\ensuremath{\left({\kappa_{ncp}},{\kappa_{ncq}}\right)}}}\,. \\ \tilde{a}_n^\prime &= 0\,, \\ d_n^\prime &= 0\,.\end{aligned}$$ Note that the arguments of the functions $\kappa_{ijk}$ are omitted. They are functions of the momenta ${\ensuremath{\boldsymbol{{p^{\prime }}}}}$ and ${\ensuremath{\boldsymbol{{\tilde{q}}}}}$. The vector ${\ensuremath{\boldsymbol{{\tilde{q}}}}}$ has the same absolute value as ${\ensuremath{\boldsymbol{q}}}$, it is introduced since ${\ensuremath{\boldsymbol{q}}}$ is already in use. For the angular integration ${\int \dd{\Omega_{{\ensuremath{\boldsymbol{{p^{\prime }}}}}}}} {\int \dd{\Omega_{{\ensuremath{\boldsymbol{{\tilde{q}}}}}}}}$ in the expression above the identities given in appendix \[subsec:ang\_simp\] can be used, so that only one angular integral has to be carried out numerically. In order to calculate $O_c$ one needs: $$\begin{aligned} a_c^\prime &= 0\,, \\ \tilde{a}_c^\prime &= 0\,, \\ d_c^\prime &= {\int \dd{{p^{\prime }}} {p^{\prime 2}}} { \prescript{}{c}{ \mel{p,q;\Omega_c}{O_c}{{p^{\prime }},q;\Omega_c}_{c}^{} } } \psi_c{{\ensuremath{\left({p^{\prime }},q\right)}}} \nonumber \\ & {\quad}+ \frac{ 1 }{\sqrt{2}} {\int \dd{{p^{\prime }}} {p^{\prime 2}}} { \prescript{}{c}{ \mel{p,q;\Omega_c}{O_c}{{p^{\prime }},q;\Omega_c}_{c}^{} } } \nonumber \\ & {\quad}{\quad}\cross \int_{-1}^{1} \dd{ \cos{\theta_{{\ensuremath{\boldsymbol{{p^{\prime }}}}},{\ensuremath{\boldsymbol{{\tilde{q}}}}}}} } \cos{\theta_{{{\ensuremath{\boldsymbol{\kappa}}}_{cnp}}, {{\ensuremath{\boldsymbol{\kappa}}}_{cnq}}}} \psi_n{{\ensuremath{\left({\kappa_{cnp}},{\kappa_{cnq}}\right)}}} \nonumber \\ & {\quad}+ \frac{ 1 }{\sqrt{2}} {\int \dd{{p^{\prime }}} {p^{\prime 2}}} { \prescript{}{c}{ \mel{p,q;\Omega_c}{O_c}{{p^{\prime }},q;\Omega_c}_{c}^{} } } \nonumber \\ & {\quad}{\quad}\cross \int_{-1}^{1} \dd{ \cos{\theta_{{\ensuremath{\boldsymbol{{p^{\prime }}}}},{\ensuremath{\boldsymbol{{\tilde{q}}}}}}} } \cos{\theta_{{{\ensuremath{\boldsymbol{\kappa}}}_{cnp}^{\prime}}, {{\ensuremath{\boldsymbol{\kappa}}}_{cnq}^{\prime}}}} \psi_n{{\ensuremath{\left({\kappa_{cnp}^\prime},{\kappa^\prime_{cnq}}\right)}}}\end{aligned}$$ Also in this case $\kappa_{ijk}$ are functions of ${\ensuremath{\boldsymbol{{p^{\prime }}}}}$ and ${\ensuremath{\boldsymbol{{\tilde{q}}}}}$. Again, $|{\ensuremath{\boldsymbol{{\tilde{q}}}}}| = |{\ensuremath{\boldsymbol{q}}}|$ holds. The density $O_{{n^\prime}}$ can be obtained with $$\begin{aligned} a^\prime_{{n^\prime}} &= 0\,, \\ \tilde{a}^\prime_{{n^\prime}} &= \Bigg( {\int \dd{{p^{\prime }}} {p^{\prime 2}}} { \prescript{}{n}{ \mel{p,q;\Omega_n}{O_n}{{p^{\prime }},q;\Omega_n}_{n}^{} } } \psi_n{{\ensuremath{\left({p^{\prime }},q\right)}}} \nonumber \\ & {\quad}+ {\int \dd{{p^{\prime }}} {p^{\prime 2}}} { \prescript{}{n}{ \mel{p,q;\Omega_n}{O_n }{{p^{\prime }},q;\Omega_n}_{n}^{} } } \nonumber \\ & {\quad}{\quad}\cross {\int \dd{\Omega_{{\ensuremath{\boldsymbol{{p^{\prime }}}}}}}} {\int \dd{\Omega_{{\ensuremath{\boldsymbol{{\tilde{q}}}}}}}} \sum_{L=0}^1 \sum_{M=-L}^{L} {\ensuremath{\left(-1\right)}}^{L} \frac{2^{1-L}}{6L+3} \nonumber \\ & {\quad}{\quad}\cross {\ensuremath{\left({\mathcal{Y}_{11}^{LM}{{\ensuremath{\left({\ensuremath{\boldsymbol{{p^{\prime }}}}},{\ensuremath{\boldsymbol{{\tilde{q}}}}}\right)}}}}\right)}}^* {\mathcal{Y}_{11}^{LM}{{\ensuremath{\left({{\ensuremath{\boldsymbol{\kappa}}}_{nnp}^{\prime}},{{\ensuremath{\boldsymbol{\kappa}}}_{nnq}^{\prime}}\right)}}}} \psi_n{{\ensuremath{\left({\kappa_{nnp}^\prime},{\kappa^\prime_{nnq}}\right)}}} \nonumber \\ & {\quad}+ \frac{ 1 }{\sqrt{2}} {\int \dd{{p^{\prime }}} {p^{\prime 2}}} { \prescript{}{n}{ \mel{p,q;\Omega_n}{O_n}{{p^{\prime }},q;\Omega_n}_{n}^{} } } \nonumber \\ & {\quad}{\quad}\cross \int_{-1}^{1} \dd{ \cos{\theta_{{\ensuremath{\boldsymbol{{p^{\prime }}}}},{\ensuremath{\boldsymbol{{\tilde{q}}}}}}} } \cos{\theta_{{\ensuremath{\boldsymbol{{p^{\prime }}}}},{\ensuremath{\boldsymbol{{\tilde{q}}}}}}} \psi_c{{\ensuremath{\left({\kappa_{ncp}},{\kappa_{ncq}}\right)}}} \Bigg) \Bigg \rvert_{p={\kappa_{nnp}},q={\kappa_{nnq}}}\,, \\ d^\prime_{{n^\prime}} &= 0\,.\end{aligned}$$ As usual, the $\kappa_{ijk}$ are functions of the momenta ${\ensuremath{\boldsymbol{{p^{\prime }}}}}$ and ${\ensuremath{\boldsymbol{{\tilde{q}}}}}$. The vector ${\ensuremath{\boldsymbol{{\tilde{q}}}}}$ has the same absolute value as ${\ensuremath{\boldsymbol{q}}}$. The only difference is now, that the whole term $\tilde{a}^\prime_{{n^\prime}}$ is evaluated at $p={\kappa_{nnp}},q={\kappa_{nnq}}$ as indicated at the end of the expression. Again, the angular integration ${\int \dd{\Omega_{{\ensuremath{\boldsymbol{{p^{\prime }}}}}}}} {\int \dd{\Omega_{{\ensuremath{\boldsymbol{{\tilde{q}}}}}}}}$ can be simplified using the remarks given in appendix \[subsec:ang\_simp\]. In this last case we deviate a bit from the definition given in Eq. (\[eq:def\_O\_i\_dens\]). The index for the spectator and the one of the operator $O$ are not equal in this case. The correct definition for this exceptional case reads $$\label{eq:def_O_i_dens_np} O_{{n^\prime}}{{\ensuremath{\left({{\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}}\right)}}} \coloneqq \mel{\Psi}{{\ensuremath{\left( { \ifthenelse{\equal{}{}}{ { \ket{{{\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}}}_{n}^{} } { \prescript{}{n}{ \bra{{{\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}}} } } } { { \ket{{{\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}}}_{n}^{} } { \prescript{}{n}{ \bra{} } } } } \otimes {\mathbbm{1}^{{{\ensuremath{\left(\mathrm{spin}\right)}}}}}\right)}}O_{{n^\prime}}}{\Psi}\,.$$ The expressions for the calculation of $O_i{{\ensuremath{\left({{\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}}\right)}}}$ can be tested by considering the cases of the potentials themselves, where there is another method for the calculation of $V_i{{\ensuremath{\left({{\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}}\right)}}}$. From the definition of the Faddeev wave function components in Eq. (\[eq:def\_fd\_wfc\]) the relation $$\label{eq:V_i_Psi} V_i \ket{\Psi} = G_0^{-1} \ket{\psi_i}$$ follows. Since $$\begin{aligned} { \prescript{}{i}{ \mel{{{\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}}}{G_0^{-1}}{{{\ensuremath{\boldsymbol{{p^{\prime }}}}},{\ensuremath{\boldsymbol{{q^{\prime }}}}}}}_{i}^{} } } &= {\ensuremath{\left(E_3 - p^2/{\ensuremath{\left(2\mu_{jk}\right)}} - q^2/{\ensuremath{\left(2\mu_{i(jk)}\right)}} \right)}} {\delta^{(3)}{{\ensuremath{\left({\ensuremath{\boldsymbol{p}}}-{\ensuremath{\boldsymbol{{p^{\prime }}}}}\right)}}}} {\delta^{(3)}{{\ensuremath{\left({\ensuremath{\boldsymbol{q}}}-{\ensuremath{\boldsymbol{{q^{\prime }}}}}\right)}}}}\end{aligned}$$ holds, the calculation of $V_i{{\ensuremath{\left({{\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}}\right)}}}$ using this method requires only simple modifications of the expression for the one-body probability density given in Eq. (\[eq:ob\_pbd\]). Note that, if the three-body force is included as described in subsection \[subsec:heft\], Eq. (\[eq:V\_i\_Psi\]) is only true at vanishing three-body couplings. We note also that, if the two-body probability density is calculated using the formulas of this section, then for some terms transformations to a Jacobi-momentum basis corresponding to a different spectator are required. Results for the probability densities {#subsec:results} ------------------------------------- The results were obtained with the parameters $B_3^{{\ensuremath{\left(0\right)}}} = 0.975$MeV from [@brodeur12], $a_{nn} = -18.7$fm from [@trotter06], $k_R = 37.4533$MeV and $r_1 = -174.0227$MeV. The latter two were calculated using the results of [@arndt73]. Initially, we show results obtained with [Heaviside form factor]{}s, where we set $\beta_0=\beta_1=\Lambda$ with the three-body cutoff $\Lambda$. Different angular projections done according to Eq. (\[eq:rhotpw\]) are shown in Fig. \[fig:mn\_pbds\]. The color plot of each of the two subfigures shows the probability density as function of the momenta $p$ and $q$. On the left and below these main plots usual 2d-plots are shown. They depict cuts of the probability density. The left plots show the probability density as function of $p$ at fixed $q$, whereas the plots below the color plots show it as function of $q$ at fixed $p$. For these fixed vaules a value of order (but above) the breakdown scale $M_\mathrm{core} \approx 140$MeV and a value of order of the dineutron binding momentum $\sqrt{2\mu_{c{\ensuremath{\left(nn\right)}}} S_{2n}} \approx \sqrt{8/3 \cross 940 \textrm{\,MeV}^2} \approx 50$MeV are chosen. We estimate that the numerical uncertainties[^5] of our result for the $\xi=0$ and $\xi=1$ projection of the (two-body) probability density are smaller than $1\%$. ![Projections on $\xi=0$ (upper panel) and $\xi=1$ (lower panel) of the approximation to the probability density ${\ensuremath{ \tilde{\rho}_{n}{{\ensuremath{\left({{\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}}\right)}}} }}p^2q^2$ are shown. These results were obtained with $\Lambda=750$MeV. Nevertheless, we plot up to $650$MeV in order to show the negative regions.[]{data-label="fig:mn_pbds"}](std_ff_pd_mn_n_0.pdf "fig:"){width="75.00000%"}\ ![Projections on $\xi=0$ (upper panel) and $\xi=1$ (lower panel) of the approximation to the probability density ${\ensuremath{ \tilde{\rho}_{n}{{\ensuremath{\left({{\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}}\right)}}} }}p^2q^2$ are shown. These results were obtained with $\Lambda=750$MeV. Nevertheless, we plot up to $650$MeV in order to show the negative regions.[]{data-label="fig:mn_pbds"}](std_ff_pd_mn_n_1.pdf "fig:"){width="75.00000%"} The upper panel of Fig. \[fig:mn\_pbds\] shows that the corrections cause negative probabilities at momenta higher than the high-momentum scale of this EFT $M_\mathrm{core} \approx 140$MeV, i.e. in regions where we cannot trust the result anyway. Indeed, the full probability density is positive semi-definite, so the appearance of negative values for ${\ensuremath{ \tilde{\rho}_{n}^{{\ensuremath{\left(0\right)}}}{{\ensuremath{\left({{\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}}\right)}}} }}p^2q^2$ in certain regions signals that the approximation we made to evaluate it has failed there. A contrast is provided by the lower panel of Fig. \[fig:mn\_pbds\] where ${\ensuremath{ \tilde{\rho}_{n}^{{\ensuremath{\left(1\right)}}}{{\ensuremath{\left({{\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}}\right)}}} }}p^2q^2$ is negative in the low-momentum regime. In this case the negative values do not signal a breakdown of the approximation we made to evaluate $\tilde{\rho}$. This projection of the probability density contains interferences between different partial waves in the wave function, and so does not have to be non-negative. That our approximation is valid, i.e. the correction is small a low momenta, can be seen from Fig. \[fig:pbd\_div\], where the quotients of the two-body and one-body probability densities are shown next to plots of the two-body probability density. This is done for $\xi=0$ and $\xi=1$. The normalization constant changes due to the modificiation of the probability density and contains probability values from regions where we do not trust the density anymore. Therefore the actual values of the quotient shown in the left panel do not contain the key point. The point is that the quotient varies very little in the domain of validity of the EFT. Accordingly the shape of the probability density is the same with and without the two-body contributions, and so they are small in this sense. ![The upper left plot shows the probability density ${\ensuremath{ \tilde{\rho}_{n}^{{\ensuremath{\left(0\right)}}}{{\ensuremath{\left(p,q\right)}}} }}p^2 q^2$, the upper right shows the quotient of probability densities ${\ensuremath{ \tilde{\rho}_{n}^{{\ensuremath{\left(0\right)}}}{{\ensuremath{\left(p,q\right)}}} }}/{\ensuremath{ \rho_{n}^{{\ensuremath{\left(0\right)}}}{{\ensuremath{\left(p,q\right)}}} }}$. The lower left plot shows ${\ensuremath{ \tilde{\rho}_{n}^{{\ensuremath{\left(1\right)}}}{{\ensuremath{\left(p,q\right)}}} }}p^2 q^2$, the lower right shows ${\ensuremath{ \tilde{\rho}_{n}^{{\ensuremath{\left(1\right)}}}{{\ensuremath{\left(p,q\right)}}} }}/{\ensuremath{ \rho_{n}^{{\ensuremath{\left(1\right)}}}{{\ensuremath{\left(p,q\right)}}} }}$. All were obtained with $\Lambda=750$MeV.[]{data-label="fig:pbd_div"}](std_ff_pd_mn_n_0_d_std.pdf "fig:"){width="49.00000%"} ![The upper left plot shows the probability density ${\ensuremath{ \tilde{\rho}_{n}^{{\ensuremath{\left(0\right)}}}{{\ensuremath{\left(p,q\right)}}} }}p^2 q^2$, the upper right shows the quotient of probability densities ${\ensuremath{ \tilde{\rho}_{n}^{{\ensuremath{\left(0\right)}}}{{\ensuremath{\left(p,q\right)}}} }}/{\ensuremath{ \rho_{n}^{{\ensuremath{\left(0\right)}}}{{\ensuremath{\left(p,q\right)}}} }}$. The lower left plot shows ${\ensuremath{ \tilde{\rho}_{n}^{{\ensuremath{\left(1\right)}}}{{\ensuremath{\left(p,q\right)}}} }}p^2 q^2$, the lower right shows ${\ensuremath{ \tilde{\rho}_{n}^{{\ensuremath{\left(1\right)}}}{{\ensuremath{\left(p,q\right)}}} }}/{\ensuremath{ \rho_{n}^{{\ensuremath{\left(1\right)}}}{{\ensuremath{\left(p,q\right)}}} }}$. All were obtained with $\Lambda=750$MeV.[]{data-label="fig:pbd_div"}](std_ff_div_pd_n_0_d_std.pdf "fig:"){width="49.00000%"}\ ![The upper left plot shows the probability density ${\ensuremath{ \tilde{\rho}_{n}^{{\ensuremath{\left(0\right)}}}{{\ensuremath{\left(p,q\right)}}} }}p^2 q^2$, the upper right shows the quotient of probability densities ${\ensuremath{ \tilde{\rho}_{n}^{{\ensuremath{\left(0\right)}}}{{\ensuremath{\left(p,q\right)}}} }}/{\ensuremath{ \rho_{n}^{{\ensuremath{\left(0\right)}}}{{\ensuremath{\left(p,q\right)}}} }}$. The lower left plot shows ${\ensuremath{ \tilde{\rho}_{n}^{{\ensuremath{\left(1\right)}}}{{\ensuremath{\left(p,q\right)}}} }}p^2 q^2$, the lower right shows ${\ensuremath{ \tilde{\rho}_{n}^{{\ensuremath{\left(1\right)}}}{{\ensuremath{\left(p,q\right)}}} }}/{\ensuremath{ \rho_{n}^{{\ensuremath{\left(1\right)}}}{{\ensuremath{\left(p,q\right)}}} }}$. All were obtained with $\Lambda=750$MeV.[]{data-label="fig:pbd_div"}](std_ff_pd_mn_n_1_d_std.pdf "fig:"){width="49.00000%"} ![The upper left plot shows the probability density ${\ensuremath{ \tilde{\rho}_{n}^{{\ensuremath{\left(0\right)}}}{{\ensuremath{\left(p,q\right)}}} }}p^2 q^2$, the upper right shows the quotient of probability densities ${\ensuremath{ \tilde{\rho}_{n}^{{\ensuremath{\left(0\right)}}}{{\ensuremath{\left(p,q\right)}}} }}/{\ensuremath{ \rho_{n}^{{\ensuremath{\left(0\right)}}}{{\ensuremath{\left(p,q\right)}}} }}$. The lower left plot shows ${\ensuremath{ \tilde{\rho}_{n}^{{\ensuremath{\left(1\right)}}}{{\ensuremath{\left(p,q\right)}}} }}p^2 q^2$, the lower right shows ${\ensuremath{ \tilde{\rho}_{n}^{{\ensuremath{\left(1\right)}}}{{\ensuremath{\left(p,q\right)}}} }}/{\ensuremath{ \rho_{n}^{{\ensuremath{\left(1\right)}}}{{\ensuremath{\left(p,q\right)}}} }}$. All were obtained with $\Lambda=750$MeV.[]{data-label="fig:pbd_div"}](std_ff_div_pd_n_1_d_std.pdf "fig:"){width="49.00000%"}\ In the left panel of Fig. \[fig:reg\_cmp\] we compare our results for the two-body probability density at low momenta for different cutoffs. Due to the modifications resulting from the energy-dependence of the potentials the normalization constant is in our case cutoff-dependent. Thus again the point is that the quotient varies little in the region where the EFT is valid. The shape of the probability density in this region is therefore, to a good approximation, cutoff independent, even though the wave function normalization is not. This observation is also true in case of ${\ensuremath{ \tilde{\rho}_{n}^{{\ensuremath{\left(1\right)}}}{{\ensuremath{\left(p,q\right)}}} }}p^2 q^2$. ![ The left plot shows the quotient of the probability density [$ \tilde{\rho}_{n}^{{\ensuremath{\left(0\right)}}}{{\ensuremath{\left(p,q\right)}}} $]{} obtained with $\Lambda = 1500$MeV and the identical quantity obtained with $\Lambda = 750$MeV (both with [Heaviside form factor]{}). The right plot shows the the quotient of [$ \tilde{\rho}_{n}^{{\ensuremath{\left(0\right)}}}{{\ensuremath{\left(p,q\right)}}} $]{} obtained with Yamaguchi form factors and [$ \tilde{\rho}_{n}^{{\ensuremath{\left(0\right)}}}{{\ensuremath{\left(p,q\right)}}} $]{} obtained with [Heaviside form factor]{}s. The right plot was obtained with $\Lambda = 750$MeV and in case of Yamaguchi form factors with $\Lambda = 950$MeV and $\beta_0 = \beta_1 = 500$MeV. It may be useful to have the plot of ${\ensuremath{ \tilde{\rho}_{n}^{{\ensuremath{\left(0\right)}}}{{\ensuremath{\left(p,q\right)}}} }} p^2 q^2$ as given in the upper left of Fig. \[fig:pbd\_div\] in mind when looking at the plots of these quotients.[]{data-label="fig:reg_cmp"}](std_ff_div_co_pd_mn_n_0_d_std.pdf "fig:"){width="49.00000%"} ![ The left plot shows the quotient of the probability density [$ \tilde{\rho}_{n}^{{\ensuremath{\left(0\right)}}}{{\ensuremath{\left(p,q\right)}}} $]{} obtained with $\Lambda = 1500$MeV and the identical quantity obtained with $\Lambda = 750$MeV (both with [Heaviside form factor]{}). The right plot shows the the quotient of [$ \tilde{\rho}_{n}^{{\ensuremath{\left(0\right)}}}{{\ensuremath{\left(p,q\right)}}} $]{} obtained with Yamaguchi form factors and [$ \tilde{\rho}_{n}^{{\ensuremath{\left(0\right)}}}{{\ensuremath{\left(p,q\right)}}} $]{} obtained with [Heaviside form factor]{}s. The right plot was obtained with $\Lambda = 750$MeV and in case of Yamaguchi form factors with $\Lambda = 950$MeV and $\beta_0 = \beta_1 = 500$MeV. It may be useful to have the plot of ${\ensuremath{ \tilde{\rho}_{n}^{{\ensuremath{\left(0\right)}}}{{\ensuremath{\left(p,q\right)}}} }} p^2 q^2$ as given in the upper left of Fig. \[fig:pbd\_div\] in mind when looking at the plots of these quotients.[]{data-label="fig:reg_cmp"}](ff_div_pd_mn_n_0_d.pdf "fig:"){width="49.00000%"} So far, all results were obtained by using Heaviside form factors. Finally, we compare results for the two-body probability density obtained with Yamaguchi form factors[^6] to those obtained with Heaviside form factors in the right panel of Fig. \[fig:reg\_cmp\]. This time, the relation between the two-body and three-body cutoffs is $\beta_0 = \beta_1 \approx \Lambda/2$. The shown ratio of $\xi=0$ projections is roughly constant. This is also true for $\xi=1$. This shows that two different regulators in the two-body potential lead to almost the same results for the ${}^6$He wave function. In summary, we have obtained a two-body probability density valid at low momenta. The corrections due to the energy dependence are small in this region. This low-momentum part is approximately regulator independent. Summary and Future Work {#sec:future} ======================= In this work we calculated the probability density of [^6^He]{} in momentum space. Within Halo Effective Field Theory, [^6^He]{} is a three-body $nn{c}$ bound state. At leading order the two-body pairwise interactions are parameterized to reproduce the $nn$ $^1S_0$ scattering length and the $n{c}$ $^2P_{3/2}$ scattering volume and effective range. A three-body force is mandatory for renormalization [@ji14] and is adjusted to reproduce the [^6^He]{} ground-state two-neutron separation energy. The [^6^He]{} ground-state three-body wave function was solved in the Faddeev formalism, with the $nn$ and $n {c}$ potentials adjusted to exactly reproduce the corresponding leading-order amplitudes of Halo EFT. Using partial-wave decomposition, we obtained the Faddeev components, each of which contains only one spin-angular-momentum channel in the corresponding Jacobi spectator representation. However, a single channel in one spectator representation cannot completely describe the three-body total wave function, due to the non-trivial overlaps between spin-angular-momentum eigenstates from two different spectator representations. The presence of energy-dependent potentials in this problem yields a modification of the orthonormality conditions used in calculations of the momentum-space density. In addition to the usual one-body piece of the probability density $\langle \Psi|\Psi\rangle$, there is a new two-body term. This is linked to the energy derivative of the two-body potential, and ensures correct normalization of the probability density when integrated over a set of momentum eigenstates. The resulting momentum-space probability density of [^6^He]{} is a function of the angle between the Jacobi momenta $\vec{p}$ and $\vec{q}$. We expand this dependence in Legendre polynomials and focus mainly on the angle-averaged (zeroth moment) and ${\ensuremath{\boldsymbol{p}}} \cdot {\ensuremath{\boldsymbol{q}}}$ (first moment) in that expansion. By comparing the density obtained with different regulators in the two-body potentials, we see that its low-momentum part is regulator independent. The energy-dependent potential has been commonly used in nuclear reaction theory as well [@Feshbach:1958nx]. In the scattering of nucleons from nuclei optical potentials are often introduced to account for channels of lower energy than the elastic energy, which can remove flux. Optical potentials are energy dependent and, strictly speaking, probabilities calculated using them should include additional two-body terms of the type discussed here. DRP and CJ thank Charlotte Elster for useful discussions during the early stages of this work. DRP thanks Jerry Yang for drawing his attention to Ref. [@mckellar83]. MG thanks Wael Elkamhawy and Fabian Hildenbrand for useful discussions during the early stages of this work. The work of DRP was supported by the US Department of Energy under contract DE-FG02-93ER-40756 and by the ExtreMe Matter Institute EMMI at the GSI Helmholtzzentrum für Schwerionenphysik, Darmstadt, Germany. HWH was supported by the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) under project number 279384907 – SFB 1245 and by the Bundesministerium für Bildung und Forschung (BMBF) through contract 05P18RDFN1. Alternative derivation of the normalization condition {#ap:alternativederivationnorm} ===================================================== In this appendix we give an alternative derivation of the normalization condition in the presence of an energy-dependent potential. This derivation focuses on the resolvent of $H(E)$. It is equivalent to the normalization conditions obtained for two-particle vertex functions in relativistic bound-state equations in Refs. [@lepage77; @adam97]. The resolvent of $H(E)$ is: $$G(E) \coloneqq \frac{1}{E - H(E)}\,.$$ This contrasts with the resolvent of the energy-independent Hamiltonian obtained by evaluating $H(E)$ at $E=E_\alpha$, $\bar{H} \coloneqq H(E_\alpha)$. Denoting that resolvent by $\bar{G}(E)$ we have: $$\bar{G}(E) \coloneqq \frac{1}{E-\bar{H}}\,.$$ By construction, $\bar{H}$ and $H(E)$ both have $|\psi_\alpha \rangle$ as a right eigenstate corresponding to energy $E_\alpha$. The rest of $\bar{H}$’s spectrum will, however, be different. And, while the eigenstate $|\psi_\alpha \rangle$ is common to the spectrum of both operators, it is not guaranteed that it should be normalized in the same way. Indeed, we will show here that it should not be. Expanding $H(E)$ around $E=E_\alpha$ yields: $$G(E)=\frac{1}{E-\bar{H} - (E-E_\alpha) V'(E_\alpha) - \ldots}\,,$$ where $'$ denotes differentiation with respect to $E$. This, in turn, allows us to write: $$G(E)=\bar{G}(E)\left\{1 + \sum_{n=1}^\infty [V'(E_\alpha) (E-E_\alpha) \bar{G}(E)]^n\right\} + \mbox{regular as $E \rightarrow E_\alpha$}\,. \label{eq:resumming}$$ Since $\bar{G}(E)$ is the resolvent of an energy-independent Hermitian operator it has a standard spectral representation. If $E_\alpha$ is an isolated bound state then that is: $$\bar{G}(E)=\frac{|\bar{\psi}_\alpha \rangle \langle \bar{\psi}_\alpha|}{E-E_\alpha} + \mbox{pieces in the space orthogonal to $| \bar{\psi}_\alpha \rangle$}\,. \label{eq:spectral}$$ where we have used $|\bar{\psi}_\alpha \rangle$ to emphasize that this eigenstates of $\bar{H}$ obey the standard normalization condition $\langle \bar{\psi}_\beta|\bar{\psi}_\alpha \rangle=\delta_{\beta \alpha}$, since $\bar{H}$ is an energy-independent, Hermitian, operator. As discussed at length in Sec. \[sec:energydeppotentials\] the states $|\psi_\alpha \rangle$ do not obey this condition, although or $\beta \neq \alpha$ this is not a surprise, since the spectrum of $\bar{H}$ differs from that of $H$ apart from the one state at $E=E_\alpha$. But for that particular state we must write: $${\cal Z}^{1/2} |\psi_\alpha \rangle=|\bar{\psi}_\alpha \rangle\,,$$ with ${\cal Z}$ an additional wave function renormalization associated with the energy dependence of $V(E)$. To fix ${\cal Z}$ we insert Eq. (\[eq:spectral\]) in Eq. (\[eq:resumming\]) and sum the geometric series. This yields: $$G(E)=\frac{|\bar{\psi}_\alpha \rangle \langle \bar{\psi}_\alpha|}{E-E_\alpha} \frac{1}{1 - \langle \bar{\psi}_\alpha| V'(E_\alpha))|\bar{\psi}_\alpha \rangle} + \mbox{regular as $E \rightarrow E_\alpha$}\,. \label{eq:GofEspectral}$$ The factor in the denominator here ensures that the consistency condition: $$G'(E)=-G(E) \left[\frac{\partial}{\partial E} G^{-1}(E)\right] G(E)\,.$$ is satisfied. The spectral decomposition (\[eq:GofEspectral\]) will then take the standard form: $$G(E)=\frac{|\psi_\alpha \rangle \langle \psi_\alpha|}{E-E_\alpha} + \mbox{regular as $E \rightarrow E_\alpha$}\,,$$ provided we identify: $${\cal Z}=1 - \langle \bar{\psi}_\alpha| V'(E_\alpha)|\bar{\psi}_\alpha\rangle=\langle \bar{\psi}_\alpha|{\mathbbm{1}}- V'(E_\alpha)|\bar{\psi}_\alpha \rangle\,. \label{eq:Z}$$ In practice this means that we solve the Hamiltonian eigenvalue problem at $E=E_\alpha$, thereby obtaining an eigenvector that is, prior to normalization, proportional to both $|\bar{\psi}_\alpha \rangle$ and $|\psi_\alpha \rangle$. Then, if we wish to compute the latter, we must normalize such that: $$\langle \psi|{\mathbbm{1}}- V'(E_\alpha)|\psi \rangle=1\,,$$ instead of using the standard $$\langle \psi|\psi \rangle=1\,.$$ Additional formulas for the probability density =============================================== Transformation of Jacobi coordinates {#subsec:jacobi} ------------------------------------ For the evaluation of overlaps of the type ${ \prescript{}{i}{ \braket{{{\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}}}{{{\ensuremath{\boldsymbol{{p^{\prime }}}}},{\ensuremath{\boldsymbol{{q^{\prime }}}}}}}_{j}^{} } }$ the following functions are useful, since they abbreviate many expressions: $$\begin{aligned} {{\ensuremath{\boldsymbol{\pi}}}_{1}{{\ensuremath{\left({\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}\right)}}}} &\coloneqq {\ensuremath{\boldsymbol{p}}} + \frac{A}{A+1} {\ensuremath{\boldsymbol{q}}}\,,\\ {{\ensuremath{\boldsymbol{\pi}}}_{2}{{\ensuremath{\left({\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}\right)}}}} &\coloneqq {\ensuremath{\boldsymbol{p}}} + \frac{1}{2} {\ensuremath{\boldsymbol{q}}}\,,\\ {{\ensuremath{\boldsymbol{\pi}}}_{3}{{\ensuremath{\left({\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}\right)}}}} &\coloneqq {\ensuremath{\boldsymbol{p}}} + \frac{1}{A+1} {\ensuremath{\boldsymbol{q}}}\,.\end{aligned}$$ The transformations can be summarized by the expression $${ \prescript{}{i}{ \braket{{{\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}}}{{{\ensuremath{\boldsymbol{{p^{\prime }}}}},{\ensuremath{\boldsymbol{{q^{\prime }}}}}}}_{j}^{} } } \eqqcolon {\delta^{(3)}{{\ensuremath{\left({\ensuremath{\boldsymbol{{p^{\prime }}}}}-\kappa_{ijp}{{\ensuremath{\left({{\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}}\right)}}}\right)}}}} \, {\delta^{(3)}{{\ensuremath{\left({\ensuremath{\boldsymbol{{q^{\prime }}}}}-\kappa_{ijq}{{\ensuremath{\left({{\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}}\right)}}}\right)}}}} \,,$$ which defines ${\ensuremath{\boldsymbol{\kappa}}}_{ijk}{{\ensuremath{\left({{\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}}\right)}}}$ with $i,j \in \{n,c\}$, $i \neq j$ and $k \in \{p,q\}$. The functions ${\ensuremath{\boldsymbol{\kappa}}}_{ijk}$ are given by $$\begin{aligned} {{\ensuremath{\boldsymbol{\kappa}}}_{ncp}}&\coloneqq {{\ensuremath{\boldsymbol{\pi}}}_{2}{{\ensuremath{\left({\ensuremath{\boldsymbol{q}}}, -{{\ensuremath{\boldsymbol{\pi}}}_{1}{{\ensuremath{\left({\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}\right)}}}}\right)}}}}\,, \\ {{\ensuremath{\boldsymbol{\kappa}}}_{ncq}}&\coloneqq -{{\ensuremath{\boldsymbol{\pi}}}_{1}{{\ensuremath{\left({\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}\right)}}}}\,, \\ {{\ensuremath{\boldsymbol{\kappa}}}_{cnp}}&\coloneqq -{{\ensuremath{\boldsymbol{\pi}}}_{1}{{\ensuremath{\left({\ensuremath{\boldsymbol{q}}},{{\ensuremath{\boldsymbol{\pi}}}_{2}{{\ensuremath{\left({\ensuremath{\boldsymbol{p}}},-{\ensuremath{\boldsymbol{q}}}\right)}}}}\right)}}}}\,, \\ {{\ensuremath{\boldsymbol{\kappa}}}_{cnq}}&\coloneqq {{\ensuremath{\boldsymbol{\pi}}}_{2}{{\ensuremath{\left({\ensuremath{\boldsymbol{p}}},-{\ensuremath{\boldsymbol{q}}}\right)}}}}\,.\end{aligned}$$ Furthermore, we define the functions ${\ensuremath{\boldsymbol{\kappa}}}^\prime_{ijk}{{\ensuremath{\left({{\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}}\right)}}}$ with $i,j \in \{n,c\}$ and $k \in \{p,q\}$ by $${ \prescript{}{i}{ \mel{{{\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}}}{{\mathcal{P}_{nn}^\mathrm{(spatial)}}}{{{\ensuremath{\boldsymbol{{p^{\prime }}}}},{\ensuremath{\boldsymbol{{q^{\prime }}}}}}}_{j}^{} } } \eqqcolon {\delta^{(3)}{{\ensuremath{\left({\ensuremath{\boldsymbol{{p^{\prime }}}}}-{\ensuremath{\boldsymbol{\kappa}}}^\prime_{ijp}{{\ensuremath{\left({{\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}}\right)}}}\right)}}}} \, {\delta^{(3)}{{\ensuremath{\left({\ensuremath{\boldsymbol{{q^{\prime }}}}}-{\ensuremath{\boldsymbol{\kappa}}}^\prime_{ijq}{{\ensuremath{\left({{\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}}\right)}}}\right)}}}} \,,$$ where the spatial part of the $nn$ permutation operator ${\mathcal{P}_{nn}}$ is given by ${\mathcal{P}_{nn}^\mathrm{(spatial)}}$. The used functions of the type ${\ensuremath{\boldsymbol{\kappa}}}^\prime_{ijk}$ read $$\begin{aligned} {{\ensuremath{\boldsymbol{\kappa}}}_{nnp}}^\prime &\coloneqq {{\ensuremath{\boldsymbol{\pi}}}_{3}{{\ensuremath{\left({\ensuremath{\boldsymbol{q}}},{{\ensuremath{\boldsymbol{\pi}}}_{3}{{\ensuremath{\left({\ensuremath{\boldsymbol{p}}},-{\ensuremath{\boldsymbol{q}}}\right)}}}}\right)}}}}\,, \\ {{\ensuremath{\boldsymbol{\kappa}}}_{nnq}}^\prime &\coloneqq {{\ensuremath{\boldsymbol{\pi}}}_{3}{{\ensuremath{\left({\ensuremath{\boldsymbol{p}}},-{\ensuremath{\boldsymbol{q}}}\right)}}}}\,, \\ {{\ensuremath{\boldsymbol{\kappa}}}_{cnp}}^\prime &\coloneqq -{{\ensuremath{\boldsymbol{\pi}}}_{1}{{\ensuremath{\left({\ensuremath{\boldsymbol{q}}},-{{\ensuremath{\boldsymbol{\pi}}}_{2}{{\ensuremath{\left({\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}\right)}}}}\right)}}}}\,, \\ {{\ensuremath{\boldsymbol{\kappa}}}_{cnq}}^\prime &\coloneqq -{{\ensuremath{\boldsymbol{\pi}}}_{2}{{\ensuremath{\left({\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}\right)}}}}\,.\end{aligned}$$ Simplification of certain combinations of coupled spherical harmonics {#subsec:ang_simp} --------------------------------------------------------------------- In this appendix identities simplifying the angular dependence of the probability density are given. The angle $\gamma$ between two vectors ${\ensuremath{\boldsymbol{v}}}_1$ and ${\ensuremath{\boldsymbol{v}}}_2$ is defined by $$\begin{aligned} \cos{\gamma{\ensuremath{\left(\theta_1, \varphi_1, \theta_2, \varphi_2 \right)}}} \coloneqq {\ensuremath{\boldsymbol{\myhat{v}_1}}} \cdot {\ensuremath{\boldsymbol{\myhat{v}_2}}}\, \\ {\ensuremath{\boldsymbol{{v}_i}}} = \begin{pmatrix} \sin{\theta_i} \cos{\varphi_i} \\ \sin{\theta_i} \sin{\varphi_i} \\ \cos{\theta_i} \end{pmatrix}\,.\end{aligned}$$ Expressed in terms of the angles of the two vectors it reads $$\label{eq:cos_gamma} \cos{\gamma{\ensuremath{\left(\theta_1,\varphi_1,\theta_2,\varphi_2\right)}}} \coloneqq \cos{\theta_1} \cos{\theta_2} + \sin{\theta_1} \sin{\theta_2} \cos{{\ensuremath{\left(\varphi_1 - \varphi_2\right)}}}\,.$$ In fact, the angular dependence of the coupled spherical harmonic ${\mathcal{Y}_{11}^{00}}$ is captured entirely by this angle $\gamma$. $$\label{eq:csh_11_00_s} {\mathcal{Y}_{11}^{00}{{\ensuremath{\left(\theta_1, \varphi_1, \theta_2, \varphi_2\right)}}}} = \frac{-\sqrt{3}}{4\pi} \cos{\gamma{\ensuremath{\left(\theta_1, \varphi_1, \theta_2, \varphi_2\right)}}}\,.$$ Also the following combination of coupled spherical harmonics can be purely expressed in terms of relative angles: $$\begin{aligned} \label{eq:csh_11_sum_s} & \sum_{M=-1}^{1} {\mathcal{Y}_{11}^{1M}{{\ensuremath{\left(\theta_1, \varphi_1, \theta_2, \varphi_2\right)}}}} {\ensuremath{\left({\mathcal{Y}_{11}^{1M}{{\ensuremath{\left(\theta_1^\prime, \varphi_1^\prime, \theta_2^\prime, \varphi_2^\prime\right)}}}}\right)}}^* \nonumber \\ & {\quad}= \frac{1}{2} {\ensuremath{\left( \frac{3}{4\pi}\right)}}^2 \bigg( \cos{\gamma{\ensuremath{\left(\theta_1, \varphi_1, \theta_1^\prime, \varphi_1^\prime \right)}}} \cos{\gamma{\ensuremath{\left(\theta_2, \varphi_2, \theta_2^\prime, \varphi_2^\prime\right)}}} \nonumber \\ & {\quad}- \cos{\gamma{\ensuremath{\left(\theta_1, \varphi_1, \theta_2^\prime, \varphi_2^\prime \right)}}} \cos{\gamma{\ensuremath{\left(\theta_1^\prime, \varphi_1^\prime, \theta_2, \varphi_2\right)}}} \bigg) \,.\end{aligned}$$ Although these relative angles are angles between different vectors, it simplifies the probability density a lot. If the identity is applied to our probability density all these relative angles will be only functions of $p$, $q$ and $\theta_{pq}$. Therefore angular integrations over the probability density simplify significantly. Instead of four angular integrals only the one over $\theta_{pq}$ has to be carried out numerically. Numerical calculation of $X_{ij}$ {#subsec:num_Xij} --------------------------------- In the following expressions for the numerical calculation of the $X_{ij}$ are given for any type of form factor. In a first step the expression for $X_{nn}$ is derived using the identity $$\begin{aligned} \label{eq:Xnn_id} &{ \prescript{}{n}{ \mel{p,q;\Omega_n}{-{\mathcal{P}_{nn}}}{{p^{\prime }},{q^{\prime }};\Omega_n}_{n}^{} } } = -{\int \dd{\Omega_{{\ensuremath{\boldsymbol{p}}}}}} {\int \dd{\Omega_{{\ensuremath{\boldsymbol{q}}}}}} {\int \dd{\Omega_{{\ensuremath{\boldsymbol{{p^{\prime }}}}}}}} {\int \dd{\Omega_{{\ensuremath{\boldsymbol{{q^{\prime }}}}}}}} \nonumber \\ &{\quad \cross}{ \prescript{}{n}{ \mel{p,q;\Omega_n}{ {\ensuremath{\left( { \ket{{{\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}}\vphantom{{{\ensuremath{\boldsymbol{{p^{\prime }}}}},{\ensuremath{\boldsymbol{{q^{\prime }}}}}}} }_{n}^{} } { \prescript{}{n}{ \mel{{{\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}}}{{\mathcal{P}_{nn}^\mathrm{(spatial)}}}{{{\ensuremath{\boldsymbol{{p^{\prime }}}}},{\ensuremath{\boldsymbol{{q^{\prime }}}}}}}_{n}^{} } } { \prescript{}{n}{ \bra{{{\ensuremath{\boldsymbol{{p^{\prime }}}}},{\ensuremath{\boldsymbol{{q^{\prime }}}}}}} } } \otimes {\mathcal{P}_{nn}^\mathrm{(spin)}}\right)}} }{{p^{\prime }},{q^{\prime }};\Omega_n}_{n}^{} } }\,.\end{aligned}$$ The calculation yields $$\begin{aligned} &X_{nn}{{\ensuremath{\left(q,{q^{\prime }};E\right)}}} \nonumber \\ & {\quad}= {\int \dd{{p^{}}} {p^{2}}} {\int \dd{{p^{\prime }}} {p^{\prime 2}}} {g_{l_n}{{\ensuremath{\left(p\right)}}}} {G_0^{{\ensuremath{\left(n\right)}}}{{\ensuremath{\left(p,q;E\right)}}}} {g_{l_n}{{\ensuremath{\left({p^{\prime }}\right)}}}} { \prescript{}{n}{ \mel{p,q;\Omega_n}{-{\mathcal{P}_{nn}}}{{p^{\prime }},{q^{\prime }};\Omega_n}_{n}^{} } } \nonumber \\ & {\quad}= - {\int \dd{\Omega_{{\ensuremath{\boldsymbol{p}}}}}} {\int \dd{\Omega_{{\ensuremath{\boldsymbol{q}}}}}} {\int \dd{\Omega_{{\ensuremath{\boldsymbol{{p^{\prime }}}}}}}} {\int \dd{\Omega_{{\ensuremath{\boldsymbol{{q^{\prime }}}}}}}} {g_{l_n}{{\ensuremath{\left({\pi_{3}{{\ensuremath{\left({\ensuremath{\boldsymbol{{q^{\prime }}}}},{\ensuremath{\boldsymbol{{q^{}}}}}\right)}}}}\right)}}}} {G_0^{{\ensuremath{\left(n\right)}}}{{\ensuremath{\left({\pi_{3}{{\ensuremath{\left({\ensuremath{\boldsymbol{{q^{\prime }}}}},{\ensuremath{\boldsymbol{{q^{}}}}}\right)}}}}, q; E\right)}}}} {g_{l_n}{{\ensuremath{\left({\pi_{3}{{\ensuremath{\left({\ensuremath{\boldsymbol{{q^{}}}}},{\ensuremath{\boldsymbol{{q^{\prime }}}}}\right)}}}}\right)}}}} \nonumber \\ & {\quad}{\quad \cross}\sum_{L=0}^1 \sum_{M=-L}^L \frac{ {\ensuremath{\left(-2\right)}}^{1-L} }{ 6L+3 } {\ensuremath{\left( {\mathcal{Y}_{11}^{L M}{{\ensuremath{\left({{{\ensuremath{\boldsymbol{{p}}}}},{{\ensuremath{\boldsymbol{{q}}}}}}\right)}}}} \right)}}^* {\mathcal{Y}_{11}^{L M}{{\ensuremath{\left({{{\ensuremath{\boldsymbol{{{p^{\prime }}}}}}},{{\ensuremath{\boldsymbol{{{q^{\prime }}}}}}}}\right)}}}} \nonumber \\ & {\quad}{\quad \cross}{\delta^{(\Omega)}{{\ensuremath{\left({{\ensuremath{\boldsymbol{{p}}}}} - {{{\ensuremath{\boldsymbol{{\pi}}}}}_{3}{{\ensuremath{\left({\ensuremath{\boldsymbol{{q^{\prime }}}}}, {\ensuremath{\boldsymbol{q}}}\right)}}}}\right)}}}} {\delta^{(\Omega)}{{\ensuremath{\left({{\ensuremath{\boldsymbol{{{p^{\prime }}}}}}} - {{{\ensuremath{\boldsymbol{{\pi}}}}}_{3}{{\ensuremath{\left({\ensuremath{\boldsymbol{q}}}, {\ensuremath{\boldsymbol{{q^{\prime }}}}}\right)}}}}\right)}}}} \nonumber \\ & {\quad}= - {\int \dd{\Omega_{{\ensuremath{\boldsymbol{q}}}}}} {\int \dd{\Omega_{{\ensuremath{\boldsymbol{{q^{\prime }}}}}}}} {g_{l_n}{{\ensuremath{\left({\pi_{3}{{\ensuremath{\left({\ensuremath{\boldsymbol{{q^{\prime }}}}},{\ensuremath{\boldsymbol{{q^{}}}}}\right)}}}}\right)}}}} {G_0^{{\ensuremath{\left(n\right)}}}{{\ensuremath{\left({\pi_{3}{{\ensuremath{\left({\ensuremath{\boldsymbol{{q^{\prime }}}}},{\ensuremath{\boldsymbol{{q^{}}}}}\right)}}}}, q; E\right)}}}} {g_{l_n}{{\ensuremath{\left({\pi_{3}{{\ensuremath{\left({\ensuremath{\boldsymbol{{q^{}}}}},{\ensuremath{\boldsymbol{{q^{\prime }}}}}\right)}}}}\right)}}}} \nonumber \\ & {\quad}{\quad \cross}\sum_{L=0}^1 \sum_{M_L=-L}^L \frac{ {\ensuremath{\left(-2\right)}}^{1-L} }{ 6L+3 } {\ensuremath{\left( {\mathcal{Y}_{11}^{L M_L}{{\ensuremath{\left({{\ensuremath{\boldsymbol{\pi}}}_{3}{{\ensuremath{\left({\ensuremath{\boldsymbol{{q^{\prime }}}}}, {\ensuremath{\boldsymbol{q}}}\right)}}}}, {\ensuremath{\boldsymbol{q}}}\right)}}}} \right)}}^* {\mathcal{Y}_{11}^{L M_L}{{\ensuremath{\left({{\ensuremath{\boldsymbol{\pi}}}_{3}{{\ensuremath{\left({\ensuremath{\boldsymbol{q}}}, {\ensuremath{\boldsymbol{{q^{\prime }}}}}\right)}}}}, {\ensuremath{\boldsymbol{{q^{\prime }}}}}\right)}}}} \,,\end{aligned}$$ where ${\delta^{(\Omega)}{{\ensuremath{\left({{\ensuremath{\boldsymbol{p}}}}- {{\ensuremath{\boldsymbol{{p^{\prime }}}}}}\right)}}}} \coloneqq {\delta{{\ensuremath{\left(\varphi-\varphi^\prime\right)}}}} \frac{{\delta{{\ensuremath{\left(\theta-\theta^\prime\right)}}}}}{\sin{\theta}}$ holds. In order to evaluate this expression only one angular integral has to be computed numerically by using the identities given in appendix \[subsec:ang\_simp\] The calculation of $X_{nc}$ is based on an identity similar to Eq. (\[eq:Xnn\_id\]): $$\begin{aligned} \label{eq:Xnc_id} &{ \prescript{}{n}{ \braket{p,q;\Omega_n}{{p^{\prime }},{q^{\prime }};\Omega_c}_{c}^{} } } = {\int \dd{\Omega_{{\ensuremath{\boldsymbol{p}}}}}} {\int \dd{\Omega_{{\ensuremath{\boldsymbol{q}}}}}} {\int \dd{\Omega_{{\ensuremath{\boldsymbol{{p^{\prime }}}}}}}} {\int \dd{\Omega_{{\ensuremath{\boldsymbol{{q^{\prime }}}}}}}} \nonumber \\ &{\quad \cross}{ \prescript{}{n}{ \mel{p,q;\Omega_n}{ {\ensuremath{\left( { \ket{{{\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}}\vphantom{{{\ensuremath{\boldsymbol{{p^{\prime }}}}},{\ensuremath{\boldsymbol{{q^{\prime }}}}}}} }_{n}^{} } { \prescript{}{n}{ \braket{{{\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}}}{{{\ensuremath{\boldsymbol{{p^{\prime }}}}},{\ensuremath{\boldsymbol{{q^{\prime }}}}}}}_{c}^{} } } { \prescript{}{c}{ \bra{{{\ensuremath{\boldsymbol{{p^{\prime }}}}},{\ensuremath{\boldsymbol{{q^{\prime }}}}}}} } } \otimes {\mathbbm{1}^{{{\ensuremath{\left(\mathrm{spin}\right)}}}}}\right)}} }{{p^{\prime }},{q^{\prime }};\Omega_c}_{c}^{} } }\,.\end{aligned}$$ With it one obtains $$\begin{aligned} &X_{nc}{{\ensuremath{\left(q,{q^{\prime }};E\right)}}} \nonumber \\ & {\quad}= {\int \dd{{p^{}}} {p^{2}}} {\int \dd{{p^{\prime }}} {p^{\prime 2}}} {g_{l_n}{{\ensuremath{\left(p\right)}}}} {G_0^{{\ensuremath{\left(n\right)}}}{{\ensuremath{\left(p,q;E\right)}}}} {g_{l_c}{{\ensuremath{\left({p^{\prime }}\right)}}}} { \prescript{}{n}{ \braket{p,q;\Omega_n}{{p^{\prime }},{q^{\prime }};\Omega_c}_{c}^{} } } \nonumber \\ & {\quad}= - {\int \dd{\Omega_{{\ensuremath{\boldsymbol{p}}}}}} {\int \dd{\Omega_{{\ensuremath{\boldsymbol{q}}}}}} {\int \dd{\Omega_{{\ensuremath{\boldsymbol{{p^{\prime }}}}}}}} {\int \dd{\Omega_{{\ensuremath{\boldsymbol{{q^{\prime }}}}}}}} {g_{l_n}{{\ensuremath{\left({\pi_{1}{{\ensuremath{\left({\ensuremath{\boldsymbol{{q^{\prime }}}}},{\ensuremath{\boldsymbol{{q^{}}}}}\right)}}}}\right)}}}} {G_0^{{\ensuremath{\left(n\right)}}}{{\ensuremath{\left({\pi_{1}{{\ensuremath{\left({\ensuremath{\boldsymbol{{q^{\prime }}}}},{\ensuremath{\boldsymbol{{q^{}}}}}\right)}}}}, q; E\right)}}}} {g_{l_c}{{\ensuremath{\left({\pi_{2}{{\ensuremath{\left({\ensuremath{\boldsymbol{{q^{}}}}},{\ensuremath{\boldsymbol{{q^{\prime }}}}}\right)}}}}\right)}}}} \nonumber \\ & {\quad}{\quad \cross}\sqrt{\frac{2}{3}} {\ensuremath{\left( {\mathcal{Y}_{11}^{00}{{\ensuremath{\left({{\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}}\right)}}}} \right)}}^* {\mathcal{Y}_{00}^{00}{{\ensuremath{\left({\ensuremath{\boldsymbol{{p^{\prime }}}}}, {\ensuremath{\boldsymbol{{q^{\prime }}}}}\right)}}}} {\delta^{(\Omega)}{{\ensuremath{\left({{\ensuremath{\boldsymbol{{p}}}}} + {{{\ensuremath{\boldsymbol{{\pi}}}}}_{1}{{\ensuremath{\left({\ensuremath{\boldsymbol{{q^{\prime }}}}}, {\ensuremath{\boldsymbol{q}}}\right)}}}}\right)}}}} {\delta^{(\Omega)}{{\ensuremath{\left({{\ensuremath{\boldsymbol{{{p^{\prime }}}}}}} - {{{\ensuremath{\boldsymbol{{\pi}}}}}_{2}{{\ensuremath{\left({\ensuremath{\boldsymbol{q}}}, {\ensuremath{\boldsymbol{{q^{\prime }}}}}\right)}}}}\right)}}}} \nonumber \\ & {\quad}= - {\int \dd{\Omega_{{\ensuremath{\boldsymbol{q}}}}}} {\int \dd{\Omega_{{\ensuremath{\boldsymbol{{q^{\prime }}}}}}}} {g_{l_n}{{\ensuremath{\left({\pi_{1}{{\ensuremath{\left({\ensuremath{\boldsymbol{{q^{\prime }}}}},{\ensuremath{\boldsymbol{{q^{}}}}}\right)}}}}\right)}}}} {G_0^{{\ensuremath{\left(n\right)}}}{{\ensuremath{\left({\pi_{1}{{\ensuremath{\left({\ensuremath{\boldsymbol{{q^{\prime }}}}},{\ensuremath{\boldsymbol{{q^{}}}}}\right)}}}}, q; E\right)}}}} {g_{l_c}{{\ensuremath{\left({\pi_{2}{{\ensuremath{\left({\ensuremath{\boldsymbol{{q^{}}}}},{\ensuremath{\boldsymbol{{q^{\prime }}}}}\right)}}}}\right)}}}} \nonumber \\ & {\quad}{\quad \cross}{\ensuremath{\left( {\mathcal{Y}_{11}^{00}{{\ensuremath{\left(-{{\ensuremath{\boldsymbol{\pi}}}_{1}{{\ensuremath{\left({\ensuremath{\boldsymbol{{q^{\prime }}}}}, {\ensuremath{\boldsymbol{q}}}\right)}}}}, {\ensuremath{\boldsymbol{q}}}\right)}}}} \right)}}^* {\mathcal{Y}_{00}^{00}{{\ensuremath{\left({{\ensuremath{\boldsymbol{\pi}}}_{2}{{\ensuremath{\left({\ensuremath{\boldsymbol{q}}}, {\ensuremath{\boldsymbol{{q^{\prime }}}}}\right)}}}}, {\ensuremath{\boldsymbol{{q^{\prime }}}}}\right)}}}} \nonumber \\ & {\quad}= \frac{1}{\sqrt{2}} \int_{-1}^{1} \dd{\cos{\theta_{q,{q^{\prime }}}}} {g_{l_n}{{\ensuremath{\left({\pi_{1}{{\ensuremath{\left({\ensuremath{\boldsymbol{{q^{\prime }}}}},{\ensuremath{\boldsymbol{{q^{}}}}}\right)}}}}\right)}}}} {G_0^{{\ensuremath{\left(n\right)}}}{{\ensuremath{\left({\pi_{1}{{\ensuremath{\left({\ensuremath{\boldsymbol{{q^{\prime }}}}},{\ensuremath{\boldsymbol{{q^{}}}}}\right)}}}}, q; E\right)}}}} {g_{l_c}{{\ensuremath{\left({\pi_{2}{{\ensuremath{\left({\ensuremath{\boldsymbol{{q^{}}}}},{\ensuremath{\boldsymbol{{q^{\prime }}}}}\right)}}}}\right)}}}} \cos{\theta_{-{{\ensuremath{\boldsymbol{\pi}}}_{1}{{\ensuremath{\left({\ensuremath{\boldsymbol{{q^{\prime }}}}}, {\ensuremath{\boldsymbol{q}}}\right)}}}},{\ensuremath{\boldsymbol{q}}}}} \,.\end{aligned}$$ Note that $$X_{nc}{{\ensuremath{\left(q,{q^{\prime }};E\right)}}} = X_{cn}{{\ensuremath{\left({q^{\prime }},q;E\right)}}}$$ holds. This can be shown using that ${ \prescript{}{n}{ \braket{p,q;\Omega_n}{{p^{\prime }},{q^{\prime }};\Omega_c}_{c}^{} } }$ is real. The implementation of these results was checked by evaluating the expressions for [Heaviside form factor]{}s, for which the analytic expressions have been calculated in [@ji14]. As in the derivation of the analytic expressions the [Heaviside form factor]{}s have been neglected, discrepancies from the numerical results can occur at momenta of the order of the regularization scale. Overlaps in partial wave basis {#subsec:ov_pwb} ------------------------------ In the following expressions for the overlaps of partial wave states with different spectators are derived. Also matrix elements of the $nn$ permutation operator ${\mathcal{P}_{nn}}= {\mathcal{P}_{nn}^\mathrm{(spatial)}}\otimes {\mathcal{P}_{nn}^\mathrm{(spin)}}$ are given in this basis. We obtain $$\begin{aligned} & { \prescript{}{c}{ \braket{p,q;\Omega_c}{{p^{\prime }},{q^{\prime }};\Omega_n}_{n}^{} } } \nonumber \\ & {\quad}= {\int \dd{\Omega_{{\ensuremath{\boldsymbol{p}}}}}} {\int \dd{\Omega_{{\ensuremath{\boldsymbol{q}}}}}} {\int \dd{\Omega_{{\ensuremath{\boldsymbol{{p^{\prime }}}}}}}} {\int \dd{\Omega_{{\ensuremath{\boldsymbol{{q^{\prime }}}}}}}} \nonumber \\ & {\quad}{\quad}\cross { \prescript{}{c}{ \mel{p,q;\Omega_c}{ {\ensuremath{\left( { \ket{{{\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}}\vphantom{{{\ensuremath{\boldsymbol{{p^{\prime }}}}},{\ensuremath{\boldsymbol{{q^{\prime }}}}}}} }_{c}^{} } { \prescript{}{c}{ \braket{{{\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}}}{{{\ensuremath{\boldsymbol{{p^{\prime }}}}},{\ensuremath{\boldsymbol{{q^{\prime }}}}}}}_{n}^{} } } { \prescript{}{n}{ \bra{{{\ensuremath{\boldsymbol{{p^{\prime }}}}},{\ensuremath{\boldsymbol{{q^{\prime }}}}}}} } } \otimes {\mathbbm{1}^{{{\ensuremath{\left(\mathrm{spin}\right)}}}}}\right)}} }{{p^{\prime }},{q^{\prime }};\Omega_n }_{n}^{} } } \nonumber \\ & {\quad}= \frac{ \sqrt{2} }{{\ensuremath{\left(4\pi\right)}}^2} {\int \dd{\Omega_{{\ensuremath{\boldsymbol{p}}}}}} {\int \dd{\Omega_{{\ensuremath{\boldsymbol{q}}}}}} {\int \dd{\Omega_{{\ensuremath{\boldsymbol{{p^{\prime }}}}}}}} {\int \dd{\Omega_{{\ensuremath{\boldsymbol{{q^{\prime }}}}}}}} \nonumber \\ & {\quad}{\quad}\cross \cos{\theta_{{{\ensuremath{\boldsymbol{{p^{\prime }}}}},{\ensuremath{\boldsymbol{{q^{\prime }}}}}}}} {\delta^{(3)}{{\ensuremath{\left( {\ensuremath{\boldsymbol{{p^{\prime }}}}} - {{\ensuremath{\boldsymbol{\kappa}}}_{cnp}}{{\ensuremath{\left({{\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}}\right)}}} \right)}}}} {\delta^{(3)}{{\ensuremath{\left( {\ensuremath{\boldsymbol{{q^{\prime }}}}} - {{\ensuremath{\boldsymbol{\kappa}}}_{cnq}}{{\ensuremath{\left({{\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}}\right)}}} \right)}}}}\,, \label{eq:cOn_aux} \\ & \hphantom{{\quad}{\quad}\cross { \prescript{}{c}{ \mel{p,q;\Omega_c}{ {\ensuremath{\left( { \ket{{{\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}}\vphantom{{{\ensuremath{\boldsymbol{{p^{\prime }}}}},{\ensuremath{\boldsymbol{{q^{\prime }}}}}}} }_{c}^{} } { \prescript{}{c}{ \mel{{{\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}}}{{\mathcal{P}_{nn}^\mathrm{(spatial)}}}{{{\ensuremath{\boldsymbol{{p^{\prime }}}}},{\ensuremath{\boldsymbol{{q^{\prime }}}}}}}_{n}^{} } } { \prescript{}{n}{ \bra{{{\ensuremath{\boldsymbol{{p^{\prime }}}}},{\ensuremath{\boldsymbol{{q^{\prime }}}}}}} } } \otimes {\mathcal{P}_{nn}^\mathrm{(spin)}}\right)}} }{{p^{\prime }},{q^{\prime }};\Omega_n }_{n}^{} } } \nonumber}\end{aligned}$$ $$\begin{aligned} & { \prescript{}{c}{ \mel{p,q;\Omega_c}{{\mathcal{P}_{nn}}}{{p^{\prime }},{q^{\prime }};\Omega_n}_{n}^{} } } \nonumber \\ & {\quad}= {\int \dd{\Omega_{{\ensuremath{\boldsymbol{p}}}}}} {\int \dd{\Omega_{{\ensuremath{\boldsymbol{q}}}}}} {\int \dd{\Omega_{{\ensuremath{\boldsymbol{{p^{\prime }}}}}}}} {\int \dd{\Omega_{{\ensuremath{\boldsymbol{{q^{\prime }}}}}}}} \nonumber \\ & {\quad}{\quad}\cross { \prescript{}{c}{ \mel{p,q;\Omega_c}{ {\ensuremath{\left( { \ket{{{\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}}\vphantom{{{\ensuremath{\boldsymbol{{p^{\prime }}}}},{\ensuremath{\boldsymbol{{q^{\prime }}}}}}} }_{c}^{} } { \prescript{}{c}{ \mel{{{\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}}}{{\mathcal{P}_{nn}^\mathrm{(spatial)}}}{{{\ensuremath{\boldsymbol{{p^{\prime }}}}},{\ensuremath{\boldsymbol{{q^{\prime }}}}}}}_{n}^{} } } { \prescript{}{n}{ \bra{{{\ensuremath{\boldsymbol{{p^{\prime }}}}},{\ensuremath{\boldsymbol{{q^{\prime }}}}}}} } } \otimes {\mathcal{P}_{nn}^\mathrm{(spin)}}\right)}} }{{p^{\prime }},{q^{\prime }};\Omega_n }_{n}^{} } } \nonumber \\ & {\quad}= -\frac{ \sqrt{2} }{{\ensuremath{\left(4\pi\right)}}^2} {\int \dd{\Omega_{{\ensuremath{\boldsymbol{p}}}}}} {\int \dd{\Omega_{{\ensuremath{\boldsymbol{q}}}}}} {\int \dd{\Omega_{{\ensuremath{\boldsymbol{{p^{\prime }}}}}}}} {\int \dd{\Omega_{{\ensuremath{\boldsymbol{{q^{\prime }}}}}}}} \nonumber \\ & {\quad}{\quad}\cross \cos{\theta_{{{\ensuremath{\boldsymbol{{p^{\prime }}}}},{\ensuremath{\boldsymbol{{q^{\prime }}}}}}}} {\delta^{(3)}{{\ensuremath{\left( {\ensuremath{\boldsymbol{{p^{\prime }}}}} - {{\ensuremath{\boldsymbol{\kappa}}}_{cnp}^{\prime}}{{\ensuremath{\left({{\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}}\right)}}} \right)}}}} {\delta^{(3)}{{\ensuremath{\left( {\ensuremath{\boldsymbol{{q^{\prime }}}}} - {{\ensuremath{\boldsymbol{\kappa}}}_{cnq}^{\prime}}{{\ensuremath{\left({{\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}}\right)}}} \right)}}}}\,, \label{eq:cOn_aux_wp} \\ & { \prescript{}{n}{ \braket{p,q;\Omega_n}{{p^{\prime }},{q^{\prime }};\Omega_c}_{c}^{} } } \nonumber \\ & {\quad}= {\int \dd{\Omega_{{\ensuremath{\boldsymbol{p}}}}}} {\int \dd{\Omega_{{\ensuremath{\boldsymbol{q}}}}}} {\int \dd{\Omega_{{\ensuremath{\boldsymbol{{p^{\prime }}}}}}}} {\int \dd{\Omega_{{\ensuremath{\boldsymbol{{q^{\prime }}}}}}}} \nonumber \\ & {\quad}{\quad}\cross { \prescript{}{n}{ \mel{p,q;\Omega_n}{ {\ensuremath{\left( { \ket{{{\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}}\vphantom{{{\ensuremath{\boldsymbol{{p^{\prime }}}}},{\ensuremath{\boldsymbol{{q^{\prime }}}}}}} }_{n}^{} } { \prescript{}{n}{ \braket{{{\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}}}{{{\ensuremath{\boldsymbol{{p^{\prime }}}}},{\ensuremath{\boldsymbol{{q^{\prime }}}}}}}_{c}^{} } } { \prescript{}{c}{ \bra{{{\ensuremath{\boldsymbol{{p^{\prime }}}}},{\ensuremath{\boldsymbol{{q^{\prime }}}}}}} } } \otimes {\mathbbm{1}^{{{\ensuremath{\left(\mathrm{spin}\right)}}}}}\right)}} }{{p^{\prime }},{q^{\prime }};\Omega_c }_{c}^{} } } \nonumber \\ & {\quad}= \frac{\sqrt{2}}{{\ensuremath{\left(4\pi\right)}}^2} {\int \dd{\Omega_{{\ensuremath{\boldsymbol{p}}}}}} {\int \dd{\Omega_{{\ensuremath{\boldsymbol{q}}}}}} {\int \dd{\Omega_{{\ensuremath{\boldsymbol{{p^{\prime }}}}}}}} {\int \dd{\Omega_{{\ensuremath{\boldsymbol{{q^{\prime }}}}}}}} \nonumber \\ & {\quad}{\quad}\cross \cos{\theta_{{{\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}}}} \frac{1}{4\pi} {\delta^{(3)}{{\ensuremath{\left( {\ensuremath{\boldsymbol{{p^{\prime }}}}} - {{\ensuremath{\boldsymbol{\kappa}}}_{ncp}}{{\ensuremath{\left({{\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}}\right)}}} \right)}}}} {\delta^{(3)}{{\ensuremath{\left( {\ensuremath{\boldsymbol{{q^{\prime }}}}} - {{\ensuremath{\boldsymbol{\kappa}}}_{ncq}}{{\ensuremath{\left({{\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}}\right)}}} \right)}}}}\,, \label{eq:nOc_aux} \\ & { \prescript{}{n}{ \mel{p,q;\Omega_n}{{\mathcal{P}_{nn}}}{{p^{\prime }},{q^{\prime }};\Omega_n}_{n}^{} } } \nonumber \\ & {\quad}= {\int \dd{\Omega_{{\ensuremath{\boldsymbol{p}}}}}} {\int \dd{\Omega_{{\ensuremath{\boldsymbol{q}}}}}} {\int \dd{\Omega_{{\ensuremath{\boldsymbol{{p^{\prime }}}}}}}} {\int \dd{\Omega_{{\ensuremath{\boldsymbol{{q^{\prime }}}}}}}} \nonumber \\ & {\quad}{\quad}\cross { \prescript{}{n}{ \mel{p,q;\Omega_n}{ {\ensuremath{\left( { \ket{{{\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}}\vphantom{{{\ensuremath{\boldsymbol{{p^{\prime }}}}},{\ensuremath{\boldsymbol{{q^{\prime }}}}}}} }_{n}^{} } { \prescript{}{n}{ \mel{{{\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}}}{{\mathcal{P}_{nn}^\mathrm{(spatial)}}}{{{\ensuremath{\boldsymbol{{p^{\prime }}}}},{\ensuremath{\boldsymbol{{q^{\prime }}}}}}}_{n}^{} } } { \prescript{}{n}{ \bra{{{\ensuremath{\boldsymbol{{p^{\prime }}}}},{\ensuremath{\boldsymbol{{q^{\prime }}}}}}} } } \otimes {\mathcal{P}_{nn}^\mathrm{(spin)}}\right)}} }{{p^{\prime }},{q^{\prime }};\Omega_n }_{n}^{} } } \nonumber \\ & {\quad}= {\int \dd{\Omega_{{\ensuremath{\boldsymbol{p}}}}}} {\int \dd{\Omega_{{\ensuremath{\boldsymbol{q}}}}}} {\int \dd{\Omega_{{\ensuremath{\boldsymbol{{p^{\prime }}}}}}}} {\int \dd{\Omega_{{\ensuremath{\boldsymbol{{q^{\prime }}}}}}}} \sum_{L=0}^1 \sum_{M=-L}^{L} {\ensuremath{\left(-1\right)}}^{1-L} \frac{2^{1-L}}{6L+3} \nonumber \\ & {\quad}{\quad}\cross {\ensuremath{\left({\mathcal{Y}_{11}^{LM}{{\ensuremath{\left({\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}\right)}}}}\right)}}^* {\mathcal{Y}_{11}^{LM}{{\ensuremath{\left({{\ensuremath{\boldsymbol{{p^{\prime }}}}},{\ensuremath{\boldsymbol{{q^{\prime }}}}}}\right)}}}} {\delta^{(3)}{{\ensuremath{\left( {\ensuremath{\boldsymbol{{p^{\prime }}}}} - {{\ensuremath{\boldsymbol{\kappa}}}_{nnp}^{\prime}}{{\ensuremath{\left({{\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}}\right)}}} \right)}}}} {\delta^{(3)}{{\ensuremath{\left( {\ensuremath{\boldsymbol{{q^{\prime }}}}} - {{\ensuremath{\boldsymbol{\kappa}}}_{nnq}^{\prime}}{{\ensuremath{\left({{\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}}\right)}}}\right)}}}}\,. \label{eq:nOn_aux_wp}\end{aligned}$$ [^1]: In memory of Ludwig Faddeev [^2]: This result for the probability density can also be derived without invoking a density matrix if we instead compute the expectation value in the ${}^6$He ground state of the modified momentum-space projector (\[eq:PpD\]) using the modified scalar product of [@formanek04]. [^3]: Note that we abbreviate the spin states in the following way: $$\begin{aligned} { \ket{L, -M}_{n}^{} } &= { \ket{{\ensuremath{\left({\frac{1}{2}}, 0\right)}}{\frac{1}{2}},{\frac{1}{2}}; L, -M}_{n}^{} }\,, \\ { \ket{0, 0}_{c}^{} } &= { \ket{{\ensuremath{\left({\frac{1}{2}}, {\frac{1}{2}}\right)}}0,0; 0, 0}_{c}^{} }\,, \end{aligned}$$ where the notation ${ \ket{{\ensuremath{\left(\nu_j,\nu_k\right)}}s_i, \sigma_i; S, M_S}_{i}^{} }$ from [@ji14] is used. [^4]: We put it in quotation marks, since it is actually only an approximation for the potential energy density. The potential energy density of $V_i$ is defined as $P_{{{\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}}} V_i$, where $P_{{{\ensuremath{\boldsymbol{p}}},{\ensuremath{\boldsymbol{q}}}}}$ projects on the Jacobi momenta, ${\ensuremath{\boldsymbol{p}}}$ and ${\ensuremath{\boldsymbol{q}}}$. The formula above represents the expectation value of this product of operators in the case of potentials which are not energy dependent. Otherwise, correction terms exist. [^5]: We define the following measure for the numerical uncertainty: $$\frac{ \max_{i,j}{{\ensuremath{\left(\left| \rho_{ij}^{{\ensuremath{\left(2n\right)}}} - \rho_{ij}^{{\ensuremath{\left(n\right)}}} \right|\right)}}} }{ \max_{i,j}{{\ensuremath{\left(\left| \rho_{ij}^{{\ensuremath{\left(2n\right)}}} \right|\right)}}}/2 }\,,$$ where the $\rho_{ij}$ is the probability density evaluated on a grid. We put the number of mesh points in the brackets in the superscript of the $\rho_{ij}$. Accordingly this is the maximum absolute difference of the result with a fixed number of mesh points and the result obtained with twice as many mesh points, divided by a “typical" density value. For the latter we choose the half of the maximum of the absolute values. [^6]: Note that we make a small error by using the analytic results for the $\lambda_i$ as given in subsection \[subsec:coupling\_strength\], since in these calculations the three-body cutoff $\Lambda$ is neglected.

ArXiv

**ON A CLASS OF C\*-PREDUALS OF $l_1$**\ *By STEFANO ROSSI* $$$$\ $$$$ **Abstract**As it is well known, the Banach space $l_1$ of absolutely summable (complex) sequences endowed with the $\|\cdot\|_1$ norm is not *unique predual*. This means that there are many different (*i.e.* non isometrically isomorphic) Banach spaces $X$ such that $X^*\cong l_1$.\ The present note is aimed to point out a simple class of C\*- preduals of $l_1$: namely the spaces $C_{\tau}(\mathbb{N})$ of continuous functions $f:\mathbb{N}\rightarrow\mathbb{C}$, where the set of natural numbers $\mathbb{N}$ is equipped with a compact Hausdorff topology $\mathcal{T}$.\ To be more concrete, we shall explicitly describe a countable collection $\{\mathcal{T}_n\}$ of such topologies.\ Finally, we also provide an abstract characterization of the previous preduals as closed subspaces $M\subset l^{\infty}$ rich of positive elements. $$$$ As commonly used in the literature, we shall denote by $l_1$ the (complex) Banach space of absolutely summable sequences, given of the norm $\|\cdot\|_1$ defined by $\|a\|_1\doteq\sum_{i=1}^{\infty}|a_i|$ for each $a\in l_1$.\ It is a very well known fact that $l_1$ is a conjugate Banach space, that is there exists at least a Banach space $X$, such that $X^*\cong l_1$ (isometric isomorphism). Such a space is usually named a *predual*. The most famous predual of $l_1$ is probably represented by the space $c_0$ of those (complex) sequences converging to $0$, endowed of the *sup*-norm. In this case, the isometric isomorphism $c_0^*\cong l_1$ is the map $\Psi: l_1\rightarrow c_0^*$ given by $\langle\Psi(y), x\rangle\doteq\sum_{i=1}^{\infty}y_ix_i$ for every $x\in c_0$ and $y\in l_1$.\ In spite of its simple definition, $l_1$ is a rather pathological[^1] Banach space: for instance the predual is not unique; there is in fact a plenty of (non isomorphic) preduals of $l_1$. Some of these are quite “irregular”: Y. Benyamini and J. Lindenstrauss [@Linde] proved in 1972 that there is a predual of $l_1$ that is not (topologically) complemented in any $C(K)$-space, $K$ being any compact Hausdorff topological space.\ On the other hand, the present paper is aimed to discuss a very nice class of C\*-preduals of $l_1$. In this spirit, the first thing that should be noticed is the following: \[prop1\] If $\mathcal{T}$ is a compact Hausdorff topology on the set of natural numbers $\mathbb{N}$, one has $C_{\tau}(\mathbb{N})^*\cong l_1$. It is possible to prove the statement by using the Riesz-Markov theorem. Here we perform a proof based on the characterization of separable conjugate spaces given in [@Rossi]. To this aim, we only have to check that $C_{\tau}(\mathbb{N})\subset l^{\infty}$ is a closed, norm-attaining and $1$-norming subspace.\ $C_{\tau}(\mathbb{N})$ is closed in $l^{\infty}$ as a complete subspace. It is norm-attaining (when it is thought as subspace of bounded linear functionals on $l_1$ ) thanks to Weierstrass’ theorem, since $(\mathbb{N},\mathcal{T})$ is a compact space by assumption.\ If $y\in l_1$ and $\varepsilon>0$, there is $n\in\mathbb{N}$ such that $\|y\|_1\leq \sum_{i=1}^n|y_i|+\varepsilon$. Let $\theta_i\in\mathbb{R}$ such that $y_i=|y_i|e^{i\theta_i}$ for each $i=1,2\dots,n$. The subset $C_n\doteq\{1, 2, \dots,n\}\subset\mathbb{N}$ is closed (and discrete), hence the function $f:C_n\rightarrow\mathbb{C}$ given by $f(i)=e^{-i\theta_i}$ for each $i\in C_n$ is continuous and $\|f\|_{\infty}=1$. Since $(\mathbb{N},\mathcal{T})$ is a compact Hausdorff space, it is a normal topological space, so Tietze extension theorem applies to get a function $g\in C_{\tau}(\mathbb{N})$ such that $\|g\|_{\infty}=1$ and $g(i)=e^{-i\theta_i}$ for each $i\in{1, 2,\dots,n}$.\ We have $|\langle g, y\rangle|=|\sum_{i=1}^{\infty} g(i)y_i|\geq\sum_{i=1}^n |y_i|-\varepsilon\geq \|y\|_1-2\varepsilon$. The last inequality easily implies that $$\sup_{g\in C_{\tau}(\mathbb{N})_1}|\langle g, y\rangle|=\|y\|_1$$ that is $C_{\tau}(\mathbb{N})\subset l^{\infty}$ is a $1$-norming subspace. This ends the proof. The previous proposition immediately leads to the following corollary in point-set topology: Every compact Hausdorff topology on the set of natural numbers $\mathbb{N}$ is metrizable. Let $\mathcal{T}$ be such a topology. We have $C_{\tau}(\mathbb{N})^*\cong l_1$, hence $C_{\tau}(\mathbb{N})$ is a separable Banach space, as a predual of the separable Banach space $l_1$, so that $(\mathbb{N},\mathcal{T})$ is metrizable. As far as I know, a simple proof of the corollary quoted above does not seem available in the general setting of point-set topology, since it is not apparent that a compact Hausdorff topology on $\mathbb{N}$ is automatically second countable.\ On the other hand, non first countable topologies on $\mathbb{N}$ are known: *Appert* topology, for instance, provides an elegant example of such a space. For the reader’s convenience, we recall here that Appert’s topology on $\mathbb{N}$ is defined as follows: a subset $A\subset\mathbb{N}$ is open if $1\notin A$ or (when $1\in A$) if $$\lim_{n\to\infty}\frac{N(n,A)}{n}=1$$ where $N(n,A)\doteq \left|\left\{k\in A: k\leq n\right\}\right|.$[^2] Appert space is Lindelöf, separable but it is not first countable, since $1$ does not have a countable basis of neighborhoods. For more details, we refer the interested reader to [@Steen] or directly to the original paper by Appert [@Appert]. Here below we shall describe explicitly a countable collection of compact Hausdorff topologies on $\mathbb{N}$. Before introducing the announced topologies, one should mention that every set $X$ can be endowed with a compact Hausdorff topology, by virtue of a straightforward application of the Axiom of Choice[^3].\ Now let $n\in \mathbb{N}$ be a fixed natural number. Given any $k\in\{1,2,\dots, n\}$, we define the sets $A_{k,l}\doteq\{k, mn+k: m\geq l\}$. The sets $A_{k,l}$ allow us to define a topology $\mathcal{T}_n$, whose basis $\mathcal{B}_n$ is given by the subset $B\subset\mathbb{N}$ of the form $A_{k,l}$ if $k\in B$ for some $k\in\{1,2,\dots, n\}$, otherwise we do not put any restriction, namely if $\{1,2\dots, n\}\cap B=\emptyset$ then $B$ is allowed to be any subset of the natural numbers.\ Since $A_{k,l}\cap A_{k,h}=A_{k,l\vee h}$[^4] and $A_{k,l}\cap A_{k',h}=\emptyset$ when $k,k'\in\{1,2,\dots, n\}$ are different, $\mathcal{B}_n$ is really a basis. It is a straightforward verification to check that $\mathcal{T}_n$ is a compact Hausdorff topology; the notion of convergence inherited by this topology is clearly the following:\ a sequence $\{n_m: m\in\mathbb{N}\}$ of integers converges to $k\in\{1,2,\dots, n\}$ iff $n_m$ is eventually in a set $A_{k,l}$, while converges to $k>n$ iff it is eventually equal to $k$.\ In the topology $\mathcal{T}_n$ the set $\{k: k\leq n\}$ is composed by non isolated points, while all the integers $k>n$ are isolated. In some sense, topologies $\mathcal{T}_n$ are as best as possible among compact Hausdorff ones, since it is a straightforward application of *Baire* category theorem that a compact Hausdorff topology on $\mathbb{N}$ cannot have an infinite set of accumulation points[^5].\ However, what is more important here is that a simple argument can be performed to prove that the topologies $\mathcal{T}_n$ are not homeomorphic: With the notations above, if $n\neq m$ the topological spaces $(\mathbb{N},\mathcal{T}_n)$ and $(\mathbb{N},\mathcal{T}_m)$ are not homeomorphic. Let us suppose that $m>n$ and let $\Phi:(\mathbb{N},\mathcal{T}_m)\rightarrow (\mathbb{N},\mathcal{T}_n)$ be a continuous injective map. If $k\in\{1,2,,\dots m\}$, we can consider a sequence $\{n_l\}$ converging to $k$. The sequence $\{\Phi(n_l)\}$ converges to $\Phi(k)$ thanks to the continuity of $\Phi$. Since $\{n_l\}$ is not constant and $\Phi$ is an injection $\Phi(k)$ is forced to be a natural number belonging to the subset $\{1,2,\dots,n\}$, against the injectivity of $\Phi$. Let us denote by $X_n$ the Banach space $C_{\tau_n}(\mathbb{N})$. Clearly we have $X_n^*\cong l_1$ and If $n\neq m$ the Banach space $X_n$ and $X_m$ are $l_1$-preduals, which are not isometrically isomorphic. If they were isometrically isomorphic, the topological space $(\mathbb{N},\mathcal{T}_n)$ and $(\mathbb{N},\mathcal{T}_m)$ should be homeomorphic according to the classical Banach-Stone theorem. The remaining part of the present paper is devoted to provide an intrinsic characterization of the spaces $C_{\tau}(\mathbb{N})$ as suitable subspaces of $l^{\infty}$. To this aim, one probably has to remind that any predual $M$ of a conjugate spaces $X$ should be sought as a closed subspace of the dual space $X^*$, which is $1$-*norming*[^6] and *norm-attaining*, namely each linear functionals belonging to the subspace is required to attain its norm on the unit ball of $X$.\ When $X$ is a separable conjugate space, the conditions above are also sufficient for a closed subspace $M\subset X^*$ to be canonically a predual of $X$ as it is shown in [@Rossi].\ Here canonically means that the isometric isomorphism $X\cong M^*$ is nothing but the restriction of the canonical injection $j:X\rightarrow X^{**}$ to $M$.\ Before stating the result announced, let us fix some notations: $e\in l^{\infty}$ is the sequence constantly equal to $1$, $M_+$ stands for the positive[^7] cone of a subspace $M\subset l^{\infty}$, while $a^{\frac{1}{2}}$ is the square root[^8] of a positive element $a\in l^{\infty}_+$.\ According to the next theorem the spaces $C_{\tau}(\mathbb{N})$ are precisely those $l_1$-predual rich of positive elements: $$$$ Let $M\subset l^{\infty}$ be a predual of $l_1$, such that:\ $(a)$ $e\in M$.\ $(b)$ $M_+$ is weakly\*-dense in $l^{\infty}_+$.\ $(c)$ If $x\in M_+$, then $x^{\frac{1}{2}}\in M_+$.\ Then $M=C_{\tau}(\mathbb{N})$ for a suitable compact Hausdorff topology on the set of natural numbers $\mathbb{N}$. Let be $\mathfrak{A}\subset l^{\infty}$ be the unital C\*-algebra[^9] generated by $M$. If $\omega$ is a *pure* (multiplicative) state on $\mathfrak{A}$, we can consider its restriction $\omega\upharpoonright_M$. Since $M^*\cong l_1$, we have $\omega(x)=\varphi_y(x)\doteq \sum_i y_ix_i$ for each $x\in M$, where $y$ is a suitable sequence in $l_1$. Now pick a positive element $a\in l_{\infty}$. Thanks to $(b)$, there is a sequence $\{x_n\}_{n\in\mathbb{N}}\subset M_+$ such that $x_n\rightharpoonup a$ (in the weak\* topology of $l^{\infty}$). Then we have $$\begin{aligned} \varphi_y(a)=\lim_n\varphi_y(x_n)=\lim_n\varphi\left(x_n^{\frac{1}{2}}x_n^{\frac{1}{2}}\right)=\nonumber\\ \lim_n\omega\left(x_n^{\frac{1}{2}}x_n^{\frac{1}{2}}\right)= \lim_n\omega\left(x_n^{\frac{1}{2}}\right)\omega\left(x_n^{\frac{1}{2}}\right)= \varphi_y(a^{\frac{1}{2}})^2\nonumber\\ \nonumber\end{aligned}$$ where the last equality holds since $x_n^{\frac{1}{2}}\rightharpoonup a^{\frac{1}{2}}$ (the weak\* convergence in $l^{\infty}$ is nothing but the bounded pointwise convergence).\ If $e_i\in l^{\infty}$ is the sequence given by $e_i(k)=\delta_{i,k}$, we get $\varphi_y(e_i)=\varphi_y(e_i)^2$, because $e_i^{\frac{1}{2}}$ is $e_i$ itself. It follows that, for each $i\in\mathbb{N}$, $\varphi_y(e_i)$ is $0$ or $1$. Since $\sum_i |y_i|= \|\varphi_y\|=1$, one has $y=e_k$ for some $k$. It easily follows that $\omega$ is the evaluation map at $k$.\ This means that $\sigma(\mathfrak{A})\cong\mathbb{N}$, hence $\mathfrak{A}=C_{\tau}(\mathbb{N})$, $\mathcal{T}$ being the weak\* topology on the spectrum of $\mathfrak{A}$.\ Thanks to proposition \[prop1\], we have $C_{\tau}(\mathbb{N})\cong l_1$; since no proper inclusion relationships are allowed between preduals, we finally get $M=\mathfrak{A}$. This concludes the proof. [8]{} A. Appert, *Proprietés des espaces abstraits le plus généraux*, Actualités Sci. Indust. No. 146, Hermann, Paris, 1934. K. R. Davidson, *$C^*$-Algebras by Example*, Fields Institute Monographs, American Mathematical Society, 1996. S. W. Golomb, *A connected topology for the integers*, Amer. Math. Montly, **66**, 663-665, 1959. Y. Benyami, J. Lindenstrauss, *A predual of $l_1$ which is not isomorphic to a $C(K)$ space*, Israel Journal of Mathematics **13**, 246-254, 1972. S. Rossi, *A characterization of separable conjugate spaces*, www.arxiv.org. L.A. Steen, J.A. Seebach, *Counterexamples in Topology*, Dover Pubblication, Inc. New York, 1995. $$$$ *DIP. MAT. CASTELNUOVO, UNIV. DI ROMA LA SAPIENZA, ROME, ITALY*\ *E-mail address:* `s-rossi@mat.uniroma1.it` [^1]: The weak topology of $l_1$ is not well behaved: every weakly convergent sequence is indeed norm-convergent, although the weak topology is strictly weaker than the norm topology. [^2]: $|X|$ is the cardinality of any set $X$. [^3]: The discrete topology $\mathcal{P}(X)$ on $X$ is locally compact and Hausdorff. The Alexandroff compactification $\hat{X}$ of $X$ is compact and Hausdorff; moreover, if $X$ is an infinite set, there is a bijection $\Phi:X\rightarrow\hat{X}$. We can use $\Phi$ to define a compact Hausdorff topology $\mathcal{T}$ on $X$, by requiring a set $U\subset X$ to be open if $\Phi(U)$ is an open subset of $\hat{X}$. [^4]: Here $l\vee h$ stands for $\max\{l,h\}$. [^5]: Whenever $\mathcal{T}$ is a compact Hausdorff topology on $\mathbb{N}$, $(\mathbb{N},\mathcal{T})$ is a Baire space as a complete metric space, hence it cannot be written as a countable union of rare sets, but every non isolated point $n\in\mathbb{N}$ gives a rare singleton $\{n\}$. In particular, the set of natural numbers $\mathbb{N}$ cannot be given of a connected compact Hausdorff topology; anyway a connected Hausdorff topology on $\mathbb{N}$ is available: for instance*Golomb* topology, see [@Golomb]. [^6]: A subspace $M\subset X^*$ is said to be $1$-norming if for each $x\in X$, one has $$\|x\|=\sup\{|\varphi(x)|: \varphi\in M_1\}$$ $M_1$ being the unit ball of $M$. [^7]: An element $x\in l^{\infty}$ is said to be positive if $x_i\geq 0$ for each $i\in \mathbb{N}$; in this case one writes $x\geq 0$. [^8]: If $x\geq 0$, then $x^{\frac{1}{2}}$ is the positive sequence given by $x^{\frac{1}{2}}(i)\doteq x_i^{\frac{1}{2}}$ for each $i\in\mathbb{N}$. [^9]: For a basic treatment of $C^*$-algebras theory, we refer the reader to [@Davidson].

ArXiv

--- abstract: | We compute radiative correction to the lightest neutral Higgs mass ($m_h$) induced by the Kaluza-Klein (KK) towers of fermions and sfermions in a minimal supersymmetric scenario embeded in a 5-dimensional warped space. The Higgs is confined to the TeV brane. The KK spectra of matter supermultiplets is tied to the explanation of the fermion mass hierarchy problem. We demonstrate that for a reasonable choice of extra-dimensional parameters, the KK-induced radiative correction can enhance the upper limit on $m_h$ by as much as 100 GeV beyond the 4d limit of 135 GeV. `PACS Nos:  12.60.Jv, 11.10.Kk `\ `Key Words:  Supersymmetry, Warped Extra Dimension, Higgs mass` --- SINP/TNP/2008/13, HRI-P-08-06-002, CU-Physics/09-2008, LPT-ORSAY 08-66 [**Radiative correction to the lightest neutral Higgs mass\ in warped supersymmetry**]{}\ [ [Gautam Bhattacharyya${}^1$]{}, [Swarup Kumar Majee${}^{2,3}$]{}, [Tirtha Sankar Ray${}^{1}$]{} ]{}\ [**Introduction**]{}:  Minimal supersymmetric standard model (MSSM) with superparticles in the 1 TeV range, primarily for its ability to settle the gauge hierarchy problem and for providing a cold dark matter candidate, has emerged as a leading candidate of physics beyond the standard model (SM). A key prediction of MSSM is the existence of a light Higgs ($m_h <$ 135 GeV). If such a light scalar exists, the CERN Large Hadron Collider (LHC) will find it hard to miss. Moreover, if a quantum picture for all interactions including gravity has to be weaved, we have to rely on string theory, which invariably includes supersymmetry (SUSY). Since string theory is fundamentally a higher dimensional theory, a reanalysis of the four-dimensional (4d) MSSM Higgs spectra by embedding the theory in an extra-dimensional set-up is a worthwhile phenomenological exercise. Randall-Sundrum (RS) type models [@Randall:1999ee] with a warped space-time geometry, where the bulk is a slice of Anti-deSitter (AdS) space in which the SM/MSSM particles can also penetrate [@bulksm; @Gherghetta:2000qt], lead to important phenomenological consequences: (i) gauge hierarchy problem is solved thanks to the warp factor, (ii) mass hierarchy of fermions can be explained by their relative localizations in the bulk [@Huber:2000ie], (iii) the smallness of neutrino masses can be explained [@Grossman:1999ra], (iv) gauge couplings unify if the warped space is supersymmetric [@Dienes:1999sz], (v) SUSY breaking can be realized with a geometrical interpretation [@Gherghetta:2000kr], (vi) light Kaluza-Klein (KK) gauge boson and fermion states can be captured at the LHC, and some other specific signals, like top flavor-violating decays, can be detected as well [@rslhc]. Since Higgs is the [*most-wanted*]{} entity at the LHC, our intention in this paper is to calculate how the upper limit on the lightest supersymmetric neutral Higgs mass changes in the warped extra-dimensional backdrop due to radiative corrections induced by the KK towers of fermions and sfermions. Before we perch on extra-dimensional details, we mention that even within the 4d set-up the Higgs mass receives additional contribution, beyond the MSSM limit of 135 GeV, in the next-to-minimal MSSM [@Drees:1988fc] and in the left-right MSSM [@Zhang:2008jm], to the tune of a few tens of a GeV in cach case. [**5d warped MSSM**]{}:  The fifth dimension $y$ is compactified on an $S^1/Z_2$ orbifold of radius $R$. Two 3-branes are located at the orbifold fixed points at $y = (0,\pi R)$. The space-time between the two branes is a slice of AdS$_5$ geometry. The 5d metric is given by, $$\label{metric} ds^2=e^{-2\sigma}\eta_{\mu\nu}dx^\mu dx^\nu+dy^2 ,~~~{\rm where}~~ \sigma=k|y|~.$$ Above, $1/k$ is the AdS curvature radius and $\eta_{\mu\nu}={\rm diag}(-1,1,1,1)$. The 3-brane at $y=0$ is called the Planck brane as the natural mass scale associated with it is $M_P$, while the 3-brane at $y=\pi R$ with an effective mass scale $M_P e^{-\pi kR}$ can be called a TeV brane provided $kR\simeq 12$. We consider a supersymmetric scenario in which not only gravity but all particles can, in principle, access the AdS bulk [@Gherghetta:2000qt]. There are quite a few boons of supersymmetrizing an ${\rm AdS_5}$ slice, e.g. a possible connection to string theory can be established, new avenues for SUSY breaking can be explored, etc. We concentrate on the structure of matter hypermultiplets and the nature of Yukawa interactions for their special relevance in the computation of radiative corrections to the Higgs mass. The hypermultiplet $\Phi=(\phi^i,\Psi)$ contains two complex scalars $\phi^i$ ($i=1,2$) and a Dirac fermion $\Psi$. The action can be written as ($g \equiv {\rm det} (g_{MN})$) [@Gherghetta:2000qt] $$\label{kinhyper} S_5=-\int d^4x\int dy\sqrt{-g}\, \Bigg[ \left|\partial_M \phi^i \right|^2+i\bar{\Psi}\gamma^MD_M\Psi +m^2_{\phi^i}|\phi^i|^2+im_\Psi\bar{\Psi}\Psi\Bigg]\, .$$ Invariance under supersymmetric transformation yields [@Gherghetta:2000qt] $$\begin{aligned} \label{h:susycon} m^2_{\phi^{1,2}}&=&(c^2\pm c-\frac{15}{4})k^2 +\left(\frac{3}{2}\mp c\right) \frac{d^2\sigma}{dy^2} \, , ~~ {\rm and} ~~ m_\Psi = c\frac{d\sigma}{dy} \, ,\end{aligned}$$ where $c$ is some arbitrary real number. A generic 5d field can be decomposed as [@Gherghetta:2000qt] $$\label{Kaluza-Klein} \Phi(x^\mu,y)={1\over\sqrt{2\pi R}}\sum_{n=0}^\infty \Phi^{(n)}(x^\mu)f_n(y) , ~~{\rm where}~~ f_n(y)=\frac{e^{s\sigma/2}}{N_n}\left[J_\alpha(\frac{m_n}{k}e^{\sigma}) +b_{\alpha}(m_n)\, Y_\alpha(\frac{m_n}{k}e^{\sigma})\right]\, ,$$ with $s = (4,1,2)$ for $\Phi=\{\phi,e^{-2\sigma}\Psi_{L,R},A_\mu\}$. For the detailed formulae of $b_{\alpha}$ and the normalization $N_n$ in terms of the Bessel functions $J_\alpha$ and $Y_\alpha$ (where $\alpha$ is an index for 5d mass and is a function of $c$), see Ref. [@Gherghetta:2000qt]. Regardless of $Z_2$ even or odd nature of KK modes, one obtains an approximate expression of the KK mass: $$\label{kkhypmassapp} m^{(n)}\simeq (n+\frac{c}{2}-\frac{1}{2}) \pi k e^{-\pi k R}~,~{\rm where}~ n=1,2,..$$ Two points are worth noting: (i) Even though 5d $N=1$ SUSY is equivalent to $N=2$ in 4d, the massless sector of the hypermultiplet (from $\phi^1$ and $\Psi_L$, in our convention) forms an $N=1$ chiral supermultiplet, and (ii) in the conformal limit ($c=1/2$), the massless modes are not localized in the bulk and couple to the two branes with equal strength. For $c<1/2$, the zero modes are confined towards the TeV brane, while for $c>1/2$, the zero modes are localized closer to the Planck brane. We now come to the Yukawa interaction. First, we assume that the Higgs boson is localized at the TeV brane, i.e. $H(x,y) = H(x) \delta(y-\pi R)$ \[this immediately solves the $\mu$ problem, as $\mu \sim \cal{O}$ (TeV)\]. Considering that for each fermion flavor $i$, there are two 5d Dirac fermions $\Psi_{iL}(x,y)$ and $\Psi_{iR}(x,y)$, one can write the Yukawa action as ($H(x) \equiv H_u(x) ~{\rm or}~ H_d(x)$) $$\label{5dimenyc} S_y=\int d^4x \int dy\, \sqrt{-g}\,\,\lambda_{ij(5d)} H(x) \Big( \bar\Psi_{iL}(x,y)\Psi_{jR}(x,y) + {\rm h.c.} \Big) \delta(y-\pi R) \, .$$ Recall that each 5d fermion field has a bulk mass term, characterized by $c_{iL}$ or $c_{iR}$. For simplicity, we assume that $c_i \equiv c_{iL}= c_{iR}$. We now expand the 5d fermion fields in zero modes and higher KK modes and obtain the corresponding 4d Yukawa couplings. For simplicity, we consider only the diagonal couplings, i.e. ignore quark mixings as their numerical effects are negligible for our calculation. The Yukawa couplings of the zero mode fermions are given by $$\label{yc} \lambda_{i}= \lambda_{i(5d)}k (1/2-c_i) \left(1- e^{(2c_i-1)\pi kR} \right)^{-1} \, .$$ Now we assume $\lambda_{i(5d)} k \sim 1$, and trade the zero mode fermion masses in favor of the corresponding $c_i$ (see Table 1). This is how the fermion mass hierarchy problem is addressed. We note here that the choice of $c_i > 1/2$ for the first two families helps evade tight constraints ($m^{(1)} >$ a few TeV) from FCNC processes [@bulksm]. For the third generation, FCNC constraints are not so stringent any way. $f_i$ $e$ $\mu$ $\tau$ $u$ $d$ $c$ $s$ $t$ $b$ ------------- ------ ------- -------- ------ ------ ------ ------ ------- ------ $c_i$ 0.61 0.52 0.40 0.62 0.57 0.52 0.52 -0.50 0.26 $m^{(1)}_i$ 1598 1508 1388 1609 1562 1504 1510 500 1249 : The $c_i$ parameters and $m^{(1)}_i$ (in GeV) for different flavors are shown for $kR=11.93$ and $\tan\beta= {\langle H_u^0\rangle}/{\langle H_d^0\rangle} = 10$. For this choice, the mass gap between the consecutive KK states is $m^{(n+1)} - m^{(n)} = 1987$ GeV, irrespective of $c_i$. The corresponding $n=1$ KK mass for gauge boson is $1490$ GeV.[]{data-label="table"} -40pt We now turn our attention to the Yukawa couplings of KK fermions. We [*assume*]{} KK number conservation at the tree level Higgs coupling with the KK fermions[^1]. Inserting the expansion in Eq. (\[Kaluza-Klein\]) in Eq. (\[5dimenyc\]), we derive (again, considering only diagonal couplings): $$\begin{aligned} \label{kkyukawa-full} {\lambda}_{i}^{(n)}& = & {\lambda}_{i(5d)} \pi m^{(n)} e^{\pi kR} {\left[ J_{\alpha}\Big(\frac{m^{(n)}}{k}e^{\pi kR}\Big)+b_{\alpha}(m^{(n)})Y_{\alpha}\Big(\frac{m^{(n)}}{k}e^{\pi kR}\Big) \right] }^2 \, .\end{aligned}$$ A set of (reasonable) approximations $m_n \ll k \sim M_P$, $kR \gg 1$ and ${\lambda}_{i(5d)}k \sim 1$ simplifies the above as, $$\label{kkyukawa} {\lambda}_{i}^{(n)} \sim {\cos}^2 \Big(\left[ n-\frac{3}{4} \mp \frac{1}{4} \right] \pi \Big) \, ,$$ where $\mp$ correspond to $Z_2$ even/odd KK modes. We draw two important conclusions from Eq. (\[kkyukawa\]): (i) all KK Yukawa couplings, regardless of their flavors (i.e. $c_i$ values) and KK numbers, are roughly equal being close to unity (more precisely, ${\lambda}_{i(5d)}k$), and (ii) the KK Yukawa couplings of $Z_2$ odd modes are vanishing (since the Higgs is brane-bound). [**Radiative correction to the Higgs mass**]{}:  We recall that in 4d MSSM the lightest neutral Higgs mass at the tree level is lighter than $M_Z$, the $Z$-boson mass. More specifically, $m_h \leq {\rm min}~(m_A, M_Z) |\cos 2\beta| \leq {\rm min}~(m_A, M_Z)$, where $M_A$ is the pseudo-scalar Higgs mass. The radiative correction to $m_h$, dominated by the top quark Yukawa coupling and the masses of the stop squarks, is given by $\Delta m_h^2 \simeq (3 m_t^4/2\pi^2 v^2) \ln (m_{\tilde{t}}^2/m_t^2)$. Here, $v^2 = {\langle H_u^0\rangle}^2 + {\langle H_d^0\rangle}^2 = 1/(\sqrt{2} G_F) $ and $m_{\tilde{t}} = \sqrt{m_{\tilde{t}_1} m_{\tilde{t}_2}}$ is an average stop mass [@dj]. Here, our goal is to compute $\Delta m_h^2$ induced by the KK fermions. We shall follow the effective potential technique to implement this correction. We shall ignore the contributions from gauge sector as they are not numerically significant. Dominant effects arise from matter sector. The full one-loop effective potential is given by $V_1(Q) = V_0(Q) + \Delta V_1(Q)$ where, in terms of the field dependent masses $M(H)$, $\Delta V_1(Q) = {(64\pi^2)}^{-1}{\rm Str} M^4(H) \left \{ \ln(M^2(H)/Q^2) - 3/2 \right \}$. The scale dependence of $\Delta V_1(Q)$ cancels against that of $V_0(Q)$ making $V_1(Q)$ scale independent up to higher loop orders. The supertrace is defined through ${\rm Str} f(m^2) = \sum_i (-1)^{2J_i} (2J_i+1) f(m_i^2)$, where the sum is taken over all members of a supermultiplet. The contribution from the chiral multiplet labeled by the fermion $f$ is given by (with $N_c$ as the color factor) $$\label{epatop} \Delta V_{f} = {N_c \over {32\pi^2}} \left\{ m_{\tilde{f_1}}^4 \left(\ln{m_{\tilde {f_1}}^2\over{Q^2}}-{3\over2}\right) + m_{\tilde {f_2}}^4 \left(\ln{m_{\tilde {f_2}}^2\over{Q^2}}-{3\over2}\right) -2 m_{f}^4 \left(\ln{m_{f}^2\over{Q^2}}-{3\over2}\right)\right\} \, .$$ For illustration, we take only the up-quark chiral multiplet. The field dependent KK quark masses are, $$\label{mthmbh} \left(m^{(n)}_{u}\right)^2(H) = \left(\lambda^{(n)}_{u}\right)^2 |H_u^0|^2 + \left(m^{(n)}\right)^2 ~,$$ while the KK mass square matrix of the up-squark is obtained as, $$\begin{aligned} && {M_{\tilde {{u}}}}^{(n)2} (H) = \left(\begin{array}{cc} m_Q^2 + |H^0_{u}|^2 \left(\lambda^{(n)}\right)^2 & {\lambda^{(n)}} (A_{{u}} H^0_{u}+\mu{H^0_d}^*)\\ {\lambda^{(n)}} (A_{{u}}{H^0_u}^* +\mu H^0_d) & m_U^2 +|H^0_u|^2 \left(\lambda^{(n)}\right)^2 \end{array}\right) + \left(\begin{array}{cc} \left(m^{(n)}\right)^2 & 0\\ 0 & \left(m^{(n)}\right)^2 \end{array}\right) \, . \label{tsquark}\end{aligned}$$ We treat the soft SUSY breaking parameters ($m_{Q,U}^2$, $A_u$), $\mu$ and $\tan\beta$ as phenomenological inputs, assuming that the soft terms add in [*quadrature*]{} (see Eq. (\[tsquark\]))[^2]. The radiative contribution to the (zero mode) CP-even Higgs mass-square matrix from the $n$th KK quark/squark mode is given by, $$\begin{aligned} \label{msqevencor} \left. \Delta{\cal{M}}^2\right|_{(n)ij} = \left(N_c/{4\pi^2 v^2}\right) \left(\Delta^u_{ij} \right)^n \, , $$ where, putting $\lambda^{(n)} = 1$ following Eq. (\[kkyukawa\]), and denoting $g(a,b) \equiv 2-\{(a+b)/(a-b)\}\ln(a/b)$, $$\begin{aligned} \label{deltaup} (\Delta_{11}^u)^n &=& {v_u^4\over{{4}{\rm sin}^2\beta}}\left(\mu (A_u+\mu {\rm cot}\beta)\over{m_{\tilde u_1^n}^2 - m_{\tilde u_2^n}^2}\right)^2g(m_{\tilde u_1^n}^2,m_{\tilde u_2^n}^2) ,\nonumber \\ (\Delta_{12}^u)^n &=& {v_u^4\over{{4}\sin^2\beta}}{\mu (A_u+\mu \cot\beta)\over{m_{\tilde u_1^n}^2 - m_{\tilde u_2^n}^2}}\left[\ln {{m_{\tilde u_1^n}^2}\over{m_{\tilde u_2^n}^2}}+{A_u(A_u+\mu \cot \beta)\over{m_{\tilde u_1^n}^2 - m_{\tilde u_2^n}^2}}g(m_{\tilde u_1^n}^2,m_{\tilde u_2^n}^2)\right] , \\ (\Delta_{22}^u)^n& =& {v_u^4\over{{4}{\rm sin}^2\beta}}\left[{ \ln}{{m_{\tilde u_1^n}^2}{m_{\tilde u_2^n}^2}\over{m_{u^n}^4}} + {2A_u(A_u+\mu \cot\beta)\over{m_{\tilde u_1^n}^2 -m_{\tilde u_2^n}^2}} \ln {{m_{\tilde u_1^n}^2}\over{m_{\tilde u_2^n}^2}} + \left(A_u(A_u+\mu \cot\beta)\over{m_{\tilde u_1^n}^2 - m_{\tilde u_2^n}^2}\right)^2g(m_{\tilde u_1^n}^2,m_{\tilde u_2^n}^2) \right] \nonumber \, .\end{aligned}$$ The expressions for $(\Delta_{11}^d)^n$ can be written [*mutatis mutandis*]{}. A comparison with what happens in flat space supersymmetric Universal Extra Dimension (UED) [@Bhattacharyya:2007te] is now in order. In UED, the KK states are equispaced (due to space-time flatness), and the KK Yukawa couplings are proportional to the corresponding zero mode masses. In the warped scenario, the KK states have a sparse spectrum following the zeros of the Bessel function, and the KK Yukawa couplings are, to a good approximation, independent of the flavor indices and are all close to unity for a reasonable choice of extra-dimensional parameters. In our warped SUSY scenario, for low and moderate $\tan\beta (\sim 10)$, only $m_t^{(1)} \sim 500$ GeV is light, which is a consequence of appropriately choosing the $c_i$ parameters for correctly reproducing zero mode fermion masses. For large $\tan\beta(\sim 40)$, although $m_b^{(1)}$ can be as light as 500 GeV, the prefactor $v_d^4$ that appears in $(\Delta_{ij}^d)^n$ (not shown explicitly) suppresses the radiative contribution to the Higgs mass induced by the down-type chiral multiplets. So in the warped case, only $u^{(1)}, c^{(1)}$ and especially $t^{(1)}$ multiplets contribute to $\Delta m_h^2$ in a numerically significant way. The contributions from higher KK states are negligible. This is in sharp contrast with the SUSY UED scenario where the first [*few*]{} $t^{(n)}$ (and [*not*]{} $u^{(n)}$ or $c^{(n)}$) chiral multiplets provide sizable contribution to $\Delta m_h^2$. The net numerical effects in the two cases are comparable. Recall that in UED, unlike in the warped case, the KK spectra are not linked to fermion mass hierarchy. {width="70.00000%"} {width="70.00000%"} [**Numerical Results**]{}:  For simplicity, we have assumed a common soft mass $M_S\equiv m_Q=m_U=m_D$. The trilinear couplings $A_u$ and $A_d$ have been varied in the range $[0.8-1.2]M_S$, which resulted in bands in the figures. In Fig. 1, we have plotted our results in the $m_h$-$m_A$ plane. We have used $m_t^{(1)}$, the lightest KK top quark mass ([*modulo*]{} the zero mode mass), to represent the extra-dimensional effect. In Fig. 1, we have displayed the effects for two choices of $m_t^{(1)}$, namely, 500 and 1000 GeV, and for a moderate $\tan\beta=10$. In Fig. 2, we have demonstrated that $m_h$ indeed falls with increasing $m_t^{(1)}$, eventually attaining its 4d value. In this plot, we have set $A_u=A_d=\sqrt{6}M_S$, which maximises not only the 4d MSSM radiative correction but also the KK-induced one, which is why we have used the symbol $m_h^{\rm max}$. The three lines correspond to three different choices of $M_S =$ 500, 750 and 1000 GeV. All in all, $m_h$ increases by a few to several tens of a GeV, depending on the choice of soft SUSY breaking parameters, the radiative contribution coming primarily from all up-type multiplets. Beyond $m_t^{(1)} = 2$ TeV (which corresponds to an average value of $\sim$ 6 TeV for other first level KK fermions and gauge bosons), the KK effects almost vanish and the three falling curves become flat lines approaching their 4d limits (of course, without ever meeting as the three $M_S$ values are different). [**Conclusions**]{}:  We have calculated one-loop correction to the lightest neutral Higgs boson mass in a generic MSSM embeded in a slice of AdS$_5$. For a reasonable choice of warped space parameters, the 4d upper limit of 135 GeV could be relaxed by as much as $\sim$ (50-100) GeV depending on $M_S$. A few other closely related highlights are the following: (i) matter KK spectra are controlled by the $c_i$ parameters, which, in turn, are determined by the zero mode fermion masses; (ii) all KK Yukawa couplings are close to unity to a very good approximation; (iii) only $m_t^{(1)}$ could be light ($\sim$ 500 GeV) for a moderate $\tan\beta$ ($m_b^{(1)}$ can also be light too for large $\tan\beta \sim 40$) - this feature can be used to discriminate this scenario from the others producing similar shift to $m_h$; (iv) small values of $\tan\beta (\ltap~3)$, which are otherwise disfavored in 4d MSSM due to nonobservation of Higgs up to 114.5 GeV [@lep], are now resurrected thanks to an additional KK-induced radiative corrections. Admittedly, the stability of the proton would require further care [@Gherghetta:2000qt]. Besides, the warped models with fermions in the bulk, in general, pass the electroweak precision tests (EWPT) with some difficulty [@Hewett:2002fe], unless the KK mass is raised to tens of a TeV. To suppress excessive contribution to $T$ (or $\Delta \rho$), gauge symmetry in the bulk is enhanced to ${\rm SU(2)_L \times SU(2)_R \times U(1)_{B-L}}$ [@Agashe:2003zs], while to keep the contributions to $Zb_L\bar b_L$ vertex and other loop corrections under control, a further discrete $L \leftrightarrow R$ symmetry has been employed [@Carena:2007ua]. This allows us to consider $m_t^{(1)}$ as light as 500 GeV (the lightest KK gauge boson is then 1.5 TeV). Furthermore, $\tan\beta$ can be tuned to reduce the contribution to $T$. Since our primary intention here has been to develop a simple analytic framework to (for the first time) compute KK-induced radiative corrections to $m_h$ in a supersymmetric warped space, we pared the scenario down to its bare minimum. The further details necessary to overcome the above constraints are unlikely to alter the essential qualitative and quantitative features we explored here. Finally, we note that in a general class of such models, KK-parity violating gauge interactions would induce the lightest KK particle decay into zero mode SM particles well before nucleosynthesis sets in, thus without disturbing any cosmological constraints. The dark matter candidate (which could very well be the lightest zero mode neutralino, provided $R$-parity is conserved) would still be decided on the basis of the 4d spectra controlled by our choice of soft masses. Moreover, the lowest KK excitations of the SM particles could be heavier or lighter than the zero mode superparticles, again depending on our choice of the zero mode supersymmetry breaking soft parameters. A detailed phenomenological study preparing a catalogue of all such possibilities is beyond the goal of this paper. [ **Acknowledgements:**]{} We thank E. Dudas and A. Raychaudhuri for valuable comments. GB acknowledges hospitality at LPT-UMR 8627, Orsay, and a CNRS fellowship during a part of this work, and a partial support through the project No. 2007/37/9/BRNS of BRNS (DAE), India. TSR acknowledges the S.P. Mukherjee fellowship of CSIR, India. [99]{} L. Randall and R. Sundrum, Phys. Rev. Lett.  [**83**]{} (1999) 3370. W. D. Goldberger and M. B. Wise, Phys. Rev. Lett.  [**83**]{} (1999) 4922; W. D. Goldberger and M. B. Wise, Phys. Lett.  B [**475**]{} (2000) 275; H. Davoudiasl, J. L. Hewett and T. G. Rizzo, Phys. Lett.  B [**473**]{} (2000) 43; A. Pomarol, Phys. Lett.  B [**486**]{} (2000) 153; S. Chang, J. Hisano, H. Nakano, N. Okada and M. Yamaguchi, Phys. Rev.  D [**62**]{} (2000) 084025. T. Gherghetta and A. Pomarol, Nucl. Phys.  B [**586**]{} (2000) 141. S. J. Huber and Q. Shafi, Phys. Lett.  B [**498**]{} (2001) 256. Y. Grossman and M. Neubert, Phys. Lett.  B [**474**]{} (2000) 361. K. R. Dienes, E. Dudas and T. Gherghetta, Nucl. Phys.  B [**567**]{} (2000) 111. T. Gherghetta and A. Pomarol, Nucl. Phys.  B [**602**]{} (2001) 3. K. Agashe, A. Belyaev, T. Krupovnickas, G. Perez and J. Virzi, Phys. Rev.  D [**77**]{} (2008) 015003; K. Agashe, G. Perez and A. Soni, Phys. Rev.  D [**75**]{} (2007) 015002. M. Drees, Int. J. Mod. Phys.  A [**4**]{} (1989) 3635; U. Ellwanger and C. Hugonie, Phys. Lett.  B [**623**]{} (2005) 93. Y. Zhang, H. An, X. d. Ji and R. N. Mohapatra, arXiv:0804.0268 \[hep-ph\]. K. Agashe, A. Falkowski, I. Low and G. Servant, JHEP [**0804**]{} (2008) 027. See, for example, A. Djouadi, Phys. Rept. [**459**]{} (2008) 1, and references therein. G. Bhattacharyya, S. K. Majee and A. Raychaudhuri, Nucl. Phys.  B [**793**]{} (2008) 114. See also, N. Uekusa, arXiv:0806.3229 \[hep-ph\]. J. Alcaraz [*et al.*]{} \[LEP Collaboration\], “A combination of preliminary electroweak measurements and constraints on the standard model,” arXiv:hep-ex/0612034. J. L. Hewett, F. J. Petriello and T. G. Rizzo, JHEP [**0209**]{} (2002) 030; C. Csaki, J. Erlich and J. Terning, Phys. Rev.  D [**66**]{} (2002) 064021. K. Agashe, A. Delgado, M. J. May and R. Sundrum, JHEP [**0308**]{} (2003) 050. M. S. Carena, E. Ponton, J. Santiago and C. E. M. Wagner, Phys. Rev.  D [**76**]{} (2007) 035006; M. S. Carena, E. Ponton, J. Santiago and C. E. M. Wagner, Nucl. Phys.  B [**759**]{} (2006) 202. [^1]: Although, unlike in UED, KK-parity is not automatic in the warped scenario, it is still possible to implement it in a slice of AdS$_5$ [@Agashe:2007jb]. We assume this parity for simplicity of our analytic computation. This also helps in evading some FCNC constraints. [^2]: SUSY breaking by twisted boundary conditions on fermions leads to a [*linear*]{} splitting ($0.5 \pi k~ e^{-\pi kR}$) [@Gherghetta:2000kr]. The numerical effects on $\Delta m_h^2$ would be covered within the parameter space we scanned (see figures later).

ArXiv

--- abstract: 'The magnetic field dependent localization in a disordered quantum wire is considered nonperturbatively. An increase of an averaged localization length with the magnetic field is found, saturating at twice its value without magnetic field. The crossover behavior is shown to be governed both in the weak and strong localization regime by the magnetic diffusion length $L_B$. This function is derived analytically in closed form as a function of the ratio of the mean free path $l$, the wire thickness $W$, and the magnetic length $l_B$ for a two-dimensional wire with specular boundary conditions, as well as for a parabolic wire. The applicability of the analytical formulas to resistance measurements in the strong localization regime is discussed. A comparison with recent experimental results is included.' address: 'I. Institut f. Theoretische Physik, Universit" at Hamburg, Germany' author: - 'Stefan Kettemann, Riccardo Mazzarello' title: ' Magneto-localization in Disordered Quantum Wires ' --- INTRODUCTION {#sec:introduc} ============ The phase coherent movement of electrons in a disorder potential can result in strong localization due to quantum interference [@reviews; @ef]. As soon as the localization length $L_c$ becomes smaller than the size of the sample $L$ and the phase coherence length, $L_{\phi}$, the resistance increases exponentially. The strong localization due to quantum interference is known to depend on the global symmetry of the disordered electron system [@larkin]. In disordered quantum wires, the localization length is $$\label{locsym} L_c = \beta \pi \hbar \nu S D_0,$$ where $\beta =1, 2, 4$, corresponding to no magnetic field, finite magnetic field, and strong spin- orbit scattering or magnetic impurities, respectively. $ \nu (E) $ is the electronic density of states in the wire. $ D_0 = \tau v_F^2/d $ is the classical diffusion constant of the electrons in the wire, with $\tau$ the elastic scattering time, $v_F$ the Fermi velocity, and $ d$ the dimension of classical diffusion. $S$ is the wire crossection. This result was first obtained by calculating the spatial decay of the density correlation function for wires with diffuse crossections and many transversal channels $N \gg 1$. It can also be obtained by calculating the transmission probability through thin, few channel wires upto a correction of order $1/N$: $ L_c = (\beta N + 2 - \beta ) l$ [@reviews], where $l= v_F \tau$ is the mean free path, and $\beta =1, 2, 4$, as defined above. This correction ensures that the localization length is for a single channel, $N=1$, independent of $\beta$, $L_c = 2 l$. Recently, the doubling of the localization length was observed in sub-micron thin wires of Si $\delta$- doped Ga As structures by Khavin, Gershenson and Bogdanov, who found a continously decreasing activation energy when the magnetic field is increased, saturating indeed at one half of its field free value [@khavin]. This symmetry dependence of the localization properties of quantum wires allows to test our present theoretical understanding by detailed comparison with the experiment. The quantum wires used in the experiment have mean free paths which are smaller than or comparable to their thickness. Also, in addition to the disorder in the bulk due to the random electrostatic potential of the donor impurities, there is an unspecified surface roughness which may influence the classical mobility of the wires as well as its quantum transport properties. Therefore, a more detailed analysis of the localization length as function of these parameters is called for, in order to be able to compare the theory with the experimental results quantitatively. In the next section we review the known weak localization corrections to the conductivity in disordered quantum wires and their magneto-sensitivity as function of mobility, wire thickness, and electron density [@altaronov; @dugaev; @aronov; @beenakker]. In the third section, the non-perturbative theory of localization in disordered electron systems[@ef] is extended, in order to allow the study of wires with ballistic crossections. In the fourth section, the magnetic phase shifting rate is introduced and identified with a correlation function of the magnetic vector potential, relating it to the coefficient of the time reversal symmetry breaking term in the nonlinear sigma model. This expression for the magnetic phase shifting rate, is calculated anayltically for arbitrary ratios of the mean free path $l$ and the width of the wire $W$, and compared with previously derived analytical and numerical results[@dugaev; @beenakker] for a wire with specular boundary scattering. Next, it is calculated for a wire with harmonic confinement which allows to extend the analysis to stronger magnetic fields, when the cyclotron radius, $l_{\rm C}$ is smaller than the the wire thickness $W$, but still larger than the elastic mean free path. In that regime a new enhancement mechanism for the magnetic phase shifting rate leading to a stronger magneto-sensitivity, is identified. In the fifth section, the autocorrelation function of spectral determinants (ASD)[@haake; @us] is considered for a coherent disordered quantum wire, which shows the expected crossover from Wigner- Dyson statistics[@dyson], typical for a spectrum of extended states in phase coherent disordered metal systems[@ef], to Poisson statistics, corresponding to a spectrum of localized states [@efetov1982; @altshk; @isa; @altfuchs; @mirlin; @guhr], as the length of the wire is increased beyond a localization length $L_C$, as reported earlier[@prb]. This crossover length scale to Possionian statistics is used to derive the averaged localization length of disordered quantum wires, and it is shown that it yields the correct symmetry dependence, Eq. (\[locsym\]). A comparison with the result of the supersymmetric theory of the two-terminal conductance of a disordered quantum wire, is given. It is concluded, that the definition of an averaged localization length, by the decay of an energy level correlation function, can be used to consider analytically the magnetic field dependence of the localization length. Thereby, analytical formulas for the localization length as a function of wire width, mean free path and magnetic field are derived. In the sixth section, the theory of finite tempreature magnetoresistance in quantum wires is discussed. In particular, the variable range hopping conductivity in quantum wires is reviewed for various temperature and dimensional regimes. It is shown that in a wide temperature regime the resistance has an activated behaviour, and that therefore, the activation gap can be directly measured and related to the localization length of the electrons in the wire. This allows a comparison of the analytical results for the magnetic field dependence of the localization length with these experimental results, as done in the seventh section. In appendix A, the functional integral representation of the ASD by Grassmann intergals is given, and the averaging over disorder is performed. In appendix B the derivation of the magnetic phase shifting rate is given. In appendic C the representation of the matrix fields $Q$ is given, and their Laplacian derived. Weak Localization =================== Classically, the transport of a disordered conductor is characterized by its mobility $\mu = q \tau/m$ and the electron density $n$ related to the the classical Drude conductivity $ \sigma_0 = n q^2 \tau/m$. Alternatively, it can be characterized by the diffusion constant $D$, which is in a metal related to the conductivity by the Einstein relation $ \sigma_0 = 2 q^2 \nu D $. When the electrons diffuse coherently, quantum interference without magnetic field results in a suppression of the conductivity of a quantum wire of order[@reviews; @ab; @gorkov; @khm; @elk; @schmid] $$\frac{\Delta \sigma}{\sigma_0} = - \frac{2 }{\sqrt{2 \pi^3 }} ( \frac{\sqrt{\tau_{\varphi}}}{\sqrt{\tau}} - 1 ),$$ where $\tau_{\phi}$ is the phase coherence time, that increases when decreasing the temperature as a power law: $$\tau_{\phi} \sim T^{-\gamma}$$ and defines the phase coherence length, which an electron diffuses coherently, $ L_{\phi} = ( D \tau_{\phi} )^{1/2} $. Quasi elastic electron-electron scattering can be the dominant low temperature dephasing mechanism and yields $\gamma = 2/3$ for a 1-d wire and $\gamma=1$ for a 2-d film [@aronov; @phase2]. At higher temperatures the exponent crosses over to $\gamma = 4$ due to electron-phonon scattering at temperatures $ k_{\rm B} T \ll ( \hbar^2/\tau \epsilon_F) \Omega_D$ where $\Omega_D$ is the optical Debye phonon frequency. This power can be smaller, due to the confinement, in quantum wires. The above definition of the phase coherence rate is not applicable when approaching the localized regime, and the phase coherence length is larger than the localization length $L_c$. Also, there are mechanisms which may lead to a saturation of $\tau_{\phi}$ below $T=1 K$, as observed in a wide range of conductors [@phase1; @buettikerphase]. A magnetic field breaks the time reversal symmetry. Therefore, the magnetic phase accumulated in a Brownian motion of electrons, enters effectively as an additive contribution to the phase coherence rate, diminishing the weak localization corrections of the conductivity[@khm]. For wires with diffusive width $W > l$, it varies quadratically with the magnetic field, $1/\tau_{\phi} (B) = 1/\tau_{\phi} + D \frac{q^2}{\hbar^2} S B^2/K_D$, where S is the crossection of the wire, and the constant $K_D$ depends on the geometry of the wire, the direction of the magnetic field and the scattering mechanisms [@altaronov]. For example, for a 2-dimensional wire of diffusive crossection in a perpendicular magnetic field, it yields, $K_D=3$. In this way, the conductivity increases to its classical value, when the magnetic field is turned on. For a wire with ballistic crossection and a magnetic field being perpendicular to its crossection, the magnetic field dependence of the weak localization correction to the conductivity is weakened by flux cancellation effects due to boundary scattering[@dugaev]. If the magnetic field is so small that less than one flux quantum $ \phi_0 = h/e $ is penetrating an area $ W l$, the effective dephasing rate $1/\tau_{\phi} (B) $ is quadratically increasing as for diffusive crossections. Its slope was found to be by at least a factor $W/l$ smaller, as a consequence of the flux cancellation effect of edge to edge skipping orbits[@dugaev; @beenakker]. When $B W l \gg \phi_0$, the effective dephasing rate $1/\tau_{\phi} (B) $ was found by a semiclassical method, to increase only linearly with the magnetic field $B$ in this regime[@dugaev; @beenakker]. In the presence of magnetic impurities, scattering the electrons with a rate $1/\tau_S$, there is no temperature dependence of the conductivity, if $1/\tau_S \gg 1/\tau_{\phi}$. Strong spin-orbit scattering reverses the sign of the quantum correction to the conductivity[@hikamiln]. The conductivity is then larger than classically expected. This can be observed by increasing an external magnetic field, which destroys time reversal invariance and acts through an effective decoherence time $1/\tau_{\phi} (B) = 1/\tau_{\phi}$ as noted above. In the case of moderately strong spin-orbit scattering, the conductivity decreases therefore when the magnetic field is turned on[@aronov]. At low temperatures, when the dephasing rate $1/\tau_{\phi}$ becomes smaller than the typical energy scale of strong localization, the local level spacing $ \Delta_C = 1/( \nu W L_C)$, a perturbation theory in the elastic scattering rate $1/\tau$ is no longer appropriate, and a nonperturbative treatment of disorder is called for, as the scaling theory of localization does indicate[@ab; @gorkov]. Nonperturbative theory of localization in disordered quantum wires =================================================================== In this section, the nonperturbative theory of disordered noninteracting electrons in quantum wires is derived[@weg; @elk; @ef]. Its action, governed by the long wave length modes corresponding to diffusion, the nonlinear sigma model is rederived, extending previous derivations, to allow the description of quantum wires with ballistic crossections. The Hamiltonian of disordered noninteracting electrons is $$H= \epsilon \left({\bf p }- q {\bf A} \right) + V({\bf x}) +{\bf \sigma b_s ( x) } + {\bf \sigma u_{SO}} \times { \bf p } ,$$ where $q$ is the electron charge. In the following, we will generally approximate the electronic dispersion $ \epsilon \left({\bf p }- q {\bf A} \right)$ by $\left({\bf p }- q {\bf A} \right)^2/(2 m)$, where $m$ is the effective electron mass, but note that higher moments are sometimes needed to regularize the correlation functions, calculated below. $V({\bf x})$ is taken to be a Gaussian distributed random function $\left<V({\bf x})\right> = 0$, and $\left<V({\bf x}) V({\bf x'})\right> =\hbar \Delta S L/(2 \pi \tau) \delta ({\bf x} - {\bf x'}),$ which models randomly distributed, uncorrelated impurities in the sample. $\Delta= 1/(\nu S L) $ is the mean level spacing. This corresponds to a Gaussian distribution function $ P( V ) = \exp ( - \frac{\pi \tau}{ \hbar \Delta } \int \frac{d {\bf x}}{Vol.} V ( {\bf x} )^2)$ of the disorder potential, defining the disorder average as $<...>_V = \int \prod_{{\bf x}} d V P(V) ... $. According to the central limit theorem, this is therefore a good description of the various sources of randomness in the electrostatic potential, in which the electrons are moving. The vector potential is used in the gauge ${\bf A} = (- B y, 0, 0 )$, where $x$ is the coordinate along the wire of length L, $y$ the one in the direction perpendicular both to the wire and the magnetic field ${\bf B}$, which is directed perpendicular to the wire. The angular brackets denote averaging over impurities. ${\bf \sigma }$ is the electronic spin operator, and ${\bf b_s ( x) }$ is a random magnetic impurity field. ${\bf u_{SO}}$ is the local electrostatic field of impurities with large atomic number $Z$, which do give a stronger spin orbit coupling to the conduction electrons. The Hamiltonian can be classified by its symmetry with respect to time reversal and spin rotation as summarized in Table 1. It has been noted that the averaged density of states or the averaged one-particle Green’s function does not contain any information on the localization of Eigenfunctions of the disordered Hamiltonian $H$[@weg]. The physical reason is, that the one-particle Green’s function describes the propagation of the wave function amplitude $\psi ( {\bf x} )$. Elastic impurity scattering randomizes the phase of the amplitude and therefore, this propagator decays on the scale of the mean free scattering time $\tau$. To catch classical diffusion and quantum localization, at least the evolution of the density or amplitude square has to be averaged over the disorder, leading to a correlation function of two one-particle Green’s functions. While weak localization corrections can be calculated within a diagrammatic perturbation expansion of such correalation functions [@altaronov; @schmid], the study of strong electron localization in a disordered potential, necessitates a nonperturbative averaging of such products of Green’s functions. This can be achieved by means of the super-symmetry method, whereby the product of Green’s functions is written as a functional integral[@ef]. Thus, the average over the form of the disorder potential can be done right at the beginning as a Gaussian integral, exactly. Here, for simplicity, we present the derivation of a simpler correlation function, which does not necessitate the use of the full super-symmetry method, but still contains some information on strong quantum localization, as shown recently[@prb; @prl; @prbr]. The statistics of discrete energy levels of a finite coherent, disordered metal particle is an efficient way to characterize its properties [@ef]. This can be studied by calculating a disorder averaged autocorrelation function between two energies at a distance $\omega$ in the energy level spectrum. Thereby, an uncorrelated spectrum of localized states can be distinguished from a correlated spectrum of extended states. The autocorrelation function of spectral determinants (ASD) is the most simple such spectral correlation function, which allows to explore complex quantum systems analytically, and still does contain nontrivial information on level statistics and, thus, on localization[@prb; @prl]. It is an oscillatory function whose amplitude decays with a power law, when the energy levels in the vicinity of the central energy $E$ are extended, while a Gaussian decay is a strong indication that all states are localized. It is defined by $ C(\omega) = \bar{C}(\omega)/ \bar{C}(0), \hspace{.5cm} \bar{C}(\omega) = \left<\mbox{det}( E + \omega/2 - H) \mbox{det} ( E- \omega/2 -H )\right>, $ where $E$ is a central energy. Since it is a product of two spectral determinants, and a spectral determinant can be written as a Gaussian functional integral over Grassmann variables $\psi$, $\psi^*$, one does need at least a 2-component Grassman field, one for each spectral determinant. In general, $4 \alpha$ -component Grassman fields are needed to get the functional integral representation of the ASD. Here, $\alpha=1$, when the Hamiltonian is independent of the spin of the electrons, and each level is doubly spin degenerate. There is one pair of Grassman fields for each determinant in the ASD and each pair is composed of a Grassman field and its time reversed one, as obtained by complex conjugation. $\alpha=2$ has to be considered, when the Hamiltonian does depend on spin, as for the case with moderately strong magnetic impurity or spin- orbit scattering. This necessitates the use of a vector of a spinor and the corresponding time reversed one. The representation as a Gaussian functional integral over Grassmann variables is given explicitly for $\alpha=1$ in appendix A. There, the averaging over disorder and the decoupling of the resulting $\psi^4$ interaction with a Gaussian integral over a matrix field $Q$ is given. Thus, the disorder averaged ASD is given by a functional integral over a matrix field $Q$. The matrix $Q$ is element of the full symmetric space, including rotations between the subspace corresponding to the left and the right spectral determinant. Therefore, the long wavelength modes of $Q$, do contain the nonperturbative information on the diffuson and Cooperon modes. In order to consider the action of long wavelength modes governing the physics of diffusion and localization, one can now expand around the saddle point solution of the action, satisfying for $\omega =0$, $$\label{saddle} Q = i/(\pi \nu) < {\bf x} \mid 1/( E - H_0 + i \hbar/(2 \tau) Q ) \mid {\bf x} >.$$ This saddle point equation is found to be solved by $ Q_0 = \Lambda$. For $\alpha=1$, and $B=0$, at $\omega =0$, the rotations $ U$, which leave $Q$ in the symplectic symmetric space yield the complete manifold of saddle point solutions as $ Q = \bar{U} \Lambda U$, where $ U \bar{U} = 1$, with $Q^T C = C Q$. The modes which leave $\Lambda$ invariant, elements of $Sp(1) \times Sp(1)$ are surplus, or spontanously broken, and can be factorized out, leaving the saddle point solutions to be elements of the symmetric space $ Sp(2)/( Sp(1) \times Sp (1)) $[@lie]. For $\alpha=2$ the matrix $C$ is, due to the time reversal of the spinor, substituted by $i \sigma_2 \tau_1$[@elk]. Both magnetic impurities and spin-orbit scattering reduce the Q matrix to unity in spin space. Thus, C has effectively the form $\tau_1$. The condition $Q^T C = C Q$ leads therefore to a new symmetry class, when the spin symmetry is broken but the time reversal symmetry remains intact. This is the case for moderately strong spin-orbit scattering. Then, $Q$ are $4 \times 4$- matrices on the orthogonal symmetric space $ O(4)/(O(2) \times O(2) )$ [@weg], which is the nonperturbative consequence of the sign change of a spinor component under time reversal operation, which leads to the positive quantum correction to the conductivity in perturbation theory [@schmid]. With magnetic impurities both the spin and time reversal symmetry is broken, and the Q- matrices are in the unitary symmetric space $U(2)/(U(1) \times U(1) )$ as for a moderate magnetic field and spin degenerate levels. The difference in the prefactor $\alpha$ remains. One can extend this approach to other compact symmetric spaces with physical realizations, see Ref. for a complete classification. In addition to these gapless transversal modes there are massive longitudinal modes with $Q^2 \neq 1$, which for $N \gg 1$, can be integrated out[@ef], and the ASD thereby reduces to a functional integral over the transverse modes $U$. Now, the action of finite frequency $\omega$ and spatial fluctuations of $Q$ around the saddle point solution can be found by an expansion of the action $F$, Eq. (\[exactfree\]). Inserting $ Q = \bar{U} \Lambda U$ into Eq. (\[exactfree\]), and performing the cyclic permutation of $U$ under the trace $Tr$, yields, $$F = - \frac{1}{2} \int d {\bf x} <{\bf x} \mid Tr \ln ( G_0^{-1} - U [ H_0, \bar{U} ] + \omega U \Lambda \bar{U} ) \mid {\bf x} >,$$ where $$G_0^{-1} = E - H_0 + \frac{i \hbar }{2 \tau} \Lambda.$$ Expansion to first order in the energy difference $\omega$ and to second order in the commutator $U [ H_0, \bar{U} ]$, yields, $$\begin{aligned} \label{free2} F[U] &=& - \frac{1}{2} \omega \int d {\bf x} <{\bf x} \mid Tr G_{0 E} U \Lambda \bar{U} \mid {\bf x} > \nonumber \\ &+& \frac{1}{2}\int d {\bf x} <{\bf x} \mid Tr G_{0 E} U [ H_0, \bar{U} ] \mid {\bf x} > \nonumber \\ &+& \frac{1}{4}\int d {\bf x} <{\bf x} \mid Tr (G_{0 E} U [ H_0, \bar{U} ])^2 \mid {\bf x} >.\end{aligned}$$ Note that $ [ H_0, \bar{U} ] = - \frac{\hbar^2}{2 m} ({\bf \nabla}^2 \bar{U} ) - \frac{\hbar^2}{ m}({\bf \nabla} \bar{U} ) {\bf \nabla} - \frac{q \hbar }{ i m c } ( \tau_3 {\bf A} {\bf \nabla} \bar{U} - \bar{U} \tau_3 {\bf A} {\bf \nabla}) $. The first order term in $U [ H_0, \bar{U} ]$ vanishes for Gaussian white noise isotropic scattering. In general, in order to account for the ballistic motion of electrons in ballistic wires, or to account for different sources of randomness, a directional dependence of the matric $ U = U ( {\bf x}. {\bf n} ) $, where $ {\bf n} = {\bf p }/\mid {\bf p} \mid$, has to be considered[@taras1; @blanter]. However, for the geometries considered in this article, we have found that the form of the action derived below remains valid for diffusive as well as ballistic crossections, when the vector fields ${\bf S}$ as intorduced in Refs. , are integrated out. This will be presented in more detail in a separate article. Then, one can keep second order terms in ${\bf \nabla} \bar{U}$ and $ {\bf A} $, which turns out to be valid for the regime of weak disorder, $ l \gg 1/k_F$ and for any magnetic field, $l_B \gg k_F$. Thus, one gets, using the saddle point equation, Eq. (\[saddle\]), $$\begin{aligned} \label{free3} F[U] &=& - \frac{\pi }{4} \frac{ \omega}{\Delta } \int \frac{ d {\bf x}}{S L} Tr \Lambda Q \nonumber \\ &+& \frac{1}{4}\int d {\bf x} <{\bf x} \mid Tr (G_{0 E} U ( \frac{\hbar^2}{2 m}({\bf \nabla} \bar{U} ) ({\bf \nabla} -\frac{i}{\hbar} q {\bf A} \tau_3) + \frac{q \hbar }{ m } [\tau_3, \bar{U} {\bf A} {\bf \nabla}]) )^2 \mid {\bf x} >.\end{aligned}$$ Next, one can separate the physics on different length scales, noting that the physics of diffusion and localization is governed by spatial variations of $U$ on length scales larger than the mean free path $l$. The smaller length scale physics, is then included in the correlation function of Green’s functions, being related to the conductivity by the Kubo-Greenwood formula, $$\label{kubo} \sigma_{\alpha \beta } ( \omega ) = \frac{\hbar}{\pi S L} \frac{q^2}{m^2}\sum_{\bf p, p'} (p_{\alpha} - q A_{\alpha}) ( p'_{\beta} - q A_{\beta} ) < {\bf p} \mid G^R_{0 E} \mid {\bf p'} > < {\bf p'} \mid G^A_{0 E+\omega} \mid {\bf p} > ,$$ where ${\bf p} =\frac{\hbar}{i} {\bf \nabla}$. The remaining averaged correlators, involve products $ G^R_{0 E} G^R_{0 E+\omega}$ and $ G^A_{0 E} G^A_{0 E+\omega}$ and are therefore by a factor $\hbar/(\tau E)$ smaller than the conductivity, and can be disregarded for small disorder $\hbar/\tau \ll E $. In the bulk of this article we are interested in the weak magnetic field limit, where $ \omega_c \tau \ll 1$, with the cyclotron frequency $ \omega_c = q B/m$. In this limit we can disregard the nondiagonal Hall conductivity and the explicit magnetic field dependence of the longitudinal conductivity. In order to insert the Kubo-Greenwood formula in the saddle point expansion of the nonlinear sigma model, it is convenient to rewrite the propagator in $F$ as $ G_{0 E} = \frac{1}{2} G^R_{0 E} ( 1 + \Lambda ) + \frac{1}{2} G^A_{0 E} ( 1 - \Lambda) $. Then, we can use, that $Tr [ \sum_{\alpha = 1}^d \sum_{s = \pm} ( 1 + s \Lambda ) U ( \nabla_{\alpha} \bar{U}) ( 1 - s \Lambda ) U ( \nabla_{\alpha} \bar{U})] = - Tr[ ( {\bf \nabla} Q)^2]$, and $ Tr [ \sum_{s= \pm} ( 1 + s \Lambda ) U [\tau_3, \bar{U}]] ( 1 - s \Lambda ) U [\tau_3, \bar{U}] = - Tr [ [\tau_3, Q ]^2 ] $. Thereby we can rewrite Eq. (\[free3\]) as $$\begin{aligned} \label{free4} F[Q] &=& - \frac{\pi }{4} \frac{ \omega}{\Delta } \int \frac{ d {\bf x}}{S L} Tr \Lambda Q \nonumber \\ &-& \frac{1}{4}\int d {\bf x} Tr[ ( {\bf \nabla} Q({\bf x} ))^2 <{\bf x} \mid G_{0 E}^R \frac{\hbar^2}{2 m} ({\bf \nabla} - \frac{i}{\hbar} q {\bf A} ) G_{0 E}^A \frac{\hbar^2}{2 m} ({\bf \nabla} - \frac{i}{\hbar} q {\bf A} ) \mid {\bf x} > \nonumber \\ &-& \frac{1}{4} (\frac{q \hbar }{ m })^2 \int d {\bf x} Tr[ [\tau_3, Q({\bf x}) ]^2 ] <{\bf x} \mid G_{0 E}^R {\bf A} {\bf \nabla} G_{0 E}^A {\bf A} {\bf \nabla} \mid {\bf x} > + c.c. .\end{aligned}$$ For wires of thickness $ W $ not exceeding the length scale $L_{C U} = L_C (\beta = 2) = 2 \pi \hbar \nu S D_0$, the variations of the field $Q$ can be neglected in the transverse direction, and the action reduces to the one of a one- dimensional nonlinear sigma model. Using the Kubo formula, Eq. (\[kubo\]), this functional of $Q$ thus simplifies, for $\omega_c \tau \ll 1$, to, $$\begin{aligned} \label{freeh} F = \frac{\pi \hbar}{16 q^2} \sigma (\omega=0) W \int_0^L d { x} ( Tr ( { \nabla_x} Q({ x} ))^2 - < A_x \bullet A_x > \frac{q^2}{\hbar^2} Tr [\tau_3, Q({ x}) ]^2 ). \end{aligned}$$ The prefactor of the time reversal symmetry breaking term, the correlation function $$\begin{aligned} < A_x \bullet A_x > &=& B^2 < y \bullet y > \nonumber \\ & = & \frac{(<{\bf x} \mid G_{0 E}^R {\bf A} {\bf \nabla} G_{0 E}^A {\bf A} {\bf \nabla} \mid {\bf x} > + c.c.)} {<{\bf x} \mid G_{0 E}^R ({\bf \nabla} - \frac{i}{\hbar} q {\bf A}) G_{0 E}^A ({\bf \nabla} - \frac{i}{\hbar} q {\bf A}) \mid {\bf x} >},\end{aligned}$$ is increasing with the magnetic field $B$, suppressing modes with $ [ Q, \tau_3 ] \neq 0 $, the Cooperon modes, arising from the self interference of closed diffusion paths. Accordingly, the symmetry of the $Q$- fields is broken from $Sp(2)/(Sp(1) \times Sp (1) )$ to $U(2)/( U(1) \times U(1) )$. In the next section it is shown that this prefactor is related to the magnetic phase shifting rate, and is evaluated for a disordered quantum wire. The Magnetic Phase Shifting Rate ================================ It can be seen that the prefactor of the symmetry breaking term in Eq. (\[freeh\]) is proportional to the effective phase shifting rate $1/\tau_B$, governing the weak localization suppression by a magnetic field. To this end, one can use the supersymmetric version of the above nonlinear sigma model, obtained by substituting the matrix $Q$ by supermatrices, and the trace over matrices $Tr$ by the supertrace $STr$, but keeping all coefficients the same as in Eq. (\[freeh\]). Then, the weak localization corrections to the conductivity can be calculated as outlined in [@ef], by an expansion of $Q$ around the classical saddle point $ Q_c = \Lambda$. Thus, the magnetic phase shifting rate $1/\tau_B$ can be identified as, $$\label{magneticphaseshifting} 1/\tau_B = 4 D \frac{q^2}{\hbar^2} < A_x \bullet A_x >,$$ where the Einstein relation $\sigma = 2 q^2 \nu D$ of the classical conductivity $\sigma$ to the classical diffusion constant $D$ has been used. 2D wire with specular boundary conditions ----------------------------------------- The general expression for the correlation function $ < y \bullet y > $, is found by inserting the momentum eigenstates of the wire and summing the correlation functions of Green’s functions for $l_B \gg W$ in Eq. (\[h\]). It is thus obtained to be given for a two dimensional wire of width $W$ in momentum representation by, $$\label{h} < y \bullet y > = \sum_{k_x,k_y,k_y'} k_x^2 ( G_{0 E}^R( k_x, k_y ) G_{0 E}^A ( k_x, k_y' ) + c.c. ) \mid < k_y \mid y \mid k_y' > \mid^2/ \sum_{k_x,k_y} (k_x-\frac{q}{\hbar} A_x )^2 G_{0 E}^R( k_x, k_y ) G_{0 E}^A ( k_x, k_y ).$$ Here, $G_{0 E}^{R/A} ( k_x, k_y ) = ( E - \hbar^2 ( k_x^2 + k_y^2 )/(2 m) \pm i/(2 \tau) $. Keeping all corrections for finite number of transverse channels $N = k_F W/\pi$ and effective mean free path $\lambda = k_F l$, in the weak disorder limit $ E \gg \hbar/\tau$, we get for $ N \gg 1$ the expression : $$\begin{aligned} \label{hexact} < y \bullet y > &=& W^2 ( \frac{1}{12} K - \frac{1}{ 2 \pi^2} K_1 - \frac{\lambda^2}{ \pi^2 N^2} K_2 \nonumber \\ &+&\frac{4}{\pi^4} \frac{\lambda^3}{ N^4} \sum_{s =1}^{N} \frac{s^2}{N^2} \sqrt{1- \frac{s^2}{N^2} } Im \sqrt{ \frac{s^2}{N^2} + i \frac{2}{\lambda} } \tan ( \frac{\pi N}{2} ( \sqrt{ \frac{s^2}{N^2} + i \frac{2}{\lambda} } - \frac{s}{N} ) ))/K_0 , \end{aligned}$$ where the definition of the constants $K_i$ is given in Appendix B . Its dependence on the mean free path parameter $\lambda = k_F l$ is shown in Fig. 1. \[fig1\] Note that, although $N \gg 1$ is required for the validity of the nonlinear sigma model, the equation (\[hexact\]) is valid for arbitrary ratios of the width of the wire $W$ and the mean free path $l$, since the motion remains diffusive along the wire axis on large length scales, even if $l \gg W$. For diffusive wire crossections, $l<W$, $< y \bullet y > \rightarrow \overline{ y^2 } = W^2/12 $ which results exactly in the known result for the magnetic phase shifting rate $1/\tau_B = 4 D \frac{q^2}{\hbar^2} \overline{y^2} B^2$ [@aronov; @beenakker]. The above derivation is more general, and applies for arbitary ratios of the wire thickness $W$ and the mean free path $l$, as long as the magnetic length $l_B$ is both larger than the width $W$ and the elastic mean free path $l$, and for a large number of transverse channels $ N = k_F W/\pi \gg 1$. For ballistic wire crossections, $l>W$, Eq. (\[hexact\]) shows, that the effect of the magnetic field becomes weaker, as $W/l$ decreases. This is a result of the flux cancellation effect, discussed in the limit of weak localization in Ref. : The matrix element of the vector potential $<{\bf k} \mid {\bf A} \mid {\bf k'}>$ vanishes for ${\bf k} = {\bf k'}$, since ${\bf A} = (-B y,0,0)$ is antisymmetric in the coordinate perpendicular to the wire, $y$. Thus, elastic impurity scattering is needed to mix different momentum states and contribute finite matrix elements of the magnetic vector potential. One can check that Eq. (\[hexact\]) is valid also in the weak disorder limit, by Taylor expanding the correlation function in $ 1/(k_{\rm F} l)$, giving $< y \bullet y>= \frac{ W^2}{10} ( N^3/\lambda^2 ) $, showing that it vanishes for $ \lambda \gg N^2$, corresponding to $\hbar/\tau \ll \pi^2 \hbar^2/(2 m W^2)$, when the disorder does not mix transversal modes, like $1/\lambda^2$, as seen in Fig. 1. In the intermediate regime, $N < \lambda$, it had been argued in Ref. , that $1/\tau_B$ should be reduced by a factor linear in $N/\lambda$ resulting for a 2 dimensional wire with perpendicular magnetic field in a disorder independent expression $$\label{ballphase} \frac{1}{\tau_B} = \frac{1}{C} \frac{W^3 v_F}{l_B^4},$$ where $l_B = ( \hbar/(q B))^{1/2} $ is the magnetic length. For specular boundary condition, as considered in this article, it was found numerically that $C= 9.5$[@beenakker]. Correspondingly, the function $ < y \bullet y>/W^2 $ should approach $ < y \bullet y >/W^2 \rightarrow (\pi/2 C) N/\lambda$ or for $N = 100$, $ < y \bullet y >/W^2 \rightarrow 16.5/\lambda$. The result Eq. ( \[magneticphaseshifting\] ) agrees indeed with this behaviour, in a regime $N \ll \lambda \ll N^2 $, although the best fit gives a different prefactor $14.5$, corresponding to $C = 10.8$. The analytical result shows, furthermore, that this behaviour is only an approximation and that there is a crossover to the perturbative regime, discussed above, where $ < y \bullet y>/W^2 $ decays like $\sim 1/\lambda^2$, see Fig. 1. Note that this result is accurate upto corrections of order $1/N$. Parabolic Wire -------------- As long as the elastic scattering rate exceeds the cyclotron frequency, $ 1/\tau \gg \omega_c$, or correspondingly, $l \ll l_{Cyc} $, where $ l_{Cyc} = k_F l_B^2$ is the cyclotron path, determining the length scale on which ballistic paths start to bend due to the Lorentz force, the magnetic field dependence of the classical diffusion constant and the density of states can be neglected, being for a 2- dimensional wire $ D= \tau v_F^2/2$ and $ \nu( E) = m/( 2 \pi \hbar^2)$, respectively. However, the cyclotron length can be small compared to the width of the wire, $l_{Cyc}< W$, while exceeding the elastic mean free path $ l_{Cyc} > l $, when the crossection of the wire is diffusive, $ l < W $ . Thus, the localization length can depend sensitively on the ratio of these length scales, even in the weak magnetic field limit, where the density of states and classical conductivity are insensitive to the magnetic field. In order to study the crossover as function of the magnetic field, the dependence of the eigen functions on the magnetic field have to be taken into account, therefore. This regime is most conveniently studied for a parabolic wire, having a harmonic confinement, $$H_0 =\frac{1}{2 m} ( {\bf p} - q {\bf A} )^2 +\frac{1}{2} m \omega_0^2 y^2,$$ having the energy eigen values $$E_{n,k} = \hbar \omega_{\rm eff} ( n+ 1/2) + \frac{1}{2 m^*} \hbar^2 k^2,$$ where the effective mass is $ m^* = m \omega_{\rm eff}^2/\omega_0^2$, and the effective frequency is $ \omega_{\rm eff} = ( \omega_{B}^2 + \omega_0^2)^{1/2} $, where $ \omega_{B} = q B/m $ is the cyclotron frequency. The spatial center of the electron eigenstates are shifted by the guding center $ y_k = k \hbar \omega_B/(m \omega_{\rm eff}^2)$. Thus, the width of the wire is at constant Fermi energy $E_{\rm F}$ dependent on the magnetic field $B$. Defining the width of the wire $W$ at fixed Fermi energy as $ W^2 = max ( < n,k \mid y^2 \mid n,k > ) $ with $ E_{n,k} = E_{\rm F}$, one finds for the parabolic wire: $$W^2 ( B) = l_{\rm eff}^2 {\rm max} ( 2 \frac{E_{\rm F}}{\hbar \omega_{\rm eff} } \frac{ \omega_{\rm B }^2}{\omega_{\rm 0}^2} + ( n +1/2 ) ( 1- \frac{ \omega_{\rm B }^2}{\omega_{\rm 0}^2})).$$ For large magnetic field, $ \omega_B \gg \omega_0$, this approaches exactly twice the value at zero magnetic field, and thus, $$W ( \omega_C \gg \omega_0 ) = \sqrt{2} W( 0 ) = (2 E_{\rm F}/ ( \hbar \omega_0 ))^{1/2} l_0.$$ Thus, the wire width is a slowly vaying function of the paramter $ \omega_c/\omega_0 = W( B=0)/l_{cyc}$. The presence of impurities smoothens this function further, and we can thus assume the width to be practically magnetic field independent: $$W = \sqrt{2 E_{\rm F}/ m_e }/\omega_0.$$ This allows us to study the various regimes of interest as a function of the wire width $W$, the magnetic length $l_{\rm B}$ and the average mean free path $ l = ( 2 E/m )^{1/2} \tau $. Naturally, the classical conductivity in such a wire is anisotropic. We find that $$\sigma_{xx} = \frac{ 1 + \omega_0^2 \tau^2 }{ 1 + \omega_{\rm eff}^2 \tau^2 } q^2 \tau n_e/ m,$$ and $$\sigma_{yy} = \frac{ 1 }{ 1 + \omega_{\rm eff}^2 \tau^2 } q^2 \tau n_e/ m,$$ where $ n_e = (2/3 \pi ) ( m_e E/\hbar^2 \omega_0) $ is the average electron density in the wire, which is taken to be approximately independent of the magnetic field. Since we consider magnetic fields where $ \omega_{\rm C} \tau \ll 1$, the classical conductivity is magnetic field independent, $ \sigma_{xx} = q^2 \tau n_e /m$, and $ \sigma_{yy} = \sigma_{xx}/(1 + \omega_0^2 \tau^2) $. Thus, the condition that the localization is governed by the one-dimensional nonlinear sigma model is changed to $ L_{C U}/( 1 + \omega_0^2 \tau^2 ) > W$. With $ \omega_0 \tau = l/ W$ follows that the one dimensional localization condition requires, $ l < 2 N W$, in the weak disorder regime, $ k_{\rm F} l \gg 1$. Rederiving the nonlinear sigm model in the representation of a clean parabolic wire, using the definition of the correlation fucntion, Eq. (\[h\]), where teh sum over transverse momenta is substituted by the sum over the band index, $n$, $ k_y \rightarrow n$, we find the result, $$\label{parabol} < y \bullet y > = W^2 (\frac{2}{5}( \frac{1}{ 1 + \omega_0^2 \tau^2 } + 3 \frac{\omega_c^2}{\omega_0^2} ) = W^2 \frac{2}{5} (\frac{1}{ 1 + l^2/ W^2 } + 3 \frac{W^2}{l_{\rm cyc}^2} ).$$ Note that, since $ \omega_0^2 \tau^2 = l^2/( W^2)$, the ballistic crossection limit $l > W$, coincides for the parabolic wire with the clean wire limit, where transversal modes are not mixed by the disorder $ \hbar/\tau < \hbar \omega_0$. Thus the flux cancellation effect leads in the parabolic wire to a supppression of the phase shifting rate by a factor $ W^2/l^2$ as found for the wire with specular boundaries in the clean wire limit as seen in the previous subsection. Thus, it is not surprising that the behaviour of the magnetic phase shifting rate, as known from weak localization corrections for a wire with ballistic crossection, $ W > l $, and hard wall boundary conditions, is not reproduced when considering a parabolic wire. In the former case, there is a regime, $W^2 < l_B^2 < W l $, implying $ l_B < l $, where the magnetic phase shifting rate is given by $$\frac{1}{\tau_B} = \frac{W^2}{ C_2 \tau l_B^2} < \frac{W^3 v_F }{C l_B^4},$$ where $C_2 = 24/5$. This is smaller than expected from Eq. (\[ballphase\]), and is not obtained for the parabolic wire. Instead, we find that there is a regime, where the magnetic field sensitivity of localization becomes stronger, when the cyclotron length $l_{\rm cyc}$, becomes comparable to the width of the wire $W$. When $ l < l_{\rm cyc} < W$ the magnetic phase shifting rate is found to increase with the magnetic field like $B^4$, $$\frac{1}{\tau_B} = \frac{24}{5} D \frac{q^2}{\hbar^2} B^2 \frac{W^4}{l_{\rm cyc}^2}.$$ When the magnetic field becomes so strong that the cyclotron length $l_{\rm cyc}$, becomes comparable or smaller than the mean free path $l$, or $\omega_c \tau > 1$, the diffusion constant and the density of states become functions of the magnetic field. Then, the spatial modes of the nonlinear sigma model perpendicular to the wire can become soft and contribute to the functional integral, and thus, the nonlinear sigma model becomes effectively two dimensional. In this limit, a quantum Hall wire, the approach used in this article can yield qualitative information on the location and size of localized states in a quantum Hall system [@prl], and will be reconsidered in a forthcoming work. Magnetolocalization in disordered quantum wires ================================================ It is known that the localization length depends on the global symmetry of the wire [@larkin]: $L_c = \beta \pi \hbar \nu S D_0 $, where $\beta =1, 2, 4$, corresponding to no magnetic field, finite magnetic field, and strong spin- orbit scattering or magnetic impurities, respectively. $ \nu (E) $ is the electronic density of states in the wire[@reviews; @ef]. $ D_0 $ is the classical diffusion constant of the electrons in the wire, and $S$ its crossection. This result was obtained by calculating the spatial decay of the density correlation function for wires whose thickness exceeds the mean free path $l$. Here, we use an extension of a recent nonperturbative calculation, to obtain the localization length as a function of the magnetic field, using the fact that the ASD shows a crossover from an oscillating behaviour, decaying with a power law[@haake; @us], typical for Wigner- Dyson energy level statistics[@dyson] to a gaussian decaying function, when the length of the wire is increased beyond the localization length[@prb], as seen in other measures of correlations in the discrete energy level spectrum of a phase coherent disordered electron system[@ef; @mirlin; @guhr; @altfuchs; @isa] . Taking the representation of the ASD derived above, Eq. (\[functionalintegral\]), $$\bar{C}(\omega) = \int \prod d Q ({\bf x}) \exp( - F [ Q ] ),$$ where the action Eq. (\[freeh\] ) can be rewritten conveniently in terms of the diffusion length, an electron would diffuse classically in the magnetic phase shifting time $\tau_B$, $L_B = \sqrt{ D \tau_B }$: $$\begin{aligned} \label{freelb} F[Q] &=& \alpha \frac{1 }{16} L_{C U} \int_0^L \mbox{d} x \mbox{Tr} \left[ ( { { \nabla_x}} Q({ x}))^2 - \frac{1}{4 L_B^2} [ Q, \tau_3]^2 \right] \nonumber \\ &+& i \alpha \frac{\pi}{4} \frac{\omega}{\Delta} \int \frac{\mbox{d} x}{ L} \mbox{Tr} \Lambda_3 Q({ x}).\end{aligned}$$ where $L_{C U} = L_C (\beta = 2) = 2 \pi \hbar \nu S D_0$ is the localization length in the wire in a moderately strong magnetic field [@larkin]. In the limit when $ L_B < L_C $, a moderately strong magnetic field, $Q$ is reduced to a $2 \times 2$- matrix by the broken time reversal symmetry. This reduces the space of Q to $ U(2)/(U(1) \times U(1) )$. For $\omega/\Delta < L_{C U}/L$, corresponding to $ \omega < E_{\rm C}$, where $ E_{\rm C} = 2 \pi D/L^2$ is the Thouless energy scale of classically free diffusion through the wire of length $L$, the spatial variation of $Q$ can be neglected and one retains the same ASD as for random matrices of orthogonal or unitary symmetry, respectively [@haake; @us]. Increasing the length of the wire $L$, a crossover in the autocorrelation function can be seen as the wire exceeds the length scale $L_c$[@prb]. In order to study quantum localization along the wire, the function $C(\omega)$ should be thus considered as a function of the finite length L of the wire and spatial variations of $Q$ along the wire have to be considered, as described by the one dimensional nonlinear sigma model derived above. The impurity averaged ASD can to this end be written as a partition function [@prl] $$\label{part} \bar{C} ( \omega ) = \mbox{Tr} \exp ( - L \bar{H} \left[ Q \right]),$$ where $\bar{ H}$ is an effective Hamiltonian of matrices Q on a compact manifold, determined by the symmetries of the Hamiltonian $H$ of disordered electrons. Thus, the problem reduces to the one of finding the spectrum of the effective Hamiltonian $\bar{H}$. We can derive the corresponding Hamiltonian $\bar{H}$ by means of the transfer matrix method, reducing the one-dimensional integral over matrix field $Q$, Eq. (\[functionalintegral\]), to a single functional integral. Thus, the ASD is obtained in the simple form of Eq. (\[part\]), with the effective Hamiltonian $$\label{effha} \bar{H}( \omega = 0) = \frac{1}{\alpha L_{CU} } (- 4 \Delta^R_Q - \frac{1}{16} X^2 Tr_Q [Q,\tau_3]^2).$$ $\Delta^R_Q$ is that part of the Laplacian on the symmetric space, which does not commute with $Tr[\Lambda_3 Q ]$. The time reversal symmetry breaking due to the external magnetic field is governed by the parameter $X = \alpha L_{CU}/( 2 L_B)$. The problem is now equivalent to a particle with “mass” $ (\alpha/8) L_{CU} (E) $ moving on the symmetric space of $Q$ in a harmonic potential with “frequency” $ 1/( 2 L_{B})$, and in an external field $i \alpha (\pi/4) \omega/(L \Delta)$, in “time” $x$, the coordinate along the wire. To find the ASD as a function of $\omega$ and the length of the wire $L$, one can do a Fourier analysis in terms of the spectrum and eigenfunctions of the effective Hamiltonian at zero frequency, $\bar{H} ( \omega =0 )$ [@zirnbauer2]. There is a finite gap $E_G$ between the ground state energy and the energy of the next excited state of $\bar{H} (\omega = 0) $. For a long wire, $ L E_G \gg 1$, the ASD becomes, $C(\omega) = \exp ( - const. L \omega^2/E_G)$, where both $ const. \omega^2 = \mid <~0~\mid \bar{H} ( \omega ) - \bar{H} (0) \mid~1~> \mid^2$, and the gap between the ground state and the first excited state, $E_G = E_1 -E_0$ do depend on the symmetry of the Hamiltonian $\bar{H}$. This exponential decay with $ L\omega^2$ is typical for a a spectrum of localized states[@prl]. In the other limit $L E_G \ll 1$, all modes of $\bar{H}$ do contribute to the trace in the partition function Eq. (\[part\]) with equal weight, yielding the correlation function of a spectrum of extended states[@prb]. Thus, the crossover length is entirely determined by the gap $E_G$, through $\xi_c = 1/E_G$, and can be identified with an averaged localization length. In order to derive the eigenvalues of the effective Hamiltonian at zero frequency, $\bar{H} ( \omega =0 )$, we need to introduce a representation of the matrix $Q$ and evaluate the Laplacian in its parameters. This is done in Appendix C. Without magnetic field, $B= 0$, the Laplacian is obtained to be $$\begin{aligned} \label{orth} \Delta^R_Q & = & \partial_{\lambda_C} ( 1 - \lambda_C^2 ) \partial_{\lambda_C} + 2 \frac{1 - \lambda_C^2}{\lambda_C} \partial_{\lambda_C} \nonumber \\ & + & \frac{1}{\lambda_C^2} \partial_{\lambda_D} ( 1 - \lambda_D^2 ) \partial_{\lambda_D},\end{aligned}$$ where $\lambda_{C,D} \in [-1,1]$. Its ground state is $1$ and the first excited state is $ \lambda_C \lambda_D$. Thus, the gap is $$E_G ( B = 0 ) = 16/L_{©C U}.$$ For moderate magnetic field, with the condition $ L_{C U} ( < y \bullet y> )^{1/2} B \gg \phi_0 = h/q$, all degrees of freedom arising from time reversal invariance are frozen out, due to the term $Tr_Q [ Q, \tau_3]^2 = 16 (\lambda_C^2-1)$ which fixes $\lambda_C^2 = 1$. Then, the Laplacian reduces to $$\begin{aligned} \label{unit} \Delta^R_Q = \partial_{\lambda_D} ( 1 - \lambda_D^2 ) \partial_{\lambda_D}.\end{aligned}$$ Its eigenfunctions are the Legendre polynomials. There is a gap above the isotropic ground state of magnitude $$E_G ( X \gg 1 ) =8/L_{C U}.$$ For moderate magnetic impurity scattering, exceeding the local level spacing, $1/\tau_s > \Delta_C$, $\alpha=2$, and the Laplacian is given by Eq.(\[unit\]). Thus, due to $\alpha=2$, the gap is reduced to $ E_G (1/\tau_S > \Delta_C) = 4/L_{C U}$. For moderately strong spin- orbit scattering $1 / \tau_{SO} > \Delta_C$, the Laplace operator is $$\label{spo} \Delta^R_Q =\sum_{l=1,2} \partial_{\lambda_l} ( 1 - \lambda_l^2 ) \partial_{\lambda_l},$$ where $\lambda_{1,2} \in [-1,1]$. The ground state is $\psi_0 = 1$, the first excited state is doubly degenerate, $\psi_{11} = \lambda_1$, $\psi_{12} = \lambda_2$. Thus, the gap is the same as for magnetic impurities, $$E_G ( 1/\tau_{SO} > \Delta_C ) = 4/L_{C U}.$$ An external magnetic field lifts this degeneracy but does not change the gap. \[table\] Class Symmetric Space Cartan class Gap $E_G$ ---------- -------- ----------------- ---------------------------------------- ----------- -------------- Ordinary T R S R Sp(2)$/$( Sp(1) $\times$ Sp(1)) CII $16/L_{C U}$ Ordinary No T R S R $ U(2)/(U(1) \times U(1) )$ ( Sphere ) AIII $8/L_{C U}$ Ordinary T R No S R $ O(4)/(O(2) \times O(2) )$ BDI $4/L_{C U}$ Ordinary No T R No S R $ U(2)/(U(1) \times U(1) )$ AIII $4/L_{C U}$ : Relation between symmetry of the Hamiltonian and the gap of the quasi-1D- NLSM Thus, using the crossover in energy level statistics as the definition of a localization length as above, we get in a quasi- 1 -dim. wire, $$\xi_c = 1/E_G(\beta)= ( 1/16 ) \beta L_{C U} ,$$ where $\beta =1, 2, 4$ corresponding to no magnetic field, finite magnetic field, and strong spin- orbit scattering or magnetic impurities, respectively. Comparing with the known equation for the localization length, $ L_c $, we find that the dependence of the ratios $\beta$ on the symmetry are in perfect agreement with the result as obtained from the spatial decay of the density- density- correlation function[@larkin], while it defers by the overall constant $ 1/ 8$. \[fig2\] This relation can be proven directly. The ASD at zero frequency $\bar{C} ( 0 )_{L}$ of the wire of length $L$, becomes, when the wire is divided into two parts, $\bar{C} ( 0 )_{L/2}^2$. For $L \rightarrow \infty$, we find that the relative difference is: $$\label{specdec} f(L) = \frac{\bar{C} ( 0 )_{L/2}^2}{ \bar{C} ( 0 )_{L}}- 1 = 2 p \exp ( - L E_G/2 ),$$ exponentially decaying with the length $L$. Here $p$ is the degeneracy of the first excited state of $\bar{H} ( \omega = 0)$. $f(L)$ can be estimated, following an argument by Mott[@mottdavis]: When the two halves of the wire get connected, see Fig. (2), the Eigenstates of the two separate halves become hybridized and the Eigenenergy of a state $ \psi_n $ is changed by $ \pm \Delta_C \exp ( - 2 x_n/L_C ) $. $x_n$ is random, depending on the position of an eigenstate with closest energy in the other half of the wire. Thus, averaging over $x_n$ gives: $$f( L ) \sim + \exp ( - 4 L/ L_C ).$$ Comparison with Eq. (\[specdec\]) yields indeed $ 1/L_C = 8 E_G$. It is thus a remarkable fact that this length scale, defined as the crossover length of the spectral autocorrelation function, and related to the excitation gap of the compact nonlinear sigma model, has exactly the same symmetry dependence as the localization length, defined through the exponential decay of the spatial density correlation function, found in Ref. . This is especially surprising, since the nonperturbative derivation of the disorder average of the quantity, $ < \rho ( {\bf r}, t ) \rho ( {\bf r'}, t' ) > - <\rho ( {\bf r} )^2 >$, necessitates the use of the supersymmetry method, resulting in a nonlinear sigma model of supermatrices, having in addition to a compact sector, the one considered here, a non compact sector, where the matrix is parametrized on a semi infinite interval. The full supersymmetry allows furthermore rotations between this compact and noncompact sector which are parametrized by Grassmann numbers $\xi$, having the property $\xi^2= 0$. Apart from this increase of the manifold of the matrix fields $Q$ to the supersymmetric space, the structure of the theory is equivalent. Especially, the free energy of the supersymmetric nonlinear sigma model, has exactly the same form as Eq. (\[freelb\]), replacing $ Q $ by supermatrices and the Trace over Q, by a supertrace $STr$, giving the opposite sign to the noncompact sector[@ef]. Studying localization in a wire with this supersymmetric nonlinear sigma model, the transfer matrix method yields an effective Hamiltonian of supermatrices $Q$, of the same form as Eq. (\[effha\]), where the Laplacian is now defined on the respective supersymmetric manifold. In full analogy, the spectrum of $\bar{H}$ determines accordingly the properties of a disordered quantum wire, and has been derived in Ref. for the pure ensembles. The partition function $ Z = STr \exp ( - L \bar{H} ) $ is a generating function of spectral correlation functions[@andreev; @guhr]. In order to derive spatial correlation functions like the density correlation function, in addition, the Eigen functions of the respective diffusion equation on the supersymmetric manifold, $$( - \partial_x + \bar{H} (Q) ) \psi(x; Q) = 0,$$ have to be found[@larkin]. In that way, a formula for the conductance of a finite disordered wire attached to two leads at a distance $L$, has been derived[@zirnbauer2], see also Ref . In the limit of a wire which is perfectly coupled to the leads, that formula for the average conductance simplifies to $$< g > = \frac{1}{2 \alpha} \int d \mu ( {l_i} ) E ( {l_i} ) \exp ( - \frac{L}{16 } E ( {l_i} ).$$ Where $ E ( {l_i} ) $ are the eigenvalues of the supersymmetric Hamitlonian $\bar{H}( \omega = 0 ) $ and $ d \mu ( {l_i} )$ the corresponding integration measure, of the discrete and continous eigenvalues of the angular momentum operator on the compact and noncompact sector, respectively. They were found to be given for $B=0$ by[@zirnbauer2] $$\label{spectrum1} E ( {l_i} ) = 0, 4/L_{C U} 2 ( \epsilon^2 +1 ), 4/L_{C U} ( l^2 + \epsilon_1^2 + \epsilon_2^2 + 1),$$ where $ l = 3,5,...$, and $ \epsilon > 0, \epsilon_1 > 0, \epsilon_2 > 0$. For time reversal symmetry broken wires $ X > 1$ the eigenvalues were found to be, $$\label{spectrum2} E ( {l_i} ) = 0, \frac{4}{\alpha L_{C U})} ( l^2 + \epsilon^2 ),$$ where $ l = 1,3,5,...$, and $ \epsilon > 0$. If spin symmetry is broken, but time reversal symmetry conserved, in the presence of spin orbit scattering, the eigenvalues were found to be, $$\label{spectrum3} E ( {l_i} ) = 0, \frac{4}{ 2 L_{C U}} ( 2 (l-1)^2, \frac{4}{ 2 L_{C U}} ( l_1^2 + l_2^2 +\epsilon^2 - 1),$$ where $ l = 3,5,...$, $ l_i = 1,3,5,.., i=1,2$ and $ \epsilon > 0$. In that case it can be seen that for a distance between the leads much exceeding the localizaion length, $ L \gg L_{C U}$, the conductance decays exponentiallly, and that this is entirely determined by the compact gap $\tilde{E}_{\rm G}$, between the lowest angular momentum eigenstates of the compact sector. The integration over the continous eigenvalues of the noncompact sector, leads only to a prefactor, decaying as a power of the length, $ \sim 1/L^{3/2}$. Indeed, the gap between the ground state value $ E=0$ and the first excited state is seen from Eqs. (\[spectrum1\],\[spectrum2\],\[spectrum3\]), to be $\tilde{E}_G = 8/L_{C U} $ for $B=0$, $\tilde{E}_G = 4/L_{C U} $ for $X > 1$, $\tilde{E}_G = 2/L_{C U} $ for magnetic impurity scattering, $\alpha=2$, and $\tilde{E}_G = 2/L_{C U} $ for moderate spin- orbit scattering, coinciding with the symmetry dependence of the compact gap derived above. However, that coincidence might appear as mere chance, since in fact, the Laplacian of the supersymmetric matrix $Q$ can not be written as a sum of the one of the respective compact nonlinear sigma model, Eqs. ( \[orth\],\[unit\],\[spo\] ), because the metric tensor $ \hat{g} $ on the supersymmetric space contains mixed factors of compact and noncompact parameters. Therefore, the discrete eigenvalues of $-\Delta_Q$, are not the Eigenvalues of the square of the angular momentum on a compact sphere[@zirnbauer2]. Only, in the limit of infinite noncompact parameters does one recover the respective Laplacian on the compact symmetric space, Eqs. ( \[orth\],\[unit\],\[spo\] ). Thus, having shown that the ASD yields the correct symmetry dependence of the localization length, we can now use this approach to get an analytical solution for the crossover behaviour of the localization length and the local level spacing as a magnetic field is turned on, and there is no spin- orbit scattering. While a self consistent approach [@bouchaud], a semiclassical analysis[@lerner] and numerical studies[@leadbeater; @crossover] showed a continous increase of the localization length, an analytical result [@kolesnikov] indicated that both limiting localization lengths $L_c(\beta = 1)$ and $L_c(\beta =2)$ are present in the crossover regime and that there is no single parameter scaling. This is explained by arguing that the far tails of the wavefunctions do cover a large enough area to have fully broken time reversal symmetry, decaying with the length scale $L_c (\beta =2)$ even if the magnetic field is too weak to affect the properties of the bulk of the wavefunction, which does decay at smaller length scales with the shorter localization length $L_c (\beta =1)$, corresponding to the time reversal symmetric case. The quantity studied there is the imprurity averaged correlation function of local wavefunction amplitudes and its momenta at a fixed energy $\epsilon$: $Y(\epsilon) = < \sum_{\alpha} \mid \psi_{\alpha} (0) \mid^2 \mid \psi_{\alpha} ( r) \mid^2 \delta (\epsilon - \epsilon_{\alpha} )>$. It is averaged over a distribution of eigenfunctions in different impurity representations. Thus, each eigenfunction could decay exponentially with a single localization length, but having a distribution which has two maxima, at $L_c(\beta=1)$ and $L_c(\beta=2)$, whose weight is a function of the magnetic field in the crossover regime. While the distribution function of $ \ln (\mid \psi_{\alpha} (0) \mid^2 \mid \psi_{\alpha} ( r) \mid^2 ) $ is known to be Gaussian in both limiting cases of conserved, and fully broken time reversal symmetry, centered around the value $ r/L_{C } ( \beta ) $, $ \beta = 1,2$, respectively, it is not yet known in this crossover regime, however[@mirlin]. The average value of moments, $\mid \psi_{\alpha} (0) \mid^k \mid \psi_{\alpha} ( r) \mid^k$, is decaying more slowly than its typical value, and does not depend on the order of the moment,$k$. This was taken as a proof that moments are determined by states with anomalously large localization lengths of the order of the system size[@mirlin]. Therefore, the result of Ref. can be a property of such rare states with anomalously large localization length, and it remains to see, if the full distribution function scales with two lengths $ L_c ( \beta )$, $\beta = 1,2$, or a single one, changing continously with the magnetic field, $ L_c ( B ) $. While we cannot resolve this question by calculating a spectral autocorrelation function like the ASD, this is another motivation to see if the energy level statistics is governed by a single parameter as the magnetic field is varied. The effective Hamiltonian for moderate magnetic fields is found, without spin dependent scattering, $\alpha=1$, using $ Tr [Q, \tau_3 ]^2 = 16 ( 1-\lambda_C^2) $ to be given by: $$\bar{H} =\frac{1}{L_{C U}} ( - 4 \Delta^R_Q + X^2 ( 1-\lambda_C^2) ) ,$$ where the Laplacian is Eq.(\[orth\]) and $X = L_{C U}/(2 L_B) $. In the limit $ X \rightarrow 0$ the ground state and first excited state approach $ 1, \lambda_C \lambda_D$, respectively. In the limit $ X \gg 1$, $\lambda_C^2$ becomes fixed to 1. Thus, the Ansatz $\psi_0(\lambda_C ) \sim \exp ( A_0 X^2 (1- \lambda_C^2))$, and $\psi_1(\lambda_C,\lambda_D ) \sim \lambda_C \lambda_D \exp ( A_1 X^2 (1-\lambda_C^2))$, where $ A_0 < 0, A_1 < 0$ are negative constants, solves $\bar{H} \psi =\bar{ E} \psi$ to first order in $z= X^2 ( 1-\lambda_C^2)$. One finds that the two lowest magnetic field dependent eigenvalues are $E_0 = 4/L_{C U} ( -5 + \sqrt{ 25 + X^2} )$, and $E_1 = 4/L_{C U} ( -3 + \sqrt{ 49 + X^2} )$, and the Eigenfunctions are given as above with $ A_0 = - L_{C U} E_0/ (16 X^2) $, and $ A_1 = ( 1 - L_{C U} E_1/ 16 )/X^2$, yielding the right limits for $ X \rightarrow 0$ and $ X \gg 1$, respectively. Thus, there is a magnetic field dependent gap $ E_G = E_1 -E_0 $ of magnitude: $$\label{smooth} E_G (X) = 4( 2 + \sqrt{49 + X^2} - \sqrt{25 + X^2})/L_{C U}.$$ This solution is valid in both the limits $X \ll 1$ and $X \gg 1$, interpolating the region $ X \approx 1$. With the magnetic diffusion length $L_B = ( D \tau_B )^{1/2}$, and the magnetic phase shifting rate, as given by Eq. (\[magneticphaseshifting\]), we obtain: $$X = L_{C U}/ (2 L_B) = L_{C U} \frac{q}{ \hbar} \sqrt{ < y \bullet y> } B,$$ which is $ \sqrt{ < y \bullet y> }/ W$ times the number of flux quanta penetrating a localization area $ L_{C U} W $. From Eq. (\[smooth\]) follows that the magnetic change of the localization length is $\delta L_C(B) \sim B^2$ for small and $\sim 1/B$ at large magnetic fields, which agrees with the result of the selfconsistent method as obtained by Bouchaud[@bouchaud]. Resistance of disordered Quantum Wires ====================================== In the limit of zero temperature, $T=0$, the resistivity of a disordered quantum wire, having only localized states at the Fermi energy, is infinite. For finite temperature, $T>0$, in the strong localization regime $k_{\rm B} T < \Delta_{\rm C}$, the mechanism of conduction is hopping of electrons between localized states. Then, the resistivity increases exponentially with temperature. According to the resistor network model[@Miller; @Shkl1], each pair of localized states $i$ and $j$ is linked by a resistance $R_{ij}$: $$\label{resistor} R_{ij}=\exp(\frac{2r_{ij}}{L_{c}}+\frac{\epsilon_{ij}}{k_{B}T})$$ where $r_{ij}=|r_{i}-r_{j}|$ and $\epsilon_{ij}=(|\epsilon_{i}-\mu|+ |\epsilon_{j}-\mu|+|\epsilon_{i}-\epsilon_{j}|)/2k_{B}T$ ($r_{i}$ and $\epsilon_{i}$ are the position and energy of the state $i$, $\mu$ being the Fermi energy). Because of the exponential dependence of $R$ on $r_{ij}$ and $\epsilon_{ij}$, percolation theory methods can be applied [@Ambeg; @Pollak; @Shkl2]. In 2-D and 3-D systems, the dependence of $R$ on temperature $T$ shows a crossover from an activated behaviour to the variable range hopping (VRH) regime. In this regime the temperature is so low that the typical resistances between neighbouring states are large because of the second term in Eq. (\[resistor\]). Therefore electrons tunnel to distant states whose energies are close to the Fermi level. If we neglect electron-electron interactions the resistivity is described by Mott’s law [@Mott; @Ambeg]: $$\label{mott} R(T)=R_{0}\exp[(\gamma T_{0}/T)^{1/(d+1)}]$$ where $d$ is the dimensionality of the system, $\gamma$ a numerical coefficient which depends on $d$, $T_{0}=1/ \nu L_{c}^{d}$ and $\nu_d$ is the dimens dependent density of states. However, in the quasi-1-D case and for sufficiently long wires the variable range hopping result, Eq. (\[mott\]), cannot used due to the presence of exponentially rare segments inside which all the localized states have energies far from the Fermi level [@Lee; @Raikh1; @Kurki]. These large resistance segments (LRS) do not strongly affect the resistivity of 2-D and 3-D systems because they can be circumvented by the current lines. In 1-D this is not possible and the total resistance of a wire is given by the sum of the resistances of all the LRS’s. This sum yields an activated type dependence of $R$ on $T$[@Raikh1] for infinite wires: $$\label{raikh} R=R_{0}\frac{L}{L_{c}}(\frac{T_{0}}{T})^{\frac{1}{2}}\exp(T_{0}/2T),$$ where $k_{\rm B} T_{0}=1/ \nu L_{c} = \Delta_{\rm c} $ coincides with the local level spacing, and $L$ is the length of the wire. Eq. (\[raikh\]) is valid provided that the number of optimal LRS’s (i.e. those LRS’s which give the largest contribution to $R$ [@Raikh1]) within the length of the sample) is large. Bur for a finite wire length this condition fails to be fulfilled at very low temperature $T$, and the resistance of the chain is determined by smaller LRS’s; in this regime Eq. (\[raikh\]) is replaced by [@Lee; @Raikh1]: $$R\approx R_{0} \exp [ \sqrt{ 2\frac{T_{0}}{T} \log(\frac{L}{L_{c}}(\frac{T}{T_{0}})^{\frac{1}{2}}\log^{\frac{1}{2}}(\frac{L}{L_{c}})) } ],$$ which is valid below a temperature $$T_{1} =\frac{T_{0}}{2 \ln(L/L_{c})},$$ approaching Mott’s law, Eq. (\[mott\]) at lower temperatures $T < T_1$. So far, electron-electron interactions have not been taken into account. This approximation is valid if the Coulomb interaction is screened over distances of the order of the hopping length, as by a metal gate electrode deposited on top of the wires at a distance smaller than the typical hopping lengths. When this is not the case, long range electron-electron interactions affect both the density of states and the resistance of the samples [@Raikh2; @Larkin]. Comparison with experimental results ==================================== The magnetic field dependent activation energy was measured recently in transport experiments of Si $\delta$- doped Ga As quantum wires[@khavin]. As an example, we discuss here the sample 5 of Ref. , with a width $ W = .2 \mu m$, a localization length $L_{CO} = .61 \mu m$ a length $ L = 40 \mu m$, and $ N = 30$ channels. The activation energy coincides with the local level spacing $k_{\rm B} T_0 = \Delta_c = 1/ (\nu W L_c) $ and is estimated for sample 5 to be $ T_0 = .34 K$. Thus, according to the theory outlined in the previous section, there is an activated reistance in an order of magnitude temperature range $ T_1 = .04 K < T < T_0 = .34 K$, allowing in good approximation the direct measurement of the magnetic field dependent activation energy $ \Delta_c ( B) $, and thus the magnetic field dependence of the localization length $L_C ( B) $. The ratio of the cyclotron frequency and the elastic scattering rate, $ \omega_C \tau = l/(k_{\rm F} l_B^2) \ll 1$, is small in the whole range of magnetic fields considered there, so that the classical conductance would be magnetic field independent, $\sigma = n e^2 \tau/m (1 +\omega_C^2 \tau^2)^{-1} \approx n e^2 \tau/m$. The mean free path $l \sim .02 \mu$ is small compared to the width of the sample $W = .2 \mu m$. The magnetic length is $l_B = .026 \mu m ( B/T )^{-1/2}$. Thus, while $ \omega_C \tau \ll 1$, the magnetic length becomes smaller than the width of the sample at magnetic fields $ B > .0165 T$. The experimental magnetic field dependence of the ratio of activation energies is shown in Fig. (3) together with the theoretical curve for the ratio of local energy level spacings $\Delta_C(B)/\Delta_C(0) = E_G(B)/E_G(0)$, as derived above, Eq. ( 36 ), using for the magnetic phase shifting rate the results for a 2-dimensional wire with specular boundary conditions, Eq. (14), and, for comparison, the one derived for a parabolic wire, Eq. (\[parabol\]). There is a quantitative discrepancy between the best fit $X = .036 B/G$, and $X = 2 \pi \phi/\phi_0 $, $\phi = \mu_0 H L_{C U} (\overline{y^2})^{(1/2)}$, when using the analytical formula Eq. ( 14). With the experimental parameters $ \alpha=1, L_{CO} = .61 \mu m$, width $W = .2 \mu m $ of sample 5 in Ref. and $ \overline{y^2} = W^2/12$ for a 2- dimensional wire, it yields rather $ X= .010 B/G$. We note that smooth confinement can give $\overline{y^2} > W^2/12$. A similar discrepancy was observed between $W$ as obtained from the sample resistance and estimated from the analysis of the weak localization magnetoresistance, which also depends on $\overline{y^2}$[@khavin2]. We note that the agreement, when using the experimental parameters, for the parabolic wire, is better. The cyclotron length $ l_{\rm cyc} = k_{\rm F} l_{\rm B}^2 = .32/(B/T) \mu m$, is found to be larger than the mean free path $l$ for $B < 15 T$ and larger than the wire width for $ B < 1.5 T$. We find for the parabolic wire: $X = .024 ( .99 + 1.33~10^{-8} ( B/G)^2 )^{1/2} B/G $. The enhancement of the magnetic phase shifting rate in a parabolic wire, Eq. (\[parabol\]), is thus too weak to be seen at the magnetic fields used in the experiment, $B < .2T $, as shown in Fig. (3), and seems thus not to be the origin of the increase in the decay of the activation gap, at about $.1T$ . An extension of the derivation given in section IV to include a dependence of the eigenfunctions on the magnetic field also for a 2-dimensional wire with specular boundary conditions has to be done, in order to make the comparison with the experiment more quantitative, and conclude from the magnetolocalization on the form of the confinement potential in these Si- $\delta$- doped Ga As quantum wires. But, our results may indicate that the harmonic confinement model of the parabolic wire is a better description of the wires in sample 5. \[fig4\] Summary and open problems {#sec:summary} ========================== A formula for the magnetic phase shifting rate has been derived, which allows its calulcation for arbitrary wire geometries and ratios of the elastic mean free path, the wire width, and the magnetic length. For a quantum wire with specular boundary conditions and harmonic confinement this formula has been evaluated explicitly, and compared with previous analytical and numerical results for the magnetic phase shifting rate. The localization length is derived as the crossover length scale from correlated to uncorrelated energy level statistics, as studied with the autocorrelation function of spectral determinants. It is shown that its symmetry dependence coincides exactly with the localization length as defined by the exponential decay of the averaged two-terminal conductance and derived with the supersymmetry method. Therefore, the ASD can be used to get analytical information on the magnetic field dependence of the localization length, which is shown to be governed by the magnetic phase shifting rate, and thus strongly dependent on the geometry of the wire and the ratios of the elastic mean free path, the wire width, and the magnetic length. A comparison with the magnetic field dependence of the activation gap, as observed in low temperature resistance measurements in Si $\delta$- doped Ga As wires, indicates, that the electrons move in a potential which is closer to a harmonic than a hard wall confinement. Enhancement of the sensitivity of the localization to a magnetic field is found analytically, when the cyclotron length is comparable with its width. The physical reason for this enhancement is found to be the magnetic field dependent shift of the guiding centers of the electronic eigenstates in the quantum wire, even at moderate magnetic fields, when the classical conductivity is still independent of the magnetic field. It remains to extend the derivation to include random surface scattering [@leadbeater] and the effect of correlated, smooth disorder[@aleiner], in order to allow for a more quantitative comparison with the experiment. Both effects necessitate a new derivation of the nonlinear sigma model, which allows for a directional dependence of the matrix field $Q$. This has been recently introduced for a system with broken time reversal symmetry in the study of localization in correlated disorder [@taras1], and the spectral statistics of quantum billards with surface scattering[@blanter]. In both cases one is lead to a nonlinear sigma model, where variations of the matrix $Q$ on ballistic length scales are taken into account[@muzy; @aasa; @simons]. The application of this approach to the magnetolocalization in disordered quantum wires will be presented in a future publication. ACKNOWLEDGMENTS {#acknowledgments .unnumbered} =============== The authors gratefully acknowledge, usefull discussions with Isa Zarekeshev, and Konstantin Efetov, thank Yuri Khavin for providing his data and usefull communications, and Bernhard Kramer for stimulating support and critical reading of the manuscript. Appendix A {#appendix-a .unnumbered} ========== Here, the derivation for spinless case, $\alpha=1$, is given in detail. We use for compactness the vectors of anticommuting variables, $$\label{psi} \psi ({\bf x}) = \left( \begin{array}{c} \xi ( {\bf x} ) \\ \xi^* ({\bf x}) \\ \eta ( {\bf x}) \\ \eta^*( {\bf x} ) \end{array} \right), \bar{\psi} ( {\bf x} ) = ( \xi^* ( {\bf x} ), - \xi ( {\bf x} ), \eta^*({\bf x}), - \eta ({\bf x}) ).$$ Note that $ \bar{\psi} = ( C \psi )^T $, where the matrix $C$ interchanges the Grassmann fields with their conjugate one, and has thus the form $ C = \left( \begin{array}{cccc} 0 & -1 & 0 & 0 \\ 1 & 0 & 0 & 0 \\ 0 & 0 & 0 & -1 \\ 0 & 0 & 1 & 0 \end{array} \right)$. Thus, the ASD is written as $$\begin{aligned} \label{grassmannrep} \bar{C}(\omega ) &=& \int \prod_{\bf x} d \psi({\bf x}) \nonumber \\ && \exp ( - \frac{1}{2}\int d {\bf x} \bar{\psi} ({\bf x}) (E + \frac{1}{2}\omega \Lambda - \hat{H_0} - V({\bf x}) )\psi({\bf x}) ), \end{aligned}$$ Here, the diagonal Pauli matrix $ \Lambda = \left( \begin{array}{cc} 1 & 0 \\ 0 & -1 \end{array} \right)$ has been introduced for compactness, its diagonal elements projecting on the respective spectral determinant of the ASD. The kinetic Hamiltonian becomes a matrix $$\hat{H_0} = (\hat{p}- q \tau_3 {\bf A} )^2/2/m,$$ where the diagonal Pauli matrix $\tau_3$ had to be introduced since each vector has elements of the Grassmann field and the time reversed one, and the diamagnetic term ${\bf p A}/m$ in the Hamiltonian changes sign, as ${\bf p} \rightarrow -{\bf p}$, breaking the time reversal invariance. To summarize the notation, here, and in the following, $\Lambda_i$ are the Pauli matrices in the subbasis of the left and the right spectral determinant, $\tau_i$ the ones in the subbasis spanned by time reversal and $\sigma_i$ the ones in the subspace spanned by the spinor, for $i=1,2,3$. Note that a global transformation of the Grassmann vectors $\psi \Rightarrow \tilde{\psi} = A \psi $ does leave the functional integral for $\omega=0$ invariant, as long as $ A^+ A = 1$, and $A^{+ T} C = C A$, restricting the matrices $A$ to be symplectic ones, being elements of $Sp(2)$, commuting with the antisymmetric matrix $C$. A finite frequency breaks this symmetry group, and only symplectic transformations of each field of a single spectral determinant separately, $ Sp(1) \times Sp(1) $, do leave the functional integral invariant. Now, the averaging over the disorder potential can be done, integrating Eq. (\[grassmannrep\]) over the Gaussian distribution function of the random potential V. Thus, the averaged ASD is found to be given by a functional integral over interacting Grassman fields $\psi$, $$\begin{aligned} \bar{C}(\omega ) &=& \int \prod_{\bf x} d \psi({\bf x}) \nonumber \\ && \exp ( - \frac{1}{2}\int d{\bf x} \bar{\psi} ({\bf x}) (E + \frac{1}{2}\omega \Lambda - \hat{p}^2/2/m )\psi({\bf x}) ) \nonumber \\ && \exp( - \frac{1}{16 \pi} \frac{\hbar \Delta}{\tau} S L \int d {\bf x} Tr (\psi({\bf x}) \times \bar{\psi}({\bf x}) )^2).\end{aligned}$$ Now, the resulting $\psi^4$-interaction term can be decoupled by introducing another Gaussian integral over an auxilliary field. Clearly, the field should not be a scalar, otherwise we would simply reintroduce the Gaussian integral over the random potential $V$. Rather, in order to go a step forward, the auxilliary field should capture the full symmetry of the autocorrelation function. Therefore, the Gaussian integral is chosen to be over a 4 by 4 matrix $Q_{4 \times 4}$, which is itself an element of the respective symmetric space, as the matrix $A$ which leaves the functional integral invariant. Thus, allowing for a spatial dependence of $Q$, one can decouple the interaction term: $$\begin{aligned} \label{hst} &&\exp( -\frac{1}{16 \pi} \frac{\hbar \Delta}{ \tau} S L \int d {\bf x} ( \psi({\bf x}) \times \bar{\psi}({\bf x}) )^2) \nonumber \\ &=& \int \prod_x d Q_{4 \times 4} ({\bf x}) \exp( - \pi \frac{ \tau}{\hbar \Delta} \int \frac{d {\bf x}}{S L} Tr Q_{4 \times 4} ({\bf x})^2 \nonumber \\ &+& i \frac{1}{2} \int d {\bf x} Tr Q_{4 \times 4} ({\bf x}) \psi({\bf x}) \times \bar{\psi}({\bf x})). \end{aligned}$$ Anticipating, however, that the functional integral over the matrices $Q$ cannot be performed exactly, but rather only an integral over slowly varying modes around a saddle point solution, it is necessary to separate fast and slowly varying modes already before the decoupling of the interaction term Eq. (\[hst\])[@elk]. It turns out that there are two equivalent slowly varying interaction terms, corresponding to diffusion, and one arrives finally after a Gaussian decoupling to a, by a factor $1/2$, shallower nonlinear coupling $ Tr Q^2$[@ef]. Next, one can perform the Gaussian integral over the Grassmann vectors $\psi({\bf x})$ and one obtains for the ASD, rescaling $ Q_{4 \times 4} \rightarrow 2 \tau/\hbar Q_{4 \times 4}$, the representation: $$\label{functionalintegral} \bar{C}(\omega) = \int \prod d Q_{4 \times 4} ({\bf x}) \exp( - F [ Q ] ),$$ with $$\begin{aligned} \label{exactfree} F[Q] = && \frac{\pi}{8} \frac{\hbar}{\Delta \tau} \int \frac{d {\bf x}}{S L} Tr Q_{4 \times 4}({\bf x})^2 ) \nonumber \\ && + \frac{1}{2} \int d {\bf x} <{\bf x} \mid Tr \ln ( G ( \hat{x}, \hat{p} ) \mid {\bf x} >,\end{aligned}$$ where $$G( \hat{x}, \hat{p} ) = 1/( \frac{1}{2} \omega \Lambda - \frac{(\hat{p}- q \tau_3 A)^2}{2 m} + i \frac{\hbar}{2 \tau} Q_{4 \times 4}(\hat{x}) ).$$ is the propagator matrix. We used the operator notation $\hat{x}$, in order to stress that the terms in the inverse propagator, do not commute with each other. Appendix B {#appendix-b .unnumbered} ========== For a clean wire with hard wall boundaries, the transversal Eigen modes are for $ -W/2 < y < W/2$, $< k_y \mid y > = \cos k_y y$ for $ k_y = \pi s/W$, $s$ being an odd integer, and $< k_y \mid y > = \sin k_y y$ for $ k_y = \pi s /W$, s being an even integer, one obtains: $$\mid < k_y \mid y \mid k_y' > \mid^2 = \frac{1}{W^2} ( \frac{1}{(k_y -k_y')^2} - \frac{1}{(k_y + k_y')^2})^2,$$ when $ k_y = \pi s/W$, and $ k_y' = \pi s' /W$, s being even, and s’ odd, or vice versa. Then, the sum over $k_y'$ in Eq. ( \[h\] ) can be performed by use of the Matsubara trick, for $s$ even, and odd integers, separately. The remaining sum over $k_x,k_y$ can be transformed as $1/(W L) \sum_{k_x,k_y} = \int d \epsilon \nu ( \epsilon ) \int \frac{d \hat{e}_k}{\Omega_k} $, noting that the unit vector $\hat{e}_k$ can point only in discrete directions. Thus, while in 2 dimensions $ \int \frac{d \hat{e}_k}{\Omega_k} = \int_0^{2 \pi} \frac{ d \theta }{ 2 \pi } = 4/(2 \pi) \int_0^1 d y 1/( 1-y^2)^(1/2) $, for finite number of transverse channels $N = k_F W/\pi$ there is a sum, $\int \frac{d \hat{e}_k}{\Omega_k} = 2/(\pi N ) \sum_{s > 0} 1/(1 - s^2/N^2)^(1/2)$. Thus, $k_y = \pi s/W= k_F s/N$ and $k_x = k_F ( 1 - s^2/N^2)^(1/2)$. Performing finally for $ E \gg \hbar/\tau$ the integral over $\epsilon$, one arrives with some patience at Eq. (\[hexact\]), where $K_0 = 2/(\pi N) \sum_{s=0}^{N} \sqrt{1- \frac{s^2}{N^2} }$, $K = 2/(\pi N) \sum_{s=1}^{N} \sqrt{1- \frac{s^2}{N^2} }$, $K_1 = 2/(\pi N) \sum_{s=1}^{N} \sqrt{1- \frac{s^2}{N^2} }/s^2$, $K_2 = 2/(\pi N) \sum_{s=1}^{N} \sqrt{1- \frac{s^2}{N^2} } s^2/N^2$. Appendix C {#appendix-c .unnumbered} ========== In order to derive the Laplacian in the respective representation of the matrix field $Q$, its general definition in an arbitrary parametrization, $$\label{laplacebeltrami} \Delta_Q = \frac{1}{\sqrt{ \bar{g} }} \sum_{i,k} \partial_k g^{ik} \sqrt{ \bar{g} } \partial_i,$$ where the matrix $g$ is the metric tensor, being defined by the quadratic form $ ds^2 = 1/4 Tr dQ^2$ of the representation $$d s^2 = d {\bf x}^T g d {\bf x},$$ where $ {\bf x}$ is the vector of parameters of the representation. For [*$B \neq 0 $*]{}, $Q$ is element of $U(2)/ ( U(1) \times U(1) )$, by enforcing the conditions $ Q^2 = 1$, $Q^T C = C Q$, and $Q^+ = Q$, $ [ Q,\tau_3 ] = 0$. It can be paramterized as $ Q = \left( \begin{array}{cc} \cos \theta & e^{i \chi} \sin \theta \\ e^{-i \chi} \sin \theta & - \cos \theta \end{array} \right)$ where $ \theta \in [ 0, \pi ] $ and $\chi \in [0, 2 \pi] $. Thus, $$d s^2 = d \theta^2 + \sin^2 \theta d \chi^2.$$ and $ g = \left( \begin{array}{cc} 1 & 0 \\ 0 & \sin^2 \theta \end{array} \right)$. Thus, with Eq. (\[laplacebeltrami\]) follows: $$\Delta_Q = \partial_{\lambda_D} ( 1 - \lambda_D^2 ) \partial_{\lambda_D} + \frac{1}{1- \lambda_D^2} d \chi^2,$$ where $ \lambda_D = \cos ( \theta ) $. Note that the autocorrelation function depends on the energy difference $\omega$ through the coupling $Tr \Lambda Q = 2 * 2 \lambda_D$, so that only that part of the Laplacian which does not commute with $Tr \Lambda Q $, $$\Delta^R_Q = \partial_{\lambda_D} ( 1 - \lambda_D^2 ) \partial_{\lambda_D}.$$ enters in the frequency dependence of the autocorrelation function of spectral determinants, Eq. ( \[part\] ). Since $U(2)/(U(1) \times U(1) ) = S_2$, the two sphere, this is equivalent to the treatment of spherically symmetric potentials, and the Laplacian can be identified with the square of the angular momentum, $ - \Delta_Q = {\bf L}^2$, and $ L_z = i \partial_{\chi}/( 1 - \lambda_D^2 )$ does commute with the Hamiltonian, $$\bar{H} = - 1/(2 m) L^2 + i \alpha \frac{\pi}{4} \frac{\omega}{\Delta} z,$$ since $ z = \cos \theta_D $ does commute with $L_z$. Therefore, $ \omega \neq 0$ does not break the azimuthal symmetry of rotations around the z-axis, $n_z$. For [*$ B= 0$*]{}, $Q$ is element of the symplectic symmetric space, $Sp(2)/ ( Sp(1) \times Sp(1) )$, by enforcing the conditions $ Q^2 = 1$, $Q^T C = C Q$, and $Q^+ = Q$. One obtains: $$Q = \left( \begin{array}{cc} c \openone & A \\ A^+ & - c \openone \end{array} \right).$$ with $A= \left( \begin{array}{cc} a & b \\ b^* & -a^* \end{array} \right)$ where $ \mid a \mid^2 + \mid b \mid^2 + c^2 = 1$. A matrix Q with the above symmetries can be represented as, $$Q= U^{-1} Q_c^0 U,$$ with $$U = V_C U_D,$$ where $$U_D = V_D^{-1} T_D^0 V_D,$$ where $$Q_c^0 = \left( \begin{array}{cc} \cos \theta_C & i \sin \theta_C \tau_2 \\ i \sin \theta_C \tau_2 & - \cos \theta_C \end{array} \right),$$ and $$T_D^0 = \left(\begin{array}{cc} \cos \theta_D/2 & i \sin \theta_D/2 \\ i \sin \theta_D/2 & \cos \theta_D/2 \end{array} \right).$$ and $$V_{C,D} = \left(\begin{array}{cc} \exp ( i \phi_{C,D} \tau_3 ) & 0 \\ 0 & \openone \end{array} \right).$$ and $\tau_i, i=1,2,3$ are the Pauli matrices. Such a representation was first given by Altland, Iida and Efetov[@altland] to study the crossover between the spectral statistics of Gaussian distributed random matrices as the time reversal symmetry is broken, within the supersymmetric nonlinear sigma model. Here, in order to study the ASD, we need to consider only the compact block of the representation given there. We find that $ ds^2 = Tr dQ^2/4 = d \theta_C^2 + \cos^2 \theta_C d \theta_D^2 + \sin^2 \theta_C \phi_C^2 +\cos^2 \theta_C \sin^2 \theta_D d \phi_D^2$ and thereby with Eq. (\[laplacebeltrami\] ), the part of the Laplace operator which does not commute with $ Tr \lambda Q = 4 \lambda_C \lambda_D $ is given by Eq. ( \[orth\]), $$\begin{aligned} \Delta^R_Q & = & \partial_{\lambda_C} ( 1 - \lambda_C^2 ) \partial_{\lambda_C} + 2 \frac{1 - \lambda_C^2}{\lambda_C} \partial_{\lambda_C} \nonumber \\ & + & \frac{1}{\lambda_C^2} \partial_{\lambda_D} ( 1 - \lambda_D^2 ) \partial_{\lambda_D},\end{aligned}$$ where $\lambda_i = \cos \theta_i, i= C,D$. For moderately strong spin- orbit scattering $1 / \tau_{SO} > \Delta_C$, in the functional integral representation of the spectral determinants by Grassman vectors the spin degree of freedom is introduced, $\alpha=2$ and the matrix $C$ is, due to the time reversal of the spinor, substituted by $i \sigma_2 \tau_1$[@elk]. The spin-orbit scattering reduces the Q matrix to unity in spin space. Thus, the matrix C has effectively the form $\tau_1$. The condition $Q^T C = C Q$ leads therefore to a new symmetry class, when the spin symmetry is broken but the time reversal symmetry remains intact. Then, $Q$ are $4 \times 4$- matrices on the orthogonal symmetric space $ O(4)/(O(2) \times O(2) )$ [@weg]. A matrix $Q$ with the above symmetries can be represented as, $$Q= V^{-1} Q_0 V,$$ with $$Q_c^0 = \left( \begin{array}{cc} \cos \hat{\theta} & \sin \hat{\theta} \\ \sin \hat{\theta} & - \cos\hat{\theta} \end{array} \right),$$ where $$\hat{\theta} = \left(\begin{array}{cc} \theta_1 & \theta_2 \\ \theta_2 & \theta_1 \end{array} \right),$$ with $\theta_i \in [0,\pi], i=1,2$, and $$V = \left(\begin{array}{cc} V_1 & 0 \\ 0 & V_2 \end{array} \right),$$ where $$V_i = \exp ( i \chi_{i} \tau_3 ),$$ with $\chi_i \in [0, 2 \pi], i=1,2$. Thus, we find $ ds^2 = Tr Q^2/4 = \sum_{i=1,2} d\theta_i^2 + d {\bf \chi}^T \hat{g_{\chi}} {\bf \chi}$, where $$\hat{g_{\chi}} = \left(\begin{array}{cc} \sin^2 \theta_1 + \sin^2 \theta_2 & - \sin^2 \theta_1 + \sin^2 \theta_2 \\ - \sin^2 \theta_1 + \sin^2 \theta_2 & \sin^2 \theta_1 + \sin^2 \theta_2 \end{array} \right),$$ Thus, the part of the Laplace operator which does not commute with $ Tr \lambda Q = 4 \lambda_1 \lambda_2 $ is given by Eq. ( \[spo\]), $$\Delta^R_Q =\sum_{l=1,2} \partial_{\lambda_l} ( 1 - \lambda_l^2 ) \partial_{\lambda_l},$$ where $\lambda_i = \cos \theta_i, i=1,2$. P.A. Lee, T.V. Ramakrishnan, Rev. of Mod. Phys. [**57**]{}, 287 (1985); [B. Kramer]{} and A. MacKinnon, Rep. Prog.Phys. [**56**]{}, 1469 (1993); C. W. J. Beenakker, Rev. Mod. Phys. [**69**]{}, 731 (1997). K. B. Efetov, [*Supersymmetry in Disorder and Chaos*]{} Cambridge University Press, Cambridge (1997). K. B. Efetov, A. I. Larkin, Zh. Eksp Teor. Fiz. [**85**]{}, 764(1983) ( Sov. Phys. JETP [**58**]{}, 444 ). M. E. Gershenson et al., Phys. Rev. Lett. [**79**]{}, 725(1997); Phys. Rev. [**B 58**]{}, 8009 (1998). B. L. Altshuler, A. G. Aronov, Pis’ma Zh. Eksp. Teor. Fiz. [**33**]{}, 515 (1981)\[JETP Lett. [**33**]{},499 (1981)\]. V. K. Dugaev, D.E. Khmel’nitskii, Sov. Phys. JETP [**59**]{}, 1038 (1984). B. L. Altshuler, A.G. Aronov in [*Electron-electron interactions in disordered systems*]{}, North Holland (1985). C. W. J. Beenakker, H. van Houten, Phys. Rev. [**B 37**]{}, 6544 (1988). F. Haake, M. Kus, H.- J. Sommers, H. Schomerus, and K. Zychowski, J. Phys. A [**29**]{}, 3641(1996). S. Kettemann, D. Klakow, U. Smilamsky, J. Phys. A,3643(1997). F. J. Dyson, J. Math. Phys. [**3**]{}, 1199 (1962). K. B. Efetov, Zh. Eksp. Teor. Fiz. [ **83**]{}, 833 (1982)\[Sov. Phys. JETP [**56**]{},467 (1982)\]; J. Phys. [**C 15**]{}, L 909 (1982). L. P. Gor’kov, O. N. Dorokhov, and F. V. Prigara, Zh. Eksp. Teor. Fiz. [ **84**]{}, 1440 (1983)\[Sov. Phys. JETP [**57**]{}, 838 (1983)\]. B. L. Alt’shuler, B. I. Shklobskii, Zh. Eksp. Teor. Fiz. [ **91**]{}, 127 (1986)\[Sov. Phys. JETP [**64**]{},127 (1986)\]. B. L. Altshuler, I. Kh. Zharekeshev, S. A. Kotochigova, and B. I. Shklovskii, Zh. Eksp. Teor. Fiz. [ **94**]{}, 343 (1988)\[Sov. Phys. JETP [**67**]{}, 625 (1988)\]; S. N. Evangelou, and E. N. Economou, Phys. Rev. Lett. [**68**]{}, 361 (1992); E. Hofstetter, and M. Schreiber, Phys. Rev. Lett. [**73**]{}, 3137 (1994); I. Kh. Zharekeshev, and B. Kramer, Phys. Rev. Lett. [**79**]{}, 717(1997). A. Altland, D. Fuchs, Phys. Rev. Lett. [**74**]{}, 4269(1995). A. D. Mirlin, Phys. Rep. 326, 259 (2000), Procy. of Intern. School of Physics “ Enrico Fermi”, Course CXLIII, IOS Press, Amsterdam ( 2000 ). T. Guhr, A. Müller-Groeling, and H.A. Weidenmüller, Phys. Rep. 299, 189 (1998). S. Kettemann, Phys. Rev. [**B 59** ]{}, 4799 (1999). E. Abrahams, P. W. Anderson, D. C. Licciardello, V. Ramakrishnan, Phys. Rev. Lett [ **42**]{} 673(1979). L. P. Gor’kov, A. I. Larkin, D. E. Khmel’nitskii, Pis’ma Zh. Eksp. Teor. Fiz. [**30** ]{}, 248 (1979) ( JETP Lett. [**30**]{}, 228 (1979)). B. L. Altshuler, D. E. Khmelmitskii, A. I. Larkin, and P. A. Lee, Phys. Rev. [**B 22**]{}, 5142 (1980). K. B. Efetov, A. I. Larkin, D. E. Khmel’nitskii, Zh. Eksp. Teor. Fiz. [**79**]{}, 1120(1980) ( Sov. Phys. JETP [**52**]{}, 568(1980) ). S. Chakravarty, A. Schmid, Physics Reports [**140**]{}, 193(1986). B. L. Altshuler, A. G. Aronov, D.E. Khmelnitskii, J. Phys. C [**15**]{}, 7367 (1982). M. E. Gershenson, Ann. Phys. [**8**]{},559 (1999). M. Buettiker, cond-mat/0105519 (2001). S. Hikami, A. I. Larkin and Y. Nagaoka, Prog. Theor. Phys. [**63**]{}, 707 (1980). F. Wegner, Z. Physik [ **B 36**]{},l209(1979); Nucl. Phys. B [**316**]{}, 663(1989); S. Hikami, Prog. Theor. Phys. [**64**]{}, 1466 (1980). S. Kettemann and A. Tsvelik, Phys. Rev. Lett. [**82**]{}, 3689 (1999). S. Kettemann, Phys. Rev. Rapid Commun. [**B 62** ]{}, R13282 (2000). S. Helgason, [*Differential Geometry and Symmetric Spaces*]{} Academic Press, New York (1962). M. R. Zirnbauer Jy. Math. Phys. [**37**]{},4986 (1996). M. Titov, P. W. Brouwer, A. Furusaki, C. Mudry, cond-mat/0011146. D. Taras- Semchuk, K. B. Efetov, Phys. Rev. Lett. [**85**]{}, 1060 (2000); cond-mat/0010282 (2001). Ya. M. Blanter, A. D. Mirlin, B. A. Muzykantskii, cond-mat/0011498 (2000). M. R. Zirnbauer Phys. Rev. Lett. [**69**]{},1584 (1992), A. D. Mirlin, A. Mullergroeling, M. R. Zirnbauer, Ann. Phys. ( New York) [**236**]{}, 325 (1994); P. W. Brouwer, K. Frahm, Phys. Rev. [**B 53**]{} ,1490 (1996); B. Rejaei, Phys. Rev. [**B 53**]{}, R13235 (1996), B. Rejaei, Phys. Rev. [**B 53**]{}, R13235 (1996). N. F. Mott, E. A. Davis, [*Electronic Processes in Non-crystalline Materials*]{}, Clarendon Press, Oxford (1971). A. V. Andreev and B. D. Simons Phys. Rev. Lett. 75, 2304-2307 (1995). J. P. Bouchaud, J. Phys. 1 (France) [**1**]{}, 985 (1991). I. V. Lerner, Y. Imry, Europhys. Lett. [**29**]{}, 49(1995). M. Leadbeater, V. I. Falko, C. J. Lambert,Phys. Rev. Lett. 81, 1274 (1998). J.-L. Pichard, M. Sanquer, K. Slevin, and P. Debray, Phys. Rev. Lett. 65, 1812 (1990); H. Schomerus and C.W. Beenakker, Phys. Rev. Lett. 84, 3927 (2000); M. Weiss, T. Kottos, T. Geisel, Phys. Rev. [**B 63**]{}, R081306(2001). A.V. Kolesnikov and K.B. Efetov, Phys. Rev. Lett. 83, 3689 (1999); A.V. Kolesnikov and K.B. Efetov, cond-mat/0005048 (to be published). A. Miller and E. Abrahams, Phys. Rev. [**120** ]{}, 745 (1960). B. I. Shklovskii and A. L. Efros, Electronic Properties of Doped Semiconductors, Springer-Verlag, New York (1984). V. Ambegaokar, B. I. Halperin and J. S. Langer, Phys. Rev. [**B4** ]{}, 2612 (1971). M. Pollak, J. Non-Crystal. Solids [**11** ]{}, 1 (1972). B. I. Shklovskii and A. L. Efros, Zh. Eksp. Teor. Fiz. [**60** ]{}, 867 (1971) \[Sov. Phys. JETP [**33** ]{}, 468 (1971)\]. N. F. Mott, J. Non-Crystal. Solids [**1** ]{}, 1 (1968). P. A. Lee, Phys. Rev. Lett.[**53** ]{}, 2042 (1984); R. A. Serota, R. K. Kalia and P. A. Lee, Phys. Rev. [**B 33** ]{}, 8441 (1986). M. E. Raikh and I. M. Ruzin, Zh. Eksp. Teor. Fiz. [**95** ]{}, 1113 (1989) \[Sov. Phys. JETP [**68** ]{}, 642 (1989)\]. J. Kurkijarvi, Phys. Rev. [**B8** ]{}, 922 (1973). M. E. Raikh and A. L. Efros, Pis’ma Zh. Eksp. Teor. Fiz. [**45** ]{}, 225 (1987) \[JETP Lett. [**45** ]{}, 280 (1987)\]. A. I. Larkin and D. E. Khmel’nitskii, Zh. Eksp. Teor. Fiz. [**83** ]{}, 140 (1982) \[Sov. Phys. JETP [**56** ]{}, 647 (1982)\]. Yu. B. Khavin, M. E. Gershenson, A. L. Bogdanov, Phys. Rev. Lett. [**81**]{}, 1066 (1998). A. Altland, S. Iida, K. B. Efetov, [*J. Phys. A*]{} [**26**]{} (1993) 3545. I. L. Aleiner, A. I. Larkin, Phys. Rev. [**B 54**]{}, 14423 (1996). B. A. Muzykantskii, D. E. Khmelnitskii, JETP Letters [**62**]{},76 (1995)\]. A. V. Andreev, O. Agam, B. D. Simons, and B. L. Altshuler, Phys. Rev. Lett. [**76**]{}, 3947 (1996). Nucl. Phys. [**B 482**]{}, 536 (1996). B. D. Simons, O. Agam, A. V. Andreev, J. Math. Phys. [**38**]{}, 1982 (1997). A. Altland, S. Iida, K. B. Efetov, [*J. Phys. A*]{} [**26**]{} (1993) 3545.

ArXiv

--- abstract: 'The general misconception regarding velocity measurements of a test particle as it approaches black hole is addressed by introducing generalized observer set. For a general static spherically symmetric metric applicable to both Einstein and alternative gravities as well as for some well known solutions in alternative gravity theories, we find that velocity of the test particle do not approach that of light at event horizon by considering ingoing observers and test particles.' author: - Sumanta Chakraborty title: Velocity Measurements in Some Classes of Alternative Gravity Theories --- Introduction {#vi} ============ The radial motion of a test particle falling in a black hole is one of the key issues in general relativity. The infalling motion has been studied specifically for Schwarzschild black hole by several authors ([@lan71],[@Wald],[@Bergmann],[@Moller]). All of them reached the same conclusion that velocity of the infalling particle approaches that of light near the event horizon, which for the Schwarzschild case is at $r=2M$, where $M$ is the mass of the black hole. The observers, called static observers, are at rest with respect to the mass creating the gravitational field. They are actually the world lines on the hypersurface of orthogonal killing vector field for the metric describing the gravitational field. However there exists a common misconception that particle approaches the speed of light as it moves to the black hole horizon for all observers, but not as a limiting procedure for a static observer at $r$ as $r \rightarrow 2M$. However if we assume that the particle approaches the event horizon at the speed of light for a static observer, as we have defined it earlier, then simple velocity composition law tells that it should approach the speed of light for all local observers as space time is locally Minkowskian. So we have to modify our notion of velocity for a test particle near a black hole for a static observer which was done for Schwarzschild black hole ([@Crawford],[@jan77]). The notion of observer is implemented and used in various co-ordinate frames by several authors ([@bol11],[@ell85],[@bol06]). However recently a progress has been made in obtaining trajectory around a general spherically symmetric non-rotating black hole by choosing a general metric ansatz [@cha11], $$\label{i1} ds^{2}=-f(r)dt^{2}+\frac{dr^{2}}{f(r)}+r^{2}d\Omega ^{2}$$ For this general case we find the velocity of the test particle with respect to a static observer ($r=$constant) to be a function of $f(r)$. While for the case of a general observer such that both the observer and the test particle moves along geodesic in $\theta =\frac{\pi}{2}$ plane then the velocity of the test particle with respect to the observer to our surprise, do not depend on the choice of the function $f(r)$ provided the particle has high energy which is the most common case for astrophysical bodies, however it depends on the angular momenta which was absent in earlier works [@Crawford]. Then we have used some classes of spherically symmetric solutions in alternative gravity theories to find the relative velocity of a test particle with respect to an observer. We have discussed spherically symmetric solution in string inspired dilaton model [@Garfinkle], and calculate motion of a test particle in this spacetime. Secondly we have considered a spherically symmetric solution in quadratic gravity obtained in a recent paper [@Yunes] to discuss the velocity profile of an object. Finally we have discussed motion in spherically symmetric solutions in Einstein-Maxwell-Gauss-Bonnet(EMGB) theory and vacuum solution in $F(R)$ gravity. Throughout the paper we shall use natural unit such that $G=c=1$. This paper is organized as follows, in section ($\ref{vc}$) we introduce the general idea of observer and co-ordinate frames which we shall use throughout this work. In section ($\ref{vs}$) we discuss the motion in spherical symmetric space-time for the general choice of metric as presented in equation ($\ref{i1}$). In the next section we discuss some classes of alternative gravity theories. The paper ends with a short discussion on the results obtained. Co-ordinate System, Reference Frames and Observers {#vc} ================================================== The mathematical beauty of general relativity is the freedom of choice of coordinates in the description of physical phenomenon. We could choose any co-ordinate system as we wish, this choice might be taken in favor of the symmetry involved in the problem. Also the co-ordinates are not sufficient we need reference frame as well. However the co-ordinate system and reference frames are not independent, for example in one reference frame one set of co-ordinates may be important while it could change in other reference frame. However in literature [@Bergmann] it is often seen that co-ordinate system and reference frames are used interchangeably. However in our discussion we find the use of “reference frame” and “co-ordinate system” to be distinct. By reference frame we shall mean a set of observers to take measurements, for example the set of all observers moving in a time like geodesic form a reference frame, whereas co-ordinate system refer to numbers specified over the whole space time manifold. In special relativity an infinite lattice work of sticks and clocks [@Wheeler] suffice to define a unique reference frame. However in general relativity we cannot have such rigid framework since the space time is Minkowskian only locally, so we replace this rigid system by a fluid [@Moller]. In a strictly mathematical sense the set of observers represents a set of future pointing time like congruence, which is a three parameter family of curves $x^{\mu}(\lambda ,y^{i})$, where $\lambda$ is an affine parameter defined over the path, and $y^{i}$ labels the spatial parts of the curve. Observer in general theory is very local and it is a material particle parameterized by proper time. An observer field i.e. its velocity field $u$ on the manifold $M$ is stationary provided there exist a smooth function $f$ greater than $0$, such that $fu=\xi$ is a killing vector field, so the lie derivative of the metric with respect to the vector field $\xi$ vanishes (i.e. $L_{\xi}g_{\mu \nu}=0$). There is a natural way for an $u$-observer to define the speed of any particle with four velocity $t^{\mu}$ as it passes an event $p\in M$, then the observer measure the square of the speed at event $p$ to yield [@Crawford], $$\label{i4} v^{2}=\frac{(g_{\mu \nu}+u_{\mu}u_{\nu})t^{\mu}t^{\nu}}{(u_{\alpha}t^{\alpha})^{2}}$$ Then we have $g_{\mu \nu}t^{\mu}t^{\nu}=-1$ and as well as $u_{\mu}u_{\nu}t^{\mu}t^{\nu}=(u_{\alpha}t^{\alpha})^{2}$. Thus the above relation can be simplified to yield, $$\label{i3} v^{2}=1-\frac{1}{(u^{\mu}t_{\mu})^{2}}$$ Note that the two velocities $u^{\mu}$ and $t^{\mu}$ are time like as observer and the test particle are both time like. Motion in a General Spherically Symmetric Space Time {#vs} ==================================================== Test Particle Geodesic ---------------------- We shall assume that our test particle is confined to a plane which is generally chosen as $\theta=\pi/2$ for calculational simplicity and as well as we have spherical symmetry so if we discuss the situation for some specified $\theta$ plane then it would be the same for all. Thus this no longer represent a radially ingoing particle but a more generalized case where the particle has two variables to specify namely, ($r,\phi$). The motion is determined by the Euler equations corresponding to the lagrangian formed as $2L=g_{\mu \nu}\dot{x}_{\mu}\dot{x}_{\nu}$, Which has the following explicit form (using equation ($\ref{i1}$)), $$\label{v1} 2L=-f(r)\dot{t}^{2}+\frac{1}{f(r)}\dot{r}^{2}+r^{2}\dot{\phi}^{2}$$ where dot denotes differentiation with respect to proper time of the particle. This equation can be written in terms of the particle proper time and then along the orbit we have $2L=-1$. This finally leads to, $$\label{v2} d\tau ^{2}=f(r)dt^{2}-\frac{1}{f(r)}dr^{2}-r^{2}d\phi ^{2}=f(r)dt^{2}[1-v^{2}]$$ where $$\label{v3} v^{2}=\frac{1}{f(r)^{2}}\left(\frac{dr}{dt}\right)^{2}+\frac{r^{2}}{f(r)}\left(\frac{d\phi}{dt}\right)^{2}$$ This is the velocity of the particle with respect to a static observer (r=constant) as illustrated by plugging $u^{\mu}=(1,0,0,0)$ in equation (\[i4\]); i.e. the particle moves through a distance $\frac{1}{\sqrt{f}}\sqrt{dr^{2}+r^{2}fd\phi ^{2}}$ in a proper time given by $\sqrt{f}dt$, where from now on we shall use simply $f$ for $f(r)$ due to notational simplicity. Since the lagrangian as given in ($\ref{v1}$) do not contain $t$ explicitly we have a constant of motion which is nothing but the energy of the particle and it is given by, $$\label{v4} -\frac{\partial L}{\partial \dot{t}}=f\dot{t}=E$$ This constant of motion actually originates from the killing vector field $\frac{\partial}{\partial t}$, this can be phrased as, if the 4-velocity of the particle $t^{a}$ is a geodesic, then we have $\triangledown _{t}t=0$. From ($\ref{v2}$) and ($\ref{v4}$) we have obtained, $$\label{v5} v=\sqrt{1-\frac{f}{E^{2}}}$$ Also the energy can be determined from the initial value of radius and velocity using ($\ref{v5}$) such that, $E^{2}=\frac{f(R)}{1-v_{0}^{2}}$. Where $R$ is the initial radial co-ordinate and $v_{0}$ is the initial velocity. We have another constant of motion in this case which corresponds to the angular momentum of the particle and could be given by, $$\label{v6} \dot{\phi}=\frac{L}{r^{2}}$$ Thus finally the velocity in proper frame on the plane $\theta =\frac{\pi}{2}$ is given by, $$\label{v7} \left(\frac{dr}{d\tau}\right)^{2}=\left(\frac{dr}{dt}\right)^{2}\left(\frac{dt}{d\tau}\right)^{2}=E^{2}-V^{2}$$ where $V^{2}=f\left[1+\frac{r^{2}}{f^{2}}E^{2}\left(\frac{d\phi}{dt}\right)^{2}\right]=f\left[1+\frac{L^{2}}{r^{2}}\right]$. Thus the 4-velocity components for the geodesic particle specified by energy and angular momentum is given by, $$\label{v8} t^{\mu}=\left(\frac{E}{f},\sqrt{E^{2}-V^{2}},0,\frac{L}{r^{2}}\right)$$ written in terms of the constants of motion $E$ and $L$. As a check we can use the identity $t_{\mu}t^{\mu}=-1$. Thus equation (\[v8\]) represents the four velocity of a test particle in a space-time metric given by equation (\[i1\]). Static Limit ------------ In some cases the velocity is measured in terms of proper time, as determined by clocks synchronized along trajectory of the particle. The velocity in case of radial particle is given by [@lan71], $$\label{v9} v^{2}=\left(g_{00}+g_{01}\frac{dx^{1}}{dx^{0}}\right)^{-2}\left(g_{01}^{2}-g_{00}g_{11}\right)\left(\frac{dx^{1}}{dx^{0}}\right)^{2}$$ When we generalize this result to our case where we have three co-ordinates $x^{0},x^{1}$ and $x^{3}$ (since $x^{2}=\theta =constant$), then velocity expression generalizes to, $$\label{v10} v^{2}=\frac{\left(g_{10}^{2}-g_{00}g_{11}\right)\left(\frac{dx^{1}}{dx^{0}}\right)^{2}+ \left(g_{30}^{2}-g_{00}g_{33}\right)\left(\frac{dx^{3}}{dx^{0}}\right)^{2}+ 2\left(g_{10}g_{30}-g_{13}g_{00}\right)\frac{dx^{1}}{dx^{0}}\frac{dx^{3}}{dx^{0}}} {\left(g_{00}+g_{10}\frac{dx^{1}}{dx^{0}}+g_{30}\frac{dx^{3}}{dx^{0}}\right)^{2}}$$ Note that if we let $\frac{dx^{3}}{dx^{0}}$ to be zero, then it reduces to equation ($\ref{v9}$). In our case keeping the non zero terms we obtain, $$\label{v11} v^{2}=\frac{1}{f^{2}}\left( \frac{dr}{dt} \right)^{2}+\frac{r^{2}}{f}\left( \frac{d\phi}{dt} \right)^{2}$$ which is completely identical to ($\ref{v3}$). This definition has co-ordinate invariance. The 4 velocity has components $u_{\mu}=(-g_{00})^{-1/2}g_{\mu 0}$ and that for the particle reduces to $t^{\mu}=(\frac{dx^{0}}{d\tau},\frac{dx^{1}}{d\tau},0,\frac{dx^{3}}{d\tau})$. Thus using equation ($\ref{i3}$) we obtain the same equation as ($\ref{v11}$). From equation ($\ref{v5}$) we see that as $f(r)=0$, the velocity is equal to 1. Hence for static observers $v$ approaches the speed of light at the event horizon and they predict faster than light speed inside event horizon. It might seem at first sight that this result has nothing to do with $f(r)=0$ but is connected to the co-ordinate system. However it has nothing to do with co-ordinate system but with the observer. So we should generalize our observer set. Also no observer can be at rest at $r=2M$ except photon, with respect to photon all particle traverse at speed of light. To get a clear view we discuss the acceleration of a static observer in the field of the gravitating body. The acceleration is necessary as in general relativity an observer at rest is not geodesic and is accelerated. The four acceleration field is defined as, $$\label{v12} a^{\eta}=u^{\eta}_{;\mu}u^{\mu}=\left(u^{\eta}_{,\mu}+u^{\alpha}\Gamma ^{\eta}_{\alpha \mu}\right)$$ The only non zero component is given by using the definition of four velocities for static observers , $u_{\mu}=\frac{g_{\mu 0}}{\sqrt{-g_{00}}}$ to yield, $$\label{v13} a^{1}=\frac{1}{2}\frac{df}{dr}$$ So acceleration depends on the function $f(r)$. Ingoing Observers ----------------- ![The figure shows variation of $v^{2}$ with radial co-ordinate r for different choice of $E_{1}$, $E_{2}$, $L_{1}$ and $L_{2}$.\[fig1\]](velocity1.eps){height="3.5in" width="3.5in"} ![The figure shows variation of $v^{2}$ with test particle energy for different observer energy and radial distance.\[fig2\]](velocity2.eps){height="3.5in" width="3.5in"} ![The figure shows variation of $v^{2}$ with test particle angular momentum for different choices of test particle energy.\[fig3\]](velocity3a.eps){height="3.5in" width="3.5in"} ![The figure shows variation of $v^{2}$ with $E_{2}$ and r.\[fig4\]](velocity3b.eps){height="3.5in" width="3.5in"} We consider motion of two particles such that the four velocities are given by, $$\begin{aligned} \label{v14} \left.\begin{array}{c} t^{\mu}=\left( \frac{E_{1}}{f}, \sqrt{E^{2}_{1}-V^{2}_{1}}, 0, \frac{L^{2}_{1}}{r^{2}} \right)\\ u^{\nu}=\left( \frac{E_{2}}{f}, \sqrt{E^{2}_{2}-V^{2}_{1}}, 0, \frac{L^{2}_{2}}{r^{2}} \right) \end{array}\right\}\end{aligned}$$ Hence we obtain the following result, $$\label{v15} t^{\mu}u_{\mu}=g_{\mu \nu}t^{\mu}u^{\nu}=-\frac{E_{1}E_{2}}{f}+ \frac{\sqrt{\left( E^{2}_{1}-V^{2}_{1} \right) \left( E^{2}_{2}-V^{2}_{1} \right)}}{f}+\frac{L_{1}L_{2}}{r^{2}}$$ Thus we obtain, $$\label{v16} \left( t^{\mu}u_{\mu} \right)^{2}=\left( \frac{E_{1}E_{2}}{f} \right)^{2} \left[ 1-\frac{fL_{1}L_{2}}{r^{2}E_{1}E_{2}}-\sqrt{\left( 1-fa_{1} \right)\left( 1-fa_{2} \right)} \right]^{2}$$ Where, $a_{1}=\frac{1}{E^{2}_{1}}\left( 1+\frac{L^{2}_{1}}{r^{2}} \right)$ and $a_{2}=\frac{1}{E^{2}_{2}}\left( 1+\frac{L^{2}_{2}}{r^{2}} \right)$. Simplifying and rearranging terms we have obtained that, $$\label{v17} \left( t^{\mu}u_{\mu} \right)^{2}=E^{2}_{1}E^{2}_{2} \left[\frac{1}{2}\left( a_{1}+a_{2} \right)-\frac{L_{1}L_{2}}{r^{2}E_{1}E_{2}} \right]^{2}$$ We know that the relative velocity could be given by, $v^{2}=1-\frac{1}{\left(u^{\mu}t_{\mu}\right)^{2}}$. Thus using equation ($\ref{v17}$) and assuming that energy of both the particle and the observer are high enough or the distance is large enough we ultimately arrive at, $$\label{v18} v^{2}=1-\frac{4}{E^{2}_{1}E^{2}_{2}\left[ \left( \frac{1}{E^{2}_{1}}+\frac{1}{E^{2}_{2}} \right)+ \frac{1}{r^{2}}\left( \frac{L_{1}}{E_{1}}-\frac{L_{2}}{E_{2}} \right)^{2} \right]^{2}}$$ Note that as $r\rightarrow 0$ the velocity approaches that of light i.e. $v=1$. However if the particle and the observer has the same impact parameter i.e. $\frac{L_{1}}{E_{1}}=\frac{L_{2}}{E_{2}}$ then even if $r\rightarrow 0$ the velocity does not approach $1$, which is a very interesting result. Also at short distance the velocities and hence energies are very high so $a_{1}$ and $a_{2}$ are small quantities, however at large distance not $a_{1}$ and $a_{2}$ but $f(r)$ become smaller and thus as they appear in product form in the velocity expression it holds good for all $r$. Thus we can say that equation ($\ref{v18}$) is a general result. This result is valid in spherically symmetric solutions for Einstein gravity like the Schwarzschild and Reissner-Nordström solutions but also for the Einstein-Maxwell-Gauss Bonnet theory. There exists two additional well known spherically symmetric solution but they do not have the form used. So we shall consider them in the next two sections. Note from figure-$\ref{fig1}$, as radial co-ordinate of the particle is decreased the velocity remain less than speed of light. As $r\rightarrow 0$ the velocity also approaches $1$ in our system of units, which is justified and shows the actual motion that happen as the particle moves within the event horizon. It is also clear that with increase of the energy of the particle the velocity increases and it also increases with increasing the angular momenta. From figure-$\ref{fig2}$ we see that as energy of the particle is increased we get a interesting behavior, at first it decreases and become zero, then it again increases. Thus here the combined quantity in the denominator becomes 4. This happens when $E_{1}$ coincides with $E_{2}$ (see the figure), as we have chosen $L_{1}=L_{2}$ (see equation $\ref{v18}$). However changing the radius has a very small effect on velocity profile. From figure-$\ref{fig3}$ we find that velocity varies with angular momentum in some what the same manner as it does with energy. However by proper choice of $E_{2}=E_{1}$ the velocity can be made zero when $L_{2}=E_{2}$ as we have chosen other parameters such that $L_{1}=E_{1}$, since under this condition the denominator in equation ($\ref{v18}$) become $4$. As well as we can eliminate that zero by changing $E_{2}$. Figure-$\ref{fig4}$ shows the variation of velocity both with radial co-ordinate and the energy of the particle. This graph merely shows combined effects of varying radius and energy as we have illustrated in earlier graphs. Motion for Some Classes of Alternative Gravity Theories {#vs2} ======================================================= Current theoretical cosmology has two fundamental problems, namely inflation and late time acceleration of the universe. The usual scenarios used to explain both these accelerating cosmology epochs are to develop acceptable dark energy model, such as: scalar, spinor, cosmological constant and higher dimensions. Even if such a scenario seems to be partially succesful it is hindered by the coupling with usual matter, compatibility with standard elementary particle theories. However another natural choice is the classical generalization of general relativity, called modified gravity or alternative gravity theory ([@cald03], [@noj03],[@noj07],[@noj11]). Thus a gravitational alternative to explain inflation and dark energy seems very reasonable on the ground of the expectation that general relativity is just an approaximation that is valid at small curvature. A sector of modified gravity containing the gravitational terms relevent at high energy produced the inflationary epoch. During evolution curvature decreases and general relativity describes to an good approaximation the intermediate universe. With a furthur decrease of curvature as sub-dominant terms grow we see a transition from deceleration to cosmic acceleration. There exists traditional $F(R)$, string inspired models, scalar tensor theory, Gauss-Bonnet theory and some other models. In the next subsections we shall discuss motion of a test particle and hence its velocity in four spherically symmetric solutions for different alternative gravity theories. Motion in Dilaton Coupled Electromagnetic Field {#va1} ----------------------------------------------- Static uncharged black hole in general relativity are described by Schwarzschild solution. If mass of the black hole is much large compared to Planck mass then this also, to a good approximation, describes the uncharged black hole in string theory except regions near singularity. However there was some departure from the schwarzschild scenario when an exact calculation is made [@Yunes]. We shall discuss this solution later in this work. From now on we shall assume that the above assertion is correct. However for Einstein-Maxwell solutions the string inspired theory differ widely from the known classical solution i.e. the Reissner-Nordström solution. The dilaton coupling with $F^{2}$ implies that every solution with non zero $F_{\mu \nu} $ will come with a non zero dilaton. Thus the charged black hole solution in general relativity (which is the Reissner-nordström solution) appears in a new form in string theory due to the presence of dilaton. The effective four dimensional low energy Lagrangian obtained from string theory is, $$S=\int d^{4}x \sqrt{-g}[-R+e^{-2\Phi}F^{2}+2(\nabla\Phi)^{2}]$$ where $F_{\mu \nu} $ is the Maxwell field associated with a $U(1)$ subgroup of $E_{8}\times E_{8}$ or $Spin(32)/Z_{2}$. We have set the remaining gauge fields and antisymmetric tensor field $H_{\mu \nu \rho}$ to zero and $\Phi $ is the dilaton field ([@Garfinkle],[@Coleman],[@Vega],[@Bekensteina],[@Bekensteinb],[@Bocha],[@Witten2]). Extremizing with respect to the $U(1)$ potential $A_{\mu}$, $\Phi$ and $g_{\mu \nu}$ leads to the following field equations, $$\begin{aligned} \label{n1} \left.\begin{array}{c} (a) \nabla _{\mu} \left(e^{-2\Phi}F^{\mu \nu} \right)=0\\ (b) \nabla ^{2}\Phi +\frac{1}{2}e^{-2\Phi}F^{2}=0\\ (c) R_{\mu \nu}=2\nabla _{\mu}\Phi \nabla _{\nu}\Phi +2e^{-2\Phi}F_{\mu \lambda}F^{\lambda}_{\nu}-\frac{1}{2}g_{\mu \nu}e^{-2\Phi}F^{2} \end{array}\right\}\end{aligned}$$ The static spherically symmetric solution corresponding to the above field equation (\[n1\]) would give the following line element as, [@Garfinkle] $$ds^{2}=-(1-\frac{2M}{r})dt^{2}+\frac{1}{(1-\frac{2M}{r})}dr^{2}+r(r-e^{2\Phi_{0}}\frac{Q^{2}}{M})d\Omega^{2}$$ where, $d\Omega^{2}=d\theta^{2}+sin^{2}\theta d\phi^{2}$. Once again due to isometry we have taken our motion in the equatorial plane such that, $d\Omega^{2}=d\phi^{2}$. Here $\Phi_{0}$ is the asymptotic value of dilaton and $Q$ represents the black hole charge. Note that this is almost identical to the Schwarzschild metric, with a difference that areas of spheres of constant r and t now depend on Q. In particular the surface $r=\frac{Q^{2}e^{2\Phi_{0}}}{M}$ is singular. Also $r=2M$ is the regular event horizon. Also the evolution of the scalar field $\Phi$ could be given by, $$\label{n2} e^{-2\Phi}=e^{-2\Phi _{0}}-\frac{Q^{2}}{Mr}$$ We can define the dilaton charge as, $$D=\frac{1}{4\pi}\int d^{2}\sigma^{\mu}\nabla_{\mu}\Phi$$ where the integral is over a two sphere at spatial infinity and $\sigma^{\mu}$ is the normal to the two sphere at spatial infinity. For charged black hole this leads to, $$\label{n3} D=-\frac{Q^{2}e^{2\Phi_{0}}}{2M}$$ Here D depends on the asymptotic value of dilaton field, which is determined once M and Q are given and is always negative. Note that the actual dependance on dilaton field is described by, $e^{-\Phi /M_{pl}}$. Since we have walked in the unit $M_{pl}\sim 1$ we have the term modified to $e^{-\Phi}$. so as $\Phi \rightarrow \Phi _{0}\sim M_{pl}$, this term is expected to become significant.\ Now we write the above metric in a generalized form, $$\label{v21} 2L=-f(r)\dot{t}^{2}+\frac{1}{f(r)}\dot{r}^{2}+g(r)\dot{\phi}^{2}$$ As usual we have $f(r)=(1-\frac{2M}{r})$ and $g(r)=r(r-e^{2\Phi_{0}}\frac{Q^{2}}{M})$, however due to notational simplicity we have taken them to be simply $f$ and $g$ respectively. Then the velocity has the following expression, $$\label{v22} v^{2}=\frac{1}{f^{2}}\left( \frac{dr}{dt} \right)^{2}+\frac{g}{f}\left(\frac{d\phi}{dt}\right)^{2}$$ The potential has the following expression which could be given by, $$\label{v23} V^{2}=f\left(1+\frac{L^{2}}{g}\right)$$ Thus 4-velocity components are given by for this potential to yield, $$\label{v24} t^{\mu}=\left(\frac{E}{f},-\sqrt{E^{2}-V^{2}},0,\frac{L}{g}\right)$$ Note that for this case as well we have the following result $t^{\mu}t_{\mu}=-1$. If we have used equation ($\ref{v10}$) then we might have obtained that the velocity has the same expression as that given by ($\ref{v22}$). Also the acceleration has no change only $a^{1}$ is non zero and has the value given by ($\ref{v13}$). If we proceed in an identical way then we obtain the following result for the velocity of a particle relative to an observer, $$\label{v25} v^{2}=1-\frac{4}{E^{2}_{1}E^{2}_{2}\left[ \left( \frac{1}{E^{2}_{1}}+\frac{1}{E^{2}_{2}} \right)+ \frac{1}{g}\left( \frac{L_{1}}{E_{1}}-\frac{L_{2}}{E_{2}} \right)^{2} \right]^{2}}$$ The most interesting part of this velocity expression corresponds to the fact that when at some finite $r$ the quantity $g=0$ then $v=1$. So even if it had not go to $r=0$ the particle is seen to move with the velocity of light. However under the same situation as above such that both the particle and the observer moves with the same impact parameter we obtain that this case is prohibited and the particle has $v$ less than $1$ for all $r$. This singularity corresponds to $r=e^{2\Phi_{0}}\frac{Q^{2}}{M}$ However this particular result is actually an artifact of our co-ordinate system. For string theory, the statement that the spacetime has singularity when $r=e^{2\Phi_{0}}\frac{Q^{2}}{M}$ is actually irrelevant. Since the strings do not couple to the metric $g_{\mu \nu}$ but rather to $e^{2\Phi} g_{\mu \nu}$. This metric appear in string $\sigma$ model. In terms of the string metric the effective lagrangian would become [@Garfinkle], $$S=\int d^{4}x \sqrt{-g}e^{-2\Phi}\left[-R-4(\nabla \Phi)^{2}+F^{2} \right]$$ Hence the charged black hole metric, $$\label{n4} ds^{2}_{string}=-\frac{1-2Me^{\Phi _{0}}/\rho}{1-Q^{2}e^{3\Phi _{0}}/M\rho}d\tau ^{2}+\frac{d\rho ^{2}}{\left(1-2Me^{\Phi _{0}}/\rho \right)\left(1-Q^{2}e^{3\Phi _{0}}/M\rho\right)}+\rho ^{2}d\Omega$$ This metric is identical to the metric given in equation $(\ref{i2})$ where we have just rescaled the metric by some conformal factor which is finite every where outside and on the horizon. With this choice of metric and the assumption that energy is high or radius is small we obtain the following expression for relative velocity, $$\label{v18a} v^{2}=1-\frac{4}{E^{2}_{1}E^{2}_{2}\left[ \left( \frac{1}{E^{2}_{1}}+\frac{1}{E^{2}_{2}} \right)+ \frac{1}{\rho ^{2}}\left( \frac{L_{1}}{E_{1}}-\frac{L_{2}}{E_{2}} \right)^{2} \right]^{2}}$$ This is completely identical to the result in equation ($\ref{v18}$), however the metric is completely different, here the general form would be $ds^{2}=-\frac{f(r)}{g(r)}dt^{2}+\frac{dr^{2}}{f(r)g(r)}+r^{2}d\Omega ^{2}$ where $f(r)=1-2Me^{\Phi _{0}}/\rho$ and $g(r)=1-Q^{2}e^{3\Phi _{0}}/M\rho$. Hence we arrive at a very important result that for both the spherically symmetric solution in section (\[vs1\]) and that for dilaton gravity has the same velocity profile. Spherically Symmetric Solution in Quadratic Gravity {#va2} --------------------------------------------------- In this section we consider a class of alternative theories of gravity in four dimensions defined by modifying the Einstein-Hilbert action through all possible quadratic, algebraic curvature scalars, multiplied by constants or non-constant couplings as ([@Yunes],[@Stewart],[@Green2]),\ $S=\int d^{4}x \sqrt{-g}[\kappa R+\alpha _{1}f_{1}(\upsilon)R^{2}+\alpha _{2}f_{2}(\upsilon)R_{ab}R^{ab}+\alpha _{3}f_{3}(\upsilon)R_{abcd}R^{abcd}$ $$\label{62} +\alpha _{4}f_{4}(\upsilon)R_{abcd}^{*}R^{abcd}-\frac{\beta}{2}\left(\nabla _{a}\upsilon \nabla ^{a}\upsilon +2V(\upsilon)\right)+L_{matter}]$$ where $g$ is the determinant of the metric $g_{ab}$; $(R,R_{ab},R_{abcd},R_{abcd}^{*})$ are the Ricci scalar and tensor, the Riemann tensor and its dual [@Yunes2], respectively; $L_{matter}$ is the lagrangian density for other matter; $\upsilon$ is a scalar field; $(\alpha _{i},\beta)$ are coupling constants; and $\kappa=(16\pi G)^{-1}$. All other quadratic curvature terms are linearly dependent e.g., the Weyl tensor squared. Theories of this type are motivated from low energy expansion of string theory ([@Deser],[@Green]). Varying equation $(\ref{62})$ with respect to the metric and setting $f_{i}(\upsilon)=1$, we find the modified field equations, $$\label{63} \kappa G_{ab}+\alpha _{1}H_{ab}+ \alpha _{2}I_{ab}+ \alpha _{3}J_{ab}=\frac{1}{2}T_{ab}^{matter}$$ where $T_{ab}^{matter}$ is the stress energy of matter, and, $$\begin{aligned} \label{64} \left. \begin{array}{c} (a) H_{ab}=2R_{ab}R-\frac{1}{2}g_{ab}R^{2}- 2 \nabla _{ab}R+ 2g_{ab}\square R\\ (b) I_{ab}=\square R_{ab}+2R_{abcd}R^{cd}-\frac{1}{2}g_{ab}R_{cd}R^{cd}+\frac{1}{2}g_{ab}\square R -\nabla _{ab}R,\\ (c) J_{ab}=8R^{cd}R_{acbd}-2g_{ab}R^{cd}R_{cd}+4\square R_{ab}-2RR_{ab}+\frac{1}{2}g_{ab}R^{2}-2\nabla _{ab}R \end{array}\right \}\end{aligned}$$ with $\nabla _{a}$, $\nabla _{ab}=\nabla _{a}\nabla _{b}$, and $\square = \nabla _{a}\nabla ^{a}$ the first and second order covariant derivative and the D’Alembertian. The scalar field equation can be given by, $$\label{64a} \beta \square \upsilon -\beta \frac{dV}{d\upsilon}=-\alpha _{1}R^{2}-\alpha _{2}R_{ab}R^{ab}-\alpha _{3}R_{abcd}R^{abcd}- \alpha _{4}R_{abcd}^{*}R^{abcd}$$ \ The spherically symmetric solution to the above field equations imposing dynamical arguments could be written using the metric ansatz as [@Yunes], $$\label{65} ds^{2}=-f_{0}\left[1+\epsilon h_{0}(r)\right]dt^{2}+ f_{0}^{-1}\left[1+\epsilon k_{0}(r)\right]dr^{2}+r^{2}d\Omega ^{2}$$ and $\upsilon = \upsilon _{0}+\epsilon \upsilon _{0}$, where $f_{0}=1-2M_{0}/r$, with $M_{0}$ the bare or GR BH mass and $d\Omega _{2}$ is the line element on two sphere. The free functions $(h_{0},k_{0})$ are small deformations about the Schwarzschild metric. The scalar field equation can be solved to yield, $$\label{66} \upsilon _{0}=\frac{\alpha _{3}}{\beta}\frac{2}{M_{0}r}\left(1+\frac{M_{0}}{r}+\frac{4M_{0}^{2}}{3r^{2}} \right)$$ We can use this scalar field solution to solve modified field equations to linear in $\epsilon$. Requiring the metric to be asymptotically flat and regular at $r=2M_{0}$, we find the unique solution $h_{0}=F\left(1+\tilde{h_{0}}\right)$ and $K_{0}=-F\left(1+\tilde{h_{0}}\right)$, where $F=-(49/40)\zeta (M_{0}/r)$ and,                                   $\tilde{h_{0}}=\frac{2M_{0}}{r}+\frac{548}{147}\frac{M_{0}^{2}}{r^{2}}+\frac{8}{21}\frac{M_{0}^{3}}{r^{3}}-\frac{416}{147}\frac{M_{0}^{4}}{r^{4}}-\frac{1600}{147}\frac{M_{0}^{5}}{r^{5}}$ $$\label{67} \tilde{k_{0}}=\frac{58}{49}\frac{M_{0}}{r}+\frac{76}{49}\frac{M_{0}^{2}}{r^{2}}-\frac{232}{21}\frac{M_{0}^{3}}{r^{3}}-\frac{3488}{147}\frac{M_{0}^{4}}{r^{4}}-\frac{7360}{147}\frac{M_{0}^{5}}{r^{5}}$$ Here we have defined the dimensionless coupling function $\zeta=\frac{\alpha _{3}^{2}}{\beta \kappa M_{0}^{4}}$, which is of the order of $\epsilon$. Such a solution is most general for all dynamical, algebraic, quadratic gravity theories, in spherical symmetry. We can define the physical mass $M=M_{0}\left[1+(49/80)\zeta\right]$, such that only modified metric components become $g_{tt}=-f(1+h)$ and $g_{rr}=f^{-1}(1+k)$ where $h=\zeta /(3f)(M/r)^{3}\tilde{h}$ and $k=-(\zeta / f)(M/r)^{2}\tilde{k}$, and $$\label{68} \tilde{h}=1+\frac{26M}{r}+\frac{66}{5}\frac{M^{2}}{r^{2}}+\frac{96}{5}\frac{M^{3}}{r^{3}}-\frac{80M^{4}}{r^{4}}$$ $$\label{69} \tilde{k}=1+\frac{M}{r}+\frac{52}{3}\frac{M^{2}}{r^{2}}+\frac{2M^{3}}{r^{3}}+ \frac{16M^{4}}{5r^{4}}- \frac{368}{3}\frac{M^{5}}{r^{5}}$$ where $f=1-2M/r$. Note from the above expression for metric element that Physical observables are related to renormalized mass $M$ not on bare mass $M_{0}$.\ ![The figure shows variation of $v^{2}$ with test particle angular momentum $L_{2}$ for different choices of $E_{2}$.\[fig5\]](velocitya1.eps){height="3.5in" width="3.5in"} ![The figure shows variation of $v^{2}$ with test particle energy for different observer energy and test particle angular momenta.\[fig6\]](velocitya2.eps){height="3.5in" width="3.5in"} ![The figure shows variation of $v^{2}$ with test particle energy and radial distance.\[fig7\]](velocitya3.eps){height="3.5in" width="3.5in"} ![The figure shows variation of $\Delta v^{2}$ with $E_{2}$.\[fig8\]](velocitya4.eps){height="3.5in" width="3.5in"} ![The figure shows variation of $\Delta v^{2}$ with $L_{2}$.\[fig9\]](velocitya5.eps){height="3.5in" width="3.5in"} In this case the lagrangian has the specific form given by, $$\label{70} 2L=-f(r)\left[1+h(r)\right]\dot{t}^{2}+\frac{\left[1+k(r)\right]}{f(r)}\dot{r}^{2}+r^{2}\dot{\phi}^{2};$$ from this we can easily found components of velocity by differentiation. Since the lagrangian does not involve time we have two conserved quantities, $E$ the energy per particle mass and $L$ the angular momentum per particle mass given by, $$\begin{aligned} \label{71} \left.\begin{array}{c} E=-\frac{\partial L}{\partial \dot{t}}=f(r)\left[1+h(r)\right]\dot{t}\\ L=\frac{\partial L}{\partial \dot{\phi}}=r^{2}\dot{\phi} \end{array}\right\}\end{aligned}$$ where the time derivatives are with respect to affine co-ordinate $\tau$. Finally the equation of motion would be given by [@Yunes], $$\label{72} \left(\frac{dr}{d\tau}\right)^{2}=V_{eff}^{GR}-\left[E^{2}h(r)+V_{eff}^{GR}k(r)\right]=V_{eff}$$ where we have obtained $V_{eff}^{GR}=E^{2}-f(r)\left[1+\frac{L^{2}}{r^{2}}\right]$. Then the 4-velocity vector could be given by, $$\label{73} t^{\mu}=\left(\frac{E}{f(1+h)},\sqrt{V_{eff}},0,\frac{L}{r^{2}}\right)$$ we can easily check that $t^{\mu}t_{\mu}=-1$. Now we can proceed in an identical way as presented in the previous two sections and that finally leads to the following expression for relative velocity of a particle with respect to an observer in this space-time to yield, $$\begin{aligned} \label{v71} \begin{array}{c} v^{2}=v^{2}_{GR}-\Delta v^{2}=1-\frac{4}{E^{2}_{1}E^{2}_{2}\left[ \left( \frac{1}{E^{2}_{1}}+\frac{1}{E^{2}_{2}} \right)+ \frac{1}{\rho ^{2}}\left( \frac{L_{1}}{E_{1}}-\frac{L_{2}}{E_{2}} \right)^{2} \right]^{2}} -\frac{2f^{2}}{E^{2}_{1}E^{2}_{2}\left[1-\frac{fL_{1}L_{2}}{E_{1}E_{2}r^{2}}- \sqrt{E^{2}_{1}- V^{2}_{1}}\sqrt{E^{2}_{2}-V^{2}_{2}}\right]^{3}} \\ \left[h\left(1-\frac{fL_{1}L_{2}}{E_{1}E_{2}r^{2}}- \sqrt{E^{2}_{1}-V^{2}_{1}}\sqrt{E^{2}_{2}-V^{2}_{2}}\right) + 2 \left(\frac{V_{1}V_{2}} {E_{1}E_{2}}\left(h+k+\frac{1}{2}\left(E_{1}\frac{\delta V_{1}}{V^{2}_{1}}+E_{2}\frac{\delta V_{2}} {V^{2}_{2}}\right)\right)+\frac{L_{1}L_{2}fh}{r^{2}E_{1}E_{2}}\right)\right] \end{array}\end{aligned}$$ where we have defined $V_{1}=V_{eff}^{GR}(E_{1},L_{1})$ and similarly $V_{2}=V_{eff}^{GR}(E_{2},L_{2})$ with similar interpretation such that $\delta V=-hE^{2}-kV_{eff}^{GR}$. Here the quantities $f,h,k$ are defined earlier, among them $f(r)=1-2M/r$ and $h,k$ are given by equations (\[68\]) and (\[69\]). Also note that first two terms are just the velocity expression we have obtained in equation (\[v18\]) for a general spherically symmetric solution and in equation (\[v18a\]) for dilaton coupled gravity and refereed to $v_{GR}^{2}$. Also note that the last term which is the correction term due to alternative gravity has a negative contribution and when $\zeta =0$ then we recover our original equation (\[v18\]). Figure-$\ref{fig5}$ and figure-$\ref{fig6}$ represents the variation of $v^{2}$ with test particle angular momentum and energy respectively, as well as figure-$\ref{fig7}$ represents the variation with both test particle energy and radial distance. We can very easily verify by comparison with previous graphs that the effect of introducing quadratic terms in the action alters the velocity profile near $r=0$ and for low test particle energy and angular momentum. The effect of test particle energy and angular momentum on the extra piece $\delta v^{2}$ is shown in the figure-$\ref{fig8}$ and figure-$\ref{fig9}$, which verifies our previous assertion. At low energy and angular momentum the velocity is mostly dictated by the gravitational effect of the source and that is when the effect of introduction of quadratic terms could be evident. Hence the above result can be interpreted as a astrophysical manifestation of the stringy signature, as these quadratic terms come from some high energy effective string theory. Motion in Einstein-Maxwell-Gauss-Bonnet Gravity {#va3} ----------------------------------------------- Theories with extra spatial dimension have been an active area of interest even since the original work of Kaluza and Klein, and the advent of string theory which predicts the presence of extra spatial dimension. Among many alternatives the Brane world scenario is considered as a strong candidate which has theoretical basis in some underlying string theory. Usually, the effect of string theory on classical gravitational physics ([@Green2],[@Davies]) is investigated by means of a low energy effective action, which in addition to the Einstein-Hilbert action contain squares and higher powers of curvature term. However the field equations become fourth order and brings in ghosts [@Zumino]. In this context Lovelock [@Lovelock] showed that if the higher curvature terms appear in a particular combination, the field equation become second order and consequently the ghosts disappear. In Einstein-Maxwell-Gauss-Bonnet (EMGB) gravity, the action in five dimensional spacetime ($M,g_{\mu \nu}$) can be written as, $$\label{46} S=\frac{1}{2}\int _{M} d^{5}x \sqrt{-g} \left[R+\alpha L_{GB}+L_{matter} \right],$$ where $L_{GB}=R_{\alpha \beta \gamma \delta}R^{\alpha \beta \gamma \delta}-4R_{\mu \nu}R^{\mu \nu}+R^{2}$ is the GB Lagrangian and $L_{matter}=F^{\mu \nu}F_{\mu \nu}$ is the Lagrangian for the electromagnetic field. Here $\alpha$ is the coupling constant of the GB term having dimension $(length)^{2}$. As $\alpha$ is regarded as inverse string tension, so $\alpha \geq 0$. The gravitational and electromagnetic field equations obtained by varying the above action with respect to $g_{\mu \nu}$ and $A_{\mu}$ we could have obtained (see [@Chakraborty]), $$\begin{aligned} \label{47} \left. \begin{array}{c} G_{\mu \nu}-\alpha H_{\mu \nu}=T_{\mu \nu}\\ \bigtriangledown _{\mu}F^{\mu}_{\nu}=0\\ H_{\mu \nu}=2\left[RR_{\mu \nu}-2R_{\mu \lambda}R^{\lambda}_{\mu}-2R^{\gamma \delta}R_{\mu \gamma \nu \delta}+R^{\alpha \beta \gamma}_{\mu}R_{\nu \alpha \beta \gamma} \right]-\frac{1}{2}g_{\mu \nu}L_{GB} \end{array}\right \}\end{aligned}$$ where $T_{\mu \nu}=2F^{\lambda}_{\mu}F_{\lambda \nu}-\frac{1}{2}F_{\lambda \sigma}F^{\lambda \sigma}g_{\mu \nu}$ is the electromagnetic field tensor. A spherically symmetric solution to the above action has been obtained by [@Dehghani] and the line element is given by, $$\label{48} ds^{2}=-g(r)dt^{2}+\frac{dr^{2}}{g(r)}+r^{2}d\Omega _{3}^{2},$$ where the metric co-efficient is, $$\label{49} g(r)=K+\frac{r^{2}}{4\alpha}\left[1\pm \sqrt{1+\frac{8\alpha \left(m+2\alpha \mid K \mid \right) }{r^{4}} -\frac{8\alpha q^{2}}{3r^{6}}} \right]$$ Here $K$ is the curvature, $m+2\alpha \mid K\mid$ is the geometrical mass and $d\Omega _{3}^{2}$ is the metric of a 3D hypersurface such that, $$\label{50} d\Omega _{3}^{2}=d\theta _{1}^{2}+sin^{2}\theta _{1}\left( d\theta _{2}^{2}+sin^{2}\theta _{2}d\theta _{3}^{2}\right)$$ The range is given by $\theta _{1},\theta _{2}:[0,\pi]$. We assume that there is a constant charge $q$ at $r=0$ and the vector potential be $A_{\mu}=\Phi (r)\delta_{\mu}^{0}$ such that $\Phi (r)=-\frac{q}{2r^{2}}$. In this metric the metric function $g(r)$ will be real for $r \geq r_{0}$ where $r_{0}^{2}$ is the largest real solution of the cubic equation, $$\label{51} 3z^{3}+24\alpha \left(m+2\alpha \mid K \mid \right)z-8\alpha q^{2}=0$$ By a transformation of the radial co-ordinates we can show that $r=r_{0}$ is an essential singularity of the spacetime. We shall choose $K=1$ and shall consider the $-$ve sign in front of square root of equation $(\ref{49})$ which leads to asymptotically flat solution. However note that the line element as presented in equation ($\ref{48}$) is exactly of the same form as we have used in equation ($\ref{i1}$). Thus the velocity of a test particle relative to an observer would have the same form as presented in equation ($\ref{v18}$). Thus all the properties of this velocity remain valid in this EMGB gravity and shows the usefulness of our definition of velocity. Motion in F(R) gravity {#va4} ---------------------- General Relativity (GR) is a widely accepted as a fundamental theory relating matter energy density to geometric properties of spacetime. The standard big-bang cosmological model can explain the evolution of the universe well except inflation and late time cosmic acceleration. Although many scalar field models have been constructed in the frame work of string theory and supergravity to explain inflation but Cosmic Microwave Background radiation still do not show any evidence in favor of a particular model. The same kind of approach is also taken to explain cosmic acceleration by introducing different dark energy models where also concrete observation is still lacking. Thus one of the simplest choice is to modify GR action by introducing a term $F(R)$ in the lagrangian, where $F$ is an arbitrary function of scalar curvature $R$. There exists two methods for deriving field equations, first, by varying the action with respect to metric tensor $g_{\mu \nu}$. The other method called Palatini method should not be discussed here .In F(R) gravity ([@nel10],[@cor10],[@bal10],[@fel10]), the scalar curvature $R$ in the Einstein-Hilbert action $$\label{va11} S_{EH}=\int d^{4}x \sqrt{-g}\left(\frac{R}{16\pi} +L_{matter}\right),$$ gets replaced by an appropriate function of scalar curvature: $$\label{va12} S_{F(R)}=\int d^{4}x \sqrt{-g}\left(\frac{F(R)}{16\pi} +L_{matter}\right)$$ Varying this action we readily obtain the corresponding field equation to be given by, $$\label{va13} \frac{1}{2}g_{\mu \nu}F(R)-R_{\mu \nu}F'(R)-g_{\mu \nu}\square F'(R)+\nabla _{\mu}\nabla _{\nu}F'(R) =-4\pi T_{matter \mu \nu}$$ Several solutions (often exact) to this field equation may be found but due to complicated nature of field equations the number of such exact solutions are much less than that in general relativity. Without any matter and assuming the Ricci tensor to be covariantly constant equation ($\ref{va13}$) reduces to the following algebraic equation, $$\label{va14} 0=2F(R)-RF'(R)$$ From the above equation we can show that Schwarzchild-(anti-)de Sitter space is an exact vacuum solution to it. Thus the respective line element would be given by, $$\label{va15} ds^{2}=-\left(1-\frac{2M}{r}\mp \frac{r^{2}}{L^{2}}\right)dt^{2}+ \left(1-\frac{2M}{r}\mp \frac{r^{2}}{L^{2}}\right)^{-1}dr^{2}+r^{2}d\Omega ^{2}$$ Here the minus and plus sign corresponds to de Sitter and anti de Sitter space respectively, $M$ is the mass of the black hole and $L$ is the length parameter of (anti-)de Sitter space, which is related to the curvature $R=\pm \frac{12}{L^{2}}$ (the plus sign corresponds to de Sitter space and minus sign corresponds to anti de Sitter space). The vacuum solution for $F(R)$ gravity also has the same form as we have used in equation ($\ref{i1}$). Thus all the results of section \[vs1\] will remain valid here as well. Hence the relative velocity will have the same characteristics in vacuum solution for $F(R)$ gravity theory as well. This justifies our assertion as stated in section $\ref{vc}$. Discussion {#vd} ========== We have shown that velocity of any ingoing particle with respect to observer sets as defined in the section ($\ref{vc}$) for a general spherically symmetric potential with unit 2-sphere is always less than that of light outside the singular point, it approaches the speed of light as $r\rightarrow 0$. However the notion of static observers are not valid for $r\leq 2M$. It is valid only for region outside the event horizon. Thus we have defined ingoing observers and determine velocity with respect to the observer. We found that velocity of the test particle always remain less than 1. For a different choice of metric with a function on 2-sphere we found that the velocity is always less than 1 which may not be self-evident in one set of co-ordinates, but by going to another set we have actually shown that the previous results are retained. Finally the spherically symmetric solution in quadratic gravity shows another instance of the correctness of our result. However there we have obtained a correction factor to the velocity expression due to presence of quadratic terms and hence this directly shows that the velocity profile of an object differ considerably in alternative theories from the result in Einstein gravity. However that particular correction term would be Planck suppressed and hence very difficult to observe, however just out side the event horizon of the BH, where the tidal effects are huge these effects can in principle be observed. For the other two theories we have obtained the same expression as for the general spherically symmetric model. Thus they follow our previous assertion connecting to the relative velocity of a test particle. Also it should be noted that the above analysis is not restricted to Einstein gravity or the solutions we have discussed, it can also be applied to other spherically symmetric black hole solutions in other modified gravity theories. Also it could be extended to higher dimensional black holes. Extension to rotating black holes would be an interesting work for the future.\ The author thanks prof. Subenoy Chakraborty of Jadavpur University and Prof. Soumitra Sengupta of IACS for helpful discussion. The author also thanks DST, Govt. of India for awarding KVPY fellowship. He gratefully thanks IUCAA, Pune, for warm hospitality where a part of this work was done. [40]{} Landau, L. & Lifschitz, E., 1971, *The Classical Theory of Fields*, 3rd ed. (Addison-Wesley, Reading, Massachusetts) Wald, R. M., 1984 *General Relativity* (University of Chicago Press, Chicago, Illinois) Bergmann, P. G., 1942 *Introduction to The Theory of Relativity* (Prentice-Hall, New York) Moller, C., 1972 *The Theory of Relativity*, 2nd ed. (Oxford University Press, Delhi) Crawford, P. and Tereno, I., 2002, Gen. Relativity Gravitation 34, 2075 Janis, A. 1977, phys. Rev. D 15, 3068 Bol$o^{'}$s, V.J. 2011, preprint gr-qc/0506032v4 Ellis, G. F. R., Nel, S. D., Maartens, R., Stoeger, W. R. and Whitman, A.P. 1985, Phys. Rep. 124, 315 Bol$o^{'}$s, V. J., 2006 J. Geom. Phys. 56, 813 Chakraborty, Sumanta and Chakraborty Subenoy, 2011 Can. J. Phys. 89, 689 \[arxiv:1109.0676 \[gr-qc\], 2011\] Garfinkle, D., Horowitz, G. T. and Strominger, A. 1991 Phys. Rev. D 43, 3140 Yunes, N. and Stein, L. C. 2011 Phys. Rev. D 83, 104002 Teller, E.F. and Wheeler, J. A. 1992 *Spacetime Physics: Introduction to Special Relativity* (W. H. Freeman, San Fransisco) Coleman, S. 1983 in *The Unity of the Fundamental Interactions*, edited by A.Zichichi (Plenum, London) De Vega, H. J. and Sanchez, N. 1988 Nucl.phys. B309, 552 Bekenstein, J. 1972 Phys. Rev. D 5, 1239 Bekensein, J. 1975 Ann.phys. (N.Y.) 91, 75 Bocharova, N., Broonikov, K. and Melnikov, V. 1970 Vestn.Mosk.Univ.Fiz.Astron. 6, 706 Witten. E. (ed.) 1962 *Gravitation: An Introduction to Current Reaserch*, (Wiley, N.Y.) Caldwell, R. R., Kamionkowski, M. and Weinberg, N. N. 2003 Phys. Rev. Lett 91, 071301 Nojiri, S. and Odinstov, S. D. 2003 Phys. Rev. D 68, 123512 Nojiri, S. and Odinstov, S. D. arXiv:0807.0685 Nojiri, S. and Odinstov, S. D. 2011 Phys. Rep. 505, 59 Mignemi, S. and Stewart, N. R. 1993 Phys. Lett. B 298, 299 Green, M. B., Schwarz, J. H. and Witten, E. 1987 *Cambridge Monograph on mathematical physics* (Cambridge university press, Cambridge, England) Alexander, S. and Yunes, N. 2009 Phys. Rep. 480, 1 Boulware, D. G. and Deser, S. 1985 Phys. Review Lett. 55, 2656 Green, M. B., Schwarz, J. H. and Witten, E. 1987 *Loop Amplitudes, Anomalies and Phenomenology*, Superstring Theory Volume-2 (Cambridge: Cambridge University Press) Zumino, B. 1986 Phys. Rep. 137, 109 Lovelock, D., 1971 J. Math. Phys. 12, 498 Birrell, N. D. and Davies, P. C. W., 1982 *Quantum Fields in Curved Space* (Cambridge: Cambridge University Press) Chakraborty, S. and Bandyopadhyay, T., 2008 Class. Quantum. Grav. 25, 245015 Dehghani, M. H., 2004 Phys. Rev. D 70, 064019 Nelson, W., 2010 Phys. Rev. D 82, 124044 Corda, C., 2010 Eur. Phys. J 65, 257 Balcerzak, A. and Dabrowski, M. P., 2010 Phys. Rev. D 81, 123527 Felice, A. D. and Tsujikawa, S., 2010 Living Rev. Relativity 13, 3

ArXiv

--- author: - | Carlos Baquero, Paulo Sérgio Almeida and Ali Shoker\ HASLab, INESC TEC and Universidade do Minho,\ Braga, Portugal. bibliography: - 'ref.bib' - 'predef.bib' - 'bib.bib' - 'shapiro-bib.bib' - 'local.bib' title: 'Pure Operation-Based Replicated Data Types[^1]' --- [^1]: The work presented was partially supported by EU FP7 SyncFree project (609551), EU H2020 LightKone project (732505), and SMILES line in project TEC4Growth (NORTE-01-0145-FEDER-000020.

ArXiv

--- abstract: '[*The critical temperature and condensate fraction of a trapped interacting Bose gas are investigated when both atom-atom interaction and finite-size effects are taken into account. Canonical ensemble is used to obtain the equations on the condensate fraction for the trapped interacting Bose gas near and below the critical temperature. In our approaches corrections due to atom-atom interaction and finite-size effects are obtained simultaneously for the critical temperature and condensate fraction of the system. Analytical high-order correction to the condensate fraction is given in this work.* ]{}' address: - '$^{1}$Department of Applied Physics, Zhejiang University of Technology, Hangzhou, 310032, P. R. China' - '$^{2}$Zhijiang College, Zhejiang University of Technology, Hangzhou, 310012, P. R. China ' - '$^{3}$Department of Physics, East China Normal University, 200062, Shanghai, P. R. China' author: - 'Hongwei Xiong$^{1,2}$, Shujuan Liu$^{1}$, Guoxiang Huang$^{3}$, Zhijun Xu$^{1}$, Cunyuan Zhang$^{1}$' title: 'Critical temperature and condensate fraction of the trapped interacting Bose gas with finite-size effects' --- Introduction ============ The experimental realization of Bose-Einstein condensation (BEC) in the dilute alkali-metal atoms [@EXP] and more recently in the atomic hydrogen [@MIT] has stimulated a new interest in the theoretical study of the inhomogeneous Bose gas. Thermodynamic properties such as critical temperature, condensate fraction have been investigated by several authors for the trapped Bose gas recently [@RMP]. In the presence of an anisotropic harmonic potential of the form ${V_{ext}(% \vec{r})=m(\omega _{x}^{2}x^{2}+\omega _{y}^{2}y^{2}+\omega _{z}^{2}z^{2})/2}$ the noninteracting model gives the critical temperature ${T_{c}^{0}=\frac{% \hbar \omega _{ho}}{k_{B}}\left( \frac{N}{\zeta (3)}\right) ^{1/3}}$ [@RMP], where $\omega _{ho}=\left( \omega _{x}\omega _{y}\omega _{z}\right) ^{1/3}$ is the geometrical average of the oscillator frequencies. $\zeta (n)$ is Riemann $\zeta$ function. In the large-N limit, the condensate fraction is $\xi =\frac{N_{0}}{N}=$ ${1-\frac{\zeta (3)}{N}\left( \frac{k_{B}T}{\hbar \omega _{ho}}\right) ^{3}}$. Finite-size effects [@FIN] and interatomic interaction will give corrections to the thermodynamic properties of the system. The correction to the critical temperature due to finite-size effects has shown to be $\delta {T_{c}^{0}/T_{c}^{0}\simeq -0.73}\overline{\omega }N^{-1/3}/\omega_{ho}$, where $\overline{\omega }% =(\omega _{x}+\omega _{y}+\omega _{z})/3$ is the mean frequency. Although the atom clouds realized in the present experiments are very dilute, the effects due to interatomic interaction are important at low temperature. Researches show that atom-atom interaction will give leading corrections when $N$ is larger than $10^{5}$ for the alkalis. In fact the question of how two-body forces affect the thermodynamic properties of these systems have been the object of several theoretical investigations [@BEF]. Using a local density approximation Giorgini, Pitaevskii, and Stringari [@RMP; @GIO] obtained the shift of the critical temperature due to interatomic interaction: $\delta {T_{c}^{0}/T_{c}^{0}\simeq -1.33}% aN^{1/6}/a_{ho}$, where $a_{ho}=\sqrt{\hbar /m\omega _{ho}}$ is the harmonic oscillator length. Recently Monte Carlo simulation [@MCS] is also used to investigate the thermodynamic properties of the trapped interacting Bose gases. In this work, we investigate the critical temperature and the condensate fraction of the system when both atom-atom interaction and finite-size effects are taken into account. Canonical ensemble is used to obtain the analytical high-order correction due to interatomic interaction. Especially, finite-size effects are obtained simultaneously. To obtain the shift of the critical temperature, we give the analytic description of the condensation fraction near the critical temperature. The paper is planed as follows. In Sec.II we outline the canonical ensemble. In Sec.III we give the condensate fraction of the system near the critical temperature. The shift of the critical temperature agrees with the usual results [@RMP; @FIN; @GIO]. In Sec.IV we obtain the analytical high-order correction to the condensate fraction due to atom-atom interaction. The lowest-order correction agrees with the well-established results [@RMP; @GIO; @NAR]. In Sec.V the correction to the condensate fraction beyond mean field theory is given. Partition Function of the Trapped Interacting Bose Gases ======================================================== Canonical ensemble and saddle-point method have been used to investigate the thermodynamic properties of the interacting uniform Bose gases [@PAT]. In this work canonical ensemble is used to discuss the trapped interacting Bose gases. The partition function of $N$ trapped interacting bosons is given by $${Q(N)=\Sigma _{n}\exp \left( -\beta E_{n}\right) =\Sigma _{\left\{ n_{x},n_{y},n_{z}\right\} }\exp \left[ -\beta \left\{ \Sigma _{n_{x},n_{y},n_{z}}N_{n_{x}n_{y}n_{z}}\varepsilon _{n_{x}n_{y}n_{z}}+E_{int}\right\} \right] }, \label{par1}$$ where $N_{n_{x}n_{y}n_{z}}$ and $\varepsilon _{n_{x}n_{y}n_{z}}$ are the occupation numbers and energy level of the state $\{n_{x},n_{y},n_{z}\}$ respectively. $E_{int}$ is the interaction energy of the system. From (\[par1\]) $$Q(N){=\Sigma _{N_{0}=0}^{N}}\left\{ {\exp }\left[ -\beta \left( E_{0}+E_{int}\right) \right] {Q}_{0}\left( N-N_{0}\right) \right\}, \label{par2}$$ where ${Q}_{0}\left( N-N_{0}\right) ={\Sigma _{\{n_{x},n_{y},n_{z}\}}^{\prime }\exp \left[ -\beta \left( \Sigma _{n_{x},n_{y},n_{z}}N_{n_{x}n_{y}n_{z}}\varepsilon _{n_{x}n_{y}n_{z}}\right) \right] }$ stands for the partition function of a fictitious system of $N-N_{0}$ trapped noninteracting atoms. $E_{0}$ is the energy of the condensate. For convenience, we have separated out the ground state $n_{x}=n_{y}=n_{z}=0$ from the states $\left\{n_{x},n_{y},n_{z}\right\} \neq 0$, so that we first carry out the primed summation over all $\left\{ n_{x},n_{y},n_{z}\right\} $ ($\left\{ n_{x},n_{y},n_{z}\right\} \neq 0$) with a fixed value $(N-N_{0})$ of the partial sum $\Sigma _{\left\{ n_{x}n_{y}n_{z}\right\} }^{\prime }$ and then carry out a summation over all possible values of $N_{0}$, [*ie.*]{} $N_{0}=0$ to $N_{0}=N$. Assume $A_{0}\left( N-N_{0}\right) $ is the free energy of the fictitious system. $A_{0}(N-N_{0})=-\frac{1}{\beta }\ln Q_{0}(N-N_{0})$. From (\[par2\]) $$Q(N){=\Sigma _{N_{0}=0}^{N}\exp }\left[ -\beta \left( E_{0}+E_{int}\right) -\beta A_{0}(N-N_{0})\right]. \label{par3}$$ The sum ${\Sigma _{N_{0}=0}^{N}}$ in (\[par3\]) maybe replaced by the largest term in the sum, for the error omitted in doing so will be statistically negligible. With this approximation we do not investigate the fluctuations of the system. However, this approximation is reasonable because we can give the occupation number of the energy level $\varepsilon _{n_{x}n_{y}n_{z}}$, which agrees with the widely used mean occupation number in the frame of grand-canonical ensemble. Neglecting all terms but the largest in (\[par3\]), the number of bosons in the condensate can be obtained as: $$-\beta \frac{\partial }{\partial N_{0}}\left( E_{0}+E_{int}\right) -\beta \frac{\partial }{\partial N_{0}}A_{0}(N-N_{0})=0. \label{eq4}$$ The calculations of the free energy $A_{0}(N-N_{0})$ of the fictitious noninteracting Bose gas is nontrivial because there is a requirement that the number of the particles is $N-N_{0}$ in the summation of the partition function. Using the saddle-point method of integration developed by Darwin and Fowler [@DAR] it is straightforward to find that $-\beta \frac{\partial }{\partial N_{0}}A_{0}(N-N_{0})=\ln z_{0}$ in which $z_{0}$ is the fugacity of the fictitious Bose gas. In addition, the fugacity $z_{0}$ is determined by $${N-N}_{0}{=\Sigma _{n_{x},n_{y},n_{z}}^{\prime}\frac{1}{\exp \left[ \beta \varepsilon _{n_{x}n_{y}n_{z}}\right] z_{0}^{-1}-1}}. \label{main1}$$ We can easily understand (\[main1\]) in terms of Bose-Einstein distribution of the trapped noninteracting Bose gas. Using the relation $-\beta \frac{\partial }{\partial N_{0}}A_{0}(N-N_{0})=\ln z_{0}$, (\[eq4\]) becomes $$-\beta \frac{\partial }{\partial N_{0}}\left( E_{0}+E_{int}\right) +lnz_{0}=0. \label{main2}$$ Equations (\[main1\]) and (\[main2\]) will be used to discuss the corrections due to atom-atom interaction and finite-size effects. Once we know the interaction energy of the system, it is easy to obtain the correction due to atom-atom interaction. In addition, finite-size effects are separated off in (\[main1\]). Omitting interactions between atoms, we can obtain $\ln z=\beta \epsilon _{000}$ from (\[main2\]). From (\[main1\]) the number of the thermal atoms $N_{T}$ is given by $$N_{T}={N-N}_{0}{=\Sigma _{n_{x},n_{y},n_{z}}^{\prime}}\frac{1}{\exp \left[\beta (\varepsilon _{n_{x}n_{y}n_{z}}-\epsilon _{000})\right] -1}. \label{eq7}$$ This is the exact conclusion in the frame of grand-canonical ensemble for the trapped noninteracting Bose gas. This shows the equivalence between the grand-canonical ensemble and canonical ensemble in the calculations of the condensate fraction. In addition, it shows that neglecting all terms but the largest in (\[par3\]) is reasonable. In the absence of interaction, from (\[eq7\]), the condensate fraction is given by [@FIN] $$\xi =1-(\frac{T}{T_{c}^{0}})^{3}-\frac{3\overline{\omega }\zeta \left( 2\right) }{2\omega _{ho}\zeta \left( 3\right) ^{2/3}}(\frac{T}{T_{c}^{0}})^{2}N^{-1/3}. \label{eq8}$$ The third term in (\[eq8\]) is the finite-size correction to the ideal Bose gas in the thermodynamic limit. In short, (\[main1\]) and (\[main2\]) account for both interatomic interaction and finite-size effects. We will use (\[main1\]) and (\[main2\]) to calculate the critical temperature and condensate fraction of the system. critical temperature and condensate fraction near $T_{c}$ ========================================================= interaction energy of the system near the critical temperature -------------------------------------------------------------- The parameter expressing the importance of the interatomic interaction compared to the kinetic energy is $\frac{E_{int}}{E_{kin}} \propto \frac{N_{0}|a|}{a_{ho}}$, where $a$ is the scattering length between bosons. Near the critical temperature $\frac{N_{0}|a|}{a_{ho}}<<1$. This means that $E_{kin}>>E_{int}$. In this case we can use the method of pseudopotentials developed by Huang and Yang [@YAN] to calculate the interaction energy of the system. In the method of pseudopotentials, the actual Hamiltonian of the system is replaced by an effective Hamiltonian, such that the ground state and the low-lying energy levels of the system are given equally well by the new Hamiltonian. With hard-sphere approximation, the boundary conditions between atoms are replaced by the pseudopotential operator $g\delta ^{(3)}(\vec{r})\frac{\partial }{\partial r}r$ in the effective Hamiltonian, where $g=\frac{4\pi a\hbar^{2}}{m}$. The effective Hamiltonian of the system may be taken to be $$\widehat{H}=-\frac{\hbar ^{2}}{2m}\left( \nabla _{1}^{2}+\cdots +\nabla _{N}^{2}\right) +\Sigma _{i=1}^{N}V_{ext}(\vec{r}_{i})+\Sigma _{i<j}\omega _{ij}, \label{eff}$$ where $\omega _{ij}=g\Sigma _{i<j}\delta ^{(3)}\left( \vec{r}_{i}-\vec{r}_{j}\right) \frac{\partial }{\partial r_{ij}}% r_{ij}$. For a trapped dilute Bose gas near the critical temperature the pseudopotentials can be regarded as perturbation terms. The energy levels to the first order in $a$ may be obtained through the usual perturbation theory. In the general case of a system containing an arbitrary number $N$ of bosons, the unperturbed normalized wave function of the system is given by $$\Psi _{N}=\left( \frac{N_{\alpha _{1}}!N_{\alpha _{2}}!\cdots }{N!}\right) ^{1/2}\Sigma_{P}\lbrack \phi _{\alpha _{1}}\phi _{\alpha _{2}}\cdots \phi _{\alpha _{N}}\rbrack,$$ where the sum is taken over all permutations of the different suffixes $% \alpha _{1}$, $\alpha _{2}$, $\cdots $, $\alpha _{N}$ and the numbers $% N_{\alpha _{i}}$ show how many of these suffixes have the same value $\alpha _{i}$ (with $\Sigma N_{\alpha _{i}}=N$). From the first-order perturbation theory the interaction energy $E_{int}$ of the system is given by $$E_{int}=<\Psi _{N}|\Sigma _{i<j}\omega _{ij}|\Psi _{N}>.$$ Thus, the operators $\left( \partial /\partial r_{ij}\right) r_{ij}$ will be operating on a set of functions which are well behaved for all values of $% r_{ij}$; accordingly, these operators may be replaced by the unit operators. Basing on the investigation of the various permutations [@PAT] of the single particle states the interaction energy takes the form, $$E_{int}=\frac{g( N_{0}^{2}-N_{0})}{2}\int \phi _{0}^{4}(\vec{r})d^{3}\vec{r}+2gN_{0}\int \phi _{0}^{2}(\vec{r})n_{T}\left( \vec{r}\right) d^{3}\vec{r} +g\int n_{T}^{2}(\vec{r})d^{3}\vec{r}. \label{eint}$$ To obtain (\[eint\]) we used the fact that $n_{T}(\vec{r})=\Sigma _{l\neq 0}N_{l}\phi _{l}^{2}(\vec{r})$ is the density distribution of the thermal atoms. Near the critical temperature, the density distribution of the condensate is $n_{0}\left( \vec{r}\right) =N_{0}\phi _{0}^{2}\left( \vec{r}\right) =N_{0}\left( \frac{m\omega _{ho}}{\pi \hbar }\right) ^{3/2}exp\lbrack -\frac{m}{\hbar }\left( \omega _{x}x^{2}+\omega _{y}y^{2}+\omega _{z}z^{2}\right) \rbrack $. The density distribution of the thermal atoms can be obtained using Bose-Einstein distribution and the semiclassical approximation of the energy level of a single particle. Near the critical temperature $\frac{k_{B}T}{\hbar\omega_{ho}} \propto N^{1/3}$. In the available traps $N$ ranges from a few thousand to several millions, thus $\hbar \omega _{ho}<<k_{B}T_{c}^{0}$, [*ie.*]{} the energy level of the thermal atoms can be approximated as continuous. In addition, this means that the semiclassical approximation for the normal gas is expected to work well on a wide range of temperatures [@RMP]. In terms of Bose-Einstein distribution $n_{T}\left( \vec{r}\right) =\int d^{3}\vec{p}\left( 2\pi \hbar \right) ^{-3}\left[ \exp \left( \beta \varepsilon \left( \vec{p},\vec{r}\right) \right) -1\right] ^{-1}$, where $\varepsilon (\vec{p},\vec{r})=\frac{\vec{p}^{2}}{2m}+V_{ext}\left( \vec{r}\right) $ is the semiclassical energy in phase space and $\beta =1/k_{B}T$. Density distribution of the thermal atoms in the coordinate space would be $n_{T}(\vec{r})=\lambda _{T}^{-3}g_{3/2}\left( \exp \lbrack -V_{ext}(\vec{r}% )/(k_{B}T)\rbrack \right)$, where $\lambda _{T}=\left[ 2\pi \hbar ^{2}/\left( mk_{B}T\right) \right] ^{1/2}$ is the thermal wavelength. $g_{3/2}\left( z\right) $ belongs to the class of functions $g_{\alpha }(z)=\Sigma _{n=1}^{\infty }z^{n}/n^{\alpha }$. By integrating the semiclassical approximation $n_{T}(\vec r)$ over space one obtain the condensate fraction $\xi =1-\left( \frac{T}{T_{c}^{0}}\right) ^{3}$, which does not account for finite-size effects. However, in this paper the semiclassical approximation $n_{T}(\vec r)$ is used to calculate the interaction energy of the system, [*ie.*]{} used to calculate the high-order correction due to interatomic interaction. Thus the adoption of the semiclassical approximation will omit only higher-order modification, which is of the order of the multiplication of the corrections due to the interatomic interaction and finite-size effects. Because the finite-size effects have been separated off in (\[main1\]), we can give finite-size effects although semiclassical approximation is used to calculate the correction due to interatomic interaction. Using the geometrical average of the oscillator frequencies $\omega _{ho}$, $n_{T}\left( r\right) =\lambda _{T}^{-3}g_{3/2}\left( \exp \left[ -m\omega _{ho}^{2}r^{2}/2k_{B}T\right] \right) $. Introducing the width of the normal gas $R_{T}=\sqrt{2k_{B}T/m\omega _{ho}^{2}}$ and defining the rescaled variable through $r=R_{T}\overline{r}$, we can obtain $n_{T}\left( \overline{r}\right) =\lambda _{T}^{-3}g_{3/2}\left( \exp \left[ -\overline{r}^{2}\right] \right) $. Near the critical temperature $k_{B}T>>\hbar \omega _{ho}$, the classical Boltzmann distribution $n_{cl}\left( \overline{r}\right) \propto \lambda _{T}^{-3}\exp \left[ -\overline{r}^{2}\right] $ gives a zero-order approximation for the density distribution of the thermal cloud. Thus we take the trial function of the form $$n_{T}^{G}\left( \overline{r}\right) =\eta _{1}\lambda _{T}^{-3}\exp \left[ -\eta _{2}\overline{r}^{2}\right] \label {normal}$$ where $\eta _{1}$ and $\eta _{2}$ are two dimensionless variational parameters. Another reason for the adoption of the Gaussian trial function lies in the fact that there is a factor $\exp \left[ -\overline{r}^{2}\right] $ in $n_{T}\left( \overline{r}\right) $. $\eta _{1}$ and $\eta _{2}$ are determined by the requirement that $\int |n_{T}^{G}\left( \vec{r}\right) -n_{T}\left( \vec{r}\right) |d^{3}\vec{r}$ obtain the minimum. Results of numerical calculations show that $\eta _{1}=2.587$, $\eta _{2}=1.146$ and $\int |n_{T}^{G}\left( \vec{r}\right) -n_{T}\left( \vec{r}\right) |d^{3}\vec{r}/\int n_{T}\left( \vec{r}\right) d^{3}\vec{r}=0.0017$. So the Gaussian distribution $n_{T}^{G}(\vec{r})$ agrees very well with $n_{T}(\vec{r})$. We should note that the Gaussian distribution $n_{T}^{G}\left( \vec{r}\right) $ is different from the classical Boltzmann distribution because of the variational parameters $\eta _{1}$ and $\eta _{2}$. Combining (\[eint\]) and (\[normal\]) we have $$E_{int}=\frac{g}{2\left( \sqrt{2\pi }a_{ho}\right) ^{3}}N_{0}^{2}+ \left( 2-\frac{1}{2^{3/2}}\right) \eta _{1}g\lambda _{T}^{-3}N_{0}+\frac{\eta _{1}g\lambda _{T}^{-3}}{2^{3/2}}N, \label{energy}$$ where we have used the fact that $a_{ho}>>\lambda_{T}$ and $k_{B}T>>\hbar \overline{\omega }$ near the critical temperature. shift of the critical temperature and condensate fraction near $T_{c}$ ---------------------------------------------------------------------- We introduce a scaling parameter which accounts for the role of the two-body repulsive interaction. This parameter $\theta$ is fixed by the ratio between $gn_{T}\left( \vec{r}=0,T_{c}^{0}\right) $ and the critical temperature $k_{B}T_{c}^{0}$. $\theta =\frac{gn_{T}\left( \vec{r}=0,T_{c}^{0}\right) }{k_{B}T_{c}^{0}}=\frac{\eta _{1}\sqrt{2/\pi }}{\zeta \left( 3\right) ^{1/6}}\frac{a}{a_{ho}}N^{1/6}$. The scaling parameter $\theta$ can also be written in the form $\theta =0.65\eta ^{5/2}$, where $\eta =\mu \left( T=0\right) /k_{B}T_{c}^{0} =\frac{\zeta \left( 3\right) ^{1/2}15^{2/5}}{2}\left( N^{1/6}\frac{a}{a_{ho}}\right) ^{2/5}$ is also an important scaling parameter accounting for the role of interatomic interaction [@RMP; @GOS]. Obviously $\theta$ stands for high-order correction due to interatomic interaction, compared to the parameter $\eta$. From (\[main1\]), (\[main2\]), and (\[energy\]) we can obtain the equation on the condensate fraction. $$1-\xi =\frac{t^{3}}{\zeta \left( 3\right) }g_{3}\left( z_{0}\right) +\frac{3\zeta \left( 2\right) }{2\zeta \left( 3\right) ^{2/3}}\frac{\overline{\omega }}{\omega _{ho}}t^{2}N^{-1/3}\label{near1},$$ $$-\frac{\zeta \left( 3\right) ^{1/2}\theta N^{1/2}}{\eta _{1}t}\xi -\left( 2-\frac{1}{2^{3/2}}\right) \theta t^{1/2}+\ln z_{0}=0, \label {near2}$$ where we have used the reduced temperature $t=T/T_{c}^{0}$. From (\[near1\]) and (\[near2\]) the condensate fraction is given by $$\xi =\frac{1-t^{3}-4.51\frac{a}{a_{ho}}N^{1/6}t^{7/2}-\frac{3\zeta \left( 2\right) }{2\zeta \left( 3\right) ^{2/3}}\frac{\overline{\omega }}{\omega _{ho}}t^{2}N^{-1/3}}{1+1.16\frac{a}{a_{ho}}N^{2/3}t^{2}}. \label{conden}$$ The third term in the numerator of (\[conden\]) represents the correction due to the interaction between atoms, while the last term in the numerator accounts for the correction due to the finite-size effects. By setting $\xi =0$ in (\[conden\]) one can obtain the shift of the critical temperature, $$\frac{\delta T_{c}^{0}}{T_{c}^{0}}=\frac{\delta T_{int}}{T_{c}^{0}}+\frac{\delta T_{finite}}{T_{c}^{0}}=-1.50\frac{a}{a_{ho}}N^{1/6}-\frac{\zeta \left( 2\right) }{2\zeta \left( 3\right) ^{2/3}}\frac{\overline{\omega }}{\omega _{ho}}N^{-1/3}. \label{shift}$$ The first term in (\[shift\]) is the shift due to interatomic interaction. It agrees with the results based on the local density approximation [@RMP; @GIO]. Because of the denominator in (\[conden\]), close to $T_{c}^{0}$ the condensate fraction will increase slowly with decreasing the temperature. Especially, it shows that there is only high-order correction of the scaling parameter $\eta$ due to interatomic interaction. We should note that (\[conden\]) holds only when $\frac{N_{0}a}{a_{ho}}<<1$. The second contribution in (\[shift\]) gives exactly the usual results due to the finite-size effects. condensate fraction of the system below $T_{c}$ =============================================== lowest-order correction due to interatomic interaction and finite-size effects ------------------------------------------------------------------------------ Below the critical temperature $\frac{N_{0}a}{a_{ho}}>>1$, $\it ie.$ $E_{int}>>E_{kin}$. In this case we can use the well-known Thomas-Fermi approximation. With the Thomas-Fermi approximation the density profile of the condensate is $n_{0}\left( \vec{r}\right) =\frac{\mu -V_{ext}\left( \vec{r}\right) }{g}\theta \left( \mu -V_{ext}\left( \vec{r}\right) \right) $. The normalization condition on $n_{0}(\vec r)$ provides the relation between chemical potential and number of particles in the condensate: $\mu =\frac{\hbar \omega _{ho}}{2}\left( \frac{15N_{0}a}{a_{ho}}\right) ^{2/5}$. From Gross-Pitaevskii (GP) equation [@GPE] and the well-known Virial theorem the energy of the condensate turns out to be $E_{0}=\left( 5/7\right) \mu N_{0}$ [@RMP]. Omitting atom-atom interactions in the normal gas and the interaction between the condensate and normal gas, from (\[main1\]) and (\[main2\]) $$N_{0}=N-N_{T}=N-\Sigma_{n_{x},n_{y},n_{z}}^{\prime}\frac{1}{\exp \left[ \beta \left( \varepsilon _{n_{x}n_{y}n_{z}}-\mu \right) \right] -1}. \label{low1}$$ From (\[low1\]) it is easy to obtain the following result $$\xi =1-t^{3}-\frac{\zeta \left( 2\right) }{\zeta \left( 3\right) }\frac{\mu }{k_{B}T}-\frac{3\overline{\omega }\zeta \left( 2\right) }{2\omega _{ho}\zeta \left( 3\right) ^{2/3}}t^{2}N^{-1/3}\label{low2},$$ where the last term in (\[low2\]) is the correction due to finite-size effects [@FIN]. Using the relation $\mu /k_{B}T=\eta \xi ^{2/5}/t$, the third term in (\[low2\]) gives exactly the usual lowest-order modification due to atom-atom interactions [@RMP; @GIO; @NAR]. $$\xi =1-t^{3}-\frac{\zeta \left( 2\right) }{\zeta \left( 3\right) }\eta \xi ^{2/5}t^{2}-\frac{3\overline{\omega }\zeta \left( 2\right) }{2\omega _{ho}\zeta \left( 3\right) ^{2/3}}t^{2}N^{-1/3}. \label{lowc}$$ This proves the equivalence between canonical ensemble and grand-canonical ensemble, even in the presence of interatomic interactions. We will give high-order modification in the following. high-order modification due to interatomic interactions ------------------------------------------------------- Equation (\[lowc\]) is obtained by omitting the interaction between thermal atoms and interaction between thermal atoms and the condensate. When these two sorts of interactions are considered $$E_{0}+E_{int}=\frac{5}{7}\mu N_{0}+2g\int n_{0}\left( \vec{r}\right) n_{T}\left( \vec{r}\right) d^{3}\vec{r}+g\int n_{T}^{2}\left( \vec{r}\right) d^{3}\vec{r}. \label{high}$$ The calculations of (\[high\]) are straightforward. The results is given by $$E_{0}+E_{int}=\frac{5\mu N_{0}}{7}+\frac{16\pi \eta _{1}\mu }{15}\frac{R_{\mu }^{3}}{\lambda _{T}^{3}}-\frac{16\pi \eta _{1}\eta _{2}}{35}\frac{R_{\mu }^{3}}{\lambda _{T}^{3}}\frac{\mu ^{2}}{k_{B}T}+\frac{\eta _{1}g\lambda _{T}^{-3}}{2^{3/2}}\left( N-N_{0}\right), \label{high1}$$ where $R_{\mu }=\sqrt{2\mu /m\omega _{ho}^{2}}$ is the radius of the condensate. $R_{\mu }^{3}/\lambda _{T}^{3}=\alpha _{R\lambda }N^{1/2}t^{3/2}N_{0}^{3/5}$, where $\alpha _{R\lambda }=0.294\left( a/a_{ho}\right) ^{3/5}$. From (\[high1\]) $$-\beta \frac{\partial }{\partial N_{0}}\left( E_{0}+E_{int}\right) =-\left( 1-0.89\eta ^{5/2}t^{1/2}\right) \frac{\mu }{k_{B}T}-1.01\eta ^{5/2}t^{1/2}. \label{high2}$$ From (\[main1\]), (\[main2\]), and (\[high2\]) we can obtain the condensate fraction below the critical temperature, $$\xi =1-t^{3}-\frac{\zeta \left( 2\right) }{\zeta \left( 3\right) }t^{3}\left[ \left( 1-0.89\eta ^{5/2}t^{1/2}\right)\frac{\eta \xi ^{2/5}}{t}+1.01\eta ^{5/2}t^{1/2}\right] -\frac{3\overline{\omega }\zeta \left( 2\right) }{2\omega _{ho}\zeta \left( 3\right) ^{2/3}}t^{2}N^{-1/3}. \label{highc}$$ In (\[highc\]) the terms comprises $\eta ^{5/2}$ represents high-order correction to the condensate fraction due to interatomic interaction. Omitting the terms comprises $\eta ^{5/2}$, we can obtain the lowest-order correction (\[lowc\]). Essentially, (\[lowc\]) and (\[highc\]) are transcendental equations on the condensate fraction. We can easily give the numerical solution of the condensate fraction. In Fig.1 and Fig.2 we used the experimental parameter by Ensher [*et al.*]{} [@ENS], where the cloud consists of $4\times 10^{4}$ atoms at the transition and $a/a_{ho}=5.4\times 10^{-3}$. According to Fig.2 the high-order correction (\[highc\]) agrees well with the condensate fraction (\[conden\]) in the region $\frac{N_{0}a}{a_{ho}}<<1$. From (\[conden\]) and (\[highc\]), close to $T_{c}^{0}$ the condensate fraction increase quite slowly with decreasing the temperature. However, the lowest-order correction (\[lowc\]) is rather different from the dashed line near the critical temperature. The difference between (\[conden\]) and (\[lowc\]) lies in the fact that there is only high-order correction due to interatomic interaction near the critical temperature, while there is only lowest-order correction in (\[lowc\]). Fig.2 demonstrates the validity of (\[highc\]) over the whole range of temperature below $T_{c}$. In the preceding calculations we neglects the role of the interaction between the condensate and the thermal atoms on the density distributions of the condensate and the thermal atoms. In the frame of Hartree-Fork model [@RMP; @NAR], the densities of the condensate and the thermal component are given as follows. $$n_{0}\left( \vec{r}\right) =\frac{\mu -V_{ext}\left( \vec{r}\right) -2gn_{T}\left( \vec{r}\right) }{g}\theta \left( \mu -V_{ext}\left( \vec{r}\right) -2gn_{T}\left( \vec{r}\right) \right), \label{density1}$$ $$n_{T}\left( \vec{r}\right) =\lambda _{T}^{-3}g_{3/2}\left[ e^{-\left( v_{ext}\left( \vec{r}\right) +2g\left( n_{0}\left( \vec{r}\right) +n_{T}\left( \vec{r}\right) \right) -\mu \right) /k_{B}T}\right]. \label{density2}$$ Using (\[main1\]), (\[main2\]), (\[high\]) and (\[density1\]), (\[density2\]) and the usual iterative procedure we can obtain the numerical conclusion of the condensate fraction. The numerical conclusion is illustrated in Fig.3. condensate fraction beyond mean-field theory ============================================ In the preceding calculations of the condensate fraction, GP equation is used to obtain the energy $E_{0}$ of the condensate. GP equation is expected to be valid if the system is dilute, [*ie.*]{} $n|a|^{3}<<1$. Obviously, correction to GP equation would modify the condensate fraction of the system. The first correction to the mean-field approximation have been investigated by Timmermans, Tommasini, and Huang [@TIM] and by Braaten and Nieto [@BRA]. For large $N_{0}$, using the local density approximation, the density distribution of the condensate is given by $$n_{0}\left( \vec{r}\right) =\frac{\mu -V_{ext}\left( \vec{r}\right) }{g}-\frac{4m^{3/2}}{3\pi ^{2}\hbar ^{2}}\left[ \mu -V_{ext}\left( \vec{r}\right) \right] ^{3/2}, \label{GP1}$$ with $\mu$ given by $$\mu =\frac{\hbar \omega _{ho}}{2}\left( \frac{15N_{0}a}{a_{ho}}\right) ^{2/5}\left( 1+\sqrt{\pi a^{3}n\left( 0\right) }\right). \label{GP2}$$ The parameter $a^{3}n\left( 0\right) $ can be directly expressed in terms of the relevant parameters of the system. $a^{3}n\left( 0\right) =\frac{15^{2/5}}{8\pi }\left( N_{0}^{1/6}\frac{a}{a_{ho}}\right) ^{12/5}=0.0078\eta ^{6}\xi ^{2/5}$. Using the relation $E_{0}=5N_{0}\mu /7$ and (\[main1\]), (\[main2\]) we can obtain the correction to the condensate fraction below $T_{c}$ $${\xi =1-t^{3}}$$ $$-\frac{\zeta \left( 2\right) }{\zeta \left( 3\right) }t^{3}\left[ \left( 1-0.89\eta ^{5/2}t^{1/2}+0.16\eta ^{3}\xi ^{1/5}\right) \frac{\eta \xi ^{2/5}}{t}+1.01\eta ^{5/2}t^{1/2}\right] -\frac{3\overline{\omega }\zeta \left( 2\right) }{2\omega _{ho}\zeta \left( 3\right) ^{2/3}}t^{2}N^{-1/3}.\label{GP3}$$ The term $0.16\eta ^{3}\xi ^{1/5}$ is the correction due to the modification of the GP equation. It gives higher-order correction of the scaling parameter $\eta$, compared to the high-order correction due to interatomic interactions. Another correction to the condensate fraction beyond mean-field approximation is given by the quantum depletion of the condensate. There are atoms which do not occupy the condensate at zero temperature because of quantum depletion effects. We can use local density approximation to write the density of atoms out of the condensate. Timmermans, Tommasini, and Huang [@TIM] gives $n_{out}\left( \vec{r}\right) =\left( 8/3\right) \left[ n\left( \vec{r}\right) a^{3}/\pi \right] ^{1/2}$. Integration of $n_{out}\left( \vec{r}\right) $ yields the result $$\xi _{out}=\int n_{out}\left( \vec{r}\right) d^{3}\vec{r}=\frac{5\sqrt{\pi }}{8}\sqrt{a^{3}n\left( 0\right)}. \label{GP4}$$ Combining (\[GP3\]) and (\[GP4\]) $$\xi =1-t^{3}-\frac{\zeta \left( 2\right) }{\zeta \left( 3\right) }t^{3}\left[ \left( 1-0.89\eta ^{5/2}t^{1/2}+0.16\eta ^{3}\xi ^{1/5}\right) \frac{\eta \xi ^{2/5}}{t}+1.01\eta ^{5/2}t^{1/2}\right]$$ $$-0.11\eta ^{3}\xi ^{1/5}-\frac{3\overline{\omega }\zeta \left( 2\right) }{2\omega _{ho}\zeta \left( 3\right) ^{2/3}}t^{2}N^{-1/3}. \label{BEY}$$ The term $-0.11\eta ^{3}\xi ^{1/5}$ is the correction due to quantum depletion. Compared to the finite-size effects, the corrections to the mean-field approximation have a different dependence on the parameter $N$ and $a/a_{ho}$. The corrections beyond mean field theory become larger than finite-size effects when $N$ is larger than about $10^{6}$. In Fig.4 we give numerical conclusion of (\[BEY\]). The corrections beyond mean field theory become important when the temperature is much lower than the critical temperature. conclusion ========== To conclude, the canonical ensemble is used to obtain the equations on the condensate fraction (\[main1\]) and (\[main2\]), which account for interaction effects. Finite-size effects can be obtained simultaneously because these effects are comprised in (\[main1\]). (\[main1\]) and (\[main2\]) are used to investigate the condensate fraction near and below the critical temperature. From the condensate fraction near the critical temperature, we obtain the shift of the critical temperature due to atom-atom interaction and finite-size effects simultaneously. In addition, the analytical high-order correction due to interatomic interaction is obtained in this work. Acknowledgments {#acknowledgments .unnumbered} =============== This work was supported by the Science Foundation of Zhijiang College, Zhejiang University of Technology. This work was also supported by the National Natural Science Foundation of China (19975019). Anderson M H, Ensher J R, Matthews M R, Wieman C E and Cornell E A 1995 [*Science*]{} [**269**]{} 198\ Davis K B, Mewes M O, Andrews M R, van Druten N J, Durfee D S, Kurn D M and Ketterle W 1995 [*Phys. Rev. Lett.*]{} [**75**]{} 3969\ Bradley C C, Sackett C A, Tollett J J and Hulet R G 1995 [*Phys. Rev. Lett.*]{} [**75**]{} 1687 Killian T C, Fried D G, Willmann L, Landhuis D, Moss S C, Greytak T J and Kleppner D 1998 [*Phys. Rev. Lett.*]{} [**81**]{} 3807\ Fried D G, Killian T C, Willmann L, Landhuis D, Moss S C, Kleppner D and Greytak T J 1998 [*Phys. Rev. Lett.*]{} [**81**]{} 3811 Dalfovo F, Giorgini S, Pitaevskii L P and Stringari S 1999 [*Rev. Mod. Phys.*]{} [**71**]{} 463 and references therein Grossmann S, Holthaus M 1995 [*Phys. Lett. A*]{} [**208**]{} 188\ Ketterle W and van Druten N J 1996 [*Phys. Rev. A*]{} [**54**]{} 656\ Kirsten K, Toms D J 1996 [*Phys. Rev. A*]{} [**54**]{} 4188\ Haugset T, Haugerud H and Andersen J O 1997 [*Phys. Rev. A*]{} [**55**]{} 2922 Goldman V V, Silvera I F and Leggett A J 1981 [*Phys. Rev. B*]{} [**24**]{} 2870\ Huse D A and Siggia E D 1982 [*J. Low Temp. Phys.*]{} [**46**]{} 137\ Bagnato V, Pritchard D E and Kleppnet D 1987 [*Phys. Rev. A*]{} [**35**]{} 4354\ Oliva J 1989 [*Phys. Rev. B*]{} [**39**]{} 4197\ Chou T T, Yang C N and Yu L H 1996 [*Phys. Rev. A*]{} [**53**]{} 4257 Giorgini S, Pitaevskii L P and Stringari S 1996 [*Phys. Rev. A*]{} [**54**]{} R4633 Krauth W 1996 [*Phys. Rev. Lett.*]{} [**77**]{} 3695\ Holzmann M, Krauth W and Naraschewski M 1998 [*e-print*]{} cond-mat/9806201\ Cerboneschi E, Menchini C and Arimondo E 2000 [*Phys. Rev. A*]{} [**62**]{} 013606\ Wu H and Foot C J 1996 [*J. Phys. B*]{} [**29**]{} L 321\ Wu H, Arimondo E and Foot C J 1997 [*Phys. Rev. A*]{} [**56**]{} 560 Naraschewski M, Stamper-Kurn D M 1998 [*e-print*]{} cond-mat/9803216 Pathria R K 1972 [*Statistical Mechanics*]{} (Pergamon Press: New York) ch 10 Darwin C G and Fowler R H 1922 [*Phil. Mag.*]{} [**44**]{} 450, 823; 1922 [*Proc. Combridge Phil. Soc.*]{} [**21**]{} 262 Huang K and Yang C N 1957 [*Phys. Rev.*]{} [**105**]{} 767 Giorgini S, Pitaevskii L P and Stringari S 1997 [*Phys. Rev. Lett.*]{} [**78**]{} 3987 Gross E P 1961 [*Nuovo Cimento*]{} [**20**]{} 454\ Gross E P 1963 [*J. Math. Phys.*]{} [**4**]{} 195\ Pitaevskii L P 1961 [*Zh. Eksp. Teor. Fiz.*]{} [**40**]{} 646 \[1961 [*Sov. Phs. JETP*]{} [**13**]{} 451\] Ensher J R, Jin D S, Matthews M R, Wieman C E and Cornell E A 1996 [*Phys. Rev. Lett.*]{} [**77**]{} 4984 Timmermans E, Tommasini P and Huang K 1997 [*Phys. Rev. A*]{} [**55**]{} 3645 Braaten E and Nieto A 1997 [*Phys. Rev. B*]{} [**56**]{} 14745

ArXiv

--- author: - 'L. Ducci' - 'M. Sasaki' - 'F. Haberl' - 'W. Pietsch' bibliography: - 'lducci\_m83.bib' title: 'X-ray source population study of the starburst galaxy M83 with XMM-*Newton* [^1] [^2] [^3] ' --- [We present the results obtained from the analysis of three XMM-*Newton* observations of M83. The aims of the paper are studying the X-ray source populations in M83 and calculating the X-ray luminosity functions of X-ray binaries for different regions of the galaxy.]{} [We detected 189 sources in the XMM-*Newton* field of view in the energy range of $0.2-12$ keV. We constrained their nature by means of spectral analysis, hardness ratios, studies of the X-ray variability, and cross-correlations with catalogues in X-ray, optical, infrared, and radio wavelengths.]{} [We identified and classified 12 background objects, five foreground stars, two X-ray binaries, one supernova remnant candidate, one super-soft source candidate and one ultra-luminous X-ray source. Among these sources, we classified for the first time three active galactic nuclei (AGN) candidates. We derived X-ray luminosity functions of the X-ray sources in M83 in the $2-10$ keV energy range, within and outside the $D_{25}$ ellipse, correcting the total X-ray luminosity function for incompleteness and subtracting the AGN contribution. The X-ray luminosity function inside the $D_{25}$ ellipse is consistent with that previously observed by *Chandra*. The Kolmogorov-Smirnov test shows that the X-ray luminosity function of the outer disc and the AGN luminosity distribution are uncorrelated with a probability of $\sim 99.3$%. We also found that the X-ray sources detected outside the $D_{25}$ ellipse and the uniform spatial distribution of AGNs are spatially uncorrelated with a significance of 99.5%. We interpret these results as an indication that part of the observed X-ray sources are X-ray binaries in the outer disc of M83.]{} Introduction {#section introduction} ============ {width="16cm"} M83 (NGC5236) is a grand-design barred spiral galaxy (SAB(s)c; @deVaucouleurs92) located at $4.5 \pm 0.3$ Mpc from the Milky Way [@Thim03]. M83 is oriented nearly face-on ($i=24^\circ$; @Rogstad74) and shows a galactic disc spanning $12.9^\prime \times 11.5^\prime$ ($17$ kpc $\times$ $15.2$ kpc; @Wofford11). M83 is experiencing a starburst activity with a present-day star formation rate (SFR) of $3-4$ M$_\odot$ yr$^{-1}$ [@Boissier05] in three regions: the nuclear region (galactocentric distance $d\lesssim 300$ pc; @Harris01), the inner disc ($300$ pc $\lesssim d \lesssim 7.5$ kpc), and the outer disc ($7.5$ kpc$\lesssim d \lesssim 20$ kpc; @Dong08). Ultraviolet (UV) images of M83 obtained with the *Galaxy Evolution Explorer* (GALEX) satellite revealed a population of young stars ($\lesssim 400$ Myr) in the outer disc of M83 [@Thilker05]. Although this would indicate recent star-forming activity, using *Spitzer* and GALEX data, @Dong08 discovered that the star formation in the outer disc started at least 1 Gyr ago. These results are confirmed by the study of AGB stars of @Davidge10. @Bigiel10 compared the HI data from the National Radio Astronomy Observatory (NRAO) Very Large Array (VLA) and far-ultraviolet (FUV) data from GALEX in the outer disc of M83, and discovered that the star formation traced by the FUV emission and HI are spatially correlated out to almost four optical radii. @Bigiel10 also found that the star formation rate in the outer disc ($\sim 0.01$ M$_\odot$ yr$^{-1}$; @Bresolin09) implies that the star formation activity is not completely consuming the HI reservoir, which will be available as fuel for star formation in the inner disc. M83 was observed in the X-ray bands by *Einstein* in 1979-1981 [@Trinchieri85], Ginga in 1987 [@Ohashi90], ROSAT in 1992-1994 (@Ehle98; @Immler99), ASCA in 1994 [@Okada97], and *Chandra* in 2000 (@SoriaWu02; @SoriaWu03, SW03 hereafter). SW03 identified 127 discrete sources near the centre of M83 ($8.3^\prime \times 8.3^\prime$) and resolved for the first time the nuclear region in X-rays. The diffuse X-ray emission of M83 has been studied by @Owen09 with an XMM-*Newton* observation performed in January 27, 2003 (obsid 0110910201). They obtained a good fit to the spectrum assuming a two-temperature thermal model, which is typical of the diffuse emission in normal and starburst galaxies. They also found that the soft X-ray emission mainly overlaps with the inner spiral arm, and shows a strong correlation with the distribution of UV emission, indicative of a correlation between X-ray emission and recent star formation. The recent high star formation activity experienced by the nucleus and the spiral arms of M83 provided an unusually large number of supernova remnants (SNRs). In fact, the optical survey performed at the Cerro Tololo Inter-American Observatory in Chile by @Blair04 identified 71 sources as SNR candidates, the *Hubble* Space Telescope (HST) observations of the nuclear region of M83 [@Dopita10] provided the identification of 60 SNR candidates, and the Magellan I survey 271 SNR candidates [@Blair12]. In a normal galaxy such as M83, X-ray binaries (XRBs) are the most prominent class of X-ray sources. XRBs show X-ray luminosities ranging from $\sim 10^{32}$ erg s$^{-1}$ to the Eddington luminosity, and sometimes they can exceed this limit (see e.g. @White78). They are composed of a compact object (a neutron star or a black hole) and a companion star, which can be a main-sequence, giant, or supergiant star, and in some cases a white dwarf (e.g. @vanParadijs98). The strong X-ray emission is produced by the accretion of matter from the companion star onto the compact object. XRBs are usually divided into two classes: low mass X-ray binaries (LMXBs), and high mass X-ray binaries (HMXBs). The companion stars of LMXBs have masses lower than $\sim 1$ M$_\odot$. The lifetime of an LMXB is determined by the nuclear evolution time-scale of the companion star to $10^8 - 10^9$ yr (e.g. @Tauris06), and their number is correlated to the total stellar mass of a galaxy [@Gilfanov04]. The companion star of LMXBs usually tranfers mass by Roche-lobe overflow, and the compact object accretes from an accretion disc (e.g. @vanParadijs98). The donors in HMXBs have masses $\gtrsim 8$ M$_\odot$, and their typical lifetime does not exceed $10^6 -10^7$ yr. Therefore, the presence of HMXBs in a particular region of a galaxy is associated with a relatively recent star formation event (e.g. @Fabbiano06). The X-ray emission from HMXBs is usually explained with the accretion of a fraction of the stellar wind ejected by the donor star onto the compact object, or through mass transfer via Roche-lobe overflow (see e.g. @Treves88 and references therein). As a first approximation, two standard models are commonly used to describe the X-ray spectra of XRBs in nearby galaxies: an absorbed disc-blackbody model, with temperatures ranging from $\sim 0.5$ to $\sim 1$ keV (e.g. @Makishima86), or an absorbed powerlaw model. X-ray spectra of LMXBs below 10 keV are described by absorbed powerlaw with photon indices $1-3$. HMXBs usually show harder X-ray spectra in the energy range $1-10$ keV, with photon indices $1-2$ and a high intrinsic absorption [@White95]. Within each of these classes, the properties of the X-ray spectra can also depend on the type of the accreting compact object. Accreting black holes can show states of high luminosity (e.g. @Jones77), with very soft spectra, with slopes steeper than those shown by accreting neutron stars (see e.g. @White84). Given the wide variety of spectral shapes shown by XRBs, they can be confused with background AGNs, whose X-ray spectra have roughly a powerlaw shape, with indices ranging from 1.6 to 2.5 (see e.g. @Walter93; @Vignali99; @Turner91). In this paper we report the results obtained from a study of the X-ray source populations of M83, using three XMM-*Newton* observations covering both the inner and outer disc regions. The higher spatial resolution and sensitivity of XMM-*Newton* compared to the previous observations of ROSAT and *Einstein* allowed an increase of the number of detected sources in M83. While the spatial coverage of the *Chandra* observation was limited to a region located at the centre of M83 with a size of $8.3^\prime \times 8.3^\prime$ (the ACIS S3 field of view), the XMM-*Newton* observations allowed us to obtain a complete coverage of M83, and to study also the outer parts of the galaxy, which in total provided us with a more representative sample of X-ray sources in M83. The paper is organised as follows: in Sect. \[section Reduction and Data Analysis\] we describe the data reduction and analysis of XMM-*Newton* observations. In Sect. \[section Astrometrical corrections\] we show the astrometrical corrections that have been applied. In Sect. \[sect. Analysis\] we present the techniques adopted to classify the X-ray sources (X-ray variability, spectral analysis, and hardness ratios). In Sect. \[sect. Source classification\] we describe the properties and classification of the detected sources. In Sect. \[sect. X-ray Luminosity Functions\] we derive the X-ray luminosity functions (XLFs) of X-ray binaries within and outside the $D_{25}$ ellipse, after correcting them for incompleteness and subtracting the AGN contribution, and we discuss our results. We examine in detail the properties of the sources that have been identified and classified in this work in Appendix \[sect. Discussion of classification and identification of the XMM-Newton sources\]. Reduction and data analysis {#section Reduction and Data Analysis} =========================== We analysed the public archival XMM-*Newton* data of M83 (PIs: M. Watson, K.D. Kuntz). Table \[Tab. OBS ID XMM\] lists the three observations that we analysed, one pointing at the centre of the galaxy (obs.1) and two in the south, which covered the outer arms with a young population of stars discovered with GALEX. The data analysis was performed using the XMM-*Newton* Science Analysis System (SAS) 12.0. For each pointing we produced PN, MOS1, and MOS2 event files. We excluded times of high background due to soft proton flares as follows. For each observation and instrument, we created background lightcurves (with sources removed) in the 7$-$15 keV energy band. Good time intervals (GTIs) were determined by selecting count rates lower than 8 cts ks$^{-1}$ arcmin$^{-2}$ and 2.5 cts ks$^{-1}$ arcmin$^{-2}$ for PN and MOS, respectively. For each observation, data were divided into five energy bands: - *R1:* 0.2–0.5 keV; - *R2:* 0.5–1 keV; - *R3:* 1–2 keV; - *R4:* 2–4.5 keV; - *R5:* 4.5–12 keV. For the PN data we used single-pixel events (PATTERN=0) in the first energy band and for the other energy bands single- and double-pixel events (PATTERN$\leq$4) were selected. For the MOS data, single-pixel to quadruple-pixel events (PATTERN$\leq$12) were used for all five bands. We ran the source detection procedure separately for images of each observation, and simultaneously for five energy bands and three instruments with the SAS task [edetect\_chain]{}. The source detection consists of three steps. The first step provides a list of source positions used to create the background maps. We adopted a minimum-detection likelihood[^4] of 7 to obtain this list of sources. After removing the sources, a two-dimensional spline with 20 nodes was fitted to the exposure-corrected image. In the second step the background maps are used to improve the detection sensitivity and hence to create a new source list, assuming a minimum-detection likelihood of 4. In the last step, a maximum-likelihood point-spread function (PSF) fit to the source count distribution is performed simultaneously in all energy bands and each EPIC instrument, from the input list of source positions obtained in the previous step (a description of this algorithm is given by @Cruddace88). For each observation we generated the final source list adopting a lower threshold for the maximum-detection likelihood of 6, which corresponds to a detection probability of $\sim 99.75$%. The source detection gives several parameters for each source, such as the coordinates, count rates, and likelihood of detection (see Table \[Tab. source list\] in the appendix \[sect. catalogue-table\]). As mentioned above, 20 nodes (more than the default 16) for the background spline map were used to better follow the central diffuse emission and to minimise false detections. We removed the remaining false detections due to diffuse emission structures by visual inspection. Fig. \[figure RGB\] shows the combined PN, MOS1, and MOS2 three-colour mosaic image obtained from the three observations. The numbers of the detected sources are overplotted on the image. The red, green, and blue colours represent the $0.2-1$ keV, $1-2$ keV, and $2-4.5$ keV energy bands. Astrometrical corrections {#section Astrometrical corrections} ========================= Corrections between XMM-Newton observations ------------------------------------------- We calculated the RA and Dec offsets of the three XMM-*Newton* observations using position of the sources detected in at least two observations. Sources were considered as detections in at least two different observations if their position was closer than $3\times$ the combined statistical positional errors. We calculated the offsets of observations 2 and 3 with respect to the reference observation 1 as the weighted mean of RA and Dec of all sources, then recalculated all X-ray positions correcting for the shifts relative to the observation 1. Correcting the position of the detected sources using X-ray and optical observations ------------------------------------------------------------------------------------ We also applied the cross-correlation procedure described above to determine the systematic errors in the X-ray positions of the XMM-*Newton* observations by calculating the offsets in the X-ray positions of the XMM-*Newton* sources with respect to the X-ray sources observed by SW03 with *Chandra*. The offset between the XMM-*Newton* and *Chandra* positions (given as the weighted mean of RA and Dec in arcsec) is $\Delta$RA$ = -1.22 \pm 0.16$, $\Delta$Dec$ = -0.72 \pm 0.16$. We point out that SW03 corrected the *Chandra* positions using only the position of the infrared nucleus of M83 deduced from HST/WFPC2 observations. Therefore, to obtain more accurate positions from possible optical counterparts, we cross-correlated the XMM-*Newton* list of sources with the optical catalogue of the United States Naval Observatory USNO-B1 [@Monet03]. For this calculation we were interested in associations between X-ray sources and foreground stars. As discussed in Sect. \[sect. Foreground stars\], we classified five sources as foreground star candidates. The offset between the X-ray positions and optical positions corrected for proper motion (given as the weighted mean of RA and Dec in arcsec) is $\Delta$RA$ = -2.02 \pm 0.43$, $\Delta$Dec$ = -0.44 \pm 0.43$. The measured offset in RA agrees with the expected precision of the XMM-*Newton* Attitude Measurement System [@Guainazzi12]. We used these systematic offsets to correct the position of all detected sources. Analysis {#sect. Analysis} ======== [lcccc|lcccc]{}\ Source & flux max. & $V_{\rm f}$ & error $V_{\rm f}$ & $S$ & Source & flux max. & $V_{\rm f}$ & error $V_{\rm f}$ & $S$\ 2 &$(3.05 \pm 1.76)\times 10^{-14}$ & 3.0 & 0.6 & 1.2 & 104 &$(8.29 \pm 3.25)\times 10^{-15}$ & 1.9 & 0.6 & 1.0\ [ ]{} [ ]{} 3 &$(3.00 \pm 0.71)\times 10^{-14}$ & 3.1 & 0.2 & 2.9 & 107 &$(1.36 \pm 0.12)\times 10^{-13}$ & 1.97 & 0.12 & 5.1\ [ ]{} [ ]{} 4 &$(1.65 \pm 0.60)\times 10^{-14}$ & 1.7 & 0.4 & 1.2 & 108 &$(1.79 \pm 0.15)\times 10^{-13}$ & 1.44 & 0.10 & 3.3\ [ ]{} [ ]{} 7 &$(1.13 \pm 0.10)\times 10^{-13}$ & 7.16 & 0.09 & 9.4 & 109 &$(3.43 \pm 0.34)\times 10^{-14}$ & 2.0 & 0.4 & 2.4\ [ ]{} [ ]{} 9 &$(2.22 \pm 0.71)\times 10^{-14}$ & 1.5 & 0.3 & 1.1 & 110 &$(1.79 \pm 0.44)\times 10^{-14}$ & 2.3 & 0.4 & 2.1\ [ ]{} 12 &$(7.31 \pm 1.08)\times 10^{-14}$ & 3.95 & 0.14 & 5.0 & 111 &$(5.91 \pm 1.28)\times 10^{-14}$ & 6.4 & 0.2 & 3.9\ [ ]{} 15 &$(8.04 \pm 1.13)\times 10^{-14}$ & 3.1 & 0.2 & 4.4 & 113 &$(3.05 \pm 1.76)\times 10^{-14}$ & 3.0 & 0.6 & 1.2\ [ ]{} 16 &$(1.41 \pm 0.13)\times 10^{-13}$ & 10.6 & 0.3 & 9.5 & 114 &$(1.27 \pm 0.17)\times 10^{-13}$ & 2.39 & 0.18 & 4.1\ [ ]{} 17 &$(7.23 \pm 1.01)\times 10^{-14}$ & 2.1 & 0.2 & 3.3 & 116 &$(3.32 \pm 0.51)\times 10^{-14}$ & 2.63 & 0.15 & 4.0\ [ ]{} 19 &$(1.86 \pm 0.63)\times 10^{-14}$ & 1.9 & 0.3 & 1.4 & 118 &$(3.41 \pm 1.21)\times 10^{-14}$ & 3.6 & 0.4 & 2.0\ [ ]{} 20 &$(5.31 \pm 1.12)\times 10^{-14}$ & 3.3 & 0.2 & 3.3 & 119 &$(1.64 \pm 0.29)\times 10^{-14}$ & 1.70 & 0.17 & 2.4\ [ ]{} 25 &$(2.62 \pm 0.82)\times 10^{-14}$ & 1.8 & 0.3 & 1.4 & 121 &$(2.82 \pm 0.50)\times 10^{-14}$ & 1.76 & 0.17 & 2.4\ [ ]{} 26 &$(2.85 \pm 0.87)\times 10^{-14}$ & 1.8 & 0.3 & 1.4 & 122 &$(2.21 \pm 0.38)\times 10^{-14}$ & 1.62 & 0.17 & 2.2\ [ ]{} 33 &$(2.64 \pm 0.55)\times 10^{-14}$ & 1.4 & 0.3 & 1.1 & 123 &$(2.76 \pm 0.49)\times 10^{-14}$ & 1.8 & 0.3 & 2.1\ [ ]{} 37 &$(7.96 \pm 0.89)\times 10^{-14}$ & 8.58 & 0.11 & 7.9 & 126 &$(4.84 \pm 0.62)\times 10^{-14}$ & 1.4 & 0.2 & 1.4\ [ ]{} 40 &$(1.39 \pm 0.37)\times 10^{-14}$ & 1.7 & 0.5 & 1.2 & 129 &$(4.96 \pm 0.58)\times 10^{-14}$ & 3.94 & 0.11 & 6.4\ [ ]{} 41 &$(1.82 \pm 0.47)\times 10^{-14}$ & 2.4 & 0.4 & 2.0 & 131 &$(2.26 \pm 0.45)\times 10^{-14}$ & 2.2 & 0.4 & 2.1\ [ ]{} 50 &$(3.50 \pm 0.51)\times 10^{-14}$ & 2.77 & 0.14 & 4.4 & 133 &$(7.64 \pm 0.29)\times 10^{-13}$ & 2.24 & 0.06 & 12.4\ [ ]{} 51 &$(2.20 \pm 0.36)\times 10^{-14}$ & 1.8 & 0.3 & 2.0 & 135 &$(5.26 \pm 1.29)\times 10^{-14}$ & 3.8 & 0.2 & 3.0\ [ ]{} 53 &$(2.43 \pm 0.51)\times 10^{-14}$ & 1.5 & 0.3 & 1.3 & 136 &$(1.57 \pm 0.51)\times 10^{-14}$ & 1.9 & 0.5 & 1.2\ [ ]{} 55 &$(2.15 \pm 0.65)\times 10^{-14}$ & 2.5 & 0.3 & 2.0 & 140 &$(2.39 \pm 0.48)\times 10^{-14}$ & 1.4 & 0.3 & 1.1\ [ ]{} 56 &$(2.09 \pm 0.69)\times 10^{-14}$ & 2.2 & 0.3 & 1.7 & 143 &$(1.50 \pm 0.39)\times 10^{-14}$ & 2.0 & 0.3 & 1.9\ [ ]{} 60 &$(7.24 \pm 1.09)\times 10^{-14}$ & 1.46 & 0.19 & 1.8 & 144 &$(9.08 \pm 3.86)\times 10^{-15}$ & 2.6 & 0.8 & 1.2\ [ ]{} 61 &$(7.25 \pm 0.64)\times 10^{-14}$ & 4.28 & 0.08 & 8.6 & 145 &$(1.92 \pm 0.56)\times 10^{-14}$ & 2.0 & 0.3 & 1.7\ [ ]{} 62 &$(1.75 \pm 0.40)\times 10^{-14}$ & 1.7 & 0.3 & 1.6 & 153 &$(3.52 \pm 0.43)\times 10^{-14}$ & 1.6 & 0.2 & 1.9\ [ ]{} 64 &$(3.74 \pm 1.14)\times 10^{-14}$ & 2.5 & 0.4 & 1.9 & 154 &$(1.08 \pm 0.12)\times 10^{-13}$ & 6.41 & 0.11 & 7.5\ [ ]{} 65 &$(1.25 \pm 0.11)\times 10^{-13}$ & 3.90 & 0.15 & 7.8 & 155 &$(4.55 \pm 0.58)\times 10^{-14}$ & 3.61 & 0.12 & 5.7\ [ ]{} 67 &$(8.92 \pm 2.93)\times 10^{-15}$ & 2.4 & 0.5 & 1.6 & 157 &$(3.34 \pm 0.66)\times 10^{-14}$ & 3.46 & 0.19 & 3.6\ [ ]{} 69 &$(3.95 \pm 0.52)\times 10^{-14}$ & 3.13 & 0.13 & 5.1 & 160 &$(8.04 \pm 1.13)\times 10^{-14}$ & 12.4 & 0.4 & 6.4\ [ ]{} 75 &$(1.46 \pm 0.36)\times 10^{-14}$ & 1.7 & 0.5 & 1.1 & 162 &$(1.97 \pm 0.56)\times 10^{-14}$ & 2.3 & 0.5 & 1.7\ [ ]{} 79 &$(3.99 \pm 0.58)\times 10^{-14}$ & 1.4 & 0.2 & 1.5 & 163 &$(1.40 \pm 0.50)\times 10^{-14}$ & 1.6 & 0.4 & 1.1\ [ ]{} 80 &$(3.47 \pm 0.41)\times 10^{-14}$ & 1.53 & 0.11 & 2.9 & 166 &$(1.96 \pm 0.39)\times 10^{-14}$ & 1.68 & 0.19 & 2.0\ [ ]{} 81 &$(5.58 \pm 0.94)\times 10^{-14}$ & 2.5 & 0.3 & 3.0 & 171 &$(7.03 \pm 0.98)\times 10^{-14}$ & 3.74 & 0.13 & 5.3\ [ ]{} 87 &$(1.55 \pm 0.18)\times 10^{-13}$ & 1.31 & 0.13 & 1.9 & 172 &$(3.32 \pm 0.68)\times 10^{-14}$ & 2.1 & 0.2 & 2.5\ [ ]{} 89 &$(1.40 \pm 0.23)\times 10^{-14}$ & 2.1 & 0.5 & 1.8 & 173 &$(3.95 \pm 0.55)\times 10^{-14}$ & 3.13 & 0.13 & 4.9\ [ ]{} 90 &$(1.34 \pm 0.04)\times 10^{-12}$ & 1.35 & 0.03 & 8.1 & 177 &$(5.01 \pm 0.83)\times 10^{-14}$ & 2.2 & 0.4 & 2.6\ [ ]{} 93 &$(2.78 \pm 0.17)\times 10^{-13}$ & 16.42 & 0.06 & 15.6 & 183 &$(2.71 \pm 0.60)\times 10^{-14}$ & 1.9 & 0.2 & 2.1\ [ ]{} 94 &$(1.73 \pm 0.59)\times 10^{-14}$ & 1.9 & 0.3 & 1.4 & 184 &$(2.83 \pm 0.60)\times 10^{-14}$ & 1.6 & 0.2 & 1.8\ [ ]{} 98 &$(4.01 \pm 0.51)\times 10^{-14}$ & 3.18 & 0.12 & 5.4 & 186 &$(2.44 \pm 0.74)\times 10^{-14}$ & 2.1 & 0.3 & 1.7\ [ ]{} 99 &$(1.47 \pm 0.28)\times 10^{-13}$ & 15.29 & 0.18 & 5.0 & 187 &$(1.78 \pm 0.63)\times 10^{-14}$ & 1.7 & 0.4 & 1.2\ 103 &$(1.21 \pm 0.15)\times 10^{-13}$ & 12.78 & 0.12 & 7.2 & 189 &$(7.96 \pm 0.89)\times 10^{-14}$ & 8.58 & 0.11 & 7.9\ \ Source& flux max. & $V_{\rm f}$ & error $V_{\rm f}$ & $S$ & Source & flux max. & $V_{\rm f}$ & error $V_{\rm f}$ & $S$\ 55 &$(2.15 \pm 0.65)\times 10^{-14}$ & 3.3 & 0.3 & 2.3 & [ ]{} 97 &$(1.30 \pm 0.12)\times 10^{-13}$ & 2.79 & 0.10 & 6.6\ 60 &$(7.24 \pm 1.09)\times 10^{-14}$ & 2.35 & 0.16 & 3.8 & [ ]{} 98 &$(4.01 \pm 0.51)\times 10^{-14}$ & 3.18 & 0.12 & 5.4\ 69 &$(3.95 \pm 0.52)\times 10^{-14}$ & 3.13 & 0.13 & 5.1 & [ ]{} 99 &$(1.47 \pm 0.28)\times 10^{-13}$ & 15.29 & 0.18 & 5.0\ 76 &$(1.29 \pm 0.60)\times 10^{-14}$ & 4.5 & 0.5 & 1.7 & 106 &$(4.21 \pm 0.52)\times 10^{-14}$ & 1.28 & 0.13 & 1.7\ 79 &$(3.99 \pm 0.58)\times 10^{-14}$ & 1.4 & 0.2 & 1.5 & 107 &$(1.36 \pm 0.17)\times 10^{-13}$ & 1.97 & 0.12 & 5.1\ 80 &$(3.47 \pm 0.41)\times 10^{-14}$ & 1.53 & 0.11 & 2.9 & 108 &$(1.79 \pm 0.15)\times 10^{-13}$ & 1.44 & 0.10 & 3.3\ 81 &$(5.58 \pm 0.94)\times 10^{-14}$ & 2.5 & 0.3 & 3.0 & 109 &$(3.43 \pm 0.34)\times 10^{-14}$ & 4.02 & 0.14 & 7.4\ 84 &$(1.95 \pm 0.42)\times 10^{-14}$ & 2.7 & 0.3 & 2.8 & 114 &$(1.27 \pm 0.17)\times 10^{-13}$ & 2.39 & 0.18 & 4.1\ 87 &$(1.55 \pm 0.18)\times 10^{-13}$ & 1.35 & 0.11 & 2.2 & 116 &$(3.32 \pm 0.51)\times 10^{-14}$ & 20.8 & 0.3 & 6.2\ 88 &$(1.03 \pm 0.33)\times 10^{-14}$ & 4.6 & 0.4 & 2.5 & 125 &$(7.23 \pm 0.27)\times 10^{-15}$ & 5.3 & 0.5 & 2.2\ 92 &$(3.74 \pm 0.19)\times 10^{-14}$ & 2.7 & 0.4 & 4.3 & 129 &$(4.96 \pm 0.58)\times 10^{-14}$ & 3.94 & 0.11 & 6.4\ 95 &$(2.00 \pm 0.44)\times 10^{-14}$ & 1.9 & 0.2 & 2.0 & & & & &\ Variability of the sources {#sect. Variability of the sources} -------------------------- ### Short-term variability For each XMM-*Newton* observation, we searched for pulsations of the brightest sources (counts$\gtrsim 200$) on time scales between $\sim 4$ s and the time duration of each observation. After extracting the event files, we applied both a Fourier transform and a $Z^2_n$ analysis [@Buccheri83]. No statistically significant variability from the analysed sources was detected. ### Long-term variability To study the long-term time variability of sources observable at least in two different observations, we calculated the average flux (or the $3\sigma$ upper limit in case of non-detection) at the source position in each observation. We considered fluxes in the $0.2-4.5$ keV energy band because, as @Pietsch04 noted, the band $4.5-12$ keV has a lower sensitivity and is contaminated by hard background. We calculated the fluxes with the energy conversion factors (ECFs) reported in Table \[Tab. ecfs\]. Then, we searched for variable sources by comparing their fluxes (or upper limits) in different observations. We measured the X-ray variability of each source by its variability factor $V_{\rm f} = F_{\rm max}/F_{\rm min}$, where $F_{\rm max}$ and $F_{\rm min}$ are the maximum and minimum (or upper-limit) fluxes. To estimate the significance of the variability between different observations, we calculated the significance parameter $S=(F_{\rm max} - F_{\rm min})/\sqrt{\sigma_{\rm max}^2 + \sigma_{\rm min}^2}$, where $\sigma_{\rm max}$ and $\sigma_{\rm min}$ are the errors of the maximum and minimum flux [@Primini93]. We also studied the X-ray variability considering the *Chandra* observation of M83. We converted the *Chandra* counts ($0.3-8$ keV) of SW03 to $0.2-4.5$ keV fluxes with the conversion factor calculated by SW03 and the distance of M83 ($d=4.5$ Mpc) assumed in this work. The conversion factor $CF=8\times 10^{37}/300$ erg s$^{-1}$ counts$^{-1}$ was calculated by SW03 assuming an absorbed powerlaw spectrum with $\Gamma=1.7$, $N_{\rm H}=10^{21}$ cm$^{-2}$, and a distance of $3.7$ Mpc. For each *Chandra* source, we obtained the flux in the energy range $0.2-4.5$ keV correcting the luminosity $L_{\rm 0.3-8\ keV} =$ counts $\times CF$ by the absorption column density, the galaxy distance, and the energy range. The results are reported in Table \[Tab. variability\]. Fig. \[fig. variab\] shows the variability factor plotted versus the maximum detected flux and the hardness ratio (R2-R1)/(R1+R2) (see section \[sect. Hardness ratios diagrams\]) for each source. The left column shows the variability factors calculated for sources observed in at least two XMM-*Newton* observations. The right column shows the variability factors calculated for sources observed with *Chandra* and in at least one XMM-*Newton* observation. Applying a variability significance threshold of $S=3$, we found 35 variable sources. Like XRBs and AGNs, SSSs can show high variability, and because of their soft spectrum (see section \[sect. super-soft sources\]), they can be distinguished from the other sources: in Fig. \[fig. variab\] (lower panels), SSSs candidates should appear on the left-hand side, while XRBs (characterized by a much harder spectrum) are expected to appear on the right-hand side. Spectral analysis {#sect. spectral analysis} ----------------- We extracted the X-ray spectra of sources with $\gtrsim 300$ counts in the energy range $0.2-12$ keV. For each source, we fitted all three EPIC spectra simultaneously with different models: powerlaw, disc-blackbody, thermal plasma model (APEC @Smith01), and blackbody, using XSPEC (ver. 12.7.0, @Arnaud96). For the absorption we used the PHABS model. A good fit with one of the above-mentioned spectral models can be used to classify the sources into one of the following classes of sources: - X-ray binaries; - supernova remnants; - super-soft sources. In total, we fitted the spectra of 12 sources (see section \[sect. Discussion of classification and identification of the XMM-Newton sources\]). For sources that are not bright enough for spectral modelling, we only calculated their hardness ratios, as described in section \[sect. Hardness ratios diagrams\]. ![Hardness-ratio diagrams of sources with error-bars smaller than 0.3. Black squares are sources classified as XRBs (section \[sect. X-ray Binaries\]), orange diamonds are SNRs (section \[sect. supernova remnants\]), violet plus signs are SSSs (section \[sect. super-soft sources\]), green crosses are ULXs (section \[sect. Observation of an ULX\]), cyan stars are foreground stars (section \[sect. Foreground stars\]), red triangles are background sources (section \[sect. Background objects\]), and blue circles are sources not classified. The lines are the hardness ratios calculated for different spectral models and column densities, as described in section \[sect. Hardness ratios diagrams\]. []{data-label="fig. col-col diagrams"}](colcol_withouterror_all_hr0-vs-hr1.ps "fig:"){width="8cm"} ![Hardness-ratio diagrams of sources with error-bars smaller than 0.3. Black squares are sources classified as XRBs (section \[sect. X-ray Binaries\]), orange diamonds are SNRs (section \[sect. supernova remnants\]), violet plus signs are SSSs (section \[sect. super-soft sources\]), green crosses are ULXs (section \[sect. Observation of an ULX\]), cyan stars are foreground stars (section \[sect. Foreground stars\]), red triangles are background sources (section \[sect. Background objects\]), and blue circles are sources not classified. The lines are the hardness ratios calculated for different spectral models and column densities, as described in section \[sect. Hardness ratios diagrams\]. []{data-label="fig. col-col diagrams"}](colcol_withouterror_all_hr1-vs-hr2.ps "fig:"){width="8cm"} ![Hardness-ratio diagrams of sources with error-bars smaller than 0.3. Black squares are sources classified as XRBs (section \[sect. X-ray Binaries\]), orange diamonds are SNRs (section \[sect. supernova remnants\]), violet plus signs are SSSs (section \[sect. super-soft sources\]), green crosses are ULXs (section \[sect. Observation of an ULX\]), cyan stars are foreground stars (section \[sect. Foreground stars\]), red triangles are background sources (section \[sect. Background objects\]), and blue circles are sources not classified. The lines are the hardness ratios calculated for different spectral models and column densities, as described in section \[sect. Hardness ratios diagrams\]. []{data-label="fig. col-col diagrams"}](colcol_withouterror_all_hr2-vs-hr3.ps "fig:"){width="8cm"} Hardness-ratio diagrams {#sect. Hardness ratios diagrams} ----------------------- We used the hardness-ratio diagrams to separate different classes of sources according to their X-ray properties. They are especially helpful for sources that are too faint, for which spectral fitting is not possible. For each source, we computed four hardness ratios, defined as $$\label{eq. hrs} HR_i = \frac{R_{i+1} - R_i}{R_{i+1} + R_i} \mbox{ \ \ for \emph{i} = 1, ... , 4,}$$ where $R_i$ are the net source counts in five energy bands. To obtain the best statistics we combined the hardness-ratios of all three instruments. When a source was detected in more than one observation, we considered the observation with the highest number of counts. Some sources can exhibit different spectral states (which can be correlated with the X-ray flux), resulting in hardness-ratio changes between different observations (see e.g. @Done07). Therefore, for some of these sources we only considered a state by adopting the highest number of counts when determining the hardness ratio. This approach allowed us to obtain the hardness ratios with small uncertainties for bright sources in their bright states. However, one has to be aware that if a source changes its state, the hardness-ratio may change as well. For fainter sources (with hardness ratio uncertainties $\gtrsim 0.2$), the hardness ratios are not sensitive to changes of the state of the source within uncertainties. The hardness ratios calculated for each source are reported in Table \[Tab. source list\]. Fig. \[fig. col-col diagrams\] shows the hardness ratios of sources with errors smaller than $0.3$, detected in the field of view of M83. We plotted sources classified as XRBs, SNRs, SSSs, ultra-luminous X-ray sources (ULXs), foreground stars, and background objects (see section \[sect. Source classification\]) with different symbols. On the same plot we also overlaid grids of hardness ratios calculated for different spectral models: three absorbed powerlaws with photon-index $\Gamma=1$, $2$, $3$ (XRBs in hard state), two absorbed disc-blackbody models with temperatures at the inner disc radius of $kT_{\rm in}=0.5$ and $1$ keV (XRBs in soft state), four thermal plasma models APEC with temperatures $kT_{\rm apec}=0.2$, $0.5$, $1$, $1.5$ keV (SNRs), and two blackbody models with temperatures $kT_{\rm bb}=50$ and $100$ eV (SSSs, see section \[sect. super-soft sources\]). The column densities range from $N_{\rm H}=10^{20}$ cm$^{-2}$ to $N_{\rm H}=10^{24}$ cm$^{-2}$. Source classification {#sect. Source classification} ===================== We cross-correlated the list of sources observed with XMM-*Newton* with existing catalogues. For this purpose we used X-ray (@Trinchieri85; @Ehle98; @Immler99; SW03; @DiStefano03), optical (@Blair04; @Dopita10; @Jones04; @Rumstay83; USNO-B1, @Monet03), radio (@Maddox06; @Cowan94; @Condon98), and infrared (2MASS, @Skrutskie06) catalogues. We considered two sources as associated to each other if their positions were closer than the $3\times$ combined statistical errors. The optical counterparts of several X-ray sources cannot be determined uniquely. In such cases we assumed as counterpart the brightest optical object within the error circle. The cross-correlations are reported in Table \[Tab. source list classification\] in appendix \[sect. catalogue-table\]. We used the previous classifications in X-rays and other wavelengths and the methods of classification described in sections \[sect. Variability of the sources\] (X-ray variability), \[sect. spectral analysis\] (spectral analysis), and \[sect. Hardness ratios diagrams\] (hardness ratios), to identify and classify sources as background objects, foreground stars, XRBs, SNRs, SSSs, and ULXs. In this section we describe the observational properties for each class of sources and define the classification criteria. Foreground stars {#sect. Foreground stars} ---------------- X-ray observations of nearby galaxies are contaminated by foreground stars, which have X-ray luminosities ranging from $\sim 10^{26}$ to $\sim 10^{30}$ erg s$^{-1}$ for stars of spectral type F to M, and $\sim 10^{29}$ to $\sim 10^{34}$ erg s$^{-1}$ for stars of spectral types O and B (@Vaiana81; @Rosner85). Stars of spectral classes F to M emit X-rays because of the intense magnetic fields that form a corona, in which the plasma is heated to temperatures of about $\sim 10^6-10^8$ K (e.g. @Guedel02). A mechanism proposed to explain the X-ray emission from stars of spectral types O-B is the formation of shocks in the coronal regions due to the instability of the wind-driven mechanism (see @Puls08 and references therein). In A-type stars, none of the above mechanisms for X-ray emission can operate efficiently. Therefore, A-type stars are expected to be weak X-ray sources [@Schroeder07] and only very few have been observed in X-rays (see e.g. @Robrade10; @Schroeder08). {width="9.1cm"} {width="9.1cm"} The X-ray spectra of foreground stars are relatively soft and can be described by models of optically thin plasma in collisional equilibrium (e.g. @Raymond77) with temperatures ranging from $10^6$ to $10^7$ K. A common method to distinguish stars from other X-ray sources is comparing the X-ray-to-optical flux ratio, as suggested by @Maccacaro88: $$\label{eq. Maccacaro} \log_{\rm 10} (f_{\rm x}/f_{\rm opt}) = \log_{\rm 10}(f_{\rm x}) + \frac{m}{2.5} +5.37 \mbox{ ,}$$ where $m$ is the visual magnitude $m_{\rm v}$. In the USNO-B1 catalogue the red and blue magnitudes are given, thus we assumed $m_{\rm v} \approx (m_{\rm red} + m_{\rm blue})/2$. We used the blue magnitude $m_{\rm blue}$ as magnitude $m$ when the red magnitude was not available. For each X-ray source with an optical counterpart, we distinguished foreground stars from other sources by plotting X-ray-to-optical flux ratios over the hardness ratios $HR_2$ and $HR_3$ (Fig. \[fig. log10-fx-fopt\]). The X-ray-to-optical flux ratios and the hardness ratios differ significantly between different classes of sources. The soft X-ray flux of early-type stars (OB type) scales with $f_{\rm x} \approx 10^{-7} f_{\rm opt}$ (@Kudritzki00 and references therein), while the ratio $f_{\rm x}/f_{\rm opt}$ of late-type stars (F to M) usually ranges from $10^{-6}$ to $10^{-1}$ (e.g. @Krautter99). In contrast, sources such as SNRs, SSSs, and XRBs radiate mainly in X-rays. We also used optical and near-infared magnitudes and colours to classify foreground stars (Figs. \[fig. b-r\_j-k.ps\] and \[fig. B-R vs R\]). Fig. \[fig. b-r\_j-k.ps\] is the colour-colour diagram for XMM-*Newton* sources with optical (USNO-B1) and infrared (2MASS) counterparts. Lines show the expected $(B-R)$ and $(J-K)$ colours for main-sequence, giant, and supergiant stars belonging to the Milky Way. We obtained these lines using intrinsic colours calculated by @Johnson66. Stars located at the Galactic latitude of M83 ($b \approx 32^\circ$) have on average a colour excess per kiloparsec of $E(B-V) = 0.05 \pm 0.05$ mag kpc$^{-1}$ [@Gottlieb69]. Therefore, the colour excesses $E(J-K)$ and $E(B-R)$ are negligible compared to the optical and infrared magnitude uncertainties [@Schild77]. Figs. \[fig. b-r\_j-k.ps\] and \[fig. B-R vs R\] allow to separate foreground stars from other classes of sources. Foreground stars are brighter in $R$ than background objects or members of M83, and sources with $J-K \lesssim 1.0$ and $B-R \lesssim 2.0$ are most likely foreground stars. From previous considerations, we classified foreground stars when these conditions were met: - $\log (f_{\rm x}/f_{\rm opt}) \lesssim -1$; - $HR_2 \lesssim 0.3$; - $HR_3 \lesssim -0.4$; - $J-K \lesssim 1.0$; - $B-R \lesssim 2.0$. The five sources classified as foreground star candidates are reported in Table \[Tab. source-offset list\]. A detailed discussion of the identification and classification of foreground stars is provided in sections \[sect. Discussion foreground stars\] and \[sect. Sources which are not foreground stars\]. Background objects {#sect. Background objects} ------------------ The identification of AGNs, normal galaxies, and galaxy clusters is based on SIMBAD and NED correlations, and is confirmed if there is an optical counterpart in the *2nd Digitized Sky Survey* (DSS2) image. New classifications are based on the radio counterpart and hardness ratio $HR_2 \geq -0.4$ [@Pietsch04]. We identified nine sources as background galaxies and AGNs (sources No. 7, 17, 31, 65, 83, 89, 144, 148, 158, see Table \[Tab. list-galaxies\]). We found radio counterparts of the sources No. 20, 37, 189 and classified them as AGN candidates for the first time (see section \[sect. Discussion Background objects\]). Based on the $\log N - \log S$ calculated by @Cappelluti09 (see section \[sect. AGN-corrected XLFs\]), about $40$ observed sources (with a $2-10$ keV flux $F_{\rm x} > 10^{-14}$ erg cm$^{-2}$ s$^{-1}$) are expected to be background objects in each XMM-*Newton* observation of Table \[Tab. OBS ID XMM\]. From a comparison with other works (e.g. @Misanovic06), we expect a large difference between the predicted number of background objects from background surveys and the number of identified/classified background objects in an XMM-*Newton* observation. This difference is due to the difficulty in classifying sources which, because of their distance, are too faint (and therefore provide little information) to be classified with the methods at our disposal. Notes:\ $^1$: @Maddox06;\ $^2$: @Bresolin09;\ $^3$: @Immler99. Nuclear sources {#sect. nuclear sources} --------------- We detected two bright sources in the nuclear region of M83 with the source detection procedure: sources No.92 and No.95. They are separated by $\sim6.3^{\prime\prime}$ and are the brightest sources detected with XMM-*Newton* in M83 ($F_{\rm No.\,90} = [1.03 \pm 0.25] \times 10^{-12}$ erg cm$^{-2}$ s$^{-1}$; $F_{\rm No.\,93} = [2.59 \pm 0.15] \times 10^{-13}$ erg cm$^{-2}$ s$^{-1}$; $0.2-12$ keV, assuming an absorbed powerlaw spectrum with index 1.8 and a foreground Galactic absorption of $N_{\rm H}=3.69 \times 10^{20}$ cm$^{-2}$). The two nuclear sources coincide with $\sim 18$ *Chandra* sources and the bright diffuse emission of the starburst nucleus, not resolved by XMM-*Newton* because of its high PSF, which causes source confusion in crowded regions, such as the nuclear region of M83. X-ray binaries {#sect. X-ray Binaries} -------------- We classified sources as XRBs if the X-ray spectra or hardness ratios were compatible with the typical spectra of XRBs and we detected a flux periodicity. We identified two X-ray binaries (Nos.81 and 120), previously classified by SW03 using *Chandra* observations (section \[sect. class. X-ray binaries\]). Supernova remnants {#sect. supernova remnants} ------------------ We assume that the X-ray spectra of SNRs are well described by the thermal plasma model APEC [@Smith01], with temperatures ranging from $0.2$ to $1.5$ keV. At this distance we are unable to resolve an SNR or to verify a more detailed spectral model assuming, e.g., a non-equilibrium ionisation. We classified an X-ray source as SNR if $HR_1 >0.1$, $HR_2 < -0.4$, the source was not a foreground star, and did not show a significant variability [@Pietsch04]. We identified the source No.79 as source \[SW03\]27, classified as a young SNR candidate by SW03 (section \[sect. class. Supernova remnants\]). #### **SN1957D** @Long12 reported the first detection of SN1957D in X-rays with *Chandra*. The source shows a luminosity of $1.7 \times 10^{37}$ erg cm$^{-2}$ s$^{-1}$ ($d=4.61$ Mpc, @Saha06; $0.3-8$ keV), and the spectrum is well modelled with an absorbed powerlaw with an index $\sim 1.4$, a foreground Galactic absorption of $N_{\rm H}=4 \times 10^{20}$ cm$^{-2}$ and an intrinsic column density of $N_{\rm H}=2 \times 10^{22}$ cm$^{-2}$. We did not detect SN1957D in the XMM-*Newton* observations. In observation 1 the source is located near to the centre of the field of view, and in the other two observations the source is located at the edge of the field of view. Assuming the spectral parameters found by @Long12, we calculated a $3\sigma$ upper-limit in observation 1 of $\sim 2.4 \times 10^{-14}$ erg cm$^{-2}$ s$^{-1}$ ($0.2-12$ keV), corresponding to a luminosity of $\sim 5.8 \times 10^{37}$ erg s$^{-1}$, well above the luminosity detected by @Long12. Super-soft sources {#sect. super-soft sources} ------------------ Super-soft sources are a class of sources that are believed to be binary systems containing a white dwarf. The white dwarf accretes matter from a Roche-lobe-filling companion at high rates ($\dot{M}_{\rm acc} \sim 10^{-7}$ M$_\odot$ yr$^{-1}$), which leads to quasi-steady nuclear burning on its surface (see e.g. @vandenHeuvel92). SSSs show soft spectra with blackbody temperatures of $15-150$ eV and X-ray luminosities ranging from $\sim 10^{35}$ erg s$^{-1}$ to $10^{38}$ erg s$^{-1}$ (@DiStefano03; @Kahabka97). An additional harder component, due to interactions of the radiation with matter near to the white dwarf or wind interactions can be observed [@DiStefano03]. Moreover, SSSs are often observed as transient X-ray sources (see @Greiner00). Other classes of sources with soft spectra can be confused with SSSs. For example, some X-ray pulsars observed outside the beam of the pulsed radiation can show a soft ($\sim 30$ eV) component (@Hughes94; @DiStefano03). Moreover, stripped cores of giant stars can be classified as SSSs [@DiStefano01]. As our classification criteria, we assumed blackbody temperatures of $kT_{\rm bb} \leq 100$ eV (in agreement with the selection procedure proposed by @DiStefano03) and hardness ratios that do not overlap with those of other classes of sources. These criteria are an $HR_1 \lesssim 0$ and $HR_2 - EHR_2 < -0.9$. We classified a source as SSS only if both criteria are fulfilled. We identified source No.91 as source M83-50, classified as an SSS candidate by @DiStefano03 using *Chandra* observations (section \[sect. class. Super-Soft Sources\]). Hard sources {#sect. main hard sources} ------------ Hard sources show hard X-ray spectra (or hard HRs, see Table 5 in @Pietsch04). Using their spectral properties and hardness ratios, we classified five hard sources (Nos.16, 61, 103, 126, and 153; see section \[section Newly classified hard sources\]) and we identified 11 hard sources (Nos.60, 80, 92, 97, 99, 106, 107, 108, 114, 116, 129; see section \[section identification hard sources\]). Ultra-luminous X-ray sources {#sect. Observation of an ULX} ---------------------------- ULXs are pointlike non-nuclear sources with X-ray luminosities in excess of the Eddington limit ($L_{\rm Edd} \simeq 10^{39}$ erg s$^{-1}$) for a stellar mass black-hole (see e.g. @FengSoria11). They are usually located in active star-forming environments [@Miller04], and their nature is still unclear; recent studies indicate that ULXs are a heterogeneous sample of objects (e.g. @Gladstone11). Several models have been proposed to explain the high X-ray luminosity of ULXs, but there are three models that are often used for this class of sources. The first model requires that ULXs are intermediate-mass black-hole systems (IMBHs) with masses $M \sim 10^{2}-10^{4}$ M$_\odot$, accreting at sub-Eddington rates (e.g. @Colbert99). The other models assume that ULXs are stellar-mass black holes (with masses $M \lesssim 100$ M$_\odot$) in a super-Eddington accretion regime [@Poutanen07] or with beamed radiation (see e.g. @King09). We identified ULX No.133, discovered by @Trinchieri85 with *Einstein* (source H2), and previously observed in X-rays with ROSAT by @Ehle98 and @Immler99 (see section \[sect. class. Ultra-Luminous X-ray sources\]). X-ray luminosity functions {#sect. X-ray Luminosity Functions} ========================== For each observation, we calculated the XLFs in the energy range $2-10$ keV excluding the softer bands to reduce the effect of incompleteness of the observed source sample due to absorption. Moreover, from an XLF calculated in this energy band, it is possible to easily subtract the contribution of the $\log N-\log S$ of the AGNs, which was calculated from several surveys performed by XMM-*Newton* and *Chandra* (see section \[sect. AGN-corrected XLFs\]). We considered for XLFs only sources with a detection likelihood greater than 6 in the energy range $2-12$ keV. For each source, we converted the count rates to the $2-10$ keV fluxes using the ECFs of Table \[Tab. ecfs\] for the energy bands R4 and R5. We excluded the region inside a circle centred on the nuclear region of M83 with radius $R = 26^{\prime\prime}$ from the XLF calculation, where the large PSF of EPIC in a crowded region causes source confusion effects (see section \[sect. nuclear sources\]). Since we were interested in obtaining XLFs of XRBs, we also excluded the sources previously classified as SNRs, SSSs, ULXs, and foreground stars (section \[sect. Source classification\]). For each observation, we calculated the XLFs of sources detected within two regions of M83: the inner disc inside the $D_{25}$ ellipse, and the outer disc outside the $D_{25}$ ellipse. XLFs corrected for incompleteness {#sect. XLFs corrected for incompleteness} --------------------------------- ![Sky coverage as a function of the X-ray flux ($2-10$ keV) for the region inside the $D_{25}$ ellipse (observation 1), calculated excluding the region within the circle centered on the nuclear region of M83 with radius $R = 26^{\prime\prime}$.[]{data-label="fig. sky coverage function"}](sky_coverage_all_2-10keV.ps){width="9cm"} The sensitivity of the EPIC instruments depends on the exposure, background, and PSF, which are not uniform across the FOV. Indeed, the exposure time is relatively high at the centre of the FOV and decreases with increasing off-axis angle (vignetting effect). The background, modelled by the task [esplinemap]{}, decreases with increasing angular distance from the nuclear region of M83 (due to the diffuse emission in the disc of M83), and the optical properties of the X-ray telescope introduce a degradation of the PSF with increasing off-axis angle. Therefore, the sensitivity also varies across the observed area, allowing the detection of the brightest sources across the entire observed area, whereas the effective area for the detection of faint sources is smaller. This effect leads to an underestimation of the number of sources observed at the faintest flux levels. We corrected the XLFs by taking into account the incompleteness effect described above by calculating the *sky coverage function*, which is the effective area covered by the observation as a function of flux. For each observation, we first created the combined sensitivity maps of PN, MOS1, and MOS2 with the SAS task [esensmap]{}, which requires as input files the exposure maps, the background images, and the detection masks created by the source detection procedure. We used the sensitivity maps to calculate the sky coverage function for each observation (Fig. \[fig. sky coverage function\]). The cumulative XLF corrected for incompleteness is given by $$\label{eq. xlf} N(>F_{\rm x}) = A_{\rm tot} \sum_{i=1}^{N_{\rm s}} \frac{1}{\Omega(F_i)} \mbox{ ,}$$ where $N(>F_{\rm x})$ is the number of sources with a flux higher than $F_{\rm x}$, weighted by the fraction of the surveyed area $\Omega(F_i)/A_{\rm tot}$ over which sources with flux $F_i$ can be detected; $A_{\rm tot}$ is the total area of the sky observed by EPIC, $\Omega(F_i)$ is the sky coverage (Fig. \[fig. sky coverage function\]), and $N_{\rm s}$ is the total number of the detected sources. Therefore, with equation (\[eq. xlf\]), every source is weighted with a factor correcting for incompleteness at its flux. The variance of the source number counts is defined as $$\label{eq. var xlf} \sigma^2 = \sum_{i=1}^{N_{\rm s}} \left( \frac{1}{\Omega_i} \right)^2 \mbox{.}$$ AGN-corrected XLFs {#sect. AGN-corrected XLFs} ------------------ The XLFs obtained in section \[sect. XLFs corrected for incompleteness\] consist of sources belonging to M83 (XRBs) and AGNs. We subtracted the AGN contribution using the AGN XLF of @Cappelluti09, who derived the XLFs from the 2 deg$^2$ of the XMM-COSMOS survey [@Scoville07]. These authors found that the XLF of AGNs in the energy range $2-10$ keV is described by a broken powerlaw: $$\label{eq xlf cappelluti09} \frac{dN}{dF} = \left\{ \begin{array}{rl} AF^{-\alpha_1} & F > F_{\rm b} \\ BF^{-\alpha_2} & F \leq F_{\rm b} \mbox{ ,} \end{array} \right.$$ where $A=BF_{\rm b}^{\alpha_1 - \alpha_2}$ is the normalisation, $\alpha_1=2.46 \pm 0.08$, $\alpha_2=1.55 \pm 0.18$, $F_{\rm b}=(1.05 \pm 0.16)\times 10^{-14}$ erg cm$^{-2}$ s$^{-1}$, and $A=413$. Fig. \[fig. xlf\] shows the XLFs of sources detected within the $D_{25}$ ellipse and outside, calculated for each XMM-*Newton* observation. Blue lines are the observed XLFs, and black lines are the XLFs corrected for incompleteness. Solid green lines are the AGN XLFs of equation (\[eq xlf cappelluti09\]) with relative uncertainties (dashed green lines). Solid red lines show the XLFs corrected for incompleteness and AGN-subtracted, and dashed red lines are the 90% confidence errors, obtained from equation (\[eq. var xlf\]) and the 90% confidence errors of the AGN distribution. Vertical black lines in the right column of Fig. \[fig. xlf\] show the level at which the survey is 90% complete (see section \[sect. outer disc\]), defined as the flux at which $$\sum_{i=1}^{N_{\rm s}} N(F_i) / \sum_{i=1}^{N_{\rm s}} A_{\rm tot}/\Omega(F_i) = 0.9 \mbox{ .}$$ Fit {#sect. Resulting XLFs} --- {width="9cm"} {width="9cm"} {width="9cm"} {width="9cm"} {width="9cm"} {width="9cm"} We fitted the differential XLFs corrected for incompleteness and AGN-subtracted with a powerlaw: $$\label{eq pl fit} A(F) = kF^{\alpha} \mbox{ ,}$$ where $k$ is the normalisation and $\alpha$ the powerlaw index. Notes:\ $^a$: see equation (\[eq pl fit\]);\ $^b$: see equation (\[eq bpo fit\]);\ We also fitted the differential XLFs with a broken powerlaw: $$\label{eq bpo fit} A(F) = \left\{ \begin{array}{lr} kF_{\rm b}^{\alpha_2 - \alpha_1} F^{\alpha_1} & F > F_{\rm b} \\ kF^{\alpha_2} & F \leq F_{\rm b} \mbox{ ,} \end{array} \right.$$ where $F_{\rm b}$ is the break point. The resulting parameters obtained from the fit are reported in Table \[Tab. all fit xlfs\]. ### Inner disc {#sect. Inner disc} From *Chandra* observation, SW03 calculated the XLFs of sources located in the inner region (distance $<60^{\prime\prime}$ from the nucleus) and outer region ($60^{\prime\prime}<d<R_{25}$) of the optical disc. They found that the inner region sources have a powerlaw luminosity distribution with an differential index of $-1.7$, while the luminosity distribution of the outer region sources shows a lack of bright sources above $\sim 10^{38}$ erg s$^{-1}$. These authors modelled the XLF of these sources with a broken powerlaw with a break around $\sim 10^{38}$ erg s$^{-1}$ and differential indices of $-1.6$ and $-2.6$. They explained the XLF of the inner region sources in terms of current starburst activity, while the XLF of the outer region may result from an older population of disc sources mixing with a younger population. We recall that we cannot study the innermost region because of poor spatial resolution of XMM-*Newton* compared to *Chandra*. We compared the best-fitting parameters of the XLF of the outer region sources ($60^{\prime\prime}<d<R_{25}$) obtained by SW03 with those obtained from the XMM-*Newton* analysis (Table \[Tab. all fit xlfs\]). In particular, we considered the broken powerlaw fit of sources detected in observation 1. Only during this observation was the whole optical disc of M83 observed. We found that the indices $\alpha_1$, $\alpha_2$ and the break $F_{\rm b}$ of equation \[eq bpo fit\] agree within the uncertainties with the parameters found by SW03. @Grimm03 studied the XLFs of a sample of galaxies and found the probable existence of a universal HMXB XLF (in the luminosity range $\sim 4 \times 10^{36}-10^{40}$ erg s$^{-1}$), described by a powerlaw with differential slope of $-1.6$. They found that the number of HMXBs with $L_{\rm x} > 2 \times 10^{38}$ erg s$^{-1}$ in a star-forming galaxy is directly proportional to the SFR, and proposed that the number and the total X-ray luminosity of HMXBs can be used to measure the star formation rate of a galaxy. Based on a much larger sample of galaxies, @Mineo12 found that the properties of populations of HMXBs and their relation with the SFR agree with those obtained by @Grimm03. We estimated the SFR in the optical disc of M83 using the $N_{\rm HMXBs} - $SFR relation of @Mineo12: $$\label{eq. mineo} N (> 10^{38} \mbox{erg s}^{-1}) = 3.22 \times \mbox{SFR\,(M}_\odot\mbox{ yr}^{-1}\mbox{)\,.}$$ We assumed that the XLF we used for this calculation provides a good approximation of the HXMB XLF in M83. The contribution of LMXBs to the XLF is negligible for a starburst galaxy such as M83 when $L_{\rm x} \gtrsim 10^{38}$ erg s$^{-1}$ [@Grimm03]. Moreover, the contribution of LMXBs to the XLF is minimized by excluding the nuclear region of the galaxy, from which a strong contribution to the total number of LMXBs is expected. Using the XLF of sources detected in observation 1 within the $D_{25}$ ellipse, from equation \[eq. mineo\] we found an SFR$\approx 3.1$ M$_\odot$ yr$^{-1}$, in agreement with the SFR estimates obtained from observations in other wavelengths (see e.g. @Boissier05; @Dong08; @Grimm03 and references therein). ### Outer disc {#sect. outer disc} The XLFs of the outer disc ($d > R_{25}$) show an excess of sources (with respect to the expected number of AGNs) in the luminosity range $\sim 10^{37}$ to $\sim 2\times 10^{38}$ erg s$^{-1}$ (Fig. \[fig. xlf\]). We are interested in calculating the probability of the luminosity distribution of the observed sources to be consistent with the luminosity distribution of equation (\[eq xlf cappelluti09\]) which represents the AGN distribution. Therefore, we compared for each observation the luminosity distribution of the sources detected in the outer disc ($d> R_{25}$) that was not corrected for incompleteness (see section \[sect. X-ray Luminosity Functions\]) with a distribution of simulated sources over the EPIC FOV obtained from a uniform spatial distribution of sources with a luminosity distribution given by equation (\[eq xlf cappelluti09\]), filtered to exclude sources with a flux below the detection threshold calculated at the position of each source in the sensitivity map. The Kolmogorov-Smirnov test applied to these source samples showed that the probabilities that the luminosity distributions of the observed sources are consistent with the luminosity distribution of AGNs (equation \[eq xlf cappelluti09\]) are almost zero, being $0.04$% in observation 1, $0.7$% in observation 2, and $0.6$% in observation 3. ![Sample of $10^4$ simulated sources, distributed over the EPIC field of view of observation 1 and located outside the $D_{25}$ ellipse.[]{data-label="fig. agn distrib. simul."}](xyunif_incomplcorr_1e4.ps){width="9cm"} To quantify the probability that the set of X-ray sources located outside the $D_{25}$ ellipse are AGNs (which are expected to be uniformly distributed across the sky) or XRBs (whose distribution should not be uniform, because the position of XRBs should correlate with the arms extending out of the optical disc), we performed a two-dimensional Kolmogorov-Smirnov test (@Fasano87; @Peacock83). This test is based on the statistic $\delta$, which in the unidimensional Kolmogorov-Smirnov test represents the largest difference between two cumulative distributions. We applied this test to two data samples: 1. all X-ray sources detected in observation 1 that are located outside the $D_{25}$ ellipse. The number of these sources is $N_1=39$; 2. a distribution of simulated sources in the EPIC FOV of observation 1, obtained from a uniform spatial distribution of sources (which represents the uniform spatial distribution of AGNs) modified to take into account the incompleteness effect described in section \[sect. XLFs corrected for incompleteness\]. We obtained this spatial distribution of sources as follows. We first generated a uniform spatial distribution of sources with fluxes given by the XLF of AGNs described in section \[sect. AGN-corrected XLFs\]. Then, we selected sources with flux higher than that corresponding to the position of each source in the sensitivity map. We additionally selected sources with luminosity $> 10^{37}$ erg s$^{-1}$ in the energy range $2-10$ keV that are located outside the $D_{25}$ ellipse. With this method, we generated a sample of $N_2=10^4$ coordinate pairs (RA, Dec) of sources (see Fig. \[fig. agn distrib. simul.\]). From the number of data points $N_1$ and $N_2$ of the two data sets, the significance level was calculated from the probability distribution of the quantity $$\label{eq. Zn} Z_n \equiv \delta \sqrt{n} \mbox{,}$$ where $n=N_1N_2/(N_1 + N_2)$. The analytical formula for calculating of the probability that the two data samples come from the same distribution is accurate enough for large data sets with $n>80$ [@Fasano87]. Since in our case $n \approx 39$, we needed to use Monte Carlo simulations. We generated many synthetic data samples simulating the uniformly distributed AGNs with the same method previously used to calculate sample 2; each of the synthetic data samples has the same number of sources as the observed data set 1 ($N_1=39$). For each data set we applied the 2D Kolmogorov-Smirnov test by comparing the synthetic data set with the set of $10^4$ sources distributed across the EPIC FOV previously described, then we calculated the quantity $Z_n$ using equation (\[eq. Zn\]). The probability of the observed $Z_n$ is given by the fraction of the times the simulated $Z_n$ are larger than the observed $Z_n$. Applying this statistical method to our data, we found a probability of $99.5$% that the observed sample 1 and the simulated homogeneously distributed sample 2 are significantly different, which suggests a non-uniform distribution of the observed X-ray sources and therefore a possible correlation between the positions of these sources and the extended arms of M83. The incompleteness correction given by equation (\[eq. xlf\]) is based on the hypothesis that sources are uniformly distributed. However, we have demonstrated that the X-ray sources located outside the $D_{25}$ ellipse have a non-uniform distribution, hence the associated XLFs corrected for incompleteness of Fig. \[fig. xlf\] (right column) are not reliable at low luminosities. Therefore we only considered the part of the XLFs with luminosities higher than the level at which the survey is 90% complete (to the right of the vertical black lines in Fig. \[fig. xlf\]). We found that the 90% complete XLFs of observations 1 and 3 (for which we have enough data points to find a good fit) are well fitted with a powerlaw with differential slopes $\alpha = -2.2 \pm 0.5$ (observation 1), and $\alpha = -1.7 \pm 0.4$ (observation 3), which are consistent with each other within errors. These are also consistent with the AGN slope of @Cappelluti09. Assuming that the spatial distribution of AGNs and their number density are not subject to strong fluctuations on small angular scales corresponding to different directions in the M83 field, the observed excess of sources (with respect to the AGN distribution) in the luminosity range $\sim 10^{37}$ to $\sim 2\times 10^{38}$ erg s$^{-1}$ (Fig. \[fig. xlf\]) can probably be ascribed to a population of XRBs located in the outer disc of M83. The recent star-forming activity discovered by GALEX in this region indicates that a large portion of the observed X-ray sources are HMXBs. However, the observed XLF slope is steeper than the slope of the universal HMXB XLF inferred by @Grimm03. A possible explanation for the difference between the two slopes could be that the observed XLFs are the result of a mix of XRB populations formed after starbursts of different ages. An alternative explanation is that the mass distribution of the population of stars in the low-density regions of the outer disc of M83 is described by a truncated initial mass function (IMF), whose existence was proposed to explain the production of fewer high-mass stars (compared to the standard IMF) in low-density environments (see e.g. @Krumholz08; @Meurer09). The universality of the IMF is still a matter of debate [@Bastian10]; in this context, a recent Subaru H$\alpha$ observation of the outer disc of M83 revealed O stars even in small clusters ($M \lesssim 10^3$ M$_\odot$), which supports the hypothesis that the IMF is not truncated in low-density environments [@Koda12]. Summary {#sect. conclusions} ======= We presented an analysis of three XMM-*Newton* observations of M83. We performed the source detection procedure separately for images of each observation, and we obtained a catalogue containing 189 sources. Based on cross-correlations with other catalogues we identified counterparts for 103 sources, 12 of which were identified or classified as background objects and 5 as foreground stars (one as candidate CV). We performed spectral analysis of the sources with the largest number of counts, as well as studies of the X-ray variability and the hardness ratio diagrams. The spectral analysis of ULX No.133 in observations 2 and 3 showed good fits with the standard IMBHs model as well as with accreting stellar-mass black-hole model, in agreement with the results obtained by @Stobbart06 from observation 1. In Sect. \[sect. X-ray Luminosity Functions\] we presented the XLFs of sources in the $2-10$ keV energy band, within and outside the $D_{25}$ ellipse. We corrected the XLFs for incompleteness and subtracted the contribution of background AGNs from the total XLF to obtain the XLFs of XRBs. The XLF of the optical disc is well fitted with a powerlaw or a broken powerlaw, while the XLF of the outer disc is well fitted with a simple powerlaw. The broken powerlaw fit parameters agree (within the uncertainties) with the parameters found by SW03 with *Chandra*. From the XMM-*Newton* XLF, we obtained an SFR$\approx 3.1$ M$_\odot$ yr$^{-1}$ in the optical disc of M83, which agree with previous estimates obtained in other wavelengths. The XLFs of these sources show an excess of sources (compared to the AGNs distribution) in the luminosity range $\sim 10^{37}$ to $\sim 2\times 10^{38}$ erg s$^{-1}$. The application of the Kolmogorov-Smirnov test to the X-ray sources detected outside the $D_{25}$ ellipse allowed us to find that this population of sources is significantly different from the population of background AGNs, which is supposed to have a homogeneous distribution. These results led us to suggest that a part of the X-ray sources observed outside the $D_{25}$ ellipse belongs to the outer disc of M83. The 90% complete XLFs of the outer disc are well fitted with a simple powerlaw with differential slope $\alpha = -2.2 \pm 0.5$ (observation 1), and $\alpha = -1.7 \pm 0.4$ (observation 3) steeper than the universal HMXB XLF discovered by @Grimm03. We proposed as a possible origin for the steep slope of the observed XLF that the observed XLFs are the result of a mix of XRB populations of different ages, or, as an alternative explanation, that the IMF in the low-density regions of the outer disc of M83 is truncated, as previously suggested by e.g. @Krumholz08 and @Meurer09 to explain the low production of high-mass stars in low-density environments. Additional X-rays and UV observations of the outer disc of M83, analysed with most effective methods such as the one used by @Bodaghee12 to measure the spatial cross-correlation of HMXBs and OB star-forming complexes in the Milky-Way, will be fundamental to confirm our hypothesis. We thank the referee Eric M. Schlegel for constructive comments, which helped to improve the manuscript. This research is funded by the Deutsche Forschungsgemeinschaft through the Emmy Noether Research Grant SA 2131/1. This research has made use of the SIMBAD database, operated at CDS, Strasbourg, France, and of the NASA/IPAC Extragalactic Database (NED), which is operated by the Jet Propulsion Laboratory, California Institute of Technology, under contract with the National Aeronautics and Space Administration. This publication has made use of data products from the Two Micron All Sky Survey, which is a joint project of the University of Massachusetts and the Infrared Processing and Analysis Center, funded by the National Aeronautics and Space Administration and the National Science Foundation. This research has made use of SAOImage DS9, developed by Smithsonian Astrophysical Observatory. Classification and identification of the XMM-*Newton* sources {#sect. Discussion of classification and identification of the XMM-Newton sources} ============================================================= Foreground stars {#sect. Discussion foreground stars} ---------------- #### **Sources No.21, 143, 182, and 174** Using the criteria in Sect. \[sect. Foreground stars\], we classified sources No.21, 143, 182, and 174 as foreground stars according to their optical and infrared properties (Figs. \[fig. b-r\_j-k.ps\], \[fig. B-R vs R\]), and their optical-to-X-ray ratios as a function of the hardness ratios (Fig. \[fig. log10-fx-fopt\]). Although the hardness ratio criterion $HR_3 \lesssim -0.4$ of source No.174 is not fulfilled, we classified this source as a foreground star because of the large uncertainty of the hardness ratio (see Fig. \[fig. log10-fx-fopt\]). #### **Source No.24** has optical and infrared counterparts and $\log_{10}(f_{\rm x}/f_{\rm opt})<-1$, but violates the hardness ratio $HR_2$ criterion (see Fig. \[fig. log10-fx-fopt\]). The optical counterpart is bright ($m_{\rm B,No.\,24}=14.1$), and the $B-R$ and $J-K$ colours are consistent with those of foreground stars (Figs. \[fig. b-r\_j-k.ps\] and \[fig. B-R vs R\]), thus this source most likely belongs to the Milky Way. It has been detected in observations 1 and 2 in all three EPIC cameras. In all cases, source No.24 shows hard HR$_2$ (Fig. \[fig. log10-fx-fopt\], left panel), inconsistent with the expected X-ray spectra of foreground stars. The properties of the optical companion and the hard X-ray spectra may indicate a cataclysmic-variable nature for this source. This class of sources can show short- and long-term time variability, therefore we produced the X-ray lightcurve in the energy range $0.5-4.5$ keV to give more evidence for this identification. However, the resulting X-ray lightcurve (with a bin-time of 2000 s) shows neither short- nor long-term variability. Sources that are not foreground stars {#sect. Sources which are not foreground stars} ------------------------------------- #### **Sources No.12, 137, 164, and 189** coincide with ROSAT sources H2, H31, H34 and H36. They were classified by @Immler99 as foreground stars based on positional coincidences with optical sources of the *APM Northern Sky Catalogue* [@Irwin94]. We found possible optical counterparts in the USNO-B1 catalogue for source No.164 (USNO$-$B1$0601-0299090$) and source No.12 (USNO$-$B1$0602-0301227$). However, their X-ray-to-optical flux ratios (equation \[eq. Maccacaro\]) are $\log(f_{\rm x}/f_{\rm opt}) \approx 0.10$ and $0.11$ respectively ($f_{\rm opt}$ of both sources was calculated using visual magnitude), hence the foreground star classification for these sources is ruled out. The refined positions of sources Nos.137 and 189 obtained with XMM-*Newton*, allowed us to exclude their association with the optical counterparts proposed by @Immler99. Source No.189 can be associated with a new optical counterpart, USNO$-$B1$0600-0300832$, which is $\sim 3$ orders of magnitude fainter than the previous one (USNO$-$B1$0600-0300831$). However, the new X-ray-to-optical flux ratio is $\log(f_{\rm x}/f_{\rm opt}) \approx 0.68$ ($f_{\rm opt}$ was calculated using visual magnitude), too high for a foreground star (see Sect. \[sect. Discussion Background objects\]). Hardness ratios of sources No.164 and 137 are consistent with a powerlaw or disk-blackbody spectrum. Therefore, the spectra of these sources are too hard to be classified as foreground stars. Background objects {#sect. Discussion Background objects} ------------------ We found radio counterparts of the sources No.20, 37, and 189 and classified them as AGN candidates for the first time. #### **Source No.20** is located outside the $D_{25}$ ellipse ($D_{25}=11.5^\prime$; @Tully88) at $\sim 0.41^\circ$ from the centre of the galaxy. It coincides with the radio source NVSSJ133618$-$301459. We detected this source with XMM-*Newton* in observations 2 and 3 in the outer disc of M83. Source No.20 shows a significant long-term variability (Table \[Tab. variability\]), and the hardness ratios are roughly consistent with a spectrum described by an APEC model with a temperature of $kT_{\rm apec} \sim 0.5$ keV ($HR_2=-0.2 \pm 0.1$; $HR_3=-0.81 \pm 0.11$). Therefore, source No.20 can be classified as an AGN candidate (with a soft spectral component) or an SNR candidate. The distance of this source from the nuclear region of M83 of $\sim32$ kpc rather indicates that source No.20 does not belong to the galaxy, therefore it is more likely an AGN than an SNR candidate. #### **Sources No.37 and 189** coincide with the radio sources NVSSJ133630$-$301651 and NVSSJ133805$-$295748, respectively. Source No.189 was previously classified as a foreground star by @Immler99 (see Sect. \[sect. Sources which are not foreground stars\]). We detected these sources with XMM-*Newton* in observation 3. Their hardness ratios are consistent with a spectrum described with a powerlaw or disc-blackbody model (No.37: $HR_2=0.62 \pm 0.12$; $HR_3=-0.37 \pm 0.13$; No.189: $HR_2=0.07 \pm 0.10$; $HR_3=-0.25 \pm 0.12$). Therefore, they can be classified as AGN candidates. X-ray binaries {#sect. class. X-ray binaries} -------------- #### **Source No.81** coincides with the *Chandra* source \[SW03\]33, classified as an accreting X-ray pulsar, with a hard spectrum ($\Gamma \approx 1.7$) and a spin period of 174.9 s. We observed source No.81 in all XMM-*Newton* observations. The hardness ratios are consistent with an absorbed powerlaw spectrum, and this source shows a significant long-term X-ray variability ($V_{\rm f}=2.5$, $S=3.0$, Table \[Tab. variability\]). We applied a Fourier transform periodicity search and a Z$_n^2$ analysis (section \[sect. Variability of the sources\]), which did not reveal any significant periodicity. We calculated the upper-limit on the pulsed fraction (defined as the semi-amplitude of the sinusoidal modulation divided by the mean count rate) using the procedure described by @Vaughan94. The upper limit on the pulsed fraction obtained from the combined PN and MOS events of observation 1 is 16% at the 99% confidence level. This upper limit is marginally compatible with the pulsed fraction of $(50 \pm 15)$% of source \[SW03\]33. #### **Source No.120** corresponds to the X-ray source \[SW03\]113. Using the spectral properties and the 201.5 s periodicity detected with *Chandra*, SW03 classified source \[SW03\]113 as an XRB in a soft state. We observed source No.120 with XMM-*Newton* in observations 1 and 3. The hardness ratios of this source are consistent with an absorbed powerlaw spectrum with $N_{\rm H} \sim 5 \times 10^ {21}$ cm$^{-2}$ and $\Gamma \sim 1.5$. Similarly to source No.81, a Fourier transform periodicity search and a $Z_n^2$ analysis did not reveal any significant periodicity. At the 99% confidence level, the upper limit on the pulsed-fraction of source No.120 derived from the MOS events is 49%. This upper limit is compatible with the $(50 \pm 19)$% pulsed fraction of \[SW03\]113. Supernova remnant candidates {#sect. class. Supernova remnants} ---------------------------- #### **Source No.79** The position of this source corresponds to the position of the ROSAT source H15 [@Immler99] and the *Chandra* source \[SW03\]27. The *Chandra* spectrum shows emission lines, suggesting the possibility of emission from optically thin thermal plasma, and has been fitted by SW03 with an absorbed powerlaw with $\Gamma \sim 1.4$ and $N_{\rm H}\sim 7 \times 10^{20}$ cm$^{-2}$. SW03 classified this source as a young SNR candidate. Another possible explanation for the hard powerlaw spectrum with superposition of emission lines of \[SW03\]27 is that the source is an XRB surrounded by a photoionised nebula (SW03). However, XRBs showing these spectral properties usually have a higher absorbing column density than that of \[SW03\]27 (see e.g. @Sako99). The XMM-*Newton* hardness ratios of source No.79 below 2 keV are consistent with an APEC model with temperature $kT_{\rm apec} \gtrsim 1.5$ keV, while at higher energies the hardness ratios are consistent with a powerlaw with photon index $\sim 2$. The spectral shape of source No.79 derived from XMM-*Newton* hardness-ratio diagrams agrees with the X-ray spectrum of \[SW03\]27 presented by SW03 (see Figure 6 in SW03) and can be interpreted as an SNR exhibiting both a thin-thermal emission (below $\sim 2$ keV) and an additional hard component, which dominates at energies above $\sim 2$ keV. Also, source No.79 does not show any significant long-term variability (see Table \[Tab. variability\]). Super-soft source candidates {#sect. class. Super-Soft Sources} ---------------------------- ![Hardness-ratio diagram of source No.91 observed with XMM-*Newton*. Thick lines are different spectral models as function of the $N_{\rm H}$, thin lines are different column densities $N_{\rm H}$ (from left to right: $10^{20}$, $10^{21}$, $10^{22}$ cm$^{-2}$) as a function of the spectral parameters.[]{data-label="fig. col-col source 93"}](91_hr0-vs-hr1.ps){width="8cm"} #### **Source No.91** coincides with *Einstein* source 3 [@Trinchieri85] and *Chandra* source \[SW03\]55 classified by @DiStefano03 as an SSS candidate (source M83-50 in @DiStefano03). @DiStefano03 fitted the X-ray spectrum of M83-50 with an absorbed blackbody with a temperature of $kT_{\rm bb}= 66{+13 \atop -24}$ eV, a column density of $N_{\rm H}=2.4{+7.4 \atop -2.4} \times 10^{20}$ cm$^{-2}$, and a luminosity of $L_{\rm x}=2.8 \times 10^{37}$ erg s$^{-1}$ ($0.3-7$ keV, $d=4.5$ Mpc). We detected source No.91 in observation 1, where the hardness ratios are consistent with a blackbody spectrum (with column density in the range $\approx 10^{20}-10^{21}$ cm$^{-2}$) and marginally compatible with an APEC spectrum with temperature in the range $\approx 0.2-0.5$ keV (Fig. \[fig. col-col source 93\]). Source No.91 has a $0.2-4.5$ keV luminosity of $L_{\rm x}=(2.2 \pm 0.2)\times 10^{37}$ erg s$^{-1}$ and does not show any significant variability compared to the *Chandra* observation. Ultra-luminous X-ray sources {#sect. class. Ultra-Luminous X-ray sources} ---------------------------- obs. 3 ---------------------------------------------------- ---------------------------------------- ---------------------------------------- --------------------------------------- model [powerlaw + diskbb]{} [bbody + diskbb]{} [powerlaw]{} $N_{\rm H}$ ($10^{22}$ cm$^{-2}$) $0.30{+0.12 \atop -0.09}$ $\leq 0.03$ $0.12 \pm 0.05$ $\Gamma$ or $kT_{\rm bb}$ (keV) $2.64{+0.19 \atop -0.18}$ $0.30{+0.02 \atop -0.04}$ $2.6 {+0.3 \atop -0.2}$ norm. $4.4{+1.0 \atop -0.4}\times 10^{-4}$ $3.6{+0.8 \atop -0.9}\times 10^{-6}$ $1.4{+0.3 \atop -0.3}\times 10^{-4}$ $kT_{\rm in}$ (keV) $0.09{+0.02 \atop -0.03}$ $1.4{+0.3 \atop -0.2}$ norm. $4.8{+47.4 \atop -4.4}\times 10^3$ $7.8{+9.7 \atop -3.1}\times 10^{-3}$ $\hat{\chi}^2_\nu$ (d.o.f.) 0.946 (184) 0.994 (184) 0.918 (62) $F_{\rm x}$ ($0.2-12$ keV, erg cm$^{-2}$ s$^{-1}$) $8.9{+57.5 \atop -3.7}\times 10^{-13}$ $8.5{+24.5 \atop -5.2}\times 10^{-13}$ $3.7{+2.2 \atop -1.4}\times 10^{-13}$ $L_{\rm x}$ ($d=4.5$ Mpc, erg s$^{-1}$) $1.3{+20.4 \atop -0.6}\times 10^{40}$ $2.2{+6.3 \atop -1.3}\times 10^{39}$ $2.2{+0.9 \atop -0.5}\times 10^{39}$ Two ULXs have been discovered in M83: H2 [@Trinchieri85], and a transient ULX discovered with *Chandra* on 23 December 2010 with a luminosity of $L_{\rm x} \sim 4 \times 10^{39}$ erg s$^{-1}$ ($0.3-10$ keV) by @Soria10, and classified as an accretion-powered black hole with mass $M_{\rm BH} \approx 40-100$ M$_\odot$ [@Soria12]. This ULX has not been detected in the XMM-*Newton* data. @Soria12 measured an upper limit to the X-ray luminosity of $\sim 10^{37}$ erg s$^{-1}$ ($0.3-10$ keV) from the three XMM-*Newton* observations. #### **Source No.133** We observed the ULX as source No.133 in all XMM-*Newton* observations. @Ehle98 and @Immler99 found a faint extended optical source within the error circle of the ROSAT source position. @Roberts08 used HST images in three Advanced Camera for Survey (ACS) filters to find the counterparts to six ULXs in different galaxies. For the ULX in M83, they compared the optical position with the X-ray position from a *Chandra* High Resolution Camera for Imaging (HRC-I) observation. They detected a counterpart to the ULX with magnitudes $B=25.66\pm 0.13$, $V=25.36\pm 0.17$. They also noticed that the ULX is located at $\sim 5^{\prime\prime}$ from the centre of a background galaxy, and although the latter is outside the error circle, @Roberts08 did not completely rule out a possible association between the ULX and the background galaxy. @Stobbart06 reported the XMM-*Newton* spectral analysis of source No.133 during observation 1. They found that the X-ray spectrum is well fitted with a cool disc-blackbody ($kT_{\rm in} \sim 0.2$ keV) plus a powerlaw ($\Gamma \sim 2.5$), or with a cool blackbody ($kT_{\rm bb} \sim 0.2$ keV) plus a warm disc-blackbody ($kT_{\rm in} \sim 1.1$ keV). The first spectral model is the standard IMBH model, where the low disc temperature is due to a black hole with mass of $\sim 1000$ M$_\odot$, while the origin of the powerlaw component is still not clear (see @Roberts05). Instead, the spectral parameters obtained with the second spectral model suggest that No.133 is a stellar-mass black hole accreting close to the Eddington limit. In this model, the cool blackbody component represents the optically thick wind from the stellar-mass black-hole accreting at or above the Eddington limit, while the high temperature of the disc follows the standard trend $L_{\rm x} \propto T^4$ shown by the Galactic stellar-mass black-hole binaries. We analysed all XMM-*Newton* observations of the ULX No.133 and fitted the PN, MOS1 and MOS2 spectra simultaneously with a model assuming an IMBH ([phabs\*\[diskbb + powerlaw\]]{} in XSPEC), and a model assuming a stellar-mass BH ([phabs\*\[bbody + diskbb\]]{}). We used two absorption components: the Galactic absorption column density ($N_{\rm H}=3.69\times 10^{20}$ cm$^{-2}$) and the absorption within M83 plus the intrinsic column density of the ULX. In all fits we obtained a good fit with both spectral models with the resulting spectral parameters in agreement with those obtained by @Stobbart06 from observation 1. However, the spectral parameters in observation 3 are only poorly constrained due to the poor statistics (only MOS1 and MOS2 data were available for this observation). Therefore, we fitted the spectrum of observation 3 with a single component model and found that an absorbed powerlaw can adequately fit the data (Fig. \[fig. ulx-obs2\], Table \[Tab. ulx spectral parameters\]). Hard sources {#section Hard sources} ------------ ### New classifications {#section Newly classified hard sources} Source -------- --------------------------- ------------------------ -------------------------------------- ----------------------------- ---------------------------------------- -------------------------------------- ------ -------------- $N_{\rm H}$ $\Gamma$ norm. $\hat{\chi}^2_\nu$ (d.o.f.) $F_{\rm x}$ $L_{\rm x}$ obs. instrument ($10^{21}$ cm$^{-2}$) (erg cm$^{-2}$ s$^{-1}$) (erg s$^{-1}$) 16 $0.8{+0.4 \atop -0.3}$ $2.6{+0.3 \atop -0.3}$ $4.1{+0.9 \atop -0.7}\times 10^{-5}$ $0.928$ $(38)$ $1.4{+0.8 \atop -0.5}\times 10^{-13}$ $6.3{+2.7 \atop -1.3}\times 10^{38}$ 1 PN,MOS1,MOS2 61 $2.01{+0.75 \atop -0.65}$ $2.4{+0.3 \atop -0.3}$ $2.9{+0.9 \atop -0.6}\times 10^{-5}$ $1.04$ $(36)$ $8.0{+6.8 \atop -3.7}\times 10^{-14}$ $4.3{+1.4 \atop -0.6}\times 10^{38}$ 1 PN,MOS1,MOS2 103 $0.7{+1.0 \atop -0.7}$ $1.8{+0.4 \atop -0.4}$ $3.7{+1.8 \atop -1.2}\times 10^{-5}$ $0.876$ $(19)$ $2.2{+3.3 \atop -1.3}\times 10^{-13}$ $6.6{+6.7 \atop -2.7}\times 10^{38}$ 1 PN,MOS2 126 $0.01{+0.56 \atop -0.01}$ $1.8{+0.4 \atop -0.2}$ $8.8{+2.7 \atop -1.2}\times 10^{-6}$ $0.743$ $(17)$ $6.3{+3.1 \atop -2.9}\times 10^{-14}$ $1.5{+0.8 \atop -0.4}\times 10^{38}$ 2 PN,MOS1,MOS2 153 $0 < N_{\rm H} \leq 1.5$ $1.4{+0.7 \atop -0.3}$ $7.6{+5.8 \atop -1.6}\times 10^{-6}$ $1.027$ $(15)$ $8.8{+13.7 \atop -6.2}\times 10^{-14}$ $2.1{+3.3 \atop -1.2}\times 10^{38}$ 1 PN,MOS1,MOS2 #### **Source No.16** coincides with the ROSAT source H3 discovered by @Immler99. This source is located outside the optical disc of M83, and its position overlaps with the outer disc of M83 observed by GALEX (e.g. @Thilker05). We detected source No.16 in all XMM-*Newton* observations, but only in observation 1 was it bright enough to allow spectral analysis. The spectrum can be well fitted with an absorbed powerlaw with $\Gamma = 2.6{+0.3 \atop -0.3}$, compatible with that of an XRB or an AGN (see Table \[Tab. spectral parameters xrbs\]). Source No.16 shows a significant long-term variability ($S=9.5$) with a variability factor of $V_{\rm f}=10.6 \pm 0.3$ (Table \[Tab. variability\]). It also shows a significant variability within observation 1, with a variability factor of $V_{\rm f}=6.6 \pm 4.5$ and significance $S=4.0$. #### **Source No.61** is in the field of view of XMM-*Newton* during observation 1, where it shows an X-ray luminosity of $L_{\rm x} \approx 4\times 10^{38}$ erg s$^{-1}$ (see Table \[Tab. spectral parameters xrbs\]). It has not been previously detected in X-ray, optical, radio, infrared, or UV. The X-ray spectrum is well fitted with an absorbed powerlaw with $\Gamma = 2.4{+0.3 \atop -0.3}$ or a disc-blackbody model with temperature $kT_{\rm in}=0.82{+0.13 \atop -0.11}$ keV (Table \[Tab. spectral parameters xrbs\]). Source No.61 shows a significant long-term variability ($S=8.6$) with a variability factor of $V_{\rm f}=4.3 \pm 0.1$ (Table \[Tab. variability\]). #### **Source No.103** is located at a distance of $\sim 6^{\prime\prime}$ from a radio source (6 in @Cowan94, 36 in @Maddox06), and at $1.6^{\prime\prime}$ from the *Chandra* source \[SW03\]84, which shows hardness ratios compatible with a powerlaw or a disc-blackbody spectrum. We detected source No.103 only in the XMM-*Newton* observation 2, with a flux of $(2.23{+3.26 \atop -1.34})\times 10^{-13}$ erg cm$^{-2}$ s$^{-1}$ ($0.2-12$ keV). The X-ray spectrum is well fitted with an absorbed powerlaw with $\Gamma =1.8{+0.4 \atop -0.4}$ (Table \[Tab. spectral parameters xrbs\]). We did not detect source No.103 in observations 1 and 3, thus we calculated the flux upper-limits and we found a significant ($S = 7.2$) long-term variability, with a variability factor of $V_{\rm f}=12.78\pm0.12$ (Table \[Tab. variability\], Fig. \[fig. variab\]). #### **Source No.126** coincides with X-ray source 30 [@Ehle98] discovered with ROSAT. Source No.126 also cross-correlates with the optical counterpart USNO-B1$0599-0300335$, but the ratio $\log_{10}(F_{\rm x}/F_{\rm opt})$ does not match the criteria previously specified to classify foreground stars. Source No.126 is located outside the optical disc of M83, and its position overlaps with an extended arm of the galaxy. We observed source No.126 in all XMM-*Newton* observations. The X-ray spectra extracted from each observation can be well fitted with an absorbed powerlaw with $\Gamma \approx 1.8$ and the flux is consistent with that measured by @Ehle98 (Table \[Tab. spectral parameters xrbs\]). #### **Source No.153** is detected in all XMM-*Newton* observations, and has not been previously detected in X-rays, optical, radio, infrared, or UV bands. It is located in the extended arms observed by *GALEX*, $\approx 10^\prime$ away from the nuclear region of M83. The spectra extracted from each observation can be well fitted with an absorbed powerlaw with $\Gamma \approx 1.5$, suggesting an XRB nature for this source (see Table \[Tab. spectral parameters xrbs\]). ### Identifications {#section identification hard sources} Source -------- ------------------------ ------------------------ -------------------------------------- ----------------------------- ---------------------------------------- --------------------------------------- ------ -------------- $N_{\rm H}$ $\Gamma$ norm. $\hat{\chi}^2_\nu$ (d.o.f.) $F_{\rm x}$ $L_{\rm x}$ obs. instrument ($10^{21}$ cm$^{-2}$) (erg cm$^{-2}$ s$^{-1}$) (erg s$^{-1}$) 97 $4.4{+1.1 \atop -0.9}$ $2.4{+0.3 \atop -0.3}$ $7.4{+2.3 \atop -1.7}\times 10^{-5}$ $1.03$ $(45)$ $1.7{+1.5 \atop -0.8}\times 10^{-13}$ $1.1{+0.4 \atop -0.2}\times 10^{39}$ 1 PN,MOS2 106 $0.3{+0.5 \atop -0.3}$ $1.8{+0.4 \atop -0.3}$ $8.3{+2.5 \atop -1.9}\times 10^{-6}$ $1.38$ $(27)$ $5.4{+5.0 \atop -2.7}\times 10^{-14}$ $1.5{+1.0 \atop -0.5}\times 10^{38}$ 1 PN,MOS1,MOS2 107 $3.3{+1.0 \atop -0.8}$ $2.8{+0.4 \atop -0.3}$ $3.9{+1.4 \atop -1.0}\times 10^{-5}$ $0.821$ $(43)$ $6.5{+6.5 \atop -3.3}\times 10^{-14}$ $6.6{+5.1 \atop -2.3}\times 10^{38}$ 1 PN,MOS1,MOS2 108 $3.3{+0.7 \atop -0.6}$ $2.7{+0.3 \atop -0.2}$ $8.0{+1.9 \atop -1.5}\times 10^{-5}$ $0.91$ $(68)$ $1.5{+0.9 \atop -0.6}\times 10^{-13}$ $1.3{+0.5 \atop -0.3}\times 10^{39}$ 1 PN,MOS1 114 $4.5{+3.8 \atop -3.1}$ $1.7{+0.4 \atop -0.4}$ $1.8{+1.3 \atop -0.7}\times 10^{-5}$ $0.952$ $(20)$ $9.6{+24.0 \atop -6.9}\times 10^{-14}$ $3.5{+5.68 \atop -1.8}\times 10^{38}$ 1 PN,MOS1,MOS2 129 $5.4{+4.3 \atop -3.0}$ $2.0{+0.6 \atop -0.5}$ $2.0{+1.9 \atop -0.9}\times 10^{-5}$ $0.867$ $(18)$ $6.7{+20.9 \atop -5.1}\times 10^{-14}$ $3.1{+5.4 \atop -1.4}\times 10^{38}$ 1 PN,MOS1 #### **Source No.60** correlates with the *Chandra* source \[SW03\]5 SW03 suggested that this source is an XRB candidate. We observed source No.60 in all XMM-*Newton* observations. The source shows a significant long-term variability ($V_{\rm f}=2.4$, $S=3.8$, Table \[Tab. variability\]) with respect to the *Chandra* observation. X-ray colours of No.60 are consistent with a powerlaw or disc-blackbody spectrum, in agreement with the spectral analysis of SW03. #### **Source No.80** correlates with the *Chandra* source \[SW03\]31. From the spectral properties, SW03 suggested that \[SW03\]31 is an XRB candidate. We observed source No.80 with XMM-*Newton* in observation 1. The hardness ratios are consistent with a powerlaw or disc-blackbody spectrum with column density of $\sim 10^{21}$ cm$^{-2}$ #### **Source No.92** coincides with the *Chandra* source \[SW03\]60. SW03 suggested that No.92 is a XRB candidate because of its hard spectrum ($\Gamma \sim 1.6$). We observed source No.92 with XMM-*Newton* in observation 1. The hardness ratios are consistent with a spectrum described by an absorbed powerlaw model with $\Gamma \sim 2$. Source No.92 also shows a high long-term variability by a factor of $V_{\rm f}=2.7$, with a variability significance of $S=4.3$ (see Table \[Tab. variability\]). #### **Source No.97** coincides with the *Chandra* source \[SW03\]72 and with a ROSAT source (source 7 in @Ehle98 and source H20 in @Immler99). We observed source No.97 in all XMM-*Newton* observations. The spectra extracted from each observation can be well fitted with an absorbed powerlaw or a disc-blackbody model (Table \[Tab. spectral parameters xrbs appendix\]), with spectral parameters in agreement with the spectral analysis of SW03. Source No.97 shows a significant long-term variability between XMM-*Newton* and *Chandra* observations ($V_{\rm f}=2.8 \pm 0.1$, $S=6.6$; Table \[Tab. variability\]). Within observation 1 we found a variability of $V_{\rm f}=6.4 \pm 2.7$ with a significance of $S=4.8$. #### **Source No.99** coincides with the *Chandra* source \[SW03\]73, and it is associated with the radio source MCK34 (@Maddox06). located in a HII region (RK137, @Rumstay83). From a spectral study, SW03 proposed that \[SW03\]73 is more likely an XRB than a young SNR. We observed source No.99 with XMM-*Newton* in observations 2 and 3. The source shows a significant variability ($S=5.0$), with a variability factor of $V_{\rm f}=15.3$ (Table \[Tab. variability\]). and the hardness ratios are consistent with an absorbed powerlaw or disk-blackbody spectrum. #### **Source No.106** corresponds to the X-ray source H25 observed by @Immler99 in a ROSAT observation and the *Chandra* source \[SW03\]85. We observed source No.106 in all the observations. During observation 1 the source was bright enough to allow spectral analysis. The spectrum can be well fitted with an absorbed powerlaw (see Table \[Tab. spectral parameters xrbs appendix\]), with spectral parameters in agreement with those previously obtained by SW03. #### **Source No.107** was detected by @Ehle98 (source 9) and @Immler99 (source H26) in ROSAT (PSPC and HRI) observations. @Immler99 found that H26 coincides with a compact radio source (source 8 in @Cowan94), and with a giant HII region [@Rumstay83]. Hence, they classified this source as an SNR candidate. Moreover, also the observation of H$\alpha$ and H$\beta$ emission anti-coincident with HI emission [@Tilanus93] supports the SNR hypothesis. Source No.107 was also observed in 2000 April 29 by *Chandra* (source \[SW03\]86). From a spectral analysis, SW03 proposed that No.110 is more likely an XRB (BH candidate) than an SNR. We detected source No.107 in all XMM-*Newton* observations with a luminosity of $\sim 7 \times 10^{38}$ erg s$^{-1}$. In observations 1 and 3 the source was bright enough to allow spectral analysis. The spectra can be well fitted with an absorbed powerlaw or a disc-blackbody (see Table \[Tab. spectral parameters xrbs appendix\]). The obtained spectral parameters are consistent with those previously found by SW03 with *Chandra*. Source No.107 shows a significant long-term variability between XMM-*Newton* observations ($V_{\rm f}=1.97 \pm 0.12$, $S=5.1$). #### **Source No.108** was first detected in X-rays by @Trinchieri85 (source 4) with the *Einstein* satellite and by @Ehle98 (source 8) and @Immler99 (source H27) with ROSAT. It also coincides with the *Chandra* source \[SW03\]88. We observed source No.108 in all XMM-*Newton* observations. During observation 1 source No.108 was in the centre of the field of view, providing enough statistic to allow spectral analysis. We extracted the PN and MOS1 spectra (the position of source No.108 was in a gap of MOS2) and we found that an absorbed powerlaw or an absorbed disc-blackbody provide acceptable fits (Table \[Tab. spectral parameters xrbs appendix\]), with spectral parameters consistents with those obtained by SW03. Source No.108 shows a significant long-term X-ray variability ($V_{\rm f}=1.4 \pm 0.1$, $S=3.3$ Table \[Tab. variability\]), and during observation 1 we found a variability of $V_{\rm f} = 4.2 \pm 1.5$, with a significance of $S=4.4$. #### **Source No.114** coincides with the *Chandra* source \[SW03\]104. We observed source No.114 in all XMM-*Newton* observations. During observation 1 source No.114 was in the centre of the field of view, providing enough statistics to allow a spectral analysis. The spectrum is well fitted with an absorbed powerlaw with spectral parameters consistent with those found by SW03 with *Chandra* (see Table \[Tab. spectral parameters xrbs appendix\]). Source No.114 also shows a significant long-term variability ($V_{\rm f} = 2.4 \pm 0.2$, $S=4.1$). #### **Source No.116** coincides with \[SW03\]105. @DiStefano03 observed \[SW03\]105 with *Chandra* (source M83-88 in @DiStefano03) and classified it as an SSS candidate. @Blair04 compared the list of *Chandra* sources of SW03 with a list of optical SNR candidates and associated \[SW03\]105 with the optical SNR candidate BL53. We observed source No.116 with XMM-*Newton* in observation 1, where it shows a significant X-ray variability ($V_{\rm f}=20.8$, $S=6.2$, Table \[Tab. variability\]) compared to the *Chandra* observation, and the X-ray hardness ratios are consistent with a hard spectrum. These properties indicate that source No.116 is most likely an XRB. We overlaid the $3\sigma$ error circles of source No.116, \[SW03\]105, and BL53 on the emission line images H$\alpha$ and SII obtained from the public *Wide Field Camera 3 (WFC3)* observation of 2009-08-20 (Fig. \[fig. HST/WFC3 source 119\]). H$\alpha$ and SII images are used in extragalactic searches of SNRs because their optical spectra show high \[SII\]:H$\alpha$ ratios compared to the spectra of normal HII regions (see e.g. @Blair04). Fig. \[fig. HST/WFC3 source 119\] shows that the shell of the optical SNR is located only in the error circles of BL53 and \[SW03\]105, indicating that source No.116 and \[SW03\]105 cannot be the same source. Therefore, source No.116 is more likely a transient source not associated with BL53. ![Emission line *HST*/WFC3 images of the region surrounding source No.116. *Left panel:* WFC3 image with the narrowband filter F657N, corresponding to H$\alpha$ line emission. *Right panel:* WFC3 image with the narrowband filter F673N, corresponding to SII line emission. The radii of the circles of *Chandra* (SW105) and XMM-*Newton* (116) sources give the 3$\sigma$ accuracy of the position of the sources. The circle labelled BL53 gives the position of the opitcal SNR candidate.[]{data-label="fig. HST/WFC3 source 119"}](f657n_histo_smooth_y.ps "fig:"){width="4.4cm"} ![Emission line *HST*/WFC3 images of the region surrounding source No.116. *Left panel:* WFC3 image with the narrowband filter F657N, corresponding to H$\alpha$ line emission. *Right panel:* WFC3 image with the narrowband filter F673N, corresponding to SII line emission. The radii of the circles of *Chandra* (SW105) and XMM-*Newton* (116) sources give the 3$\sigma$ accuracy of the position of the sources. The circle labelled BL53 gives the position of the opitcal SNR candidate.[]{data-label="fig. HST/WFC3 source 119"}](f673n_histosmooth_y.ps "fig:"){width="4.4cm"} #### **Source No.129** coincides with the *Chandra* source \[SW03\]121 and with a ROSAT source (source 12 in the catalogue of @Ehle98, source H29 in the catalogue of @Immler99). We observed source No.129 during observation 1, where it was in the XMM-*Newton* field of view. The spectrum is well fitted with an absorbed powerlaw with spectral parameters consistent with those found by SW03 with *Chandra* (see Table \[Tab. spectral parameters xrbs appendix\]). We did not detect source No.129 in observations 2 and 3, thus we calculated the flux upper limits and we found a significant ($S\gtrsim 6.4$) long-term variability with a variability factor of $V_{\rm f}=3.94\pm0.11$ (Table \[Tab. variability\]). X-ray source catalogue of the XMM-*Newton* EPIC M83 observation {#sect. catalogue-table} ===============================================================   [lccccccccccccc]{} \ No. & RA(2000) & DEC(2000) & pos. err. & rate (obs1) & rate (obs2) & rate (obs3) & & HR1 & HR2 & HR3 & HR4\ & & & & & & & (obs1) & (obs2) & (obs3) & & & &\ \ No. & RA(2000) & DEC(2000) & pos. err. & rate (obs1) & rate (obs2) & rate (obs3) & & HR1 & HR2 & HR3 & HR4\ & & & & & & & (obs1) & (obs2) & (obs3) & & & &\ \ 1 & 13 35 48.83 & -29 56 39.5 & $ 1.57^{\prime\prime}$ & $0.012 \pm 0.002$ & & & $ 33.2$ & & & $ -0.07 \pm 0.21$ & $ -0.06 \pm 0.23$ & $ -0.43 \pm 0.34$ & $ -0.58 \pm 1.11$\ 2 & 13 35 49.82 & -30 00 38.4 & $ 2.94^{\prime\prime}$ & & $0.016 \pm 0.004$ & & & $ 21.2$ & & $ 0.06 \pm 0.31$ & $ 0.27 \pm 0.25$ & $ -0.60 \pm 0.32$ & $ 0.50 \pm 0.44$\ 3 & 13 35 52.30 & -30 02 59.8 & $ 1.72^{\prime\prime}$ & & $0.019 \pm 0.004$ & $0.0048 \pm 0.0008$ & & $ 28.2$ & $ 57.0$ & $ 1.00 \pm 0.08$ & $ 0.06 \pm 0.17$ & $ -0.64 \pm 0.20$ & $ -1.00 \pm 0.76$\ 4 & 13 35 55.46 & -30 13 23.3 & $ 1.96^{\prime\prime}$ & & $0.013 \pm 0.004$ & & & $ 16.2$ & & $ 0.22 \pm 0.28$ & $ -0.79 \pm 0.31$ & $ 0.64 \pm 0.50$ & $ 0.41 \pm 0.40$\ 5 & 13 35 59.23 & -29 59 05.5 & $ 2.19^{\prime\prime}$ & $0.010 \pm 0.003$ & & & $ 6.7$ & & & $ 1.00 \pm 0.25$ & $ 0.13 \pm 0.41$ & $ 0.15 \pm 0.37$ & $ 0.39 \pm 0.31$\ 6 & 13 36 03.03 & -30 03 21.3 & $ 1.37^{\prime\prime}$ & & & $0.0067 \pm 0.0013$ & & & $ 25.5$ & $ 0.45 \pm 0.31$ & $ 0.36 \pm 0.19$ & $ -0.48 \pm 0.20$ & $ -0.50 \pm 0.61$\ 7 & 13 36 04.66 & -30 08 30.8 & $ 0.77^{\prime\prime}$ & & $0.036 \pm 0.004$ & $0.080 \pm 0.004$ & & $ 158.6$ & $1247.9$ & $ 0.52 \pm 0.06$ & $ 0.17 \pm 0.05$ & $ -0.27 \pm 0.05$ & $ -0.40 \pm 0.10$\ 8 & 13 36 05.29 & -29 52 52.8 & $ 2.13^{\prime\prime}$ & $0.006 \pm 0.002$ & & & $ 9.1$ & & & $ 0.28 \pm 0.31$ & $ -0.28 \pm 0.27$ & $ -0.32 \pm 0.48$ & $ -0.15 \pm 1.08$\ 9 & 13 36 05.95 & -29 52 00.8 & $ 1.30^{\prime\prime}$ & $0.012 \pm 0.002$ & & & $ 30.3$ & & & $ 0.11 \pm 0.22$ & $ 0.10 \pm 0.20$ & $ -0.28 \pm 0.23$ & $ -0.59 \pm 0.61$\ 10 & 13 36 06.77 & -30 06 51.7 & $ 1.60^{\prime\prime}$ & & $0.0078 \pm 0.0018$ & & & $ 13.8$ & & $ 1.00 \pm 0.56$ & $ 0.44 \pm 0.21$ & $ -0.62 \pm 0.22$ & $ -1.00 \pm 1.04$\ 11 & 13 36 10.66 & -29 54 03.1 & $ 1.65^{\prime\prime}$ & & $0.0097 \pm 0.0029$ & & & $ 14.0$ & & $ 0.27 \pm 0.29$ & $ -0.10 \pm 0.29$ & $ -0.47 \pm 0.42$ & $ 0.25 \pm 0.80$\ 12 & 13 36 11.72 & -29 43 36.6 & $ 0.47^{\prime\prime}$ & $0.058 \pm 0.004$ & & & $ 414.4$ & & & $ 0.17 \pm 0.08$ & $ -0.12 \pm 0.08$ & $ -0.48 \pm 0.11$ & $ -0.26 \pm 0.31$\ 13 & 13 36 12.43 & -30 04 24.1 & $ 1.10^{\prime\prime}$ & & & $0.0064 \pm 0.0012$ & & & $ 23.0$ & $ 0.51 \pm 0.21$ & $ -0.50 \pm 0.22$ & $ -0.29 \pm 0.43$ & $ 0.64 \pm 0.26$\ 14 & 13 36 12.71 & -29 58 37.8 & $ 1.74^{\prime\prime}$ & $0.015 \pm 0.004$ & & $0.0069 \pm 0.0014$ & $ 9.3$ & & $ 16.1$ & $ 0.66 \pm 0.31$ & $ 0.18 \pm 0.21$ & $ -0.45 \pm 0.22$ & $ 0.02 \pm 0.47$\ 15 & 13 36 13.25 & -29 59 40.4 & $ 1.26^{\prime\prime}$ & $0.013 \pm 0.002$ & $0.015 \pm 0.002$ & $0.0182 \pm 0.0018$ & $ 35.6$ & $ 55.8$ & $ 138.5$ & $ 0.12 \pm 0.13$ & $ -0.03 \pm 0.12$ & $ -0.35 \pm 0.14$ & $ 0.21 \pm 0.20$\ 16 & 13 36 13.88 & -29 56 13.4 & $ 0.32^{\prime\prime}$ & $0.105 \pm 0.004$ & $0.010 \pm 0.002$ & $0.0043 \pm 0.0009$ & $1522.4$ & $ 15.0$ & $ 21.7$ & $ 0.24 \pm 0.05$ & $ -0.09 \pm 0.05$ & $ -0.58 \pm 0.05$ & $ -0.93 \pm 0.14$\ 17 & 13 36 15.42 & -29 57 58.2 & $ 0.64^{\prime\prime}$ & $0.049 \pm 0.005$ & $0.033 \pm 0.003$ & $0.028 \pm 0.002$ & $ 243.6$ & $ 171.8$ & $ 253.2$ & $ 0.44 \pm 0.11$ & $ -0.08 \pm 0.10$ & $ -0.45 \pm 0.11$ & $ -0.19 \pm 0.25$\ 18 & 13 36 17.09 & -30 13 55.4 & $ 1.76^{\prime\prime}$ & & & $0.0018 \pm 0.0006$ & & & $ 9.2$ & $ 0.38 \pm 0.42$ & $ 0.17 \pm 0.31$ & $ -1.00 \pm 0.16$ &\ 19 & 13 36 18.13 & -29 40 13.0 & $ 1.71^{\prime\prime}$ & $0.008 \pm 0.003$ & & & $ 6.2$ & & & $ -1.00 \pm 0.58$ & $ 1.00 \pm 0.33$ & $ 0.40 \pm 0.32$ & $ -0.02 \pm 0.42$\ 20 & 13 36 18.21 & -30 15 00.5 & $ 0.47^{\prime\prime}$ & & $0.053 \pm 0.005$ & $0.0062 \pm 0.0009$ & & $ 302.7$ & $ 89.2$ & $ 0.08 \pm 0.09$ & $ -0.20 \pm 0.10$ & $ -0.81 \pm 0.11$ & $ -0.06 \pm 0.83$\ 21 & 13 36 18.73 & -30 01 38.1 & $ 0.56^{\prime\prime}$ & $0.023 \pm 0.003$ & $0.015 \pm 0.002$ & $0.0190 \pm 0.0017$ & $ 138.5$ & $ 71.1$ & $ 222.4$ & $ 0.42 \pm 0.09$ & $ -0.36 \pm 0.09$ & $ -0.82 \pm 0.13$ & $ 0.42 \pm 0.43$\ 22 & 13 36 18.96 & -30 06 12.6 & $ 1.87^{\prime\prime}$ & & & $0.0033 \pm 0.0009$ & & & $ 7.0$ & $ 0.41 \pm 0.67$ & $ 0.52 \pm 0.31$ & $ -0.15 \pm 0.28$ & $ -0.17 \pm 0.44$\ 23 & 13 36 19.84 & -30 05 18.3 & $ 1.17^{\prime\prime}$ & & & $0.0047 \pm 0.0009$ & & & $ 15.9$ & $ 0.28 \pm 0.42$ & $ 0.37 \pm 0.24$ & $ -0.11 \pm 0.22$ & $ -0.44 \pm 0.32$\ 24 & 13 36 19.94 & -29 51 08.4 & $ 1.09^{\prime\prime}$ & $0.0101 \pm 0.0016$ & $0.008 \pm 0.003$ & & $ 40.8$ & $ 8.4$ & & $ 0.03 \pm 0.35$ & $ 0.55 \pm 0.19$ & $ -0.45 \pm 0.17$ & $ 0.11 \pm 0.31$\ 25 & 13 36 19.96 & -29 41 11.6 & $ 1.15^{\prime\prime}$ & $0.014 \pm 0.003$ & & & $ 41.7$ & & & $ 0.34 \pm 0.20$ & $ -0.06 \pm 0.20$ & $ -0.30 \pm 0.25$ & $ -0.70 \pm 0.68$\ 26 & 13 36 24.05 & -30 14 56.6 & $ 0.89^{\prime\prime}$ & & $0.019 \pm 0.003$ & & & $ 59.9$ & & $ 0.70 \pm 0.15$ & $ -0.20 \pm 0.15$ & $ -0.25 \pm 0.21$ & $ -1.00 \pm 0.55$\ 27 & 13 36 24.17 & -29 54 00.6 & $ 1.20^{\prime\prime}$ & $0.0132 \pm 0.0019$ & $0.010 \pm 0.002$ & $0.0154 \pm 0.0019$ & $ 51.9$ & $ 8.9$ & $ 71.7$ & $ 0.47 \pm 0.22$ & $ 0.31 \pm 0.14$ & $ -0.39 \pm 0.14$ & $ 0.03 \pm 0.25$\ 28 & 13 36 26.53 & -30 06 31.8 & $ 1.56^{\prime\prime}$ & & & $0.0047 \pm 0.0009$ & & & $ 19.5$ & $ 0.90 \pm 0.31$ & $ 0.50 \pm 0.19$ & $ -0.43 \pm 0.19$ & $ -0.15 \pm 0.42$\ 29 & 13 36 26.62 & -29 53 09.4 & $ 1.65^{\prime\prime}$ & & & $0.0049 \pm 0.0012$ & & & $ 9.3$ & $ 0.79 \pm 0.63$ & $ 0.08 \pm 0.30$ & $ 0.18 \pm 0.26$ & $ -0.94 \pm 0.40$\ 30 & 13 36 26.64 & -29 55 35.6 & $ 0.90^{\prime\prime}$ & $0.0108 \pm 0.0014$ & & $0.0082 \pm 0.0015$ & $ 59.3$ & & $ 34.3$ & $ 0.02 \pm 0.19$ & $ 0.28 \pm 0.16$ & $ -0.44 \pm 0.16$ & $ -0.83 \pm 0.46$\ 31 & 13 36 28.13 & -29 42 27.9 & $ 1.61^{\prime\prime}$ & $0.010 \pm 0.002$ & & & $ 9.7$ & & & $ 0.25 \pm 0.39$ & $ -0.20 \pm 0.39$ & $ -0.09 \pm 0.48$ & $ 0.73 \pm 0.17$\ 32 & 13 36 28.58 & -29 57 16.3 & $ 1.58^{\prime\prime}$ & & $0.0062 \pm 0.0017$ & & & $ 6.9$ & & $ 0.06 \pm 1.11$ & $ 0.74 \pm 0.37$ & $ 0.07 \pm 0.27$ & $ -0.01 \pm 0.36$\ 33 & 13 36 28.94 & -29 55 39.4 & $ 0.83^{\prime\prime}$ & $0.0152 \pm 0.0016$ & $0.0136 \pm 0.0021$ & $0.0121 \pm 0.0015$ & $ 112.4$ & $ 45.5$ & $ 62.2$ & $ 0.59 \pm 0.15$ & $ 0.09 \pm 0.12$ & $ -0.27 \pm 0.13$ & $ -0.42 \pm 0.27$\ 34 & 13 36 28.95 & -29 51 22.9 & $ 0.57^{\prime\prime}$ & $0.025 \pm 0.004$ & $0.022 \pm 0.004$ & $0.0070 \pm 0.0012$ & $ 197.3$ & $ 52.9$ & $ 66.3$ & $ 0.96 \pm 0.40$ & $ 0.41 \pm 0.44$ & $ 0.58 \pm 0.16$ & $ -0.04 \pm 0.16$\ 35 & 13 36 29.63 & -29 42 47.3 & $ 1.18^{\prime\prime}$ & $0.0097 \pm 0.0017$ & & & $ 32.1$ & & & $ 0.68 \pm 0.19$ & $ -0.20 \pm 0.18$ & $ -0.57 \pm 0.24$ & $ -0.14 \pm 0.72$\ 36 & 13 36 30.28 & -29 41 56.9 & $ 0.99^{\prime\prime}$ & $0.0110 \pm 0.0018$ & & & $ 31.5$ & & & $ 0.47 \pm 0.23$ & $ 0.09 \pm 0.18$ & $ -0.37 \pm 0.21$ & $ -0.51 \pm 0.49$\ 37 & 13 36 30.53 & -30 16 57.0 & $ 0.88^{\prime\prime}$ & & & $0.0095 \pm 0.0013$ & & & $ 112.1$ & $ 0.48 \pm 0.29$ & $ 0.62 \pm 0.12$ & $ -0.37 \pm 0.13$ & $ -0.51 \pm 0.33$\ 38 & 13 36 30.80 & -29 45 51.7 & $ 1.23^{\prime\prime}$ & $0.0078 \pm 0.0016$ & & & $ 18.2$ & & & $ 0.37 \pm 0.70$ & $ 0.26 \pm 0.53$ & $ 0.50 \pm 0.24$ & $ 0.16 \pm 0.20$\ 39 & 13 36 31.10 & -29 49 24.9 & $ 1.48^{\prime\prime}$ & $0.0051 \pm 0.0011$ & & & $ 12.1$ & & & $ 0.40 \pm 0.23$ & $ -0.45 \pm 0.24$ & $ -0.29 \pm 0.44$ & $ 0.20 \pm 0.56$\ 40 & 13 36 31.64 & -29 57 11.4 & $ 1.19^{\prime\prime}$ & $0.0072 \pm 0.0014$ & & $0.0061 \pm 0.0011$ & $ 18.3$ & & $ 26.4$ & $ 0.16 \pm 0.40$ & $ 0.60 \pm 0.19$ & $ -0.34 \pm 0.19$ & $ -0.08 \pm 0.35$\ 41 & 13 36 33.45 & -30 00 18.2 & $ 1.10^{\prime\prime}$ & & $0.0073 \pm 0.0013$ & $0.0036 \pm 0.0016$ & & $ 38.6$ & $ 10.2$ & $ 1.00 \pm 0.86$ & $ 0.80 \pm 0.16$ & $ -0.34 \pm 0.16$ & $ -0.40 \pm 0.34$\ 42 & 13 36 33.77 & -29 45 22.5 & $ 1.57^{\prime\prime}$ & $0.0042 \pm 0.0012$ & & & $ 6.2$ & & & $ -0.43 \pm 0.62$ & $ 0.65 \pm 0.42$ & $ -0.23 \pm 0.31$ & $ -0.04 \pm 0.50$\ 43 & 13 36 34.92 & -30 09 01.2 & $ 0.99^{\prime\prime}$ & & $0.0128 \pm 0.0019$ & $0.0037 \pm 0.0008$ & & $ 64.2$ & $ 17.9$ & $ 0.03 \pm 0.21$ & $ 0.33 \pm 0.16$ & $ -0.39 \pm 0.17$ & $ -0.37 \pm 0.49$\ 44 & 13 36 35.07 & -30 11 07.4 & $ 1.58^{\prime\prime}$ & & & $0.0028 \pm 0.0008$ & & & $ 8.2$ & $ -0.34 \pm 0.52$ & $ 0.56 \pm 0.38$ & $ -0.62 \pm 0.24$ & $ 0.52 \pm 0.37$\ 45 & 13 36 35.82 & -29 41 17.0 & $ 1.44^{\prime\prime}$ & $0.0050 \pm 0.0015$ & & & $ 8.1$ & & & $ 0.23 \pm 0.43$ & $ 0.27 \pm 0.33$ & $ -0.39 \pm 0.31$ & $ -0.32 \pm 0.95$\ 46 & 13 36 35.83 & -29 51 20.2 & $ 0.77^{\prime\prime}$ & $0.0098 \pm 0.0012$ & & & $ 71.2$ & & & $ 0.06 \pm 0.20$ & $ 0.24 \pm 0.16$ & $ -0.16 \pm 0.15$ & $ -0.78 \pm 0.27$\ 47 & 13 36 36.02 & -30 03 37.7 & $ 1.49^{\prime\prime}$ & & & $0.0030 \pm 0.0006$ & & & $ 10.6$ & $ 0.18 \pm 0.29$ & $ 0.09 \pm 0.25$ & $ -0.72 \pm 0.26$ & $ -0.75 \pm 1.08$\ 48 & 13 36 36.76 & -29 50 18.1 & $ 0.98^{\prime\prime}$ & $0.0090 \pm 0.0018$ & & & $ 39.8$ & & & $ 0.61 \pm 0.24$ & $ 0.22 \pm 0.20$ & $ -0.83 \pm 0.11$ & $ 0.53 \pm 0.41$\ 49 & 13 36 38.46 & -29 55 51.5 & $ 1.48^{\prime\prime}$ & $0.0061 \pm 0.0011$ & & $0.0057 \pm 0.0011$ & $ 28.5$ & & $ 28.4$ & $ 1.00 \pm 0.55$ & $ 0.53 \pm 0.17$ & $ -0.39 \pm 0.17$ & $ -0.68 \pm 0.44$\ 50 & 13 36 38.95 & -29 47 43.3 & $ 0.43^{\prime\prime}$ & $0.0263 \pm 0.0018$ & & & $ 384.0$ & & & $ 0.35 \pm 0.10$ & $ 0.22 \pm 0.07$ & $ -0.46 \pm 0.08$ & $ -0.42 \pm 0.19$\ 51 & 13 36 39.18 & -30 04 07.8 & $ 1.38^{\prime\prime}$ & $0.009 \pm 0.002$ & $0.0069 \pm 0.0011$ & $0.0131 \pm 0.0011$ & $ 22.1$ & $ 46.6$ & $ 258.2$ & $ 0.59 \pm 0.17$ & $ 0.36 \pm 0.09$ & $ -0.31 \pm 0.09$ & $ -0.50 \pm 0.15$\ 52 & 13 36 39.42 & -30 10 08.6 & $ 1.80^{\prime\prime}$ & & & $0.004 \pm 0.0010$ & & & $ 9.8$ & $ -0.49 \pm 0.41$ & $ 0.34 \pm 0.51$ & $ 0.18 \pm 0.35$ & $ -1.00 \pm 0.58$\ 53 & 13 36 40.14 & -29 59 52.6 & $ 0.82^{\prime\prime}$ & $0.0154 \pm 0.0017$ & $0.0130 \pm 0.0015$ & $0.0134 \pm 0.0012$ & $ 92.7$ & $ 103.6$ & $ 188.2$ & $ 0.22 \pm 0.12$ & $ 0.03 \pm 0.11$ & $ -0.29 \pm 0.11$ & $ -0.39 \pm 0.22$\ 54 & 13 36 40.52 & -30 05 47.0 & $ 1.34^{\prime\prime}$ & & & $0.0029 \pm 0.0007$ & & & $ 10.6$ & $ 0.45 \pm 0.35$ & $ -0.17 \pm 0.28$ & $ -0.56 \pm 0.46$ & $ 0.63 \pm 0.40$\ 55 & 13 36 40.73 & -29 51 09.1 & $ 1.33^{\prime\prime}$ & & $0.0121 \pm 0.0025$ & $0.015 \pm 0.002$ & & $ 12.2$ & $ 29.4$ & $ 0.89 \pm 0.18$ & $ 0.24 \pm 0.17$ & $ -0.44 \pm 0.17$ & $ 0.06 \pm 0.34$\ 56 & 13 36 41.41 & -30 13 26.8 & $ 1.38^{\prime\prime}$ & & $0.010 \pm 0.002$ & $0.009 \pm 0.003$ & & $ 22.2$ & $ 22.2$ & $ 0.19 \pm 0.27$ & $ -0.05 \pm 0.25$ & $ 0.00 \pm 0.27$ & $ -0.37 \pm 0.52$\ 57 & 13 36 41.79 & -30 11 17.4 & $ 1.64^{\prime\prime}$ & & & $0.0032 \pm 0.0008$ & & & $ 9.1$ & & $ 1.00 \pm 0.26$ & $ -0.03 \pm 0.26$ & $ -0.09 \pm 0.34$\ 58 & 13 36 42.24 & -30 03 31.4 & $ 1.40^{\prime\prime}$ & & & $0.0020 \pm 0.0006$ & & & $ 6.5$ & $ 1.00 \pm 1.35$ & $ 0.63 \pm 0.31$ & $ -0.41 \pm 0.29$ & $ -0.49 \pm 0.67$\ 59 & 13 36 42.49 & -30 09 34.5 & $ 1.40^{\prime\prime}$ & & & $0.0030 \pm 0.0007$ & & & $ 9.4$ & $ 0.74 \pm 0.26$ & $ 0.00 \pm 0.26$ & $ -0.73 \pm 0.25$ & $ 0.55 \pm 0.43$\ 60 & 13 36 43.44 & -29 51 06.5 & $ 0.50^{\prime\prime}$ & $0.03 \pm 1.58$ & $0.034 \pm 0.004$ & $0.0382 \pm 0.0031$ & $ 306.0$ & $ 94.7$ & $ 272.8$ & $ 0.72 \pm 0.92$ & $ 0.89 \pm 0.10$ & $ -0.04 \pm 0.11$ & $ -0.69 \pm 7.21$\ 61 & 13 36 44.16 & -29 48 41.8 & $ 0.28^{\prime\prime}$ & $0.051 \pm 0.002$ & & & $1211.9$ & & & $ 0.49 \pm 0.06$ & $ 0.18 \pm 0.05$ & $ -0.46 \pm 0.05$ & $ -0.97 \pm 0.07$\ 62 & 13 36 45.26 & -30 00 38.0 & $ 1.10^{\prime\prime}$ & $0.0113 \pm 0.0016$ & $0.0124 \pm 0.0014$ & $0.0094 \pm 0.0009$ & $ 50.7$ & $ 112.3$ & $ 125.3$ & $ 0.39 \pm 0.11$ & $ -0.26 \pm 0.11$ & $ -0.83 \pm 0.16$ & $ -0.46 \pm 0.97$\ 63 & 13 36 45.34 & -30 14 41.4 & $ 1.61^{\prime\prime}$ & & & $0.0060 \pm 0.0013$ & & & $ 16.4$ & $ 0.15 \pm 0.26$ & $ -0.12 \pm 0.26$ & $ -0.16 \pm 0.31$ & $ -0.43 \pm 0.77$\ 64 & 13 36 45.56 & -30 03 27.5 & $ 1.12^{\prime\prime}$ & $0.021 \pm 0.003$ & $0.0116 \pm 0.0013$ & $0.0135 \pm 0.0010$ & $ 81.5$ & $ 93.6$ & $ 256.9$ & $ 0.44 \pm 0.10$ & $ -0.05 \pm 0.09$ & $ -0.30 \pm 0.10$ & $ -0.54 \pm 0.16$\ 65 & 13 36 45.78 & -29 59 13.0 & $ 0.26^{\prime\prime}$ & $0.114 \pm 0.004$ & $0.048 \pm 0.002$ & $0.0189 \pm 0.0015$ & $2376.9$ & $ 878.9$ & $ 314.8$ & $ 0.59 \pm 0.04$ & $ 0.11 \pm 0.04$ & $ -0.60 \pm 0.04$ & $ -0.33 \pm 0.11$\ 66 & 13 36 45.94 & -30 00 00.2 & $ 1.69^{\prime\prime}$ & $0.0043 \pm 0.0013$ & & $0.0046 \pm 0.0007$ & $ 7.9$ & & $ 30.4$ & $ 0.55 \pm 0.21$ & $ -0.11 \pm 0.18$ & $ -0.33 \pm 0.22$ & $ -1.00 \pm 0.29$\ 67 & 13 36 47.76 & -30 02 44.4 & $ 1.37^{\prime\prime}$ & & $0.004 \pm 0.001$ & $0.0023 \pm 0.0006$ & & $ 14.2$ & $ 6.2$ & $ 0.15 \pm 0.30$ & $ -0.10 \pm 0.31$ & $ -0.15 \pm 0.34$ & $ -0.84 \pm 0.53$\ 68 & 13 36 47.79 & -29 46 47.2 & $ 1.50^{\prime\prime}$ & $0.0040 \pm 0.0008$ & & & $ 16.3$ & & & $ 1.00 \pm 0.44$ & $ 0.24 \pm 0.26$ & $ -0.06 \pm 0.21$ & $ -1.00 \pm 0.32$\ 69 & 13 36 49.07 & -29 52 58.7 & $ 0.50^{\prime\prime}$ & $0.032 \pm 0.002$ & $0.009 \pm 0.002$ & & $ 271.8$ & $ 9.0$ & & $ 0.53 \pm 0.11$ & $ 0.20 \pm 0.08$ & $ -0.68 \pm 0.06$ & $ -0.76 \pm 0.18$\ 70 & 13 36 50.59 & -30 14 35.1 & $ 1.45^{\prime\prime}$ & & & $0.0061 \pm 0.0014$ & & & $ 13.1$ & $ 0.16 \pm 0.44$ & $ -0.20 \pm 0.52$ & $ 0.51 \pm 0.36$ & $ 0.24 \pm 0.25$\ 71 & 13 36 51.15 & -29 41 54.7 & $ 1.84^{\prime\prime}$ & $0.0058 \pm 0.0014$ & & & $ 6.8$ & & & $ 0.34 \pm 0.63$ & $ 0.47 \pm 0.31$ & $ -0.13 \pm 0.26$ & $ -0.08 \pm 0.37$\ 72 & 13 36 51.39 & -30 18 01.7 & $ 2.03^{\prime\prime}$ & & & $0.0022 \pm 0.0008$ & & & $ 6.2$ & $ 0.53 \pm 0.62$ & $ 0.51 \pm 0.33$ & $ -0.47 \pm 0.46$ & $ 0.36 \pm 0.60$\ 73 & 13 36 51.59 & -29 53 34.5 & $ 1.12^{\prime\prime}$ & $0.0041 \pm 0.0012$ & & $0.0089 \pm 0.0020$ & $ 11.2$ & & $ 8.2$ & $ 1.00 \pm 0.70$ & $ 0.10 \pm 0.44$ & $ -0.17 \pm 0.26$ & $ -0.12 \pm 0.31$\ 74 & 13 36 52.38 & -29 51 43.6 & $ 1.72^{\prime\prime}$ & $0.0079 \pm 0.0016$ & & $0.010 \pm 0.002$ & $ 7.6$ & & $ 6.3$ & $ 0.22 \pm 0.21$ & $ -0.66 \pm 0.20$ & $ -0.96 \pm 0.49$ & $ 0.86 \pm 1.75$\ 75 & 13 36 53.43 & -30 08 40.3 & $ 1.49^{\prime\prime}$ & & $0.005 \pm 0.001$ & $0.005 \pm 0.002$ & & $ 18.3$ & $ 36.9$ & $ 0.93 \pm 1.68$ & $ -0.43 \pm 0.70$ & $ 0.77 \pm 0.28$ & $ -0.71 \pm 0.40$\ 76 & 13 36 53.53 & -29 55 59.1 & $ 1.58^{\prime\prime}$ & & $0.0039 \pm 0.0012$ & & & $ 7.7$ & & $ 0.82 \pm 0.18$ & $ -0.57 \pm 0.31$ & $ -1.00 \pm 0.40$ & $ 1.00 \pm 0.35$\ 77 & 13 36 53.78 & -29 48 49.5 & $ 1.63^{\prime\prime}$ & $0.0038 \pm 0.0009$ & & & $ 6.2$ & & & $ 0.15 \pm 0.23$ & $ -0.68 \pm 0.23$ & $ -1.00 \pm 0.88$ & $ 1.00 \pm 13.45$\ 78 & 13 36 54.86 & -30 09 46.7 & $ 1.63^{\prime\prime}$ & & & $0.0034 \pm 0.0008$ & & & $ 9.6$ & $ 0.65 \pm 0.37$ & $ -0.02 \pm 0.27$ & $ -0.10 \pm 0.29$ & $ -0.20 \pm 0.45$\ 79 & 13 36 55.43 & -29 55 09.0 & $ 0.69^{\prime\prime}$ & $0.015 \pm 0.002$ & $0.024 \pm 0.002$ & $0.0221 \pm 0.0019$ & $ 105.5$ & $ 146.5$ & $ 192.6$ & $ 0.11 \pm 0.22$ & $ 0.51 \pm 0.12$ & $ -0.08 \pm 0.09$ & $ -0.33 \pm 0.15$\ 80 & 13 36 56.56 & -29 49 12.2 & $ 0.55^{\prime\prime}$ & $0.0232 \pm 0.0018$ & & & $ 254.3$ & & & $ 0.65 \pm 0.16$ & $ 0.28 \pm 0.11$ & $ -0.20 \pm 0.08$ & $ -0.62 \pm 0.11$\ 81 & 13 36 57.22 & -29 53 38.3 & $ 0.36^{\prime\prime}$ & $0.0322 \pm 0.0019$ & $0.024 \pm 0.003$ & $0.0203 \pm 0.0024$ & $ 392.4$ & $ 97.4$ & $ 54.1$ & $ 0.50 \pm 0.11$ & $ 0.21 \pm 0.08$ & $ -0.33 \pm 0.06$ & $ -0.33 \pm 0.10$\ 82 & 13 36 57.29 & -29 47 28.1 & $ 1.13^{\prime\prime}$ & $0.0041 \pm 0.0008$ & & & $ 20.6$ & & & $ -0.79 \pm 1.60$ & $ 0.97 \pm 0.23$ & $ 0.02 \pm 0.19$ & $ -0.45 \pm 0.28$\ 83 & 13 36 58.26 & -29 51 04.3 & $ 0.86^{\prime\prime}$ & $0.0053 \pm 0.0009$ & & & $ 30.0$ & & & & $ 1.00 \pm 0.23$ & $ 0.24 \pm 0.20$ & $ -0.31 \pm 0.17$\ 84 & 13 36 58.26 & -29 48 32.8 & $ 0.78^{\prime\prime}$ & $0.0115 \pm 0.0013$ & & & $ 90.9$ & & & $ 0.85 \pm 0.51$ & $ 0.54 \pm 0.18$ & $ -0.14 \pm 0.11$ & $ -0.31 \pm 0.13$\ 85 & 13 36 58.67 & -29 43 35.7 & $ 0.85^{\prime\prime}$ & $0.0162 \pm 0.0016$ & & & $ 144.7$ & & & $ 0.40 \pm 0.13$ & $ -0.10 \pm 0.11$ & $ -0.30 \pm 0.14$ & $ -0.08 \pm 0.22$\ 86 & 13 36 58.84 & -30 05 18.1 & $ 1.82^{\prime\prime}$ & $0.007 \pm 0.002$ & $0.007 \pm 0.001$ & $0.0070 \pm 0.0008$ & $ 13.5$ & $ 64.6$ & $ 82.5$ & $ 0.78 \pm 0.16$ & $ 0.21 \pm 0.14$ & $ -0.31 \pm 0.14$ & $ -0.18 \pm 0.21$\ 87 & 13 36 59.35 & -29 49 58.4 & $ 0.24^{\prime\prime}$ & $0.075 \pm 0.003$ & $0.025 \pm 0.003$ & $0.057 \pm 0.004$ & $1750.7$ & $ 197.4$ & $ 266.3$ & $ 0.52 \pm 0.06$ & $ 0.17 \pm 0.04$ & $ -0.12 \pm 0.04$ & $ -0.53 \pm 0.05$\ 88 & 13 36 59.51 & -29 54 13.9 & $ 0.97^{\prime\prime}$ & $0.0057 \pm 0.0009$ & & & $ 32.1$ & & & $ 0.62 \pm 4.08$ & $ 0.95 \pm 0.25$ & $ -0.12 \pm 0.15$ & $ -0.48 \pm 0.20$\ 89 & 13 36 59.68 & -30 00 58.8 & $ 1.60^{\prime\prime}$ & $0.0051 \pm 0.0012$ & $0.0040 \pm 0.0009$ & $0.008 \pm 0.001$ & $ 15.5$ & $ 13.0$ & $ 81.4$ & $ 0.99 \pm 0.29$ & $ 0.59 \pm 0.13$ & $ -0.18 \pm 0.13$ & $ -0.75 \pm 0.25$\ 90 & 13 37 00.35 & -29 51 57.8 & $ 0.16^{\prime\prime}$ & $0.881 \pm 0.009$ & $0.959 \pm 0.016$ & $1.173 \pm 0.014$ & $ $ & $7696.9$ & $ $ & $ 0.62 \pm 0.01$ & $ -0.21 \pm 0.01$ & $ -0.62 \pm 0.01$ & $ -0.57 \pm 0.03$\ 91 & 13 37 00.47 & -29 50 52.3 & $ 1.23^{\prime\prime}$ & $0.0051 \pm 0.0011$ & & & $ 7.8$ & & & $ -0.41 \pm 0.24$ & $ -1.00 \pm 0.30$ & $ 1.00 \pm 22.70$ & $ 0.44 \pm 3.24$\ 92 & 13 37 00.51 & -29 53 18.7 & $ 0.99^{\prime\prime}$ & $0.0089 \pm 0.0018$ & & & $ 11.5$ & & & $ 0.06 \pm 0.56$ & $ 0.60 \pm 0.30$ & $ -0.29 \pm 0.21$ & $ -0.29 \pm 0.26$\ 93 & 13 37 00.83 & -29 51 59.9 & $ 0.19^{\prime\prime}$ & $0.243 \pm 0.006$ & & & $1121.0$ & & & $ 0.55 \pm 0.03$ & $ -0.41 \pm 0.03$ & $ -0.51 \pm 0.04$ & $ -0.63 \pm 0.07$\ 94 & 13 37 00.97 & -30 16 43.0 & $ 1.07^{\prime\prime}$ & & & $0.014 \pm 0.002$ & & & $ 41.8$ & $ 0.28 \pm 0.21$ & $ 0.08 \pm 0.17$ & $ -0.44 \pm 0.19$ & $ 0.35 \pm 0.28$\ 95 & 13 37 01.12 & -29 52 46.2 & $ 1.07^{\prime\prime}$ & $0.004 \pm 0.002$ & & & $ 15.0$ & & & $ -1.00 \pm 1.60$ & $ 1.00 \pm 1.27$ & $ 0.21 \pm 0.67$ & $ 0.01 \pm 0.70$\ 96 & 13 37 01.16 & -30 00 36.2 & $ 1.00^{\prime\prime}$ & & $0.0040 \pm 0.0009$ & $0.0036 \pm 0.0008$ & & $ 23.1$ & $ 12.5$ & $ 0.98 \pm 0.48$ & $ -0.66 \pm 0.49$ & $ 0.73 \pm 0.39$ & $ 0.25 \pm 0.22$\ 97 & 13 37 01.36 & -29 53 25.0 & $ 0.23^{\prime\prime}$ & $0.078 \pm 0.003$ & $0.044 \pm 0.003$ & $0.083 \pm 0.004$ & $1708.1$ & $ 282.9$ & $ 853.5$ & $ 0.97 \pm 0.07$ & $ 0.65 \pm 0.04$ & $ -0.30 \pm 0.03$ & $ -0.78 \pm 0.04$\ 98 & 13 37 01.49 & -29 47 42.7 & $ 0.38^{\prime\prime}$ & $0.0235 \pm 0.0015$ & & & $ 462.1$ & & & $ 0.87 \pm 0.11$ & $ 0.53 \pm 0.07$ & $ -0.19 \pm 0.07$ & $ -0.56 \pm 0.10$\ 99 & 13 37 01.75 & -29 51 26.4 & $ 1.47^{\prime\prime}$ & & $0.074 \pm 0.012$ & $0.036 \pm 0.004$ & & $ 39.6$ & $ 84.0$ & $ 0.67 \pm 0.15$ & $ 0.12 \pm 0.12$ & $ -0.27 \pm 0.10$ & $ -0.94 \pm 0.16$\ 100 & 13 37 01.99 & -29 55 17.5 & $ 0.58^{\prime\prime}$ & $0.0211 \pm 0.0017$ & $0.014 \pm 0.003$ & $0.023 \pm 0.002$ & $ 215.7$ & $ 38.9$ & $ 153.3$ & $ 0.41 \pm 0.16$ & $ 0.21 \pm 0.12$ & $ -0.08 \pm 0.10$ & $ -0.31 \pm 0.12$\ 101 & 13 37 02.10 & -30 12 28.3 & $ 1.72^{\prime\prime}$ & & $0.0068 \pm 0.0015$ & $0.0060 \pm 0.0011$ & & $ 12.0$ & $ 29.3$ & $ 0.91 \pm 0.21$ & $ 0.44 \pm 0.21$ & $ -0.20 \pm 0.19$ & $ -0.16 \pm 0.35$\ 102 & 13 37 02.25 & -29 44 26.8 & $ 1.35^{\prime\prime}$ & $0.0075 \pm 0.0015$ & & & $ 16.7$ & & & $ -0.09 \pm 0.55$ & $ 0.35 \pm 0.42$ & $ 0.42 \pm 0.25$ & $ 0.13 \pm 0.22$\ 103 & 13 37 02.72 & -29 52 25.5 & $ 0.46^{\prime\prime}$ & & $0.065 \pm 0.005$ & & & $ 234.4$ & & $ 0.46 \pm 0.13$ & $ 0.19 \pm 0.10$ & $ -0.24 \pm 0.08$ & $ -0.51 \pm 0.15$\ 104 & 13 37 02.79 & -29 57 36.6 & $ 1.59^{\prime\prime}$ & $0.0032 \pm 0.0008$ & $0.004 \pm 0.001$ & & $ 7.0$ & $ 8.7$ & & $ -0.64 \pm 0.40$ & $ 0.73 \pm 0.31$ & $ -0.27 \pm 0.29$ & $ -0.63 \pm 0.52$\ 105 & 13 37 03.69 & -30 06 31.0 & $ 0.95^{\prime\prime}$ & & $0.0085 \pm 0.0015$ & $0.0054 \pm 0.0008$ & & $ 54.7$ & $ 47.7$ & $ 0.69 \pm 1.10$ & $ 0.81 \pm 0.14$ & $ -0.24 \pm 0.18$ & $ -0.40 \pm 0.27$\ 106 & 13 37 03.80 & -29 49 29.9 & $ 0.40^{\prime\prime}$ & $0.0288 \pm 0.0018$ & $0.0055 \pm 0.0014$ & $0.013 \pm 0.003$ & $ 354.0$ & $ 17.0$ & $ 21.4$ & $ 0.26 \pm 0.10$ & $ 0.03 \pm 0.08$ & $ -0.25 \pm 0.07$ & $ -0.63 \pm 0.11$\ 107 & 13 37 04.21 & -29 54 03.1 & $ 0.30^{\prime\prime}$ & $0.045 \pm 0.002$ & $0.054 \pm 0.004$ & $0.087 \pm 0.004$ & $ 853.6$ & $ 350.7$ & $1165.7$ & $ 0.69 \pm 0.07$ & $ 0.43 \pm 0.04$ & $ -0.46 \pm 0.04$ & $ -0.78 \pm 0.09$\ 108 & 13 37 04.32 & -29 51 21.0 & $ 0.20^{\prime\prime}$ & $0.089 \pm 0.003$ & $0.016 \pm 0.003$ & $0.123 \pm 0.005$ & $1854.7$ & $ 47.2$ & $ 852.1$ & $ 0.60 \pm 0.05$ & $ 0.35 \pm 0.04$ & $ -0.43 \pm 0.03$ & $ -0.88 \pm 0.04$\ 109 & 13 37 05.12 & -29 52 26.5 & $ 0.46^{\prime\prime}$ & $0.0224 \pm 0.0016$ & & $0.0058 \pm 0.0017$ & $ 220.1$ & & $ 8.6$ & $ 0.52 \pm 0.14$ & $ 0.31 \pm 0.09$ & $ -0.25 \pm 0.07$ & $ -0.71 \pm 0.10$\ 110 & 13 37 05.49 & -29 57 56.5 & $ 1.35^{\prime\prime}$ & $0.0039 \pm 0.0009$ & $0.0053 \pm 0.0011$ & $0.0058 \pm 0.0010$ & $ 10.4$ & $ 23.4$ & $ 21.6$ & $ 0.97 \pm 0.25$ & $ 0.19 \pm 0.23$ & $ -0.18 \pm 0.23$ & $ -0.67 \pm 0.47$\ 111 & 13 37 06.30 & -29 51 03.2 & $ 1.04^{\prime\prime}$ & & & $0.031 \pm 0.005$ & & & $ 108.3$ & $ -0.01 \pm 1.06$ & $ 0.85 \pm 0.28$ & $ 0.16 \pm 0.17$ & $ -0.56 \pm 0.20$\ 112 & 13 37 06.32 & -30 17 24.5 & $ 2.88^{\prime\prime}$ & & $0.007 \pm 0.002$ & $0.010 \pm 0.002$ & & $ 6.1$ & $ 18.6$ & $ 0.34 \pm 0.25$ & $ -0.13 \pm 0.24$ & $ 0.09 \pm 0.26$ & $ -0.46 \pm 0.43$\ 113 & 13 37 06.45 & -30 13 40.4 & $ 1.99^{\prime\prime}$ & & $0.009 \pm 0.003$ & & & $ 8.6$ & & $ 1.00 \pm 3.86$ & $ 0.90 \pm 0.31$ & $ -0.03 \pm 0.29$ & $ -0.36 \pm 0.54$\ 114 & 13 37 06.99 & -29 51 02.1 & $ 0.40^{\prime\prime}$ & $0.0291 \pm 0.0017$ & $0.016 \pm 0.002$ & $0.033 \pm 0.004$ & $ 436.0$ & $ 118.4$ & $ 96.7$ & $ 0.98 \pm 0.15$ & $ 0.68 \pm 0.09$ & $ -0.14 \pm 0.06$ & $ -0.44 \pm 0.07$\ 115 & 13 37 07.09 & -29 46 49.5 & $ 0.87^{\prime\prime}$ & $0.006 \pm 0.001$ & & & $ 45.1$ & & & $ 0.14 \pm 0.34$ & $ 0.55 \pm 0.18$ & $ -0.37 \pm 0.17$ & $ -0.40 \pm 0.30$\ 116 & 13 37 07.23 & -29 51 32.9 & $ 0.52^{\prime\prime}$ & $0.0179 \pm 0.0014$ & & & $ 218.6$ & & & $ 1.00 \pm 0.26$ & $ 0.72 \pm 0.13$ & $ -0.11 \pm 0.08$ & $ -0.49 \pm 0.09$\ 117 & 13 37 08.13 & -30 03 05.0 & $ 1.11^{\prime\prime}$ & & $0.0046 \pm 0.0009$ & $0.0039 \pm 0.0010$ & & $ 20.6$ & $ 12.7$ & $ 0.32 \pm 0.32$ & $ 0.14 \pm 0.24$ & $ -0.04 \pm 0.23$ & $ -0.53 \pm 0.38$\ 118 & 13 37 08.26 & -29 53 36.1 & $ 2.90^{\prime\prime}$ & $0.020 \pm 0.004$ & $0.008 \pm 0.002$ & & $ 34.0$ & $ 11.4$ & & $ 0.85 \pm 0.57$ & $ 0.65 \pm 0.34$ & $ -0.09 \pm 0.19$ & $ -0.77 \pm 0.15$\ 119 & 13 37 10.65 & -30 11 18.9 & $ 1.11^{\prime\prime}$ & & $0.0136 \pm 0.0017$ & $0.011 \pm 0.003$ & & $ 90.2$ & $ 21.1$ & $ 0.03 \pm 0.14$ & $ -0.19 \pm 0.15$ & $ -0.61 \pm 0.20$ & $ -0.91 \pm 0.82$\ 120 & 13 37 12.54 & -29 51 53.2 & $ 0.78^{\prime\prime}$ & $0.011 \pm 0.003$ & & $0.008 \pm 0.002$ & $ 44.4$ & & $ 11.7$ & $ 0.97 \pm 0.49$ & $ -0.25 \pm 0.37$ & $ -0.25 \pm 0.36$ & $ 0.35 \pm 0.28$\ 121 & 13 37 12.60 & -30 09 01.0 & $ 0.74^{\prime\prime}$ & & $0.0147 \pm .0015$ & $0.0177 \pm 0.0016$ & & $ 125.8$ & $ 178.7$ & $ 0.24 \pm 0.15$ & $ 0.38 \pm 0.10$ & $ -0.42 \pm 0.10$ & $ -0.03 \pm 0.19$\ 122 & 13 37 12.66 & -29 43 10.0 & $ 1.02^{\prime\prime}$ & $0.0147 \pm 0.0017$ & & & $ 87.0$ & & & $ 0.15 \pm 0.16$ & $ 0.13 \pm 0.14$ & $ -0.14 \pm 0.15$ & $ -0.29 \pm 0.25$\ 123 & 13 37 12.80 & -30 05 33.0 & $ 1.51^{\prime\prime}$ & $0.0102 \pm 0.0020$ & $0.0135 \pm 0.0013$ & $0.0076 \pm 0.0009$ & $ 23.6$ & $ 172.3$ & $ 94.8$ & $ 0.27 \pm 0.19$ & $ 0.35 \pm 0.11$ & $ -0.23 \pm 0.11$ & $ -0.60 \pm 0.20$\ 124 & 13 37 12.91 & -29 45 08.9 & $ 1.14^{\prime\prime}$ & $0.0071 \pm 0.0011$ & & & $ 33.7$ & & & $ -0.03 \pm 0.22$ & $ 0.12 \pm 0.21$ & $ -0.37 \pm 0.22$ & $ -0.10 \pm 0.40$\ 125 & 13 37 14.65 & -29 54 28.8 & $ 1.05^{\prime\prime}$ & $0.0051 \pm 0.0009$ & & & $ 25.9$ & & & $ 0.80 \pm 0.47$ & $ 0.56 \pm 0.22$ & $ -0.48 \pm 0.16$ & $ -0.24 \pm 0.31$\ 126 & 13 37 15.83 & -30 02 56.3 & $ 0.62^{\prime\prime}$ & $0.028 \pm 0.002$ & $0.0339 \pm 0.0019$ & $0.0240 \pm 0.0017$ & $ 259.7$ & $ 597.8$ & $ 334.1$ & $ 0.22 \pm 0.08$ & $ -0.03 \pm 0.07$ & $ -0.37 \pm 0.08$ & $ -0.52 \pm 0.14$\ 127 & 13 37 16.06 & -29 56 55.5 & $ 1.24^{\prime\prime}$ & $0.004 \pm 0.001$ & $0.0057 \pm 0.0014$ & & $ 14.0$ & $ 8.7$ & & $ 0.98 \pm 0.71$ & $ 0.70 \pm 0.20$ & $ -0.84 \pm 0.17$ & $ 0.28 \pm 0.82$\ 128 & 13 37 16.22 & -29 41 56.2 & $ 1.37^{\prime\prime}$ & $0.0072 \pm 0.0014$ & & & $ 19.3$ & & & $ 0.73 \pm 0.42$ & $ 0.41 \pm 0.21$ & $ -0.50 \pm 0.20$ & $ -0.09 \pm 0.44$\ 129 & 13 37 16.27 & -29 49 38.3 & $ 0.36^{\prime\prime}$ & $0.0295 \pm 0.0017$ & & & $ 595.3$ & & & $ 0.48 \pm 0.14$ & $ 0.49 \pm 0.07$ & $ -0.17 \pm 0.06$ & $ -0.53 \pm 0.09$\ 130 & 13 37 17.22 & -29 51 53.1 & $ 0.88^{\prime\prime}$ & $0.0093 \pm 0.0011$ & & $0.008 \pm 0.002$ & $ 65.1$ & & $ 6.0$ & $ 0.80 \pm 0.12$ & $ -0.38 \pm 0.11$ & $ -1.00 \pm 0.17$ & $ 1.00 \pm 1.88$\ 131 & 13 37 19.23 & -29 57 09.6 & $ 0.74^{\prime\prime}$ & $0.0115 \pm 0.0013$ & $0.0071 \pm 0.0016$ & $0.0063 \pm 0.0013$ & $ 103.8$ & $ 16.0$ & $ 12.9$ & $ 0.63 \pm 0.19$ & $ 0.27 \pm 0.13$ & $ -0.21 \pm 0.12$ & $ -0.33 \pm 0.21$\ 132 & 13 37 19.54 & -30 04 29.2 & $ 1.08^{\prime\prime}$ & $0.014 \pm 0.002$ & $0.0176 \pm 0.0015$ & $0.0172 \pm 0.0014$ & $ 56.3$ & $ 199.3$ & $ 244.4$ & $ 0.30 \pm 0.11$ & $ 0.04 \pm 0.09$ & $ -0.40 \pm 0.10$ & $ -0.28 \pm 0.23$\ 133 & 13 37 19.73 & -29 53 48.1 & $ 0.12^{\prime\prime}$ & $0.294 \pm 0.004$ & $0.473 \pm 0.009$ & $0.261 \pm 0.007$ &$13526.54$ & $8642.0$ & $4230.5$ & $ 0.37 \pm 0.02$ & $ 0.04 \pm 0.02$ & $ -0.46 \pm 0.02$ & $ -0.73 \pm 0.03$\ 134 & 13 37 19.98 & -30 09 03.7 & $ 1.90^{\prime\prime}$ & & $0.0045 \pm 0.0014$ & $0.0073 \pm 0.0012$ & & $ 6.3$ & $ 31.7$ & $ 0.38 \pm 0.22$ & $ -0.29 \pm 0.21$ & $ 0.28 \pm 0.21$ & $ -0.33 \pm 0.29$\ 135 & 13 37 22.41 & -29 40 31.3 & $ 1.07^{\prime\prime}$ & $0.014 \pm 0.002$ & & & $ 79.2$ & & & $ 0.44 \pm 0.22$ & $ 0.22 \pm 0.17$ & $ -0.40 \pm 0.17$ & $ -0.68 \pm 0.37$\ 136 & 13 37 22.46 & -30 08 23.7 & $ 1.34^{\prime\prime}$ & & $0.0046 \pm 0.0013$ & $0.0036 \pm 0.0011$ & & $ 13.3$ & $ 10.1$ & $ 0.65 \pm 0.99$ & $ 0.14 \pm 0.51$ & $ 0.54 \pm 0.26$ & $ -0.27 \pm 0.34$\ 137 & 13 37 24.66 & -29 58 56.3 & $ 0.65^{\prime\prime}$ & $0.0168 \pm 0.0016$ & $0.0073 \pm 0.0012$ & $0.0099 \pm 0.0013$ & $ 156.0$ & $ 33.9$ & $ 53.0$ & $ 0.53 \pm 0.14$ & $ 0.19 \pm 0.11$ & $ -0.35 \pm 0.11$ & $ -0.54 \pm 0.27$\ 138 & 13 37 25.36 & -30 01 59.9 & $ 1.53^{\prime\prime}$ & & $0.0045 \pm 0.0011$ & $0.0058 \pm 0.0011$ & & $ 11.4$ & $ 24.4$ & $ 0.59 \pm 0.31$ & $ 0.36 \pm 0.19$ & $ -0.28 \pm 0.20$ & $ -0.43 \pm 0.44$\ 139 & 13 37 25.47 & -29 54 33.6 & $ 0.99^{\prime\prime}$ & $0.006 \pm 0.001$ & & & $ 27.4$ & & & $ 0.23 \pm 0.25$ & $ 0.17 \pm 0.20$ & $ -0.50 \pm 0.20$ & $ -0.28 \pm 0.48$\ 140 & 13 37 26.20 & -30 00 30.2 & $ 0.84^{\prime\prime}$ & $0.0114 \pm 0.0015$ & $0.0154 \pm 0.0018$ & $0.0157 \pm 0.0015$ & $ 78.2$ & $ 92.3$ & $ 146.3$ & $ 0.76 \pm 0.13$ & $ 0.25 \pm 0.10$ & $ -0.33 \pm 0.11$ & $ -0.45 \pm 0.25$\ 141 & 13 37 26.36 & -29 48 33.0 & $ 1.24^{\prime\prime}$ & $0.0057 \pm 0.0009$ & & & $ 34.6$ & & & $ 0.61 \pm 0.20$ & $ 0.18 \pm 0.17$ & $ -0.71 \pm 0.16$ & $ 0.11 \pm 0.51$\ 142 & 13 37 26.68 & -30 01 47.4 & $ 1.50^{\prime\prime}$ & $0.0066 \pm 0.0014$ & & & $ 14.7$ & & & $ 0.35 \pm 0.43$ & $ 0.46 \pm 0.23$ & $ -0.30 \pm 0.23$ & $ -0.50 \pm 0.50$\ 143 & 13 37 27.24 & -29 55 47.6 & $ 1.33^{\prime\prime}$ & $0.0112 \pm 0.0026$ & $0.0110 \pm 0.0018$ & & $ 31.6$ & $ 50.7$ & & $ 0.47 \pm 0.18$ & $ -0.05 \pm 0.16$ & $ -0.99 \pm 0.17$ & $ 0.95 \pm 1.08$\ 144 & 13 37 27.46 & -30 02 28.3 & $ 1.80^{\prime\prime}$ & $0.0061 \pm 0.0015$ & $0.0037 \pm 0.0009$ & & $ 11.3$ & $ 10.6$ & & $ 0.60 \pm 0.24$ & $ -0.32 \pm 0.22$ & $ -0.93 \pm 0.53$ & $ 0.50 \pm 3.29$\ 145 & 13 37 27.47 & -30 13 56.1 & $ 1.17^{\prime\prime}$ & & $0.014 \pm 0.002$ & $0.009 \pm 0.005$ & & $ 35.6$ & $ 16.0$ & $ 0.50 \pm 0.25$ & $ 0.14 \pm 0.18$ & $ -0.42 \pm 0.20$ & $ 0.07 \pm 0.38$\ 146 & 13 37 28.34 & -29 54 25.3 & $ 1.55^{\prime\prime}$ & $0.0048 \pm 0.0012$ & & & $ 12.8$ & & & $ 0.82 \pm 0.32$ & $ 0.02 \pm 0.26$ & $ 0.03 \pm 0.26$ & $ -0.75 \pm 0.62$\ 147 & 13 37 28.78 & -29 49 43.1 & $ 1.20^{\prime\prime}$ & $0.005 \pm 0.001$ & & & $ 26.7$ & & & $ 0.13 \pm 0.60$ & $ 0.61 \pm 0.25$ & $ 0.01 \pm 0.20$ & $ -0.12 \pm 0.27$\ 148 & 13 37 29.36 & -29 50 27.4 & $ 1.35^{\prime\prime}$ & $0.0039 \pm 0.0009$ & & & $ 8.2$ & & & $ 0.01 \pm 0.29$ & $ -0.04 \pm 0.29$ & $ -0.37 \pm 0.35$ & $ 0.07 \pm 0.53$\ 149 & 13 37 29.48 & -29 50 08.5 & $ 1.00^{\prime\prime}$ & $0.0093 \pm 0.0012$ & & & $ 57.8$ & & & $ 0.25 \pm 0.18$ & $ 0.07 \pm 0.15$ & $ -0.29 \pm 0.16$ & $ -0.32 \pm 0.34$\ 150 & 13 37 29.51 & -30 01 47.6 & $ 1.44^{\prime\prime}$ & $0.0042 \pm 0.0012$ & $0.0064 \pm 0.0013$ & $0.0095 \pm 0.0017$ & $ 6.8$ & $ 14.7$ & $ 29.0$ & $ 0.76 \pm 0.32$ & $ 0.39 \pm 0.18$ & $ -0.24 \pm 0.19$ & $ 0.17 \pm 0.27$\ 151 & 13 37 29.52 & -30 04 16.6 & $ 1.06^{\prime\prime}$ & & $0.0106 \pm 0.0019$ & & & $ 38.9$ & & $ 0.83 \pm 0.16$ & $ -0.02 \pm 0.17$ & $ -0.88 \pm 0.16$ & $ 0.53 \pm 0.63$\ 152 & 13 37 29.91 & -29 48 27.6 & $ 1.40^{\prime\prime}$ & $0.0064 \pm 0.0017$ & & & $ 26.1$ & & & $ 0.13 \pm 1.22$ & $ 0.81 \pm 0.32$ & $ -0.64 \pm 0.21$ & $ -0.22 \pm 0.64$\ 153 & 13 37 30.47 & -29 59 37.5 & $ 0.58^{\prime\prime}$ & $0.0253 \pm 0.0020$ & $0.0247 \pm 0.0022$ & $0.0189 \pm 0.0018$ & $ 229.3$ & $ 233.8$ & $ 145.0$ & $ 0.61 \pm 0.11$ & $ 0.07 \pm 0.10$ & $ -0.40 \pm 0.11$ & $ -0.27 \pm 0.23$\ 154 & 13 37 30.97 & -29 42 34.4 & $ 0.49^{\prime\prime}$ & $0.072 \pm 0.004$ & & & $ 728.3$ & & & $ 0.94 \pm 0.05$ & $ 0.31 \pm 0.07$ & $ -0.27 \pm 0.06$ & $ -0.26 \pm 0.11$\ 155 & 13 37 31.17 & -29 51 56.8 & $ 0.47^{\prime\prime}$ & $0.029 \pm 0.002$ & $0.0033 \pm 0.0011$ & & $ 375.3$ & $ 11.4$ & & $ 0.52 \pm 0.13$ & $ 0.32 \pm 0.09$ & $ -0.22 \pm 0.08$ & $ -0.24 \pm 0.13$\ 156 & 13 37 31.98 & -30 05 58.8 & $ 1.18^{\prime\prime}$ & & $0.0064 \pm 0.0013$ & & & $ 25.0$ & & $ 0.47 \pm 0.23$ & $ -0.09 \pm 0.20$ & $ -0.30 \pm 0.25$ & $ -0.81 \pm 1.10$\ 157 & 13 37 32.39 & -30 10 19.9 & $ 1.18^{\prime\prime}$ & & $0.013 \pm 0.002$ & $0.0196 \pm 0.0021$ & & $ 41.9$ & $ 126.6$ & $ 0.44 \pm 0.13$ & $ -0.05 \pm 0.12$ & $ -0.10 \pm 0.13$ & $ -0.72 \pm 0.34$\ 158 & 13 37 32.94 & -29 51 01.2 & $ 1.20^{\prime\prime}$ & $0.0059 \pm 0.0012$ & & & $ 19.8$ & & & $ 0.78 \pm 0.22$ & $ -0.10 \pm 0.23$ & $ -0.13 \pm 0.25$ & $ -0.20 \pm 0.41$\ 159 & 13 37 33.01 & -30 06 40.6 & $ 1.75^{\prime\prime}$ & & $0.0051 \pm 0.0013$ & $0.0080 \pm 0.0016$ & & $ 9.9$ & $ 14.9$ & $ 0.14 \pm 0.38$ & $ 0.48 \pm 0.22$ & $ -0.26 \pm 0.21$ & $ -0.06 \pm 0.40$\ 160 & 13 37 33.32 & -29 55 15.6 & $ 1.27^{\prime\prime}$ & $0.004 \pm 0.001$ & & $0.052 \pm 0.003$ & $ 18.5$ & & $ 416.3$ & $ 0.64 \pm 0.10$ & $ 0.29 \pm 0.07$ & $ -0.25 \pm 0.07$ & $ -0.23 \pm 0.13$\ 161 & 13 37 33.35 & -29 57 02.6 & $ 0.84^{\prime\prime}$ & $0.0114 \pm 0.0014$ & $0.0099 \pm 0.0019$ & $0.0078 \pm 0.0015$ & $ 84.5$ & $ 39.0$ & $ 25.3$ & $ 0.26 \pm 0.17$ & $ 0.06 \pm 0.14$ & $ -0.20 \pm 0.15$ & $ -0.76 \pm 0.40$\ 162 & 13 37 33.80 & -29 59 58.8 & $ 1.08^{\prime\prime}$ & $0.0087 \pm 0.0015$ & & $0.004 \pm 0.001$ & $ 35.6$ & & $ 10.1$ & $ 0.57 \pm 0.45$ & $ 0.58 \pm 0.21$ & $ -0.12 \pm 0.17$ & $ -0.49 \pm 0.30$\ 163 & 13 37 36.42 & -30 10 52.1 & $ 1.28^{\prime\prime}$ & & $0.0075 \pm 0.0018$ & $0.0097 \pm 0.0021$ & & $ 15.1$ & $ 21.1$ & $ 0.83 \pm 0.23$ & $ 0.26 \pm 0.22$ & $ -0.30 \pm 0.22$ & $ -0.22 \pm 0.53$\ 164 & 13 37 36.72 & -29 48 18.3 & $ 0.51^{\prime\prime}$ & $0.0191 \pm 0.0017$ & & & $ 195.4$ & & & $ 0.29 \pm 0.11$ & $ -0.07 \pm 0.10$ & $ -0.33 \pm 0.12$ & $ -0.41 \pm 0.27$\ 165 & 13 37 39.04 & -30 03 32.5 & $ 1.29^{\prime\prime}$ & & $0.0061 \pm 0.0013$ & & & $ 19.6$ & & $ 0.27 \pm 0.48$ & $ 0.59 \pm 0.21$ & $ -0.45 \pm 0.20$ & $ -1.00 \pm 0.51$\ 166 & 13 37 39.27 & -29 43 21.5 & $ 1.09^{\prime\prime}$ & $0.0129 \pm 0.0018$ & & & $ 47.6$ & & & $ 0.70 \pm 0.19$ & $ 0.16 \pm 0.16$ & $ -0.22 \pm 0.16$ & $ -0.41 \pm 0.33$\ 167 & 13 37 40.17 & -30 03 16.7 & $ 1.38^{\prime\prime}$ & & $0.0059 \pm 0.0014$ & & & $ 12.5$ & & $ 0.67 \pm 0.27$ & $ -0.10 \pm 0.24$ & $ -0.52 \pm 0.31$ & $ 0.19 \pm 0.58$\ 168 & 13 37 40.30 & -29 51 23.9 & $ 1.42^{\prime\prime}$ & $0.0050 \pm 0.0011$ & & & $ 11.6$ & & & $ 0.25 \pm 0.23$ & $ -0.25 \pm 0.24$ & $ -0.50 \pm 0.36$ & $ -0.09 \pm 0.97$\ 169 & 13 37 41.38 & -30 06 04.3 & $ 1.34^{\prime\prime}$ & & $0.010 \pm 0.002$ & $0.0081 \pm 0.0015$ & & $ 20.4$ & $ 23.5$ & $ 0.80 \pm 0.31$ & $ 0.19 \pm 0.19$ & $ -0.32 \pm 0.21$ & $ -0.29 \pm 0.41$\ 170 & 13 37 42.47 & -29 51 36.9 & $ 1.19^{\prime\prime}$ & $0.0060 \pm 0.0012$ & & & $ 22.9$ & & & $ 0.76 \pm 0.29$ & $ 0.23 \pm 0.19$ & $ -0.43 \pm 0.23$ & $ -0.38 \pm 0.61$\ 171 & 13 37 42.88 & -30 05 17.3 & $ 0.48^{\prime\prime}$ & & $0.041 \pm 0.003$ & $0.030 \pm 0.002$ & & $ 392.5$ & $ 221.6$ & $ 0.24 \pm 0.09$ & $ -0.03 \pm 0.09$ & $ -0.21 \pm 0.10$ & $ -0.67 \pm 0.16$\ 172 & 13 37 43.32 & -30 06 01.4 & $ 0.84^{\prime\prime}$ & & $0.019 \pm 0.002$ & $0.018 \pm 0.002$ & & $ 90.5$ & $ 102.2$ & $ 0.15 \pm 0.14$ & $ 0.01 \pm 0.13$ & $ -0.40 \pm 0.15$ & $ -0.33 \pm 0.40$\ 173 & 13 37 44.45 & -29 53 06.4 & $ 0.55^{\prime\prime}$ & $0.029 \pm 0.002$ & & & $ 364.4$ & & & $ 0.20 \pm 0.09$ & $ 0.09 \pm 0.08$ & $ -0.40 \pm 0.09$ & $ -0.48 \pm 0.24$\ 174 & 13 37 44.72 & -30 07 47.3 & $ 1.72^{\prime\prime}$ & & $0.0063 \pm 0.0015$ & $0.0074 \pm 0.0018$ & & $ 10.5$ & $ 10.7$ & $ 0.81 \pm 0.19$ & $ -0.17 \pm 0.23$ & $ -1.00 \pm 0.32$ & $ 1.00 \pm 0.33$\ 175 & 13 37 49.64 & -29 55 40.9 & $ 0.92^{\prime\prime}$ & $0.0124 \pm 0.0019$ & $0.0031 \pm 0.0009$ & $0.007 \pm 0.002$ & $ 53.8$ & $ 13.1$ & $ 12.7$ & $ 0.79 \pm 0.24$ & $ 0.38 \pm 0.18$ & $ -0.20 \pm 0.16$ & $ -0.60 \pm 0.29$\ 176 & 13 37 49.97 & -29 52 17.4 & $ 0.75^{\prime\prime}$ & $0.0206 \pm 0.0019$ & & & $ 171.3$ & & & $ 0.34 \pm 0.13$ & $ 0.11 \pm 0.10$ & $ -0.33 \pm 0.12$ & $ -0.66 \pm 0.28$\ 177 & 13 37 50.30 & -29 56 43.3 & $ 0.76^{\prime\prime}$ & $0.0126 \pm 0.0014$ & $0.0061 \pm 0.0014$ & $0.024 \pm 0.005$ & $ 139.6$ & $ 23.3$ & $ 35.9$ & $ 0.59 \pm 0.16$ & $ 0.22 \pm 0.13$ & $ -0.31 \pm 0.14$ & $ -0.18 \pm 0.24$\ 178 & 13 37 51.98 & -29 48 27.4 & $ 1.74^{\prime\prime}$ & $0.0073 \pm 0.0016$ & & & $ 12.5$ & & & $ -0.22 \pm 0.37$ & $ 0.32 \pm 0.33$ & $ 0.05 \pm 0.25$ & $ -0.20 \pm 0.38$\ 179 & 13 37 52.98 & -29 44 06.6 & $ 1.29^{\prime\prime}$ & $0.0022 \pm 0.0007$ & & & $ 8.8$ & & & $ 0.20 \pm 0.47$ & $ 0.10 \pm 0.42$ & $ -0.41 \pm 0.45$ & $ 0.48 \pm 0.44$\ 180 & 13 37 55.42 & -29 55 00.6 & $ 1.51^{\prime\prime}$ & $0.0024 \pm 0.0007$ & & & $ 7.1$ & & & $ 0.31 \pm 0.33$ & $ -0.42 \pm 0.37$ & $ 0.26 \pm 0.44$ & $ -0.76 \pm 0.67$\ 181 & 13 37 56.50 & -29 47 26.6 & $ 1.48^{\prime\prime}$ & $0.005 \pm 0.002$ & & & $ 16.9$ & & & $ 0.52 \pm 0.69$ & $ 0.00 \pm 0.50$ & $ -0.17 \pm 0.66$ & $ -0.02 \pm 0.75$\ 182 & 13 37 57.58 & -30 01 39.9 & $ 1.51^{\prime\prime}$ & & $0.0025 \pm 0.0008$ & $0.009 \pm 0.002$ & & $ 14.2$ & $ 8.6$ & $ 0.42 \pm 0.31$ & $ -0.29 \pm 0.31$ & $ -1.00 \pm 0.61$ & $ $\ 183 & 13 37 58.34 & -29 55 31.3 & $ 1.14^{\prime\prime}$ & $0.0062 \pm 0.0011$ & & & $ 40.3$ & & & $ 0.14 \pm 0.24$ & $ 0.25 \pm 0.19$ & $ -0.57 \pm 0.19$ & $ -0.47 \pm 0.62$\ 184 & 13 37 58.94 & -30 09 10.5 & $ 1.79^{\prime\prime}$ & & & $0.015 \pm 0.003$ & & & $ 33.0$ & $ 0.69 \pm 0.30$ & $ 0.28 \pm 0.19$ & $ -0.26 \pm 0.20$ & $ 0.00 \pm 0.43$\ 185 & 13 38 00.44 & -30 02 58.3 & $ 2.64^{\prime\prime}$ & & & $0.008 \pm 0.003$ & & & $ 10.2$ & $ 0.42 \pm 0.29$ & $ -0.88 \pm 0.29$ & $ 0.82 \pm 0.44$ & $ 0.18 \pm 0.49$\ 186 & 13 38 01.14 & -30 05 38.1 & $ 1.59^{\prime\prime}$ & & $0.0064 \pm 0.0014$ & & & $ 23.3$ & & $ 0.50 \pm 0.31$ & $ 0.25 \pm 0.25$ & $ -0.19 \pm 0.27$ & $ -0.07 \pm 0.42$\ 187 & 13 38 02.67 & -29 56 45.4 & $ 8.25^{\prime\prime}$ & & & $0.013 \pm 0.004$ & & & $ 12.1$ & $ -0.33 \pm 0.31$ & $ -0.24 \pm 0.55$ & $ 0.50 \pm 0.42$ & $ 0.43 \pm 0.29$\ 188 & 13 38 02.87 & -29 49 40.5 & $ 2.10^{\prime\prime}$ & $0.0022 \pm 0.0007$ & & & $ 12.1$ & & & $ 1.00 \pm 0.22$ & $ 0.33 \pm 0.29$ & $ -0.43 \pm 0.34$ & $ -1.00 \pm 0.89$\ 189 & 13 38 05.57 & -29 57 45.4 & $ 0.74^{\prime\prime}$ & & & $0.041 \pm 0.004$ & & & $ 222.4$ & $ 0.27 \pm 0.12$ & $ 0.07 \pm 0.10$ & $ -0.25 \pm 0.12$ & $ -0.40 \pm 0.30$\ [lcccccc]{} \ & & & & & &\ \ & & & & & &\ \ [ ]{} [ ]{} 1 & 0600-0299949 (A) & & & & &\ 2 & 0599-0299809 (B) & & & & &\ 3 & & & & & &\ 4 & & & & & &\ 5 & & & & & &\ 6 & & & & & &\ 7 & 0598-0300930 (C) & 13360480-3008319 (C) & QSO1333-298 (C) & & & AGN\ 8 & & & & & &\ 9 & & 13360571-2952048 (C) & & & &\ 10 & 0598-0300943 (C) & 13360646-3006475 (C) & & & &\ 11 & & & & & &\ 12 & 0602-0301227 (C) & & & & \[I1999\]2 star (B), \[E1998\]17 (A) &\ 13 & & & & & &\ 14 & & & & & \[I1999\]4 (C) &\ 15 & 0600-0300086 (C) & & & & &\ 16 & 0600-0300087 (C) & & & & \[I1999\]3 (B), \[E1998\]18 (B) & hard source\ 17 & 0600-0300099 (C) & & & & \[I1999\]5 gal. (B), \[E1998\]19 (B) & AGN\ 18 & & & & & &\ 19 & & & & & &\ 20 & & & & NVSSJ133618-301459 (B) & & (AGN)\ 21 & 0599-0299962 (C) & 13361884-3001381 (C) & & & \[E1998\]20 (B) & fg. star\ 22 & & & & & &\ 23 & & & & & &\ 24 & 0601-0298625 (C) & 13362007-2951058 (C) & & & & (CV)\ 25 & & & & & &\ 26 & 0597-0300800 (C) & & & & &\ 27 & & & & & &\ 28 & 0598-0301061 (A) & & & & &\ 29 & 0601-0298670 (A) & & & & &\ 30 & & & & & \[E1998\]22 (B) &\ 31 & 0602-0301338 (C) & 13362821-2942266 (B) & & & & GAL\ 32 & & & & & &\ 33 & 0600-0300208 (C) & & & & \[E1998\]22 (B) &\ 34 & 0601-0298683 (C) & 13362901-2951232 (B) & & & &\ 35 & & & & & &\ 36 & & & & & &\ 37 & & & & NVSSJ133630-301651 (C) & & (AGN)\ 38 & & & & & &\ 39 & & & & & &\ 40 & & & & & &\ 41 & & & & & &\ 42 & & & & & &\ 43 & & & & & &\ 44 & & & & & &\ 45 & & & & & &\ 46 & & & & & &\ 47 & & & & & &\ 48 & & & & & &\ 49 & & & & & &\ 50 & & & & & \[I1999\]7 (C), \[E1998\]24 (B) &\ 51 & & & & & &\ 52 & & & & & &\ 53 & 0600-0300263 (B) & & & & \[E1998\]25 (B) &\ 54 & 0599-0300108 (B) & & & & &\ 55 & & & & & \[SW2003\]2 (C) &\ 56 & & & & & &\ 57 & & & & & &\ 58 & & & & & &\ 59 & & & & & &\ 60 & & & & & \[SW2003\]5 xrb cand. (C), \[I1999\]8 (C), \[E1998\]1 (C) & hard source\ 61 & & & & & & hard source\ 62 & & & & & &\ 63 & 0597-0300945 (B) & & & & &\ 64 & & & & & \[E1998\]26 (B) &\ 65 & 0600-0300305 (C) & 13364580-2959124 (C) &6dFGSgJ133645.8-295913, gal (C)& & \[I1999\]10 gal. (B), \[E1998\]27 (B) & GAL\ 66 & & & & & &\ 67 & & & & & &\ 68 & & & & & &\ 69 & & & & & \[SW2003\]8 snr or xrb cand. (A), \[SW2003\]9 (C), \[I1999\]12 (B), \[E1998\]3 (B) &\ 70 & & & & & &\ 71 & & & & & &\ 72 & & & & & &\ 73 & & & & & \[SW2003\]14 (A), \[I1999\]14 (B) &\ 74 & & & & & &\ 75 & 0598-0301228 (C) & & & & &\ 76 & & & & & \[SW2003\]19 (B) &\ 77 & & & & & \[SW2003\]20 (A), \[DK03\]20 (A) &\ 78 & 0598-0301233 (A) & & & & &\ 79 & & & & & \[SW2003\]27 snr cand. (B), \[I1999\]15 (B) & (SNR)\ 80 & & & & & \[SW2003\]31 xrb cand. (A) & hard source\ 81 & & & & & \[SW2003\]33 xrb (C), \[I1999\]16 (B) & XRB\ 82 & & & & & \[SW2003\]34 (C) &\ 83 & & & &\[CRB1994\]3 (B), \[MCK2006\]28 (A), radio-gal.& \[SW2003\]39 gal. (B) & GAL\ 84 & & & & & \[SW2003\]40 (B) &\ 85 & 0602-0301563 (C) & & & & \[E1998\]28 (B) &\ 86 & & & & & &\ 87 & & & & & \[SW2003\]44 (C), \[I1999\]17 (B), \[E1998\]5 (B) &\ 88 & & & & & \[SW2003\]46 (B) &\ 89 & 0599-0300236 (A) & & \[BRK2009\] 7, gal (A) & & & GAL\ 90 & & & & & \[I1999\]19 (A), \[E1998\]6 (A), \[T1985\]1 (A) &\ 91 & & & & & \[SW2003\]55 (B), \[T1985\]3 (A), \[DK03\]50 sss cand. (B) & (SSS)\ 92 & & & & & \[SW2003\]60 xrb cand. (C) & hard source\ 93 & & & & & \[E1998\]6 (A), \[T1985\]1 (A) &\ 94 & 0597-0301054 (B) & & & & &\ 95 & & & & & \[SW2003\]65 (A), \[I1999\]21 (B) &\ 96 & 0599-0300246 (C) & 13370132-3000361 (C) & & & &\ 97 & & & & & \[SW2003\]72 xrb cand. (C), \[I1999\]20 (A), \[E1998\]7 (A) & hard source\ 98 & & & & & \[SW2003\]76 (C) &\ 99 & & 13370140-2951257 (C) & & \[MCK2006\]34 (C) & \[SW2003\]73 xrb or snr cand. (B) & hard source\ 100 & & & & & \[SW2003\]78 (A), \[I1999\]22 (C) &\ 101 & & & & & &\ 102 & 0602-0301590 (B) & & & & &\ 103 & & & &\[CRB1994\]6 (C), \[MCK2006\]36 (C) & \[SW2003\]84 (C) & hard source\ 104 & & & & & &\ 105 & 0598-0301292 (B) & & & & &\ 106 & & 13370433-2949306 (C) & & & \[SW2003\]85 xrb cand. (C), \[I1999\]25 (B) & hard source\ 107 & & & &\[CRB1994\]8 (C), \[MCK2006\]38 (C) & \[SW2003\]86 snr or xrb cand. (C), \[I1999\]26 (B), \[E1998\]9 (B) & hard source\ 108 & & & & & \[SW2003\]88 xrb cand. (C), \[I1999\]27 (C), \[E1998\]8 (C), \[T1985\]4 (A) & hard source\ 109 & & & & & \[SW2003\]93 (C) &\ 110 & & & & & &\ 111 & & & & & &\ 112 & & & & & &\ 113 & & & & & &\ 114 & & & & & \[SW2003\]104 xrb cand. (C) & hard source\ 115 & & & & & &\ 116 & & & \[BL2004\]53, snr cand. (C)& & \[SW2003\]105 (C), \[DK03\]88 sss cand. (C) & hard source\ 117 & & & & & &\ 118 & 0601-0298889 (C) & 13370870-2953313 (C) & & & &\ 119 & & & & & &\ 120 & & & & & \[SW2003\]113 xrb (B) & XRB\ 121 & & & & & &\ 122 & 0602-0301643 (B) & & & & &\ 123 & 0599-0300322 (B) & & & & &\ 124 & & & & & &\ 125 & & & & & \[SW2003\]119 (B) &\ 126 & 0599-0300335 (C) & & & & \[E1998\]30 (C) & hard source\ 127 & & & & & &\ 128 & & & & & &\ 129 & & & & & \[SW2003\]121 xrb cand. (C), \[I1999\]29 (B), \[E1998\]12 (B)& hard source\ 130 & & & & & \[SW2003\]122 (A) &\ 131 & & & & & &\ 132 & & & & & &\ 133 & & & & & \[I1999\]30 ULX (C), \[E1998\]13 (C), \[T1985\]2 (A) & ULX\ 134 & & & & & &\ 135 & 0603-0300266 (A) & & & & &\ 136 & & & & & &\ 137 & & & & & \[I1999\]31 star (B), \[E1998\]31 (B) &\ 138 & & & & & &\ 139 & & & & & &\ 140 & & & & & &\ 141 & & & & & &\ 142 & & & & & &\ 143 & 0600-0300561 (A) & 13372725-2955475 (A) & & & \[I1999\]32 star (B) & fg. star\ 144 & 0599-0300410 (A) & 13372747-3002283 (A) &6dFGSgJ133727.5-300228, gal (A)& & & GAL\ 145 & 0597-0301232 (A) & & & & &\ 146 & & & & & &\ 147 & 0601-0298991 (C) & & & & &\ 148 & 0601-0298995 (B) & 13372947-2950283 (B) &6dFGSgJ133729.5-295028, gal (B)& & & GAL\ 149 & & & & & &\ 150 & & & & & &\ 151 & & & & & &\ 152 & & & & & &\ 153 & & & & & &\ 154 & & & & & \[E1998\]32 (A) &\ 155 & & & & & \[I1999\]33 (B), \[E1998\]33 (B) &\ 156 & & & & & &\ 157 & & & & & &\ 158 & 0601-0299030 (C) & 13373327-2951007 (C) & ESO444-85, gal (C) & & & GAL\ 159 & 0598-0301544 (A) & & & & &\ 160 & 0600-0300590 (C) & 13373342-2955182 (C) & & & &\ 161 & & & & & &\ 162 & & & & & &\ 163 & 0598-0301583 (C) & & & & &\ 164 & 0601-0299090 (C) & & & & \[I1999\]34 (A), \[E1998\]34 (A) &\ 165 & & & & & \[E1998\]35 (B) &\ 166 & 0602-0301832 (C) & & & & \[I1999\]35 (B), \[E1998\]36 (A) &\ 167 & & & & & \[E1998\]35 (C) &\ 168 & 0601-0299123 (B) & 13374032-2951233 (A) & & & &\ 169 & & & & & &\ 170 & & & & & \[E1998\]37 (A) &\ 171 & & & & & &\ 172 & 0598-0301628 (C) & & & & \[E1998\]38 (B) &\ 173 & 0601-0299160 (B) & & & & \[E1998\]39 (B) &\ 174 & 0598-0301638 (A) & 13374472-3007463 (A) & & & & fg. star\ 175 & 0600-0300711 (B) & 13374956-2955409 & & & &\ 176 & & & & & &\ 177 & & 13375055-2956418 (C) & & & \[E1998\]40 (B) &\ 178 & 0601-0299209 (B) & & & & &\ 179 & & 13375276-2944038 (C) & & & &\ 180 & & & & & &\ 181 & & & & & &\ 182 & 0599-0300696 (B) & 13375770-3001406 (B) & & & & fg. star\ 183 & & & & & &\ 184 & 0598-0301763 (A) & & & & &\ 185 & & & & & &\ 186 & & & & & &\ 187 & & & & & &\ 188 & & & & & &\ 189 & 0600-0300832 (A) & & & NVSSJ133805-295748 (C) & \[I1999\]36 star (B), \[E1998\]42 (B) & (AGN)\ [^1]: Based on observations obtained with XMM-Newton, an ESA science mission with instruments and contributions directly funded by ESA Member States and NASA. [^2]: Appendix \[sect. Discussion of classification and identification of the XMM-Newton sources\] is available in electronic form at http://www.aanda.org [^3]: Tables \[Tab. source list\] and \[Tab. source list classification\] are only available in electronic form at the CDS via anonymous ftp to cdsarc.u-strasbg.fr (130.79.128.5) or via http://cdsweb.u-strasbg.fr/cgi-bin/qcat?J/A+A/ [^4]: The detection likelihood $L$ is defined by the relationship $L = - \ln (p)$, where $p$ is the probability that a Poissonian fluctuation in the background is detected as a spurious source.

ArXiv

--- author: - | \ Department of Physics and Astronomy, San Francisco State University,\ San Francisco, CA 94132, USA\ E-mail: - | \ Raymond and Beverly Sackler School of Physics and Astronomy\ Tel-Aviv University, Ramat Aviv, 69978 Israel\ E-mail: title: Effective field theory for pions and a dilatonic meson --- Introduction ============ Asymptotically free gauge theories with relatively few fermion degrees of freedom exist in a chirally broken and confining phase, associated with a coupling that grows toward the infrared. Increasing the number of fermion degrees of freedom can bring the running of the coupling to a halt. An infrared-attractive fixed point (IRFP) appears [@IRFP] and the theory exists in an infrared-conformal phase. The smallest number of flavors where the theory admits an IRFP is generally referred to as the “sill” of the so-called conformal window. With a number of flavors slightly below the sill, the theory is still chirally broken and confining. But it is different from QCD in being nearly conformal. More precisely, the beta function is very small near the energy scale where chiral symmetry breaking sets in. We say that the theory has a “walking,” rather than “running,” coupling. Lattice simulations of walking theories have revealed the presence of a flavor-singlet scalar meson that can be as light as the pions over a wide fermion-mass range (for a recent review, see Ref. [@TD]). Notable examples include the $SU(3)$ gauge theory with $N_f=8$ Dirac fermions in the fundamental representation [@LatKMI; @LSD] or with two flavors of sextet fermions [@LatHC]. We stress that, when dealing with a theory with a very small beta function, deciding whether the theory is chirally broken and confining, or, alternatively, infrared conformal, can be very challenging. Here we will assume that the models mentioned above are indeed chirally broken in the continuum limit. Walking theories have features which are attractive for extensions of the Standard Model that involve a new strong interaction. The renormalized coupling is changing very slowly with energy scale even when its value is rather large. As a result, one sometimes finds large anomalous dimensions, which, in turn, can lead to a very large enhancement of the corresponding operator. This feature is desired when trying to reconcile flavor physics with experiment (for reviews, see Refs. [@HS; @BSM; @JG]). Having a very light scalar is an added benefit, because, within the context of technicolor-like theories, it is a natural candidate for the Higgs particle. Walking theories are also theoretically interesting. In particular, it is natural to ask if the presence of the light singlet scalar meson is somehow connected to the smallness of the beta function. Indeed, the running of the coupling reflects the breaking of classical scale invariance by the quantum theory. When the beta function is small, the quantum breaking of dilatation symmetry is in some sense also small. Here we will discuss the construction of a low-energy effective action for the pions together with the light singlet scalar meson [@EFT]. A consistent low-energy description must account for all the light states, and must incorporate the scalar meson which can be as light as the pions. More generally, even if the pions will eventually become lighter than the scalar meson in the chiral limit, such an effective description is appropriate whenever the scalar meson is much lighter than all other states in the theory. The main challenge facing the construction is that, in order to build a systematic low-energy expansion, one has to quantify the violations of dilatation symmetry in the effective theory, and to be able to relate them to the microscopic theory in such a way that these violations are controlled by a small parameter. The light scalar, or “dilatonic meson,” then becomes a pseudo Nambu-Goldstone boson of the approximate dilatation symmetry. Building an effective field theory \[build\] ============================================ We start by reviewing the ingredients of standard chiral perturbation theory (for a review, see Ref. [@MGrev]). The massless microscopic theory has chiral symmetry, whose spontaneous breaking gives rise to Nambu-Goldstone bosons, the pions. When the fermions are given a non-zero mass, the pions become massive, too, but they remain the lightest asymptotic states as long as the fermion mass is small enough. Let us assume that we have $N_f$ Dirac fermions in the fundamental representation. This is a complex representation (when $N_c\ge 3$), and the symmetry breaking pattern is $SU(N_f)_L\times SU(N_f)_R \to SU(N_f)_V$, where $SU(N_f)_V$ is the diagonal subgroup. The lagrangian of the microscopic theory is $$\cl^{\rm MIC}(\c) = \frac{1}{4}F^2 + \bj \Sl{D}\j + \bj_R\c^\dagger \j_L + \bj_L \c \j_R \label{Lmic}$$ Here $\c$ is an $N_f\times N_f$ matrix-valued spurion, , an external source field. As usual, $\j_{R,L}=\half(1\pm\g_5)\j$ and $\bj_{R,L}=\half\bj(1\mp\g_5)$. Under a chiral rotation, the (dynamical) fermion fields and the (external) spurion field transform according to $$\begin{aligned} \j_{L,R} &\to& g_{L,R}\, \j_{L,R} \ , \qquad \bj_{L,R} \ \to\ \bj_{L,R}\, g_{L,R}^\dagger \ , \qquad \c \ \to\ g_L \,\c\, g_R^\dagger \ , \label{chiralrot}\end{aligned}$$ where $g_{L,R} \in SU(N_f)_{L,R}$. The lagrangian $\cl^{\rm MIC}(\c)$ is invariant when we apply the chiral transformation to all the fields including the spurion field. The lagrangian is also chirally invariant when we turn off the external source by setting $\c(x)=0$, and $\cl^{\rm MIC}(0)$ is recognized as the lagrangian of the massless theory. But we can also choose to set the chiral source to some non-zero “expectation value,” $\c(x)=m$. Now $\cl^{\rm MIC}(m)$ is no longer chirally invariant, and instead, under an infinitesimal chiral transformation we have $\d\cl^{\rm MIC}(m) = m \d(\bj\j)$, which exhibits the explicit (soft) breaking of chiral symmetry by the fermion mass term. We see the dual role of the chiral spurion. On the one hand, it encodes the explicit breaking of chiral symmetry coming from the mass term. On the other hand, it does so in a manner that assigns certain chiral transformation properties to the mass matrix itself, thereby rendering the lagrangian of the [*massive*]{} theory formally invariant. These same transformation properties will next be used to constrain the structure of the chiral lagrangian. At the leading order, the lagrangian of the low-energy effective theory is $$\cl^{\rm EFT} = \frac{f^2}{4}\, \tr(\partial_\m \S^\dagger \partial_\m \S) -\frac{f^2 B}{2} \, \tr\Big(\c^\dagger \S + \S^\dagger \c\Big) \ . \label{LChPT}$$ It depends on two low-energy constants (LECs): $f$ and $B$. The dynamical effective field $\S$ takes values in the coset $SU(N_f)_L\times SU(N_f)_R / SU(N_f)_V$, which is isomorphic to $SU(N_f)$. The effective field $\S_{ij}$ is loosely identified with the fermion bilinear $\tr(\j_{L,i} \bj_{R,j})$, and inherits its transformation properties, $$\S \to g_L \,\S\, g_R^\dagger \ . \label{transS}$$ It is easy to check that the chiral lagrangian (\[LChPT\]) is invariant under the combined transformation of Eqs. (\[chiralrot\]) and (\[transS\]). Setting $\c(x)=m>0$, it becomes $$\cl^{\rm EFT} = - f^2 B m N_f + \tr\big( (\partial_\m \p)^2 + 2m B\, \p^2 \big) + O(\p^4) \ , \label{treeprop}$$ where we have expanded the non-linear field $\S(x)=\exp(2i\p(x)/f)$ around its classical vacuum $\svev{\S}=1$. We see that at tree level, the pion mass is given by $M^2=2mB$. The other LEC, $f$, is the pion decay constant in the chiral limit (up to normalization conventions), as can be seen by coupling the effective theory to an external axial gauge field. Why does the leading-order chiral lagrangian (\[LChPT\]) contain just two terms? The chiral lagrangian provides a systematic expansion in the external momenta and in the fermion mass. Denoting by $\d$ the small expansion parameter, the power counting is $$p^2/\L^2 \ \sim \ m/\L \ \sim \ \d \ . \label{pcm}$$ Here $p^2$ stands for the inner product of any two external momenta. The reference scale is usually taken to be $\L =4\p f$. While being a dynamical, infrared scale of the microscopic theory, $\L$ may be identified with the ultraviolet cutoff of the chiral lagrangian. This works because the mass of the pions, which sets the energy scale probed by the effective lagrangian, tends to zero in the chiral limit. At the leading order, we allow for terms of order $\d^1$, and, after imposing the invariance under chiral symmetry, this leaves us with just the two operators we have in Eq. (\[LChPT\]). We have seen how the spurion $\c$ communicates information about the explicit breaking of chiral symmetry between the microscopic and the effective theories. More generally, by taking derivatives with respect to $\c(x)$ and $\c^\dagger(x)$ one defines a set of correlation functions that can be computed in both theories and compared. The LECs of the effective theory are fixed order by order in the chiral expansion (\[pcm\]) by requiring that the effective theory reproduce the correlation functions of the microscopic theory. We now turn our attention to scale transformations, which act on both the coordinates and the fields. Given some field $\F(x)$, its variation under an infinitesimal dilatation is $$\d\F = x_\m \partial_\m \F + s\, \F \ , \label{diltrans}$$ where $s$ is the scaling dimension of $\F$. In a theory containing gauge and fermion fields (but no elementary scalar fields) the dilatation current is given by $$S_\m = x_\n T_{\m\n} \ , \label{SxT}$$ where $T_{\m\n}$ is the energy-momentum tensor. Classically, the lagrangian of the massless theory transforms into a total derivative under an infinitesimal dilatation, and the dilatation current is conserved. Quantum mechanically, the dilatation current is not conserved. On shell, its divergence is equal to the trace of the energy-momentum tensor [@CDJ] $$\partial_\m S_\m = T_{\m\m} \equiv -T \ , \label{dS}$$ where $T = T_{cl} + T_{an}\,,$ and $$T_{cl}(m) = m \bj\j \ , \hspace{5ex} T_{an}(m) = \, \frac{\b(g^2)}{4g^2}\, F^2 + \g_m\, m\, \bj\j \ . \label{rentrace}$$ All quantities occurring on the right-hand side are the renormalized ones. $\b(g^2)$ is the familiar beta function, while $\g_m=\g_m(g^2)$ is the mass anomalous dimension. $T_{cl}$ is the classical divergence of the dilatation current, which vanishes if the fermion mass does. $T_{an}$ quantifies the quantum breaking of scale symmetry, reflected primarily in the running of the coupling. Following the example of chiral perturbation theory, our first task is to formally recover dilatation invariance of the microscopic theory. To this end we introduce a new spurion field $\s(x)$, which we will call the dilaton. Unlike the homogeneous transformation rule (\[diltrans\]), the infinitesimal variation of the dilaton field is $$\d\s = x_\m \partial_\m\s + 1 \ . \label{dildil}$$ The inhomogeneous term will play a crucial role below. The renormalized chiral source transforms like an ordinary field, with the same anomalous dimension as the renormalized mass, $$\d\c = x_\m \partial_\m\c + (1+\g_m) \c \ . \label{dilchi}$$ The lagrangian of the microscopic theory becomes $$\cl^{\rm MIC}(\s,\c) = \cl^{\rm MIC}(\c) + \s T_{an}(\c) + O(\s^2) \ , \label{Lmicsig}$$ where $T_{an}(\c)$ is obtained by the replacement $m\to\c(x)$ in Eq. (\[rentrace\]). The classical variation of the lagrangian is absent thanks to the scale transformation properties of the chiral source $\c$. Disregarding total derivatives, the variation of $\cl^{\rm MIC}(\c)$ is thus $-T_{an}(\c)$, which in turn is cancelled by the inhomogeneous term in Eq. (\[dildil\]) when we vary $\s T_{an}(\c)$. In order to cancel the terms proportional to $\s$ (as well as to higher powers of $\s$) in the variation of $\cl^{\rm MIC}(\s,\c)$, we would need the $O(\s^2)$ terms on the right-hand side of Eq. (\[Lmicsig\]). We will not attempt to derive these higher order terms, because they do not play any role in the following. In the case of the chiral lagrangian, we have seen that setting $\c(x)=0$ reproduces the massless theory, and, hence, exact chiral symmetry. The same is not true for scale symmetry. Setting $\c(x)=\s(x)=0$, the quantum variation of the massless theory becomes $-T_{an}(0)$, namely, the trace anomaly is $(\b(g^2)/(4g^2))\, F^2$. The massless quantum theory is not scale invariant, because the coupling runs. Moving on to the effective theory, we introduce a new effective field for the dilatonic meson, denoted $\t(x)$. Its transformation rule is similar to that of the external dilaton source, $$\d\t = x_\m \partial_\m\t + 1 \ , \label{diltau}$$ and again contains an inhomogeneous piece. Both $\s$ and $\t$ are inert under chiral transformations. As for the non-linear chiral field $\S$, its scaling dimension must be zero because it is unitary, and its variation under an infinitesimal dilatation is thus $$\d\S = x_\m \partial_\m\S \ . \label{dilS}$$ The next step is to construct the leading-order effective lagrangian. We are to write down all possible operators that depend on the effective fields, $\S$ and $\t$, and on the source fields, $\c$ and $\s$, which are invariant under chiral and scale transformations. As a first attempt, we follow the same power counting as for the chiral lagrangian, , we allow for all terms which are of order $\d^1$ according to Eq. (\[pcm\]). The resulting leading-order lagrangian is $$\tcl = \tcl_\p + \tcl_\t + \tcl_m + \tcl_d \ , \label{LeffL}$$ where $$\begin{aligned} \tcl_\p &=& \frac{f_\p^2}{4}\, V_\p(\t-\s)\, e^{2\t} \, \tr(\partial_\m \S^\dagger \partial_\m \S) \ , \label{LpV}\\ \tcl_\t &=& \frac{f_\t^2}{2}\, V_\t(\t-\s)\, e^{2\t} (\partial_\m \t)^2 \ , \label{LtV}\\ \tcl_m &=& -\frac{f_\p^2 B_\p}{2} \, V_M(\t-\s)\, e^{y\t} \, \tr\Big(\c^\dagger \S + \S^\dagger \c\Big) \ , \label{LmV}\\ \tcl_d &=& f_\t^2 B_\t \, V_d(\t-\s)\, e^{4\t} \ . \label{LdV}\end{aligned}$$ $\tcl_\p$ and $\tcl_\t$ are the kinetic terms for pions and for the dilatonic meson, respectively. $\tcl_m$ is a generalized chiral mass term, whereas $\tcl_d$ accounts for the self-interactions of the dilatonic meson. The presence of a separate set of $f$ and $B$ parameters for the pions and for the dilatonic meson is to be expected. As we discuss below, the exponent $y$ in Eq. (\[LmV\]) compensates for the dependence of the transformation rule of the renormalized chiral source on the mass anomalous dimension. The trouble with this new effective lagrangian is the occurrence of the potentials $V_\p$, $V_\t$, $V_M$ and $V_d$, each of which is an arbitrary function of its argument. The reason why these potentials are there is that the inhomogeneous terms in the variations of $\s$ and $\t$ cancel out in the difference $\t-\s$. As a result, any function $V(\t-\s)$ transforms homogeneously and has a scaling dimension equal to zero, much like the non-linear field $\S$. But unlike the $\S$-dependent terms, whose structure is constrained algebraically both by the unitarity of $\S$ and by the non-abelian nature of chiral symmetry, the [*abelian*]{} dilatation symmetry places no algebraic constraints on the form of the $V(\t-\s)$ potentials. At this point, our effort seems to have reached a dead end. The four potentials occurring in the leading-order lagrangian can be Taylor expanded, and the expansion coefficients amount to an infinite set of parameters. If all of them would remain in the leading-order lagrangian, then we will have lost any predictive power. To remedy this, we will reexamine the dynamics, seeking a way to extend the chiral power counting (\[pcm\]) to a more powerful one that will impose a power-counting hierarchy on the Taylor coefficients of these potentials. {width="8cm"} > [Two-loop beta function of the $SU(3)$ gauge theory with varying numbers $N_f$ of fundamental-representation flavors. The dashed vertical line at $g^2=\p^2\simeq 9.87$ marks the critical value $g_c^2$ of the coupling where, according to the gap equation, chiral symmetry breaking takes place in a walking theory.]{} A crude model \[model\] ======================= In this section we consider a crude model for the dynamics of $SU(N_c)$ gauge theories with $N_f$ fermions in the fundamental representation. As an approximation for the beta function we will consider the familiar two-loop expression [@IRFP], $$\frac{\partial g^2}{\partial \log\m} = -\frac{b_1}{16\p^2}\, g^4 -\frac{b_2}{(16\p^2)^2}\, g^6 \ . \label{2loop}$$ In Fig. \[betafn\] we have plotted the two-loop beta function for $N_c=3$ and various values of $N_f$. The $N_f=2$ curve shows how the beta function looks in a QCD-like theory. In this case the coefficients $b_1,b_2$ in Eq. (\[2loop\]) are both positive, and the running becomes faster with growing $g$. As the number of flavors $N_f$ increases, we reach a range where $b_1>0>b_2$ (for $N_c=3$ this range is given by $8.05 \ \leqx\ N_f < 16.5$). With $b_1>0$ the theory is still asymptotically free, and the beta function starts off negative. But as the coupling grows the screening effect of the fermions takes over. The beta function turns back and crosses the axis. The crossing point $g=g_*$ defines an IRFP. When $N_f$ is only slightly above the minimum needed to produce a negative $b_2$, the value of $g_*$ is very large. But it decreases monotonically with increasing $N_f$. As an analytic handle on chiral symmetry breaking we will use the gap equation. It predicts that in a walking theory, chiral symmetry breaking sets in when the coupling reaches the critical value [@HS] $$g_c^2 = \frac{4\p^2}{3 C_2} = \p^2 \ , \label{gcrit}$$ where the last equality is valid for the fundamental representation of $SU(3)$. Note that $g_c$ does not depend on the number of flavors. We are now ready to determine the “phase diagram.” First assume that $N_f$ is small enough that either there is no two-loop IRFP, or, if it exist, that $g_*(N_f)>g_c$. As we go down in energy scales, the coupling $g$ will grow, and chiral symmetry breaking (ultimately accompanied by confinement) will set in when $g$ reaches $g_c$. If, on the other hand, $N_f$ is large enough that $g_*(N_f)<g_c$, the running will come to a halt at the IRFP $g_*$. The renormalized coupling will never reach $g_c$, and the infrared physics will be conformal. Our crude dynamical model predicts that the conformal window occupies the range $N_f^* \le N_f \le (11/2) N_c$, where the sill of the conformal window, $N_f^*$, is the solution of $g_*(N_f^*)=g_c$. (In general $N_f^*$ is not an integer. The model suggests that $N_f^*$ is close to 12 for $N_c = 3$, but whether this is indeed the case is still under investigation.) Moreover, the dynamical model reveals an interesting feature of the chirally broken phase. As can be seen from Fig. \[betafn\], when $N_f<N_f^*$ and $N_f^*-N_f$ is not too large, the (negative) beta function at the critical coupling, $\b(g_c^2)$, is roughly proportional to $N_f-N_f^*$. This is the hint that will lead us to the desired power counting. Power counting \[showpc\] ========================= According to the model of the previous section, the beta function at the chiral symmetry breaking scale is $\b(g_c^2)$. This is a measure of the explicit breaking of dilatation symmetry felt by the low-energy sector. As $N_f$ is increased towards the sill of the conformal window, we expect this explicit breaking to vanish; for $N_f>N_F^*$, the infrared theory has an emergent conformal symmetry. Loosely speaking, what this means is that the small parameter controlling the explicit breaking of dilatation symmetry in the low-energy theory is $N_f-N_f^*$. But there is an obvious problem. $N_f$ takes integer values, and, unlike the fermion mass, we cannot tune $N_f-N_f^*$ continuously, nor can we actually reach the critical point $N_f=N_f^*$ since $N_f^*$ is not an integer. This problem can be solved for fermions in the fundamental representation by taking a suitable large-$N$ limit, the Veneziano limit. We assume that the number of flavors $N_f$ grows in proportion with the number of colors $N_c$, while the ratio $$n_f = N_f/N_c \ , \label{nf}$$ is held fixed. Based on the behavior of the two-loop beta function, we expect that the limit $$n_f^* = \lim_{N_c\to\infty} \frac{N_f^*(N_c)}{N_c} \ , \label{nfstar}$$ will be finite, where now $N_f^*(N_c)$ is an integer: the actual smallest number of flavors where the $SU(N_c)$ theory is infrared conformal. The small parameter we seek for our power counting is $n_f-n_f^*$. In the Veneziano limit, $n_f$ has effectively become a continuous parameter, and the Veneziano-limit sill of the conformal window can be reached by letting $n_f\to n_f^*$ from below. Of course, we must not forget that the increments we can make in $n_f$ cannot be parametrically smaller than $1/N_c$. The complete power counting we need is thus given by (with $N\equiv N_c$) $$p^2/\L^2 \ \sim \ m/\L \ \sim \ 1/N \ \sim \ |n_f-n_f^*| \ \sim \ \d \ . \label{pc}$$ For any large-$N$ limit, the appropriate coupling is the ’t Hooft coupling, which we take to be $\ta=g^2 N_c/(16\p^2)$. Notice that $\b(g^2)/(4g^2)=\b(\ta)/(4\ta)$. Our central [*hypothesis*]{} is that at the dynamical scale $\L$ where chiral symmetry breaks spontaneously, the beta function behaves like $$\b(\ta(\L)) = O(n_f-n_f^*) + O(1/N) \ . \label{tbgN}$$ As a consequence, $\b(\ta(\L))$ vanishes when the Veneziano limit followed by the limit $n_f\nearrow n_f^*$ are taken. We need to spend a moment to explain what $\L$ is. Let us reexamine Eqs. (\[LpV\]) and (\[LtV\]). If we disregard the potentials $V_\p$ and $V_\t$ (the justification for doing this will be explained shortly), the pion decay constant in the chiral limit is $\hf_\p = e^{v_0} f_\p$, where $v_0$ is the expectation value of the dilatonic meson field in the chiral limit. Similarly, the decay constant of the dilatonic meson itself is $\hf_\t = e^{v_0} f_\t$. Much like $\hf_\p$, the decay constant of the dilatonic meson is defined by the matrix element of the dilatation current between the vacuum and a one dilatonic-meson state. Alternatively, it can be defined from the matrix element of the energy-momentum tensor between the same states. Taking into account the behavior of these matrix elements in the Veneziano limit, we let $$\L \sim \frac{4\p \hf_\p}{\sqrt{N}} \sim \frac{4\p \hf_\t}{N} \ . \label{LIR}$$ Being $O(1)$ in large-$N$ counting, $\L$ is the characteristic scale for the masses of the lightest [*non-Goldstone*]{} mesons, which, in turn, provides the ultraviolet cutoff of the chiral lagrangian. How does the power counting (\[pc\]) constrain the potentials? Let us differentiate the lagrangian of the microscopic theory, Eq. (\[Lmicsig\]), with respect to the dilaton source $\s(x)$, and then set the sources to zero. We obtain $$\frac{\partial}{\partial \s(x)}\, \cl^{\rm MIC} \bigg|_{\s=\c=0} \ = \ T_{an}(x)\bigg|_{\c=0} \ = \ \frac{\b(\ta)}{4\ta}\, F^2(x) \ = \ O(\d) \ , \label{dsMIC}$$ where the last equality follows from our central assumption (\[tbgN\]). More generally, if we differentiate the partition function $Z^{\rm MIC}$ with respect to the $\s$ field $n$ times, and we are careful to do this at non-coinciding points, the resulting correlation function will be parametrically of order $\d^n$. On the effective field theory side, taking $n$ derivatives of the lagrangian with respect to $\s$ probes the $n$-th derivative of the potentials, $V^{(n)}$. In terms of the Taylor expansion $$V = \sum_{n=0}^\infty \frac{c_n}{n!}\, (\t-\s)^n \ , \label{expandV}$$ this probes $c_k$ for $k\ge n$. The idea is to match suitable correlation functions of the microscopic and the effective theory, setting $\s=0$ (and, if desired, $\c=0$ as well) in the end. It takes a detailed study to verify that one can constrain all the expansion coefficients of the potentials this way [@EFT]. The end result is that the Taylor coefficients are subject to the power-counting hierarchy $$c_n = O(\d^n) \ . \label{cn}$$ The alert reader will have noticed that we must allow for multiple $\s$ derivatives at the same spacetime point in the effective theory, but we disallow them in the microscopic theory. In fact, this is not a problem, because the effective theory deals with hadrons, which are not point-like objects; the effective theory cannot resolve spacetime distances smaller than $1/\L$. We use this opportunity to draw the attention of the reader to a subtle point concerning the power-counting proof of Ref. [@EFT]. While we expect the hierarchy (\[cn\]) to hold for generic (small) values of all of the expansion parameters (\[pc\]), the proof we have given in Ref. [@EFT] effectively invokes the Veneziano limit, in that it neglects all the $1/N$ corrections in Eq. (\[tbgN\]). Some other places in Ref. [@EFT] also tacitly neglect $1/N$ corrections, notably Sec. 4.4, where we discuss the tree-level theory in the limit $n_f\nearrow n_f^*$. The final result is that the leading-order lagrangian now consists of terms of order $\d$ according to the power counting (\[pc\]), with the expansion coefficients of the potentials subject to Eq. (\[cn\]). This allows us to discard $V_\p$, $V_\t$ and $V_M$, because $\tcl_\p$, $\tcl_\t$ and $\tcl_M$ are already $O(\d)$ without them. Only in $V_d$ do we need to go to the first non-trivial order in its expansion. After setting $\s=0$ and $\c=m$, the leading order lagrangian reads $$\cl = \cl_\p + \cl_\t + \cl_m + \cl_d \ , \label{Leff}$$ where $$\begin{aligned} \cl_\p &=& \frac{f_\p^2}{4}\, e^{2\t} \, \tr(\partial_\m \S^\dagger \partial_\m \S) \ , \label{Lp}\\ \cl_\t &=& \frac{f_\t^2}{2}\, e^{2\t} (\partial_\m \t)^2 \ , \label{Lt}\\ \cl_m &=& -\frac{f_\p^2 B_\p}{2} \, e^{y\t} \, m \, \tr(\S + \S^\dagger) \ , \label{Lm}\\ \cl_d &=& f_\t^2 B_\t \, e^{4\t} (c_0 + c_1\t) \ . \label{Ld}\end{aligned}$$ It remains to discuss the exponent $y$ in Eq. (\[Lm\]). Assuming that the transition into the conformal window is sufficiently smooth for $\g_m$, one can show that we need $\g_m=\g_m^*$ in the transformation rule of the renormalized chiral source, Eq. (\[dilchi\]), where $\g_m^*$ is the IRFP value of the mass anomalous dimension at the sill of the conformal window. As a result, $$y = 3 - \g_m^* \ . \label{gammay}$$ Present day numerical evidence suggests that $0\le \g_m^* \leqx 1$, and, therefore, $2\leqx y \le 3$. Tree level ========== In this section we consider the leading-order lagrangian for a given theory with fixed $N_c$ and $N_f$. We first discuss the classical vacuum of the dilatonic meson in the chiral limit. As follows from Eq. (\[Ld\]), for $m=0$ the dilatonic meson’s potential is $U(\t) = e^{4\t} (c_0 + c_1 \t)$ up to a dimensionful constant. This potential is bounded from below provided that $c_1>0$. The unique, global minimum of $U(\t)$ is $$v_0 = -1/4 - c_0/c_1 \ . \label{minV}$$ \[Like all LECs, the actual value of $c_1$ must be determined by matching the effective theory to the microscopic theory. Note that only products such as $c_0 B_\t$ or $c_1 B_\t$ have an invariant meaning, much like $m B_\p$ in the case of the standard chiral lagrangian. We use this freedom to assume $B_\t>0$. Self-consistency of the low-energy description then excludes a negative value for $c_1$.\] Observe that the classical vacuum would become ill-defined for $c_1=0$. This has the following interesting interpretation. The potentials $V(\t-\s)$ introduced in Sec. \[build\] originate from the explicit breaking of scale invariance in the massless microscopic theory. This is true, in particular, for $c_1$, which is the [*only*]{} LEC in the leading-order lagrangian coming from the expansion of the potentials (note that the lagrangian (\[Leff\]) becomes scale invariant if we set $m=c_1=0$). Thus, the stable classical vacuum of the effective theory ultimately owes its existence to the running of the coupling in the microscopic theory. This should not come as a surprise, because, if the vacuum has a preferred scale (as opposed to a vacuum with no characteristic scale, or a continuous manifold of vacua with a gradually changing characteristic scale), then the theory cannot have exact scale invariance. The tree-level mass of the dilatonic meson in the chiral limit is $$m_\t^2 = 4 c_1 e^{2v_0} B_\t \ . \label{mtau}$$ If we consider the ratio of the dilatonic meson’s mass and decay constant $\hf_\t = e^{v_0} f_\t$, we get $$% \frac{N^2 m_\t^2}{\hf_\t^2} = 4 c_1 \frac{N^2 B_\t}{f_\t^2} \ , N^2 m_\t^2/\hf_\t^2 = 4 c_1 N^2 B_\t/f_\t^2 \ , \label{mftau}$$ in which the dependence on $v_0$ cancels out. \[The role of the factor of $N^2$ on both sides is to undo the large-$N$ dependence of the decay constant of the dilatonic meson, thereby keeping the ratio finite in the Veneziano limit (compare Eq. (\[LIR\])).\] Recall that $c_1=O(\d)$ according to Eq. (\[cn\]). It follows that $m_\t \sim \d^{1/2}$. This resembles the familiar behavior of the pion mass in ordinary chiral perturbation theory, $m_\p\sim m^{1/2}$. We next consider the classical vacuum $v(m)$ for $m>0$. It is implicitly given by $$\frac{f_\p^2 B_\p N_f y m}{f_\t^2 B_\t c_1} = 4 v_1(m) e^{(4-y)v(m)} \ , \label{solvesdl}$$ where $v_1(m) = v(m)-v_0$. Generically, $v_1(m)$ is $O(1)$, because $c_1 \sim m \sim \d$ by the power counting. One can check that $v_1(m)>0$ for $m>0$, and that $v(m)$ is a monotonically increasing function. Using Eq. (\[gammay\]), the tree-level masses of the dilatonic meson and the pion are $$\begin{aligned} m_\t^2 &=& 4c_1 B_\t e^{2v(m)} (1+(1+\g_m^*)v_1(m)) \ , \label{mintreetau}\\ m_\p^2 &=& 2m B_\p e^{(1-\g_m^*)v(m)} \ = \ \frac{8 c_1 f_\t^2 B_\t}{y f_\p^2 N_f}\, e^{2v(m)} v_1(m) \ . \label{mintreepion}\end{aligned}$$ Both $m_\t$ and $m_\p$ are monotonically increasing with $m$. Interestingly, the dependence of the tree-level pion mass on the fermion mass $m$ would reduce to that of ordinary chiral perturbation theory, if $\g_m^*$ happened to be equal to 1, which is the favored value according to the gap-equation analysis. For any other value of $\g_m^*$, Eq. (\[mintreepion\]) furnishes us with a prediction of the low-energy theory that distinguishes it from ordinary chiral perturbation theory. Approaching the sill of the conformal window \[sill\] ===================================================== In this section we study the tree-level predictions of the effective theory as the sill of the conformal window is approached. To avoid technical complications, we will consider only the chiral limit, $m=0$. Also, as was done in Ref. [@EFT], we will take the Veneziano limit, thereby neglecting the $1/N$ corrections in Eq. (\[tbgN\]). In the Taylor series for the potentials (\[expandV\]), each coefficient $c_n$ can in itself be expanded as a power series in $n_f-n_f^*$, $$c_n = \sum_{k=n}^\infty \tc_{nk} (n_f-n_f^*)^k \ . \label{expandcn}$$ The lower limit of the summation comes from the power-counting hierarchy (\[cn\]) (remember that $n_f-n_f^* \sim \d$). In particular, the tree-level potential in Eq. (\[Ld\]) becomes $$V_d(\t) = c_0 + c_1 \t = \tc_{00} + (n_f-n_f^*)(\tc_{01} + \tc_{11} \t) \ . \label{Vdnf}$$ Since $n_f<n_f^*$ for chirally broken theories, the constraint $c_1>0$ translates into $\tc_{11}<0$. We may ask what happens if we attempt to apply the low-energy expansion to a theory that lives [*inside*]{} the conformal window. Assuming $n_f>n_f^*$, we see that $c_1=(n_f-n_f^*)\tc_{11}$ becomes negative. A a result, the classical potential becomes unbounded from below. The conclusion is that the effective theory breaks down inside the conformal window. This is as it should be, because there is no spontaneous breaking of chiral symmetry inside the conformal window. In this sense, the limit $n_f\nearrow n_f^*$ is qualitatively different, and more singular, than the chiral limit $m\to 0$. Let us next examine the dependence of a few observables on $n_f-n_f^*$. Since we will be comparing observables belonging to different theories, we must compare dimensionless quantities. The dependence on $n_f-n_f^*$ may come directly from $c_1=(n_f-n_f^*)\tc_{11}$, or it can also arise from the behavior of the classical vacuum $v_0$. In fact, we already have one such example, namely, the ratio $N m_\t/\hf_\t$ in the chiral limit, given in Eq. (\[mftau\]). In this case there is no dependence on $v_0$, and the dependence on $n_f-n_f^*$ comes only from $c_1$. Before moving on, it is convenient to use the freedom to shift the $\t$ field by a constant, $\t \to \t + \D$, in order to simplify the expression for $v_0$. Given that $n_f-n_f^*$ is one of the small expansion parameters, we take $\D$ to be independent of $n_f-n_f^*$ so as not to obscure the power counting. Substituting in Eq. (\[Vdnf\]) we see that the shift has the effect of changing $\tc_{01} \to \tc_{01} + \tc_{11} \D$, while $\tc_{00}$ and $\tc_{11}$ are unchanged. We will use this freedom to set $\tc_{01}=0$. (The remaining dependence of the lagrangian (\[Leff\]) on the shift $\D$ is absorbed into redefinitions of the $f$’s and $B$’s.) The classical vacuum of the $m=0$ theory thus becomes (compare Eq. (\[minV\])) $$% v_0 = -\frac{1}{4} -\frac{\tc_{00}}{\tc_{11}(n_f-n_f^*)} \ . v_0 = -1/4 -\tc_{00}/(\tc_{11}(n_f-n_f^*)) \ . \label{minVnf}$$ We comment in passing that the dependence of the physical decay constants, $\hf_\p = e^{v_0} f_\p$ and $\hf_\t = e^{v_0} f_\t$, on $v_0$ suggests that we should have $v_0\to -\infty$ for $n_f\nearrow n_f^*$, which in turn requires $\tc_{00}>0$. Appealing as this may be, however, we have not been able to prove this assertion, basically because it involves the comparison of dimensionful quantities of different theories. As our second example we consider the fermion condensate, measured in units of $\hf_\p$. We find $$\frac{\svev{\bj\j}}{\hf_\p^3} = - \frac{B_\p N_f}{f_\p}\, e^{-\g_m^* v_0} \ , \label{condscale}$$ where $v_0$ is now given by Eq. (\[minVnf\]), and where we have used that the tree-level condensate is $$\svev{\bj\j} = -f_\p^2 B_\p N_f\, e^{yv_0} \ . \label{eftbjj}$$ Assuming that $\tc_{00}>0$ (and that $\g_m^*>0$ as well), Eq. (\[condscale\]) predicts an enhancement of the fermion condensate for $n_f\nearrow n_f^*$, which, apart from the familiar dependence on the mass anomalous dimension, depends also on the LECs $\tc_{00}$ and $\tc_{11}$ through Eq. (\[minVnf\]). The low-energy effective theory provides us with a quantitative description of the (pseudo) Nambu-Goldstone sector in the chirally broken phase. But it does not give us any access to physics inside the conformal window, nor to the dynamics of a chirally broken theory at any energy scale which is comparable to or larger than $\L$. We may gain some qualitative understanding of the transition into the conformal window by using the dynamical model of Sec. \[model\]. This consists of using the two-loop beta function, combined with the prediction of the gap equation for the critical coupling that triggers chiral symmetry breaking. Here we add a new element, namely, we will use this dynamical model in the Veneziano limit, where, in terms of the ’t Hooft coupling introduced in Sec. \[showpc\], the critical coupling is $\ta=1/6$ (for fermions in the fundamental representation). In the Veneziano limit, one can express the two-loop beta function as $$\b(\ta) = -\left( \frac{1}{6}-\ha \right)^2 \left( \ha + \frac{\hn}{3} \left( \frac{25}{6} - 13\ha \right) \right) \ , \label{betasill}$$ where we wrote $\ta = 1/6-\ha = \ta_c-\ha$ and $n_f = 4-\hn$. At the chiral symmetry breaking scale $\ta(\L)=\ta_c$, which corresponds to $\ha=0$. The beta function then satisfies $\b(\ta_c)\propto\hn$. It follows that the sill of the conformal window is at $n_f^* = 4$, and that the conformal window is $4<n_f<11/2$. (For $n_f>11/2$ asymptotic freedom is lost.) We next introduce a new reference scale denoted $\L_{nc}$, where the subscript “$nc$” stands for “nearly-conformal.” It is defined in the massless theory by the condition that $$\b(\ta(\L_{nc})) = -\e_0 \ , \label{Lnc}$$ for some fiducial value $0<\e_0\ll 1$. Eq. (\[Lnc\]) is supplemented by the additional instruction that $\L_{nc}$ is to be found by starting in the deep infrared, and then increasing the scale till Eq. (\[Lnc\]) is satisfied. \[This additional instruction is needed to avoid the second occurrence of $\b(\ta) = -\e_0$ in the vicinity of the gaussian fixed point, as is visible, for example, in the $N_f=12$ or $N_f=13$ curves in Fig. 1.\] Because it relies on the beta function, the criterion (\[Lnc\]) make sense only if its solution $\L_{nc}$ is large compared to any dynamical infrared scale that may be induced in the massless theory. The scale $\L_{nc}$ thus always exists for theories inside the conformal window, where no dynamical infrared scale is generated. In the chirally broken phase, our dynamical model predicts that $\L_{nc}$ exists provided that $n_f$ is close enough to $n_f^*$, so that at the critical coupling, $|\b(\ta_c)|<\e_0$. Moreover, because $\b(\ta_c)$ tends to zero when $n_f$ tends to $n_f^*$, it follows that the ratio $\L/\L_{nc}$ also tends to zero in this limit. Let us now distinguish three regions for the fermion mass: $$\begin{aligned} {\rm I}:\ \L \ll m \ll \L_{nc} \ , \qquad {\rm II}:\ m \sim \L \ , \qquad {\rm III}:\ m \ll \L \ . \nonumber\end{aligned}$$ Region III is where the low-energy expansion is valid. The theory has both approximate chiral symmetry and approximate dilatation symmetry, both of which are spontaneously broken. In Region I, chiral symmetry and dilatation symmetry are both explicitly broken by the fermion mass, but this breaking is soft. Because of the smallness of the beta function, what we expect to see in Region I is the characteristic behavior of a [*mass-perturbed conformal system*]{}. This implies that the masses of all mesons behave like (see  Ref. [@DDZ]) $$M \sim \L \left(m/\L\right)^{\frac{1}{1+\g_m^*}} \ . \label{Mscale}$$ The transition between the conformal and chirally broken behavior occurs in Region II. Once $m$ goes below $\L$, we enter the chiral regime. The masses of all non-Goldstone mesons freeze out at $\otherM\sim\L$, while the masses of the pseudo Nambu-Goldstone mesons behave like $$M^2_{\rm pNGB} \ = \ \left[ O(n_f-n_f^*) + O(m/\L)\right] \L^2 \ \ll \ \L^2 \ \sim \ \otherM^2 \ . \label{M2L}$$ We see that as $n_f$ tends to $n_f^*$ from below, the masses of [*all*]{} mesons in the massless theory tend to zero, if measured in units of $\L_{nc}$. But the masses of the pseudo Nambu-Goldstone mesons vanish faster; the smallness of the ratio $M_{\rm pNGB}/\otherM$ is what allows for the existence of a systematic low-energy description. Notice that in order to stay in the chiral regime when $n_f$ gets closer to $n_f^*$ we must keep decreasing $m$. This is because we must maintain $m/\L\ll 1$, and $\L/\L_{nc}$ vanishes at the conformal sill. It is also useful to consider what happens if we hold $m$ fixed in units of $\L_{nc}$. Regardless of whether $n_f$ is smaller or larger than $n_f^*$, all theories where $|n_f-n_f^*|\ll 1$ then have a wide region where the theory exhibits the typical behavior of a mass-perturbed conformal system. The difference between $n_f>n_f^*$ and $n_f<n_f^*$ is that in the former case, the mass-perturbed conformal behavior exists for any $m\ll \L_{nc}$, regardless of how small $m$ is. By contrast, for $n_f<n_f^*$ this behavior exists only in Region I: $\L \ll m \ll \L_{nc}$, which is bounded from below. As $n_f$ approaches the sill $n_f^*$, the range of fermion mass where the theory exhibits a mass-perturbed conformal behavior keeps expanding because $\L/\L_{nc}$ gets smaller, until eventually at $n_f=n_f^*$ we have $\L/\L_{nc}\to 0$, and the chirally broken behavior is completely lost. The physical picture that emerges is that, if we always use $\L_{nc}$ as the reference scale, and the fermion mass is kept at some fixed value in units of $\L_{nc}$, then the physical spectrum will vary continuously as we dial $n_f$ upwards, across $n_f^*$ and into the conformal window. In this sense, the transition into the conformal window is smooth. [**Acknowledgments**]{} We thank David B. Kaplan for raising questions about the behavior of the theory at the transition to the conformal phase. This material is based upon work supported by the U.S. Department of Energy, Office of Science, Office of High Energy Physics, under Award Number DE-FG03-92ER40711. YS is supported by the Israel Science Foundation under grant no. 449/13. [99]{} W. E. Caswell, Phys. Rev. Lett.  [**33**]{}, 244 (1974). T. Banks and A. Zaks, Nucl. Phys.  B [**196**]{}, 189 (1982). T. DeGrand, Rev. Mod. Phys.  [**88**]{}, 015001 (2016) \[arXiv:1510.05018 \[hep-ph\]\]. Y. Aoki [*et al.*]{} \[LatKMI Collaboration\], Phys. Rev. D [**89**]{}, 111502 (2014) \[arXiv:1403.5000 \[hep-lat\]\]. T. Appelquist [*et al.*]{}, Phys. Rev. D [**93**]{}, no. 11, 114514 (2016) \[arXiv:1601.04027 \[hep-lat\]\]. Z. Fodor, K. Holland, J. Kuti, S. Mondal, D. Nogradi and C. H. Wong, PoS LATTICE [**2015**]{}, 219 (2016) \[arXiv:1605.08750 \[hep-lat\]\]. C. T. Hill and E. H. Simmons, Phys. Rept.  [**381**]{}, 235 (2003) Erratum: \[Phys. Rept.  [**390**]{}, 553 (2004)\] \[hep-ph/0203079\], and references therein. R. Contino, arXiv:1005.4269 \[hep-ph\]. B. Bellazzini, C. Csáki and J. Serra, Eur. Phys. J. C [**74**]{}, 2766 (2014) \[arXiv:1401.2457 \[hep-ph\]\]. G. Panico and A. Wulzer, Lect. Notes Phys.  [**913**]{}, 1 (2016) \[arXiv:1506.01961 \[hep-ph\]\]. J. Giedt, Int. J. Mod. Phys. A [**31**]{}, no. 10, 1630011 (2016) \[arXiv:1512.09330 \[hep-lat\]\]. M. Golterman and Y. Shamir, Phys. Rev. D [**94**]{}, no. 5, 054502 (2016) \[arXiv:1603.04575 \[hep-ph\]\]. M. Golterman, arXiv:0912.4042 \[hep-lat\]. J. C. Collins, A. Duncan and S. D. Joglekar, Phys. Rev. D [**16**]{}, 438 (1977). L. Del Debbio and R. Zwicky, Phys. Rev. D [**82**]{}, 014502 (2010) \[arXiv:1005.2371 \[hep-ph\]\].

ArXiv

arXiv:yymm.nnnn \[hep-th\] [Roberto Casalbuoni$^1$, Federico Elmetti$^2$,\ Simon Knapen$^{2,3}$ and Laura Tamassia$^2$]{} [\ [**`casalbuoni@fi.infn.it`**]{}\ $^2$ [*Instituut voor Theoretische Fysica, Katholieke Universiteit Leuven,\ Celestijnenlaan 200D B-3001 Leuven, Belgium*]{}\ [**`federico.elmetti@alice.it`**]{}, [**`laura.tamassia@fys.kuleuven.be`**]{},\ $^3$ [*Department of Physics and Astronomy, Rutgers University,\ Piscataway, NJ 08855-0849, USA*]{}\ [**`knapen@physics.rutgers.edu`**]{}\ ]{} [**Abstract**]{} We study some field representations of vector supersymmetry with superspin $Y=0$ and $Y=1/2$ and nonvanishing central charges. For $Y=0$, we present two multiplets composed of four spinor fields, two even and two odd, and we provide a free action for them. The main differences between these two multiplets are the way the central charge operators act and the compatibility with the Majorana reality condition on the spinors. One of the two is related to a previously studied spinning particle model. For $Y=1/2$, we present a multiplet composed of one even scalar, one odd vector and one even selfdual two-form, which is a truncation of a known representation of the tensor supersymmetry algebra in Euclidean spacetime. We discuss its rotation to Minkowski spacetime and provide a set of dynamical equations for it, which are however not derived from a Lagrangian. We develop a superspace formalism for vector supersymmetry with central charges and we derive our multiplets by superspace techniques. Finally, we discuss some representations with vanishing central charges. Introduction\[sec:0\] ===================== In this paper we study some field representations of the vector supersymmetry algebra, a graded extension of the Poincar[é]{} algebra in four dimensions. The extension is realized by adding to the Poincaré algebra two fermionic operators, an odd Lorentz vector and an odd Lorentz scalar. Furthermore, two central charges are allowed. The anticommutator between vector and scalar odd generators gives the four-momentum vector, from which the name vector supersymmetry (VSUSY). To our knowledge, this algebra was first introduced in [@Barducci:1976qu] in 1976, with the purpose of obtaining a pseudoclassical description of the Dirac equation. Its general algebraic properties have been studied in [@Casalbuoni:2008ez] (see also [@Casalbuoni:2009en]). The VSUSY algebra, or, better, an extension of it, arises in the context of topological field theories. In fact, an Euclidean version of VSUSY appears as a subalgebra of the symmetry algebra underlying topological $\mathcal{N}=2$ Yang-Mills theories. Supersymmetry with odd vector generators was studied after Witten [@Witten:1988ze], who, in 1988, introduced topological $\mathcal{N}=2$ Yang-Mills theories by performing a topological twist. After this twist, the fermionic generators become a vector, a scalar and an anti-selfdual tensor [@Alvarez:1994ii; @Kato:2005fj]. After truncation of the anti-selfdual sector, the twisted algebra coincides with the Euclidean VSUSY algebra, in the special case when the two central charges of VSUSY are identified. Twisted topological algebras have proven to be useful in the study of renormalization properties of topological field theories [@Birmingham:1988bx; @Delduc:1989ft]. Moreover, a superspace formalism has been developed for these topological theories, see for example [@Alvarez:1994ii; @Kato:2005fj; @Baulieu:2008at] and references therein. The main difference between vector and ordinary supersymmetry is that the odd generators of VSUSY have integer spin and so they do not satisfy the usual spin-statistics relation. This implies that in any representation of VSUSY some of the component fields, counting for half of the degrees of freedom of the multiplet, necessarily violate the usual spin-statistics relation and in a field theory setting should be identified with ghosts. VSUSY then unifies physical fields with ghosts and not fields of integer spin with fields of half-integer spin. In particular, a VSUSY multiplet always contains either only fields of integer spin or only fields of half-integer spin. Therefore, while VSUSY’s algebraic structure is very similar to the one of ordinary supersymmetry, it has completely different implications. Ghost fields are not observed but are nonetheless a very important technical tool in field and string theories. For this reason, VSUSY representations and dynamical models with underlying VSUSY are worth exploring. In any case, it is interesting to compare this alternative to ordinary supersymmetry to understand what the essential ingredients in supersymmetry are. In [@Casalbuoni:2008ez], it has been shown that the irreducible representations of VSUSY can be classified according to the value assumed by the superspin Casimir operator, in complete analogy to the case of standard supersymmetry. An irreducible multiplet with a given value of the superspin $Y$ contains components of Lorentz spin $s=|Y\pm \frac 12|$ for $Y\not =0$, whereas for $Y=0$ the components have Lorentz spin $\frac 1 2$. In this paper we explicitly construct field representations corresponding to the two lowest superspin values, $Y=0$ and $Y= \frac 12$. In the case $Y=0$, we find two off-shell multiplets both composed of four Dirac spinors, two even and two odd. The first one is compatible with Dirac-type equations of motion and is realized in the spinning particle model in [@Casalbuoni:2008iy]. The second one differs from the first because of the nontrivial action of the central charge operator on the fields. Moreover, it is not compatible with Dirac equations of motion, but only with Klein-Gordon ones. In the case $Y=\frac 12$, in Euclidean space, we construct a multiplet with a scalar, a vector and a selfdual two-form. The scalar and the two-form are even and the vector is odd, or the other way around. We show that this is a truncation of a representation of the topologically twisted $\mathcal {N}=2$ theory given in [@Kato:2005gb] (see also [@Kato:2008dw]). Since in this paper we are mainly interested in representations in Minkowski spacetime, we discuss the existence of a similar multiplet in Minkowskian signature. We find that in Minkowski spacetime the number of degrees of freedom must be doubled, since the fields must be necessarily complex. The resulting Minkowskian multiplet features a complex scalar, a complex vector and a selfdual and an antiselfdual two-forms related to each other by complex conjugation. This is of course equivalent to having two real scalars, two real vectors and one real two-form. For all multiplets we give the invariant free dynamical equations, which for $Y=0$ multiplets can also be derived from an action. Furthermore, we develop a superspace setup for VSUSY and we rederive our results in components from superfields. To do that, we adapt the superspace with central charge first introduced by Sohnius in [@Sohnius:1978fw] to VSUSY. This kind of extended superspace allows us to also derive multiplets where the central charge operators act nontrivially on the fields. Concerning the derivation of actions from superspace, we only take a first step by choosing to work with a fixed value of the central charge. In this way we derive the action for one of our $Y=0$ multiplets. Finally, we consider the case of vanishing central charges and we find some examples of field representations by superspace techniques. In the most interesting case of nonvanishing central charges, we limit our study to free fields. We leave interacting theories for future work. In the case with vanishing central charge we find an interacting action for one of our multiplets. However, this action has the unusual property of being odd. The need for this kind of actions for some supersymmetric models in the case of ordinary supersymmetry has already been pointed out in [@Soroka:1995et], [@Soroka:2001jg]. However, the quantization of these models remains to our knowledge problematic and goes in any case beyond the scope of this paper. Part of the results presented in this paper are also discussed in the master thesis of Simon Knapen [@thesis]. The paper is organized as follows.\ In Section 2 we state our conventions and we review the results obtained in [@Casalbuoni:2008ez] with particular emphasis on the classification of irreducible representations of VSUSY in the case of nonvanishing central charges.\ In Section 3 we present two multiplets with $Y=0$. Furthermore, we introduce VSUSY superspace with central charges and we derive both multiplets by superspace techniques.\ In Section 4 we present a $Y=\frac 12$ Euclidean on-shell multiplet, we discuss its rotation to Minskowski spacetime and its superspace origin.\ In Section 5 we initiate the study of the construction of VSUSY invariant actions by superspace techniques by considering the case of the spinning particle $Y=0$ multiplet.\ In Section 6 we study some representations of VSUSY with vanishing central charges.\ In Section 7 we give our conclusions and outlook.\ In Appendix A we give more details on the relation between one of our multiplets with $Y=0$ and the spinning particle model of [@Casalbuoni:2008iy].\ In Appendix B we explicitly show how to solve the first superspace constraint equation encountered in Section 3.\ In Appendix C we discuss the relation between the $Y=1/2$ VSUSY multiplet of Section 4 and a tensor supersymmetry multiplet found in [@Kato:2005gb] in the context of ${\cal N}=2$ twisted topological models. VSUSY algebra and Casimir operators =================================== Conventions ----------- Vector supersymmetry (VSUSY) algebra is a graded extension of the Poincar[é]{} algebra in four dimensions. The extension is realized by adding to the Poincaré algebra two fermionic operators, an odd Lorentz vector and an odd Lorentz scalar. Furthermore, two central charges are allowed. We work with Minkowski metric unless stated otherwise. The VSUSY commutation relations used are: $$\begin{aligned} &&[M_{\mu\nu},M_{\rho\sigma}]=-i\eta_{\nu\rho}M_{\mu\sigma}- i\eta_{\mu\sigma}M_{\nu\rho}+i\eta_{\nu\sigma}M_{\mu\rho}+ i\eta_{\mu\rho}M_{\nu\sigma}\,; \nonumber\\ &&[M_{\mu\nu},P_\rho]=i\eta_{\mu\rho}P_\nu-i\eta_{\nu\rho}P_\mu\,;\qquad [M_{\mu\nu},Q_\rho]=i\eta_{\mu\rho}Q_\nu-i\eta_{\nu\rho}Q_\mu\,; \nonumber\\ &&\{Q_{\mu},Q_{\nu}\}=Z \eta_{\mu\nu}\,;\quad \quad \{Q_{5},Q_{5}\}=\tilde{Z}\,;\quad \quad \{Q_{\mu},Q_{5}\}=-P_{\mu}\,. \label{VSUSYalgebra}\end{aligned}$$ In [@Casalbuoni:2008ez] it was shown that the VSUSY algebra can be derived by contraction from the simple orthosymplectic algebra $\rm OSp(3,2|2)$. The name we have chosen for the scalar odd generator, $Q_{5}$, is a reminder of its five-dimensional origin. The algebra will be realized in the form (\[VSUSYalgebra\]) in terms of differential operators acting on superfields. As usual, it will be realized with opposite signs in the RHS on component fields.\ We follow the conventions: $$\begin{aligned} &\eta_{\mathcal{\mu\nu}} = \rm diag\{-1,+1,+1,+1\}\,;\quad \quad {\epsilon}^{0123}=-{\epsilon}_{0123}=1\,; \nonumber\\ &P_{\mu}= i\partial_{\mu}\,;\quad \quad \Box \equiv \partial^{\mu} \partial_{\mu}= {\not\hspace{-.05cm}\partial}{\not\hspace{-.05cm}\partial}= -P^{2}\,;\quad\quad\{\gamma_{\mu},\gamma_{\nu}\}=2\eta_{\mathcal{\mu\nu}}\,;\nonumber\\ &\gamma_{0}^{\dag}=-\gamma_{0}\,; \quad \quad \gamma_{i}^{\dag}=\gamma_{i}\,; \quad \quad \gamma_{0}^{-1}\gamma_{\mu}^{\dag}\gamma_{0}=-\gamma_{\mu}\,; \quad\quad \gamma_{5}\equiv \frac{i}{4!}\epsilon_{\mu\nu\rho\sigma}\gamma^{\mu}\gamma^{\nu}\gamma^{\rho}\gamma^{\sigma}\,.\cr & \label{conv}\end{aligned}$$ Therefore we have the mass shell condition $P^2=-m^2$ and the Dirac equation is of the form $({\not\hspace{-.05cm}\partial}+m)\psi=0$.\ In this paper we discuss multiplets including both physical fields and ghosts. In general, to avoid confusion, we will denote fermionic fields by a tilde .\ For our conventions concerning spinor fields, discussed next, we follow [@VanProeyen:1999ni].\ For the complex conjugation of a bilinear in two fermionic fields $\tilde A$ and $\tilde B$, we adopt the convention of exchanging the position of the fields, i.e. $$\left(\tilde{A}\tilde{B}\right)^\dagger=\tilde{B}^\dagger \tilde{A}^\dagger\,. \label{fermionbil}$$ We define the infinitesimal VSUSY transformations of the fields as $$\delta_{Q_{5}} \psi=i{\epsilon}_5 Q_{5}\psi~~~{\rm and}~~~ \delta_{Q_{{\mu}}}\psi=i{\epsilon}^{{\mu}} Q_{{\mu}} \psi. \label{fulltrans}$$ with ${\epsilon}_5$ and $\epsilon_{\mu}$ real odd parameters. For the central charge operators, we define $$\delta_{Z} \psi=i{\epsilon}_Z Z\psi~~~{\rm and}~~~ \delta_{\tilde Z}\psi=i{\epsilon}_{\tilde Z} \tilde{Z} \psi. \label{fulltransZ}$$ where the even parameters ${\epsilon}_Z$ and ${\epsilon}_{\tilde Z}$ are also taken to be real. For Dirac conjugate fields and their transformations, we have the following conventions $$\begin{aligned} &\bar\psi= i\psi^\dagger \gamma_0\,, \nonumber\\ &\delta_{Q} \psi^\dagger=(\delta_{Q} \psi)^\dagger=-i (Q\psi)^\dagger {\epsilon}^\dagger\,. \label{daggerconv}\end{aligned}$$ We use the following definition of the Majorana condition for spinors: $$i\psi^{\dag}\gamma_{0}=\psi^{T}\mathcal{C}, \label{majorana}$$ where the charge conjugation matrix $\mathcal{C}$ satisfies $$\mathcal{C}^{T}=-\mathcal{C}\,; \quad \quad \mathcal{C}^{-1}\gamma_{\mu}^{T}\mathcal{C}=-\gamma_{\mu}.$$ As a result, for anti-commuting Majorana spinors we have the following identities: $$\overline{\tilde{\chi}}\tilde{\xi}=\overline{\tilde{\xi}}\tilde{\chi}\,; \quad \quad \quad \quad \overline{\tilde{\chi}}\gamma_{\mu}\tilde{\xi}=-\overline{\tilde{\xi}}\gamma_{\mu}\tilde{\chi}\,; \quad \quad \quad \quad\int \overline{\tilde{\chi}}{\not\hspace{-.05cm}\partial}\tilde{\xi}=\int \overline{\tilde{\xi}}{\not\hspace{-.05cm}\partial}\tilde{\chi},$$ while for commuting Majorana spinors we have: $$\overline{\psi}\lambda=-\overline{\lambda}\psi\,; \quad \quad \quad \quad \overline{\psi}\gamma_{\mu}\lambda=\overline{\lambda}\gamma_{\mu}\psi\,; \quad \quad \quad \quad \int \overline{\psi}{\not\hspace{-.05cm}\partial}\lambda=-\int \overline{\lambda}{\not\hspace{-.05cm}\partial}\psi.$$ In practice, it is useful to rephrase (\[majorana\]) in terms of the C-operation defined as follows on a spinor $\psi$ $$\psi^C=iB^{-1}\psi^* ~,~ {\rm where}~~~ B=-\mathcal{C} \gamma_0\,. \label{Cop}$$ The Majorana condition is then simply rewritten in the form $$\psi=\psi^C \,.$$ In this case we say that the spinor $\psi$ is real. For a general matrix $M$ in spinor space the C-operation is defined as follows: $$M^C= B^{-1}M^* B\,.$$ The gamma matrices behave as real matrices $$(\gamma_{\mu})^C=\gamma_{\mu}$$ and the matrix $\gamma_5$ as purely imaginary $$(\gamma_5)^C=-\gamma_5\,.$$ In the following, we will be interested in checking whether our spinorial VSUSY multiplets are compatible with a Majorana reality condition on the spinors. A multiplet is compatible with the Majorana condition when for a real spinor $\psi$ the variation $\delta\psi$ is also real. Since we are dealing with both fermionic and bosonic spinors in this paper, the Majorana condition must be investigated separately in the two cases. If $\psi$ is real and [*odd*]{} (physical), then requiring compatibility with the Majorana condition is equivalent to asking that $Q_{\mu}\psi$ and $Q_5 \psi$ are purely imaginary. If $\psi$ is real and [*even*]{} (ghost), then requiring compatibility with the Majorana condition is equivalent to asking that $Q_{\mu}\psi$ and $Q_5 \psi$ are real. From the algebra (\[VSUSYalgebra\]), $\{Q_{\mu},Q_{\nu}\}=\eta_{{\mu}{\nu}} Z$, one can easily derive that $Z\psi$ has to be purely imaginary when $\psi$ is real, both in the even and odd cases. An analogue conclusion can be drawn for $\tilde Z$. Therefore the factor of $i$ present in the definitions (\[fulltransZ\]) is needed for consistency of the reality conditions on the spinors with the algebra. Equivalence classes of irreducible representations with different values of the central charges ----------------------------------------------------------------------------------------------- As already noted in [@Casalbuoni:2008ez], for irreducible representations where, in a suitable basis, the central charges can be treated as numbers, it is possible to make a rescaling such that in general only the value of one central charge and the relative sign between the two is relevant. In fact, by implementing into the algebra (\[VSUSYalgebra\]) the rescalings $$Q_{\mu}\rightarrow \frac{1}{{\alpha}} Q_{\mu}\,, \quad\quad Q_5 \rightarrow {\alpha}Q_5\,,$$ one obtains $$\{Q_{\mu},Q_{\nu}\}\rightarrow \frac{1}{{\alpha}^2} \eta_{{\mu}{\nu}} Z, \quad\quad \{Q_5,Q_5\}={\alpha}^2 \tilde Z\,,$$ while the other (anti)commutation relations remain unchanged. By choosing ${\alpha}^2=\sqrt{\frac{\vert Z\vert}{\vert \tilde Z\vert}}$, one sees that only the absolute value of one of the two central charges and their relative signs are relevant. A similar rescaling can be performed when one of the central charges is zero and the other is not, to fix the value of the nonzero central charge to $\pm 1$.\ To summarize, one gets the different equivalence classes of irreducible representations given in Table \[tableeqclass\]. Central charge values Equivalent with ----------------------------------------------------------------------- ---------------------------- $Z\neq 0$, $\tilde Z\neq 0$ and ${\rm sign}(Z)={\rm sign}(\tilde Z)$ $Z=\tilde Z\neq 0$ $Z\neq 0$, $\tilde Z\neq 0$ and ${\rm sign}(Z)=-{\rm sign}(\tilde Z)$ $Z=-\tilde Z\neq 0$ $Z=0$, $\tilde Z\neq 0$ $Z=0$ and $\tilde Z=\pm 1$ $Z\neq0$, $\tilde Z= 0$ $Z=\pm 1$ and $\tilde Z=0$ $Z=\tilde Z= 0$ $Z=\tilde Z=0$ : Equivalence classes of irreducible representations according to the values of the central charges\[tableeqclass\] In some field representations we discuss in this paper, the two central charges of the VSUSY algebra are identified, for simplicity. One should then remember that this assumption is in fact a restriction only for reducible representations characterized by a nontrivial action of the central charge operators. Casimir operators ----------------- In [@Casalbuoni:2008ez], the Casimir operators of the VSUSY algebra (\[VSUSYalgebra\]) have been derived.\ Besides the square of the momentum $P^2$, another spin-related Casimir has been found, $W^{2}$, which is the square of the spin vector $W^{\mu}$: $$W^{\mu} = \frac{1}{2}\epsilon^{\mu\nu\rho\sigma}\left(i\,Z M_{\rho\sigma}-Q_{\rho}Q_{\sigma}\right).$$ This is the VSUSY analogue of the superspin operator of ordinary supersymmetry. The structure of the superspin Casimir allows us to derive the Lorentz spin content of an irreducible VSUSY multiplet.\ If one denotes the superspin eigenvalues by $W^{2}=m^2Y(Y+1)$, it turns out that an irreducible VSUSY multiplet of superspin $Y$ always contain fields of Lorentz spin $s=\vert Y\pm \frac{1}{2}\vert$. Therefore, for $Y>0$ the multiplets contain components of spin $(s,s+1)$, while in the degenerate case $Y=0$ the components have spin $\frac{1}{2}$. In this paper we will explicitly construct multiplets with $Y=0$ and $Y=\frac{1}{2}$.\ In [@Casalbuoni:2008ez], it was also observed that VSUSY, due to its structure with scalar and vectorial odd generators, could have an odd Casimir operator as well. In fact, it was shown that, for representations satisfying the following BPS-like condition relating the mass and the values of the central charges $$P^{2}=Z\tilde Z\,, \label{pzz}$$ the odd operator $$\mathcal{Q}=Q^{\mu}P_{\mu}+ Q_{5} Z, \label{oddcas}$$ (anti)commutes with all other operators in the algebra. Therefore, $\mathcal{Q}$ is technically not a Casimir operator, it should be called surface invariant. However, we still choose to call it ‘odd Casimir’ in this paper. $Y=0$ multiplets ================ According to the discussion in the previous section, the irreducible VSUSY multiplets with superspin $Y=0$ are doublets of spin $\frac{1}{2}$ fields. These doublets are realized in a field theory setting by on-shell fields. Off-shell multiplet have in general more fields. In the following we will present two inequivalent off-shell multiplets, both featuring four spinors, two even and two odd. For one of the multiplets we will also show a reduction, featuring only two spinors, one even and one odd, which closes only on shell. Furthermore, we will discuss whether these spinorial multiplets are compatible with a reality (Majorana) condition and we will give a VSUSY invariant free action for them. Finally, we will provide a superspace setup for VSUSY and we will derive our multiplets by superspace techniques. $Y=0$ multiplet from the spinning particle ------------------------------------------ Consider the four Dirac spinor fields $\psi_1$, $\tilde\psi_1$, $\psi_2$, $\tilde\psi_2$, two even and two odd. As mentioned earlier, we denote fermionic fields by a tilde. The VSUSY charges act on the fields as given in Table \[tablespinning\]. $Q_{5}$ $Q_{\mu}$ ---------------- ------------------------------------ --------------------------------------------------------------------------------------------------------------------------------- $\psi_1$ $-\sqrt{\frac{a}{2}} \tilde\psi_2$ $\frac{1}{\sqrt{2a}}\left[\left(-m\gamma_{\mu}+ \partial_{\mu}\right)\gamma_5\tilde\psi_1- i\partial_{\mu}\tilde\psi_2\right]$ $\tilde\psi_1$ $-\sqrt{\frac{a}{2}} \psi_2$ $\frac{1}{\sqrt{2a}}\left[\left(-m\gamma_{\mu}+ \partial_{\mu}\right)\gamma_5\psi_1- i\partial_{\mu}\psi_2\right]$ $\psi_2$ $-\sqrt{\frac{a}{2}} \tilde\psi_1$ $\frac{1}{\sqrt{2a}}\left[- i\partial_{\mu}\tilde\psi_1-\gamma_5\left(m\gamma_{\mu}+ \partial_{\mu}\right)\tilde\psi_2\right]$ $\tilde\psi_2$ $-\sqrt{\frac{a}{2}} \psi_1$ $\frac{1}{\sqrt{2a}}\left[- i\partial_{\mu}\psi_1-\gamma_5\left(m\gamma_{\mu}+ \partial_{\mu}\right)\psi_2\right]$ : Action of the VSUSY charges on the fields (four-spinor multiplet with fixed value of the central charges). \[tablespinning\] The central charges $Z$ and $\tilde Z$ are represented diagonally on this multiplet, with values $Z=\frac{m^2}{a}$ and $\tilde Z=-a$. The following set of Dirac-type equations of motion is invariant under VSUSY: $$\begin{aligned} &\left({\not\hspace{-.05cm}\partial}+m\right)\psi_1=0\,;~~~~~~~ \left({\not\hspace{-.05cm}\partial}+m\right)\tilde\psi_1=0\,;\cr &\left({\not\hspace{-.05cm}\partial}+m\right)\psi_2=0\,;~~~~~~~ \left({\not\hspace{-.05cm}\partial}+m\right)\tilde\psi_2=0\,. \label{Diraceqs}\end{aligned}$$ Now that the parameter $m$ in the representation has been identified with the mass of the spinors, we see that the relation between the values of the two central charges $Z\tilde{Z}=-m^2$ is the one allowing for the presence of the odd Casimir on the mass shell (\[pzz\]). Indeed, one can explicitly check that the odd Casimir (\[oddcas\]) is zero for this representation when $P^2=-m^2$, as expected in the absence of a natural odd constant in the model.\ Following the conventions (\[daggerconv\]), one can compute the transformations of the barred fields and the result is given in Table \[tablespinningbar\]. $\delta_{Q_{5}}$ $\delta_{Q_{\mu}}$ ---------------------- ---------------------------------------------------- ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- $\bar\psi_1$ $i\sqrt{\frac{a}{2}}\bar{\tilde\psi_2} \epsilon $ $\frac{1}{\sqrt{2a}}\left[i\bar{\tilde\psi_1}\gamma_5\left(m\gamma_{\mu}+ \overleftarrow\partial_{\mu}\right)+ \bar{\tilde\psi_2}\overleftarrow\partial_{\mu}\right]{\epsilon}^{\mu}$ $\bar{\tilde\psi_1}$ $i\sqrt{\frac{a}{2}} \bar\psi_2 {\epsilon}$ $\frac{1}{\sqrt{2a}}\left[i\bar\psi_1\gamma_5\left(m\gamma_{\mu}+ \overleftarrow\partial_{\mu}\right)+ \bar\psi_2\overleftarrow\partial_{\mu}\right] {\epsilon}^{\mu}$ $\bar\psi_2$ $i\sqrt{\frac{a}{2}} \bar{\tilde\psi_1}{\epsilon}$ $\frac{1}{\sqrt{2a}}\left[ \bar{\tilde\psi_1}\overleftarrow\partial_{\mu}+i\bar{\tilde\psi_2}\left(m\gamma_{\mu}- \overleftarrow\partial_{\mu}\right)\gamma_5\right] {\epsilon}^{\mu}$ $\bar{\tilde\psi_2}$ $i\sqrt{\frac{a}{2}}\bar\psi_1 {\epsilon}$ $\frac{1}{\sqrt{2a}}\left[ \bar\psi_1\overleftarrow\partial_{\mu}+i\bar\psi_2\left(m\gamma_{\mu}- \overleftarrow\partial_{\mu}\right)\gamma_5\right] {\epsilon}^{\mu}$ : VSUSY transformations of the Dirac-conjugate spinor fields (four-spinor multiplet with fixed value of the central charges).\[tablespinningbar\] One can check that the following Dirac-type action is VSUSY invariant $$S=\int d^4x \left[-\bar\psi_2 \left({\not\hspace{-.05cm}\partial}+ m\right)\psi_1 + \bar{\tilde\psi_2}\left({\not\hspace{-.05cm}\partial}+m\right)\tilde\psi_1 - \bar\psi_1\left({\not\hspace{-.05cm}\partial}+m\right)\psi_2 + \bar{\tilde\psi_1}\left({\not\hspace{-.05cm}\partial}+m\right)\tilde\psi_2 \right]. \label{actionstandard}$$ In [@Casalbuoni:2008iy], a VSUSY invariant action was constructed for the massive spinning particle with the method of nonlinear realizations. The representation found there can further be rewritten, as shown in Appendix A, in terms of four Dirac spinors $\psi_1$, $\tilde\psi_1$, $\psi_2$, $\tilde\psi_2$, two even and two odd. One can easily show that that representation is the same as the one given in Table \[tablespinning\]. The action of the VSUSY charges $Q_{5}$ and $Q_{\mu}$ on the spinors of [@Casalbuoni:2008iy] is given in Table \[tablespinorsold\]. $Q_{5}$ $Q_{\mu}$ ---------------- -------------------------------------------- --------------------------------------------------------------------------------------------------------------------------------- $\psi_1$ $-\sqrt{\frac{a}{2}}\gamma_5 \tilde\psi_2$ $\frac{1}{\sqrt{2a}}\gamma_5\left[\left(-m\gamma_{\mu}+ \partial_{\mu}\right)\tilde\psi_1- i\partial_{\mu}\tilde\psi_2\right]$ $\tilde\psi_1$ $-\sqrt{\frac{a}{2}}\gamma_5 \psi_2$ $\frac{1}{\sqrt{2a}}\gamma_5\left[\left(-m\gamma_{\mu}+ \partial_{\mu}\right)\psi_1- i\partial_{\mu}\psi_2\right]$ $\psi_2$ $-\sqrt{\frac{a}{2}}\gamma_5 \tilde\psi_1$ $\frac{1}{\sqrt{2a}}\gamma_5\left[- i\partial_{\mu}\tilde\psi_1-\left(m\gamma_{\mu}+ \partial_{\mu}\right)\tilde\psi_2\right]$ $\tilde\psi_2$ $-\sqrt{\frac{a}{2}}\gamma_5 \psi_1$ $\frac{1}{\sqrt{2a}}\gamma_5\left[- i\partial_{\mu}\psi_1-\left(m\gamma_{\mu}+ \partial_{\mu}\right)\psi_2\right]$ : Action of the VSUSY charges on the four-spinor multiplet in [@Casalbuoni:2008iy]. \[tablespinorsold\] To see that the representation in Table \[tablespinorsold\] is in fact the same as the one given in Table \[tablespinning\], it is enough to implement the following rescaling $$\psi_{1} \rightarrow \gamma_{5}\psi_{1}\,, \quad \quad \quad \quad \tilde\psi_{1} \rightarrow \gamma_{5}\tilde\psi_{1}. \label{rescal}$$ A natural question is whether the number of degrees of freedom in this multiplet could be reduced by half by imposing a reality (Majorana) condition on the spinors. The considerations in Section 2.1 applied to the action of the VSUSY generators on the spinorial fields given in Table \[tablespinning\] lead to the conclusion that this is not possible in this case. Another $Y=0$ multiplet ----------------------- Inspired by Refs. [@Kato:2005gb] and [@Kato:2008dw], we construct a multiplet with the same field content as in the previous section but with a nontrivial action of the central charge operators on the fields.\ We are interested in multiplets satisfying (\[pzz\]). For simplicity, we work in the case $Z=-\tilde{Z}$[^1]. Since the central charges are real, the case $Z= \tilde{Z}$ corresponds to fields with an imaginary mass (tachyons) and for this reason we do not consider it.\ As in the multiplet presented in the previous section, the field content is four Dirac spinors, $\psi$, $\tilde\chi$, $\lambda$, $\tilde\xi$, two even and two odd. They transform as in Table \[y0\]. $Q_5$ $Q_{\mu}$ $Z$ ---------------- -------------------------------------------------- -------------------------------------------------------------- ------------------------------------------------ $\psi$ $-\tilde{\chi}$ $-\gamma_{\mu}\tilde{\xi}$ $i\lambda$ $\tilde{\chi}$ $-\frac{i}{2}\lambda$ $-\frac{i}{2}{\not\hspace{-.05cm}\partial}\gamma_{\mu} \psi$ $ -i{\not\hspace{-.05cm}\partial}\tilde{\xi}$ $\lambda$ ${\not\hspace{-.05cm}\partial}\tilde{\xi}$ $-\gamma_{\mu}{\not\hspace{-.05cm}\partial}\tilde{\chi }$ $ -i\Box\psi$ $\tilde{\xi}$ $-\frac{i}{2} {\not\hspace{-.05cm}\partial}\psi$ $\frac{i}{2}\gamma_{\mu}\lambda$ $i{\not\hspace{-.05cm}\partial}\tilde{\chi}$ : Action of the VSUSY charges on the fields (four-spinor multiplet with nontrivial action of the central charge). \[y0\] One can check that this representation indeed satisfies the constraint $P^{2}=-Z^{2}$. Since this constraint allows for the presence of the odd Casimir but there is no natural odd constant, the odd Casimir vanishes on all fields. The transformations of the barred field are given in Table \[tablespinorsbar2\].\ $\delta_{Q_5}$ $\delta_{Q_{\mu}}$ $\delta_{Z}$ --------------------- ---------------------------------------------------------------------------------- ------------------------------------------------------------------------------------------------- --------------------------------------------------------------------------------- $\bar\psi$ $i\bar{\tilde\chi}{\epsilon}_5$ $-i\bar{\tilde{\xi}}\gamma_{\mu}{\epsilon}^{\mu}$ $-\bar\lambda{\epsilon}_Z$ $\bar{\tilde\chi}$ $\frac{1}{2}\bar\lambda {\epsilon}_5$ $\frac{1}{2}\bar\psi \gamma_{\mu}\overleftarrow{{\not\hspace{-.05cm}\partial}}{\epsilon}^{\mu}$ $-\bar{\tilde{\xi}}\overleftarrow{{\not\hspace{-.05cm}\partial}}{\epsilon}_Z $ $\bar\lambda$ $i\bar{\tilde\xi}\overleftarrow{{\not\hspace{-.05cm}\partial}}{\epsilon}_5$ $i\bar{\tilde\chi}\overleftarrow{{\not\hspace{-.05cm}\partial}}\gamma_{\mu}{\epsilon}^{\mu}$ $\bar\psi\overleftarrow{\Box} {\epsilon}_Z$ $\bar{\tilde{\xi}}$ $-\frac{1}{2}\bar\psi \overleftarrow{{\not\hspace{-.05cm}\partial}}{\epsilon}_5$ $\frac{1}{2}\bar\lambda\gamma_{\mu}{\epsilon}^{\mu}$ $\bar{\tilde\chi}\overleftarrow{{\not\hspace{-.05cm}\partial}}{\epsilon}_Z$ : Transformations of the Dirac-conjugate spinor fields (four-spinor multiplet with nontrivial action of the central charge) \[tablespinorsbar2\] One can check that the following Klein-Gordon type free action is invariant under VSUSY: $$S=\int d^4 x~ \left(\bar{\tilde{\xi}}\Box \tilde{\xi} + \bar{\tilde\chi}\Box \tilde{\chi} +\frac{i}{2}\bar\psi \Box \lambda - \frac{i}{2}\bar \lambda \Box \psi+ m^2( \bar{\tilde{\xi}} \tilde{\xi} + \bar{\tilde\chi} \tilde{\chi} +\frac{i}{2}\bar\psi \lambda - \frac{i}{2}\bar \lambda \psi)\right). \label{action4}$$ Following the discussion at the end of Section 2.1, we see that for this multiplet it is possible to impose the Majorana condition (\[majorana\]) on the spinors to reduce the number of degrees of freedom of the representation by half, by requiring all of them to be ‘real’ in the sense stated there.\ A comment is due on the unexpected form of the action. It is of course unnatural to have an action for spinor fields with Klein-Gordon kinetic terms. A first check is that a set of Dirac-like equations for the spinor fields indeed cannot be invariant under the VSUSY transformations given in Table \[y0\]. The puzzling issue is however that, by inspection of action (\[action4\]) only, one cannot see that the Lorentz group under which the spinors transform is the same as the spacetime Lorentz group, so apparently the symmetry group of the action is larger than the VSUSY algebra (\[VSUSYalgebra\]). This feature is related to the fact that no spinor generators are present in the VSUSY algebra, so spinors are introduced by hand in the representations and do not naturally arise as in the case of ordinary SUSY as superpartners of scalar or vectorial fields. However, the VSUSY transformations under which the action is invariant do show that the two Lorentz groups have to be identified. This puzzle is technically due to the absence of a coupling between $\gamma_{\mu}$ and $\partial_{\mu}$ in the kinetic terms for the spinors and could be solved by possible VSUSY-invariant interaction terms where this coupling would be present. This issue deserves further investigation. The search for a VSUSY invariant interacting field theory involving this multiplet is left for future work. ### On-shell $Y=0$ multiplet It is also possible to construct a multiplet with only two spinors, one even and one odd, being a realization of the irreducible doublet with $Y=0$ discussed in Section 2.2.\ We work again in the simpler case $Z=-\tilde{Z}$.\ Consider the multiplet given in Table \[tableonshell\], where $\hat z$ is the value of the central charge. $Q_5$ $Q_{\mu}$ ---------------- ------------------------- -------------------------------------------------------------------------- $\psi$ $-\tilde{\chi}$ $-\frac{i}{\hat z}\gamma_{\mu}{\not\hspace{-.05cm}\partial}\tilde{\chi}$ $\tilde{\chi}$ $-\frac{\hat z}{2}\psi$ $-\frac{i}{2}{\not\hspace{-.05cm}\partial}\gamma_{\mu}\psi$ : Action of the VSUSY charges on the fields (on-shell two-spinor multiplet).\[tableonshell\] The transformations of the barred fields are given in Table \[tableonshell2\]. $\delta_{Q_5}$ $\delta_{Q_{\mu}}$ ---------------------- --------------------------------------- ------------------------------------------------------------------------------------------------------------- $\bar\psi$ $i\bar{\tilde\chi}{\epsilon}$ $\frac{1}{\hat z}\bar{\tilde\chi}\overleftarrow{{\not\hspace{-.05cm}\partial}}\gamma_{\mu}{\epsilon}^{\mu}$ $\bar{\tilde{\chi}}$ $\frac{i\hat z}{2}\bar\psi{\epsilon}$ $\frac{1}{2}\bar\psi\gamma_{\mu}\overleftarrow{{\not\hspace{-.05cm}\partial}}{\epsilon}^{\mu}$ : Transformations of the Dirac-conjugate spinor fields (on-shell two-spinor multiplet).\[tableonshell2\] One can easily see that this table is a component realization of the VSUSY algebra (\[VSUSYalgebra\]) in the special case $Z=-\tilde{Z}$ when one imposes the equations of motion $\Box \psi=\hat z^2 \psi$ and $\Box \tilde{\chi}=\hat z^2 \tilde{\chi}$, which are a realization of the constraint $P^2=-Z^2$ on the fields. Note also that the odd Casimir is realized on this multiplet with the value zero. Actually, this short multiplet can be derived from the one presented in Table \[y0\] by requiring that the $Z$ operator acts as a number $\hat z$ on all fields. The resulting four constraint equations amount to $$\lambda=-i\hat z\psi\,, \quad\quad\quad\quad \tilde\xi=\frac{i}{\hat z} {\not\hspace{-.05cm}\partial}\tilde\chi\,, \label{reduction}$$ reducing the number of fields from four to two, together with the dynamical equations $$\Box \psi= \hat z^2 \psi\,, \quad\quad\quad\quad \Box \tilde \xi= \hat z^2 \tilde \xi\,.$$ Therefore the truncation procedure puts the remaining fields on-shell and so it is not correct to derive an action for the new multiplet by directly applying this truncation to action (\[action4\]). However, there exists a VSUSY invariant action for this shorter multiplet: $$S=\int d^4 x~ \Big(2 \bar{\tilde\chi}\Box \tilde{\chi} + \hat z\bar \psi \Box \psi+ m^2(2 \bar{\tilde\chi} \tilde{\chi} +\hat z\bar \psi \psi)\Big).$$ Concerning the possibility of reducing the number of degrees of freedom of this multiplet by imposing a Majorana condition on the spinors, inspection of constraints (\[reduction\]) shows that these are not compatible with such condition and therefore the number of degrees of freedom cannot be reduced further. This can be also seen directly at the level of the action since, for Majorana spinors, the terms involving the field $\psi$ in the action would vanish. $Y=0$ multiplets from superspace a la Sohnius --------------------------------------------- In this section we would like to derive the spinor multiplets presented in Tables \[tablespinning\], \[y0\] and \[tableonshell\] via a superspace approach.\ Since the multiplet in Table \[y0\] is characterized by a nontrivial action of the central charge on the component fields, it will not be possible to derive it by using a superspace of the standard type, spanned only by the bosonic coordinates $x^{\mu}$ associated to the generator $P^{\mu}$ and the fermionic coordinates $\theta^A$ associated to the fermionic generators $Q^A$. Two extra bosonic coordinates $z$ and $\tilde z$ associated to the central charge operators $Z$ and $\tilde Z$ must be present as well. For simplicity, we consider the case where $\tilde Z=-Z$ so that we have to add only one extra bosonic coordinate $z$. In contrast to the case of extra fermionic coordinates, automatically leading to a superfield expansion with a finite number of components, the presence of an extra bosonic coordinate leads to an infinite number of component fields. To obtain a multiplet with a finite number of fields, one must necessarily impose a set of covariant constraints in superspace forcing the coefficients of the $z$ expansion for higher powers of $z$ to be functions of a finite number of lower coefficients.\ This procedure has been discussed first by Sohnius in [@Sohnius:1978fw], in the context of $\mathcal{N}=2$ supersymmetric theories with one real central charge. There he introduced a superspace with supercoordinates $(x^{\mu},\theta^{\alpha}_{i}, \overline{\theta}^{\dot\alpha i}, z)$, where $\alpha$ and $\dot\alpha$ are spinor indices and $i$ is the internal $SU(2)$ R-symmetry index. Starting from the most general expansion of a superfield $\Phi_{i}(x^{\mu},\theta^{\alpha}_{i}, \overline{\theta}^{\dot\alpha i}, z)$ with an extra $SU(2)$ index $i$, he imposed two covariant constraints on it in order to reduce the number of degrees of freedom. In practice, he imposed that the covariant spinor derivatives $D_{\alpha}^{j}$ and $\overline{D}_{\dot\alpha j}$, when acting on $\Phi_{i}$, produce something proportional to the only two structures with two indices available in the $SU(2)$ space, the Kronecker delta $\delta^{\,j}_{i}$ and the antisymmetric symbol ${\epsilon}_{ij}$: $$D_{\alpha}^{j}\Phi_{i} = \delta^{\,j}_{i} \Psi_{\alpha}\,,\quad \quad \quad \quad \overline{D}_{\dot\alpha j}\Phi_{i}={\epsilon}_{ij}\overline{\Psi}_{\dot\alpha}\,. \label{sohnius}$$ In order for (\[sohnius\]) to be covariant constraints, $\Psi_{\alpha}$ and $\overline{\Psi}_{\dot\alpha}$ must be full superfields with a spinor index. Moreover, these two constraints automatically imply that $P^2~\sim~Z^2$ on $\Phi_{i}$, which means that the superfield satisfies $\Box \Phi_{i}=\frac{\partial^{2}}{\partial z^{2}}\Phi_{i}$. As a result, the range in spin covered by the supermultiplet is only $\Delta s=\frac{1}{2}$. This ensures that the supermultiplet contains only four independent component fields ($A$, $\psi$, $\overline{\varphi}$, $F$) and that the higher order terms in the series expansion in $z$ are simply higher derivatives of those fields. We would like to follow this approach to construct our VSUSY spinor multiplets. However, as we will discuss in detail in the next section, in our case we have an odd vectorial superspace covariant derivative, $D_{\mu}$, and an odd scalar one, $D_5$, both carrying no R-symmetry indices. Since we want to derive multiplets containing only spinor components, but no spinor indices appear in the VSUSY algebra (\[VSUSYalgebra\]), we must necessarily start from a superfield carrying a spinor index $\Phi^{\alpha}$. Moreover, since the only structure containing both vector and spinor indices at our disposal is the Dirac $(\gamma_{\mu})^{\alpha}_{\,\,\beta}$, a natural covariant constraint to be imposed on $\Phi^{\alpha}$ is $$D_{\mu}\Phi^{\alpha} = (\gamma_{\mu})^{\alpha}_{\,\,\beta}\tilde\Lambda^{\beta}\,, \label{first}$$ where $D_{\mu}$ is the odd vectorial superspace covariant derivative, to be explicitly given in the next section, and $\tilde\Lambda^{\alpha}$ is a generic spinor superfield.\ As we will show in the following, this constraint indeed does the job of eliminating all higher spin components in the superfield, leaving only spinor components in the superfield expansion. However, this constraint by itself is not enough to “terminate” the expansion in $z$. As much as Sohnius needed two contraints, one involving $D_{\alpha}^{j}$ and the other $\overline{D}_{\dot\alpha j}$, to obtain a finite number of components fields, we will also need to impose a second constraint on $\Phi^{\alpha}$ containing the covariant derivative $D_{5}$ in order to obtain a finite number of component fields. This second constraint will have the following property: combined with (\[first\]) it will imply that $P^2=-Z^2$ on $\Phi^{\alpha}$, ensuring that the infinite series expansion in $z$ actually depends only on a finite number of lower components. We will see that the right choice for a second constraint is of the form $${\not\hspace{-.1cm}D}D_{5}\Phi^{\alpha} \sim {\not\hspace{-.05cm}\partial}\Phi^{\alpha}\,. \label{second}$$ In principle, one could also think of a constraint of the form $D_5 \Phi \sim \tilde \Lambda$. In fact, one can check that this also implies that $P^2=-Z^2$ on $\Phi^{\alpha}$. However, inspection of the component constraint equations one gets from it shows that it is stronger than (\[second\]), in fact too strong, since it directly implies equations of motions for the component fields and not constraints reducing the number of components. ### Superspace We consider a superspace with supercoordinates $x^A\equiv(x^{\mu},\theta^{\mu}, \theta_{5}, z, \tilde z)$.\ The VSUSY charges have the following form: $$Q_{\mu}= \frac{\partial}{\partial \theta^{\mu}}-\frac{i}{2}\theta_{5}\frac{\partial}{\partial x^{\mu}}+\frac{i}{2}\theta_{\mu}\frac{\partial}{\partial z}\,, \quad \quad \quad Q_{5}= \frac{\partial}{\partial \theta_{5}}-\frac{i}{2}\theta_{\mu}\frac{\partial}{\partial x^{\mu}}+\frac{i}{2}\theta_{5}\frac{\partial}{\partial \tilde z}\,. \label{chargessuper}$$ They satisfy the VSUSY algebra: $$\{Q_{\mu},Q_{\nu}\}=\eta_{\mu\nu}Z\,, \quad\quad\quad \{Q_{\mu},Q_{5}\}=-P_{\mu}\,,\quad\quad\quad \{Q_{5},Q_{5}\}=\tilde Z\,. \label{algebra}$$ where $$Z= i\frac{\partial}{\partial z}\,, \quad\quad \tilde Z = i \frac{\partial}{\partial \tilde z}\,,\quad \quad P_{\mu}=i\frac{\partial}{\partial x^{\mu}}\,.$$ The superspace covariant derivatives are[^2] $$D_A=\left(\frac{\partial}{\partial x^{\mu}}, D_{\nu}, D_5, \frac{\partial}{\partial z}, \frac{\partial}{\partial \tilde z}\right), \label{covdevI}$$ where $D_{\mu}$ and $D_5$ are given by $$D_{\mu}= \frac{\partial}{\partial \theta^{\mu}}+\frac{i}{2}\theta_{5}\frac{\partial}{\partial x^{\mu}}-\frac{i}{2}\theta_{\mu}\frac{\partial}{\partial z}\,, \quad \quad \quad D_{5}= \frac{\partial}{\partial \theta_{5}}+\frac{i}{2}\theta_{\mu}\frac{\partial}{\partial x^{\mu}}-\frac{i}{2}\theta_{5}\frac{\partial}{\partial \tilde z}\,. \label{covdevII}$$ It is easy to see that these anticommute with the VSUSY charges and satisfy: $$\{D_{\mu},D_{\nu}\}=-\eta_{\mu\nu}Z\,, \quad\quad\quad \{D_{\mu},D_{5}\}=P_{\mu}\,,\quad\quad\quad \{D_{5},D_{5}\}=-\tilde Z\,.$$ Note that, similarly as in ordinary superspace, one can perform the following change of variables $$x^{{\mu}(\pm)}=x^{\mu}\pm \frac{i}{2} \theta^{\mu}\theta_5\,, \label{covdertrans}$$ where the other supercoordinates remain unchanged, so that the odd covariant derivatives take the asymmetric form $$D^{(+)}_{\mu}= \frac{\partial}{\partial \theta^{\mu}}+i\theta_{5}\frac{\partial}{\partial x^{\mu}}-\frac{i}{2}\theta_{\mu}\frac{\partial}{\partial z}\,,\quad\quad D^{(+)}_5=\frac{\partial}{\partial \theta_{5}}-\frac{i}{2}\theta_{5}\frac{\partial}{\partial \tilde z}\,. \label{newcovderivI}$$ or $$D^{(-)}_{\mu}= \frac{\partial}{\partial \theta^{\mu}}-\frac{i}{2}\theta_{\mu}\frac{\partial}{\partial z}\,,\quad\quad D^{(-)}_5=\frac{\partial}{\partial \theta_{5}}+i\theta_{\mu}\frac{\partial}{\partial x^{\mu}}-\frac{i}{2}\theta_{5}\frac{\partial}{\partial \tilde z}\,. \label{newcovderivII}$$ This is the VSUSY analogue of the chiral and antichiral superspace representations vs. the vector representation of ordinary supersymmetry.\ In the following we will write constraints in superspace to reduce the number of degrees of freedom of a general superfield. We remind the reader that all constraints written in terms of the superspace covariant derivatives (\[covdevI\]), (\[covdevII\]) and in terms of superfields are automatically VSUSY invariant. ### Spinor Superfield For simplicity, let us first restrict ourselves to the case $\tilde Z=-Z$.\ As said before, we expect to be able to obtain a supermultiplet containing only spinor components, so we choose to consider a superfield with an extra spinor index $\Phi^{\alpha}$ expanded in $\theta^{\mu}$, $\theta_{5}$ and $z$. We also choose $\Phi^{\alpha}$ to be bosonic (as usual, we put a tilde on its fermionic components): $$\begin{aligned} \Phi^{\alpha}(x^{\mu},\theta^{\mu},\theta_{5},z)&=&\phi^{\alpha}(x,z)+ \tilde{\phi}^{\alpha}(x,z)\theta_{5} +\tilde\psi_{\mu}^{\alpha}(x,z)\theta^{\mu}+ \psi_{\mu}^{\alpha}(x,z)\theta^{\mu}\theta_{5}\cr &&+\xi_{\mu\nu}^{\alpha}(x,z)\theta^{\mu}\theta^{\nu}+ \tilde{\xi}_{\mu\nu}^{\alpha}(x,z)\theta^{\mu}\theta^{\nu}\theta_{5} +\tilde\chi_{\mu\nu\rho}^{\alpha}(x,z)\theta^{\mu}\theta^{\nu} \theta^{\rho}\cr &&+\chi_{\mu\nu\rho}^{\alpha}(x,z) \theta^{\mu}\theta^{\nu}\theta^{\rho}\theta_{5} +\zeta_{\mu\nu\rho\sigma}^{\alpha}(x,z) \theta^{\mu}\theta^{\nu}\theta^{\rho}\theta^{\sigma}+ \tilde{\zeta}_{\mu\nu\rho\sigma}^{\alpha}(x,z) \theta^{\mu}\theta^{\nu}\theta^{\rho}\theta^{\sigma}\theta_{5}.\cr && \label{spinorsuperfield}\end{aligned}$$ where we assume that every component field admits a Taylor expansion in $z$, for instance $$\phi^{\alpha}(x,z)=\sum_{n=0}^{\infty}\frac{1}{n!}z^{n}\,\phi^{{\alpha}(n)}(x) \label{zetaexp}$$ and similar for the other components.\ Since the $\theta$’s anticommute, it is clear that $\xi_{\mu\nu}$, $\tilde{\xi}_{\mu\nu}$, $\chi_{\mu\nu\rho}$, $\tilde{\chi}_{\mu\nu\rho}$, $\zeta_{\mu\nu\rho\sigma}$ and $\tilde{\zeta}_{\mu\nu\rho\sigma}$ must be totally antisymmetric in their vectorial indices.\ In order to avoid a cumbersome notation, we will often omit the spinor indices on the components. ### Supercovariant constraints Saying that the constraint (\[first\]) does the job of removing all higher-spin Rarita-Schwinger-like fields and leaving only spinors components in the expansion (\[spinorsuperfield\]), can technically be re-expressed by saying that the components ($\psi_{\mu}$, $\tilde\psi_{\mu}$, $\xi_{\mu\nu}$, $\tilde{\xi}_{\mu\nu}$, $\chi_{\mu\nu\rho}$, $\tilde{\chi}_{\mu\nu\rho}$, $\zeta_{\mu\nu\rho\sigma}$, $\tilde{\zeta}_{\mu\nu\rho\sigma}$) solving the constraint only contain the following two structures $$A^{\alpha}_{\mu_{1}\mu_{2}...\mu_{n}}(x)=\gamma_{\mu_{1}\mu_{2}...\mu_{n}}B^{\alpha}(x)+ \gamma_{[\mu_{1}\mu_{2}...\mu_{n-1}}\partial_{\mu_{n}]}C^{\alpha}(x)\,, \label{forms}$$ where $$\gamma_{\mu_{1}\mu_{2}...\mu_{n}}=\frac{1}{n!}\sum_{perm} (-)^{\sigma(\mu_{1}...\mu_{n})}\, \gamma_{\mu_{1}}\gamma_{\mu_{2}}...\gamma_{\mu_{n}}$$ and with $\sigma(\mu_{1}...\mu_{n})$ we denote a permutation of the indices $\mu_{1}...\mu_{n}$. As shown in Appendix B, the general solution of constraint (\[first\]) is given by the following relations for $n=0,1,2,...$: $$\begin{aligned} &\tilde\psi_{\mu}^{(n)}=\gamma_{\mu} \tilde\lambda^{(n)}\quad\quad\quad &\psi_{\mu}^{(n)}=\gamma_{\mu} \lambda^{(n)}-\frac{i}{2}\partial_{\mu}\phi^{(n)}\nonumber\\ &\xi_{\mu\nu}^{(n)}=-\frac{i}{4}\gamma_{\mu\nu} \phi^{(n+1)}\quad\quad\quad &\tilde\xi_{\mu\nu}^{(n)}=-\frac{i}{4}\gamma_{\mu\nu} \tilde\phi^{(n+1)}-\frac{i}{2}\gamma_{[\mu}\partial_{\nu]}\tilde\lambda^{(n)}\nonumber\\ &\tilde\chi_{\mu\nu\rho}^{(n)}=-\frac{i}{12}\gamma_{\mu\nu\rho} \tilde\lambda^{(n+1)}\quad\quad\quad &\chi_{\mu\nu\rho}^{(n)}=-\frac{i}{12}\gamma_{\mu\nu\rho} \lambda^{(n+1)}-\frac{1}{8}\gamma_{[\mu\nu}\partial_{\rho]}\phi^{(n+1)}\nonumber\\ &\zeta_{\mu\nu\rho\sigma}^{(n)}=-\frac{1}{96}\gamma_{\mu\nu\rho\sigma} \phi^{(n+2)}\quad\quad\quad &\tilde\zeta_{\mu\nu\rho\sigma}^{(n)}=-\frac{1}{96}\gamma_{\mu\nu\rho\sigma} \tilde\phi^{(n+2)}-\frac{1}{24}\gamma_{[\mu\nu\rho}\partial_{\sigma]}\tilde\lambda^{(n+1)}. \label{exp}\end{aligned}$$ This is indeed of the expected form (\[forms\]).\ We are left with four infinite sets of independent field components, $$\phi^{(n)}, \quad\quad\tilde\phi^{(n)},\quad\quad\lambda^{(n)},\quad\quad\tilde\lambda^{(n)}, \label{infty}$$ whose VSUSY transformations are given in Table \[table\].\ $Q_{5}$ $Q_{\mu}$ $Z$ ----------------------- ------------------------------------- --------------------------------------------------------------------------------------------- ---------------------------- $\phi^{(n)}$ $-\tilde\phi^{(n)}$ $-\gamma_{\mu}\tilde\lambda^{(n)}$ $i\,\phi^{(n+1)}$ $\tilde\phi^{(n)}$ $-\frac{i}{2}\phi^{(n+1)}$ $\gamma_{\mu}\lambda^{(n)}-i\partial_{\mu}\phi^{(n)}$ $i\, \tilde\phi^{(n+1)}$ $\tilde\lambda^{(n)}$ $-\lambda^{(n)}$ $\frac{i}{2}\gamma_{\mu}\phi^{(n+1)}$ $i\,\tilde\lambda^{(n+1)}$ $\lambda^{(n)}$ $-\frac{i}{2}\tilde\lambda^{(n+1)}$ $\,\,\,\,\,\,-\frac{i}{2}\gamma_{\mu}\tilde\phi^{(n+1)}-i\partial_{\mu}\tilde\lambda^{(n)}$ $i\, \lambda^{(n+1)}$ : Action of the VSUSY charges on the four infinite sets of spinor components (\[infty\]) solving constraint (\[first\]). \[table\] If we want to recover the case with $Z$ acting like a number we need to further impose the constraint $$Z\Phi=\hat{z}\,\Phi\,, \label{zeta}$$ with $\hat z$ number. This constraint implies: $$\phi^{(n+1)}=-i\hat z\phi^{(n)}\,,\quad \quad \tilde\phi^{(n+1)}=-i\hat z\tilde\phi^{(n)}\,,\quad \quad \lambda^{(n+1)}=-i\hat z\lambda^{(n)}\,,\quad \quad \tilde \lambda^{(n+1)}=-i\hat z\tilde\lambda^{(n)}\,. \label{sup}$$ We are then left with only four independent field components, $\phi^{(0)}$, $\tilde\phi^{(0)}$, $\lambda^{(0)}$ and $\tilde\lambda^{(0)}$, whose VSUSY transformations are given in Table \[tablenine\]. $Q_{5}$ $Q_{\mu}$ ----------------------- ---------------------------------------- ------------------------------------------------------------------------------------------------ $\phi^{(0)}$ $-\tilde\phi^{(0)}$ $-\gamma_{\mu}\tilde\lambda^{(0)}$ $\tilde\phi^{(0)}$ $-\frac{\hat z}{2}\phi^{(0)}$ $\gamma_{\mu}\lambda^{(0)}-i\partial_{\mu}\phi^{(0)}$ $\tilde\lambda^{(0)}$ $-\lambda^{(0)}$ $\frac{\hat z}{2}\gamma_{\mu}\phi^{(0)}$ $\lambda^{(0)}$ $-\frac{\hat z}{2}\tilde\lambda^{(0)}$ $\,\,\,\,\,\,-\frac{\hat z}{2}\gamma_{\mu}\tilde\phi^{(0)}-i\partial_{\mu}\tilde\lambda^{(0)}$ : Action of the VSUSY charges on the four spinor multiplet obtained from the components given in Table \[table\] by further diagonalizing the central charge. \[tablenine\] One can check that these transformations are not compatible with the Dirac equation.\ However, it is easy to show that the following linear combinations: $$\begin{aligned} &\psi_{1}\equiv-\frac{i}{2}(\phi^{(0)}+\frac{2i}{\hat z}\lambda^{(0)})\,; \quad \quad \psi_{2}\equiv\frac{1}{2}\gamma_5(\phi^{(0)}-\frac{2i}{\hat z}\lambda^{(0)})\,;\nonumber\\ &\tilde \psi_{1}\equiv\frac{1}{\sqrt{2\hat z}}\gamma_5(\tilde\phi^{(0)}-i\tilde\lambda^{(0)})\,;\quad\quad\tilde \psi_{2}\equiv-\frac{i}{\sqrt{2\hat z}}(\tilde\phi^{(0)}+i\tilde\lambda^{(0)})\,, \label{lincomb}\end{aligned}$$ exactly reproduce the multiplet previously given in Table \[tablespinning\] with the following identifications: $$\hat z=m=a ~~~~~~(Z=-\tilde Z)\,.$$ Now we go back to the supermultiplet (\[infty\]) where the operator $Z$ is still not diagonal and we want to find a second constraint that, together with the first given in (\[first\]), automatically implies the multiplet shortening conditions $P^2=-Z^2$. As already mentioned in the introduction to this section, a natural choice is: $${\not\hspace{-.05cm}\partial}\Phi=-\frac{i}{2}D_{5}{\not\hspace{-.1cm}D}\,\Phi\,,$$ or, equivalently, $${\not\hspace{-.05cm}\partial}\,\Phi=-2i\,D_{5}\tilde\Lambda\,,$$ where $\tilde\Lambda$ is the spinor superfield appearing in (\[first\]).\ After some algebraic manipulations one can prove that (\[second\]), combined with (\[first\]), implies that $P^{2}=-Z^{2}$, that is to say: $$\Box\,\Phi^{\alpha}=-\frac{\partial^{2}}{\partial z^{2}}\Phi^{\alpha}\, . \label{box}$$ The constraint (\[box\]) reduces the number of independent components down to eight: $$\phi^{(0)}, \quad \phi^{(1)},\quad\tilde\phi^{(0)},\quad\tilde\phi^{(1)},\quad\lambda^{(0)}, \quad \lambda^{(1)},\quad\tilde\lambda^{(0)},\quad\tilde\lambda^{(1)}.$$ Higher order components (for $n=1,2...$) are related to the previous ones via: $$\begin{aligned} &\phi^{(2n)}=(-\Box)^{n}\phi^{(0)}\quad\quad&\tilde\phi^{(2n)}=(-\Box)^{n}\tilde\phi^{(0)}\nonumber\\ &\lambda^{(2n)}=(-\Box)^{n}\lambda^{(0)}\quad\quad&\tilde\lambda^{(2n)}=(-\Box)^{n}\tilde\lambda^{(0)}\nonumber\\ &\phi^{(2n+1)}=(-\Box)^{n}\phi^{(1)}\quad\quad&\tilde\phi^{(2n+1)}=(-\Box)^{n}\tilde\phi^{(1)}\nonumber\\ &\lambda^{(2n+1)}=(-\Box)^{n}\lambda^{(1)}\quad\quad&\tilde\lambda^{(2n+1)}=(-\Box)^{n}\tilde\lambda^{(1)}. \label{higher}\end{aligned}$$ Now, if we implement (\[second\]), which is stronger than (\[box\]), we obtain further constraints on the component fields: $$\lambda^{(0)}=\frac{i}{2}{\not\hspace{-.05cm}\partial}\phi^{(0)}\,, \quad \quad \lambda^{(1)}=\frac{i}{2}{\not\hspace{-.05cm}\partial}\phi^{(1)}\,, \quad \quad \tilde\lambda^{(1)}=\,{\not\hspace{-.05cm}\partial}\tilde\phi^{(0)}\,,\quad \quad \tilde \phi^{(1)}=-{\not\hspace{-.05cm}\partial}\tilde\lambda^{(0)}.$$ So, by imposing both (\[first\]) and (\[second\]), one reduces the number of independent components down to four: $$\phi^{(0)}, \quad \quad \phi^{(1)},\quad\quad\tilde\phi^{(0)},\quad\quad \tilde\lambda^{(0)}. \label{comp}$$ The superfield expansion in $\theta^{\mu}$, $\theta_{5}$ and $z$ reads: $$\begin{aligned} \Phi^{\alpha}=&\phi^{(0)} + \tilde{\phi}^{(0)}\theta_{5}+\gamma_{\mu}\tilde\lambda^{(0)}\theta^{\mu}+z\,\phi^{(1)}- z\,{\not\hspace{-.05cm}\partial}\tilde\lambda^{(0)}\theta_{5}+z\,\gamma_{\mu}{\not\hspace{-.05cm}\partial}\tilde\phi^{(0)}\theta^{\mu} +\frac{i}{2}\gamma_{\mu}{\not\hspace{-.05cm}\partial}\phi^{(0)}\theta^{\mu}\theta_{5}+\nonumber\\ &- \frac{i}{2}\partial_{\mu}\phi^{(0)}\theta^{\mu}\theta_{5} - \frac{i}{4} \gamma_{\mu\nu} \phi^{(1)}\theta^{\mu}\theta^{\nu} -\frac{1}{2}z^{2}\,\Box\,\phi^{(0)}+{\rm higher\,\, order\,\, terms},\end{aligned}$$ where the higher order terms are at least trilinear in $\theta^{\mu}$, $\theta_{5}$ and $z$ and contain only the fields appearing in (\[comp\]).\ The VSUSY transformations for this supermultiplet are the ones already given in Table \[y0\], once one makes the following identification: $$\phi^{(0)}\equiv\psi\,, \quad \quad \tilde{\phi}^{(0)}\equiv\tilde\chi\,, \quad \quad \phi^{(1)}\equiv\lambda\,, \quad \quad \tilde{\lambda}^{(0)}\equiv\xi\,.$$ Now, if we diagonalize $Z$ as we did before, we get the following relations: $$\phi^{(1)}=-i\hat z \phi^{(0)} \quad\quad\quad\quad \tilde\lambda^{(0)}=\frac{i}{\hat z} {\not\hspace{-.05cm}\partial}\tilde\phi^{(0)}.$$ So we are left with the multiplet of Table \[tabletwofields\], containing just two fields, which closes by imposing the constraint $P^2=-Z^2$ by hand. $Q_{5}$ $Q_{\mu}$ -------------------- ------------------------------- ------------------------------------------------------------------------------- $\phi^{(0)}$ $-\tilde\phi^{(0)}$ $-\frac{i}{\hat z} \gamma_{\mu}{\not\hspace{-.05cm}\partial}\tilde\phi^{(0)}$ $\tilde\phi^{(0)}$ $-\frac{\hat z}{2}\phi^{(0)}$ $-\frac{i}{2}{\not\hspace{-.05cm}\partial}\gamma_{\mu}\phi^{(0)}$ : Action of the VSUSY charges on the two spinor fields obtained from the Table \[tablenine\] by further imposing the constraint (\[second\]) \[tabletwofields\] One can check that the previous table reduces to Table \[tableonshell\] when the following identification is implemented: $$\tilde\phi^{(0)}= \tilde\chi\,, \quad\quad\quad\quad \phi^{(0)}=\psi\,.$$ $Y=1/2$ multiplet and twisted topological models ================================================ In [@Kato:2005gb] and [@Kato:2008dw], an ${\mathcal N}=2$ twisted superspace formalism with a central charge in four-dimensional Euclidean space is constructed, by introducing a Dirac-Kähler twist. This twist leads to the following supertranslation algebra $$\begin{aligned} &&\{Q_{\mu},Q_{\nu}\}=Z \delta_{\mu\nu}\,,\quad \quad \{Q_{5},Q_{5}\}=Z\,,\quad \{Q_{\mu},Q_{5}\}=-P_{\mu},\nonumber\\ &&\{Q^+_{{\mu}{\nu}},Q_5\}=0\,,\quad \{Q^+_{{\mu}{\nu}},Q_\rho\}=\delta^+_{{\mu}{\nu},\rho{\sigma}}P^{\sigma}\,,\quad \{Q^+_{{\mu}{\nu}},Q^+_{\rho{\sigma}}\}=\delta^+_{{\mu}{\nu},\rho{\sigma}}Z\,, \label{TSUSY}\end{aligned}$$ which is clearly an extension of the supertranslation sector of the VSUSY algebra (\[VSUSYalgebra\]) in the case where the two VSUSY central charges are identified. In (\[TSUSY\]) $Q^{+}_{{\mu}{\nu}}$ denotes an antiselfdual two-form in Euclidean space and $\delta^+_{{\mu}{\nu},\rho{\sigma}}=\delta_{{\mu}\rho}\delta_{{\nu}{\sigma}} -\delta_{{\nu}\rho}\delta_{{\mu}{\sigma}}-{\epsilon}_{{\mu}{\nu}\rho{\sigma}}$ is, up to a constant, a projector on the selfdual space. We will call this algebra tensor supersymmetry algebra (TSUSY), due to the fact that the extra odd generator present there with respect to VSUSY is an antiselfdual two-form $Q^{+}_{{\mu}{\nu}}$.\ In [@Kato:2005gb] and [@Kato:2008dw], multiplets which are a representation of TSUSY are constructed. These are clearly also representations of VSUSY, but as such they are in general not irreducible. However, there is a main difference between what has been done there and the discussion in this paper. While, as previously mentioned, all multiplets of [@Kato:2005gb] and [@Kato:2008dw] are constructed in Euclidean space, which is the natural setting in the context of the twisted topological theories, in this paper we are interested in representations of VSUSY in Minkowski space. Therefore, in this section we will first exhibit an on-shell VSUSY multiplet with $Y=1/2$, which will be shown in Appendix C to arise from a truncation of the Euclidean multiplet given in [@Kato:2005gb]. Then we will discuss whether this multiplet could be consistently rotated to Minkowski space. Finally, we will rederive the VSUSY multiplet directly in VSUSY superspace. The fields in the $Y=1/2$ VSUSY multiplet satisfy dynamical equations of motion with mass, which however cannot be derived from a Lagrangian. Comments about this issue from the point of view of the truncation of the TSUSY multiplet are given in Appendix C. On-shell $Y=1/2$ multiplet -------------------------- Consider the set of fields $(a, \tilde b_{\mu}, c^-_{{\mu}{\nu}})$ where the first one is an even scalar, the second an odd vector and the third an even selfdual two-tensor in Euclidean or Minkowski space, according to the definitions: $$\begin{aligned} &&c^{\pm}_{{\mu}{\nu}}=\mp\frac{1}{2}{\epsilon}_{{\mu}{\nu}\rho{\sigma}} c^{\pm \rho{\sigma}} ~~~~{\rm (Euclidean)}\,, \cr &&c^{\pm}_{{\mu}{\nu}}=\mp\frac{i}{2}{\epsilon}_{{\mu}{\nu}\rho{\sigma}} c^{\pm \rho{\sigma}} ~~~~{\rm (Minkowski)}\,, \label{selfdualityI}\end{aligned}$$ or equivalently $$c^{\pm}_{{\mu}{\nu}}=\frac{1}{4} \delta^{\pm}_{{\mu}{\nu},\rho{\sigma}} c^{\pm \rho{\sigma}}\,, \label{selfdualityII}$$ where $$\begin{aligned} &&\delta^{\pm}_{{\mu}{\nu},\rho{\sigma}} =\delta_{{\mu}\rho}\delta_{{\nu}{\sigma}}-\delta_{{\mu}{\sigma}}\delta_{{\nu}\rho}\mp {\epsilon}_{{\mu}{\nu}\rho{\sigma}}~~~~{\rm (Euclidean)}\,,\cr &&\delta^{\pm}_{{\mu}{\nu},\rho{\sigma}} =\eta_{{\mu}\rho}\eta_{{\nu}{\sigma}}-\eta_{{\mu}{\sigma}}\eta_{{\nu}\rho}\mp i {\epsilon}_{{\mu}{\nu}\rho{\sigma}}~~~~{\rm (Minkowski)} \label{projectors}\end{aligned}$$ is, up a constant, a projector on the (anti)selfdual space. Therefore, the main difference between the Minkowski and the Euclidean case is that, with Minkowskian signature, (anti)selfdual fields are necessarily complex and the complex conjugate of a selfdual field is antiselfdual. As a result, a model featuring only selfdual or only antiselfdual fields in Minkowski space cannot exist. Forgetting for a moment about this issue, we consider the VSUSY multiplet in Table \[tablevectormultipletI\]. ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ $Q_{5}$ $Q_{\mu}$ ------------------- ------------------------------------------------------------------------------------------- -------------------------------------------------------------------- $a$ $\frac{i}{\hat z}\partial^{\sigma}\tilde b_{\sigma}$ $-\tilde b_{\mu}$ $\tilde b_{\nu}$ $i\left(\frac{2}{\hat z}\partial^{\rho}c^{-}_{\nu\rho}-\frac{1}{2}\partial_{\nu}a\right)$ $2c_{\mu\nu}^{-}+ \frac{\hat z}{2}\eta_{\mu\nu}a$ $c_{\nu\rho}^{-}$ $\frac{i}{4}\delta_{\nu\rho,\sigma\tau}^{-}\partial^{\sigma}\tilde b^{\tau}$ $-\frac{\hat z}{4}\delta_{\nu\rho,\mu\sigma}^{-}\tilde b^{\sigma}$ ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ : Action of the VSUSY charges on the fields (on-shell multiplet with scalar, vector and selfdual two-form).\[tablevectormultipletI\] It is important to observe that this multiplet closes under VSUSY, in both Euclidean and Minkowski signature, when the constraint $P^2=-Z^2$ is imposed by hand, which amounts to imposing on every component field that $\Box=\hat z^2$. Moreover, the odd Casimir is present and has value zero. The closure of the algebra on the field $c_{\nu\rho}^{-}$ is nontrivial. To check it one has to use the selfduality of the field (\[selfdualityI\]) and the Schouten identity ${\epsilon}_{[{\mu}{\nu}\rho{\sigma}} \eta_{\tau]\beta}=0$.\ It is interesting to note that, if one replaces the selfdual field $c^-_{{\mu}{\nu}}$ with an antiselfdual field $c^+_{{\mu}{\nu}}$, the multiplet closes exactly in the same way, i.e. with no difference in signs or factors.\ The set of dynamical equations with mass $$\begin{aligned} &&\Box a=\hat z^2 a\,;\cr &&\Box \tilde b_{\mu}=\hat z^2 \tilde b_{\mu}\,; \cr &&\Box c^-_{{\mu}{\nu}}=\hat z^2 c^-_{{\mu}{\nu}}\end{aligned}$$ is consistent with VSUSY but cannot be naturally derived from an action.\ Finally, we would like to discuss the possibility of rotating this multiplet to Minkowski space. Since, as mentioned before, selfdual fields in Minkowski space are necessarily complex and a model featuring only selfdual fields cannot exist, one expects the whole multiplet to be complex in Minkowski space. This means that there will be a doubling of the number of degrees of freedom and that in fact two multiplets will be present, $(a, \tilde b_{\mu}, c^-_{{\mu}{\nu}})$ and $(a^*,\tilde b_{\mu}^*, c^+_{{\mu}{\nu}} )$ where of course $c^+_{{\mu}{\nu}}=(c^-_{{\mu}{\nu}})^*$. These two multiplets do not mix in the VSUSY transformations and are only related by complex conjugation. This situation is analogous to what happens with the chiral multiplet of ordinary supersymmetry in four dimensions written in complex notation. One can check that, by taking the complex conjugate of the multiplet of Table \[tablevectormultipletI\], given in Table \[tablevectormultipletII\], and by taking into account the conventions for the transformations of the fields (\[fulltrans\]) and (\[fulltransZ\]), one indeed obtains a copy of the VSUSY algebra for the complex conjugate fields, with the only change in the selfduality property of the tensor field, i.e. everything is consistent. $\delta_{Q_5}$ $\delta_{Q_{\mu}}$ -------------------- ----------------------------------------------------------------------------------------------- --------------------------------------------------------------------------------------------- $a^*$ $\frac{1}{\hat z}{\epsilon}\partial^{\sigma}\tilde b_{\sigma}^*$ $-i{\epsilon}^{\mu}\tilde b_{\mu}^*$ $\tilde b_{\nu}^*$ ${\epsilon}(-\frac{2}{\hat z}\partial^\rho c_{{\nu}\rho}^{+} + \frac{1}{2}\partial_{\nu}a^*)$ $-{\epsilon}^{\mu}(2i c_{{\mu}{\nu}}^{+} + i \frac{\hat z}{2}\eta_{{\mu}{\nu}} a^*)$ $c^+_{{\nu}\rho}$ $\frac{1}{4}{\epsilon}\delta^+_{{\nu}\rho,{\sigma}\tau}\partial^{\sigma}\tilde b^{*\tau}$ $-i \frac{\hat z}{4}{\epsilon}^{\mu}\delta^+_{{\nu}\rho,{\mu}{\sigma}}\tilde b^{*{\sigma}}$ : VSUSY transformations of the complex conjugate fields (scalar, vector, antiselfdual two-form). \[tablevectormultipletII\] $Y=1/2$ multiplet from superspace --------------------------------- In this section we derive the $Y=\frac{1}{2}$ multiplet of Table \[tablevectormultipletI\] by superspace techniques using the superspace setup of Section 3.3. We have seen that this multiplet closes exactly in the same way under vector supersymmetry in Euclidean and Minkowski spacetime. In Minkowski the fields must be complex and the complex conjugate fields transform into each other according to Table \[tablevectormultipletII\]. However, since VSUSY does not mix the fields and their complex conjugates in Minkowski space, in this section we do not have to worry about this doubling and simply consider the multiplet of Table \[tablevectormultipletI\] in either Euclidean or Minkowski spacetime. ### Scalar Superfield Let us consider a generic scalar superfield: $$\begin{aligned} &T(x^{\mu},\theta^{\mu},\theta_{5},z)=a(x,z)+ \tilde{a}(x,z)\theta_{5}+ \tilde b_{\mu}(x,z)\theta^{\mu}+ b_{\mu}(x,z)\theta^{\mu}\theta_{5}+\nonumber\\ &+c_{\mu\nu}(x,z)\theta^{\mu}\theta^{\nu}+ \tilde{c}_{\mu\nu}(x,z) \theta^{\mu}\theta^{\nu}\theta_{5} +\tilde d_{\sigma}(x,z) \epsilon^{\sigma}_{\,\,\,\mu\nu\rho}\,\theta^{\mu}\theta^{\nu} \theta^{\rho} +d_{\sigma}(x,z) \epsilon^{\sigma}_{\,\,\,\mu\nu\rho}\, \theta^{\mu}\theta^{\nu} \theta^{\rho}\theta_{5}+\nonumber\\ &+f(x,z) \epsilon_{\mu\nu\rho\sigma}\, \theta^{\mu}\theta^{\nu} \theta^{\rho} \theta^{\sigma}+ \tilde{f}(x,z) \epsilon_{\mu\nu\rho\sigma}\, \theta^{\mu}\theta^{\nu} \theta^{\rho} \theta^{\sigma}\theta_{5},\cr & \label{scalarsuperfield}\end{aligned}$$ where we assume as before that all components can be Taylor-expanded in $z$ as in (\[zetaexp\]) and where $c^{(n)}_{\mu\nu}$ and $\tilde c^{(n)}_{\mu\nu}$ are two-forms. It is useful to split them in their self-dual and anti-self-dual parts $$c^{(n)}_{\mu\nu}=c^{(n)+}_{\mu\nu}+c^{(n)-}_{\mu\nu}\,, \quad \quad \quad \quad \tilde c^{(n)}_{\mu\nu}=\tilde c^{(n)+}_{\mu\nu}+\tilde c^{(n)-}_{\mu\nu}, \label{split}$$ according to equations (\[selfdualityI\]) (\[selfdualityII\]) and (\[projectors\]). ### Supercovariant constraints The idea is now to impose a covariant constraint on $T$ that will let us work with only the self-dual parts of $c^{(n)}_{\mu\nu}$ and $\tilde c^{(n)}_{\mu\nu}$. We choose the following constraint: $$\epsilon^{\mu\nu\rho\sigma}D_{\mu}D_{\nu}D_{\rho}D_{\sigma}T=-6i\, \Box T. \label{c1}$$ After a quite long calculation one can show that the following relations must hold for any integer $n=0,1,2...$: $$\begin{aligned} c^{(n)+}_{\mu\nu}=0 &\quad\quad\quad\quad &\tilde c^{(n)+}_{\mu\nu}=\frac{i}{8}\delta_{\mu\nu,\rho\sigma}^{+}\partial^{\rho}\tilde b^{\sigma (n)} \nonumber\\ \tilde d^{(n)}_{\mu}=\frac{1}{12}\tilde b^{(n+1)}_{\mu} &\quad\quad\quad\quad& d^{(n)}_{\mu}=\frac{1}{12}b_{\mu}^{(n+1)}+\frac{i}{24}\partial_{\mu}a^{(n+1)}+\frac{1}{6}\partial^{\nu}c_{\mu\nu}^{(n)-}\nonumber\\ f^{(n)}=\frac{i}{96}\,\Box a^{(n)}&\quad\quad\quad\quad &\tilde f^{(n)}=\frac{i}{96}\,\left(\Box\tilde a^{(n)}+ \partial^{\sigma}\tilde b^{(n+1)}_{\sigma}\right)\, , \label{expscalar}\end{aligned}$$ We are left with six infinite sets of independent complex component fields whose VSUSY transformations are given in Table \[tableinfinite\]. --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- $Q_{5}$ $Q_{\mu}$ $Z$ ------------------------------- -------------------------------------------------------------------------------------------------------------- --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- ------------------------------------ $a^{(n)}$ $-\tilde a^{(n)}$ $-\tilde b_{\mu}^{(n)}$ $i\,a^{(n+1)}$ $\tilde a^{(n)}$ $-\frac{i}{2}a^{(n+1)}$ $b_{\mu}^{(n)}-\frac{i}{2}\partial_{\mu}a^{(n)}$ $i\,\tilde a^{(n+1)}$ $\tilde b_{\nu}^{(n)}$ $-b_{\nu}^{(n)}-\frac{i}{2}\partial_{\nu}a^{(n)}$ $2c_{\mu\nu}^{(n)-}+\frac{i}{2}\eta_{\mu\nu}a^{(n+1)}$ $i\, \tilde b_{\nu}^{(n+1)}$ $b_{\nu}^{(n)}$ $\frac{i}{2}\partial_{\nu}\tilde a^{(n)}-\frac{i}{2}\tilde b_{\nu}^{(n+1)}$ $-\frac{i}{4}\delta^{+}_{\mu\nu,\rho\sigma}\partial^{\rho}\tilde b^{\sigma (n)}-\frac{i}{2}\partial_{\mu}\tilde b_{\nu}^{(n)}-2\tilde c_{\mu\nu}^{(n)-}-\frac{i}{2}\eta_{\mu\nu}\tilde a^{(n+1)}$ $i\,b_{\nu}^{(n+1)}$ $c_{\nu\rho}^{(n)-}$ $\frac{i}{8}\delta^{-}_{\nu\rho,\sigma\tau}\partial^{\sigma}\tilde b^{\tau (n)}-{\tilde c}_{\nu\rho}^{(n)-}$ $-\frac{i}{4}\delta^{-}_{\nu\rho,\mu\sigma}\tilde b^{\sigma(n+1)}$ $i\,c_{\nu\rho}^{(n+1)-}$ ${\tilde c}_{\nu\rho}^{(n)-}$ $-\frac{i}{8}\delta^{-}_{\nu\rho,\sigma\tau}\partial^{\sigma} b^{\tau (n)}-\frac{i}{2}c_{\nu\rho}^{(n+1)-}$ $\frac{i}{4}\delta^{-\,\,\,\,\,\,\,\sigma}_{\nu\rho,\mu}(\frac{i}{4}\partial_{\sigma}a^{(n+1)}+b_{\sigma}^{(n+1)}+\partial^{\tau}c_{\sigma\tau}^{(n)-}) $i\,{\tilde c}_{\nu\rho}^{(n+1)-}$ -\frac{i}{2}\partial_{\mu}c_{\nu\rho}^{(n)-}$ --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- : Action of VSUSY charges on the six infinite sets of component fields solving constraint (\[c1\]) (two sets of scalars, two of vectors and two of selfdual two-forms).\[tableinfinite\] Now we impose another covariant constraint involving $D_{5}$ in order to further reduce our multiplet: $$Z D_{5}T=-i\partial^{\mu}D_{\mu}T\, . \label{c2}$$ It is not hard to prove that this constraint implies that $\Box T=Z^{2}T$, so that the higher order components of the fields are related to the lower ones via relations similar to those appearing in (\[higher\]) and the odd Casimir is present and has value zero. In practice, we need to consider only the components of order 0 and 1 in the $z$ expansion. One can show that the constraint (\[c2\]) yields: $$\begin{aligned} &\tilde a^{(1)}=-\partial^{\mu}\tilde b_{\mu}^{(0)}\,; \quad\quad\quad& \Box \tilde a^{(0)}=-\partial^{\mu}\tilde b_{\mu}^{(1)}\,;\nonumber\\ &b_{\mu}^{(1)}=-2\partial^{\nu}c_{\mu\nu}^{(0)-}\,; \quad\quad\quad& \Box b_{\mu}^{(0)}=-2\partial^{\nu}c_{\mu\nu}^{(1)-}\,; \nonumber\\ &\tilde c_{\mu\nu}^{(0)-}=-\frac{i}{8}\delta_{\mu\nu,\rho\sigma}^{-} \partial^{\rho}\tilde b^{\sigma (0)}\,;\quad\quad\quad& \tilde c_{\mu\nu}^{(1)-}=-\frac{i}{8}\delta_{\mu\nu,\rho\sigma}^{-} \partial^{\rho}\tilde b^{\sigma (1)}\,.\end{aligned}$$ Now, if we want to consider $Z$ not as an operator but as a number, indicated as usual with $\hat z$, it is sufficient to impose that $ZT=\hat z T$. We have then reduced our multiplet to three independent on-shell component fields, the scalar $a^{(0)}$, the vector $\tilde b_{\mu}^{(0)}$ and the selfdual two-form $c_{\mu\nu}^{(0)-}$. Their VSUSY transformations coincide with the ones in Table \[tablevectormultipletI\] with the identification $a^{(0)}\equiv a$ and similar for the other fields. Deriving actions from superspace ================================ In this section we initiate the study of VSUSY-invariant actions from the superspace point of view. As a simplification, we choose as before $\tilde Z=-Z$. Moreover, we will always consider $Z$ as a number, denoted by $\hat z$ to distinguish it from the bosonic coordinate $z$. Therefore, we will not derive the action for the four-spinor multiplet with nontrivial central charge operator discussed in Section 3.2. To derive that, the method of Sohnius [@Sohnius:1978fw] should be implemented and we choose to leave this issue for future work. Let us first set the dimensions (in mass units) of the coordinates and the covariant derivatives: $$\begin{aligned} &[x^{\mu}]=-1\,;\quad[\theta^{\mu}]=-\frac{1}{2}\,;\quad [\hat z]=1\,;\nonumber\\ &[dx^{\mu}]=-1\,;\quad [d\theta]=\frac{1}{2}\,;\quad[\partial_{\mu}]=1\,;\quad [D_{\mu}]=\frac{1}{2}\,; \quad [D_{5}]=\frac{1}{2}\,.\end{aligned}$$ We define a general action in the superspace as $$S=\int \,d^{4}x\,d^{4}\theta\,d\theta_{5}\,\mathcal{L}\,. \label{action}$$ Since $S$ has to be even and dimensionless, we have that $\mathcal{L}$ must be odd and of dimension $3/2$. We choose the dimension of the spinorial superfields $\Phi$, defined in (\[spinorsuperfield\]), and $\tilde \Lambda$, related to $\Phi$ by (\[first\]), as $$[\Phi]=\frac{3}{2}\,, \quad \quad \quad \quad [\tilde\Lambda]=2\,,$$ from which one can derive the dimensions of their component fields $$[\phi^{(0)}]=\frac{3}{2}\,;\quad[\tilde\phi^{(0)}]=2\,;\quad [\tilde\lambda^{(0)}]=2\quad[\lambda^{(0)}]=\frac{5}{2}\,.$$ In the same way, we choose a dimension for the scalar superfield $T$ defined in (\[scalarsuperfield\]) $$[T]=1 ,$$ from which one can derive the dimensions of its component fields $$[a^{(0)}]=1\,,\quad[\tilde{a}^{(0)}]=\frac{3}{2}\,,\quad [\tilde{b}_{\mu}^{(0)}]=\frac{3}{2}\,,\quad[b_{\mu}^{(0)}]=2\,,\quad[c_{\mu\nu}^{(0)-}]=2\,,\quad[\tilde{c}_{\mu\nu}^{(0)-}]=\frac{5}{2}\,.$$ Now, since our goal is to build a VSUSY invariant Lagrangian $\mathcal{L}$, we need to set the conventions for the Dirac conjugates of the superfields. As much as we have chosen, in Section 2.1, our supersymmetry parameters ${\epsilon}$ and ${\epsilon}_{\mu}$ to be real by convention, we now choose the superspace coordinates to be real $$(\theta_{\mu})^*=\theta_{\mu}\,,\quad \quad \quad \quad (\theta_{5})^*=\theta_{5}\,.$$ With the conventions adopted in (\[conv\]) and (\[fermionbil\]), one can see that the expansion of the bar of the spinor superfield $\Phi$ becomes $$\begin{aligned} &\bar\Phi\equiv i\Phi^{\dagger}\gamma_{0}=\bar{\phi}^{(0)}-\bar{\tilde\phi}^{(0)}\theta_{5}-\bar{\tilde\psi}^{(0)}_{\mu}\theta^{\mu}- {\bar\psi}^{(0)}_{\mu}\theta^{\mu}\theta_{5} -\bar{\xi}^{(0)}_{\mu\nu}\theta^{\mu}\theta^{\nu} + \bar{\tilde\xi}^{(0)}_{\mu\nu}\theta^{\mu}\theta^{\nu}\theta_{5}+\nonumber\\ &+\bar{\tilde\chi}_{\mu\nu\rho}^{(0)}\theta^{\mu}\theta^{\nu}\theta^{\rho}+\bar{\chi}_{\mu\nu\rho}^{(0)}\theta^{\mu}\theta^{\nu}\theta^{\rho}\theta_{5} +\bar{\zeta}_{\mu\nu\rho\sigma}^{(0)}\theta^{\mu}\theta^{\nu}\theta^{\rho}\theta^{\sigma}- \bar{\tilde\zeta}_{\mu\nu\rho\sigma}^{(0)}\theta^{\mu}\theta^{\nu}\theta^{\rho}\theta^{\sigma}\theta_{5}\,,\end{aligned}$$ where (cfr. (\[exp\]) and (\[sup\])) $$\begin{aligned} \bar{\tilde\psi}^{(0)}_{\mu}=-\bar{\tilde\lambda}^{(0)}\gamma_{\mu}&\quad\quad\quad\quad&{\bar\psi}^{(0)}_{\mu}= -\bar{\lambda}^{(0)}\gamma_{\mu}+\frac{i}{2}\partial_{\mu}\bar{\phi}^{(0)}\nonumber\\ \bar{\xi}^{(0)}_{\mu\nu}=\frac{\hat z}{4}\bar{\phi}^{(0)}\gamma_{\mu\nu}&\quad\quad\quad\quad& \bar{\tilde\xi}^{(0)}_{\mu\nu}=\frac{\hat z}{4}\bar{\tilde\phi}^{(0)}\gamma_{\mu\nu} +\frac{i}{2}\partial_{[\mu}\bar{\tilde\lambda}^{(0)}\gamma_{\nu]}\nonumber\\ \bar{\tilde\chi}_{\mu\nu\rho}^{(0)}=-\frac{\hat z}{12} \bar{\tilde\lambda}^{(0)}\gamma_{\mu\nu\rho}&\quad\quad\quad\quad& \bar{\chi}_{\mu\nu\rho}^{(0)}=-\frac{\hat z}{12} \bar{\lambda}^{(0)}\gamma_{\mu\nu\rho}+\frac{\hat z}{8} i\partial_{[\mu}\bar{\phi}^{(0)}\gamma_{\nu\rho]}\nonumber\\ \bar{\zeta}_{\mu\nu\rho}^{(0)}=\frac{\hat{z}^{2}}{96}\bar{\phi}^{(0)}\gamma_{\mu\nu\rho\sigma}&\quad\quad\quad\quad& \bar{\tilde\zeta}_{\mu\nu\rho}^{(0)}=\frac{\hat{z}^{2}}{96}\bar{\tilde\phi}^{(0)}\gamma_{\mu\nu\rho\sigma} +\frac{\hat z}{24}i\partial_{[\mu}\bar{\tilde\lambda}^{(0)}\gamma_{\nu\rho\sigma]}\,.\end{aligned}$$ Analogously, for the complex conjugate of the scalar superfield $T$ we have $$\begin{aligned} &\bar T=\bar{a}^{(0)}-\bar{\tilde a}^{(0)}\theta_{5}-\bar{\tilde b}^{(0)}_{\mu}\theta^{\mu}- {\bar b}^{(0)}_{\mu}\theta^{\mu}\theta_{5} -\bar{c}^{(0)}_{\mu\nu}\theta^{\mu}\theta^{\nu} + \bar{\tilde c}^{(0)}_{\mu\nu}\theta^{\mu}\theta^{\nu}\theta_{5}+ \bar{\tilde d}_{\sigma}^{(0)}\epsilon^{\sigma}_{\,\,\mu\nu\rho}\theta^{\mu}\theta^{\nu}\theta^{\rho}\nonumber\\ &+\bar{d}_{\sigma}^{(0)}\epsilon^{\sigma}_{\,\,\mu\nu\rho}\theta^{\mu}\theta^{\nu}\theta^{\rho}\theta_{5} +\bar{f}^{(0)}\epsilon_{\mu\nu\rho\sigma}\theta^{\mu}\theta^{\nu}\theta^{\rho}\theta^{\sigma}- \bar{\tilde f}^{(0)}\epsilon_{\mu\nu\rho\sigma}\theta^{\mu}\theta^{\nu}\theta^{\rho}\theta^{\sigma}\theta_{5}\,,\end{aligned}$$ where (cfr. (\[expscalar\])) $$\begin{aligned} \bar{c}^{(0)-}_{\mu\nu}=0 &\quad\quad\quad\quad &\bar{\tilde c}^{(0)-}_{\mu\nu}=-\frac{i}{8}\delta_{\mu\nu,\rho\sigma}^{-}\partial^{\rho} \bar{\tilde b}^{\sigma (0)}\nonumber\\ \bar{\tilde d}^{(0)}_{\mu}=\frac{i}{12}\hat z\bar{\tilde b}^{(0)}_{\mu} &\quad\quad\quad\quad& \bar{d}^{(0)}_{\mu}=\frac{i}{12} \hat z \bar{b}_{\mu}^{(0)}+\frac{1}{24}\hat z \partial_{\mu}\bar{a}^{(0)}+\frac{1}{6}\partial^{\nu}\bar{c}_{\mu\nu}^{(0)+}\nonumber\\ \bar{f}^{(0)}=-\frac{i}{96}\,\Box \bar{a}^{(0)}&\quad\quad\quad\quad &\bar{\tilde f}^{(0)}=-\frac{i}{96}\,\left(\Box\bar{\tilde a}^{(0)}+ i\hat z\partial^{\sigma}\bar{\tilde b}^{(0)}_{\sigma}\right)\, .\end{aligned}$$ Klein-Gordon-type action ------------------------ Here we work with the superfields $\Phi$ and $\tilde\Lambda$ defined in Section 4 and satisfying only constraint (\[first\]). Since $Z$ is treated as a number, we are actually working with four component fields, $\phi^{(0)}$, $\tilde\phi^{(0)}$, $\tilde\lambda^{(0)}$ and $\lambda^{(0)}$, transforming as shown in Table \[tablenine\].\ From (\[action\]) it is clear that the only term of the Lagrangian that survives integration is the one of highest order in $\theta_{\mu}$ and $\theta_{5}$, denoted by ${\mathcal L}|_{\Theta^{4}\theta_{5}}$: $$S=\int \,d^{4}x\,d^{4}\theta\,d\theta_{5}\,{\mathcal L}|_{\Theta^{4}\theta_{5}} \,\Theta^{4}\theta_{5}=\int \,d^{4}x\,{\mathcal L}|_{\Theta^{4}\theta_{5}}\,,$$ where we define $$\Theta^{4}\equiv\frac{1}{4!}{\epsilon}_{\mu\nu\rho\sigma}\theta^{\mu}\theta^{\nu}\theta^{\rho}\theta^{\sigma}.$$ We start by building the simplest combinations of $\Phi$ and $\tilde\Lambda$ that give odd Lagrangian terms of dimension $3/2$ $${\mathcal L}_{mass}=\frac{i}{\hat{z}^{2}}\bar\Phi\tilde\Lambda\,, \quad \quad \quad \quad {\mathcal L}_{kin}=\frac{1}{\hat{z}^{3}}\bar\Phi{\not\hspace{-.05cm}\partial}D_{5}\Phi\,. \label{lagr1}$$ After some lengthy calculations one can show that these two terms give the following contributions $$\begin{aligned} &{\mathcal L}_{mass}|_{\Theta^{4}\theta_{5}}=2\left(\bar{\tilde\phi}^{(0)}\gamma_{5}\tilde\lambda^{(0)} - \bar{\phi}^{(0)}\gamma_{5}\lambda^{(0)} + \bar{\lambda}^{(0)}\gamma_{5}\phi^{(0)} - \bar{\tilde\lambda}^{(0)}\gamma_{5}\tilde\phi^{(0)}\right)\,,\\ &{\mathcal L}_{kin}|_{\Theta^{4}\theta_{5}}=-\frac{2}{{\hat z}^{2}}\left(\bar{\tilde\phi}^{(0)}\gamma_{5}\,\Box\,\tilde\lambda^{(0)} - \bar{\phi}^{(0)}\gamma_{5}\,\Box\,\lambda^{(0)} + \bar{\lambda}^{(0)}\gamma_{5}\,\Box\,\phi^{(0)} - \bar{\tilde\lambda}^{(0)}\gamma_{5}\,\Box\,\tilde\phi^{(0)}\right)\,.\end{aligned}$$ Now we take the following Lagrangian $${\mathcal L_{1}}\equiv {\mathcal L}_{kin}+{\mathcal L}_{mass}\,.$$ Integrating it over the supercoordinates and removing the constant factor $\frac{2}{\hat z^2}$, we obtain the action $$S_{1}=\int d^{4}x~ (\bar{\phi}^{(0)}\gamma_{5}(\Box-\hat{z}^{2})\lambda^{(0)} - \bar{\tilde\phi}^{(0)}\gamma_{5}(\Box-\hat{z}^{2})\tilde\lambda^{(0)} + \bar{\tilde\lambda}^{(0)}\gamma_{5}(\Box-\hat{z}^{2})\tilde\phi^{(0)} - \bar{\lambda}^{(0)}\gamma_{5}(\Box-\hat{z}^{2})\phi^{(0)}), \label{actiontable}$$ which gives rise, as expected, to Klein-Gordon equations of motion for the component fields with the usual identification $\hat z=m$.\ To avoid confusion, some comments are in order. First of all we remind the reader that (\[actiontable\]) is the action for the multiplet given in Table \[tablenine\]. This multiplet was for us an intermediate step in deriving the spinning particle multiplet, given in Table \[tablespinning\], compatible with a set of Dirac equations. This step will be done at the level of superfields in the next section. However, as the reader might remember, another $Y=0$ multiplet with a Klein-Gordon type action appears in this paper, namely in Section 3.2. We would like to stress that action (\[action4\]) presented there is different from (\[actiontable\]), since it has diagonal kinetic terms for the physical fields and off-diagonal ones for the ghost fields, while in (\[actiontable\]) all terms are off-diagonal. As a result, action (\[action4\]) is compatible with Majorana condition for the spinors while (\[actiontable\]) is not. As already mentioned before, the method of Sohnius [@Sohnius:1978fw] for deriving actions from an extended superspace with central charge must be used to derive (\[actiontable\]) and we leave this for future work. Spinning particle Dirac-type action ----------------------------------- We know that the component fields appearing in the previous subsection do not satisfy Dirac-type equations of motion. Their VSUSY transformations are not compatible with Dirac equations. However, we have also shown that, if we take some linear combinations of those fields (\[lincomb\]), the resulting multiplet is compatible with the Dirac equations. The idea is to now understand how these linear combinations can be implemented at the level of superfields.\ In order to do so, we start by considering a new superfield $\Psi$, built out of $\Phi$ and $\tilde\Lambda$ $$\Psi= -\frac{i}{2}\Phi+\frac{1}{\hat z}D_{5}\tilde\Lambda\,.$$ Now, we make a redefinition of the field components inside $\Psi$ according to (\[lincomb\]). This leads to the following expansion for $\Psi$ $$\begin{aligned} &\Psi=\psi_{1}+\sqrt{\frac{\hat z}{2}}\tilde\psi_{2}\theta_{5}+\frac{1}{\sqrt{2\hat z}}\left((z\gamma_{\mu}-\partial_{\mu})\gamma_{5}\tilde\psi_{1} +i\partial_{\mu}\tilde\psi_{2}\right)\theta^{\mu}+\frac{1}{2}(z\gamma_{\mu}-\partial_{\mu})\gamma_{5}\psi_{2}\theta^{\mu}\theta_{5}+\nonumber\\ &-\left(\frac{\hat z}{4}\gamma_{\mu\nu}\psi_{1}\+\frac{1}{2}\gamma_{\mu}\partial_{\nu}\psi_{1}+\frac{i}{2}\gamma_{5}\gamma_{\mu}\partial_{\nu}\psi_{2}\right) \theta^{\mu}\theta^{\nu} - \left(\frac{\hat z}{4}\sqrt{\frac{\hat z}{2}}\gamma_{\mu\nu}\tilde\psi_{2}+\frac{1}{2}\sqrt{\frac{\hat z}{2}}\gamma_{\mu}\partial_{\nu}\tilde\psi_{2}\right)\theta^{\mu}\theta^{\nu}\theta_{5}+\nonumber\\ &+\sqrt{\frac{\hat z}{2}}\left(\frac{\hat z}{12}\gamma_{5}\gamma_{\mu\nu\rho}\tilde\psi_{1}+\frac{1}{4}\gamma_{5}\gamma_{\mu\nu}\partial_{\rho}\tilde\psi_{1} -\frac{i}{4}\gamma_{\mu\nu}\partial_{\rho}\tilde\psi_{2} \right)\theta^{\mu}\theta^{\nu}\theta^{\rho} +\frac{\hat z}{8}\gamma_{5}\left(\frac{\hat z}{3}\gamma_{\mu\nu\rho}\psi_{2} + \gamma_{\mu\nu}\partial_{\rho}\psi_{2}\right)\theta^{\mu}\theta^{\nu}\theta^{\rho}\theta_{5}+\nonumber\\ &+\frac{\hat{z}}{4}\left(\hat{z} i\gamma_{5}\psi_{1}+i\gamma_{5}{\not\hspace{-.05cm}\partial}\psi_{1}-{\not\hspace{-.05cm}\partial}\psi_{2}\right)\Theta^{4}+ \frac{\hat{z}}{4}\left(\hat{z}\sqrt{\frac{\hat z}{2}}i\gamma_{5}\tilde\psi_{2}+\sqrt{\frac{\hat z}{2}}i\gamma_{5}{\not\hspace{-.05cm}\partial}\tilde\psi_{2}\right)\Theta^{4}\theta_{5}\,.\end{aligned}$$ Notice that the dimension of $\Psi$ is $3/2$ so for the component fields we have $$[\psi_{1}]=[\tilde\psi_{1}]=[\tilde\psi_{2}]=[\psi_{2}]=\frac{3}{2}\,.$$ Inspired by (\[lagr1\]), we now construct other Lagrangian terms of the form $${\mathcal L_{box}}=\frac{1}{\hat {z}^{3}}\bar\Psi{\not\hspace{-.05cm}\partial}D_{5}\Psi\,, \quad \quad \quad \quad {\mathcal L_{slash}}=\frac{i}{\hat{z}^{2}}\bar\Psi {\not\hspace{-.1cm}D}\Psi\,. \label{lagr2}$$ After some lengthy calculations one can show that these two terms give the following contributions: $$\begin{aligned} &{\mathcal L}_{box}|_{\Theta^{4}\theta_{5}}=\frac{i}{\hat z}\left(\bar{\tilde\psi}_{1}\gamma_{5}\,\Box\,\tilde\psi_{1} - \bar{\psi}_{1}\gamma_{5}\,\Box\,\psi_{1} - \bar{\psi}_{2}\gamma_{5}\,\Box\,\psi_{2} + \bar{\tilde\psi}_{2}\gamma_{5}\,\Box\,\tilde\psi_{2}\right)\nonumber\\ &{\mathcal L}_{slash}|_{\Theta^{4}\theta_{5}}=-\frac{i}{\hat z}\left(\bar{\tilde\psi}_{1}\gamma_{5}\,\Box\,\tilde\psi_{1} - \bar{\psi}_{1}\gamma_{5}\,\Box\,\psi_{1} - \bar{\psi}_{2}\gamma_{5}\,\Box\,\psi_{2} + \bar{\tilde\psi}_{2}\gamma_{5}\,\Box\,\tilde\psi_{2}\right)+\nonumber\\ &+4\left(-\bar{\psi}_{1}({\not\hspace{-.05cm}\partial}+\hat z)\psi_{2} + \bar{\tilde\psi}_{1}({\not\hspace{-.05cm}\partial}+\hat z)\tilde\psi_{2} + \bar{\tilde\psi}_{2}({\not\hspace{-.05cm}\partial}+\hat z)\tilde\psi_{1} - \bar{\psi}_{2}({\not\hspace{-.05cm}\partial}+ \hat z)\psi_{1}\right).\end{aligned}$$ Now we take the following Lagrangian: $${\mathcal L_{2}}\equiv \frac{1}{4}\left({\mathcal L}_{box}+{\mathcal L}_{slash}\right).$$ By integrating it over the supercoordinates, we obtain the Dirac-type action (\[actionstandard\]), with the usual identification $\hat z=m$. Multiplets with $Z=\tilde Z=0$ ============================== In [@Casalbuoni:2008ez], the Casimir operators in the case of vanishing central charges were not studied and left for future work. However, it is easy to derive VSUSY representations in the case of vanishing central charge via a superspace approach. It is then worth to discuss them here. Dropping the $z$ coordinate in superspace, the supercharges (\[chargessuper\]) become $$\label{VSUSYsuperspaceRealisation} Q_\mu=\frac{\partial}{\partial\theta^\mu}-\frac{i}{2}\theta_5\frac{\partial}{\partial x^\mu} \,,~~~~~~~~~~~~~~~~ Q_5=\frac{\partial}{\partial\theta_5}-\frac{i}{2}\theta^\mu\frac{\partial}{\partial x^\mu}$$ and the associated fermionic covariant derivatives are $$\label{CovariantDerivatives} D_\mu=\frac{\partial}{\partial\theta^\mu}+\frac{i}{2}\theta_5\frac{\partial}{\partial x^\mu} \,,~~~~~~~~~~~~~~~~~ D_5=\frac{\partial}{\partial\theta_5}+\frac{i}{2}\theta^\mu\frac{\partial}{\partial x^\mu}\,.$$ Field representations can be obtained in a straightforward way by imposing the constraints $D_\mu \Phi=0$ and $D_5 \Phi=0$ on a generic superfield $\Phi$. The obtained representations will be the VSUSY analogue of the chiral and antichiral superfields for ordinary supersymmetry. We start with the case of a scalar superfield, but it will be then clear that exactly the same procedure works for a superfield with another Lorentz structure. $D_\mu\Phi=0$ Multiplet ----------------------- The constraint $D_\mu \Phi=0$ removes all non-scalar component fields and as a result we get the multiplet in components given in Table \[VectorMultiplet\]. $Q_5$ $Q_\mu$ -------------- -------------- ------------------- \[-2ex\] $A$ $i\tilde{B}$ $0$ $\tilde{B}$ $0$ $-\partial_\mu A$ \[0.5ex\] : Transformation rules for the $D_\mu\Phi=0$ multiplet. \[VectorMultiplet\] When one starts with a superfield with extra vectorial or a spinorial indices, both component fields inherit the Lorentz index structure from the superfield. The interacting action $$S=\int d^{4}x \left( \partial_\mu\bar{A}\partial^\mu\tilde{B} +\partial_\mu\bar{\tilde{B}} \partial^\mu A\right) + (\bar A\tilde{B} + \bar{\tilde{B}}A)V(\bar A A)$$ is invariant under the transformations of Table \[VectorMultiplet\], where $V$ is an arbitrary analytical function and the bar denotes complex conjugation for scalars, vectors etc. and the Dirac conjugation defined in (\[daggerconv\]) for spinors. Observe that this action has odd Grassmann parity. A discussion and examples of odd actions at the classical level can be found in [@Soroka:1995et], [@Soroka:2001jg]. At the level of quantization odd actions seem problematic, so further investigations would be needed in that direction.\ One might wonder whether it could be possible to construct an even action for the multiplet given in Table \[VectorMultiplet\]. Since the fields are complex, a diagonal kinetic term for the ghost of the form $\partial_{\mu}\bar{\tilde B} \partial^{\mu}\tilde B$ is not zero. However, the combination of this with a term of the form $\partial_{\mu}\bar{A} \partial^{\mu}A$ cannot be rendered VSUSY invariant. Therefore, the only possibility to have both a dynamical physical field and a dynamical ghost is to construct mixed, off-diagonal, kinetic terms for them. Due to the general structure of this multiplet, containing only one physical field and one ghost, the mixed, off-diagonal kinetic terms necessarily feature one physical field and one ghost and are therefore odd. $D_5\Phi=0$ Multiplet --------------------- A similar procedure with the scalar covariant derivative can be applied, resulting in the multiplet in Table \[ScalarMultiplet\]. A similar multiplet is derived by Kato et al. [@Kato:2005fj], starting from a twisted topological theory. $Q_5$ $Q_\mu$ -------------------- -------------------------------------------------------------------------------- ------------------------------------------------------ \[-2ex\] $A$ $0$ $i\tilde{F}_\mu$ $\tilde{F}_\alpha$ $-\partial_\alpha A$ $-iM_{\mu\alpha}$ $M_{\alpha\beta}$ $\partial_{[\alpha}\tilde{F}_{\beta]}$ $ -i\epsilon_{\alpha\beta\mu\gamma}\tilde{K}^\gamma$ $\tilde{K}_\alpha$ $\frac{1}{2}\epsilon_{\alpha\beta\gamma\delta}\partial^\beta M^{\gamma\delta}$ $-iH\eta_{\alpha\mu}$ $H$ $\partial_\alpha\tilde{K}^\alpha$ $0$ \[0.5ex\] : Action of the VSUSY generators on the multiplet with vanishing central charge and $D_5\Phi=0$.\[ScalarMultiplet\] One can see by inspection that there exists a submultiplet of the multiplet in Table \[ScalarMultiplet\], consisting of a vector and a scalar, shown in Table \[IrrScalMultiplet\]. $Q_5$ $Q_\mu$ ------------------------------------------------------------- ------------------- ------------------ $A$ $0$ $i\tilde{F}_\mu$ $\tilde{F}_\nu$, with $\partial_{[\mu}\tilde{F}_{\nu]}=0 $ $-\partial_\nu A$ $0$ : Action of the VSUSY charges on the irreducible vector-scalar multiplet with vanishing central charge.\[IrrScalMultiplet\] In order to close the algebra on this submultiplet, one has to impose the condition that $\tilde{F}_\nu$ has zero curl, and thus $\partial_\mu\tilde{F}_\nu=\partial_\nu\tilde{F}_\mu$. This implies that this multiplet can be written in terms of the one in Table \[VectorMultiplet\]. Conclusions and outlook ======================= In this paper we have constructed some field representations of vector supersymmetry with nonvanishing central charge, characterized by superspin $Y=0$ and $Y=1/2$. We have discussed their free dynamics in terms of equations of motion and, when possible, of an action. Furthermore, we have developed a superspace setup for vector supersymmetry and we have derived our multiplets in this setting. Concerning the construction of actions by superspace techniques, we have only discussed the simplest case of multiplets where the central charge operator acts diagonally on the fields. We leave the other, more involved case for future work. We have pointed out and worked out in detail the connection between two of our multiplets and some existing results in the literature, in one case in the context of spinning particle models [@Casalbuoni:2008iy] and in the other in the context of topological theories with twisted supersymmetry [@Kato:2005gb]. Finally, we have constructed some representations with vanishing central charge by superspace techniques. For one of those we have succeeded in writing an invariant interacting action, which is however quite bizzarre, first of all because it is odd. The need of odd actions for certain supersymmetric systems had been pointed out in the literature before [@Soroka:1995et] [@Soroka:2001jg], but their quantization remains to our knowledge an open problem. Another unsettling issue concerning one of our multiplets, with central charge this time, is that, despite the fact the fields are spinors, the equations of motion have to be of Klein-Gordon type to be compatible with vector supersymmetry. Therefore, by just looking at the action the spacetime group seems to be decoupled from the group with respect to which the fields are spinors and the symmetry group seems to be larger, containing two different Lorentz sectors. In fact, the vector supersymmetry transformations mix the two kinds of Lorentz indices, breaking this apparent larger symmetry. In general, the representations of even superspin contain only fields of half-integer spin, while the representations of odd superspin contain only fields of integer spin. All representations contain an equal number of physical and ghost degrees of freedom and the vector supersymmetry mixes these two sectors. Therefore, while vector supersymmetry has a very similar algebraic structure compared to ordinary supersymmetry, it has totally different physics and possible applications. It does not unify fields of half-integer spin with fields of integer spin, but physical fields with ghosts instead. Due to this fact, we mainly think of its role as a technical one, for instance in the context of the renormalization of gauge theories, since its presence will likely lead to cancellation of some divergences. However, ghost fields play a leading role in the quantization of string theory and in that context there is also an interesting interplay between spacetime and worldsheet supersymmetry in the RNS formulation. Therefore, vector supersymmetry could be expected to arise in some string models. Finally, as already mentioned in the introduction, vector supersymmetry serves as a good comparison to understand what the essential ingredients in supersymmetry are. The work presented in this paper is of course only the very first step in the direction of constructing and studying interacting theories with underlying VSUSY. The representations we have constructed here are possibly the simplest ones and a more general approach allowing for a classification of VSUSY representations could also be interesting work for the future. \[ss:conclusions\] Acknowledgements {#acknowledgements .unnumbered} ================ We would like to thank M. Caldarelli, J. Gomis, W. Troost and A. Van Proeyen for useful discussions. L.T. would like to thank the Galileo Galilei Institute for Theoretical Physics for the hospitality offered during the workshop “New Perspectives in String Theory" and the INFN for partial support during the completion of this work. $Y=0$ multiplet from the spinning particle multiplet ==================================================== In [@Casalbuoni:2008iy], rigid VSUSY is used to construct the action of the massive spinning particle with the method of nonlinear realizations. A quantization procedure respecting VSUSY shows that the degrees of freedom of the system, two four-dimensional Dirac spinors $\Psi_1$ and $\Psi_2$, satisfy two decoupled Dirac equation with the same mass. The VSUSY transformations, given in Table \[tablegrass\], mix the components of the two spinors.\ $Q_5$ $Q_\mu$ ------------------- ------------------------------------------- ----------------------------------------------------------------------------------------------------------- \[-2ex\] $\Psi_1$ $-\sqrt{\frac{\hat z}{2}}\gamma_5 \Psi_2$ $-\frac{1}{\sqrt{2\hat z}}\gamma_5\Big((m\gamma_{\mu}-\partial_{\mu})\Psi_1 +i\partial_{\mu}\Psi_2\Big)$ $\Psi_2$ $-\sqrt{\frac{\hat z}{2}}\gamma_5 \Psi_1$ $-\frac{1}{\sqrt{2\hat z}}\gamma_5\Big(i\partial_{\mu}\Psi_1 + (m\gamma_{\mu}+\partial_{\mu})\Psi_2\Big)$ : Action of the VSUSY charges on the spinning particle multiplet introduced in [@Casalbuoni:2008iy]. \[tablegrass\] The VSUSY invariant action for this multiplet is $$S=\int d^4x \Big(\bar\Psi_2 \gamma_5 ({\not\hspace{-.05cm}\partial}- m)\Psi_1 + \bar \Psi_1 \gamma_5 ({\not\hspace{-.05cm}\partial}+m)\Psi_2\Big)\,. \label{actiongrass}$$ Inspection of the transformations in Table \[tablegrass\], together with the fact that $Q_{\mu}$ and $Q_5$ are anticommuting generators in the VSUSY algebra and are therefore expected to be odd, leads to the conclusion that the two spinors $\Psi_1$ and $\Psi_2$ cannot have a definite Grassmann parity and should be further decomposed in an even and an odd part as follows $$\begin{aligned} &\Psi_1 = \psi_1+ \tilde\psi_1\,;\cr &\Psi_2 = \psi_2+ \tilde\psi_2\,, \label{decomp}\end{aligned}$$ where as usual $\psi_1$ and $\psi_2$ are even and $\tilde\psi_1$ and $\tilde\psi_2$ are odd.\ By using (\[decomp\]) in Table \[tablegrass\] and then decomposing the transformations in their even and odd parts one obtains the VSUSY transformations of the multiplet previously given in Table \[tablespinorsold\]. The action (\[actiongrass\]) has also no definite Grassmann parity and should be decomposed into its even and odd parts $$S=S_{\rm even}+ S_{\rm odd}\,,$$ where $$S_{\rm even}=\int d^4x \left[\bar\psi_2 \gamma_5 ({\not\hspace{-.05cm}\partial}- m)\psi_1 + \bar{\tilde\psi_2} \gamma_5 ({\not\hspace{-.05cm}\partial}- m)\tilde\psi_1+ \bar \psi_1 \gamma_5 ({\not\hspace{-.05cm}\partial}+m)\psi_2+ \bar {\tilde\psi_1} \gamma_5 ({\not\hspace{-.05cm}\partial}+m)\tilde\psi_2\right]$$ and $$S_{\rm odd}=\int d^4x \left[\bar{\tilde\psi_2} \gamma_5 ({\not\hspace{-.05cm}\partial}- m)\psi_1 + \bar{\psi_2} \gamma_5 ({\not\hspace{-.05cm}\partial}- m)\tilde\psi_1+ \bar {\tilde \psi_1}\gamma_5 ({\not\hspace{-.05cm}\partial}+m)\psi_2+ \bar {\psi_1} \gamma_5 ({\not\hspace{-.05cm}\partial}+ m)\tilde\psi_2\right]\,.$$ One can check that both parts are separately invariant under the transformations in Table \[tablegrass\] and generate all the equations of motion of the multiplet. It is then natural to choose the even part as the action for the field theoretical model. One can easily check that $S_{\rm even}$ is just the rescaling (\[rescal\]) of action (\[actionstandard\]). Solution of the superspace constraint equation ============================================== In this Appendix we would like to sketch how to solve constraint (\[first\]) introduced in Section 3.3 to obtain the general solution (\[exp\]).\ Let us expand the spinorial superfield $\tilde\Lambda$ introduced in (\[first\]) as follows: $$\begin{aligned} \tilde\Lambda(x^{\mu},\theta^{\mu},\theta_{5},z)&=&\,\,\,\,\tilde\Delta(x,z)+ \Delta(x,z)\theta_{5}+ \Delta_{\mu}(x,z)\theta^{\mu}+ \tilde\Delta_{\mu}(x,z)\theta^{\mu}\theta_{5}\cr &&+\tilde\Delta_{\mu\nu}(x,z)\theta^{\mu}\theta^{\nu}+ \Delta_{\mu\nu}(x,z)\theta^{\mu}\theta^{\nu}\theta_{5} +\Delta_{\mu\nu\rho}(x,z)\theta^{\mu}\theta^{\nu} \theta^{\rho}\cr &&+\tilde\Delta_{\mu\nu\rho}(x,z) \theta^{\mu}\theta^{\nu}\theta^{\rho}\theta_{5} +\tilde\Delta_{\mu\nu\rho\sigma}(x,z) \theta^{\mu}\theta^{\nu}\theta^{\rho}\theta^{\sigma}+ \Delta_{\mu\nu\rho\sigma}(x,z) \theta^{\mu}\theta^{\nu}\theta^{\rho}\theta^{\sigma}\theta_{5}\,.\cr && \label{lambdasuperfield}\end{aligned}$$ Constraint (\[first\]), expanded order by order in the odd cordinates $\theta^{\mu}$ and $\theta_5$, gives the following set of equations: $$\begin{aligned} & -\tilde\psi_{\mu}=\gamma_{\mu}\tilde\Delta\,;\quad &\psi_{\mu}+\frac{i}{2} \partial_{\mu}\phi=\gamma_{\mu}\Delta\,; \nonumber\\ & 2\xi_{{\mu}{\nu}} -\frac{i}{2}\eta_{{\mu}{\nu}} \partial_z \phi =\gamma_{\mu}\Delta_{\nu}\,;\quad &-2\tilde\xi_{{\mu}{\nu}}+\frac{i}{2}\partial_{\mu}\tilde\psi_{\nu}+\frac{i}{2}\eta_{{\mu}{\nu}}\partial_z \tilde\phi=\gamma_{\mu}\tilde\Delta_{\nu}\,;\nonumber\\ &-3\tilde\chi_{{\mu}{\nu}\rho}+\frac{i}{2}\eta_{{\mu}{\nu}}\partial_z \tilde\psi_\rho=\gamma_{\mu}\tilde\Delta_{\nu\rho}\,;\quad &3\chi_{{\mu}{\nu}\rho}+\frac{i}{2}\partial_{\mu}\xi_{{\nu}\rho}-\frac{i}{2}\eta_{{\mu}{\nu}}\partial_z \psi_\rho=\gamma_{\mu}\Delta_{\nu\rho}\,;\nonumber\\ &4\zeta_{{\mu}{\nu}\rho{\sigma}}-\frac{i}{2} \eta_{{\mu}{\nu}}\partial_z \xi_{\rho{\sigma}}=\gamma_{\mu}\Delta_{\nu\rho\sigma}\,;\quad &-4\tilde\zeta_{{\mu}{\nu}\rho{\sigma}}+\frac{i}{2}\partial_{\mu}\tilde\zeta_{{\nu}\rho{\sigma}}+\frac{i}{2}\eta_{{\mu}{\nu}}\partial_z\tilde{\rho{\sigma}}=\gamma_{\mu}\tilde\Delta_{\nu\rho\sigma}\nonumber\\ &\frac{i}{2}\eta_{{\mu}{\nu}}\partial_z \tilde\chi_{\rho{\sigma}\tau}=\gamma_{\mu}\tilde\Delta_{\nu\rho\sigma\tau}\,;\quad &-\frac{i}{2}\eta_{{\mu}{\nu}}\partial_z \chi_{\rho{\sigma}\tau}+\frac{i}{2}\partial_{\mu}\zeta_{{\nu}\rho{\sigma}\tau}=\gamma_{\mu}\Delta_{\nu\rho\sigma\tau}\,. \label{expandedcons}\end{aligned}$$ Let us start by considering all the terms of order $0$ in $\theta_5$, given in the left column above. At order $0$ in $\theta^{\mu}$, we have the first equation in the left column in (\[expandedcons\]), which is solved by requiring $\tilde\psi_{\mu}$ to be of the form $$\tilde\psi_{\mu}^{\alpha}=(\gamma_{\mu})^{\alpha}_{~{\beta}} \tilde\lambda^{\beta}\,, \label{eqpsi}$$ with $\tilde\lambda^{\alpha}(x,z)$ a generic spinor.\ At order $1$ in $\theta^{\mu}$, we have the second equation in the left column in (\[expandedcons\]). To possibly solve this equation, $\xi_{{\mu}{\nu}}$ must contain at least one gamma matrix. Therefore we take the following general ansatz $$\xi_{{\mu}{\nu}}(z)=\gamma_{[{\mu}} \omega_{{\nu}]}\,, \label{ansatz}$$ where $\omega^{\alpha}_{{\nu}}$ is a Rarita-Schwinger field to be determined.\ By plugging in this ansatz we obtain that the following equality must be valid $$\gamma_{\nu}\omega_{\mu}= -\frac{i}{2}\eta_{{\mu}{\nu}}\partial_z \phi + \gamma_{\mu}\delta_{\nu}$$ for some Rarita-Schwinger field $\delta^{\alpha}_{\nu}$. Clearly, this happens for $$\omega_{\mu}= -\frac{i}{4}\gamma_{\mu}\partial_z \phi\,.$$ Note that a term in $\xi_{{\mu}{\nu}}$ of the form $\gamma_{[{\mu}} \partial_{{\nu}]} \hat\omega(z)$ would be allowed by symmetry considerations but in general does not solve the equation, apart from the case when it reduces to the structure we have already considered before.\ To summarize, the solution of the constraint at order $1$ in $\theta^{\mu}$ leads to the following relation between the components $\xi_{{\mu}{\nu}}^{\alpha}$ and $\phi^{\alpha}$: $$\xi_{{\mu}{\nu}}=-\frac{i}{4}\gamma_{{\mu}{\nu}}\partial_z \phi\,. \label{theta1}$$ A similar procedure works up to order $3$ in $\theta^{\mu}$.\ At order $4$, we see from the last equation in the first column in (\[expandedcons\]) that only one term is present in the LHS. However, by using the gamma matrix identity $$\gamma_{\mu\nu\rho}=-i\epsilon_{\mu\nu\rho\sigma}\gamma^{\sigma}\gamma_{5}\,. \label{gamma5stuff}$$ and the Schouten identity, one can rewrite the order 4 term as follows $$\eta_{{\mu}{\nu}}\gamma_{\rho{\sigma}\delta}~\partial_z^2 \tilde\lambda~\sim~ \gamma_{\mu}\gamma_5{\epsilon}_{{\nu}\rho{\sigma}\delta}~\partial_z^2 \tilde \lambda \label{correctform}$$ It is then clear that the order 4 term directly satisfies the constraint equation. Let us now consider the terms of order $1$ in $\theta_5$.\ The difference with respect to the previous case is that the term $\frac{i}{2}\theta_{5}\partial_\mu$ in the covariant derivative $D_{\mu}$ now plays a role. However, one could in principle decide to work in the VSUSY analogue of the chiral representation for the superspace covariant derivatives given in (\[newcovderivII\]). In that case the $\theta_5$ term in $D_{\mu}$ would simply not be present. Therefore, in that basis the solution of the constraint equations coming from the terms of order $1$ in $\theta_5$ is completely analogue to the case of order $0$ in $\theta_5$. It is then clear that the solution rewritten in the more symmetric representation for the covariant derivatives (\[covdevII\]) will just contain an extra correction term coming from the corresponding change of basis (\[covdertrans\]). Explicitly, the analogues of eqs. (\[eqpsi\]) and (\[theta1\]) become $$\begin{aligned} &\psi_{\mu}=\gamma_{\mu}\lambda -\frac{i}{2}\partial_{\mu}\phi\,,\cr &\tilde\xi_{{\mu}{\nu}}=-\frac{i}{4}\gamma_{{\mu}{\nu}}\partial_z \tilde\phi -\frac{i}{2}\gamma_{[{\mu}}\partial_{{\nu}]} \tilde\lambda\,. \end{aligned}$$ The correction is of course the second term in the RHS, which can be easily computed by hand. Note that it also fits the general prescription given in (\[forms\]).\ The result is the general solution of constraint (\[first\]) given in (\[exp\]). $Y=1/2$ multiplet from the Kato-Miyake TSUSY multiplet ====================================================== The multiplet in Table \[tablevectormultipletI\] is a consistent truncation of the TSUSY multiplet found by Kato and Miyake in Euclidean spacetime in [@Kato:2005gb]. The Kato-Miyake multiplet contains three vectors $\tilde F^{\mu}$,$V_{\mu}$ and $K^{\mu}$, one odd and two even, an odd scalar $\tilde A$ and an odd selfdual two-form $\tilde M^-_{{\mu}{\nu}}$. The action of TSUSY on these fields is given in Table \[tablekatomiyake\]. $Q_5$ $Q_\mu$ $Q^+_{\mu\nu}$ ----------------------------- ------------------------------------------------------------------------- ------------------------------------------------------------------------------------------------------------------------------------------ -------------------------------------------------------------------------------------------------------------------------------------- $V_\alpha$ $\tilde{F}_\alpha$ $\delta_{\alpha\mu}\tilde{A}-\tilde{M}^-_{\mu\alpha}$ $-\delta^+_{{\mu}{\nu},{\alpha}\rho}\tilde F^\rho$ $\tilde{F}_\alpha$ $\frac{1}{2}K_\alpha$ $\frac{i}{2}(\delta^{-}_{\mu\alpha,\rho\sigma}\partial^\rho V^\sigma +\delta_{\alpha\mu}\partial^\rho V_\rho - 2\partial_\mu V_\alpha) $ $\frac{1}{2}\delta^+_{\mu\nu,\alpha\rho} K^\rho $ $\tilde{A}$ $-\frac{i}{2}\partial^\mu V_\mu$ $\frac{1}{2}K_\mu$ $-\frac{i}{2}\delta^+_{\mu\nu,\rho\sigma}\partial^\rho V^\sigma $ $\tilde{M}^-_{\alpha\beta}$ $\frac{i}{2}\delta^{-}_{\alpha\beta,\rho\sigma}\partial^\rho V^\sigma $ $-\frac{1}{2}\delta^{-}_{\alpha\beta,\mu\rho}K^\rho$ $-\frac{i}{2}\delta^+_{\mu\nu,\rho\sigma}\delta_{\alpha\beta,\tau\phantom{\sigma}}^{-\phantom{\mu\nu,}\sigma} \partial^\rho V^\tau $ $K_\alpha$ $-i(\partial_\alpha\tilde{A}+\partial^\rho\tilde{M}^-_{\alpha\rho})$ $-i(\delta^{-}_{\mu\alpha,\rho\sigma}\partial^\rho\tilde{F}^\sigma +\delta_{\alpha\mu}\partial^\rho \tilde{F}_\rho)$ $i\delta^+_{\mu\nu,\alpha\rho}(\partial^\rho \tilde{A}+\partial^\sigma \tilde{M}^{-\rho}_{~~~\sigma} )$ Z ----------------------------- ------------------------------------------------------------------------ $V_\alpha$ $K_{\alpha}$ $\tilde{F}_\alpha$ $-i(\partial_\alpha \tilde{A}+\partial^\rho \tilde{M}^-_{\alpha\rho})$ $\tilde{A}$ $-i\partial^\rho\tilde{F}_\rho$ $\tilde{M}^-_{\alpha\beta}$ $i\delta^{-}_{\alpha\beta,\rho\sigma}\partial^\rho \tilde{F}^\sigma$ $K_\alpha$ $-\partial^2 V_\alpha$ : Action of the TSUSY charges on the Kato-Miyake multiplet [@Kato:2005gb].\[tablekatomiyake\] In this multiplet $P^2=Z^2$ (note the different convention with respect to this paper) is satisfied on all fields, but no field equations have to be used to close the algebra, the multiplet closes off-shell. Moreover, the odd Casimir is present with value zero.\ We can shorten the multiplet by imposing that the operator $Z$ acts as a number $\hat z$ on all fields. As a result we obtains the set of constraints: $$\begin{aligned} && K_{\mu}= \hat z V_{\mu}\,;\cr && -i(\partial_{\mu}\tilde A + \partial^\rho \tilde M^-_{{\mu}\rho})= \hat z \tilde F_{\mu}\,;\cr && -i \partial^\rho \tilde F_\rho =\hat z \tilde A\,;\cr && i \delta^-_{{\mu}{\nu},\rho{\sigma}}\partial^\rho \tilde F^{\sigma}= \hat z \tilde M^-_{{\mu}{\nu}}\,;\cr && -\Box V_{\mu}= \hat z K_{\mu}\,. \label{setcons}\end{aligned}$$ The first two equations above tell us that the vectors $K_{\mu}$ and $\tilde F_{\mu}$ can be expressed in terms of $\tilde A$, $V_{\mu}$ and $\tilde{M}^-_{{\mu}{\nu}}$. The remaining three equations imply that $\tilde A$, $V_{\mu}$ and $\tilde{M}^-_{{\mu}{\nu}}$ must satisfy the following dynamical equations with mass: $$\begin{aligned} &&\Box V_{\mu}= -\hat z^2 V^{\mu}\,;\cr &&\Box \tilde A = - \hat z^2 \tilde A \,;\cr &&\Box \tilde M^-_{{\mu}{\nu}}= - \hat z^2 \tilde M^-_{{\mu}{\nu}}\,. \label{eqns}\end{aligned}$$ By implementing the first two constraints of (\[setcons\]) in Table \[tablekatomiyake\], one obtains the corresponding table for the truncated multiplet $(\tilde A,V_{\mu}, \tilde{M}^-_{{\mu}{\nu}})$. In order to have a VSUSY sector that is exactly identical (up to signs due to the different conventions) to our VSUSY multiplet given in Table \[tablevectormultipletI\], we rescale the fields as follows $$\tilde a=i\tilde A, ~~~~~~~b_{\mu}=-\frac{i}{2} \hat z V_{\mu}, ~~~~~~~\tilde{c}^-_{{\mu}{\nu}}= -i\frac{\hat z}{4}\tilde M^-_{{\mu}{\nu}}\,$$ and we switch the roles of ghosts and physical fields.\ The result is our VSUSY multiplet of Table \[tablevectormultipletI\], with the extra tensor transformations $Q^+_{\mu{\nu}}$ --------------------------- ---------------------------------------------------------------------------------------------------------------------------- $a$ $\frac{i}{\hat z}\delta^+_{{\mu}{\nu},\rho{\sigma}}\partial^\rho \tilde b^\sigma$ $\tilde b_{{\alpha}}$ $-\frac{i}{2}\delta^+_{{\mu}{\nu},{\alpha}\rho}(-\partial^\rho a + \frac{4}{\hat z}\partial^\gamma c^{-\rho}_{~~~\gamma}$) $c_{{\alpha}{\beta}}^{-}$ $-\frac{i}{4} \delta^{+}_{{\mu}{\nu},\rho{\sigma}}\delta^{-~~~{\sigma}}_{{\alpha}{\beta},\tau}\partial^\rho \tilde b^\tau$ : Action of the antiselfdual tensor charge on the truncated Kato-Miyake multiplet The dynamics of the Kato-Miyake multiplet is given by the action $$S=\int d^4 x \left(V^{\mu}\Box V_{\mu}+ 4i \tilde F^{\mu}(\partial_{\mu}\tilde A + \partial^{\nu}\tilde M^-_{{\mu}{\nu}}) + K^{\mu}K_{\mu}\right)\,. \label {KMaction}$$ The field $K^{\mu}$ is auxiliary and all fields satisfy equations of motion with zero mass. Note that the scalar and two-form fields appear in the action with an off-diagonal first order term. The fields remaining after our truncation satisfy dynamical equations with mass instead, related to the value of the central charge by a BPS-like constraint. The truncation procedure leads to constraints directly imposing the equations of motions on the fields, so it is not possible to apply it to action (\[KMaction\]) to obtain an action for the truncated multiplet. We have checked whether it was possible to build an action for the truncated multiplet leading to the equations of motion (\[eqns\]), but we find no suitable VSUSY invariant quadratic structure. On the other hand, one could think of an off-diagonal first order structure as in (\[KMaction\]), in this case necessarily odd. However, this would not lead to the correct dynamical equations and in any case direct inspection shows that a combination of this kind of terms cannot be rendered invariant under the odd vectorial VSUSY charge. Our conclusion is that the $Y=\frac{1}{2}$ multiplet satisfies dynamical equations of motion with mass that cannot be derived by an action. [10]{} A. Barducci, R. Casalbuoni and L. Lusanna, *[Supersymmetries and the pseudoclassical relativistic electron]{}*, Nuovo Cim. [**A35**]{} (1976) [377](http://dx.doi.org/10.1007/BF02730291) R. Casalbuoni, F. Elmetti, J. Gomis, K. Kamimura, L. Tamassia and A. Van Proeyen, *[Vector Supersymmetry: Casimir operators and contraction from $OSp(3,2|2)$]{}*, JHEP [**0901**]{} (2009) 035 [[arXiv:0812.1982 \[hep-th\]]{}]{}. R. Casalbuoni, F. Elmetti, J. Gomis, K. Kamimura, L. Tamassia and A. Van Proeyen, *[Vector Supersymmetry from $OSp(3,2|2)$: Casimir Operators]{}*, Fortsch. Phys.  [**57**]{} (2009) 521 [[arXiv:0901.4862 \[hep-th\]]{}]{}. E. Witten, *[Topological quantum field theory]{}*, Commun. Math. Phys. [ **117**]{} (1988) [353](http://dx.doi.org/10.1007/BF01223371) M. Alvarez and J. M. F. Labastida, *[Topological matter in four-dimensions]{}*, Nucl. Phys. [**B437**]{} (1995) [356–390](http://dx.doi.org/10.1016/0550-3213(94)00512-D), [[arXiv:hep-th/9404115]{}](http://arxiv.org/abs/hep-th/9404115) J. Kato, N. Kawamoto and A. Miyake, *[N = 4 twisted superspace from Dirac-Kaehler twist and off- shell SUSY invariant actions in four dimensions]{}*, Nucl. Phys. [**B721**]{} (2005) [229–286](http://dx.doi.org/10.1016/j.nuclphysb.2005.05.024), [[arXiv:hep-th/0502119]{}](http://arxiv.org/abs/hep-th/0502119) D. Birmingham, M. Rakowski and G. Thompson, *[Topological field theories, Nicolai maps and BRST quantization]{}*, Phys. Lett. [**B214**]{} (1988) [381](http://dx.doi.org/10.1016/0370-2693(88)91381-0) F. Delduc, F. Gieres and S. P. Sorella, *[Supersymmetry of the d = 3 Chern-Simons action in the Landau gauge]{}*, Phys. Lett. [**B225**]{} (1989) [367](http://dx.doi.org/10.1016/0370-2693(89)90584-4) L. Baulieu, G. Bossard and A. Martin, *[Twisted superspace]{}*, Phys. Lett. [**B663**]{} (2008) [275–280](http://dx.doi.org/10.1016/j.physletb.2008.03.054), [[arXiv:0802.1980 \[hep-th\]]{}](http://arxiv.org/abs/0802.1980) R. Casalbuoni, J. Gomis, K. Kamimura and G. Longhi, *[Space-time vector supersymmetry and massive spinning particle]{}*, JHEP [**02**]{} (2008) [094](http://dx.doi.org/10.1088/1126-6708/2008/02/094), [[arXiv:0801.2702 \[hep-th\]]{}](http://arxiv.org/abs/0801.2702) J. Kato and A. Miyake, *[Topological hypermultiplet on N = 2 twisted superspace in four dimensions]{}*, Mod. Phys. Lett.  A [**21**]{}, 2569 (2006) \[arXiv:hep-th/0512269\]. J. Kato and A. Miyake, *[Vafa-Witten theory on N=2 and N=4 twisted superspace in four dimensions]{}*, JHEP [**0903**]{}, 087 (2009) \[arXiv:0808.2538 \[hep-th\]\]. M. F. Sohnius, *[Supersymmetry and central charges]{}*, Nucl. Phys.  B [**138**]{} (1978) 109. V. A. Soroka, *[On action with Grassmann-odd Lagrangian]{}*, Phys. Atom. Nucl.  [**59**]{}, 1270 (1996) \[Yad. Fiz.  [**59**]{}, 1327 (1996)\] \[arXiv:hep-th/9507030\]. D. V. Soroka, V. A. Soroka and J. Wess, *[Supersymmetric D = 1, N=1 model with Grassmann-odd Lagrangian]{}*, Phys. Lett.  B [**512**]{}, 197 (2001). S. Knapen, *[Field representations of vector supersymmetry]{}*, master thesis, K.U.Leuven, 2008-2009. A. Van Proeyen, *[Tools for supersymmetry]{}*, arXiv:hep-th/9910030. [^1]: This is the same kind of restriction as choosing to have a real central charge instead of a complex one in $\mathcal{N}=2$ standard SUSY. This choice is made for instance in the $\mathcal{N}=2$ SUSY and related twisted SUSY papers we are inspired by in this paper, like [@Kato:2005gb] [@Kato:2008dw] and [@Sohnius:1978fw]. [^2]: Due to the fact that in VSUSY superspace both the even coordinate $x^{\mu}$ and the odd coordinate $\theta^{\mu}$ carry a vectorial index, the notation for the covariant derivatives could lead to some misunderstanding. We choose to denote with $D_{\mu}$ the odd vectorial covariant derivative and simply with $\frac{\partial}{\partial x^{\mu}}\equiv\partial_{\mu}$ the even one.

ArXiv

--- abstract: 'We discuss two lagrangian interacting dark energy models in the context of the holographic principle. The potentials of the interacting fields are constructed. The models are compared with CMB distance information, baryonic acoustic oscilations, lookback time and the Constitution supernovae sample. For both models the results are consistent with a non vanishing interaction in the dark sector of the universe - with more than three standard deviations of confidence for one of them. Moreover, in both cases, the sign of coupling is consistent with dark energy decaying into dark matter, alleviating the coincidence problem.' author: - 'Sandro M. R. Micheletti [^1]' title: Holographic field theory models of dark energy in interaction with dark matter --- Introduction ============ In the last years, there have been several papers where an interaction in the dark sector of the universe is considered [@1] - [@sandro3]. A motivation to considering the interaction is that dark energy and dark matter will evolve coupled to each other, alleviating the coincidence problem [@1]. A further motivation is that, assuming dark energy to be a field, it would be more natural that it couples with the remaining fields of the theory, in particular with dark matter, as it is quite a general fact that different fields generally couple. In other words, it is reasonable to assume that there is no symmetry preventing such a coupling between dark energy and dark matter fields. Using a combination of several observational datasets, as supernovae data, CMB shift parameter, BAO, etc., it has been found that the coupling constant is small but non vanishing within at least $1\sigma$ confidence level [@1] - [@sandro2]. In two recent works, the effect of an interaction between dark energy and dark matter on the dynamics of galaxy clusters was investigated through the Layser-Irvine equation, the relativistic equivalent of virial theorem [@peebles]. Using galaxy cluster data, it has been shown that a non vanishing interaction is preferred to describe the data within several standard deviations [@virial]. However, in most of these papers, the interaction term in the equation of motion is derived from phenomenological arguments. It is interesting to obtain the interaction term from a field theory. Some works have already taken a step in such a direction [@sandro2], [@sandro3], [@bean]. On the other hand, scalar fields have been largely used as candidates to dark energy. They naturally arise in particle physics and string theory. Good reviews on this subject can be found in [@copeland]. A motivation to use scalar fields as candidates to dark energy is that their pressure can be negative, making possible to reproduce the recent period of accelerated expansion of the universe. For example, for canonical scalar field, the equation of state parameter varies between $1$ and $-1$. Dark energy modeled as a canonical scalar field is called quintessence and was investigated, for example, in [@bean], [@canonico]. For tachyon scalar field, the equation of state is always negative. The tachyon field has been studied in recent years in the context of string theory, as a low energy effective theory of D-branes and open strings [@sen]. Tachyon field as dark energy was studied, for example, in [@sandro2], [@taquions], [@taqholo]. The first question about scalar fields concerns the choice of the potential. Common choices are the power law and the exponential potentials. However, these choices are in fact arbitrary. In principle, any other form for the potential which leads to recent accelerated expansion would be acceptable. On the other hand, it is possible that a complete understanding of the nature of dark energy will only be possible within a quantum gravity theory context. Although results for quantum gravity are still missing, or at least premature, it is possible to introduce, phenomenologically, some of its principles in a model of dark energy. Recently combinations of quintessence, quintom and tachyon models with holographic dark energy had been proposed - in [@escalarholo], [@quintomholo] and [@taqholo], respectively. Specifically, by imposing that the energy density of the scalar field must match the holographic dark energy density, namely $\rho_{\Lambda}=3c^{2}M_{Pl}^{2}L^{-2}$, where $c$ is a numerical constant and $L$ is the infrared cutoff, it was demonstrated that the equation of motion of fields for the non-interacting case reproduces the equation of motion for holographic dark energy. In fact, to impose that the energy density of scalar field must match the holographic dark energy density corresponds to specify its potential**.** This can be seen as a physical criterion to choose the potential. Here, we generalize this idea for two kinds of interacting scalar fields. The models ========== We consider the general action$$S=\int d^{4}x\sqrt{-g}\left\{ \frac{M_{Pl}^{2}}{2}R+\mathcal{L}_{\varphi }\left( x\right) +\frac{i}{2}[\bar{\Psi}\gamma^{\mu}\nabla_{\mu}\Psi -\bar{\Psi}\overleftarrow{\nabla}_{\mu}\gamma^{\mu}\Psi]-(M-\beta\varphi )\bar{\Psi}\Psi+\sum\limits_{j}\mathcal{L}_{j}\left( x\right) \right\} \label{action}$$ where $M_{Pl}\equiv\left( 8\pi G\right) ^{-1/2}$ is the reduced Planck mass, $R$ is the curvature scalar, $\mathcal{L}_{\varphi}\left( x\right) $ is, unless of the coupling term, the lagrangian density for the scalar field, which we will identify with dark energy, $\Psi$ is a massive fermionic field, which we will identify with dark matter, $\beta$ is the dimensionless coupling constant and $\sum\limits_{j}\mathcal{L}_{j}\left( x\right) $ contains the lagrangian densities for the ramaining fields. Notice that, in this work, we will only consider an interaction of dark energy with dark matter. If there was a coupling between the scalar field and baryonic matter, the corresponding coupling constant $\beta_{b}$ should satisfy the solar system constraint [@solarsystem]$$\beta_{b}\lesssim10^{-2}\text{ .} \label{ss}$$ We assume $\beta_{b}\equiv0$, which trivially satisfy the constraint (\[ss\]). We consider two kinds of scalar fields: the canonical scalar field, or quintessence field, for which$$\mathcal{L}_{\varphi}\left( x\right) =\frac{1}{2}\partial_{\mu}\varphi\partial^{\mu}\varphi-V(\varphi)\text{ ,} \label{canonical}$$ and the tachyon scalar field, for which$$\mathcal{L}_{\varphi}\left( x\right) =-V(\varphi)\sqrt{1-\alpha\partial ^{\mu}\varphi\partial_{\mu}\varphi}\text{ ,} \label{tachyonic}$$ where $\alpha$ is a constant with dimension $MeV^{-4}$. Notice that in both cases, we assume a Yukawa coupling with the dark matter field $\Psi$. Quintessence field ------------------ For the quintessence field, $\mathcal{L}_{\varphi}\left( x\right) $ in the action (\[action\]) is given by (\[canonical\]). From a variational principle, we obtain $$i\gamma^{\mu}\nabla_{\mu}\Psi-M^{\ast}\Psi=0\text{ ,} \label{dirac}$$$$i(\nabla_{\mu}\bar{\Psi})\gamma^{\mu}+M^{\ast}\bar{\Psi}=0\text{ ,} \label{diracadj}$$ where $M^{\ast}\equiv M-\beta\varphi$, and$$\nabla_{\mu}\partial^{\mu}\varphi+\frac{dV(\varphi)}{d\varphi}=\beta\bar{\Psi }\Psi\text{ .} \label{eqmov_scalar}$$ Eqs. (\[dirac\]) and (\[diracadj\]) are, respectively, the covariant Dirac equation and its adjoint, in the case of a non vanishing interaction between the Dirac field and the scalar field $\varphi$. For homogeneous fields and adopting the Friedmann-Robertson-Walker metric, $g_{\mu\nu}$=diag$\left( 1,-a^{2}\left( t\right) ,-a^{2}\left( t\right) ,-a^{2}\left( t\right) \right) $, where $a^{2}\left( t\right) $ is the scale factor, eqs. (\[dirac\]) and (\[diracadj\]) lead to$$\frac{d(a^{3}\bar{\Psi}\Psi)}{dt}=0$$ which is equivalent to$$\bar{\Psi}\Psi=\bar{\Psi}_{0}\Psi_{0}\left( \frac{a_{0}}{a}\right) ^{3} \label{conser_psibarpsi}$$ and (\[eqmov\_scalar\]) reduces to$$\ddot{\varphi}+3H\dot{\varphi}+\frac{dV(\varphi)}{d\varphi}=\beta\bar{\Psi }\Psi\text{ ,} \label{homoscalar}$$ where $H\equiv\frac{\dot{a}}{a}$ is the Hubble parameter. From the energy-momentum tensor, we get$$\begin{aligned} \rho_{\varphi} & =\frac{1}{2}\dot{\varphi}^{2}+V(\varphi)\text{ ,}\label{rofiscalar}\\ P_{\varphi} & =\frac{1}{2}\dot{\varphi}^{2}-V(\varphi)\text{ ,}\label{pfiscalar}\\ \rho_{\Psi} & =M^{\ast}\bar{\Psi}\Psi\text{ ,}\label{ropsi}\\ P_{\Psi} & =0\text{ .}\nonumber\end{aligned}$$ From (\[rofiscalar\]) and (\[pfiscalar\]) we have $\omega_{\varphi}\equiv\frac{P_{\varphi}}{\rho_{\varphi}}=\frac{\frac{1}{2}\dot{\varphi}^{2}-V(\varphi)}{\frac{1}{2}\dot{\varphi}^{2}+V(\varphi)}$. Deriving (\[rofiscalar\]) and (\[ropsi\]) with respect to time and using (\[conser\_psibarpsi\]) and (\[homoscalar\]), we obtain$$\dot{\rho}_{\varphi}+3H\rho_{\varphi}(\omega_{\varphi}+1)=\beta\dot{\varphi }\bar{\Psi}_{0}\Psi_{0}\left( \frac{a_{0}}{a}\right) ^{3} \label{conser_rofi}$$ and$$\dot{\rho}_{\Psi}+3H\rho_{\Psi}=-\beta\dot{\varphi}\bar{\Psi}_{0}\Psi _{0}\left( \frac{a_{0}}{a}\right) ^{3}\text{ ,} \label{conser_ropsi}$$ where the dot represents derivative with respect to time. For baryonic matter and radiation, we have respectively$$\dot{\rho}_{b}+3H\rho_{b}=0 \label{conserbaryon}$$ and$$\dot{\rho}_{r}+3H\rho_{r}(\omega_{r}+1)=0\text{ ,} \label{conserrad}$$ where $\omega_{r}=\frac{1}{3}$. Eqs. (\[conserbaryon\]) and (\[conserrad\]) implies that $\rho_{b}=\frac{\rho_{b0}}{a^{3}}$ and $\rho_{r}=\frac {\rho_{r0}}{a^{4}}$, respectively. The subscript $0$ denotes the quantities today. We are considering the radiation as composed by photons and massless neutrinos, so that $\rho_{r0}=\left( 1+0.2271N_{eff}\right) \rho_{\gamma0}$, where $N_{eff}=3.04$ is the effective number of relativistic degrees of freedom and $\rho_{\gamma0}$ is the energy density of photons, given by **** $\rho_{\gamma0}=\frac{\pi^{2}}{15}T_{CMB}$, being $T_{CMB}=2.725K$ the CMB temperature today. The Friedmann equation for a flat universe reads$$H^{2}=\frac{1}{3M_{Pl}^{2}}\left[ M^{\ast}\bar{\Psi}_{0}\Psi_{0}\left( \frac{a_{0}}{a}\right) ^{3}+\frac{1}{2}\dot{\varphi}^{2}+V(\varphi )+\frac{\rho_{b0}}{a^{3}}+\frac{\rho_{r0}}{a^{4}}\right] \text{ .} \label{friedmann_sc}$$ In order to determine the dynamics of the interacting quintessence field, it is necessary to specify the potential $V(\varphi)$. Instead of choosing an explicit form for $V(\varphi)$, we will specify it implicitly, by imposing that the energy density of the quintessence field, given by (\[rofiscalar\]), must match the holographic dark energy density, $\rho_{\Lambda}=3c^{2}M_{Pl}^{2}L^{-2}$, where $c$ is a numerical constant and $L$ is the infrared cutoff. The evolution of the interacting quintessence field with redshift will be given by the equation of evolution for the holographic dark energy density, with a certain expression for the equation of state parameter $\omega_{\varphi}$. In fact, we will see that imposing the energy density of the quintessence field to match the holographic dark energy density leads to an expression for the potential. In [@Li] it has been argued that, in order that holographic dark energy drives the recent period of accelerated expansion, the IR cutoff $L$ must be the event horizon $R_{h}=a\left( t\right) \int_{t}^{\infty}\frac{dt^{\prime }}{a\left( t^{\prime}\right) }$. Substituting $R_{h}$ in the expression of the holographic dark energy, we get $R_{h}=\frac{c}{H\sqrt{\Omega_{\varphi}}}$, therefore,$$\int_{t}^{\infty}\frac{dt^{\prime}}{a\left( t^{\prime}\right) }=\frac {c}{a\left( t\right) H\sqrt{\Omega_{\varphi}}}\text{ .}$$ Differentiating both sides with respect to time, using the Friedmann equation (\[friedmann\_sc\]) together with conservation equations (\[conser\_rofi\]) and (\[conser\_ropsi\]), we obtain$$\frac{d\Omega_{\varphi}}{dz}=-\frac{\Omega_{\varphi}}{1+z}\left( 2\frac {\sqrt{\Omega_{\varphi}}}{c}+3\Omega_{\varphi}\omega_{\varphi}+\Omega _{r}+1\right) \text{ .} \label{eq_mov_holo}$$ Equation (\[eq\_mov\_holo\]) is just the equation of evolution for the holographic dark energy [@Li]. Using the Friedmann equation (\[friedmann\_sc\]), the conservation equations (\[conserbaryon\]) and (\[conserrad\]) can be written as$$\frac{d\Omega_{b}}{dz}=-\frac{\Omega_{b}}{1+z}\left( 3\Omega_{\varphi}\omega_{\varphi}+\Omega_{r}\right) \label{eq_movbar}$$ and$$\frac{d\Omega_{r}}{dz}=-\frac{\Omega_{r}}{1+z}\left( 3\Omega_{\varphi}\omega_{\varphi}+\Omega_{r}-1\right) \text{ .} \label{eq_movrad}$$ We define $r\equiv\frac{\rho_{\Psi}}{\rho_{\varphi}}$. Deriving $r$ with respect to time, using (\[conser\_rofi\]), (\[conser\_ropsi\]), (\[rofiscalar\]) and (\[pfiscalar\]) we obtain$$\dot{r}=3Hr\omega_{\varphi}-sign\left[ \dot{\varphi}\right] \frac {\beta\left( 1+r\right) }{\sqrt{3}M_{Pl}H}\sqrt{\frac{1+\omega_{\varphi}}{\Omega_{\varphi}}}\bar{\Psi}_{0}\Psi_{0}\left( \frac{1+z}{1+z_{0}}\right) ^{3}\text{ .} \label{rdotscalar}$$ We can rewrite $\bar{\Psi}_{0}\Psi_{0}$ in terms of observable quantities. In fact, by imposing that the dark matter density today matches the observed value, we obtain $\bar{\Psi}_{0}\Psi_{0}=\frac{3M_{Pl}^{2}H_{0}^{2}\left( 1-\Omega_{\varphi0}-\Omega_{b0}-\Omega_{r0}\right) }{M-\beta\phi_{0}}$. The sign of $\dot{\varphi}$ is arbitrary, as it can be modified by redefinitions of the field, $\varphi\rightarrow-\varphi$, and of the coupling constant, $\beta\rightarrow-\beta$. Noticing that $r=\frac{1-\Omega_{\varphi}-\Omega _{b}-\Omega_{r}}{\Omega_{\varphi}}$, we can substitute $r$ and $\dot{r}$ in (\[rdotscalar\]) by $\Omega_{\varphi}$, $\Omega_{b}$, $\Omega_{r}$, $\dot{\Omega}_{\varphi}$, $\dot{\Omega}_{b}$ and $\dot{\Omega}_{r}$. Using (\[eq\_mov\_holo\]), (\[eq\_movbar\]) and (\[eq\_movrad\]) we obtain, after some algebra$$\omega_{\varphi}\left( z\right) =-\frac{1}{3}-\frac{2\sqrt{\Omega_{\varphi}}}{3c}+\frac{\gamma\left( z\right) }{3}\left[ \gamma\left( z\right) +\sqrt{\gamma\left( z\right) ^{2}+4\left( 1-\frac{\sqrt{\Omega_{\varphi}}}{c}\right) }\right] \text{ ,} \label{wfi_final_sc}$$ where$$\gamma\left( z\right) =\frac{1}{\sqrt{2}}\delta M_{Pl}\frac{\left( 1-\Omega_{\varphi0}-\Omega_{b0}-\Omega_{r0}\right) }{E\left( z\right) ^{2}\sqrt{\Omega_{\varphi}}}\left( \frac{1+z}{1+z_{0}}\right) ^{3}\text{ ,} \label{gama_sc}$$ with$$E\left( z\right) \equiv\frac{H\left( z\right) }{H_{0}}=\sqrt{\frac{\left[ \left( 1-\delta\Delta\varphi\right) \left( 1-\Omega_{\varphi0}-\Omega _{b0}-\Omega_{r0}\right) +\Omega_{b0}\right] }{1-\Omega_{\varphi}}\left( \frac{1+z}{1+z_{0}}\right) ^{3}+\frac{\Omega_{r0}}{1-\Omega_{\varphi}}\left( \frac{1+z}{1+z_{0}}\right) ^{4}}\text{ ,} \label{E_final_sc}$$ where $\Delta\varphi\left( z\right) \equiv\varphi\left( z\right) -\varphi_{0}$ and $\delta\equiv\frac{\beta}{M-\beta\varphi_{0}}$ is an effective coupling constant. Notice that, if $\delta=0$, (\[wfi\_final\_sc\]) reproduces the equation of state parameter obtained in [@Li]. The evolution of the quintessence field is given by$$\frac{d\varphi}{dz}=-\frac{\sqrt{3}M_{Pl}\sqrt{\Omega_{\varphi}\left( z\right) \left( 1+\omega_{\varphi}\left( z\right) \right) }}{1+z} \label{eq_fi_sc}$$ From (\[eq\_mov\_holo\]), (\[eq\_movbar\]), (\[eq\_movrad\]) and (\[eq\_fi\_sc\]) we can calculate the evolution with redshift of all observables in the model. If we wish to calculate the time dependence, we need to integrate the Friedmann equation (\[friedmann\_sc\]), which can be written in the form$$\frac{dt}{dz}=-\frac{1}{H_{0}E(z)(1+z)}\text{ .}$$ From (\[rofiscalar\]), we can compute the potential $V\left( z\right) $ as$$\frac{V\left( z\right) }{\rho_{c0}}=\frac{E^{2}(z)\Omega_{\varphi}(z)}{2}\left( 1-\omega_{\varphi}(z)\right) \text{ ,} \label{potential_sc}$$ where $\rho_{c0}=3M_{Pl}^{2}H_{0}^{2}$, $E\left( z\right) $ is given by (\[E\_final\_sc\]), $\omega_{\varphi}\left( z\right) $ is given by (\[wfi\_final\_sc\]) and $\Omega_{\varphi}\left( z\right) $ is the solution of (\[eq\_mov\_holo\]). From (\[potential\_sc\]) and (\[eq\_fi\_sc\]), we can compute $V(\varphi)$. Here it is worth saying that in the holographic dark energy model, in the non interacting case - (\[wfi\_final\_sc\]) with $\delta=0$ - $\omega_{\varphi}$ can be less than $-1$. However, as already mentioned in [@escalarholo], if we wish that the holographic dark energy is the quintessence field, then because (\[eq\_fi\_sc\]), $\omega_{\varphi}$ must be more than $-1$. Nevertheless, in the interacting case considered here, due to the fact that $\omega_{\varphi}$ depends explicitly on $\varphi$, $\omega_{\varphi}$ can not be less than $-1$. On the other hand, the square root in (\[wfi\_final\_sc\]) must be real. We can verify that $\omega_{\varphi}$ is real and $\omega _{\varphi}>-1$ if (i) $\frac{\sqrt{\Omega_{\varphi0}}}{c}<1$ or (ii) $\frac{\sqrt{\Omega_{\varphi0}}}{c}>1$ and $\left\vert \delta\right\vert M_{Pl}\geq\frac{2\sqrt{2\Omega_{\varphi0}}}{1-\Omega_{\varphi0}-\Omega _{b0}-\Omega_{r0}}\sqrt{\frac{\sqrt{\Omega_{\varphi0}}}{c}-1}$. However, case (ii) is irrelevant, as it corresponds to large values of $\left\vert \delta\right\vert M_{Pl}$. For example, if $\Omega_{\varphi0}=0.7$ and $c=0.8$, we have $\left\vert \delta\right\vert M_{Pl}\gtrsim1.69$. Below, we will see that the observational data constrain $\left\vert \delta\right\vert M_{Pl}\sim10^{-1}$. In order that $\omega_{\varphi}$ be real for all future times, as $\Omega_{\varphi}\rightarrow1$, it is necessary that $c\geq1$. It is interesting to notice that the condition $\frac{\sqrt{\Omega_{\varphi0}}}{c}<1$ is precisely the same one for which the entropy of the universe increases [@Li]. As $\Omega_{\varphi}\rightarrow1$ in the future, it is necessary that $c\geq1$. Therefore, the condition for $\omega_{\varphi}$ be real is precisely the same one for the entropy to increase. So the model respects the second law of thermodynamics. In figure \[fig\_w\_escrad\] we see the evolution of the equation of state parameter $\omega_{\varphi}$ with the scale factor $a$. For the non interacting case, $\delta=0$, we have $\omega_{\varphi}\rightarrow-1/3$, as $\Omega_{\varphi}(z)\ll1$ for $z\gg1$. For $\delta<0$, $\omega_{\varphi}>-1$ in the matter era, then approaches $-1$ in the radiation era. For $\delta>0$, $\omega_{\varphi}$ will eventually turns out positive and possibly $\omega_{\varphi}\gg1$ in very early times, as in the case showed in figure \[fig\_w\_escrad\]. This behaviour is explained as follows. In the matter era, $E^{2}(z)\sim\left( 1+z\right) ^{3}$ so that $\gamma\left( z\right) \sim\frac{1}{\sqrt{\Omega_{\varphi}}}$. From (\[eq\_mov\_holo\]) $\frac{d\Omega_{\varphi}}{dz}<0$, so $\left\vert \gamma\left( z\right) \right\vert $ increases with redshift. This increasing of $\left\vert \gamma\left( z\right) \right\vert $ will continue until the radiation era, when $E^{2}(z)\sim\left( 1+z\right) ^{4}$ and $\Omega_{\varphi}(z)\sim\left( 1+z\right) ^{-2}$ so that $\left\vert \gamma\left( z\right) \right\vert \rightarrow cte$. Tipically this constant will be much more than one. Therefore, for high redshifts $\omega_{\varphi}\left( z\right) \simeq-\frac{1}{3}+\frac{\gamma}{3}\left[ \gamma+\left\vert \gamma\right\vert +\frac{2}{\left\vert \gamma\right\vert }\right] $. If $\delta<0$ then $\gamma<0$ and $\omega_{\varphi}\left( z\right) \simeq-1$ in the radiation era. If $\delta>0$ then $\gamma>0$ and $\omega_{\varphi}\left( z\right) \simeq\frac{1}{3}+\frac{2}{3}\gamma^{2}\rightarrow cte$. Notice from (\[potential\_sc\]) that if $\omega_{\varphi}\left( z\right) >1$ then $V<0$. In order to avoid it, we impose the condition $\omega_{\varphi}\leq1$ for all $z$. This condition is satisfied if $\delta M_{Pl}\lesssim\frac {\sqrt{2}\left( 1-\Omega_{\varphi0}\right) }{1-\Omega_{\varphi0}-\Omega _{b0}}\sqrt{\Omega_{\varphi}}$. As we have $\Omega_{\varphi}\ll1$ as $z$ increase, there will be an abrupt decrease in the positive tail of the probability distribuction of $\delta$, as well as in the confidence regions of $\delta$ with other parameters. ![(a) Equation of state parameter of the holographic quintessence model, for $c=0.85$ and $\delta M_{Pl}=-0.1$ (red dashed line), $\delta M_{Pl}=0$ (black solid line) and $\delta M_{Pl}=+0.1$ (green dotted line). (b) Full range of the equation of state parameter for $\delta M_{Pl}=+0.1$.[]{data-label="fig_w_escrad"}](w_escrad_a "fig:"){width="6.23cm" height="5.0cm"}  ![(a) Equation of state parameter of the holographic quintessence model, for $c=0.85$ and $\delta M_{Pl}=-0.1$ (red dashed line), $\delta M_{Pl}=0$ (black solid line) and $\delta M_{Pl}=+0.1$ (green dotted line). (b) Full range of the equation of state parameter for $\delta M_{Pl}=+0.1$.[]{data-label="fig_w_escrad"}](w_escrad_b "fig:"){width="6.23cm" height="5.0cm"} In figure \[potential\_canonico\]**,** $V(\varphi)$ is shown for some values of $\delta$ and $c$. Notice that there is a region where $V(\varphi)$ is almost constant, that is, there is a slow-roll region. As we chose $\dot{\varphi}$ positive, then $\varphi$ evolves to this slow-roll region. However, if we had chosen $\dot{\varphi}$ negative, then because the right hand side of (\[eq\_fi\_sc\]) would have the opposite sign, so $\frac {dV(\varphi)}{d\varphi}$ would have also the opposite sign and again $\varphi$ would evolve to the slow-roll region. Notice also that for $\delta M_{Pl}=+0.1$, the potential is negative in the past. The equation for evolution of $\varphi$ (\[eq\_fi\_sc\]) can be written in an integral form as$$\Delta\varphi\left( z\right) =-\sqrt{3}M_{Pl}\int_{0}^{z}\frac{\sqrt {\Omega_{\varphi}(z)\left( 1+\omega_{\varphi}\left( z\right) \right) }}{1+z}dz\text{ .}$$ Since the model depends on $\Delta\varphi$ - through $E\left( z\right) $ - and neither on $\varphi$ nor on $\varphi_{0}$, then it is independent of $\varphi_{0}$. In other words, $\varphi_{0}$ is not a parameter of the model and can be chosen arbitrarily. Therefore, the parameters of the model are $\delta$, $c$, $h,$ $\Omega_{b0}$ and $\Omega_{\varphi0}$. ![Potential of the holographic quintessence field $V(\varphi)$, in units of $\rho_{c0}=3M_{Pl}^{2}H_{0}^{2}$. $\varphi$ is in units of $M_{Pl}$. The solid lines are for $c=0.95$, the dashed ones are for $c=1.1$ and the dotted are for $c=1.25$. For each value of $c$ the curves from right to left are for $\delta M_{Pl}=-0.1$ (red), $\delta M_{Pl}=0$ (black) and $\delta M_{Pl}=+0.1$ (green), respectively.[]{data-label="potential_canonico"}](potential_can){width="6.23cm" height="5.0cm"} Tachyon scalar field -------------------- In the case of dark energy modeled as the tachyon scalar field, $\mathcal{L}_{\varphi}\left( x\right) $ in the action (\[action\]) is given by (\[tachyonic\]). From a variational principle, we obtain$$i\gamma^{\mu}\nabla_{\mu}\Psi-M^{\ast}\Psi=0\text{ ,} \label{dirac_tac}$$$$i(\nabla_{\mu}\bar{\Psi})\gamma^{\mu}+M^{\ast}\bar{\Psi}=0\text{ ,} \label{diracadj_tac}$$ where $M^{\ast}\equiv M-\beta\varphi$, and$$\nabla_{\mu}\partial^{\mu}\varphi+\alpha\frac{\partial^{\mu}\varphi (\nabla_{\mu}\partial_{\sigma}\varphi)\partial^{\sigma}\varphi}{1-\alpha \partial_{\mu}\varphi\partial^{\mu}\varphi}+\frac{1}{\alpha}\frac {dlnV(\varphi)}{d\varphi}=\frac{\beta\bar{\Psi}\Psi}{\alpha V(\varphi)}\sqrt{1-\alpha\partial^{\mu}\varphi\partial_{\mu}\varphi}\text{ .} \label{eqmov_taquions}$$ The equations of motion for $\Psi$ and $\bar{\Psi}$ (\[dirac\_tac\]) and (\[diracadj\_tac\]) are the interacting covariant Dirac equation and its adjoint, respectively, i. e., (\[dirac\_tac\]) and (\[diracadj\_tac\]) are almost the same as eqs. (\[dirac\]) and (\[diracadj\]), the only difference is that the scalar field $\varphi$ in $M^{\ast}$ now is the tachyon field. For homogeneous fields and adopting the Friedmann-Robertson-Walker metric, $g_{\mu\nu}$=diag$\left( 1,-a^{2}\left( t\right) ,-a^{2}\left( t\right) ,-a^{2}\left( t\right) \right) $, where $a^{2}\left( t\right) $ is the scale factor, (\[eqmov\_taquions\]) reduces to$$\ddot{\varphi}=-(1-\alpha\dot{\varphi}^{2})\left[ \frac{1}{\alpha}\frac{dlnV(\varphi)}{d\varphi}+3H\dot{\varphi}-\frac{\beta\bar{\Psi}\Psi }{\alpha V(\varphi)}\sqrt{1-\alpha\dot{\varphi}^{2}}\right] \text{ ,} \label{homotaq}$$ whereas for the fermions, the equations of motion will reduce to eq. (\[conser\_psibarpsi\]), as already obtained above:$$\bar{\Psi}\Psi=\bar{\Psi}_{0}\Psi_{0}\left( \frac{a_{0}}{a}\right) ^{3}\text{ .} \tag{8}$$ From the energy-momentum tensor, we get$$\begin{aligned} \rho_{\varphi} & =\frac{V(\varphi)}{\sqrt{1-\alpha\dot{\varphi}^{2}}}\text{ ,}\label{rofi}\\ P_{\varphi} & =-V(\varphi)\sqrt{1-\alpha\dot{\varphi}^{2}}\text{ ,}\label{pfi}\\ \rho_{\Psi} & =M^{\ast}\bar{\Psi}\Psi\text{ ,}\nonumber\\ P_{\Psi} & =0\text{ .}\nonumber\end{aligned}$$ From (\[rofi\]) and (\[pfi\]) we have $\omega_{\varphi}\equiv \frac{P_{\varphi}}{\rho_{\varphi}}=\alpha\dot{\varphi}^{2}-1$. Deriving (\[rofi\]) and (\[pfi\]) with respect to time and using (\[homotaq\]) and (\[conser\_psibarpsi\]), we get $$\dot{\rho}_{\varphi}+3H\rho_{\varphi}(\omega_{\varphi}+1)=\beta\dot{\varphi }\bar{\Psi}_{0}\Psi_{0}\left( \frac{a_{0}}{a}\right) ^{3} \label{conser_rotac}$$ and$$\dot{\rho}_{\Psi}+3H\rho_{\Psi}=-\beta\dot{\varphi}\bar{\Psi}_{0}\Psi _{0}\left( \frac{a_{0}}{a}\right) ^{3}\text{ ,} \label{conser_psitac}$$ where the dot represents derivative with respect to time. For baryonic matter and radiation, the conservation equations are the same as in the quintessence model. We have $$\dot{\rho}_{b}+3H\rho_{b}=0 \tag{15}$$ and$$\dot{\rho}_{r}+3H\rho_{r}(\omega_{r}+1)=0\text{ ,} \tag{16}$$ where $\omega_{r}=\frac{1}{3}$. Eqs. (\[conserbaryon\]) and (\[conserrad\]) implies that $\rho_{b}=\frac{\rho_{b0}}{a^{3}}$ and $\rho_{r}=\frac {\rho_{r0}}{a^{4}}$, respectively. The subscript $0$ denotes the quantities today. We are considering the radiation as composed by photons and massless neutrinos, so that $\rho_{r0}=\left( 1+0.2271N_{eff}\right) \rho_{\gamma0}$, where $N_{eff}=3.04$ is the effective number of relativistic degrees of freedom and $\rho_{\gamma0}$ is the energy density of photons, given by $\rho_{\gamma0}=\frac{\pi^{2}}{15}T_{CMB}$, being $T_{CMB}=2.725K$ the CMB temperature today. The Friedmann equation for a flat universe reads $$H^{2}=\frac{1}{3M_{Pl}^{2}}\left[ M^{\ast}\bar{\Psi}_{0}\Psi_{0}\left( \frac{a_{0}}{a}\right) ^{3}+\frac{V(\varphi)}{\sqrt{1-\alpha\dot{\varphi}^{2}}}+\frac{\rho_{b0}}{a^{3}}+\frac{\rho_{r0}}{a^{4}}\right] \text{ .} \label{friedmann_tac}$$ Now, as done in the case of the quintessence field, we will identify the tachyon energy density (\[rofi\]) with the holographic dark energy density $\rho_{\Lambda}=3c^{2}M_{Pl}^{2}L^{-2}$. From a similar reasoning, we obtain again the equations (\[eq\_mov\_holo\]), (\[eq\_movbar\]) and (\[eq\_movrad\]),$$\frac{d\Omega_{\varphi}}{dz}=-\frac{\Omega_{\varphi}}{1+z}\left( 2\frac {\sqrt{\Omega_{\varphi}}}{c}+3\Omega_{\varphi}\omega_{\varphi}+\Omega _{r}+1\right) \text{ ,} \tag{18}$$$$\frac{d\Omega_{b}}{dz}=-\frac{\Omega_{b}}{1+z}\left( 3\Omega_{\varphi}\omega_{\varphi}+\Omega_{r}\right) \tag{19}$$ and$$\frac{d\Omega_{r}}{dz}=-\frac{\Omega_{r}}{1+z}\left( 3\Omega_{\varphi}\omega_{\varphi}+\Omega_{r}-1\right) \text{ .} \tag{20}$$ Also we obtain$$\dot{r}=3Hr\omega_{\varphi}-sign\left[ \dot{\varphi}\right] \frac {\beta\left( 1+r\right) ^{2}\sqrt{1+\omega_{\varphi}}}{3M_{Pl}^{2}\sqrt{\alpha}H^{2}}\bar{\Psi}_{0}\Psi_{0}\left( \frac{1+z}{1+z_{0}}\right) ^{3}\text{ ,} \label{rdottac}$$ with $r\equiv\frac{\rho_{\Psi}}{\rho_{\varphi}}$. The sign of $\dot{\varphi}$ is arbitrary, as it can be modified by redefinitions of the field, $\varphi\rightarrow-\varphi$, and of the coupling constant, $\beta \rightarrow-\beta$. We can rewrite $\bar{\Psi}_{0}\Psi_{0}$ in terms of observable quantities, by imposing that the dark matter density today matches the observed value. We obtain $M\bar{\Psi}_{0}\Psi_{0}=\frac{3M_{Pl}^{2}H_{0}^{2}\left( 1-\Omega_{\phi0}-\Omega_{b0}-\Omega_{r0}\right) }{1-\frac{\beta}{M\sqrt{\alpha}}\phi_{0}}$, where we defined $\phi\equiv \sqrt{\alpha}\varphi$. Furthermore, noticing that $r=\frac{1-\Omega_{\varphi }-\Omega_{b}-\Omega_{r}}{\Omega_{\varphi}}$, we can eliminate $r$ and $\dot {r}$ in favor of $\Omega_{\varphi}$, $\Omega_{b}$, $\Omega_{r}$, $\dot{\Omega }_{\varphi}$, $\dot{\Omega}_{b}$ and $\dot{\Omega}_{r}$ in (\[rdottac\]). Using (\[eq\_mov\_holo\]), (\[eq\_movbar\]) and (\[eq\_movrad\]) we obtain, after some algebra$$\omega_{\phi}\left( z\right) =-\frac{1}{3}-\frac{2\sqrt{\Omega_{\phi}\left( z\right) }}{3c}+\frac{\gamma\left( z\right) }{3}\left[ \gamma\left( z\right) +\sqrt{\gamma\left( z\right) ^{2}+4\left( 1-\frac{\sqrt {\Omega_{\phi}\left( z\right) }}{c}\right) }\right] \text{ ,} \label{wfi_final_tac}$$ where$$\gamma\left( z\right) \equiv\frac{1}{\sqrt{6}}\frac{\delta}{H_{0}}\frac{1-\Omega_{\phi0}-\Omega_{b0}-\Omega_{r0}}{\Omega_{\phi}\left( z\right) E^{3}\left( z\right) }\left( \frac{1+z}{1+z_{0}}\right) ^{3}\text{ ,} \label{gama_tac}$$ with$$E\left( z\right) \equiv\frac{H\left( z\right) }{H_{0}}=\sqrt{\frac{\left[ \left( 1-\delta\Delta\phi\right) \left( 1-\Omega_{\phi0}-\Omega_{b0}-\Omega_{r0}\right) +\Omega_{b0}\right] }{1-\Omega_{\phi}}\left( \frac {1+z}{1+z_{0}}\right) ^{3}+\frac{\Omega_{r0}}{1-\Omega_{\phi}}\left( \frac{1+z}{1+z_{0}}\right) ^{4}}\text{ ,} \label{E_final_tac}$$ where $\Delta\phi\left( z\right) \equiv\phi\left( z\right) -\phi_{0}$ and $\delta\equiv\frac{\frac{\beta}{M\sqrt{\alpha}}}{1-\frac{\beta}{M\sqrt{\alpha }}\phi_{0}}$ is an effective coupling constant. As in the quintessence field case, if $\delta=0$, (\[wfi\_final\_tac\]) reproduces the equation of state parameter obtained in [@Li]. The evolution of the tachyon scalar field is given by$$\frac{d\phi}{dz}=-\frac{\sqrt{1+\omega_{\phi}\left( z\right) }}{H_{0}E\left( z\right) \left( 1+z\right) }\text{ .} \label{eq_fi_tac}$$ From (\[eq\_mov\_holo\]), (\[eq\_movbar\]), (\[eq\_movrad\]) and (\[eq\_fi\_tac\]) we can calculate the evolution with redshift of all observables in the model. If we wish to calculate the time dependence, we need to integrate the Friedmann equation (\[friedmann\_tac\]), which can be written in the form$$\frac{dt}{dz}=-\frac{1}{H_{0}E(z)(1+z)}\text{ .}$$ From (\[rofi\]), we can compute the potential $V\left( z\right) $ as$$\frac{V\left( z\right) }{\rho_{c0}}=E^{2}\left( z\right) \Omega_{\phi }\left( z\right) \sqrt{-\omega_{\phi}\left( z\right) }\text{ ,} \label{potential}$$ where $\rho_{c0}=3M_{Pl}^{2}H_{0}^{2}$, $E\left( z\right) $ is given by (\[E\_final\_tac\]), $\omega_{\phi}\left( z\right) $ is given by (\[wfi\_final\_tac\]) and $\Omega_{\phi}\left( z\right) $ is the solution of (\[eq\_mov\_holo\]). From (\[potential\]) and (\[eq\_fi\_tac\]), we can compute $V(\phi)$. The square root in (\[wfi\_final\_tac\]) must be real. Furthermore, in analogous manner to the quintessence model, $\omega_{\phi}$ must be more than $-1$ because (\[eq\_fi\_tac\]). We can verify that $\omega_{\phi}$ is real and $\omega_{\phi}>-1$ if (i) $\frac{\sqrt{\Omega_{\phi0}}}{c}<1$ or (ii) $\frac{\sqrt{\Omega_{\phi0}}}{c}>1$ and $\frac{\left\vert \delta\right\vert }{H_{0}}>2\sqrt{6}\frac{\Omega_{\phi0}}{1-\Omega_{\phi0}-\Omega_{b0}-\Omega_{r0}}\sqrt{\frac{\sqrt{\Omega_{\phi0}}}{c}-1}$. However, case (ii) is irrelevant, as it corresponds to large values of $\frac{\left\vert \delta\right\vert }{H_{0}}$. For example, if $\Omega_{\phi0}=0.7$ and $c=0.8$, we have $\frac{\left\vert \delta\right\vert }{H_{0}}\gtrsim2.45$. Below, we will see that the observational data constrain $\frac{\left\vert \delta\right\vert }{H_{0}}\sim10^{-1}$. In order that $\omega_{\phi}$ be real for all future times, as $\Omega_{\phi}\rightarrow1$, it is necessary that $c\geq1$. As already mentioned in the quintessence case, this is also the condition for the entropy to increase for all future times, so the tachyon model also respects the second law of thermodynamics. The evolution of the equation of state parameter $\omega_{\phi}$ is showed in figure \[fig\_w\_taqrad\]. We have $\omega_{\phi}\rightarrow-1/3$ for $z\gg1$. In the non interacting case - $\delta=0$ - this occurs simply because $\Omega_{\phi}(z)\ll1$ in high redshifts. The behaviour for the interacting case - $\delta\neq0$ - is explained as follows. In the matter era $E^{2}(z)\sim\left( 1+z\right) ^{3}$ so that $\gamma\left( z\right) \sim\frac{1}{\Omega_{\phi}\left( z\right) \left( 1+z\right) ^{1,5}}$. Using (\[eq\_mov\_holo\]) we infer $\frac{d\left\vert \gamma\left( z\right) \right\vert }{dz}>0$, that is $\left\vert \gamma\left( z\right) \right\vert $ increases with redshift $z$. Therefore, if $\delta<0$ then $\gamma\left( z\right) <0$ and $\omega_{\phi}$ increases slower than in the non interacting case. If $\delta>0$ then $\gamma\left( z\right) >0$ and $\omega_{\phi}$ increases faster than in the non interacting case. In the radiation era $\gamma\left( z\right) \sim\frac{1}{1+z}$ and it turns out to be negligible, so that $\omega_{\phi}\simeq-1/3$. ![Equation of state parameter of the holographic tachyon model, for $c=0.85$ and $\frac{\delta}{H_{0}}=-0.1$ (red dashed line), $\frac{\delta }{H_{0}}=0$ (black solid line) and $\frac{\delta}{H_{0}}=+0.1$ (green dotted line).[]{data-label="fig_w_taqrad"}](w_tacrad){width="6.23cm" height="5.0cm"} In figure \[potential\_taq\], $V(\phi)$ is shown for some values of $\delta$ and $c$. Notice that - as in the case of the quintessence field potential - there is a region where $V(\phi)$ is almost constant, that is, there is a slow-roll region. As we chose $\dot{\phi}$ positive, then $\phi$ evolves to this slow-roll region. However, if we had chosen $\dot{\phi}$ negative, then because the right hand side of (\[eq\_fi\_tac\]) would have the opposite sign, so $\frac{dV(\phi)}{d\phi}$ would have also the opposite sign and again $\phi$ would evolve to the slow-roll region. ![Potential of the holographic tachyon field $V(\phi)$, in units of $\rho_{c0}=3M_{Pl}^{2}H_{0}^{2}$. $\phi$ is in units of $H_{0}^{-1}$. The solid lines are for $c=0.85$, the dashed ones are for $c=1.1$ and the dotted are for $c=1.35$. For each value of $c$ the curves from right to left are for $\frac{\delta}{H_{0}}=-0.1$ (red), $\frac{\delta}{H_{0}}=0$ (black) and $\frac{\delta}{H_{0}}=+0.1$ (green), respectively.[]{data-label="potential_taq"}](potential_tacrad){width="6.23cm" height="5.0cm"} The equation for evolution of $\phi$ (\[eq\_fi\_tac\]) can be written in an integral form as$$\Delta\phi=-\frac{1}{H_{0}}\int_{0}^{z}\frac{\sqrt{1+\omega_{\phi}\left( z\right) }}{E(z)(1+z)}dz\text{ .}$$ As before, the model depends on $\Delta\phi$ - through $E\left( z\right) $ - and neither on $\phi$ nor on $\phi_{0}$, then it is independent of $\phi_{0}$. Therefore, the parameters of the model are $\delta$, $c$, $h,$ $\Omega_{b0}$ and $\Omega_{\phi0}$. Below, we discuss the comparison with observational data and the results obtained. Constraints from observational data =================================== In [@lookback], the lookback time method has been discussed. Given an object $i$ at redshift $z_{i}$, its age $t(z_{i})$ is defined as the difference between the age of the universe at $z_{i}$ and the age of the universe at the formation redshift of the object, $z_{F}$, that is, $$\begin{aligned} t(z_{i}) & =H_{0}^{-1}\left[ \int_{z_{i}}^{\infty}\frac{dz^{\prime}}{(1+z^{\prime})E(z^{\prime})}-\int_{z_{F}}^{\infty}\frac{dz^{\prime}}{(1+z^{\prime})E(z^{\prime})}\right] \nonumber\\ & =H_{0}^{-1}\int_{z_{i}}^{z_{F}}\frac{dz^{\prime}}{(1+z^{\prime})E(z^{\prime})}=t_{L}(z_{F})-t_{L}(z_{i})\text{ ,} \label{age}$$ where $t_{L}$ is the lookback time, given by $$t_{L}(z)=H_{0}^{-1}\int_{0}^{z}\frac{dz^{\prime}}{(1+z^{\prime})E(z^{\prime})}\text{ .}$$ Using (\[age\]), the observational lookback time $t_{L}^{obs}(z_{i})$ is $$\begin{aligned} t_{L}^{obs}(z_{i}) & =t_{L}(z_{F})-t(z_{i})=[t_{0}^{obs}-t(z_{i})]-[t_{0}^{obs}-t_{L}(z_{F})]\nonumber\\ & =t_{0}^{obs}-t(z_{i})-df\text{ ,} \label{lookobs}$$ where $t_{0}^{obs}$ is the estimated age of the universe today and $df$ is the delay factor, $$df\equiv t_{0}^{obs}-t_{L}(z_{F})\ .$$ We now minimize $\chi_{lbt}^{2}$, $$\chi_{lbt}^{2}=\sum_{i=1}^{N}\frac{[t_{L}(z_{i},\vec{p})-t_{L}^{obs}(z_{i})]^{2}}{\sigma_{i}^{2}+\sigma_{t_{0}^{obs}}^{2}}\text{ ,}$$ where $t_{L}(z_{i},\vec{p})$ is the theoretical value of the lookback time in $z_{i}$, $\vec{p}$ denotes the theoretical parameters, $t_{L}^{obs}(z_{i})$ is the corresponding observational value given by (\[lookobs\]), $\sigma_{i}$ is the uncertainty in the estimated age $t(z_{i})$ of the object at $z_{i}$, which appears in (\[lookobs\]) and $\sigma_{t_{0}^{obs}}$ is the uncertainty in getting $t_{0}^{obs}$. The delay factor $df$ appears because of our ignorance about the redshift formation $z_{F}$ of the object and has to be adjusted. Note, however, that the theoretical lookback time does not depend on this parameter, and we can marginalize over it. In [@age35] and [@age32] the ages of 35 and 32 red galaxies are respectively given. For the age of the universe one can adopt $t_{0}^{obs}=13.75\pm0.11Gyr$ [@wmap7yr]. Although this estimate for $t_{0}^{obs}$ has been obtained assuming a $\Lambda CDM$ universe, it does not introduce systematical errors in the calculation: any systematical error eventually introduced here would be compensated by the adjust of $df$, in (\[lookobs\]). On the other hand, such an estimate is in perfect agreement with other estimates, which are independent of the cosmological model, as for example $t_{0}^{obs}=12.6_{-2.4}^{+3.4}Gyr$, obtained from globular cluster ages [@krauss] and $t_{0}^{obs}=12.5\pm3.0Gyr$, obtained from radioisotopes studies [@cayrel]. The WMAP distance information used by the WMAP colaboration includes the shift parameter“ $R$, the acoustic scale” $l_{A}$ and the redshift of decoupling $z_{\ast}$. These quantities are very weakly model dependent [@liR]. $R$ and $l_{A}$ are given by$$R=\sqrt{\Omega_{m0}}H_{0}r\left( z_{\ast}\right)$$ and$$l_{A}=\pi\frac{r\left( z_{\ast}\right) }{r_{s}\left( z_{\ast}\right) }\text{ ,}$$ where $r\left( z_{\ast}\right) $ is the comoving distance to $z_{\ast}$ and $r_{s}\left( z_{\ast}\right) $ is the comoving sound horizon at $z_{\ast}$. For a flat universe, $r\left( z_{\ast}\right) $ and $r_{s}\left( z_{\ast }\right) $ are given by$$r\left( z_{\ast}\right) =\frac{1}{H_{0}}\int_{0}^{z_{\ast}}\frac{dz}{E(z)}$$ and$$r_{s}\left( z_{\ast}\right) =\frac{1}{H_{0}}\int_{0}^{z_{\ast}}\frac {dz}{E(z)\sqrt{3\left( 1+\bar{R}_{b}/(1+z)\right) }}\text{ ,}$$ where $\bar{R}_{b}=3\Omega_{b0}/\left( 4\Omega_{\gamma0}\right) $. For the redshift of decoupling $z_{\ast}$ we use the fitting function proposed by Hu and Sugiyama [@sugiyama]:$$z_{\ast}=1048\left[ 1+0.00124\left( \Omega_{b0}h^{2}\right) ^{-0.738}\right] \left[ 1+g_{1}\left( \Omega_{m0}h^{2}\right) ^{g_{2}}\right] \text{ ,}$$ where$$g_{1}=\frac{0.0783\left( \Omega_{b0}h^{2}\right) ^{-0.238}}{1+39.5\left( \Omega_{b0}h^{2}\right) ^{0.763}}$$ and$$g_{2}=\frac{0.560}{1+21.1\left( \Omega_{b0}h^{2}\right) ^{1.81}}\text{ .}$$ Thus we add to $\chi^{2}$ the term$$\chi_{CMB}^{2}={\displaystyle\sum\limits_{ij}} \left( x_{i}^{th}-x_{i}^{data}\right) \left( C^{-1}\right) _{ij}\left( x_{j}^{th}-x_{j}^{data}\right) \text{ ,}$$ where $x=\left( l_{A},R,z_{\ast}\right) $ is the parameter vector and $\left( C^{-1}\right) _{ij}$ is the inverse covariance matrix for the seven-year WMAP distance information [@wmap7ykomatsu]. Baryonic Acoustic Oscilations (BAO) are described in terms of the parameter $$A=\sqrt{\Omega_{M}}E(z_{BAO})^{-1/3}\left[ \frac{1}{z_{BAO}}\int_{0}^{z_{BAO}}\frac{dz^{\prime}}{E(z^{\prime})}\right] ^{2/3}\text{ ,}$$ where $z_{BAO}=0.35$. It has been estimated that $A_{obs}=0.493\pm0.017$ [@BAO1]. We thus add to $\chi^{2}$ the term $$\chi_{BAO}^{2}=\frac{\left( A-A_{obs}\right) ^{2}}{\sigma_{A}^{2}}\ \text{.}$$ The BAO distance ratio $r_{BAO}\equiv D_{V}\left( z=0.35\right) /D_{V}\left( z=0.20\right) =1.812\pm0.060$, estimated from the joint analysis of the 2dFGRS (Two Degree Field Galaxy Redshift Survey) and SDSS (Sloan Digital Sky Survey) data [@BAO2], has also been included. It was demonstrated in [@BAO2] that this quantity is weakly model dependent. The quantity $D_{V}\left( z_{BAO}\right) $ is given by$$D_{V}\left( z_{BAO}\right) =c\left[ \frac{z_{BAO}}{H\left( z_{BAO}\right) }\left( \int_{0}^{z_{BAO}}\frac{dz^{\prime}}{H\left( z^{\prime}\right) }\right) ^{2}\right] ^{1/3}\text{ .}$$ So we have the contribution$$\chi_{r_{BAO}}^{2}=\frac{\left( r_{BAO}-r_{BAO}^{obs}\right) ^{2}}{\sigma_{r_{BAO}}^{2}}\text{ .}$$ Finally, we add the 397 supernovae data from Constitution compilation [@constitution]. Defining the distance modulus $$\mu(z)=5log_{10}\left[ c(1+z)\int_{0}^{z}\frac{dz^{\prime}}{E(z^{\prime})}\right] +25-5log_{10}H_{0}\text{ ,}$$ we have the contribution $$\chi_{SN}^{2}=\sum_{j=1}^{397}\frac{[\mu(z_{j})-\mu_{obs}(z_{j})]^{2}}{\sigma_{j}^{2}}\text{ .}$$ Using the expression $\chi^{2}=\chi_{lbt}^{2}+\chi_{CMB}^{2}+\chi_{BAO}^{2}+\chi_{r_{BAO}}^{2}+\chi_{SN}^{2}$, the likelihood function is given by$$\mathcal{L}(\delta,c,h,\Omega_{b0},\Omega_{\phi_{0}})\propto exp[-\frac {\chi^{2}(\delta,c,h,\Omega_{b0},\Omega_{\phi_{0}})}{2}]\ \text{.}$$ Quintessence field ------------------ In table 1 we present the values of the individual best fit parameters, with respective $1\sigma$, $2\sigma$ and $3\sigma$ confidence intervals. **Table 1**: Values of the holographic quintessence model parameters from lookback time, CMB, BAO and SNe Ia. In the last line, $\chi_{\min}^{2}/dof$ is the minimum $\chi^{2}$ per degree of freedom. \[c\][|l|l|]{}$\delta M_{Pl}$ & $-0.170_{-0.072-0.123-0.225}^{+0.067+0.148.+0.187}$\ $c$ & $0.891_{-0.016-0.028-0.035}^{+0.048+0.134+0.234}$\ $\Omega_{\phi0}$ & $0.7733_{-0.0087-0.0217-0.0353}^{+0.0092+0.0162+0.0264}$\ $\Omega_{b0}$ & $0.0450_{-0.0028-0.0048-0.0066}^{+0.0026+0.0061+0.0089}$\ $h$ & $0.687\pm0.013\pm0.026\pm0.039$\ $\chi_{\min}^{2}/dof$ & $1.185$\ Figure \[distributions\_esc\] shows the marginalized probability distributions for $\delta$ and $c$. The coupling constant $\delta$ is non vanishing at $2\sigma$ confidence level. Figure \[bidimensionals\_esc\] shows some joint confidence regions of two parameters. Notice the effect of the condition **** $\omega_{\phi}\leq1$ **** on the positive tail of the probability distribuction of $\delta$ and also in the confidence regions of $\delta$ with other parameters. ![Probability distribuctions of the coupling constant $\delta$ (left panel) and of the parameter $c$ (right panel) of the holographic quintessence model.[]{data-label="distributions_esc"}](pdelta_escrad "fig:"){width="6.23cm" height="5.0cm"}  ![Probability distribuctions of the coupling constant $\delta$ (left panel) and of the parameter $c$ (right panel) of the holographic quintessence model.[]{data-label="distributions_esc"}](pc_escrad "fig:"){width="6.23cm" height="5.0cm"} ![Two parameters confidence regions of $1\sigma$, $2\sigma$ and $3\sigma$ of the holographic quintessence model.[]{data-label="bidimensionals_esc"}](bideltac_escrad "fig:"){width="5.66cm" height="4.55cm"}![Two parameters confidence regions of $1\sigma$, $2\sigma$ and $3\sigma$ of the holographic quintessence model.[]{data-label="bidimensionals_esc"}](bideltaomfi0_escrad "fig:"){width="5.66cm" height="4.55cm"}![Two parameters confidence regions of $1\sigma$, $2\sigma$ and $3\sigma$ of the holographic quintessence model.[]{data-label="bidimensionals_esc"}](bideltaomb0_escrad "fig:"){width="5.66cm" height="4.55cm"} ![Two parameters confidence regions of $1\sigma$, $2\sigma$ and $3\sigma$ of the holographic quintessence model.[]{data-label="bidimensionals_esc"}](bideltah_escrad "fig:"){width="5.66cm" height="4.55cm"}![Two parameters confidence regions of $1\sigma$, $2\sigma$ and $3\sigma$ of the holographic quintessence model.[]{data-label="bidimensionals_esc"}](bicomfi0_escrad "fig:"){width="5.66cm" height="4.55cm"}![Two parameters confidence regions of $1\sigma$, $2\sigma$ and $3\sigma$ of the holographic quintessence model.[]{data-label="bidimensionals_esc"}](bicomb0_escrad "fig:"){width="5.66cm" height="4.55cm"} Tachyon scalar field -------------------- In table 2 we present the values of the individual best fit parameters, with respective $1\sigma$, $2\sigma$ and $3\sigma$ confidence intervals. **Table 2**: Values of the holographic tachyon model parameters from lookback time, CMB, BAO and SNe Ia. In the last line, $\chi_{\min}^{2}/dof$ is the minimum $\chi^{2}$ per degree of freedom. \[c\][|l|l|]{}$\frac{\delta}{H_{0}}$ & $-0.201_{-0.059-0.138-0.249}^{+0.063+0.117+0.176}$\ $c$ & $0.868_{-0.020-0.030-0.038}^{+0.059+0.165+0.235}$\ $\Omega_{\phi0}$ & $0.724_{-0.012-0.026-0.040}^{+0.013+0.025+0.036}$\ $\Omega_{b0}$ & $0.0480\pm0.0021\pm0.0041\pm0.0062$\ $h$ & $0.669\pm0.012\pm0.025\pm0.037$\ $\chi_{\min}^{2}/dof$ & $1.148$\ Figure \[distributions\_tac\] shows the marginalized probability distributions for $\delta$ and $c$. The coupling constant $\delta$ is non vanishing at more than $3\sigma$ confidence level. So for this model we obtained strong evidence for interaction. Figure \[bidimensionals\_tac\] shows some confidence regions of two parameters for this model. ![Probability distribuctions of the coupling constant $\delta$ (left panel) and of the parameter $c$ (right panel) of the holographic tachyon model.[]{data-label="distributions_tac"}](pdelta_tacrad "fig:"){width="6.23cm" height="5.0cm"}  ![Probability distribuctions of the coupling constant $\delta$ (left panel) and of the parameter $c$ (right panel) of the holographic tachyon model.[]{data-label="distributions_tac"}](pc_tacrad "fig:"){width="6.23cm" height="5.0cm"} ![Two parameters confidence regions of $1\sigma$, $2\sigma$ and $3\sigma$ of the holographic tachyon model.[]{data-label="bidimensionals_tac"}](bideltac_tacrad "fig:"){width="5.66cm" height="4.55cm"}![Two parameters confidence regions of $1\sigma$, $2\sigma$ and $3\sigma$ of the holographic tachyon model.[]{data-label="bidimensionals_tac"}](bideltaomfi0_tacrad "fig:"){width="5.66cm" height="4.55cm"}![Two parameters confidence regions of $1\sigma$, $2\sigma$ and $3\sigma$ of the holographic tachyon model.[]{data-label="bidimensionals_tac"}](bideltaomb0_tacrad "fig:"){width="5.66cm" height="4.55cm"}  ![Two parameters confidence regions of $1\sigma$, $2\sigma$ and $3\sigma$ of the holographic tachyon model.[]{data-label="bidimensionals_tac"}](bideltah_tacrad "fig:"){width="5.66cm" height="4.55cm"}![Two parameters confidence regions of $1\sigma$, $2\sigma$ and $3\sigma$ of the holographic tachyon model.[]{data-label="bidimensionals_tac"}](bicomfi0_tacrad "fig:"){width="5.66cm" height="4.55cm"}![Two parameters confidence regions of $1\sigma$, $2\sigma$ and $3\sigma$ of the holographic tachyon model.[]{data-label="bidimensionals_tac"}](bicomb0_tacrad "fig:"){width="5.66cm" height="4.55cm"} Discussion and conclusions ========================== The minimum $\chi^{2}$ per degree of freedom values indicate that both models fit well the observational data. Moreover, we see that the tachyon model is a bit more favored by the observational data than the quintessence one. **** The dimensionless coupling constants, $\delta M_{Pl}$ for the quintessence field and $\frac{\delta}{H_{0}}$ for the tachyon, agree at $1\sigma$ level and for both models we obtained significative evidence for a non vanishing interaction in the dark sector. Furthermore, both the results implies in dark energy decaying into dark matter, alleviating the coincidence problem. **** These results are consistent with previous ones, as for example those obtained in [@sandro2], where an interacting tachyonic dark energy model with a power law potential was assumed and the results were consistent with the interaction at $90\%$ confidence level. We must also mention that in [@sandro] and [@virial] evidence for interaction was found using completely different models and data sets. So those results combined with the present ones furnish sensible evidence in favor of an interaction in the dark sector of the universe. The results obtained for $c$, $\Omega_{b0}$ and $h$ for the two models also agree at $1\sigma$ confidence level. The values obtained for $\Omega_{\phi0}$ agree only at $2\sigma$ level. For the quintessence model, $\Omega_{\phi0}$ is almost superestimated, corresponding to a matter relative density today $\Omega_{m0}=0.231\pm0.010$. This value agrees at inferior limit of $1\sigma$ with the cosmological model independent estimative $\Omega_{Mobs}=0.28\pm0.04$ [@riess]. For the tachyon model, $\Omega_{\phi0}$ corresponds to a matter relative density today $\Omega_{m0}=0.276_{-0.012}^{+0.013}$, which is in perfect agreement with that observational estimative. For both models, the baryionic density and the Hubble parameter today are very reasonable. We have obtained $\Omega_{b0}h^{2}=0.0212\pm0.0015$ and $\Omega_{b0}h^{2}=0.0215\pm0.0012$ from quintessence and tachyon models, respectively. Let’s compare these values, for example, with that obtained from deuterium to hydrogen abundance ratio [@omeara], $\Omega_{b0}h^{2}=0.0213\pm 0.0013\pm0.0004$, where the errors terms represents the $1\sigma$ errors from deuterium to hydrogen abundance ratio and the uncertainties in the nuclear reaction rates, respectively. For the Hubble parameter, we have obtained $h=0.687\pm0.013$ and $h=0.669\pm0.012$ from the quintessence and tachyon models, respectively, both in excellent agreement with observational values, independent of cosmological model, as for example $h_{obs}=0.69\pm0.12$ [@age35] and $h_{obs}=0.72\pm0.08$ [@key]. We can also compare the ratios $\Omega_{b0}/\Omega_{m0}=0.248_{-0.023}^{+0.022}$, from the quintessence model, and $\Omega_{b0}/\Omega_{m0}=0.211\pm0.016$, from the tachyon model, with the observational value of 2dFGRS colaboration, $\Omega_{b0}/\Omega_{m0}=0.185\pm0.046$ [@2dFGRS]. We have obtained $c<1$ at $1\sigma$ confidence level for both models. As already said above, this implies that the equation of state parameter $\omega_{\phi}$ will not be real for all future times. However, this is not a very serious problem, because $c$ is compatible with values above unit at $2\sigma$ confidence level. Moreover, one could say that $c<1$ is only an effect due to lack of more precise observational data. Anyway, the very simple models presented here are expected to be only alternatives to an effective description of a more sophisticated subjacent theory of dark energy. In principle, nothing guarantees that they will be good descriptions for all future times. Figures \[bidimensionals\_esc\] and \[bidimensionals\_tac\] shows some joint confidence regions of two parameters for both models. In the confidence regions for $\delta$ versus $c$ and for $c$ versus $\Omega_{\phi0}$, we see that there is a lower limit on $c$ $\approx0.8$. This also can be seen in the marginalized probability distributions of $c$, which dies for $c\lesssim0.85$. This lower limit is explained by the condition $\frac{\sqrt{\Omega_{\phi0}}}{c}<1$, necessary for $\omega_{\phi}$ to be real and $\omega_{\phi}>-1$, discussed above. This limit can be seen more clearly in $c$ versus $\Omega_{\phi0}$ confidence regions. Moreover, we have $c\simeq\sqrt {\Omega_{\phi0}}$ for the best fit values of these parameters. This implies that $\omega_{\phi0}\simeq-1$ and both models approaches $\Lambda CDM$ today. This is consistent with the fact that, as $\Lambda CDM$ fits all observational data, then any alternative model must not deviates much from $\Lambda CDM$ for $z\approx0$. However, for $z>0$, both models are qualitatively different from $\Lambda CDM$. For the quintessence field we have very different qualitative behaviours for $\delta<0$, $\delta=0$ and $\delta>0$, as showed in figure \[fig\_w\_escrad\] and discussed in II A. For the tachyon model its behaviour is qualitatively the same in the three cases, $\omega_{\phi}$ approximates $-1/3$, as showed in figure \[fig\_w\_taqrad\] and discussed in II B. It is interesting to compare the results obtained in the present work for the holographic tachyon model with the previous ones, presented in [@sandro3], where a simpler version of the holographic tachyon model, without barions nor radiation, had been compared with observational data. **** The values obtained here for $c,$ $\Omega_{\phi0}$ and $h$ are the same as before and with minor incertainty intervals, despite the fact that here we have one more parameter - $\Omega_{b0}$ -, as we can see by comparing table 2 in the present work with table 1 in [@sandro3]. This is because now it was possible to use all WMAP distance information $R,$ $l_{A}$ and $z_{\ast}$, as the model was generalized to include barions and radiation. **** However the dark energy coupling constant $\delta$ now is non vanishing and compatible with dark energy decaying into dark matter with more than $3\sigma$ confidence level, whereas in [@sandro3] no evidence of interaction had been found. This can be understood as follows. In low redshifts, the universe is dominated by dark energy. The dynamics of dark energy in low redshifts is essentially determined by the equation of state parameter $\omega_{\phi}\left( z\right) $, as can be seen in (\[eq\_mov\_holo\]). As in this period the data sets were the same in both the works, so $\omega_{\phi}\left( z\right) $ must be almost the same in this period in both the works. But $\omega_{\phi}\left( z\right) $ explicitly depends on the product $\delta\Omega_{\Psi0}$, were $\delta$ is the coupling constant and $\Omega_{\Psi0}=1-\Omega_{\varphi 0}-\Omega_{b0}-\Omega_{r0}$ is the dark matter relative density, as we can see in (\[wfi\_final\_tac\]) and (\[gama\_tac\]). As in the present work $\Omega_{\Psi0}$ is less than that in [@sandro3] - where $\Omega _{b0}=\Omega_{r0}\equiv0$ -, it turns out that $\delta$ in the present work is bigger (in modulus) than that in [@sandro3]. However it is important to point out that in [@sandro3] the model was less realistic, as it wasn’t include barions nor radiation. Therefore, the present result favorable to interaction is more robust. In summary, combinations of holographic dark energy model and scalar fields were implemented. It was showed that it is possible to fix the potential of interacting scalar fields by imposing that the energy density of the scalar field must match the energy density of the holographic dark energy. A comparison of the models with recent observational data was made and the coupling is non vanishing at more than $2\sigma$ for the quintessence field and at more than $3\sigma$ for the tachyon. In both cases the results are consistent with dark energy decaying into dark matter, alleviating the coincidence problem. **Acknowledgements** This work has been supported by CNPq (Conselho Nacional de Desenvolvimento Científico e Tecnológico) of Brazil. [99]{} W. Zimdahl and D. Pavon *Phys. Lett.* **B521** (2001) 133; L. P. Chimento, A. S. Jakubi, D. Pavon and W. Zimdahl *Phys. Rev.* **D67** (2003) 083513. J.-H. He and B. Wang *JCAP* **06** (2008) 010; C. Feng, B. Wang, E. Abdalla and R.-K. Su *Phys. Lett.* **B665** (2008) 111; J.-H. He, B. Wang and E. Abdalla, Phys. Lett. **B671** (2009), 139. B. Wang, J. Zang, C.-Y. Lin, E. Abdalla and S. Micheletti, *Nucl. Phys.* **B778** (2007) 69. S. Micheletti, E. Abdalla and B. Wang, *Phys. Rev.* **D79** (2009) 123506. B. Gumjudpai, T. Naskar, M. Sami and S. Tsujikawa *JCAP* **06** (2005) 007; B. Wang, Y.-G. Gong and E. Abdalla, *Phys. Lett.* **B624** (2005) 141; M. R. Setare, *Phys. Lett.* **B642** (2006) 1; *Eur. Phys. J.* **C50** (2007) 991; idem, *Phys. Lett.* **B654** (2007) 1; E. Abdalla and B. Wang *Phys. Lett.* **B651** (2007) 89; R. Rosenfeld *Phys. Rev.* **D75** (2007) 083509; M. Quartin, M. O. Calvao, S. E. Joras, R. R. R. Reis and I. Waga *JCAP* **05** (2008) 007; Q. Wu, Y. Gong, A. Wang and J.S. Alcaniz *Phys. Lett.* **B659** (2008) 34; M.R. Setare and E. C. Vagenas *Phys. Lett.* **B666** (2008) 111; M. Jamil, M. A. Rashid *Eur. Phys. J.* **C56** (2008) 429; *Eur. Phys. J.* **C58** (2008) 111; M.R. Setare and E. C. Vagenas, *Int. J. Mod. Phys.* **D18** (2009) 147; X.-M. Chen, Y.-G. Gong and E. N. Saridakis,*JCAP* **04** (2009) 001; Z.-K. Guo, N. Ohta and S. Tsujikawa, *Phys. Rev.* **D76** (2007) 023508; O. Bertolami, F. Gil Pedro and M. Le Delliou, *Phys. Lett.* **B654** (2007) 165; O. Bertolami, F.Gil Pedro and M.Le Delliou, *Gen. Rel. Grav.* **41** (2009) 2839; L. P. Chimento, *Phys. Rev.* **D81** (2010) 043525. S. Micheletti, *JCAP* **05** (2010) 009. P. J. E. Peebles, *Physical Cosmology*, (Princeton U. Press, 1993). E. Abdalla, L. R. W. Abramo, L. Sodre Jr. and B. Wang, *Phys. Lett.* **B673**, (2009) 107; E. Abdalla, L. R. W. Abramo and J. C. C. de Souza, *Phys. Rev.* **D82** (2010) 023508. R. Bean, E. E. Flanagan, I. Laszlo and M. Trodden *Phys. Rev.* **D78** (2008) 123514. E. J. Copeland, M. Sami and S. Tsujikawa, *Int. J. Mod. Phys.* **D15** (2006) 1753; M. Sami *Curr. Sci.* **97** (2009) 887. I. Zlatev, L. Wang and P. J. Steinhardt, *Phys. Rev. Lett.* **82** (1999) 896; P. J. Steinhardt, L. Wang and I. Zlatev, *Phys. Rev.* **D59** (1999) **** 123504; L. Amendola *Phys. Rev.* **D62** (2000) 043511; R.R. Caldwell and E. V. Linder, *Phys. Rev. Lett.***95** (2005) 141301; R. J. Scherrer and A. A. Sen, *Phys. Rev.* **D77** (2008) 083515; A. A. Sen, G. Gupta and S. Das, *JCAP* **09** (2009) 027. A. Sen *JHEP* **04** (2002) 048; *JHEP* **07** (2002) 065; *Mod. Phys. Lett.* **A17** (2002) 1797. T. Padmanabhan *Phys. Rev.* **D66** (2002) 021301; A. Feinstein *Phys. Rev.* **D66** (2002) 063511; J. S. Bagla, H. K. Jassal and T. Padmanabhan *Phys. Rev.* **D67** (2003) 063504; L. R. W. Abramo and F. Finelli *Phys. Lett.* **B575** (2003) 165; R. Herrera, D. Pavon and W. Zimdahl, *Gen. Rel. Grav.* vol. **36** n${{}^o}$ **9** (2004) 2161; A. Ali, M. Sami and A. A. Sen, *Phys. Rev.* **D79** (2009) 123501. J. Zhang, X. Zhang and H. Liu, *Phys. Lett.* **B651**, (2007) 84; M. R. Setare, *Phys. Lett.* **B653**, (2007) 116. X. Zhang, *Phys. Lett.* **B648** (2007) 1. X. Zhang, *Phys. Rev.* **D74** (2006) 103505. T. Damour, G. W. Gibbons and C. Gundlach, *Phys. Rev. Lett.* **64** (1990) 123. M. Li, *Phys. Lett.* **B603**, (2004) 1; Q.-G. Huang and M. Li, *JCAP* **08**, **** (2004) 013. S. Capozziello, V. F. Cardone, M. Funaro and S. Andreon *Phys. Rev.* **D70** (2004) 123501. R. Jimenez, L. Verde, T. Treu and D. Stern *Astrophys. J*. **593** (2003) 622. J. Simon, L. Verde and R. Jimenez *Phys. Rev.* **D71** (2005) 123001. N. Jarosik et. al., *Astrophys. J. Suppl.* **192**, (2011) 14. *WMAP Cosmological Parameters Model/Dataset Matrix homepage,* <http://lambda.gsfc.nasa.gov/product/map/current/best_params.cfm> L. M. Krauss *astro-ph/0301012*. R. Cayrel et. al. *Nature* **409** (2001) 691. H. Li, J.-Q. Xia, G.-B. Zhao, Z.-H. Fan and X. Zhang, *Astrophys. J.* **683** (2008) L1; Y. Wang and P. Mukherjee *Phys. Rev.* **D76** (2007) 103533. W. Hu and N. Sugiyama, *Astrophys. J.* **471** (1996) 542. E. Komatsu, et. al., *Astrophys. J. Suppl.* **192** (2011) 18. B. A. Reid et. al. *0907.1659 \[astro-ph.CO\]*. W. J. Percival et. al. *Mon. Not. Roy. Astron. Soc.* **381** (2007) 1053. M. Hicken et. al. *Astrophys. J*. **700** (2009) 1097. A. G. Riess et. al., *Astrophys. J.* **659** (2007) 98. J. M. O’Meara, et. al., *Astrophys. J.* **649** (2006) L61. W. L. Freedman et. al., *Astrophys. J.* **553** (2001) 47. S. Cole, et. al., *Mon. Not. Roy. Astron. Soc.* **362** (2005) 505 [^1]: smrm@fma.if.usp.br

ArXiv

--- abstract: 'This is a written discussion of the paper “Optimal treatment allocations in space and time for on-line control of an emerging infectious disease” by E. B. Laber, N. J. Meyer, B. J. Reich, K. Pacifici, J. A. Collazo and J. Drake, contributed to the Journal of the Royal Statistical Society Series C.' address: University of Manchester author: - Johan Koskinen title: 'Discussion of the paper “Optimal treatment allocations in space and time for on-line control of an emerging infectious disease” by E. B. Laber, N. J. Meyer, B. J. Reich, K. Pacifici, J. A. Collazo and J. Drake[^1] ' --- Disease spread on social networks ================================= This very thorough and welcomed contribution demonstrates the use of simulation in modern inference, and engages with several academic traditions, one of which, social network analysis, I would like to add some reflections on. Spread on networks has a long history in social network analysis (Coleman et al., 1957) and, particularly in the health science, there has been much recent attention paid to interventions (Valente, 2012; Morris, 2004). Estimating these processes has prompted the development of statistical models (such as Greenan, 2015) and simulation routines (e.g. Jenness et al., 2016). Rolls et al. (2012) investigate the behaviour of hepatitis C transmission on an empirical network (Aitken et al., 2008) and demonstrate effects of different types of network topologies on disease outcomes. Information on these disease-relevant networks is by necessity patchy but can be learned through sampling techniques such as in Rolls et al (2013a) or Handcock and Gile (2010). In fact, key features can to some extent even be estimated from respondent interviews (Krivitsky and Morris, 2017). Many of these methods rely on exponential random-graph models (Lusher et al., 2013), a class of log-linear models that model interactions between links that correspond to features such as the prevalence of hubs and the tendency towards clustering of ties. Figure 1 illustrates potential consequences of such features in a ‘realistic’ network (b) relative to an ‘unrealistic’ network (a) with the same number of links. The local clustering of network (b), means that links, rather than carrying the disease further from the seed node (black), out into the network, are used up linking to nodes that are already connected. Real life networks – both human and animal (even bats, see e.g. Willis, & Brigham, 2004) – typically demonstrate high clustering (not merely artefacts of geography, Daraganova et al., 2012) and relatively short path-lengths. The example network N3 is meant to have these features (Robins et al., 2005) but to the naked eye appears to have rather long pathways. In fact the results, pairwise for S1 and N1, S2 and N2, and S3 and N3, respectively, seem to reflect the fact that the network topologies are more stringent forms of linkage than their spatial equivalents, resulting in similar behaviour but with lower uncertainty. How would the spread and interventions be affected by the more distinct network features? Obviously, these are big questions that have many moving parts (Rolls et al., 2013b) but the proposed allocation scheme is also very flexible. biblabel\[1\] Aitken, C.K., Lewis, J., Tracy, S.L., Spelman, T., Bowden, D.S., Bharadwaj, M., Drummer, H. and Hellard, M. (2008). High incidence of hepatitis C virus reinfection in a cohort of injecting drug users. [*Hepatology*]{}, 48(6), 1746–1752. Coleman, J., Katz, E., & Menzel, H. (1957). The diffusion of an innovation among physicians. [*Sociometry*]{}, 20(4), 253–270. Daraganova, G., Pattison, P., Koskinen, J., Mitchell, B., Bill, A., Watts, M., & Baum, S. (2012). Networks and geography: modelling community network structures as the outcome of both spatial and network processes. [*Social Networks*]{}, 34(1), 6–17. Greenan, C. C. (2015). Diffusion of innovations in dynamic networks. [*Journal of the Royal Statistical Society: Series A (Statistics in Society)*]{}, 178(1), 147–166. Handcock, M. S., and Gile, K. (2010). Modeling social networks from sampled data, [*The Annals of Applied Statistics*]{}, vol. 4, pp. 5–25. Jenness S.M., Goodreau S.M., Morris M. (2016). EpiModel: Mathematical Modeling of Infectious Disease. R Package Version 1.2.6. <http://epimodel.org/>. DOI: 10.5281/zenodo.16767 Krivitsky, P.N., and Morris, M. (2017). Inference for social network models from egocentrically sampled data, with application to understanding persistent racial disparities in HIV prevalence in the US. [*The Annals of Applied Statistics*]{}, Vol. 11 (1), 427-455. Lusher, D., Koskinen, J., Robins, G., (2013). [*Exponential Random Graph Models for Social Networks: Theory, Methods and Applications*]{}, New York: Cambridge University Press. Morris, M., (2004). [*Network Epidemiology: A Handbook for Survey Design and Data Collection*]{}. Oxford: Oxford University Press. Robins, G. L., Pattison, P. E., and Woolcock, J. (2005), Small and other worlds: Global network structures from local processes, American Journal of Sociology, 110, 894–936. Rolls, D. A., G. Daraganova, R. Sacks-Davis, M. Hellard, R. Jenkinson, E. McBryde, P. E. Pattison, and G. L. Robins. (2013a). Modelling a disease-relevant contact network of people who inject drugs. [*Social Networks*]{}, 35(4), 699-710. Rolls, D. A., Sacks-Davis, R., Jenkinson, R., McBryde, E., Pattison, P., Robins, G., & Hellard, M. (2013b). Hepatitis C transmission and treatment in contact networks of people who inject drugs. [*PloS one*]{}, 8(11), e78286. Valente, T.W. (2012) Network Interventions. [*Science*]{}, 337, 49-53. Willis, C. K., & Brigham, R. M. (2004). Roost switching, roost sharing and social cohesion: forest-dwelling big brown bats, Eptesicus fuscus, conform to the fission–fusion model. Animal Behaviour, 68(3), 495–505. [^1]: This is the pre-peer reviewed version of the following article: Koskinen, J.H. (2018), Discussion of “Optimal treatment allocations in space and time for on-line control of an emerging infectious disease” by Laber, N. J. Meyer, B. J. Reich, K. Pacifici, J. A. Collazo and J. Drake, Journal of the Royal Statistical Society Series C, 67, p779, which has been published in final form at https://rss.onlinelibrary.wiley.com/doi/full/10.1111/rssc.12266

ArXiv

--- abstract: 'We study the two-dimensional massless Dirac equation for a potential that is allowed to depend on the energy and on one of the spatial variables. After determining a modified orthogonality relation and norm for such systems, we present an application involving an energy-dependent version of the hyperbolic Scarf potential. We construct closed-form bound-state solutions of the associated Dirac equation.' --- $^\dagger$ and $^\ddagger$ \ $\dagger$ Department of Mathematics and Actuarial Science and Department of Physics, Indiana University Northwest, 3400 Broadway, Gary IN 46408, USA, e-mail: axgeschu@iun.edu, xbataxel@gmail.com\ \ $\ddagger$ Physics and Applied Mathematics Unit, Indian Statistical Institute, Kolkata 700108, India, e-mail: pinaki@isical.ac.in\ \ \ \ Keywords: Dirac equation, energy-dependent potential, hyperbolic Scarf potential Introduction ============ The two-dimensional massless Dirac equation can be used to model electron transport phenomena in graphene, an atomically thin conducting material that consists of carbon atoms forming a honeycomb lattice structure [@novo]. This structure gives rise to many unusual properties of graphene, such as its very high electric conductivity, where both electrons and holes serve as charge carriers [@neto]. Low-energy states of electrons or holes in graphene can be modeled by the two-dimensional massless Dirac equation [@gonz]. In order to confine charge carriers, the Dirac equation must be coupled to a suitable scalar or vector potential. While for the vast majority of such external potentials the Dirac equation will not be solvable, there are some exceptions. Such exceptional cases include coupling to scalar potentials [@ghosh] [@ho] [@hartmann] [@xbat], as well as to vector potentials [@check] [@jak] [@roy]. The purpose of this work is to extend the latter context to the case where the external potential depends on the energy. Quantum systems that feature energy-dependent potentials have been studied in the nonrelativistic case, for a comprehensive theoretical review and introductory examples the reader may refer to [@formanek] and [@sazdjian]. Energy-dependent potentials appear in a variety of applications, such as hydrodynamics [@hydro1], confined quantum systems [@lombard] [@yekken2], or multi-nucleon systems [@miya]. Theoretical applications include the generation of nonrelativistic models with energy-dependent potentials by means of the supersymmetry formalism [@yekken] and through point transformations [@jesus] [@xbatpct]. In the nonrelativistic context, the presence of energy-dependent potentials require a modification of the underlying quantum theory, principally affecting orthogonality relation and norm [@formanek]. In this note we show that a similar type of modification is also necessary in the relativistic context if the potential is energy-dependent. There is very few literature on the topic, particular systems were studied for example in [@hassan] [@ikot], [@ikot2]. We consider here a particular case of the two-dimensional, massless Dirac equation for an external scalar potential that we assume to depend on the energy and on a single spatial variable. In section 2 we derive a modified orthogonality relation and norm for systems governed by such Dirac equations. Afterwards, we introduce an energy-dependent version of the hyperbolic Scarf potential. The conventional, energy-independent version of this potential was shown to support closed-form zero-energy states [@ghosh]. In section 3 we construct bound-state solutions of our Dirac equation for the energy-dependent hyperbolic Scarf potential and give several examples. The relativistic model ====================== We start out by introducing the two-dimensional, massless Dirac equation for an energy-dependent potential. The time-dependent version of this equation that we consider here features a potential that depends only on one of the spatial coordinates. It can be written in atomic units as follows $$\begin{aligned} \left[-i~\alpha \cdot \nabla+ V \hspace{-.1cm}\left(x,i \frac{\partial}{\partial t} \right) \right] \hat{\Psi}(x,y,t) &=& i~\frac{\partial \hat{\Psi}(x,y,t)}{\partial t},~~~~~(x,y,t) \in \mathbb{R}^2 \times (0,\infty), \label{tdirac} \end{aligned}$$ where $\alpha=(\sigma_1,\sigma_2)$ has the Pauli spin matrices $\sigma_1,\sigma_2$ as components and the potential $V$ is a continuous function of two variables. We can generate energy-dependence in the potential upon setting $$\begin{aligned} \hat\Psi(x,y,t) &=& \exp\left(-i~E~t \right) \Psi(x,y), \label{ts}\end{aligned}$$ introducing the wave number $k_y$ that describes free motion in the $y$-direction, and the real-valued constant $E$ that will represent the stationary energy of our system. Insertion into (\[tdirac\]) gives a stationary Dirac equation of the form $$\begin{aligned} \left\{-i~\alpha \cdot \nabla+ \left[V(x,E)-E \right] \right\} \Psi(x,y) &=& 0,~~~(x,y) \in \mathbb{R}^2. \label{dirac} \end{aligned}$$ We observe that the potential now depends on the energy $E$. Note further that, following the usual convention, we suppress the dependence on the energy $E$ in the solution $\Psi$ and its components. Derivation of orthogonality relation and norm --------------------------------------------- It was shown [@formanek] that in the nonrelativistic scenario the presence of an energy-dependent potential forces a modification of the orthogonality relation and the norm, defined in the usual $L^2$-sense. For this reason, a similar modification is in order to accomodate systems governed by the Dirac equation (\[dirac\]). Our starting point for the construction of a modified orthogonality relation and norm is the continuity equation $$\begin{aligned} \frac{\partial P(x,y,t)}{\partial t} &=& - \nabla J(x,y,t), \label{cont}\end{aligned}$$ where $P$ and $J$ denote the relativistic probability density and probability current, respectively, that depend on the two spatial coordinates and on the time $t$. Let us now assume that $E_n$ and $E_m$ are two nonequal energies, for which the stationary Dirac equation (\[dirac\]) admits solutions $\Psi_n$ and $\Psi_m$, respectively. The associated solutions of the time-dependent Dirac equation (\[tdirac\]) can be found through (\[ts\]). In addition, since the potential in our Dirac equation depends only on the $x$-coordinate, we can separate the $y$-coordinate off. More precisely, we set for $j=m$ and $j=n$ $$\begin{aligned} \hat{\Psi}_j(x,y,t) ~=~ \exp\left(-i ~E_j~ t+i~k_y~y\right) \Psi_j(x), \label{hatpsi}\end{aligned}$$ Keeping this relation in mind, we will now replace our standard continuity equation (\[cont\]) by a modified version that satisfies the requirements imposed by an energy-dependent potential. Let us first define the probability density $P$ and the probability current $J$. These objects take the same form as in the conventional case, where the potential does not depend on the energy. We have $$\begin{aligned} P(x,y,t) &=& \hat{\Psi}_m^\dagger(x,y,t)~\hat{\Psi}_n(x,y,t) \label{prob} \\[1ex] J(x,y,t) &=& \hat{\Psi}_m^\dagger(x,y,t)~\alpha~\hat{\Psi}_n(x,y,t). \label{curr}\end{aligned}$$ Recall that the time-dependency of the expressions on the right sides is governed by (\[hatpsi\]). We will now show that (\[prob\]) and (\[curr\]) satisfy the following modified continuity equation $$\begin{aligned} \frac{\partial P(x,y,t)}{\partial t} +i\left[V(x,E_n)-V(x,E_m)\right] \hat{\Psi}_m^\dagger(x,y,t)~\hat{\Psi}_n(x,y,t) &=& -\nabla J(x,y,t), \label{contmod}\end{aligned}$$ where the symbol $\dagger$ denotes the hermitian adjoint. We observe that in contrast to the standard continuity equation (\[cont\]), the modified version (\[contmod\]) contains an additional term. The presence of this term is similar to the nonrelativistic scenario that was discussed in [@formanek] [@sazdjian]. Let us briefly show that (\[prob\]) and (\[curr\]) indeed satisfy the modified continuity equation (\[contmod\]). To this end, we substitute (\[prob\]) into the left side of the latter equation. For the sake of legibility we first evaluate the derivative with respect to the time variable. Taking into account the Dirac equation (\[dirac\]) and using standard properties of the $\alpha$-matrix, this gives $$\begin{aligned} \frac{\partial P(x,y,t)}{\partial t} &=& \frac{\partial}{\partial t} \left[ \hat{\Psi}_m^\dagger(x,y,t)~\hat{\Psi}_n(x,y,t) \right] \nonumber \\[1ex] &=& \frac{\partial \hat{\Psi}_m^\dagger(x,y,t)}{\partial t}~\hat{\Psi}_n(x,y,t)+ \hat{\Psi}_m^\dagger(x,y,t) ~\frac{\partial \hat{\Psi}_n(x,y,t)}{\partial t} \nonumber \\[1ex] &=& \left[-\alpha \cdot \nabla \hat{\Psi}_m^\dagger(x,y,t)+i~V(x,E_m)~\hat{\Psi}_m^\dagger(x,y,t) \right] \hat{\Psi}_n(x,y,t) + \nonumber \\[1ex] &+& \hat{\Psi}_m^\dagger(x,y,t) \left[-\alpha \cdot \nabla \hat{\Psi}_n(x,y,t)-i~V(x,E_n)~\hat{\Psi}_n(x,y,t) \right] \nonumber \\[1ex] & & \hspace{-3cm}=~-\left[\alpha \cdot \nabla \hat{\Psi}_m^\dagger(x,y,t) \right] \hat{\Psi}_n(x,y,t) -\left[\alpha \cdot \nabla \hat{\Psi}_n(x,y,t) \right] \hat{\Psi}_m^\dagger(x,y,t) + i\left[V(x,E_m)-V(x,E_n) \right] \nonumber \\[1ex] & & \hspace{-3cm}=~-\nabla \left[\hat{\Psi}_m^\dagger(x,y,t)~\alpha~\hat{\Psi}_n(x,y,t) \right] + i\left[V(x,E_m)-V(x,E_n) \right] \nonumber \\[1ex] & & \hspace{-3cm}=~-\nabla J(x,y,t) + i\left[V(x,E_m)-V(x,E_n) \right]. \label{contmodcheck}\end{aligned}$$ If we replace the derivative with respect to the time variable on the left side of (\[contmod\]) by expression (\[contmodcheck\]), we see that our modified continuity equation is satisfied. We are now able to construct an orthogonality relation for the solutions $\Psi_m$ and $\Psi_n$ of the stationary Dirac equation (\[dirac\]). To this end, we integrate our continuity equation (\[contmod\]) with respect to the time variable. Recall that the time-dependence of our spinors lies entirely in an exponential function as shown in (\[hatpsi\]). Taking into account the latter definition in combination with (\[prob\]), the left side of (\[contmod\]) can be integrated as follows $$\begin{aligned} \int\limits^t \left\{ \left[\frac{\partial}{\partial s}~ \hat{\Psi}_m^\dagger(x,y,s)~\hat{\Psi}_n(x,y,s) \right]+i\left[V(x,E_n)-V(x,E_m)\right] \hat{\Psi}_m^\dagger(x,y,s)~\hat{\Psi}_n(x,y,s) \right\} ds ~=\nonumber \\[1ex] & & \hspace{-9.7cm} =~ \left[1-\frac{V(x,E_m)-V(x,E_n)}{E_m-E_n}\right] \hat{\Psi}_m^\dagger(x,y,t)~\hat{\Psi}_n(x,y,t). \label{intprob}\end{aligned}$$ At this point it is convenient to make use of the relation (\[hatpsi\]) by substituting it into (\[intprob\]). Similar to the nonrelativistic case [@formanek] [@sazdjian], this leads to the sought orthogonality relation $$\begin{aligned} \int\limits_{\mathbb{R}} \left[1-\frac{V(x,E_m)-V(x,E_n)}{E_m-E_n}\right] \Psi_m^\dagger(x)~\Psi_n(x)~dx &=& C~\delta_{mn}, \label{ortho}\end{aligned}$$ where $C$ is a constant. From the orthogonality relation (\[ortho\]) we can now construct the modified norm $N$ by taking the limit $m \rightarrow n$, resulting in $$\begin{aligned} N \hspace{-.1cm} \left(\Psi_n \right) &=& \int\limits_{\mathbb{R}} \left[1-\frac{\partial V(x,E_n)}{\partial E_n}\right] \Psi_n^\dagger(x)~\Psi_n(x)~dx. \label{norm}\end{aligned}$$ For a solution $\Psi_n$ of the Dirac equation (\[dirac\]) to represent a bound state, two conditions must be fulfilled: the norm integral (\[normx\]) must exist and at the same time its integrand must be a nonnegative function. Since the sign of the integral is entirely determined by the factor involving the potential’s derivative, the condition $$\begin{aligned} 1-\frac{\partial V(x,E_n)}{\partial E_n} &\geq& 0, \label{sign}\end{aligned}$$ must be satisfied for all real numbers $x$ and energies $E_n$ associated with the system governed by the Dirac equation (\[dirac\]). Finally let us note that (\[norm\]) does not constitute a norm in the mathematical sense because it can become negative due to the term containing the energy derivative of the potential. Decoupling the Dirac system --------------------------- Before we can consider applications involving specific energy-dependent potentials, we must solve the Dirac equation (\[tdirac\]). Since this equation has two components, it can be written as a system of two equations that must be decoupled. To this end, we represent the solution spinor $\hat{\Psi}$ in the form (\[hatpsi\]) and split it up into its two components. We set $$\begin{aligned} \hat{\Psi}(x,y,t) &=& \frac{1}{2}~\exp\left(-i~E_n~t+i~k_y~y \right) \left( \begin{array}{ll} \psi_+(x) \\[1ex] \psi_-(x) \end{array} \right) \nonumber \\[1ex] &=& \frac{1}{2}~\exp\left(-i~E_n~t+i~k_y~y \right) \left( \begin{array}{ll} \psi_{1}(x)+\psi_{2}(x) \\[1ex] \psi_{1}(x)-\psi_{2}(x) \end{array} \right). \label{psinew}\end{aligned}$$ Note that the factor $1/2$ was introduced merely to facilitate calculations. Upon substitution of (\[psinew\]) into (\[tdirac\]), we obtain the following system of equations [@ghosh] for the spinor components $\psi_{1}$ and $\psi_{2}$. $$\begin{aligned} \psi_{1}''(x)+\Big\{\left[V(x,E)-E \right]^2+i~\frac{\partial V(x,E)}{\partial x}-k_y^2\Big\}~ \psi_{1}(x) ~=~0 \label{eq1} \end{aligned}$$ $$\begin{aligned} \hspace{-.85cm} \psi_{2}(x) ~=~ \frac{1}{k_y} ~\Big\{\psi_{1}'(x)+i \left[V(x,E)-E \right] \psi_{1}(x)\Big\}, \label{psi2}\end{aligned}$$ where we assume without restriction that $k_y \neq 0$. Once the first solution component $\psi_{1}$ has been found from the Schrödinger-type equation (\[eq1\]), the remaining counterpart $\psi_{2}$ is generated by means of (\[psi2\]). These two functions are then substituted into (\[psinew\]) in order to obtain the solution spinor of the stationary Dirac equation. Before we apply the results of this section to a specific model, we rewrite the orthogonality relation (\[ortho\]) and norm (\[norm\]) in terms of the solutions to the system (\[eq1\]), (\[psi2\]). To this end, we introduce two pairs of functions $\psi_{1,m}$, $\psi_{2,m}$ and $\psi_{1,n}$, $\psi_{2,n}$ that are solutions to the latter system for $E=E_m$ and $E=E_n$, respectively. Upon substituting relation (\[psinew\]) into our orthogonality relation (\[ortho\]) and norm (\[norm\]), we obtain the results $$\begin{aligned} & & \int\limits_{\mathbb{R}} \left[1-\frac{V(x,E_m)-V(x,E_n)}{E_m-E_n}\right] \Bigg\{\left[\psi^\ast_{1,m}(x)+\psi^\ast_{2,m}(x) \right] \left[\psi_{1,n}(x)+\psi_{2,n}(x) \right]+ \nonumber \\[0ex] & & \hspace{4.55cm}+\left[\psi^\ast_{1,m}(x)-\psi^\ast_{2,m}(x) \right] \left[\psi_{1,n}(x)-\psi_{2,n}(x) \right]\Bigg\}~ dx ~=~ C~\delta_{mn} \nonumber \\[1ex] & & N \hspace{-.1cm} \left(\psi_n \right) ~=~ \int\limits_{\mathbb{R}} \left[1-\frac{\partial V(x,E_n)}{\partial E_n}\right] \Big[|\psi_{1,n}(x)+\psi_{2,n}(x)|^2+ |\psi_{1,n}(x)-\psi_{2,n}(x)|^2\Big] dx. \label{normx}\end{aligned}$$ Observe that the dependence on the spatial variable $y$ is gone because the corresponding exponential term from (\[hatpsi\]) cancels out, leaving a single integral. Note further that we left some irrelevant overall constant factors out. Application: hyperbolic Scarf potential ======================================= We will now introduce a particular energy-dependent potential, for which our stationary Dirac equation (\[dirac\]) admits bound-state solutions that can be given in closed form. The potential we will focus on reads $$\begin{aligned} V(x,E) &=& -\lambda(E)~\mbox{sech}(x)+\mu(E)~\tanh(x)+E, \label{v}\end{aligned}$$ where $\lambda \neq 0$ and $\mu$ are real-valued functions of the energy parameter $E$. We see that the function (\[v\]) is an energy-dependent generalization of the hyperbolic Scarf potential. It is known [@ghosh] that the conventional, energy-indepedent version of our potential represents a well for electrons if $\lambda>0$ and a well for holes if $\lambda<0$. In what follows we will show that this interpretation can be maintained if the potential is energy-dependent and of the form (\[v\]), provided certain constraints are met. Observe that the last term on the right side of (\[v\]) will cancel with the same term in our Dirac equation (\[dirac\]). As such, solutions of the latter equation are formally equivalent to zero-energy solutions for the scenario of an energy-independent potential. Let us further remark that the potential (\[v\]) has a formal similarity with the potential discussed in [@hartmann2], as far as the shape of its graph is concerned. However, the latter reference considers the conventional, energy-independent context. General solution of the governing equation ------------------------------------------ Our starting point is the observation that the Dirac equation (\[dirac\]) for our potential (\[v\]) is exactly-solvable. The general solution provided in [@ghosh] persists under the generalization regarding the energy-dependent parameters. Since the explicit form of the solution spinor (\[psinew\]) is too long to be shown here, we focus on the function $\psi_1$ that is determined by the Schrödinger-type equation (\[eq1\]). This equation reads after incorporation of (\[v\]) $$\begin{aligned} \psi_1''(x)+\Bigg\{-k_y^2+\mu(E)^2+\mbox{sech}^2(x)\Big[\lambda(E)^2+i~\mu(E)-\mu(E)^2 \Big] +\mbox{sech}(x)~\tanh(x) \Big[i~\lambda(E)- & & \nonumber \\[1ex] & & \hspace{-6.0cm} -~2~\lambda(E)~\mu(E) \Big] \Bigg\} ~\psi_1(x) ~=~ 0. \label{eqpre}\end{aligned}$$ The general solution of this equation for $\psi_{1}$ is given by $$\begin{aligned} \psi_{1}(x) \hspace{-.1cm} &=& \hspace{-.1cm}c_1~\left[1-i~\sinh(x) \right]^{\frac{c}{2}-\frac{1}{4}} ~\left[1+i~\sinh(x) \right]^{\frac{a}{2}+\frac{b}{2}-\frac{c}{2}+\frac{1}{4}}~ {}_2F_1\left[ a,b,c,\frac{1}{2}-\frac{i}{2}~\sinh(x)\right]+ \nonumber \\[1ex] & & \hspace{-1.5cm}+~c_2~ \left[1-i~\sinh(x) \right]^{\frac{3}{4}-\frac{c}{2}} ~ \left[1+i~\sinh(x) \right]^{\frac{1}{4}-\frac{a}{2}-\frac{b}{2}+\frac{c}{2}}~ {}_2F_1\left[ 1-a,1-b,2-c,\frac{1}{2}-\frac{i}{2}~\sinh(x)\right] \hspace{-.1cm}. \nonumber \\[1ex] \label{psi1x}\end{aligned}$$ Here, $c_1, c_2$ are arbitrary constants and ${}_2F_1$ stands for the hypergeometric function [@abram]. Furthermore, the following abbreviations are in use $$\begin{aligned} a&=&\frac{1}{2}-\frac{1}{4}~\sqrt{\left[-1+2~\lambda(E)-2~i~\mu(E)\right]^2} -\frac{1}{4}~\sqrt{\left[1+2~\lambda(E)+2~i~\mu(E)\right]^2}+\sqrt{k_y^2-\mu(E)^2} \nonumber \\[1ex] b&=&\frac{1}{2}-\frac{1}{4}~\sqrt{\left[-1+2~\lambda(E)-2~i~\mu(E)\right]^2} -\frac{1}{4}~\sqrt{\left[1+2~\lambda(E)+2~i~\mu(E)\right]^2}-\sqrt{k_y^2-\mu(E)^2} \nonumber \\[1ex] c&=&1-\frac{1}{2}~\sqrt{\left[1+2~\lambda(E)+2~i~\mu(E)\right]^2}. \label{abc}\end{aligned}$$ These expressions can be simplified further once the sign of the radicands is known. We will discuss this in detail further below. Observe that in (\[abc\]) we did not include an argument to indicate the dependency of $a,b$ and $c$ on the energy $E$. We note that the function $\psi_2$ in (\[psinew\]) can now be obtained from (\[psi1x\]) through the relation (\[psi2\]), determining the general solution of our Dirac equation (\[dirac\]) for the potential (\[v\]). Construction of bound states ---------------------------- We will now impose additional conditions on (\[psi1x\]) in order to extract bound-state solutions and their corresponding energies. For such solutions, the norm integral (\[normx\]) must exist and the sign condition (\[sign\]) must be fulfilled. We will investigate these two aspects separately. #### Existence of the norm integral. Let us first ensure that the norm (\[normx\]) exists in the present case. To this end, we observe that our solution (\[psi1x\]) becomes in general unbounded at the infinities due to the hypergeometric functions it contains. We modify the latter solution by setting $c_1=1$ and $c_2=0$, removing its second term on the right side. Next, we recall that the hypergeometric function degenerates to a polynomial if its first argument equals a nonpositive integer. Taking into account that this argument is given by $a$ and defined in (\[abc\]), we obtain the constraint $$\begin{aligned} \frac{1}{2}-\frac{1}{4}~\sqrt{\left[1+2~\lambda(E)+2~i~\mu(E)\right]^2} -\frac{1}{4}~\sqrt{\left[-1+2~\lambda(E)-2~i~\mu(E)\right]^2}+\sqrt{k_y^2-\mu(E)^2} = -n, \nonumber \\ \label{eigenpre}\end{aligned}$$ for a nonnegative integer $n$. Since the first two complex roots on the left side can take two values each, we can simplify (\[eigenpre\]) by distinguishing four possible cases, depending on the sign of the two roots. These cases are $$\begin{aligned} \frac{1}{2}-\lambda(E)+\sqrt{k_y^2-\mu(E)^2} = -n \label{eigen1} \\[1ex] \frac{1}{2}+\lambda(E)+\sqrt{k_y^2-\mu(E)^2} = -n \label{eigen2} \\[1ex] 1+i~\mu(E)+\sqrt{k_y^2-\mu(E)^2} = -n \label{eigen3} \\[1ex] -i~\mu(E)+\sqrt{k_y^2-\mu(E)^2} = -n. \label{eigen4}\end{aligned}$$ Next, let us show that the last two cases can be discarded. To this end, we first assume that the root in (\[eigen3\]) and (\[eigen4\]) is real-valued. This implies $\mu=0$, such that the energy $E$ completely disappears from the condition. As a consequence, no stationary energies can be determined. If we assume that the root in (\[eigen3\]) is imaginary, then (\[eigen3\]) results in $n=-1$, which is not a valid assignment due to the restriction that $n$ must be a nonnegative integer. Finally, if the root in (\[eigen4\]) takes imaginary values, we obtain $n=0$ and $|\mu(E)|=k_y/\sqrt{2}$. While this is in principle acceptable, we will see below that bound states of our system can only be constructed if $\mu$ is a constant. As such, the energy $E$ will once more disappear from our condition (\[eigen4\]). For these reasons we only retain the conditions (\[eigen1\]) and (\[eigen2\]) that were obtained by assuming that the first two complex roots on the left side of (\[eigenpre\]) take the same sign. Now, our condition (\[eigenpre\]) can be rewritten using (\[eigen1\]) and (\[eigen2\]) as follows $$\begin{aligned} \frac{1}{2}-\epsilon~ \lambda(E_n)+\sqrt{k_y^2-\mu(E_n)^2} ~=~ -n \qquad\mbox{and} \qquad |k_y|~\geq~|\mu(E_n)|, \label{eigen}\end{aligned}$$ where $n$ is a nonnegative integer and the new parameter $\epsilon$ can take either the value positive one or negative one. Since our potential coefficients $\lambda$ and $\mu$ depend on the stationary energy, we cannot determine an explicit formula for those energies from (\[eigen\]), unless more information about the coefficients is known. Before we continue, a general remark on the role of the parameter $k_y$ is in order. We observe that a condition is placed on $k_y$ in order for (\[eigen\]) to deliver real-valued energies. While this condition on $k_y$ is relatively mild, there are cases where stronger constraints are imposed. A typical example of such a case is the work [@ghosh], where bound-state solutions of the Dirac equations are sought at zero energy. It is found that bound states can be constructed provided $k_y$ is constrained to attain certain values. This type of constraint does not appear in the present case because we do not set the energy to a fixed value. Let us for now assume that (\[eigen\]) is satisfied. The stationary energies defined in the latter condition belong to bound-state type solutions of (\[eqpre\]), given by the functions $$\begin{aligned} \psi_{1,n}(x) &=& \left[1-i~\sinh(x)\right]^{\frac{1}{4}-\frac{\epsilon}{4} \left[1+2 \lambda(E)+2 i \mu(E)\right]} \left[1+i~\sinh(x)\right]^{\frac{1}{4}- \frac{\epsilon}{4} \left[-1+2 \lambda(E)-2 i \mu(E)\right]} \times \nonumber \\[1ex] && \hspace{-1.8cm} \times~ {}_2F_1\left[-n,-n-2~\sqrt{k_y^2-\mu(E)^2},1-\frac{\epsilon}{2}~\sqrt{\left[1+2~\lambda(E)+2~i~\mu(E)\right]^2}, \frac{1}{2}-\frac{i}{2} \sinh(x) \right], \label{psihyp}\end{aligned}$$ where irrelevant overall factors were omitted. Since the first argument of the hypergeometric function in (\[psihyp\]) is a nonpositive integer, we can express it as follows $$\begin{aligned} \psi_{1,n}(x) &=& \left[1-i~\sinh(x)\right]^{\frac{1}{4}-\frac{\epsilon}{4}\left[1+2 \lambda(E)+2 i \mu(E)\right]} \left[1+i~\sinh(x)\right]^{\frac{1}{4}-\frac{\epsilon}{4}\left[-1+2 \lambda(E)-2 i \mu(E)\right]} \times \nonumber \\[1ex] && \hspace{-1.5cm} \times~ P_n^{\left(-\frac{\epsilon}{2}\sqrt{\left[1+2~\lambda(E)+2~i~\mu(E)\right]^2},-\frac{\epsilon}{2}\sqrt{\left[1-2~\lambda(E)+2~i~\mu(E)\right]^2} \right)}\left[i \sinh(x) \right]. \label{psi1bound}\end{aligned}$$ Here, the symbol $P_n$ stands for a Jacobi polynomial of degree $n$. Before we continue, let us point out that the functions (\[psi1bound\]) do not lead to bound states of our Dirac equation (\[dirac\]) unless the parameters satisfy certain conditions. In particular, existence and positiveness of our norm (\[normx\]) is not guaranteed in general. In order to find out more about this, let us now analyze the asymptotic behavior of the solution (\[psi1bound\]) at the infinities. For the sake of simplicity we first assume that $n=0$, such that the Jacobi polynomial becomes equal to one and only two factors on the right side of (\[psi1bound\]) remain. The asymptotics of these factors for large values of $|x|$ is determined by the real parts of their exponents. In particular, at least one of the exponents must have negative real part. In case one of the exponents has positive real part, it must be less than the absolute value of its counterpart. For $n=0$ and $\epsilon=1$, (\[psi1bound\]) simplifies to $$\begin{aligned} \psi_1(x) &=& \left[1-i~\sinh(x)\right]^{-\frac{\lambda(E)}{2}-\frac{i \mu(E)}{2}} \left[1+i~\sinh(x)\right]^{\frac{1-\lambda(E)}{2}+\frac{i \mu(E)}{2}}. \label{psic1}\end{aligned}$$ We see that the exponents satisfy our requirements if the condition $\lambda(E) > \frac{1}{2}$ is satisfied. Let us now evaluate (\[psi1bound\]) for $n=0$ and $\epsilon=-1$. This gives $$\begin{aligned} \psi_1(x) &=& \left[1-i~\sinh(x)\right]^{\frac{1+\lambda(E)}{2}+\frac{i \mu(E)}{2}} \left[1+i~\sinh(x)\right]^{\frac{\lambda(E)}{2}-\frac{i \mu(E)}{2}}. \label{psic2}\end{aligned}$$ This time we see that our requirements on imply $\lambda(E) < -\frac{1}{2}$. If we now drop the assumption of vanishing $n$, we must consider the effect that the Jacobi polynomial in (\[psi1bound\]) has on our conditions for $\lambda$. To this end, we observe that the Jacobi polynomial depends on the variable $x$ merely through the term $1-i \sinh(x)$. As such, the degree of the polynomial adds $n$ to the exponent of the first factor on the right side of (\[psic1\]) and (\[psic2\]), respectively. As a direct consequence, our constraint on $\lambda$ becomes for $\epsilon=1$ $$\begin{aligned} \lambda(E) &>& n+\frac{1}{2}. \label{l1}\end{aligned}$$ This generalizes a condition found in [@ghosh] to the energy-dependent potential (\[v\]). If (\[l1\]) is satisfied, bound-state solutions of the Dirac equation (\[dirac\]) describe the behavior of electrons. Next, if we choose $\epsilon=-1$, we arrive at the condition $$\begin{aligned} \lambda(E) &<& -n-\frac{1}{2}, \label{l2}\end{aligned}$$ Similar to (\[l1\]), this is a generalization of a constraint valid for the energy-independent version of (\[v\]). If (\[l2\]) holds, then bound-state solutions of the Dirac equation (\[dirac\]) model the behavior of holes [@ghosh]. Hence, if either condition (\[l1\]) or (\[l2\]) are satisfied, then the corresponding solution in (\[psi1bound\]) vanishes at both infinities. Furthermore, the derivative of (\[psi1bound\]) with respect to $x$ shows the same behavior, implying that the remaining component (\[psi2\]) forming our solution spinor (\[psinew\]) vanishes at both infinities. For the sake of brevity we omit to show this rigorously, as it would require a similar series of considerations as done above for the function (\[psi1bound\]). It now follows that the density $|\psi_+|^2+|\psi_-|^2$ also vanishes at the infinities, establishing existence of the norm integral in (\[normx\]). #### Sign of the norm integral. It now remains to study the sign of the norm (\[normx\]) in order to ensure that condition (\[sign\]) is fulfilled. The expression on the left side of this condition reads after substitution of our potential (\[v\]) $$\begin{aligned} 1-\frac{\partial V(x,E)}{\partial E} &=& \lambda'(E)~\mbox{sech}(x)-\mu'(E)~\tanh(x). \label{ve}\end{aligned}$$ Note that for a fixed value of $E$, this expression stays bounded on the whole real line, such that it cannot affect existence of the integral (\[normx\]). In order to satisfy condition (\[sign\]), the right side of (\[ve\]) must be nonnegative for all $x$ and all admissible values of $E$. This is only possible if the coefficient of the hyperbolic secant is positive and if the hyperbolic tangent term is not present. In other words, we must impose the simultaneous conditions $$\begin{aligned} \lambda'(E)~>~0 ~~~~\mbox{and}~~~~ \mu(E)~=~\mbox{constant}, \label{normex}\end{aligned}$$ for all values of the energy $E$. In summary, if one of the conditions (\[l1\]) or (\[l2\]) is met and if in addition the sign condition (\[normex\]) is fulfilled, then the functions (\[psi1bound\]) generate bound-state solutions of the Dirac equation (\[dirac\]) with energy-dependent potential (\[v\]) by means of (\[psi2\]) and (\[psinew\]). #### The case . Before we conclude this section, let us briefly comment on a particular case of our potential (\[v\]) that arises if the parameter $\mu$ vanishes. The resulting potential, consisting of a single secant term, satisfies the sign condition (\[normex\]) and as such allows for the construction of bound-state solutions to our Dirac equation (\[dirac\]). The secant potential is of importance especially in applications of graphene, as it was shown to match the graphene top-gate structure [@hart] [@hartmann]. In the latter references, bound-state solutions of the Dirac equation for an energy-independent secant potential were obtained, within a quasi-exactly solvable setting and at zero energy, respectively. Let us add that zero-energy solutions of the Dirac equation have also been found for different types of potentials, see for example [@down]. In the present case of energy-dependence in the potential, bound-state solutions and their associated stationary energies can be obtained directly from (\[psi1bound\]) and (\[eigen\]), respectively, by setting $\mu=0$. We will comment on this case below when discussing examples. For small values of $\mu$, the hyperbolic Scarf potential (\[v\]) is a deformation of the secant potential, which is an even function. Due to this property, the effective complex potential in the Schrödinger-type equation (\[eqpre\]) features ${\cal PT}$-symmetry [@bender]. Applications ------------ Even though we were able to construct the general form of bound-state solutions through (\[psi1bound\]), we can only give an implicit equation (\[eigen\]) for the associated stationary energies. This changes once more information is known about the parameter functions $\lambda$ and $\mu$. Therefore, we will now choose particular cases of those functions and apply the results of the previous sections. It will turn out that, depending on the parameter values, the resulting stationary energies form infinite sequences that can be unbounded or have an accumulation point. ### Linear energy-dependence In our first example let us employ the following settings $$\begin{aligned} \lambda(E) ~=~ \alpha~E \qquad \qquad \qquad \mu(E) ~=~ \beta, \label{set1}\end{aligned}$$ where $\alpha>0$ and $\beta$ are real-valued constants. We observe that these settings are compatible with the requirement (\[normex\]), a necessary condition for the construction of bound states. Furthermore, the constant $\beta$ is allowed to vanish, in which case (\[v\]) turns into the hyperbolic secant potential. The remaining conditions for norm existence will be verified further below. Upon substitution of (\[set1\]) into the potential (\[v\]), we obtain $$\begin{aligned} V(x,E) &=& -\alpha~E~\mbox{sech}(x)+\beta~\tanh(x)+E. \label{v1}\end{aligned}$$ This potential depends linearly on the energy in its first and third term. Next, let us determine the stationary energies of the system governed by the Dirac equation (\[dirac\]) and the potential (\[v1\]). These energies can be found from equation (\[eigen\]), where we must first provide a value for $\epsilon$. This value depends on which of the two conditions (\[l1\]), (\[l2\]) we want to satisfy. We choose the first of these conditions, implying that $\epsilon=1$. Upon substitution of this value in combination with (\[set1\]) into (\[eigen\]), we obtain the following condition $$\begin{aligned} \frac{1}{2}- \alpha~E+\sqrt{k_y^2-\beta^2} ~=~ -n,~~~|k_y|~\geq~\beta. \nonumber\end{aligned}$$ Solving for $E$ will give an explicit formula for our stationary energies. In order to indicate this, we amend $E$ by an index $n$, arriving at $$\begin{aligned} E_n &=& \frac{1}{\alpha} \left(n+\frac{1}{2}+\sqrt{k_y^2-\beta^2}\right), ~~~|k_y|~\geq~\beta. \label{eigenx1}\end{aligned}$$ These values are positive and increase linearly with $n$. Figure \[fig3a\] shows the lowest stationary energies for a particular setting of our parameters in (\[set1\]). Next, we must check for which values of the parameters $\alpha$, $\beta$ these energies comply with the existence condition (\[l1\]) of the norm. Note that due to our parameter choice $\epsilon=1$, we do not consider the second condition (\[l2\]). Taking into account the definition of $\lambda$ in (\[set1\]), we substitute (\[eigenx1\]) into (\[l1\]). This gives $\lambda(E_n) > n + 1/2$, so that after simplification we obtain $$\begin{aligned} \sqrt{k_y^2-\beta^2} &>& 0. \label{true}\end{aligned}$$ This constraint is satisfied since it coincides with our requirement in (\[eigenx1\]). Since condition (\[normex\]) is already satisfied, it follows that the stationary energies (\[eigenx1\]) are associated with solutions that are normalizable in the sense (\[normx\]). There are infinitely many such bound-state solutions because the constraint (\[true\]) is fulfilled for all values of $n$. Since the closed form of the bound-state solutions is very long, we restrict ourselves to showing only the function $\psi_1$. To this end, we insert the settings (\[set1\]) and (\[eigenx1\]) into (\[psi1bound\]), arriving at the result $$\begin{aligned} \psi_{1,n}(x) &=& \left[1-i~\sinh(x)\right]^{\frac{1}{4}-\frac{1}{2}\left(1+n+\sqrt{k_y^2-\beta^2}+4 i \beta\right)} \left[1+i~\sinh(x)\right]^{\frac{1}{4}-\frac{1}{2}\left(n+\sqrt{k_y^2-\beta^2}-4 i \beta\right)} \times \nonumber \\[1ex] && \hspace{-1.5cm} \times~ P_n^{\left(-\frac{1}{2}\sqrt{\left(2+2n+2\sqrt{k_y^2-\beta^2})+2 i \beta\right)^2}, -\frac{1}{2}\sqrt{\left(-2 n-2\sqrt{k_y^2-\beta^2}+2~i~\beta\right)^2} \right)}\left[i \sinh(x) \right]. \label{psi1}\end{aligned}$$ Note that we included the parameter $n$ as an index in order to indicate the bound-state character. In order to construct the solution to the Dirac equation (\[dirac\]), we calculate $\psi_{2,n}$ by means of (\[psi2\]). After that, we can calculate the norm of these bound-state solutions by means of (\[normx\]). Since we know that the norm integral exists, it remains to ensure that it gives a nonnegative result. To this end, we recall that the sign of the norm is determined by expression (\[ve\]). In the present case, this expression is obtained by substituting (\[set1\]) and evaluating the derivatives, giving $$\begin{aligned} 1-\frac{\partial V(x,E_n)}{\partial E_n} &=& \alpha~\mbox{sech}(x). \label{ve1}\end{aligned}$$ Since both the constant $\alpha$ and the hyperbolic secant function are positive, the norm (\[normx\]) of our bound-state solutions generated by (\[psi1\]) will also be positive. Now that we have found the functions $\psi_{1,n}$ and $\psi_{2,n}$, we can determine the components $\psi_+$ and $\psi_-$ of the Dirac spinor (\[psinew\]) through addition and subtraction, respectively. Figure \[fig1\] shows these components for a particular parameter setting and the first values of $n$. Note that we normalized the functions shown in the figure, such that $|\psi_+|^2=|\psi_-|^2=1$. ### Inverse-power energy-dependence We will now employ a new set of parameter values for our energy-dependent potential (\[v\]). Even though we are using the same form of the potential, it will turn out that in this example the sequence of stationary energies does not increase linearly, but converges to zero. We make the following parameter definitions $$\begin{aligned} \lambda(E) ~=~ -\frac{\alpha}{E} \qquad \qquad \qquad \mu(E) ~=~ \beta, \label{set2}\end{aligned}$$ where the $\alpha>0$ and $\beta$ are real constants. These settings comply with the condition (\[sign\]) that ensures the integrand in the norm (\[normx\]) to be a nonnegative function. Let us add that $\beta$ can be zero, such that this example includes the case of a hyperbolic secant potential. Next, upon substitution of the parameters (\[set2\]) into our potential (\[v\]) we obtain $$\begin{aligned} V(x,E) &=& \frac{\alpha}{E}~\mbox{sech}(x)+\beta~\tanh(x)+E, \label{v2}\end{aligned}$$ In contrast to its counterpart (\[v1\]) from the previous example, this potential has inverse energy dependence in its first term. We will now construct the stationary energies supported by the Dirac equation (\[dirac\]) with potential (\[v2\]). To this end, we must solve equation (\[eigen\]) with respect to $E$, where we again choose the parameter value $\epsilon=1$. After incorporation of the settings (\[set2\]) we get the condition $$\begin{aligned} \frac{1}{2}+\frac{\alpha}{E}+\sqrt{k_y^2-\beta^2} &=& -n,~~~|k_y|~\geq~\beta. \nonumber\end{aligned}$$ We now obtain our stationary energies by solving for $E$. Upon renaming $E=E_n$ we arrive at $$\begin{aligned} E_n &=& -\frac{2~\alpha}{2~n+1+2~\sqrt{k_y^2-\beta^2}},~~~|k_y|~\geq~\beta. \label{eigenx2}\end{aligned}$$ These energy values are negative and increase monotically with $n$, as shown in figure \[fig3b\]. Next we need to find out how many stationary energies are provided by (\[eigenx2\]), let us verify that our parameter $\lambda$ satisfies the condition (\[l2\]), guaranteeing existence of the norm (\[normx\]). We substitute the settings (\[set2\]) and (\[eigenx2\]) into (\[l2\]), arriving at $$\begin{aligned} \sqrt{k_y^2-\beta^2}&>&0. \nonumber\end{aligned}$$ Since we are assuming that $|k_y|~\geq~\beta$, this condition is fulfilled for all values of our parameters. Therefore we have an infinite numbers of stationary energies (\[eigenx2\]) that accumulate at $E=0$. In addition, these energies belong to solutions that are normalizable in the sense of our norm (\[normx\]). These solutions can be constructed from the function $\psi_1$ that is defined in (\[psi1bound\]). Upon substitution of our current parameter setting (\[set2\]) and the stationary energies (\[eigenx2\]) we obtain its explicit form $$\begin{aligned} \psi_{1,n}(x) &=& \left[1-i~\sinh(x)\right]^{-\frac{1}{4}\left[1+2n+2\sqrt{k_y^2-\beta^2}-2i\beta\right]} \left[1+i~\sinh(x)\right]^{\frac{1}{4}\left[1+2n-2\sqrt{k_y^2-\beta^2}+2 i \beta\right]} \times \nonumber \\[1ex] && \hspace{-1.5cm} \times~ P_n^{\left(-\frac{1}{2}\sqrt{\left[2+2n+2\sqrt{k_y^2-\beta^2}+2i\beta\right]^2},-\frac{1}{2}\sqrt{\left[ -2n-2\sqrt{k_y^2-\beta^2}+2i\beta\right]^2} \right)}\left[i \sinh(x) \right]. \label{psix2}\end{aligned}$$ where we included an index $n$ to emphasize the bound-state character. After calculating the function $\psi_{2,n}$ from (\[psix2\]), we can determine the solution spinor (\[psinew\]) of our Dirac equation (\[dirac\]) by calculating the functions $\psi_+$ and $\psi_-$. Since the explicit expressions of these functions are too long to be shown here, we restrict ourselves to present their graphs for a particular parameter setting, see figure \[fig2\]. Recall that we do not need to verify normalizability according to (\[normx\]), as this is guaranteed by (\[l2\]) and the fact that $\mu$ is independent of the energy. Concluding remarks ================== In this work we have demonstrated how to construct bound-state solutions of the massless Dirac equation for an energy-dependent potential. Our approach of decoupling the Dirac equation relies on the potential depending on only one of the spatial variables. If this condition is fulfilled, bound states for energy-dependent potentials different from (\[v\]) can be constructed, provided the Schrödinger-type equation (\[eq1\]) renders exactly-solvable. In addition, the sign condition (\[sign\]) for the modified norm must be verified. In most cases, this condition will either give restrictions on the parameters of the potential or dictate that the problem’s domain must be restricted. [99]{} M. Abramowitz and I. Stegun, $"$Handbook of Mathematical Functions with Formulas, Graphs, and Mathematical Tables$"$, (Dover Publications, New York, 1964) C.M. Bender, $"$Introduction to PT-Symmetric Quantum Theory$"$, Contemp. Phys. 46 (2005), 277-292 A.H. Castro Neto, F. Guinea, N.M.R. Peres, K.S. Novoselov and A.K.Geim, $"$The electronic properties of graphene$"$, Rev. Mod. Phys. 81 (2009), 109-162 J.G. Checkelsky, L. Li and N.P. Ong, $"$The zero-energy state in graphene in a high magnetic field$"$, Phys. Rev. Lett. 100 (2008), 206801 C.A. Downing, D.A. Stone and M.E. Portnoi, $"$Zero-energy states in graphene quantum dots and rings$"$, Phys. Rev. B 84 (2011), 155437 J. Formanek, R.J. Lombard and J. Mares, J., $"$Wave equations with energy-dependent potentials$"$, Czechoslov. J. Phys. 54 (2004), 289-315 J. Garcia-Martinez, J. Garcia-Ravelo, J.J. Pena and A. Schulze-Halberg, $"$Exactly solvable energy-dependent potentials$"$, Phys. Lett. A 373 (2009), 3619-3623 P. Ghosh and P. Roy, $"$An analysis of the zero energy states in graphene$"$, Phys. Lett. A 380 (2016), 567-569 J. Gonzalez, F. Guinea and M.A.H. Vozmediano, $"$The electronic spectrum of fullerenes from the Dirac equation$"$, Nucl. Phys. B 406 (1993), 771-794 R.R. Hartmann and M.E. Portnoi, $"$Quasi-exact solution to the Dirac equation for the hyperbolic-secant potential$"$, Phys. Rev. A 89 (2014), 012101 R.R. Hartmann and M.E. Portnoi, $"$Two-dimensional Dirac particles in a Pöschl-Teller waveguide$"$, Sci. Rep. 7 (2017), 11599 R.R. Hartmann, N.J. Robinson and M.E. Portnoi, $"$Smooth electron waveguides in graphene$"$, Phys. Rev. B 81 (2010), 245431 H. Hassanabadi, E. Maghsoodi, R. Oudi, S. Zarrinkamar, and H. Rahimov, $"$Exact solution Dirac equation for an energy-dependent potential$"$, Eur. J. Phys. Plus 127 (2012), 120 C.L. Ho and P. Roy, $"$On zero energy states in graphene$"$, Europhys. Lett. 108 (2014), 20004 A.N. Ikot, H. Hassanabadi, E. Maghsoodi, and S. Zarrinkamar, $"$D-dimensional Dirac equation for energy-dependent pseudoharmonic and Mie-type potentials via SUSYQM$"$, Commun. Theor. Phys. 61 (2014), 436 A.N. Ikot, P. Hooshmand, H. Hassanabadi, and E.J. Ibanga, $"$Dirac equation in minimal length quantum mechanics with energy- dependent harmonic potential$"$, Journal of Information and Optimization Sciences 37 (2016), 101 V. Jakubsky, $"$Spectrally isomorphic Dirac systems: graphene in electromagnetic field$"$, Phys. Rev. D 91 (2015), 045039 Y. Li, $"$Some water wave equations and integrability$"$, J. Nonlin. Math. Phys. 12 (2005), 466-481 R.J. Lombard, J. Mares and C. Volpe, $"$Wave equation with energy-dependent potentials for confined systems$"$, J. Phys. G 34 (2007), 1-11 K. Miyahara and T. Hyodo, $"$Structure of $\Lambda(1405)$ and construction of $\bar{K} N$ local potential based on chiral $SU(3)$ dynamics$"$, Phys. Rev. C 93 (2016), 015201 K.S. Novoselov, A.K. Geim, S.M. Morozov, Y. Zhang, S.V. Dubonos, I.V. Grigorieva and A.A. Firsov, $"$Electric field effect in atomically thin carbon films$"$, Science 306 (2004), 666-669 P. Roy, T.K. Ghosh and K. Bhattacharya, $"$Localization of Dirac-like excitations in graphene in the presence of smooth inhomogeneous magnetic fields$"$, J. Phys.: Condens. Matter 24 (2012), 055301 H. Sazdjian, $"$Relativistic wave equations for the dynamics of two interacting particles$"$, Phys. Rev. D 33 (1986), 3401 A. Schulze-Halberg and P. Roy, $"$Construction of zero-energy states in graphene through the supersymmetry formalism$"$, J. Phys. A 50 (2017), 365205 A. Schulze-Halberg and P. Roy, $"$Quantum models with energy-dependent potentials solvable in terms of exceptional orthogonal polynomials$"$, Ann. Phys. 378 (2017), 234-252 R. Yekken, M. Lassaut and R.J. Lombard, $"$Applying supersymmetry to energy dependent potentials$"$, Ann. Phys. 338 (2013), 195-206 R. Yekken and R.J. Lombard, $"$Energy-dependent potentials and the problem of the equivalent local potential$"$, J. Phys. A 43 (2010), 125301

ArXiv

--- abstract: 'We experimentally study the fluid flow induced by a broad, penetrable barrier moving through an elongated dilute gaseous Bose-Einstein condensate. The barrier is created by a laser beam swept through the condensate, and the resulting dipole potential can be either attractive or repulsive. We examine both cases and find regimes of stable and unstable fluid flow: At slow speeds of the barrier, the fluid flow is stationary due to the superfluidity of the condensate. At intermediate speeds, we observe a non-stationary regime in which the condensate gets filled with dark solitons. At faster speeds, soliton formation completely ceases and a remarkable absence of excitation in the condensate is seen again.' author: - 'P.' - 'C.' title: 'Stationary and non-stationary fluid flow of a Bose-Einstein condensate through a penetrable barrier' --- The fluid flow past an obstacle is one of the most prototypical experiments studying superfluidity. Accordingly, moving an obstacle through a superfluid has been met with considerable interest, both in the context of superfluid helium [@Donnelly1991] as well as in the context of dilute gaseous Bose-Einstein condensates (BECs) [@Leggett2001]. In experiments with BECs, the role of the obstacle can be played by a laser beam that creates a dipole potential for the atoms. By moving a small, strongly repulsive dipole beam through a BEC, evidence for a critical velocity above which superfluidity breaks down was obtained [@Raman1999; @Onofrio2000]. In those experiments, the laser was impenetrable for the atoms, and the diameter of the laser beam was chosen smaller than the size of the BEC. The atoms could therefore flow past the sides of the moving beam. The critical velocity observed in the experiments was much lower than that given by the well-known Landau criterion [@Landau1941]. Vortices can be produced by such narrow obstacles, and indeed they were observed experimentally [@Inouye2001], and in numerical simulations [@vortexnumerics]. Likewise, in the context of optical lattices, heating due to dissipative motion has been used as a tool to study superfluidity in the presence of periodic potentials [@lattice]. In this Letter we consider a situation that is complementary to the experiments described in [@Raman1999; @Onofrio2000]: In our case, the BEC is contained in a narrow, elongated trap and a barrier much wider than the radial extent of the BEC is moved through it (Fig. \[sweepspeed\](a)). The atoms cannot flow past the sides of the barrier, and we use a penetrable barrier allowing atoms to flow *through* the barrier. The barrier is created by the dipole potential of an elliptically shaped laser beam that is swept along the long axis of the elongated BEC. Such a geometry approximates a one-dimensional problem, and it is theoretically expected that in this case solitons play a similar role to the role played by vortices in higher dimensions [@Hakim1996; @Pavloff2002; @Radouani2004; @Astrakharchik2004]. Recently, shedding of solitons has also been discussed in terms of Cerenkov-radiation [@El2006; @Carusotto2006; @Susanto2007]. In this Letter we provide experimental evidence for the existence of several different flow regimes in the wake of the moving barrier. For slow and for fast velocities, the wake behind the moving barrier appears unperturbed. In an intermediate velocity regime solitons appear. This behavior is observed with repulsive barriers as well as with attractive potentials. The starting point for the experiments described in this paper are elongated $^{87}$Rb BECs of about $4.5\cdot10^{5}$ atoms in the $|F=1,m_{F}=-1\rangle$ state, held in a Ioffe-Pritchard type trap with trapping frequencies $\{\omega_{x}/(2\pi),\omega_{yz}/(2\pi)\}=\{7.1,203\}$Hz. The axis of weak confinement (x-axis, see Fig. \[sweepspeed\](a)) is oriented horizontally. Evaporative cooling is performed until no thermal cloud of uncondensed atoms is visible. A laser that is far detuned from the Rb absorption lines is directed along the second horizontal axis (y-axis). Depending on the detuning, it can create an attractive or repulsive dipole potential for the atoms. This dipole beam can be moved along the x-axis at various speeds, inducing flow in the BEC. The dipole beam waist in the vertical z-direction is much larger than the radial extent of the BEC so that the intensity variation of the beam along the z-direction can be neglected. The imaging direction coincides with the direction of the dipole beam. A notch filter prevents the dipole beam from being detected by the camera. As we show in this Letter, solitons can be produced by sweeping the dipole beam through the condensate. The width of these solitons is below our spatial imaging resolution when the BEC is held in the trap. Therefore we employ a fast antitrapping technique to expand the BEC before imaging, similar to the procedure described in [@Lewandowski2003]. The expansion imaging procedure starts with a microwave adiabatic rapid passage, converting the atoms from the $|F=1,m_{F}=-1\rangle$ state to the antitrapped $|F=2,m_{F}=-2\rangle$ state. We then set the magnetic bias field to 48 G and let the cloud expand for 3 ms. During this expansion, the magnetic field is positioned such that the antitrapping potential keeps the cloud from falling under the influence of gravity. The atoms are then imaged in a 100 G bias field using resonant absorption imaging. The antitrapping predominantly leads to a radial expansion, so that the aspect ratio of the BEC changes from $R_{yz}/R_{x}$ = 29 for the trapped BEC to a value of about 1.6 at the end of the expansion time. =3.375in In a first set of experiments, shown in Fig. \[sweepspeed\], we study the flow induced by a weak repulsive potential swept through the BEC at various speeds. The potential was produced by a laser beam with a wavelength of 660 nm and a power of 37 $\mu W$. The beam shape was elliptical with waists $w_{x} = 7.6~\mu m$ and $w_{z} = 25.8~\mu m$. Such a laser produces a weak repulsive potential whose height is about 24% of the chemical potential of the condensate. The healing length in the center of our typical condensates is 0.17 $\mu m$ so that the dipole beam can be considered wide with respect to the healing length even in the direction of the sweep. The dipole beam starts out on the right, outside the BEC, and is then swept towards the left at a constant speed as given in the figure caption. The sweep is stopped at the position indicated by the white arrow in Fig. \[sweepspeed\](c), and the BEC is imaged after antitrapped expansion as described above. We can identify several different regimes, depending on the sweep speed. For very slow sweep speeds, such as in Fig. \[sweepspeed\](c), the part of the BEC through which the dipole beam has been swept appears essentially unaffected, and the only visible effect of the beam is the density suppression that it leaves at its end position. Since the BEC is superfluid, it is not surprising that it remains unaffected by a slowly moving penetrable barrier. At faster sweep speeds (Fig. \[sweepspeed\](d)-(g)), dark notches appear in the wake of the dipole beam. We interpret these notches as the solitons predicted in theoretical studies of the one-dimensional problem [@Hakim1996; @Pavloff2002; @Radouani2004; @Astrakharchik2004; @Susanto2007] (see below for an experimental verification). Some of the notches observed in the wake of the dipole beam show slight bending or nonuniform contrast along the plane, while others are very straight and uniform. In the darkest solitons, we observe a suppression of the central density by about 50%. However, it is conceivable that the true depth of some of the solitons is larger than this because slight misalignments of the imaging direction with respect to the soliton planes can reduce the apparent depth. Interestingly, fewer solitons are observed when the sweep speed is increased beyond 0.9 mm/s (Fig. \[sweepspeed\](h). This effect is accompanied by the appearance of an increasingly wider region of low density directly behind the dipole beam (Fig. \[sweepspeed\](h),(i)). At faster sweep speeds, the condensate appears again unaffected by the obstacle, as seen in Fig. \[sweepspeed\](k). The apparent absence of excitations in the wake of the beam is very striking. Suppression of phonons and solitons at high obstacle velocities was theoretically discussed in [@Law2000; @Pavloff2002; @Radouani2004]. In particular, [@Radouani2004] describes an explanation of related numerical findings by using an analogy to the radiation of capillary-gravity waves in a classical fluid. To quantify the amount of excitation in the BEC, we first calculate the smooth, overall shape of each BEC by considering only the lowest Fourier components of each image. Then the root-mean-square deviation of each original image from the smooth, overall shape is used as a measure of the excitation present in the BEC. To reduce the influence of the inhomogeneous density distribution along the x-axis, we only consider a region in the center of the BEC that is 35 $\mu m$ wide in the x-direction and 70 $\mu m$ wide along the z-direction. The resulting quantity, $\xi$, is plotted versus the barrier speed in Fig. \[sweepspeed\](b). The points at zero velocity are taken from unperturbed BECs in the absence of a dipole beam, and mostly indicate imaging noise. The onset of excitations above a critical velocity of roughly 0.3 mm/s can clearly be seen in this plot. We do not extend this plot to velocities above 0.8 mm/s because at higher speeds the broad low-density region that develops in the wake of the barrier (as seen in Fig. \[sweepspeed\](h), (i)) affects our measure. For comparison, the bulk speed of sound in the center of our BECs is 3 mm/s. In a cigar shaped BEC, averaging over the tightly confined direction leads to a lower speed of sound [@Zaremba1998; @Stringari1998; @Kavoulakis1998]. In our case, this results in a speed of 2.1 mm/s which is still significantly larger than the observed critical velocity. This situation is reminiscent of the observation of a surprisingly low critical velocity in vortex shedding experiments in [@Raman1999; @Onofrio2000]. For a more precise analysis, the complicated fluid flow through the barrier would have to be taken into account. In addition, the nonuniform radial density profile of the BEC is expected to lead to a deviation from the Bogoliubov dispersion relation and to a lower critical velocity, as discussed in [@Fedichev2001]. When a more strongly repulsive potential is used, we observe similar behaviors. However, solitons appear already at lower sweep velocities, and the velocity above which the flow becomes stationary again increases. Also, a more pronounced accumulation of atoms in front of the dipole beam is seen, delimited by a steep wavefront (Fig. \[xsections\]). For weaker potentials the number of solitons in the BEC is reduced. For example, when using dipole beams with an intensity of 25 $\mu W$, often only one or two individual solitons are produced. =3.375in In a second set of experiments we now show that the dark notches in the wake of the dipole beam are not just soundwaves but do indeed behave like solitons. To prove this, we compare the stability of the observed notches with the stability of a simple density suppression in the BEC (Fig. \[evotime\]). To obtain a very clearly visible density suppression, a relatively strong dipole beam with a power of 129 $\mu W$ was swept through half of the BEC. The height of the dipole potential was about 85% of the chemical potential of the condensate so that a deep density suppression was left in the BEC at the end position of the sweep. For clarity this suppression is marked by the white arrow in Fig. \[evotime\](a). The sweep velocity was adjusted to 0.2 mm/s, leading to pronounced soliton formation in the wake of the laser. At the end of the sweep, the dipole beam was switched off, and the BEC was allowed to evolve in trap for a variable evolution time before starting the expansion imaging procedure. Typical images for various evolution times are shown in Fig. \[evotime\]. At very short evolution times (Fig. \[evotime\](a)), the right part of the cloud, through which the dipole beam has been swept, shows stacks of dark notches, and the density suppression produced by the dipole beam at its end position is clearly visible. This density suppression gets filled in by the BEC over a time scale of about 10 ms (Fig. \[evotime\](b)). Considering that the longitudinal speed of sound in our cigar-shaped BEC is 2.1 mm/s, it is very plausible that the density suppression with an initial width of roughly 20 $\mu m$ closes during 10 ms. However, the much narrower dark notches in the right part of the cloud remain visible for much longer times and slowly spread out (Fig. \[evotime\](b)-(f)). After evolution times of over 50 ms they are still visible and have spread over the entire BEC (Fig. \[evotime\](f)). We often observe very straight dark notches at this time. Then their contrast gradually fades and they disappear. These measurements demonstrate the stability of the produced dark notches, suggesting that they are indeed very distinct from sound waves and correspond to the solitons predicted by theory for the one-dimensional problem. =3.375in Our condensates are elongated, but not one-dimensional. Therefore, the question arises which influence the transverse dimension has on the existence of solitons. In previous experiments, individual solitons in a BEC with repulsive interactions have been engineered using phase imprinting [@Burger1999; @Denschlag2000] or wavefunction engineering [@Anderson2001]. They have also been observed in the evolution of a density gap in a BEC produced by a stopped light technique [@Dutton2001]. One important outcome of the phase imprinting studies is that the exact nature of the solitary texture depends on the strength of the radial trapping. The numerical studies given in [@Muryshev2002] reveal that if the chemical potential exceeds ten times the level spacing in the tightly confined direction, i.e. if $\mu /\hbar \omega_{yz}> 10$, phase imprinting produces a nonstationary kink that decays into phonons on a time scale of $1/\omega_{yz}$. If this ratio is below five, the kink decays into a proper dark soliton. Experimentally, solitons produced by phase imprinting were found in experiments with $\mu /\hbar \omega_{yz} = 7$ [@Burger1999]. While our mechanism for soliton formation is not based on phase imprinting, a similar influence of the transverse confinement on the existence of solitons can be expected. In particular, for our BECs $\mu /\hbar \omega_{yz} = 9.7$, which is not too different from the experiments in [@Burger1999]. By changing the laser wavelength to 830 nm we can also study the effects of an attractive potential, in which case the density is increased instead of decreased in the region of the potential. As in the case of repulsive potentials, theories predict soliton formation at intermediate speeds and vanishing excitation at very slow and very fast speeds [@Pavloff2002]. We indeed observe such regimes. However, deviating from our observations with repulsive potentials, we have not found a region of suppressed density behind an attractive dipole laser that would correspond to Fig. \[sweepspeed\](i). Results obtained with an attractive dipole beam are shown in Fig. \[excitation830\]. A dipole beam with a wavelength of 830 nm and a waist of 8.2 $\mu m$ in the x-direction and 31.7 $\mu m$ in the z-direction was used. For the analysis in Fig. \[excitation830\](a), a laser power of 10 $\mu W$ was chosen, leading to a potential depth of about 17% of the chemical potential of the BEC. As before we quantify the excitation by plotting $\xi$ versus the sweep speeds. Since we do not observe a low-density region corresponding to Fig. \[sweepspeed\](i), we can now use $\xi$ as a measure for the excitation over the whole range of velocities studied. Despite the weakness of the perturbing potential, soliton formation again occurs at surprisingly low velocities. For a laser power of 10 $\mu W$, only a few solitons are produced, as shown in Fig. \[excitation830\](b). If the laser power is increased to 19 $\mu W$, the soliton formation can become quite pronounced (Fig. \[excitation830\](c). This demonstrates that not only a repulsive potential, but also an attractive potential can be an efficient tool to produce solitons. One-dimensional theories for the case of attractive potentials predict the onset of soliton formation to occur right at the speed of sound [@Pavloff2002]. However, we see that already at a speed of 1.25 mm/s, below the central speed of sound, the central part of the BEC can be filled with solitons. This discrepancy may hint at an influence of the inhomogeneous density across the condensate in our experiments. =3.375in In conclusion, we have observed stationary and non-stationary fluid flow regimes when a penetrable barrier is moved through a BEC. A slowly moving barrier leads to stationary flow. At intermediate velocities, the flow is non-stationary and is marked by the emergence of solitons in the wake of the barrier. At fast velocities, the soliton shedding stops and the flow appears stationary again. In a one-dimensional geometry, the solitons play a role analogous to vortices present in flow around obstacles in higher dimensions [@Raman1999; @Onofrio2000; @Inouye2001; @vortexnumerics]. Thus our experiments complement these previous studies that have been hallmark experiments for investigating the superfluidity in dilute gaseous BECs. See, e.g., R. J. Donnelly, *Quantized Vortices in Helium II* (Cambridge University Press, Cambridge, 1991), and references therein. See, e.g., A. J. Leggett, Rev. Mod. Phys. [**73**]{}, 307 (2001); S. Ianeselli, C. Menotti, and A. Smerzi, J. Phys. B: At. Mol. Opt. Phys. [**39**]{}, S135 (2006), and references therein. C. Raman *et al.*, Phys. Rev. Lett. [**83**]{}, 2502 (1999). R. Onofrio *et al.*, Phys. Rev. Lett. [**85**]{}, 2228 (2000). L. D. Landau, J. Phys. U.S.S.R. [**5**]{}, 71 (1941). S. Inouye *et al.*, Phys. Rev. Lett. [**87**]{}, 080402 (2001). T. Frisch, Y. Pomeau, and S. Rica, Phys. Rev. Lett. [**69**]{}, 1644 (1992); B. Jackson, J. F. McCann, and C. S. Adams, Phys. Rev. Lett. [**80**]{}, 3903 (1998);T. Winiecki, J. F. McCann, and C. S. Adams, Phys. Rev. Lett. [**82**]{}, 5186 (1999), N. G. Berloff, cond-mat/0412743. See, e.g., O. Morsch and M. Oberthaler, Rev. Mod. Phys. [**78**]{}, 179 (2006) and references therein; B. Wu and J. Shi, cond-mat/0607098. V. Hakim, Phys. Rev. E [**55**]{}, 2835 (1997). P. Leboeuf and N. Pavloff, Phys. Rev. A [**64**]{}, 033602 (2001); N. Pavloff, Phys. Rev. A [**66**]{}, 013610 (2002). A. Radouani, Phys. Rev. A [**70**]{}, 013602 (2004). G. E. Astrakharchik and L. P. Pitaevskii, Phys. Rev. A [**70**]{}, 013608 (2004). G. A. El, A. Gammal, and A. M. Kamchatnov, Phys. Rev. Lett. [**97**]{}, 180405 (2006). I. Carusotto *et al.*, Phys. Rev. Lett. [**97**]{}, 260403 (2006). H. Susanto *et al.*, cond-mat/0701435. H. J. Lewandowski *et al.*, J. Low Temp. Phys. [**132**]{}, 309 (2003) C. K. Law *et al.*, Phys. Rev. Lett. [**85**]{}, 1598 (2000); M. Haddad and V. Hakim, Phys. Rev. Lett. [**87**]{}, 218901 (2001); C. K. Law *et al.*, Phys. Rev. Lett. [**87**]{}, 218902 (2001). E. Zaremba, Phys. Rev. A [**57**]{}, 518 (1998). S. Stringari, Phys. Rev. A [**58**]{}, 2385 (1998). G. M. Kavoulakis and C. J. Pethick, Phys. Rev. A [**58**]{}, 1563 (1998). P. O. Fedichev and G. V. Shlyapnikov, Phys. Rev. A [**63**]{}, 045601 (2001). S. Burger *et al.*, Phys. Rev. Lett. [**83**]{}, 5198 (1999). J. Denschlag *et al.*, Science [**287**]{}, 97 (2000). B. P. Anderson *et al.*, Phys. Rev. Lett. [**86**]{}, 2926 (2001). Z. Dutton *et al.*, Science [**293**]{}, 663 (2001). A. Muryshev *et al.*, Phys. Rev. Lett. [**89**]{}, 110401 (2002).

ArXiv

--- abstract: 'A graph $G$ is said to be $k$-distinguishable if the vertex set can be colored using $k$ colors such that no non-trivial automorphism fixes every color class, and the distinguishing number $D(G)$ is the least integer $k$ for which $G$ is $k$-distinguishable. If for each $v\in V(G)$ we have a list $L(v)$ of colors, and we stipulate that the color assigned to vertex $v$ comes from its list $L(v)$ then $G$ is said to be ${\mathcal{L}}$-distinguishable where ${\mathcal{L}}=\{L(v)\}_{v\in V(G)}$. The list distinguishing number of a graph, denoted $D_l(G)$, is the minimum integer $k$ such that every collection of lists ${\mathcal{L}}$ with $|L(v)|=k$ admits an ${\mathcal{L}}$-distinguishing coloring. In this paper, we prove that $D_l(G)=D(G)$ when $G$ is a Kneser graph.' author: - | Niranjan Balachandran[^1] and Sajith Padinhatteeri[^2],\ Department of Mathematics,\ Indian Institute of Technology Bombay,\ Mumbai, India. title: The List Distinguishing Number of Kneser Graphs --- **Keywords:** List Distinguishing Number, Distinguishing number, Kneser graphs\ 2010 AMS Classification Code: 05C15, 05C25, 05C76, 05C80. Introduction ============ Let $G$ be a graph and let $Aut(G)$ denote the full automorphism group of $G$. By an $r-$vertex labelling of $G$, we shall mean a map $f:V(G) \rightarrow \{1,2, \dots, r\}$, and the sets $f^{-1}(i)$ for $i\in\{1,2\ldots,r\}$ shall be referred to as the color classes of the labelling $f$. An automorphism $\sigma \in Aut(G)$ is said to fix a color class $C$ of $f$ if $\sigma(C) = C$, where $\sigma(C) = \{\sigma(v):v \in C\}$. Albertson and Collins [@AK] defined the distinguishing number of graph $G$, denoted $D(G)$, as the minimum $r$ such that $G$ admits an $r-$ vertex labelling with the property that no non-trivial automorphism of $G$ fixes every color class. An interesting variant of the distinguishing number of a graph, due to Ferrara, Flesch, and Gethner [@MF] goes as follows. Given an assignment ${\mathcal{L}}=\{ L(v) \}_{v \in V(G)}$ of lists of available colors to vertices of $G$, we say that $G$ is ${\mathcal{L}}-$distinguishable if there is a distinguishing coloring $f$ of $G$ such that $f(v) \in L(v)$ for all $v$. The list distinguishing number of $G$, $D_l(G)$ is the minimum integer $k$ such that $G$ is $L-$distinguishable for any list assignment $L$ with $|L(v)|=k$ for all $v$. The list distinguishing number has generated a bit of interest recently (see [@MF; @tree; @interval] for some relevant results) primarily due to the following conjecture that appears in [@MF]: For any graph $G$, $D_l(G)=D(G)$. The paper [@MF], in which this notion was introduced and the conjecture was made, proves the same for cycles of size at least $6$, cartesian products of cycle, and for graphs whose automorphism group is the Dihedral group. The paper [@tree] proves the validity of the conjecture for trees, and [@interval] establishes it for interval graphs. Let $r\ge 2$, and $n\ge 2r+1$. The Kneser graph $K(n,r)$ is defined as follows: The vertex set of $K(n,r)$ consists of all $k-$element subsets of $[n]$; vertices $u,v$ in $K(n,r)$ are adjacent if and only if $u\cap v=\emptyset$. The Distinguishing number of the Kneser graphs is well known (see [@AB]): $D(K(n,r))=2$ when $n \neq 2r+1$ and $r\ge 3$; for $r=2$, $D(K(5,2))=3$, and $D(K(n,2))=2$ for all $n\ge 6$. Our main theorem in this short paper establishes the validity of the list distinguishing conjecture for the family of Kneser graphs. $D_l(K(n,r))=D(K(n,r))$ for all $r\ge 2, n\ge 2r+1$. Before we proceed to the proof of the theorem, we describe the main idea of the proof. We choose randomly (uniformly) and independently for each vertex $v$, a color from its list $L(v)$, and we calculate/bound the expected number of non-trivial automorphisms that fix every color class for this random set of choices. This line of argument features in some other related contexts, for e.g., [@BP; @contuck; @IRL; @RS] most notably under the umbrella of what is called the ‘Motion Lemma’, and some of its variants. These methods however do not work in the cases $r=2$ and $n=6$ or $n=7$, so we settle these cases by different arguments. As it turns out, the case with $r\ge 3$ is much simpler in contrast to the case $r=2$. The rest of the paper is organized as follows. In the next couple of sections, we detail the proof of the conjecture for $r=2$ with $n\ge 8$, and the case for $r\ge 3$ respectively. The cases $r=2, n=6,7$ are dealt with in the appendix. We conclude with a few remarks. List-distinguishing number of $K(n,2)$ {#r=2} ====================================== Firstly, recall the following $D(K(n,2)) = 2$ for $n\ge 6$, and $D(K(5,2))=3$. Since $n\ge 2r+1$, it follows from the Erdős-Ko-Rado theorem that the full automorphism group of $K(n,r)$ is precisely $S_n$, the symmetry group on $n$ symbols. Suppose $n\ge 6$ and suppose $\{L(v)\}_{v \in V(K(n,2))}$ is a collection of color lists of size $2$ corresponding to the vertices of $K(n,2)$. It is simpler to think of these as color lists on the edges of $K_n$. In other words, let $\{L(e)\}_{e \in E(K_n)}$ be lists of colors of size $2$ for the edges of $K_n$. For each edge of $K_n$ we choose a color uniformly and independently at random from its given list of colors. As mentioned in the introduction, we seek to compute the expected number of non-trivial automorphisms that fix all the colors class of this random coloring. Firstly we set up some notation. - If the disjoint cycle decomposition of a permutation $\sigma\in S_n$ consists of $l_i$ cycles of length $\lambda_i,$ for $i=1,2,\ldots,t$ with $\lambda_1<\lambda_2<\cdots<\lambda_t$, then we say $\sigma$ is of type $\Lambda$ where $\Lambda := ( \lambda_1^{l_1}, \lambda_2^{l_2}, \dots, \lambda_t^{l_t})$. Note that $\sum_i l_i\lambda_i = n$. - $CT^{(n)}$ shall denote the set of all permutation types of permutations in $S_n$, i.e., $$CT^{(n)}:=\{ ( \lambda_1^{l_1}, \lambda_2^{l_2}, \dots, \lambda_t^{l_t})\textrm{\ with\ }\sum_i l_i\lambda_i = n\textrm{\ and\ }\lambda_1<\lambda_2<\cdots<\lambda_t\}.$$ - $CT^{(n)}_{\ge r}$, $CT^{(n)}_{r}$ shall denote the sets of all cycle types with minimum cycle length at least $r$, and with minimum cycle length exactly $r$, respectively. - For positive integers $a,b$, we shall denote by $(a,b)$, the g.c.d. of $a$ and $b$. We start with two simple observations. Firstly, note that if a nontrivial automorphism $\sigma$ fixes each of the color classes (as sets) in the random coloring of $E(K_n),$ then for each $e \in E(K_n)$ all the edges in the orbit of $e$ under the action of $\sigma$ get the same color. Also, the probability that $\sigma$ preserves every color class depends only on the cycle type of $\sigma$. For a nontrivial $\sigma\in S_n$, we first obtain an upper bound $P(\sigma)$ on the probability that $\sigma$ fixes all the color classes (as sets) in the random coloring. We set $P(\Lambda) := \displaystyle\sum_{\sigma\textrm{\ of\ type\ }\Lambda} P(\sigma)$, so this gives an upper bound on $P(\Lambda)$ as well. \[lemcore\] Let $\sigma\in S_n$ be a nontrivial permutation of type $\Lambda = ( \lambda_1^{l_1}, \lambda_2^{l_2}, \dots, \lambda_t^{l_t})$. Let $g(\lambda_i) := \left \lfloor \frac{(\lambda_i -1)^2}{2}\right \rfloor$ and $g(\lambda_i, \lambda_j ):= \lambda_i \lambda_j - ( \lambda_1, \lambda_j)$. Furthermore, for $i\le j$ write $l_j^* := l_i(l_i-1) /2$ when $i=j$ and $l_j^* = l_il_j$ for all $j>i$. Then $$P(\sigma) = \frac{1}{2^{\mu}},$$ where $$\mu = \sum\limits_{i=1}^t \left(g(\lambda_i)l_i +\sum\limits_{j\ge i}^t g(\lambda_i, \lambda_j )l_j^* \right).$$ Consequently, for $\Lambda \in CT^{(n)},$ $$P(\Lambda) \le n!2^{-\mu} \prod_{i=1}^t \frac{\lambda_i^{-l_i}}{(l_i)!}.$$ If $\sigma$ is an automorphism that fixes every color class then as observed earlier, for each edge $e$, every edge in $\{e, \sigma(e), \sigma^2(e), \dots, \sigma^k(e) \}$ has the same color. Here, the integer $k\ge 1$ is the smallest integer satisfying $\sigma^{k+1}(e) = e.$ Let $\sigma=C_1C_2\cdots C_u$ be its disjoint cycle decomposition. Writing $C=(12\cdots r)$, the prior observation implies that for each $1\le i\le \left \lfloor r/2\right \rfloor$, the set of edges $\{(1,i), (2,i+1),\ldots, (r,i+r-1)\}$ is monochrome, where the addition is performed modulo $r$. Moreover, as these form a pairwise disjoint partition of the edges of the clique on $C$, the probability that all these sets of edges are monochrome is at most $2^{-g(r)}$ where $g(r) = \left \lfloor\frac{(r -1)^2}{2}\right \rfloor$ as in the statement of the lemma. Now, by a similar argument, if $\sigma$ fixes every color class then the set of edges between the vertices of two disjoint cycles $C_i$ and $C_j$ of size $r, s$ respectively is partitioned into monochrome sets of size equal to the least common multiple of $r,s$. Hence the probability that such an event occurs is $2^{-{g(r,s)}}$ with $g(r,s)=rs-(r,s)$ as in the statement of the lemma. Moreover, these events (i.e., partitioning of the edges within each cycle $C_i$ and also across a pair of cycles $C_i,C_j$) are pairwise independent, and since $\sigma$ is of type $\Lambda$, it follows that the probability that $\sigma$ fixes every color class is at most $2^{-\mu}$, where $\mu$ is as described in the statement of the lemma. As for $P(\Lambda),$ we use the first part of the lemma in conjunction with the observation that there are $n!\prod_{i=i}^t \frac{\lambda_i^{-l_i}}{(l_i)!}$ permutations of type $\Lambda$ in $S_n.$ \[probzero\] If $\sigma \in S_n$ is of type $\Lambda$, then the bound in the preceding lemma occurs if and only if all the lists are identical. If in fact, for some $i$, the lists for the edges $\{(1,i), (2,i+1),\ldots, (r,i+r-1)\}$ has empty intersection, then $P(\sigma)$ is zero. A similar remark about the lists of edges between the vertices of disjoint cycles $C_i,C_j$ also holds. If $\Lambda, \Gamma$ are cycle types in $CT^{(n)}$ and $CT^{(n-\lambda_1)}$ respectively, we say that $\Lambda$ ‘extends’ $\Gamma$ if $$\Lambda = ( \lambda_1^{l_1}, \lambda_2^{l_2}, \dots, \lambda_t^{l_t})\textrm{\ and\ } \Gamma = ( \lambda_1^{l_1-1}, \lambda_2^{l_2}, \dots, \lambda_t^{l_t}).$$ Note that $$P(\Lambda) = R_{\lambda_1}(\Lambda) P( \Gamma)$$ where $$R_{\lambda_1}(\Lambda) = \frac{n(n-1)(n-2) \dots (n-\lambda_1+1)}{\lambda_1 l_1 }2^{-g(\lambda_1) - g(\lambda_1, \lambda_1 )(l_1-1) - \sum\limits_{j\ge 2}^t g(\lambda_1, \lambda_j )l_j}.$$ This is a straightforward consequence of lemma \[lemcore\]. \[r&lt;1\] Let $\Lambda = ( \lambda_1^{l_1}, \lambda_2^{l_2}, \dots, \lambda_t^{l_t})$ be a cycle type in $CT^{(n)}$. Then for $n\ge 9$ $$\begin{aligned} R_{\lambda_1}(\Lambda) &<&2^{-n \lambda_1/7}\textrm{\ if\ }\lambda_1\ge 2,\\ R_{1}(\Lambda) &\le &\frac{n}{2(n-2)}\textrm{\ and\ equality\ is\ achieved\ precisely\ if\ }\Lambda=(1^{n-2},2),\\ R_{1}(\Lambda) &\le &\frac{n}{4(n-3)}\textrm{\ if\ }\Lambda\neq(1^{n-2},2).\end{aligned}$$ Firstly, suppose $\lambda_1\ge 2$. Set $s = \frac{ \lambda_1( 2n+ 7 \lambda_1)}{14}$; observe that $$\begin{aligned} \log n<\frac{5n}{14}=\frac{n}{2}- \frac{s}{\lambda_1} - \frac{\lambda_1}{2} \end{aligned}$$ holds for $n\ge 9$. Here $\log$ is the logarithm to the base $2$. As $\lambda_1\ge 2$, we have $\sum_{i\ge 1} l_i\le n/2$, so we may write $$\begin{aligned} & \frac{n}{2}- \frac{s}{\lambda_1} - \frac{\lambda_1}{2}< n - \sum\limits_{j\ge 1} l_j - \frac{s}{\lambda_1} -\frac{\lambda_1}{2} + \frac{1}{\lambda_1} \left( \log(l_1)+ \log(\lambda_1)\right ). \nonumber $$ Since $n= \sum_i \lambda_i l_i$ we have (by rearranging the terms) $$\begin{aligned} & n - \sum\limits_{j\ge 1} l_j - \frac{s}{\lambda_1} -\frac{\lambda_1}{2} + \frac{1}{\lambda_1} \left( \log(l_1)+ \log(\lambda_1)\right ) \label{eq1}\\ &= \sum_i \lambda_i l_i - \left(l_1 + \frac{1}{\lambda_1}\sum\limits_{j\ge 2} \lambda_1 l_j \right)- \frac{s}{\lambda_1} -\frac{\lambda_1}{2} + \frac{1}{\lambda_1} \left( \log(l_1)+ \log(\lambda_1)\right ) \label{eq2}\\ &= \frac{-s}{\lambda_1}+ \frac{\lambda_1}{2} + l_1 \lambda_1 -\lambda_1 -l_1 + \sum\limits_{j\ge 2}^t \lambda_jl_j + \frac{1}{\lambda_1} \left( \log(l_1)+ \log(\lambda_1) - \sum\limits_{j\ge 2}^t \lambda_1 l_j \right ) \label{eq3}\\ & = \frac{-s}{\lambda_1}+ \frac{\lambda_1}{2} + l_1 \lambda_1 -\lambda_1 -l_1 + \frac{1}{\lambda_1} \left( \sum\limits_{j\ge 2}^t (\lambda_1\lambda_j - \lambda_1)l_j + \log(l_1)+ \log(\lambda_1) \right ). \label{eq4}\end{aligned}$$ To elaborate, we rewrite $n=\sum_j \lambda_j l_j$ in and write $\sum_j l_j$ as $l_1 + \frac{1}{\lambda_1}\sum\limits_{j\ge 2} \lambda_1 l_j$ to get ; results from by rearranging terms, writing $-\frac{\lambda_1}{2}$ as $\frac{\lambda_1}{2}-\lambda_1$ and also isolating the term $\lambda_1 l_1$ from $\sum_j \lambda_j l_j$, and finally is again a suitable rearrangement of . Since $\lambda_1 \ge ( \lambda_1, \lambda_j)$, we have for $n\ge 9$, $$\log n< \frac{-s}{\lambda_1}+ \frac{\lambda_1 -2}{2} + (\lambda_1 -1)(l_1-1) + \frac{1}{\lambda_1} \left( \sum\limits_{j\ge 2}^t g(\lambda_1, \lambda_j )l_j + \log(l_1)+ \log(\lambda_1) \right )$$ Since $g(x)\le (x^2-2x)/2$ we have $$\lambda_1 \log n < -s+ g(\lambda_1) + g(\lambda_1, \lambda_1 )(l_1-1) + \sum\limits_{j\ge 2}^t g(\lambda_1, \lambda_j )l_j + \log(l_1)+ \log(\lambda_1)$$ and thus, $$n^{\lambda_1}< 2^{-s} l_1 \lambda_1 2^{g(\lambda_1) + g(\lambda_1, \lambda_1 )(l_1-1) + \sum\limits_{j\ge 2}^t g(\lambda_1, \lambda_j )l_j}$$ which achieves the first part of the lemma since $n\lambda_1/7 < \frac{ \lambda_1( 2n+ 7 \lambda_1)}{14}$. When $\lambda_1=1$, then note that $$R_1(\Lambda)\le\frac{n}{l_1}2^{(\sum\limits_{j\ge 2} (1-\lambda_j)l_j)}=\frac{n}{l_1}2^{L-n},$$ where $L=\sum_{j\ge 1} l_j$. Since $\lambda_2\ge 2$, it follows that $n-L\ge (n-l_1)/2$, so we have $$R_1(\Lambda)\le\frac{n}{l_12^{(n-l_1)/2}}.$$ It follows by elementary calculus (for instance) that since the function $h(x)=x2^{(n-x)/2}$ defined on $[1,n-2]$ achieves its minimum value of $2(n-2)$ at $x=n-2$, we have $$R_1(\Lambda)\le\frac{n}{2(n-2)}$$ as required.\ If $\Lambda$ corresponds to a permutation type of a non-trivial permutation and $\Lambda\ne(1^{n-2},2)$ then arguing as before, we observe that in that case, $n-L\ge 4$, and among such permutation types, $R_1(\Lambda)$ is maximum for $\Lambda=(1^{n-3},3)$, and for this $\Lambda$, $R_1(\Lambda)\le\frac{n}{4(n-3)}$. This completes the proof. Set $f(n) := \sum\limits_{\sigma \in S_n }P( \sigma)$. Let $f_{\ge i}(n) $ denote the corresponding sum over all those permutations $\sigma\in S_n$ in which every cycle has size at least $i$. Also set $P(n) := P(\Lambda =n).$ \[f(n)&lt;1\] For any $n$, $$f(n) < \frac{n}{2(n-2)} f(n-1) + \sum\limits_{i=2}^{\left \lfloor n/2\right \rfloor} 2^{-ni/7}f_{\ge i}(n-i) + P(n).$$ Observe that any cycle type in $CT^{(n)}_{i}$ is an extension of a unique cycle type in $CT^{(n-i)}_{\ge i}$. Also, since for a fixed cycle type $\Lambda \in CT^{(n)},$ there are exactly $N(\Lambda) = n! \prod_{i=1}^t \frac{\lambda_i^{-l_i}}{(l_i)!}$ permutations of type $\Lambda$, we have $$\begin{aligned} f(n) =&\sum\limits_{\Lambda \in CT^{(n)} }N(\Lambda)P(\Lambda) = \sum\limits_{i=1}^{\left \lfloor n/2 \right \rfloor} \sum\limits_{\Lambda \in CT^{(n)}_{i} }N(\Lambda) P(\Lambda) + P(n) \nonumber\\ =& \sum\limits_{i=1}^{\left \lfloor n/2 \right \rfloor} \sum\limits_{\Lambda \in CT^{(n)}_{\ge i} }N(\Lambda) R_i(\Lambda) P(\Lambda) +P(n) \nonumber \label{eqth2} \end{aligned}$$ By the bounds for $R_{\lambda_1}(\Lambda)$ from lemma \[r&lt;1\], we have $$\begin{aligned} f(n) < & \sum\limits_{\Lambda \in CT^{(n-1)} } \frac{n}{2(n-2)} N(\Lambda) P(\Lambda) + \sum\limits_{i=2}^{\left \lfloor n/2 \right \rfloor}\sum\limits_{\Lambda \in CT^{(n)}_{\ge i} } 2^{-in/7} N(\Lambda) P(\Lambda) \nonumber\\ <&\ \ \frac{n}{2(n-2)} f(n-1) + \sum\limits_{i=2}^{\left \lfloor n/2\right \rfloor} 2^{-ni/7}f_{\ge i}(n-i) + P(n).\end{aligned}$$ $f(n)<1$ for all $n\ge 8$. In fact, for all $n\ge 8$, $$f(n) \le \frac{kn^2}{2^{n}}$$ for some absolute constant $k$. Consequently, we have $D_l(K(n,2))=2$ for $n\ge 8$. This proof is by induction on $n.$ It is straightforward, though a little tedious to check $f(8) \approx 0.874 < 1 $ by calculating $\sum\limits_{\Lambda \in CT_8}N(\Lambda)P(\Lambda)$ directly; we also check that $f_{\ge 4}(5),$ $f_{\ge 3}(6)$ and $f_{\ge 2}(7)$ are strictly less than one. Furthermore, $$\begin{aligned} f_{\ge 2}(7) =& \frac{7!}{2^2 2! 3}2^{-12} + \frac{7!}{12}2^{-2-4-11} + \frac{7!}{10}2^{-8-9} + \frac{7!}{7}2^{-18} \approx 0.061. \label{7=n>2} \\ f_{\ge 3}(6) =&\frac{6!}{3^2 2!}2^{-2-2-6} + \frac{6!}{6}2^{-12}\approx 0.0683.\label{6=n>3} \\ f_{\ge 4}(5) = &\frac{5!}{5}2^{-8}\nonumber \approx 0.0937.\label{5=n>4}\end{aligned}$$ Also, $P(n) = \frac{n!}{n}2^{-\lfloor \frac{(n-1)^2}{2} \rfloor}.$ Since $P(n)$ is monotonically strictly decreasing for $n\ge 3$, we may bound $P(n)<P(9) = 8! 2^{-32} \approx 0.0000093.$ Assume $f(k) < 1$ for $8 \le k \le n-1.$ By lemma \[f(n)&lt;1\] we have $$f(n)\le \frac{n}{2(n-2)} f(n-1) + \sum\limits_{r=2}^{\left \lfloor n/2\right \rfloor} 2^{\frac{-nr}{7}}f_{\ge r}(n-r) + P(n)$$ so $$f(n) \le \frac{n}{2(n-2)} +\sum\limits_{r=2}^{\lfloor n/2\rfloor} 2^{\frac{-nr}{7}} + 0.0000093.$$ Since $S_n < (2^{n/7}(2^{n/7}-1))^{-1}<0.3$ for $n\ge 9$, we have $f(n) <1$ when $n \ge 9.$ For the exponentially decaying upper bound, we again proceed to do so inductively. The only difference is that this time, we are slightly more careful with our bounds, though we do not attempt to optimize for the constant $k$. We shall show that $f(n)\le 20n^2/2^n$ holds for all $n\ge 8$. It is easy to see that this statement holds for $n\le 11$ since $20n^2/2^n$ is greater than $1$ for all these values of $n$. In computing $f(n)$ through the application of lemma \[f(n)&lt;1\], we isolate the terms arising from permutations of type $(1^{n-2},2)$ and note that their contribution to the sum $f(n)$ is precisely $n(n-1)/2^{n-1}$. For the remaining $\Lambda$ with $\lambda_1=1$, as observed in lemma \[r&lt;1\], we have $R_1(\Lambda)\le n/4(n-3)$. Piecing these together, and by induction, we have $$f(n) < \frac{n^2}{2^{n-1}} + \frac{n}{4(n-3)}\frac{20(n-1)^2}{2^{n-1}} + \frac{20n^2}{2^n}\sum\limits_{i\ge 2} \left(\frac{2}{2^{n/7}}\right)^i+ \left(\frac{4n}{2^n}\right)^n.$$ Now, our choice of constants gives us that for $n\ge 12$, $\left(\frac{(n-1)^2}{2n(n-3)}+\frac{1}{2^{n/7}(2^{n/7}-2)}\right) < 0.8$, so, the right hand side of the expression above is at most $18n^2/2^n + \left(\frac{4n}{2^n}\right)^n < 19n^2/2^n$, and the induction is complete. **Remark:** As observed in the proof, $f(n)\ge \displaystyle\frac{\binom{n}{2}}{2^{n-2}}$, so we actually have $f(n)=\Theta(n^2/2^n)$. List distinguishing number of $K(n,r)$ when $r\ge 3$ {#r>2} ==================================================== In this section we show that $D_l(K(n,r))=2$ for $r\ge 3, n\ge 2r+1$ holds with positive probability for a random coloring, where as before the random coloring is obtained by choosing for each vertex $v$, a color uniformly from its list, and independently across the vertices. Recall that the vertices of $K(n,r)$ correspond to $r$-subsets of $[n]:=\{1,2\ldots,n\}$ and vertices $u,v\in V(K(n,r))$ are adjacent if and only if $u\cap v=\emptyset$. As before, suppose that the vertex $v$ is assigned a color list of size $2$. \[2cycle\] Consider the random coloring of $G= K(n,r), r \ge 3$. Let $\sigma$ be a nontrivial permutation of type $\Lambda$ that fixes every color class. Then $$P(\sigma) < \frac{1}{2^m} \textrm{\ where\ } m= {n-2 \choose r-1}.$$ Without loss of generality suppose $\sigma$ has the cycle $(1,2, \hdots, t)$ for some $2\le t \le n.$ Let $v$ be a vertex corresponding to a set containing the element $1$, but not the element $2$ in $[n]$. Then since $2\in\sigma(v)$ it follows that $v\ne\sigma (v)$. Therefore, if $\sigma$ fixes every color class, since $v$ and $\sigma(v)$ for each vertex $v$ are assigned the same color, each set of size $r$ containing $1$ but not containing $2$ must get mapped to a distinct subset, not of the same form, and each of these pairs of vertices are monochrome pairs. The probability of the aforementioned event is precisely $2^{-m}$ as stated in the lemma. If $r \ge 3$ and $n > 2r+1,$ then $D_l(K(n,r))= 2.$ Consider the random coloring of $K(n,r)$ as described earlier. By Lemma \[2cycle\], the probability that there exists a non-trivial automorphism that fixes every color class under this random coloring is at most $$\frac{|Aut((K(n,r))|}{2^m}=\frac{n!}{2^{\binom{n-2}{r-1}}}\le\frac{n!}{2^{\binom{n-2}{2}}}$$ since $r\ge 3.$ It is straightforward to check that the last expression is less than $1$ for $n \ge 9.$ Since $n \ge 2r+1$ and $r\ge3$ the remaining cases are $n=7$ and $n=8.$ In these cases we look at the corresponding expression a little closer. We bifurcate the set of non trivial automorphisms into two categories: We say a permutation $\sigma\in S_n$ is of Category I if all the cycles in the cycle decomposition of $\sigma$ have size at most $2$, otherwise we say $\sigma$ is a category II permutation. For $n=7$ there are $\frac{7!}{2.5!} + \frac{7!}{2^2.2!.3!} + \frac{7!}{2^3.3!} = 231$ nontrivial permutations in Category I and $4808$ permutations in Category II. Let $E_I$ and $E_{II})$ denote the events that a nontrivial automorphism of Category I, Category II respectively, fixes every color class, then $$P(E) = P(E_I) + P(E_{II}) < \frac{231}{2^{10}} + \frac{4808}{2^{20}} < 1.$$ Similarly when $n=8$ $$P(E) < \frac{973}{2^{15}} + \frac{39346}{2^{30}}<1$$ and this completes the proof. Concluding Remarks ================== - The lone case of $r=2, n=5$ has not been considered in the preceding sections. In this case, in fact, $D(K(5,2))=3$. It is a simple calculation (again using a randomized coloring) to show that in this case too, $D_l(K(5,2))=3$. We omit the (simple) details. - While we were content with showing that the with positive probability, a random list-coloring of the vertices of $K(n,r)$ (for $r\ge 3$) actually is distinguishing, it is easy to see that in fact, these are asymptotically almost sure events, like in the case of $r=2$. In particular, these give very efficient randomized algorithms for distinguishing list colorings. - Our methods may possibly also extend to yield other results of the same kind. An instructive instance would be to consider an $r$-fold cartesian product of complete graphs; the distinguishing number of cartesian products of complete graphs was shown to be $2$ in [@KX] though it is not yet known if the list distinguishing number also equals $2$, and we believe that the same ideas may turn out to be useful there (though the computations can get more complicated). - As observed in remark \[probzero\], the expressions for the probabilities as calculated in most sections are non-zero only if certain lists are identical, otherwise the probabilities are in fact much lower. We believe that the following strengthening of the List Distinguishing Conjecture is also true: For a graph $G$, with a collection of equal sized (size $k$, say) lists ${\mathcal{L}}=\{L_v|v\in V\}$, if $p({\mathcal{L}})$ denotes the probability that a random coloring (obtained by choosing for each vertex, a color from its list uniformly and independently) admits a non-trivial automorphism which preserves all the color classes, then $p({\mathcal{L}})$ is maximized when the lists are identical. Our results, while not quite proving this stronger statement exactly (since computing these probabilities exactly would be cumbersome) in fact proves that the expected number of non-trivial automorphisms that fix all the color classes is actually maximized when the lists are identical. [AAAA]{} M. O. Albertson and K. L. Collins, Symmetry Breaking in Graphs, *Electron. J. Combin.*, [**3**]{} (1996), \#R18. Michael O. Albertson and Debra L. Bouti, Using Determining sets to Distinguish Kneser Graphs, *Electron. J. Combin.*, [**14**]{} (2007), \#R20. Niranjan Balachandran and Sajith Padinhatteeri, $\chi_D(G), |Aut(G)|$, and a variant of the Motion lemma, arXiv:1505.03396v1, to appear in *Ars Math. Contemp.* M. Condor, and T. Tucker, Motion and distinguishing number two, *Ars Math. Contemp.*, [**4**]{} (2011), 63-72. Wilfried Imrich, Rafał Kalinowski, Florian Lehner and Monika Pilśniak, Endomorphism breaking in Graphs, *Electron. J. Combin.*, [**21(1)**]{} (2014), \#P 1.16. Michael Ferrara, Breeann Flesch and Ellen Gethner, List-Distinguishing Coloring of Graphs, *Electron. J. Combin.*, [**18**]{} (2011), \#P161. Michael Ferrara, E. Gethner, S. Hartke, D. Stolee and P. Wenger, List-Distinguishing Parameters of Trees, *Discrete Appl. Math.*, [**161**]{} (2013), 864-869. Poppy Immel and Paul S. Wenger, The List Distinguishing Number Equals the Distinguishing Number for Interval Graphs, *http://arxiv.org/abs/1509.04327v1*. S. Klavzar and X. Zhu, Cartesian powers of graphs can be distinguished by two labels, *European J. of Combin.*, [**28**]{} (2007), 303-310. Alexander Russell and Ravi Sundaram, A Note on the Asymptotics and Computational Complexity of Graph Distinguishability, *Electron. J. Combin.*, [**5**]{} (1998), \#R23. Appendix: $D_l(K(n,2))=2$ for $n=6,7$ {#appendix-d_lkn22-for-n67 .unnumbered} ===================================== Consider a graph $G$ with a collection of lists ${\mathcal{L}}=\{L(e)|e\in E(G)\}$. By the **color palette of a vertex $v$** in a graph $G$, we mean the multi-set of colors assigned to the incident edges of $v$ in a list coloring of the edges of $G$. A monochromatic path $P$ shall refer to a maximum sized path in $G$ such that $\mathop{\cap}\limits_{e\in E(P)} L(e) \neq \emptyset$. We use $l(P)$ to denote the length of $P$ and $|P|$ to denote the number of vertices in $P$. Hereafter the word path shall also refer only to monochromatic paths. \[random\] Let $n\ge 6$ Suppose we have a collection of lists ${\mathcal{L}}=\{L(e)|e\in E(K_n\}$ of size $2$. If there is no monochrome path in $K_n$ of length two then there is a distinguishing list coloring of the edges of $K_n$ from the lists in ${\mathcal{L}}$. For each edge $e \in E(K_n)$, choose a random coloring of the edges as before. Observe that for any color fixing automorphism $\sigma$, the color pallettes of $u$ and $\sigma(u)$ are the same. More over the edges $uv$ and $\sigma(u)\sigma(v)$ have the same color. The probability that there exist vertices $u,v$ with the same color pallettes is at most $\binom{n}{2}/2^{n-2}$ since for any color incident at vertex $u$, and not on the edge $uv$, there is at most one edge incident with $v$ that can have that color in its list, by the hypothesis. Now, for $\binom{n}{2}/2^{n-2}<1$ for $n\ge 6$, so we are through. By the virtue of lemma \[random\], we may assume that $K_n$ contains a monochromatic path of length at least two. We introduce some further terminology. $K_n$ shall be the complete graph on the vertex set $[n]$, and $G'$ we shall denote the complete subgraph on $[n]\setminus V(P)$. The edges between $G'$ and $P$ will be referred to as crossing edges. $e_{ij}$ is the edge between vertex $i$ and $j$ and $c_{ij}$ shall denote the color assigned to the edge $e_{ij}.$ The available common color on the edges of $P$ is denoted $c_1.$ Without loss of generality we assume $V(P) = \{ 1, 2, \hdots, |P|\}.$ \[n=6,2\] $D_l(K(6,2))= 2$. As observed before, we may assume that if $P$ is a monochromatic path, then $|P| \ge 3$. Consider the following cases - $|P| =6:$ Color $E(P)$ using $c_1,$ avoid $c_1$ from all other edges except $e_{24}$ and $e_{35}$. Also ensure that $c_{24} \neq c_{35}$. This coloring is distinguishing since the color class $c_1$ is fixed (as a set) only by two maps - the identity and the permutation $\sigma=(16)(25)(34)$. But since $\sigma(e_{24}) = e_{35}$, and they are colored differently, $\sigma$ does not fix every color class. - $|P|=5$: Assign $c_1$ to $E(P)$ and avoid $c_1$ from all other edges. Again, ensure that $c_{16} \neq c_{56}$; $G'$ consists of the lone vertex $6$ and $|P|=5$, so $c_1$ does not appear on the lists of both $e_{16}$ and $e_{56}$, so this arrangement is possible. By our choices, no crossing edge is colored $c_1$, so the monochrome set of edges colored $c_1$ is again precisely $P$. This coloring is distinguishing for very similar reasons as above. - $|P|= 4$: Assign $c_1$ to $E(P)$ and avoid $c_1$ from all other edges. Ensure that $c_{45} \neq c_{14}, c_{45} \neq c_{16}$, and $c_{45} \neq c_{46}$; again, these arrangements are possible by the maximality of $P$ as none of the crossing edges from the end vertices of $P$ contain $c_1$ in the given lists. It is now easy to check that this coloring is distinguishing. - $|P|= 3$: We start by coloring the edges of $P$ using $c_1$. Color the edges $e_{16}$ and $e_{46}$ arbitrarily from their lists, and for the remaining edges, impose a restriction on the color that needs to be assigned to it as in Table \[colortable6.1\] below. Again, note that the maximality of $P$ ensures that all these avoidances are permissible. Edges Restriction on the color choice -------------------------------------- --------------------------------- -- $e_{12}, e_{23} $ Assign $c_1$ $e_{24},e_{25},e_{26},e_{13},e_{45}$ Avoid $c_1$ $e_{34},e_{35},e_{36},e_{14},e_{15}$ Avoid $c_{16}$ $e_{56}$ Avoid $c_{46}$ : Coloring Scheme when $n=6$[]{data-label="coloring table"} \[colortable6.1\] To see why this is distinguishing, suppose $\sigma$ is an automorphism that fixes each of these color classes. By the avoidance choices, the only edges that are colored $c_1$ are the edges of $P$. Our choices also ensure that the pallettes of vertices $1$ and $3$ are different, so it follows that $\sigma $ fixes $1,2,3$. Since $c_{46}\ne c_{56}$, $\sigma\neq (45), (456),(465)$ and since $c_{14}, c_{15}\ne c_{16}$, $\sigma\neq (46), (56)$, so $\sigma$ is the identity map on $[6]$. $D_l(K(7,2)) = 2.$ We proceed as we did in the theorem \[n=6,2\] and consider the following cases. - When $|P| \ge 5$ the coloring scheme is similar to that of $|P| \ge 4$ in theorem \[n=6,2\]. If $|P| = 7,$ assign $c_1$ to all the edges of $E(P),$ ensure $c_{24} \neq c_{46}$ and avoid $c_1$ from all other edges. For $|P| = 6,$ assign $c_1$ to $E(P),$ ensure $c_{24} \neq c_{35}$ and avoid $c_1$ from all other edges. For $|P| = 5,$ assign $c_1$ to $E(P),$ ensure $c_{56} \notin \{c_{16}, c_{17}, c_{57} \}$ and avoid $c_1$ from all other edges. The proofs that these give distinguishing colorings is similar to the arguments that appear in theorem \[n=6,2\], so we omit those details. - $|P| = 4:$ Assign $c_1$ to all the edges of $P$, and ensure that $c_{56}\neq c_1$. Also, avoid $c_{56}$ from $e_{67}$ and $e_{57}.$ Further ensure $c_{17} \neq c_{47}$ and $c_{16}\ne c_{15}$ from all other edges. As always, avoid $c_1$ on any crossing edge.\ Our choice of coloring guarantees that any automorphism $\sigma$ that fixes every color class necessarily maps the set $\{1,2,3,4\}$ and $\{5,6,7\}$ into themselves respectively. Since $c_{57}, c_{67}\ne c_{56}$, $\sigma(7)=7$ and since $c_{17} \neq c_{47}$ it follows that $\sigma$ fices each of $1,2,3,4$. Finally, since $c_{16}\ne c_{15}$, $\sigma$ fixes $5,6$ as well. - $|P|= 3$: Color the edges of $P$ using $c_1$. Color the edges $e_{16}$ and $e_{46}$ arbitrarily from their lists, and for the remaining edges, we consider two sub cases and in each sub case we impose a different type of restriction on the color that needs to be assigned to the edges; see Tables \[colortable7.1\] and \[colortable7.2\] for the details on the restrictions. Again, note that the maximality of $P$ ensures that all these avoidances are permissible. [Sub case 1.]{} $c_1 \in L(e_{27}).$ Edges Restriction on the color choice ---------------------------------------------- --------------------------------- -- $e_{12}, e_{23} $ Assign $c_1$ $e_{24},e_{25},e_{26},e_{13},e_{45},e_{27}$ Avoid $c_1$ $e_{34},e_{35},e_{36},e_{14},e_{15}, e_{37}$ Avoid $c_{16}$ $e_{47},e_{56}, e_{67}$ Avoid $c_{46}$ $e_{17}$ Avoid $c_{15}$ $e_{37}$ Avoid $c_{36}$ : Coloring Scheme when $n=7$[]{data-label="colortable7.1"} [Sub case 2.]{} $c_1 \notin L(e_{27}).$ Edges Restriction on the color choice ----------------------------------------------------------- --------------------------------- -- $e_{12}, e_{23} $ Assign $c_1$ $e_{24},e_{25},e_{26},e_{13},e_{45},e_{47},e_{57},e_{67}$ Avoid $c_1$ $e_{34},e_{35},e_{36},e_{14},e_{15}, e_{37}$ Avoid $c_{16}$ $e_{56}$ Avoid $c_{46}$ $e_{17}$ Avoid $c_{15}$ $e_{27}$ Avoid $c_{24}$ $e_{37}$ Avoid $c_{36}$ : Coloring Scheme when $n=7$[]{data-label="colortable7.2"} In sub case 1, Firstly we observe that by our choices, we in fact have $c_{37}\neq c_{36}$ because by the hypothesis of sub case 1, $L(e_{36})$ and $L(e_{37})$ cannot both have the color $c_{16}$, otherwise $|P|\ge 4$. Further, the hypothesis of sub case 1 guarantees that $c_1$ is not present in the lists of $e_{47}$ and $e_{67}$, so our avoidances in fact give us that $P$ is the unique path of length $2$ colored $c_1$. Since the pallettes of $1$ and $3$ are different, it follows that any $\sigma$ that preserves all the color classes must necessarily fix $1,2,3$. Now we first show that $7$ is also fixed. Indeed, if $\sigma(6)=7$, then $\sigma(e_{36})=e_{37}$ but by choice, these are colored differently. Similarly, $\sigma(5)\neq 7$ since $c_{15}\neq c_{17}$. Now, if $\sigma(4)=7$, then $\sigma(6)=5$ as a consequence of our color avoidances. But then $c_{15}\ne c_{16}$, so this shows that $\sigma$ fixes $7$ as well. Finally, by following similar arguments as in theorem \[n=6,2\], it follows that $\sigma$ fixes $4,5,6$ as well, so $\sigma $ is the identity map. In sub case 2, the crucial difference is in the color choice of $e_{27}$. The color avoidance here ensures that $\sigma(4)=7$ or $\sigma(7)=4$ is not possible since $c_{24}\ne c_{27}$. The rest of the proof is similar to sub case 1. [^1]: niranj (at) math.iitb.ac.in, Supported by grant 12IRCCSG016, IRCC, IIT Bombay [^2]: sajith(at)math.iitb.ac.in, Supported by grant 09/087(0674)/2011-EMR-I, Council of Scientific & Industrial Research, India

ArXiv

--- abstract: 'A computational approach is developed for the design of continuous low thrust transfers in the planar circular restricted three-body problem. The transfer design method of invariant manifolds is extended with the addition of continuous low thrust propulsion. A reachable region is generated and it is used to determine transfer opportunities, analogous to the intersection of invariant manifolds. The reachable set is developed on a lower dimensional section and used to design transfer trajectories. This is solved numerically as a discrete optimal control problem using a variational integrator. This provides for a geometrically exact and numerically efficient method for the motion in the three-body problem. A numerical simulation is provided developing a transfer from a $ L_1 $ periodic orbit in the Earth-Moon system to a target orbit about the Moon.' author: - 'Shankar Kulumani[^1], Taeyoung Lee[^2]' bibliography: - 'library.bib' title: 'Systematic Design of Optimal Low-Thrust Transfers for the Three-Body Problem' --- Introduction {#sec:introduction} ============ Designing spacecraft trajectories is a classic and ongoing topic of research. With the current fiscal constraints there is an increased focus on technologies with a critical impact on mass. Optimal expenditure of onboard propellant is critical to allowing a mission to continue for a longer period of time or to enable the launch of a less massive spacecraft. Electric propulsion systems offer a much greater specific impulse than chemical systems and are able to operate for extended periods of time. However, these electric propulsion systems typically have much less thrust than their chemical counterparts and therefore must operate over longer durations in order to impart the desired momentum change. Recent developments in miniature electric propulsion offer the potential for new research opportunities for small spacecraft [@haque2013]. With reduced development intervals and decreased launch costs, small satellites have gained increased attention as a cost effective means of scientific and technologic development. The merger of small satellites with miniature electric propulsion enables inexpensive and responsive missions requiring large changes in orbital energy or extended mission lifespan. With the potential for more demanding missions, even greater importance is placed on the mission design to ensure that optimal trajectories satisfy mission requirements. In addition, non-Keplerian orbits and multi-body dynamics have been shown to allow for a much greater range of potential missions at a reduced energy cost [@koon2000]. Future space missions are increasing in complexity and will require new classes of orbits that are not possible via the traditional conic approach [@ross2006; @gomez2001]. Optimally combining the dynamical structure of the three-body problem with low-thrust propulsion systems is vital for future mission success. There has been extensive research focused on optimal control for spacecraft orbital transfers in the three-body problem [@mingotti2011; @grebow2011]. Typically, the optimal control problem is solved via direct methods, which approximate the continuous problem as a parameter optimization problem. The state and/or control trajectories are parameterized and solved in the form of a nonlinear optimization problem. References  and  use this direct approach in designing low-thrust transfers in the three-body problem. Alternatively, indirect methods apply calculus of variations to derive the necessary conditions for optimality. This yields a lower dimensioned problem than the direct approach and algebraic conditions that, when satisfied, guarantee local optimality in contrast to direct methods which result in sub-optimal solutions. The application of optimal control methods for orbital trajectory design is nontrivial. The three-body system dynamics are nonlinear and exhibit chaotic behaviors. Small changes in initial conditions result in large variations of the resulting system trajectory. Therefore, any optimization routine is highly sensitive to the initial guess. In addition, insight into the problem or intuition on the part of the designer is often required to determine initial conditions that will converge which achieve satisfactory results. Efficient numerical implementation is dependent on correct initial conditions as well as accurate numerical integration. Additionally, References  and  implement the solutions using conventional Runge-Kutta integration techniques. These techniques suffer from numerical instability and energy drift behaviors which make them ill-suited for long-term propagation. These dissipative effects are even more detrimental with the addition of low-thrust propulsion to the dynamic equations of motion. Conventional integration techniques fail to capture the physical laws and geometric properties of the dynamic system. As a result, the long term effects of low-thrust on the spacecraft trajectory are not accurately captured. References  and  have illustrated the rich structure that exists in the three-body problem. Within the three-body problem, a spacecraft’s feasible region of motion is constrained by its energy, or Jacobi integral. It has been shown that there exist multi-dimensional tubes, or invariant manifolds, of constant energy trajectories that span the state space. Associated with periodic solutions of the three-body dynamics, these invariant manifolds allow for the spacecraft to traverse vast expanses of the state space with zero energy change. However, the results presented are highly case specific and difficult to generalize to arbitrary transfers. Also, these results are based on control-free trajectories which rely on the underlying structure of the three-body system. In addition, transfer orbits along an invariant manifold require a longer time of flight which may be undesirable for time critical missions. The addition of low-thrust propulsion offers the potential of reduced transit times and the ability to depart from the free motion trajectory to allow for increased transfer opportunities. In order to address these issues, the authors propose an accurate and numerically stable method for long-term optimal orbital transfers. This approach will avoid the instability and dissipative effects of conventional integration schemes. In addition, the effects of low-thrust propulsion will be directly included in the integration method to ensure that extended duration optimal trajectories are computed correctly. This improved method will enable the derivation of a systematic method of generating optimal transfer orbits between arbitrary states. Indirect optimal control, based on the calculus of variations, will be implemented in order to generate optimal orbital transfers. This will avoid the approximation issues inherent in the previous work, which utilized direct optimal control methods. With this proposed method, the previous research on control-free trajectories will be generalized with the addition of low-thrust propulsion systems. To achieve these objectives, computational geometric optimal control techniques are applied. The dynamics of the three-body system are derived from the discrete Lagrangian, which approximates the integral of the continuous time Lagrangian over a fixed discrete step. Application of the discrete Euler-Lagrange equations, or the discrete Legendre transform, results in the discrete equations of motion. This discrete update map, or variational integrator, shares the same geometric properties of the continuous time system and exhibits much better energy behavior than the traditional integration methods, especially over long time periods. A discrete optimal control problem is formulated from the discrete equations of motion. This approach, where explicit discretization occurs prior to optimization, is in contrast to the typical method, where the equations of motion are implicitly discretized during the optimization procedure. Formulating the problem in this manner results in more stable and accurate optimal solutions. In indirect methods the optimal control problem is expressed as a two-point boundary value problem. Optimal solutions are generally sensitive to small variations in the initial multipliers. As a result, the numerical stability of sensitivity derivatives is critical to accuracy and computational performance. The use of geometric integrators, which do not suffer the numerical dissipation of conventional integration methods, results in a more robust and efficient solution. A discrete optimal control problem is formulated to determine the reachability set on a section. Given an initial condition and fixed time horizon, the reachable set is the set of states attainable, subject to the operational constraints of the spacecraft. Generation of this reachability set allows for the extension of the previous control-free missions in the three-body problem. In addition, the generation of the reachable set allows for a more systematic method of determining initial conditions and eases the burden on the designer. The addition of low-thrust propulsion affords the ability to enlarge the reachable set as compared to the control-free case. Maximization of the reachability set, on an appropriately chosen section, allows for a greater space of potential transfer trajectories. The use of the section allows for design on a lower dimensional space and simplifies the design process. Once an intersection is determined on the section a transfer is calculated. In short, the authors present a systematic method of generating optimal transfer orbits in the three-body problem. Previous results in the design of optimal transfers have relied on suboptimal direct optimization methods and conventional integration techniques. This paper provides a discrete optimal control formulation to generate the reachability set on a section. The use of a geometric integrators ensures numerical stability for long-duration orbit transfers. A numerical example is presented demonstrating a transfer trajectory from $ L_1 $ to the vicinity of the Moon and compared to the control-free design methodology. System Model {#sec:pcrtbp} ============ The system model is based on the planar circular restricted three body problem (PCRTBP). The Earth is assumed to be the more massive primary, $ m_1 $, while the Moon is the second, smaller primary $ m_2$. The equations of motion are developed in a rotating reference frame which allows for much greater insight into the structure of the dynamics. The $ \hat{x} $ axis is directed along the vector from the Earth to the Moon. The $ \hat{y} $ axis lies in the orbital plane and is orthogonal to $ \hat{x} $. The rotating reference frame is centered at the system barycenter. It is assumed that the $\left( \hat{x}, \hat{y}\right)$ rotates with a constant angular velocity equal to the mean motion of the Moon. Following convention, the system is also non-dimensionalized by the characteristic mass, length, and time [@koon2000]. As a result, the system can be characterized by a single mass parameter $ \mu $, $$\mu = \frac{m_2}{m_1+m_2} \, . \label{eq:mass_param}$$ The larger primary, $m_1$, is located at $ \left( -\mu , 0 \right)$ and the smaller $ m_2$ is located at $ \left( 1-\mu , 0 \right)$. In the rotating reference frame the Lagrangian is given by $$L = \frac{1}{2} \left( \left( \dot{x} -y \right)^2 + \left( \dot{y} + x \right)^2 \right) + \frac{1-\mu}{r_1} + \frac{\mu}{r_2}\, , \label{eq:lagrangian}$$ where the distance $r_1$ and $r_2$ of the spacecraft to each primary is defined in rotating coordinates as $$\begin{aligned} r_1 &= \sqrt{\left( x + \mu\right)^2 + y^2}\, , \\ r_2 &= \sqrt{\left( x - 1 + \mu\right)^2 + y^2}\, . \label{eq:distances}\end{aligned}$$ Application of the Euler-Lagrange equations results in following equations of motion defined in the rotating reference frame $$\begin{aligned} \ddot{x} - 2 \dot{y} + \frac{\partial}{\partial x} U &= u_x \, ,\nonumber \\ \ddot{y} + 2 \dot{y} + \frac{\partial}{\partial y} U &= u_y \, , \label{eq:cont_eom}\end{aligned}$$ where the effective potential $ U$ is defined as $$U = \frac{1}{2} \left( x^2 + y^2\right) + \frac{1-\mu}{r_1} + \frac{\mu}{r_2}\, , \label{eq:eff_pot}$$ and the control inputs are defined as $ u_x$ and $u_y$. The state is defined as $ \bar{x} = \begin{bmatrix}\bar{r} &\bar{v} \end{bmatrix}$ with $\bar{r} \in \R^{2\times1}$ and $\bar{v} \in \R^{2\times1}$ representing the position and velocity with respect to the system barycenter, respectively. The equations of motion may be rewritten in state space form as $$\left[\begin{array}{c} \dot{\bar{r}} \\ \dot{\bar{v}} \end{array} \right] = \left[ \begin{array}{c} \bar{v} \\ A \bar{v} + \nabla U + \bar{u}(t) \end{array} \right] = f\left( t,x, u\right) \, ,$$ where the matrix $ A $ and psuedo gravitational potential gradient $ \nabla U$ are $$\label{eq:A_mat} A = \left[ \begin{array}{ccc} 0 & 2 & 0 \\ -2 & 0 & 0 \\ 0 & 0 & 0 \end{array} \right] \, ,$$ $$\label{eq:grav_pot} \nabla U = \left[ \begin{array}{c} x - \frac{ \left(1 - \mu\right) \left(x + \mu\right)}{r_1^3} - \frac{\mu \left( x - 1 + \mu \right)}{r_2^3} \\ y - \frac{ \left(1 - \mu\right) y}{r_1^3} - \frac{\mu y}{r_2^3} \\ - \frac{ \left(1 - \mu\right) z}{r_1^3} - \frac{\mu z}{r_2^3}\end{array}\right] = \left[\begin{array}{c} U_x \\ U_y \\ U_z\end{array} \right] \, .$$ Jacobi Integral --------------- There exists a single integral, or constant of motion for the three-body problem [@lanczos1970; @szebehely1967]. This energy constant is analogous to the total mechanical energy, however is a non-physical quantity arising from the problem formulation [@szebehely1967]. Also known as the Jacobi constant, it is defined as a function of the position and velocity in the rotating frame and given by $$E\left( \bar{r} , \bar{v} \right) = \frac{1}{2}\left( \dot{x}^2 + \dot{y}^2\right) - U\left(x,y \right) \, . \label{eq:jacobi}$$ divides the phase space into distinct regions based on the energy level of the spacecraft. Fixing the Jacobi integral to a constant defines zero velocity curves, which are the locus of points where the kinetic energy, and hence velocity vanishes. As seen in Figure \[fig:energy\_contour\], the phase space is divided into distinct realms based on the energy level. In the vicinity of $ m_1$ or $m_2$ there exists a potential well. As the energy level increases there are five critical points of the effective potential \[eq:eff\_pot\] where the slope is zero. Three collinear saddle points on the $x$ axis and two equilateral points. These equilibrium, or Lagrange points, are labeled $ L_i, i = 1, \hdots, 5 $ and are shown in Figure \[fig:energy\_contour\]. The Jacobi integral is a valuable invariant property of the three-body system that allows for greater insight into the motion of the spacecraft. ![Contour Plot of Jacobi Integral[]{data-label="fig:energy_contour"}](energy_contours){width="50.00000%"} Variational Integrator {#sec:discrete_var} ====================== Geometric numerical integration deals with numerical integration methods which preserve the geometric properties of the flow of a differential equation, such as invaraint properties and symplecticity. Variational integrators are constructed by discretizing Hamilton’s principle rather than the continuous Euler-Lagrange equations [@marsden2001]. As a result, integrators developed in this manner have the desirable properties that they are symplectic and momentum preserving. In addition, they exhibit improved energy behavior over long integration periods. A short background on the variational principle for mechanical systems is presented. A discrete approximation of the action integral is presented and allows for construction of a variational integrator for the PCRTBP. Variational Principle --------------------- Consider a continuous mechanical system described by the Lagrangian, $ L( q, \dot{q} ) $, for the generalized position, $ q$, and velocity, $ \dot{q} $. In the standard approach of variational mechanics the action integral is formed by integrating the continuous Lagrangian along a path $ q(t) $ that the system follows from time $ t = 0 $ to $ t = T $ [@greenwood1988]. In the continuous time the action integral is defined as $$\begin{aligned} \label{eq:action_integral} S = \int_{0}^T L\left( q, \dot{q}\right) \, dt \, .\end{aligned}$$ Hamilton’s principle states that the actual path followed by a holonomic system results in a stationary action integral with respect to path variations for fixed endpoints. Taking the variation of \[eq:action\_integral\] gives \[eq:var\_principle\] $$\begin{aligned} \delta S &= \int_{0}^T \deriv{L}{q} \delta q + \deriv{L}{\dot{q}} \delta \dot{q} \, dt \\ &= \int_{0}^T \deriv{L}{q} \delta q - \frac{d}{dt} \left( \deriv{L}{\dot{q}}\right) \delta q \, dt - \left. \left[ \deriv{L}{\dot{q}} \delta q\right] \right|_0^T \\ &= \int_{0}^T \deriv{L}{q} - \frac{d}{dt} \left( \deriv{L}{\dot{q}} \right) \, dt \, ,\end{aligned}$$ where we have used integration by parts and the conditions $ \delta q(0) = 0 $ and $ \delta q(T) = 0$. For Hamilton’s principle to be valid for all admissible variations $ \delta q $, the integrand of \[eq:var\_principle\] must be zero for all $ t$, giving the continuous Euler-Lagrange Equations [@lanczos1970]. $$\begin{aligned} \label{eq:euler-lagrange} 0 = \deriv{L}{q} - \frac{d}{dt} \left( \deriv{L}{\dot{q}} \right) \, .\end{aligned}$$ Hamilton’s equations are derivable through the use of the Legendre transformation which is a mapping $ \left( q, \dot{q},t\right) \rightarrow \left(q, p, t \right) $ where $ p_i$ is the generalized momenta. $$\begin{aligned} \label{eq:legendre_transform} p_i = \deriv{L}{\dot{q}_i} \, .\end{aligned}$$ In the continuous time case the Hamiltonian is defined as $$\begin{aligned} \label{eq:hamiltonian} H &= \sum_{i = 1}^N p_i \dot{q}_i - L \left( q_i,\dot{q}_i, t \right) \, .\end{aligned}$$ Applying \[eq:legendre\_transform\] and taking the variation of \[eq:hamiltonian\] allows us to derive the equations of motion in Hamiltonian form \[eq:hamilton\_eq\] $$\begin{aligned} \dot{q}_i &= \deriv{H}{p_i} \, ,\\ \dot{p}_i &= - \deriv{H}{q_i} \, , \\ \deriv{L}{t} &= -\deriv{H}{t} \, .\end{aligned}$$ Both \[eq:euler-lagrange,eq:hamilton\_eq\] result in equations of motion for the mechanical system and are equivalent via the Legendre transform. results in $ n $ second order differential equations while \[eq:hamilton\_eq\] results in $ 2n $ first order differential equations. Discrete Variational Mechanics ------------------------------ A discrete analogue of Hamilton’s principle and the action integral is formed. Rather than taking a position, $ q $, and velocity, $ \dot{q} $, consider two positions $ q_0 $ and $ q_1 $ and a fixed time step $ h \in \R $. The two positions are points on the curve $ q(t) $ such that $ q_0 \approx q(0) $ and $ q_1 \approx q(h) $. A discrete time Lagrangian $ L_d( q_0, q_1) $ is formed which approximates the action integral between $ q_0 $ and $ q_1 $ as $$\begin{aligned} \label{eq:discrete_lagrangian} L_d\left( q_0 , q_1 \right) \approx \int_{0}^{h} L \left( q , \dot{q} \right) \, dt \, .\end{aligned}$$ Since \[eq:discrete\_lagrangian\] is calculated as a numerical integral, an appropriate quadrature rule is required. There are multiple possible methods one can use to approximate the integral in \[eq:discrete\_lagrangian\]. An appropriate approximation rule is determined based on the ease of implementation and accuracy desired. Rectangle $ L_d(q_0,q_1) =L(q_0,\frac{q_1-q_0}{h}) h $ ------------- -------------------------------------------------------------------------------------------------------------- Midpoint $ L_d(q_0,q_1) = L(\frac{q_0 + q_1}{2},\frac{q_1 - q_0}{h}) h $ Trapezoidal $ L_d(q_0, q_1) = \frac{1}{2} \left[ L(q_0, \frac{q_1 - q_0}{h} ) + L(q_1, \frac{q_1 - q_0 }{h} )\right] h $ : Selected Quadrature Rules\[tab:quadrature\] shows several possible approximation rules that are typically applied. The rectangle rule is a first order accurate method and offers a straightforward implementation. The midpoint and trapezoidal rules are both second order accurate methods. However, the midpoint rule results in an implicit form which adds further complexity to the equations of motion. In this work, the trapezoidal approximation is applied to the PCRTBP. Once an appropriate discrete Lagrangian is formed a discrete action sum is formed as the discrete analogue of \[eq:action\_integral\] $$\begin{aligned} \label{eq:action_sum} S_d = \sum_{k=0}^{N-1} L_d(q_k, q_{k+1}) \, .\end{aligned}$$ Once again a discrete version of Hamilton’s principle is applied to \[eq:action\_sum\]. Applying summation by parts yields $$\begin{aligned} \label{eq:dis_var_principle} \delta S_d &= \sum_{k=0}^{N-1} \deriv{L_d(q_k, q_{k+1})}{q_k} \delta q_k + \deriv{L_d(q_k, q_{k+1})}{q_{k+1}} \delta q_{k+1} \\ &= \sum_{k=1}^{N-1} \bracket{ \deriv{L_d(q_k, q_{k+1})}{q_k} + \deriv{L_d(q_{k-1}, q_k)}{q_{k+1}}} \delta q_k \, .\end{aligned}$$ For the discrete action sum to be stationary with respect to all admissible path variations, with fixed endpoints, the discrete Euler-Lagrange equations must be satisfied for $ k = 1, \cdots, N-1 $ resulting in $$\begin{aligned} \label{eq:discrete_euler-lagrange} 0 = \deriv{L_d(q_k, q_{k+1})}{q_k} + \deriv{L_d(q_{k-1}, q_k)}{q_{k+1}} \, .\end{aligned}$$ A discrete version of the Legendre transformation, referred to as a discrete fiber derivative, results in the equivalent Hamiltonian form expression. The discrete fiber derivative is given as \[eq:discrete\_legendre\] $$\begin{aligned} p_k &= \deriv{L_d(q_{k-1},q_k)}{q_{k+1}} = - \deriv{L_d(q_k, q_{k+1})}{q_k} \, ,\\ p_{k+1} &= \deriv{L_d(q_k, q_{k+1})}{q_{k+1}} \, .\end{aligned}$$ This yields a discrete Hamiltonian map $ (q_k, p_k) \to (q_{k+1}, p_{k+1}) $. A more extensive development of variational integrators can be found in Reference . Discrete Equations of Motion ---------------------------- The discrete equations of motion for the PCRTBP are derived by choosing an appropriate quadrature rule to discretize the Lagrangian in \[eq:lagrangian\]. In this work, the trapezoidal approximation is applied. The trapezoid rule allows for an explicit second order accurate approximation. The discrete Lagrangian is given by $$\begin{aligned} \label{eq:discrete_lagrangian} L_d = &\frac{h}{2} \left( \frac{1}{2} \bracket{\left( \frac{\xkp - \xk}{h} -\yk \right)^2 + \left( \frac{\ykp - \yk}{h} + \xk \right)^2} + \frac{1 - \mu}{r_{1_k}} + \frac{\mu}{r_{2_k}} \right. \nonumber \\ & + \left. \frac{1}{2} \bracket{\left( \frac{\xkp - \xk}{h} -\ykp \right)^2 + \left( \frac{\ykp - \yk}{h} + \xkp \right)^2} + \frac{1-\mu}{r_{1_{k+1}}} + \frac{\mu}{r_{2_{k+1}}} \right) \, .\end{aligned}$$ Applying a discrete version of the Lagrange-d’Alembert principle allows for inclusion of an external control force on the system [@marsden2001]. Using \[eq:discrete\_legendre,eq:discrete\_lagrangian\] and some manipulation, the equations of motion are given by \[eq:discrete\_eoms\] $$\begin{aligned} \xkp &= \frac{1}{1+ h^2} \bracket{h \dot{x}_k + h^2 \dot{y}_k + \xk \parenth{1+ \frac{3h^2}{2}} + \frac{h^3}{2} \yk - \frac{h^3}{2} U_{y_k} - \frac{h^2}{2} U_{x_k} } \label{eq:xkp} \, ,\\ \ykp &= h \dot{y}_k + h \xk - h \xkp + \yk + \frac{h^2 \yk}{2} - \frac{h^2 }{2} U_{y_k} \label{eq:ykp} \, ,\\ \dot{x}_{k+1} &= \dot{x}_k - 2 \yk + 2 \ykp + \frac{h}{2} \parenth{\xkp + \xk} - \frac{h}{2} U_{\xkp} - \frac{h}{2} U_{\xk} + h u_x \label{eq:xdotkp}\, ,\\ \dot{y}_{k+1} &= \dot{y}_{k} + 2 \xk - 2 \xkp + \frac{h}{2} \parenth{\ykp + \yk} - \frac{h}{2} U_{\ykp} - \frac{h}{2} U_{\yk} + h u_y \label{eq:ydotkp} \, .\end{aligned}$$ The discrete equations of motion are given in the Lagrangian form after applying the discrete fiber derivative from \[eq:discrete\_legendre\] as $ p_{\xk} = \dot{x}_k - \yk $ and $ p_{\yk} = \dot{y}_k + \xk $. The state is defined as $ \bar{ x}_k = \begin{bmatrix} \xk & \yk & \dot{x}_k & \dot{y}_k \end{bmatrix}^T$ and the control input is $ \bar{u} = \begin{bmatrix} u_x & u_y \end{bmatrix}^T $. The discrete potential gradients are given by \[eq:discrete\_potential\_grad\] $$\begin{aligned} U_{\xk} &= \frac{\parenth{1 -\mu} \parenth{\xk + \mu}}{\distonek^3} + \frac{ \mu \parenth{\xk -1 + \mu}}{\disttwok^3} \label{eq:Uxk} \, ,\\ U_{\yk} &= \frac{\parenth{1 -\mu} \yk}{\disttwok^3} + \frac{ \mu \yk}{\disttwok^3} \label{eq:Uyk} \, ,\\ U_{\xkp} &= \frac{\parenth{1 -\mu} \parenth{\xkp + \mu}}{\distonekp^3} + \frac{ \mu \parenth{\xkp -1 + \mu}}{\disttwokp^3} \label{eq:Uxkp}\, ,\\ U_{\ykp} &= \frac{\parenth{1 -\mu} \ykp}{\distonekp^3} + \frac{ \mu \ykp}{\disttwokp^3} \label{eq:Uykp} \, .\end{aligned}$$ The distances to each primary are defined as \[eq:discrete\_distance\] $$\begin{aligned} \distonek &= \sqrt{\left( \xk + \mu\right)^2 + \yk^2} \label{eq:distonek}\, ,\\ \disttwok &= \sqrt{\left( \xk - 1 + \mu\right)^2 + \yk^2} \label{eq:disttwok}\, ,\\ \distonekp &= \sqrt{\left( \xkp + \mu\right)^2 + \ykp^2} \label{eq:distonekp}\, ,\\ \disttwokp &= \sqrt{\left( \xkp - 1 + \mu\right)^2 + \ykp^2} \label{eq:disttwokp} \, .\end{aligned}$$ Care must be taken during the implementation of \[eq:discrete\_eoms\]. As \[eq:discrete\_potential\_grad,eq:discrete\_distance\] are defined at both step $ k $ and $ k+1 $ they must be evaluated at both time steps. is implemented by first defining an initial state $ \bar{x}_k $ and control $ \bar{u}_k $. The distances and gravitational potential at step $ k $ are evaluated from \[eq:distonek,eq:disttwok,eq:Uxk,eq:Uyk\]. The discrete update steps in \[eq:xkp,eq:ykp\] are evaluated to generate $ \xkp $ and $ \ykp$. Next, the distances and gravitational potential at step $ k+1 $ are evaluated from \[eq:distonekp,eq:disttwokp,eq:Uxkp,eq:Uykp\]. Finally, the update steps in \[eq:xdotkp,eq:ydotkp\] are evaluated. This results in the complete discrete update map $ \bar{x}_k \to \bar{x}_{k+1} $ given $ \bar{u}_k $. [0.3]{} ![Integrator Comparison[]{data-label="fig:integrator_compare"}](trajectory "fig:"){width="\textwidth"}   [0.3]{} ![Integrator Comparison[]{data-label="fig:integrator_compare"}](components "fig:"){width="\textwidth"}   [0.3]{} ![Integrator Comparison[]{data-label="fig:integrator_compare"}](energy "fig:"){width="\textwidth"} A simulation comparing the variational integrator to a conventional Runge-Kutta method is given in \[fig:integrator\_compare\]. A particle is simulated from an initial condition of $ \bar{x}_0 = \begin{bmatrix} 0.75 & 0 & 0 & 0.2883\end{bmatrix}^T $ for $ t_f = 200 \approx 15$ years in the Earth-Moon system. The variational integrator uses a step size of while the Runge-Kutta method uses a variable step size implemented via ODE45 in Matlab. shows the trajectory of the spacecraft in the rotating reference frame for both integration schemes. Both integration schemes result in trajectories that are initially nearly identical. The discrete equations of motion are an accurate approximation for the continuous dynamics as they closely match the solution of ODE45 over the initial portion of the simulation. However, as time progresses the trajectories begin to diverge due to the differences in system energy. shows the evolution of the Jacobi integral. The variational integrator exhibits a bounded behavior about the initial energy with a mean variation of . However, the conventional Runge-Kutta method demonstrates a clear energy drift of . Over long simulation horizons or with the addition of small control inputs this poor energy behavior limits the applicability of conventional techniques. Invariant Manifolds {#sec:invariant_manifold} =================== There exists a rich structure of tubes, or invariant manifolds, in the three-body problem [@koon2000; @conley1968]. The manifold structure associated with periodic orbits about the $ L_1 $ and $ L_2 $ Lagrange points are critical to the understanding of the motion of spacecraft as well as comets/asteroids. In addition, the stable and unstable manifolds serve as the boundaries of the phase space region that allow for the transport between realms in a single three-body system or between multiple three-body systems. These invariant manifolds only exist as a result of the dynamic formulation of the three-body problem in a rotating reference frame. In this way, it is possible to design trajectories that intersect the invariant manifolds and connect widely separated regions of the state space. shows some of the tubes, projected from the phase space onto the configuration space, associated with $ L_1 $ and $ L_2 $ periodic orbits in the Earth-Moon system. This technique has been heavily investigated in Reference , applied to operational missions in Reference , and applied to potential multi-moon orbiter missions in Reference . In addition, much recent work has focused on potential return missions to the Moon [@zanzottera2012; @campagnola2012; @mingotti2011; @ozimek2010a; @mingotti2009]. [0.25]{} ![Invariant Manifolds for Planar Earth-Moon three-body system[]{data-label="fig:invariant_manifolds"}](U1_Manifolds "fig:"){width="\columnwidth"}   [0.25]{} ![Invariant Manifolds for Planar Earth-Moon three-body system[]{data-label="fig:invariant_manifolds"}](U2_Manifolds "fig:"){width="\columnwidth"}   [0.25]{} ![Invariant Manifolds for Planar Earth-Moon three-body system[]{data-label="fig:invariant_manifolds"}](U3_Manifolds "fig:"){width="\columnwidth"}   [0.25]{} ![Invariant Manifolds for Planar Earth-Moon three-body system[]{data-label="fig:invariant_manifolds"}](U4_Manifolds "fig:"){width="\columnwidth"} [0.25]{} ![Sections for Planar Earth-Moon three-body system[]{data-label="fig:poincare_sections"}](U1_poincare "fig:"){width="\columnwidth"}   [0.25]{} ![Sections for Planar Earth-Moon three-body system[]{data-label="fig:poincare_sections"}](U2_poincare "fig:"){width="\columnwidth"}   [0.25]{} ![Sections for Planar Earth-Moon three-body system[]{data-label="fig:poincare_sections"}](U3_poincare "fig:"){width="\columnwidth"}   [0.25]{} ![Sections for Planar Earth-Moon three-body system[]{data-label="fig:poincare_sections"}](U4_poincare "fig:"){width="\columnwidth"} Another useful technique in the analysis of the free motion of dynamical systems is that of the section. The section is the intersection of a periodic orbit of a dynamical system with a lower dimensional sub-space transverse to the flow. This allows for greater insight into the stability and dynamics of periodic solutions of dynamic system. Points are drawn as the periodic solution intersects the section. Great insight and structure into the dynamical system is typically available through use of section. For example, \[fig:poincare\_sections\] shows the intersections of the invariant manifolds of \[fig:invariant\_manifolds\] with the sections defined by the black line segments. The section and the Jacobi integral reduce the state space from four to two dimensions for the planar three-body problem [@koon2001]. This greatly eases the design process and allows for a simple planar visualization of the intersection of the invariant manifolds. Intersecting states are easy to determine and allow for transfers between invariant manifolds. The intersection of the invariant manifolds on a section is central to determining the desired control-free trajectory. However, the results previously developed are highly case specific and difficult to generalize to arbitrary transfers. Also, these results are based on control-free trajectories which rely on the underlying structure of the three-body system. In addition, transfer orbits along an invariant manifold require large time of flights which may be undesirable for time critical missions. The addition of low-thrust propulsion offers the potential of reduced transit times and the ability to depart from the free motion trajectory to allow for increased transfer opportunities. In this work, an optimal control problem is formulated to generate a reachable set of the spacecraft. The reachable set is computed on an appropriate section and is used to design an appropriate transfer trajectory. Optimal Control Formulation for Reachability Set {#sec:optimal_control} ================================================ In this section, an optimal control formulation is presented to determine and design transfers within the three-body problem. The application of variational integrators to optimal control problems is referred to as computational geometric optimal control. This formulation is based on the concept of the reachability set on a section. This method allows for one to easily determine potential transfer opportunities by finding set intersections on a lower dimensional space and greatly reduces the design process. The addition of continuous low thrust propulsion extends the control free design process presented in Reference  and allows for a greater range of potential transfers with a reduced time of flight. Reachability theory provides a framework to evaluate control capability and safety. The reachable set contains all possible trajectories that are achievable over a fixed time horizon from a defined initial condition, subject to the operational constraints of the system. Reachability theory has been applied to aerospace systems such as collision avoidance, safety planning, and performance characterization. The theory formally supporting reachability has been extensively developed and is directly derivable from optimal control theory [@varaiya2000; @lygeros2002; @lygeros2004]. Computation of the reachable set for a system involves solving the Hamilton-Jacobi partial differential equation or satisfying a dynamic programming principle. Analytical computation of reachable sets is an ongoing problem and is only possible for certain classes of systems. Typically, numerical methods are used to generate approximations of the reachability set, but are generally limited by the dimensionality of the problem. Reachability theory has recently been applied to space systems [@holzinger2009; @komendera2012a]. Computation of reachable sets is critical to space situational awareness, rendezvous and proximity operations, and orbit determination operations. Specifically maintaining accurate estimates of a spacecraft state over extended periods is not trivial. The challenge is increased for multiple spacecraft operating in close proximity or when there are long periods of time between measurements. Coupling the ability for continuous low-thrust propulsion between measurements increases the measurement association complexity. Computing the reachability set given estimated states and control authorities allows one to better correlate subsequent measurements or determine sensor pointing regions in the event of a lost spacecraft. The cost function is defined as $$J = -\frac{1}{2} \left( \bar{x}(N) - \bar{x}_{n}(N)\right)^T \left[ \begin{array}{cccc} 1 & 0& 0& 0 \\ 0& 0& 0& 0\\ 0 & 0 & 1 &0\\ 0 & 0& 0& 0 \end{array} \right] \left( \bar{x}(N) - \bar{x}_{n}(N)\right) \, . \label{eq:cost}$$ The term $ \bar{x}_n(N) $ is the final state of a control-free trajectory while the term $ \bar{x}(N) $ is the final state under the influence of the control input. In this fashion, the aim is to maximize the distance of the final state from that of the control-free trajectory. A chosen section is defined through the use of appropriate terminal constraints given by $$\begin{aligned} m_1 &= 0 = \frac{y(N) - L_{1y}}{x(N) - L_{1x}} - \tan{\alpha_d} \, , \\ m_2&= 0 = \frac{\dot{x}(N) - \dot{x_n}(N) }{x(N) -x_n(N) } - \tan{\theta_d} \, , \\ 0 &\geq\bar{u}^T \bar{u} - u_{max}^2 \, ,\end{aligned}$$ \[eq:constraints\] where the angles $ \alpha_d$ and $ \theta_d$ define the section and a specific direction upon the section, respectively. The goal is to determine the control input $ \bar{u}_k$ such that the cost function \[eq:cost\] is minimized subject to the state equations of motion \[eq:discrete\_eoms\] and constraints \[eq:constraints\]. Application of the Euler-Lagrange equations allows us to derive the necessary conditions for optimality [@bryson1975]. The discrete variational integrator in \[eq:discrete\_eoms\] is used rather than the continuous time counterpart. This results in a discrete optimal control problem and the discrete necessary conditions are given as \[eq:necc\_cond\] $$\begin{aligned} \lambda_{k+1}^T &= \lambda_k^T \deriv{f_k}{\bar{x}_k}^{-1} \, , \\ 0 &= \deriv{H_k}{\bar{u}_k} \, ,\\ 0 &= \deriv{\phi}{\bar{x}_k}^T + \deriv{m}{\bar{x}_k}^T\beta - \lambda^T(N) \, , \end{aligned}$$ where the Hamiltonian $H$ is defined as $$H_k = \lambda_k^T f(\bar{x}_k, \bar{u}_k) \, , \label{eq:hamiltonian_opt}$$ and $\lambda \in \R^{4 \times 1}$ is the costate and $\beta \in \R^{2 \times 1} $ are the additional Lagrange multipliers associated with the terminal constraints in \[eq:constraints\]. This indirect optimal control formulation leads to a two point boundary value problem with split boundary conditions. By sweeping the angle $ \theta_d $ one can approximate the reachable set on the section subject to the bounded control input. The costate equation of motion requires the Jacobian of \[eq:discrete\_eoms\] and is given by $$\begin{aligned} \label{eq:costate_eom} \lambda_{k+1}^T = \lambda_k^T \begin{bmatrix} f_{1_x} & f_{1_y} & f_{1_{\dot{x}}} & f_{1_{\dot{y}}} \\ f_{2_x} & f_{2_y} & f_{2_{\dot{x}}} & f_{2_{\dot{y}}} \\ f_{3_x} & f_{3_y} & f_{3_{\dot{x}}} & f_{3_{\dot{y}}} \\ f_{4_x} & f_{4_y} & f_{4_{\dot{x}}} & f_{4_{\dot{y}}} \end{bmatrix} ^ {-1} \, .\end{aligned}$$ The derivation of \[eq:costate\_eom\] is given in Appendix A. In addition, the computation of \[eq:costate\_eom\] requires inversion of the Jacobian matrix. This is a computationally expensive operation that is prone to numerical error and instability. A method is presented in Appendix B to avoid this inversion and determine an explicit update map $ \lambda_k \to \lambda_{k+1} $. The optimal control formulation presented in this section results in a two point boundary value problem (TPBVP). There exist many methods to solve TPBVPs such as gradient, quasilinearization, and shooting methods [@bryson1975]. In this work, a multiple shooting method is implemented. Shooting methods are common in astrodynamic trajectory design problems and relatively simple to implement. In the shooting method, initial conditions are varied such that a terminal constraint is satisfied, similar to the way an archer modifies the bow in order to more accurately strike a target. Rather than numerical integration over the entire time interval, multiple shooting segments the interval into several smaller sub-arcs. Additional interior constraints are used to ensure a continuous solution. This has the benefit of decreasing the numerical sensitivity of the final states to changes in the initial conditions along each sub-arc. Numerical Example {#sec:simulation} ================= A numerical simulation is presented to demonstrate the transfer procedure. The goal is to design a transfer trajectory from a planar periodic orbit about the $ L_1$ Lagrange point to a bounded orbit in the vicinity of the Moon. The target region is created by choosing an initial condition of $ x_0 = \begin{bmatrix}1.05 & 0 & 0 & 0.35 \end{bmatrix} $ with $ \mu = 0.0125 $. The initial state is propagated over a period of $ t = \num{20} $ in non-dimensional units which corresponds to approximately years. ![Target Orbit Region[]{data-label="fig:moon_orbit"}](moon_orbit){width="50.00000%"} shows that the desired trajectories remain in the vicinity of the Moon in the rotating reference frame. This type of orbit would be useful for a variety of mission scenarios. For example, a series of communication satellites could be placed in this type of orbit. The bounded trajectories of the vehicles and constant line of sight to both the Moon and the Earth would allow for constant communication for future manned missions and potential habitats. As a baseline, the method introduced in Reference  is implemented. The method is based on the invariant manifolds associated with the periodic orbits of the three body system. These invariant manifolds are a set of trajectories that either asymptotically arrive or depart the periodic orbit. Linking the invariant manifolds of several periodic orbits, or coupled three body systems, has been used to generate orbital transfers. shows the unstable invariant manifold associated with the $ L_1$ periodic orbit. The invariant manifolds are globalized using the eigenvectors of the Monodromy matrix along the periodic orbit. The eigenvectors serve as a local approximation of the stable and unstable directions. Perturbations along these directions serve to approximate the invariant manifolds. The unstable invariant manifold is numerically propagated to the same section defined on the $ \hat{x} $ axis. The intersection of the invariant manifold and the section is denoted by the green markers in \[fig:poincare\_compare\]. Only a single branch of the invariant manifold intersects with the ascending region of the target orbit. There are no intersections of the invariant manifold with the descending region of the target orbit. The numerical values associated with the green points denote the required time of flight along the invariant manifold in non-dimensional units. A transfer along the invariant manifold requires on average $ t_f \approx 3.1 $ as compared to $ t_f \approx 1.4 $ for a transfer using low thrust propulsion and the reachable set. Finally, it should be noted that an additional maneuver would be required for a transfer using the invariant manifold. The intersection on the section only shows that the $ x \text{ and } \dot{x} $ components intersect. An additional instantaneous $ \Delta V $ would be required to transfer from the invariant manifold to the target. A periodic orbit is generated about the $ L_1 $ Lagrange point. A section is chosen to allow for design on a lower dimensional subspace. The section is defined along the $ \hat{x} $ axis, or defined with $ \alpha = \ang{0} $ and intersects both the initial and target orbits. From the periodic orbit, a series of optimal trajectories are generated to approximate the reachable set. The trajectories are generated from a fixed initial state of $ x_0 = \begin{bmatrix}0.8156 & 0 & 0 & 0.1922 \end{bmatrix} $ over a fixed time span of $ t_f = 1.4 $. By varying the angle $ \theta_d$ in \[eq:constraints\], a different direction along the section is maximized. The intersection of the optimal trajectories as well as those of the target Moon orbit with the section are shown in \[fig:transfer\_orbit\]. [0.5]{} ![Transfer Trajectory[]{data-label="fig:transfer_orbit"}](reach_trajectory "fig:"){width="\textwidth"}   [0.5]{} ![Transfer Trajectory[]{data-label="fig:transfer_orbit"}](manifold_trajectory "fig:"){width="\textwidth"} [0.5]{} ![Transfer Trajectory[]{data-label="fig:transfer_orbit"}](poincare_compare "fig:"){width="\textwidth"} The optimal trajectories, under the influence of the control input $ \bar{u} $, are shown in \[fig:reach\_trajectory\]. Initially, the spacecraft is assumed to lie on the periodic orbit. As a result, the intersection of this periodic orbit with the section are two points corresponding to the two crossing of the orbit. The use of the continuous low thrust propulsion expands the reachable set to region bounded by the red markers in \[fig:poincare\_compare\]. The reachable set is an ellipsoidal region with a major axis aligned along $ \theta \approx \ang{70} $. The blue points in \[fig:poincare\_compare\] are the intersections of the target Moon orbit and the section. The two circular regions are the ascending (right) and descending (left) intersections of the target orbit and section. shows that the reachable set and those of the descending target region intersect. As both regions are discretely approximated a linear interpolation is used to determine the exact intersection state on the section. This intersection generates a partial target state of $ x_t \text{ and } \dot{x}_t $. Using \[eq:jacobi\] and the intersection state the final component $ \dot{y} $ is calculated. This results in a complete target state $ \bar{x}_t $ which lies on the reachable set and on the target orbit. A final optimal trajectory is generated such that the $ \bar{x}(N) = \bar{x}_t $. This transfer trajectory is denoted by the green path in \[fig:reach\_trajectory\]. Conclusions {#sec:conclusion} =========== In this paper, an optimal transfer process which combines concepts of reachability and section is used to generate transfer between planar periodic orbits in the three-body problem. The section allows for trajectory design on a lower dimensional phase space and simplifies the process. The indirect optimal control formulation enables straightforward method of incorporating additional path and control constraints. However, the use of optimal control techniques leads to open loop trajectories that are not robust to model uncertainties or disturbances. Lyapunov control theory, which has previously been applied to the two-body problem, is being investigated in the hope of designing closed loop control schemes for this three-body scenario [@chang2002]. This analysis has also assumed perfect attitude control and the ability to orient thrust in any direction. The addition of attitude dynamics and realistic pointing constraints would significantly improve the applicability. Appendix A: Costate Equations of Motion {#sec:costate_appendix .unnumbered} ======================================= The development of the costate equations of motions begins with determining the second order partial derivatives of the gravitational potential. Due to the symmetry of partial derivatives only three terms are required and are given by $$\begin{aligned} \label{eq:second_discrete_potential_grad} U_{x\xk} &= \parenth{1-\mu} \bracket{\frac{1}{\distonek^3} - \frac{3 \parenth{\xk +\mu}^2}{\distonek^5}} + \mu \bracket{\frac{1}{\disttwok^3} - \frac{3 \parenth{\xk -1 + \mu}^2}{\disttwok^5}} \, ,\\ U_{y\yk} &= \parenth{1-\mu} \bracket{\frac{1}{\distonek^3} - \frac{3 \yk^2}{\distonek^5}} + \mu \bracket{\frac{1}{\disttwok^3} - \frac{3 \yk^2}{\disttwok^5}} \, ,\\ U_{x\yk} &= U_{y\xk} = \frac{-3 \parenth{1-\mu} \parenth{\xk +\mu} \yk}{\distonek^3} - \frac{3\mu\yk\parenth{\xk-1+\mu}}{\disttwok^5} \, .\end{aligned}$$ The gradient of \[eq:xkp\] is given as \[eq:xkp\_grad\] $$\begin{aligned} f_{1_x} &= \frac{1}{1+h^2} \bracket{h^2 + 1 + \frac{h^2}{2} -\frac{h^3}{2} U_{y\xk} - \frac{h^2}{2}U_{x\xk}} \, ,\\ f_{1_y} &= \frac{1}{1+h^2} \bracket{ \frac{h^3}{2} -\frac{h^3}{2} U_{y\yk} - \frac{h^2}{2}U_{x\yk}} \, ,\\ f_{1_{\dot{x}}} &= \frac{h}{1+h^2} \, ,\\ f_{1_{\dot{y}}} &= \frac{h^2}{1+h^2} \, .\end{aligned}$$ The gradient of \[eq:ykp\] is given as \[eq:ykp\_grad\] $$\begin{aligned} f_{2_x} &= h -h f_{1_x} - \frac{h^2}{2} U_{y\xk} \, , \\ f_{2_y} &= -h f_{1_y} + 1 + \frac{h^2}{2} - \frac{h^2}{2} U_{y\yk} \, ,\\ f_{2_{\dot{x}}} &= -h f_{1_{\dot{x}}} \, ,\\ f_{2_{\dot{y}}} &= h - h f_{1_{\dot{y}}} \, .\end{aligned}$$ The gradients of \[eq:distonekp,eq:disttwokp\] are given as \[eq:distkp\_grad\] $$\begin{aligned} \deriv{\distonekp}{\bar{x}} &= \parenth{\left( \xkp + \mu\right)^2 + \ykp^2}^{-\frac{1}{2}} \bracket{\parenth{\xkp + \mu} f_{1_{\bar{x}}} + \ykp f_{2_{\bar{x}}}} \, ,\\ \deriv{\disttwokp}{\bar{x}} &= \parenth{\left( \xkp - 1 + \mu\right)^2 + \ykp^2}^{-\frac{1}{2}} \bracket{\parenth{\xkp -1 + \mu} f_{1_{\bar{x}}} + \ykp f_{2_{\bar{x}}}} \, .\end{aligned}$$ The second order partial derivatives of the gravitational potential at $ k+1$ are given as $$\begin{aligned} \deriv{U_{x\xkp}}{\bar{x}} &= \parenth{1-\mu}\bracket{\frac{1}{\distonekp^3} f_{1_{\bar{x}}} - \frac{3 \parenth{\xkp +mu}}{\distonekp^4} \deriv{\distonekp}{\bar{x}}} + \mu \bracket{\frac{1}{\disttwokp^3} f_{1_{\bar{x}}} - \frac{-3 \parenth{\xkp -1 + \mu}}{\disttwokp^4} \deriv{\disttwokp}{\bar{x}}} \, ,\\ \deriv{U_{y\xkp}}{\bar{x}} &= \parenth{1-\mu}\bracket{\frac{1}{\distonekp^3} f_{2_{\bar{x}}} - \frac{3 \ykp}{\distonekp^4} \deriv{\distonekp}{\bar{x}}} + \mu \bracket{\frac{1}{\disttwokp^3} f_{2_{\bar{x}}} - \frac{-3 \ykp}{\disttwokp^4} \deriv{\disttwokp}{\bar{x}}} \, .\end{aligned}$$ The gradient of \[eq:xdotkp,eq:ydotkp\] are given as \[eq:xdotkp\_grad\] $$\begin{aligned} f_{3_x} &= 2 f_{2_x} + \frac{h}{2} \parenth{f_{1_x} + 1} - \frac{h}{2} U_{x\xkp} - \frac{h}{2} U_{x\xk} \, ,\\ f_{3_y} &= -2 + 2 f_{2_y} + \frac{h}{2} f_{1_y} - \frac{h}{2} U_{x\ykp} -\frac{h}{2} U_{x\yk} \, ,\\ f_{3_{\dot{x}}} &= 1 + 2 f_{2_{\dot{x}}} + \frac{h}{2} f_{1_{\dot{x}}} - \frac{h}{2} U_{x\xdotkp}\, ,\\ f_{3_{\dot{y}}} &= 2 f_{2_{\dot{y}}}\, ,\end{aligned}$$ \[eq:ydotkp\_grad\] $$\begin{aligned} f_{4_x} &= 2 - 2 f_{1_x} + \frac{h}{2} f_{2_x} - \frac{h}{2} U_{y\xkp} - \frac{h}{2} U_{y\xk} \, ,\\ f_{4_y} &= -2f_{1_y} - \frac{h}{2} \parenth{f_{2_y} + 1} - \frac{h}{2} U_{y\ykp} -\frac{h}{2} U_{y\yk} \, ,\\ f_{4_{\dot{x}}} &= - 2 f_{1_{\dot{x}}} + \frac{h}{2} f_{2_{\dot{x}}} - \frac{h}{2} U_{y\xdotkp}\, ,\\ f_{4_{\dot{y}}} &= 1 - 2 f_{1_{\dot{y}}} + \frac{h}{2} f_{2_{\dot{y}}} - \frac{h}{2} U_{y\ydotkp}\, .\end{aligned}$$ These gradient equations are in a cascade type structure. are functions of \[eq:xdotkp,eq:ydotkp\]. As a result, the accuracy of the Jacobian will tend to decrease as the first order approximation errors accumulate. Appendix B: Gauss Jordan Elimination {#sec:costate_gauss_jordan .unnumbered} ==================================== The costate equations of motion are given by \[eq:costate\_eom\] and repeated here as $$\begin{aligned} \label{eq:costate_eom_transpose} \begin{bmatrix} \fonex & \ftwox & \fthreex & \ffourx \\ \foney & \ftwoy & \fthreey & \ffoury \\ \fonexd & \ftwoxd & \fthreexd & \ffourxd \\ \foneyd & \ftwoyd & \fthreeyd & \ffouryd \end{bmatrix} \begin{bmatrix} \lambda_{\xkp} \\ \lambda_{\ykp} \\ \lambda_{\xdotkp} \\ \lambda_{\ydotkp} \end{bmatrix} = \begin{bmatrix} \lambda_{\xk} \\ \lambda_{\yk} \\ \lambda_{\xdotk} \\ \lambda_{\ydotk} \end{bmatrix} \, .\end{aligned}$$ To determine the discrete update map $ \lambda_k \to \lambda_{k+1}$ the inverse of the Jacobian matrix is required. In order to avoid the need of an explicit inversion a Gauss Jordan method is implemented. To begin, several terms are defined which are required to carry out the row operations and are defined as \[eq:ref\_scale\] $$\begin{aligned} a &= -\frac{\foney}{\fonex} \, ,\\ b &= -\frac{\fonexd}{\fonex} \, , \\ c &= -\frac{\foneyd}{\fonex} \, ,\\ e &= -\frac{\ftwoxd + b \ftwox}{\ftwoy + a \ftwox} \, ,\\ f &= -\frac{\ftwoyd + c \ftwox}{\ftwoy + a \ftwox} \, ,\\ g &= -\frac{\fthreeyd + c \fthreex + f \parenth{\fthreey + a \fthreex}}{\fthreexd + b \fthreex + e \parenth{\fthreey + a \fthreex}}\, .\end{aligned}$$ is transformed to row echelon form using elementary row operations and is defined as $$\begin{aligned} \label{eq:costate_ref} \begin{bmatrix} \alpha_{11} & \alpha_{12} & \alpha_{13} & \alpha_{14} \\ 0 & \alpha_{22} & \alpha_{23} & \alpha_{24} \\ 0 & 0 & \alpha_{33} & \alpha_{34} \\ 0 & 0 & 0 & \alpha_{44} \end{bmatrix} \begin{bmatrix} \lambda_{\xkp} \\ \lambda_{\ykp} \\ \lambda_{\xdotkp} \\ \lambda_{\ydotkp} \end{bmatrix} = \begin{bmatrix} \beta_1 \\ \beta_2 \\ \beta_3 \\ \beta_4 \end{bmatrix} \, ,\end{aligned}$$ where the terms $ \alpha_{ij} $ and $ \beta_{i} $ are defined as follows $$\begin{aligned} \alpha_{11} &= \fonex \, ,\\ \alpha_{12} &= \ftwox \, ,\\ \alpha_{13} &= \fthreex \, ,\\ \alpha_{14} &= \ffourx \, ,\\ \alpha_{22} &= \ftwoy + a \ftwox \, ,\\ \alpha_{23} &= \fthreey + a\fthreex \, ,\\ \alpha_{24} &= \ffoury + a \ffourx \, ,\\ \alpha_{33} &= \fthreexd + b \fthreex + e \parenth{\fthreey + a \fthreex}\, ,\\ \alpha_{34} &= \ffourxd + b \ffourx + e \parenth{\ffoury + a \ffourx} \, ,\\ \alpha_{44} &= \ffouryd + c \ffourx + f \parenth{\ffoury + a \ffourx} + g \parenth{\ffourxd + b \ffourx + e \parenth{\ffoury + a \ffourx}} \, ,\\ \beta_1 &= \lambda_{\xk} \, ,\\ \beta_2 &= \lambda_{\yk} + a \lambda_{\xk} \, ,\\ \beta_3 &= \lambda_{\xdotk} + b \lambda_{\xk} + e \parenth{\lambda_{\yk} + a \lambda_{\xk}} \, ,\\ \beta_4 &= \lambda_{\ydotk} + c \lambda_{\xk} + f \parenth{\lambda_{\yk}+a \lambda_{\xk}} + g \parenth{\lambda_{\xdotk} + b \lambda_{\xk} + e \parenth{\lambda_{\yk} + a \lambda_{\xk}}} \, .\end{aligned}$$ Finally, backsubstituion is used to determine explicit equations for the discrete update map $ \lambda_k \to \lambda_{k+1} $ which is defined as \[eq:costate\_update\] $$\begin{aligned} \lambda_{\ydotkp} &= \frac{\beta_4}{\alpha_{44}} \, ,\\ \lambda_{\xdotkp} &= \frac{\beta_3}{\alpha_{33}} - \frac{\alpha_{34}}{\alpha_{33}} \lambda_{\ydotkp} \, ,\\ \lambda_{\ykp} &= \frac{\beta_2}{\alpha_22} - \frac{\alpha_{33}}{\alpha_{22}}\lambda_{\xdotkp} - \frac{\alpha_{24}}{\alpha_{22}} \lambda_{\ydotkp} \, ,\\ \lambda_{\xkp} &= \frac{\beta_1}{\alpha_{11}} - \frac{\alpha_{12}}{\alpha_{11}} \lambda_{\ykp} - \frac{\alpha_{13}}{\alpha_{11}} \lambda_{\xdotkp} - \frac{\alpha_{14}}{\alpha_{11}} \lambda_{\ydotkp} \, .\end{aligned}$$ [^1]: Doctoral Student, Mechanical and Aerospace Engineering, George Washington University, 800 22nd St NW, Washington, DC 20052, Email: <skulumani@gwu.edu>. [^2]: Associate Professor, Mechanical and Aerospace Engineering, George Washington University, 800 22nd St NW, Washington, DC 20052, Tel: 202-994-8710, Email: <tylee@gwu.edu>.

ArXiv

--- abstract: 'There is debate if phoneme or viseme units are the most effective for a lipreading system. Some studies use phoneme units even though phonemes describe unique short sounds; other studies tried to improve lipreading accuracy by focusing on visemes with varying results. We compare the performance of a lipreading system by modeling visual speech using either 13 viseme or 38 phoneme units. We report the accuracy of our system at both word and unit levels. The evaluation task is large vocabulary continuous speech using the TCD-TIMIT corpus. We complete our visual speech modeling via hybrid DNN-HMMs and our visual speech decoder is a Weighted Finite-State Transducer (WFST). We use DCT and Eigenlips as a representation of mouth ROI image. The phoneme lipreading system word accuracy outperforms the viseme based system word accuracy. However, the phoneme system achieved lower accuracy at the unit level which shows the importance of the dictionary for decoding classification outputs into words.' bibliography: - 'mybib.bib' title: 'Comparing phonemes and visemes with DNN-based lipreading' --- Introduction {#sec:intro} ============ As lipreading transitions from GMM/HMM-based technology to systems based on Deep Neural Networks (DNNs) there is merit in re-examining the old assumption that phoneme-based recognition outperforms recognition with viseme-based systems. Also, given the greater modeling power of DNNs, there is value in considering a range of rather primitive features such as Discrete Cosine Transform (DCT) [@ahmed1974discrete] and Eigenlips [@bregler1994eigenlips] which had previously been disparaged due to their poor performance. Visual speech units divide into two broad categories; phonemes and visemes. A phoneme is the smallest unit of speech that distinguishes one word sound from another [@international1999handbook]. Therefore it has a strong relationship with an acoustic speech signal. In contrast, a viseme is the basic visual unit of speech that represents a gesture of the mouth, face and visible parts of the teeth and tongue, the visible articulators. Generally speaking, mouth gestures have less variation than sounds and several phonemes may share the same gesture so a class of visemes may contain many different phonemes. There are many choices of visemes [@bear2014phoneme] and Table \[TB:P2V\] shows one of those mappings [@neti2000audio]. [width=1]{} ---- --------------------- --------------------- ----- ------------------------------------ ------------------------------------------------------------------ ---- --------------- --------- TIMIT phonemes TIMIT phoneme TIMIT phoneme /A /l/ /el/ /r/ /y/ Alveolar-semivowels /V1 /ao/ /ah/ /aa/ /er/ /oy/ /aw/ /hh/ Lip-rounding based vowels /S /sil/ /sp/ Silence /B /s/ /z/ Alveolar-fricatives /V2 /uw/ /uh/ /ow/ “ & & &\ /C & /t/ /d/ /n/ /en/ & Alveolar & /V3 & /ae/ /eh/ /ey/ /ay/ & ” /D /sh/ /zh/ /ch/ /jh/ Palato-alveolar /V4 /ih/ /iy/ /ax/ " & & &\ /E & /p/ /b/ /m/ & Bilabial & & & & & &\ /F & /th/ /dh/ & Dental & & & & & &\ /G & /f/ /v/ & Labio-dental & & & & & &\ /H & /ng/ /g/ /k/ /w/ & Velar & & & & & &\ ---- --------------------- --------------------- ----- ------------------------------------ ------------------------------------------------------------------ ---- --------------- --------- : Neti [@neti2000audio] Phoneme-to-Viseme mapping.[]{data-label="TB:P2V"} Developing DNN-HMM based lipreading system ========================================== ![Lipreading system construction techniques.[]{data-label="Fig:Lip-method"}](AutomaticLipreadingMethod.pdf) Conventional techniques to model visual speech are based around Hidden Markov Models (HMMs) [@rabiner1986introduction]. The aim of the model is to find the most likely word or unit sequence corresponding to the visual observation. HMMs comprise two probability distributions: the transition probability and the probability density function (PDF) associated with the continuous outputs. The transition probabilities represent a first-order Markov process. The PDF of speech feature vectors is modeled by a Gaussian Mixture Model (GMM) that is parameterised by the mean and the variance of each component. There are some weaknesses of GMM, that have been found in acoustic modeling  [@Hinton2012]. First, it is statistically inefficient for modeling data that lie on or near a non-linear manifold in the data space. Second, in order to reduce the computational cost by using a diagonal rather than a full covariance matrix, uncorrelated features are needed. These deficiencies motivate the consideration of alternative learning techniques. The deep network structure can be considered as a feature extractor by using the number of neurons in multiple hidden layers to learn the essential patterns from the input features [@mao1995artificial]. In addition, the backpropagation algorithm [@hecht1988theory] with its appropriate learning criterion is essentially optimizing the model to fit to the training data discriminatively. However, to decode a speech signal, temporal features and models that can capture the sequential information in speech such as an observable Markov sequence in the HMM is still necessary. Thus arises the DNN-HMM hybrid structure in which the DNN is used instead of the GMM in the HMM. The method essentially combines the advantages from these two algorithms. Feature extraction ------------------ The literature provides a variety of feature extraction methods, often combined with tracking (which is essential if the head of the talker is moving). Here we focus on features that have been previously described as “bottom-up” [@Matthews02] meaning that they are derived directly from the pixel data and require only a Region-Of-Interest, or ROI. Figure \[rois\] illustrates a typical ROI taken from the TCD-TIMIT dataset described later in Section \[sec:data\] plus two associated feature representations which we now describe. -- -- -- -- -- -- ### Discrete Cosine Transform (DCT) The DCT is a popular transform in image coding and compression. DCT aims to represent the frequency domain of signal periodically and symmetrically using the cosine function. In particular, the DCT is a part of the Fourier Transform family but contains only the real part (Cosine). Because of its popularity most modern processors execute it very quickly (roughly $\mathcal{O}(N)$ for modern algorithms) so this also explains its ubiquity. For strongly correlated Markov processes the DCT approaches the Karhunen-Loeve transform in its compaction efficiency. Possibly this explains its popularity as a benchmark feature [@neti]. Here we use DCT II with zigzag property [@2ddct2], which means that the first elements of the feature vector contain the low-frequency information. The resulting feature vector has $44$ dimensions. ### Eigenlip The Eigenlip feature is another appearance-based approach[@KIRBY199363]. The Eigenlips feature has been generated via Principal Component Analysis (PCA) [@wold1987principal]. Here, we use PCA to extract the Eigenlip feature from grey-scale lip images, where we retain only $30$-dimensions of PCA. To construct the PCA, $25$-ROIs images of each training utterance were randomly selected to be the set of training images. Almost about $100k$ images in total were used to compute the Eigenvector and Eigenvalue and use it for extracting training and testing utterances. Only $30$ dimensions of principal components with high variation were retained. Feature transformation ---------------------- ![FMLLR feature pre-processing pipeline.[]{data-label="Fig:FMLLR"}](FMLLR.pdf) The raw features are $30$-dimensional Eigenlip and $44$-dimensional DCT features. These raw features are then normalized by subtracting the mean of each speaker and the $15$ consecutive frames are spliced onto the feature to add dynamic information. Second, Linear Discriminant Analysis (LDA) [@fisher1936use] and Maximum Likelihood Linear Transform (MLLT) [@Gales1998] are applied to reduce and map the features to a new space to minimize the within-class distance and maximize the between-class distance, where the class is the HMM-state, whilst simultaneously maximizing the observation likelihood in the original feature space. Finally, feature-space Maximum Likelihood Linear Regression (fMLLR) [@fMLLR2006],[@Gales1998], also known as the feature-space speaker adaptation technique, is employed to normalize the variation within a speaker. These new $40$-dimensional fMLLR features are used as inputs (labeled as feature pre-processing pipeline in Figure \[Fig:FMLLR\]) to the subsequent machine learning. The use of LDA is quite commonplace in lipreading and is derived in the HiLDA framework [@Potamianos2001]. MLLT and fMLLR are commonplace in acoustic speech recognition but have only recently been applied to visual speech recognition [@LipreadingSAT2016] albeit on smallish datasets. Visual speech model training ---------------------------- Our DNN-HMM visual speech model training involves all five successive stages. Here, we detail the development of the visual speech model that we employ in this work including all steps and parameters. ### Context-Independent Gaussian Mixture Model (CI-GMM) The first step is to initialise a model by creating a simple CI-GMM model. This step creates a time alignment for the entire training corpus by simply constructing a mono phoneme/viseme model that contains $3$-state GMM-HMMs for each speech unit. The CI-GMMs are trained on the raw features, which are DCT and PCA, along with its first and second derivative coefficients ($\Delta +\Delta\Delta$). We use $3$-state GMM-HMMs on each visual speech unit. Instead of setting the fixed number for increasing Gaussian mixture, we have set the maximum number of Gaussian to be $1000$ so that each state will keep increasing independently until their variances reach the maximum. When the training process starts, the time alignment of training data is equally segmented and updated in every iteration for the first ten iterations, then updated every two iterations until a maximum of $40$ iterations. ### Context-Dependent Gaussian Mixture Model (CD-GMM) The context-dependent viseme models (CD-GMMs) is specified on the same feature as in CI-GMM system. Here we use tied-state of $3$-context visual speech model, where the tied-states are obtained from the data-driven approach tree-clustering [@Povey11]. We have specified the maximum number of leaf nodes to be $2000$, which limits the number of states. The maximum number of Gaussians is set to $10K$. The training iterations continue until there is convergence which in practice is fewer than $35$ iterations. We realign every $10$ iterations. ### CD-GMM with LDA-MLLT feature transformation This training step also uses CD-GMMs, but trained on the LDA-MLLT features. The $40$-dimensional LDA-MLLT features are formed by splicing $15$ frames of the current frame (seven on the left and seven on the right) then reducing, via LDA, to $40$ dimensions per frame. This compact set of LDA-MLLT feature parameterizes to the $40$-dimension that best associates with the visual speech unit and also comprises the dynamic of visual speech over $150$ms. Again, the different set of tied-state CD-GMM has been constructed considered to the current feature. The maximum number of leaf nodes is set to $2,500$, and the total number of Gaussians is $15K$. This step utilises the equivalent number of training iterations and the realignment as those used in the previous step. ### CD-GMM with Speaker Adaptive Training (SAT) In a Speaker Adaptive Training (SAT) system, the CD-GMM are built on an fMLLR transformation on top of LDA-MLLT features by estimating a transform for each speaker. The same training process in the preceding step is then applied on the $40$-dimensions of fMLLR feature, where the number of leaf nodes and Gaussian are identical. ### Context-Dependent Deep Neural Networks (CD-DNN) We construct the CD-DNNs model on the hybrid DNN-HMMs architecture. The CD-DNNs are trained and optimized by minimizing frame-based cross-entropy between the prediction and the PDF target. The PDF refers to the tied-state context-dependent label, which is generated from the SAT system, that aligned every frame. The feature we adopted for all DNN training is based on LDA+MLLT+fMLLR features with mean and variance normalization. The CD-DNNs model is trained on six hidden layers with $2048$ neurons per layer, where we use the sigmoid non-linearity function in each neuron. The input layer is the fMLLR feature with temporally spliced $11$ consecutive frames. The model is initialized by a stacking of Recurrent Boltzman Machines (RBM) with three iterations on a single-GPU machine. The learning rate for RBM training is $0.4$ and applying L2 penalty (weight decay) at $0.0002$. The learning rate for fine-tuning has been set to $0.008$ with dropout of $0.1$. We use the minibatch-Stochastic Gradient Descent (SGD) for fine-tuning with minibatch size of $256$. We produce a development set for tuning the network by randomly selecting $10\%$ of training data. Every DNN training iteration is required to have a cross-validation loss is lower than the previous training iteration. If a iteration is rejected then one retries with a new stochastic gradient descent parameter. The terminating condition is that the new loss is little different from the old loss (specifically we use a difference smaller than $0.001$ of the loss as a suitable terminating condition). Decode lipreading with WFST Decoder ----------------------------------- Weighted Finite-state Transducer decoders have been increasingly used to decode speech signal in Large Vocabulary Continuous Speech Recognition (LVCSR) tasks and have also become a state-of-the-art decoder [@howell2016visual]. To decode a visual speech signal, we need a visual speech model, a language model, and a lexicon or as so called, a pronunciation dictionary. Our lipreading decoder comprises the visual speech DNN-HMM model,the TCD-TIMIT pronunciation dictionary and the word bi-gram language model. We generate the decoding graph as a finite-state transducer (FST) via the Kaldi toolkit  [@Povey11] .Beam width pruning is applied every $25$ frames where we use $13.0$ for the Viterbi pruning beam [@viterbi1967error] and $8.0$ for the lattice beam and the visual speech model scale is $0.1$. The lattice that contains the entire surviving path is re-scored by applying the bigram language model with the scaling factor over the range $5-15$. Only the lowest word error rates after LM re-scoring are used. Analysis of the pronunciation dictionary ======================================== Reducing the set of speech units, such as reducing a set of phonemes to a set of visemes, reduces the discriminative power of the classification model whilst increasing the complexity of pronunciation dictionary by increasing the volume of homophonic words. This suggests that word accuracy of a viseme based system will be lower than a phoneme based system. The counter argument is that visemes might be simpler to classify (because there are fewer of them and they are meant to be better matched to the visual signal) so there is clearly a trade-off between homopheny and unit accuracy [@cox2008challenge]. [width=0.7]{} Word Entry IPA Symbol Phoneme Dictionary Viseme Dictionary ------------ ------------ -------------------- ------------------- TALK t k t ao k C V1 H TONGUE t t ah ng C V1 H DOG d g d ao g C V1 H DUG d g d ah g C V1 H CARE k e r k eh r H V3 A WELL w e l w eh l H V3 A WHERE w e r w eh r H V3 A WEAR w e r w eh r H V3 A WHILE w ai l w ay l H V3 A : Example of phoneme and viseme dictionary with its corresponding IPA symbols.[]{data-label="tb-homophone"} Table \[tb-homophone\] shows examples of the homophoneme and homoviseme words that occur in the TCD-TIMIT dictionary. Figure 4 describes the homophone problem in two ways. On the left words are binned according to how many homophones they have. Thus the column labelled “1 occur” is the count of all unique words, the column labelled “2 occur” is the count of words that have one other homophone and so on. It is evident the switch to visemes causes more homophones particularly large numbers of high-multiplicity homophones. This effect can also be seen in the dictionary size (right of Figure 4). Homophones cause dictionary entries to merge so the visual dictionary is smaller than the acoustic one. {width="0.85\linewidth"} {width="\linewidth"} \[Fig:Occur\] Experiment methodology ====================== Data and Benchmarks {#sec:data} ------------------- We use the TCD-TIMIT [@TCDTIMIT] corpus containing $59$ volunteer speakers. We chose this dataset because it is the largest vocabulary audio-visual speech corpus available in the public domain. The WFST operates on a vocabulary of almost $6,000$ words from a dictionary of $160,000$ entries. This dataset provides lists of non-overlapping utterances for training and evaluation in two scenarios: speaker-dependent (SD) and speaker-independent (SI). In the SD scenario, visual models are trained on $3,752$ utterances and evaluated on $1,736$ utterances. Whereas in the SI experiment, $3,822$ utterances from $39$ talkers are in the training set and we evaluate on the remaining $17$ talkers containing a total $1,666$ utterances. The TCD-TIMIT release includes a baseline viseme accuracy for both speaker dependent and speaker independent settings using the Neti visemes [@neti2000audio] used here. The best viseme accuracy of recognizing $12$ viseme units reported on TCD-TIMIT is $34.77\%$ in speaker independent tests and $34.54\%$ on speaker dependent tests. The context independent viseme models (referred to as mono-viseme in the paper) were trained on $44$-coefficient DCT feature with $4$-state HMMs and $20$ Gaussian mixtures per state. Results ======= Viseme-based lipreading experiment ---------------------------------- One fundamental measure of the performance of an automatic lipreading system is viseme accuracy. Since the viseme recognizer requires no dictionary or language model, it is quicker to build and optimise. Table \[tab:visemeaccuracy\] lists the accuracies achieved with our viseme based lipreading system. In comparison to the viseme accuracies benchmarked with the TCD-TIMIT corpus, our best SD viseme accuracy is $46.61\%$ with Eigenlips, compared to $34.54\%$, an improvement of $12.07\%$. Our best SI viseme accuracy is $44.61\%$ which improves on the benchmark $34.77\%$ by $10.16\%$, again with the Eigenlips features. [width=0.7]{} -------------- -- ------- ------- ------- ------- SD SI SD SI CD-GMM + SAT 44.66 42.48 14.37 10.47 CD-DNN 43.67 38.00 23.89 9.17 CD-GMM + SAT 45.59 44.61 16.71 12.15 CD-DNN 46.61 44.60 33.06 19.15 -------------- -- ------- ------- ------- ------- : Viseme-based lipreading accuracy (%).[]{data-label="TB:Beseline-wordacc"} \[tab:visemeaccuracy\] Word accuracy achieved with visemes, albeit lower than the viseme accuracy, also shows that Eigenlip features outperform the DCT: we achieved $33.06\%$ in speaker dependent tests, and $19.15\%$ in speaker independent tests. Phoneme-based lipreading experiment ----------------------------------- Table \[tab:phoemeaccuracy\] shows the word and phoneme accuracies achieved with our phoneme-based lipreading system. This system achieved the most accurate lipreading with a word accuracy of $48.74\%$. It is interesting that with the phoneme recogniser, word accuracy is greater than phoneme accuracy, because in the viseme recogniser, this is vice versa. Again, highest accuracy is achieved with Eigenlip features rather than DCT. [width=0.7]{} -------------- -- ------- ------- ----------- ----------- SD SI SD SI CD-GMM + SAT 28.22 27.37 21.88 17.72 CD-DNN 29.18 28.08 37.40 33.87 CD-GMM + SAT 31.14 29.59 28.79 24.57 CD-DNN 33.44 31.10 **48.74** **42.97** -------------- -- ------- ------- ----------- ----------- : Phoneme-based lipreading accuracy(%).[]{data-label="TB:Beseline-wordacc"} \[tab:phoemeaccuracy\] One interesting observation apparent in Tables \[tab:visemeaccuracy\] and \[tab:phoemeaccuracy\] is that the introduction of the DNN makes little difference to the unit accuracy but a bigger difference to a word accuracy for both DCT and eignlips features. Discussion ---------- Figure \[Fig:LipReadingReview\] plots all of our experimental results comparing unit accuracies (along the $x$-axis) against the word accuracies (on the $y$-axis) along with errorbars showing $\pm1$ standard error. ![Lipreading system performance in GMM system.[]{data-label="Fig:LipReadingReview"}](WordvsUnit.pdf) Figure \[Fig:LipReadingReview\] has two clusters: one, in the bottom right, represents the viseme experiments and the other, on the upper left the phonemes. Here we are representing viseme classifiers with circles (filled represents the DNN, open the GMM) and the phonemes with squares (either filled or open depending on the classifier). The colours represent the various SI/SD or DCT/Eigenlips combinations. The phoneme recogniser naturally obtained lower unit accuracy scores because it has three times more phoneme classes than viseme classes ($13$ to $38$ respectively). But this does not mean that phoneme classes have less power to model a visual gesture. This is visualised in the confusion matrices in Figure \[fig:confusionmatrixes\] where the colour patterns are consistent between phoneme classes (on the left of Fig \[fig:confusionmatrixes\] and between viseme classes on the right of Fig \[fig:confusionmatrixes\]). ![Comparison of visemes confusion matrix (left) vs phomemes confusion matrix (right).[]{data-label="fig:confusionmatrixes"}](VisemeCon2.pdf "fig:"){width="0.92\linewidth"} ![Comparison of visemes confusion matrix (left) vs phomemes confusion matrix (right).[]{data-label="fig:confusionmatrixes"}](PhoneCon2.pdf "fig:"){width="0.92\linewidth"} We note that reducing the set of visual speech units also reduces the discriminant power of the classification model whilst increasing the complexity of pronunciation dictionary by increasing the volume of homophone (homoviseme) words. This suggests that word accuracy of a viseme based system will be less likely to outperform the phoneme based system. One of the disadvantages of the DNN is that it is not easy to examine to internals of the network to discover from where it is getting its performance. However there is a clue in the previous observation which is that the DNN appears to make the most difference to word accuracy rather than unit accuracy. Visual speech is notorious for extensive co-articulation so the implication is that either there are significant differences in the window length between the GMM and the DNN or the DNN is better able to model co-articulation than the GMM. Although there are variations in the window length, here the GMM has a slightly longer span of $150$ms compared to $110$ms for the DNN, it is latter explanation is the most likely. In other work [@LipreadingIS2017] we were able to use identical features and we also found the DNN superior furthermore we know the DNN to be better able to learn data structured on non-linear manifolds so we believe this is the most likely explanation for the success of the DNN. One caveat is that we have not optimised the scaling factor used in our language models so there is probably more performance to come when measured as word accuracy. Conclusion ========== One observation we found is that DNN-HMM viseme recognizers can easily overfit to the training observations, this is shown in the performance disparity between SD and SI configurations. It could potentially be interesting to use visemes as an initialisation for phoneme recognition in a hierarchical training method similar to that in [@bear2016decoding] in the future. We have added more evidence to the argument that phoneme classifiers can outperform those of visemes. Whilst there is still debate about visemes, we can not forget them, but given the evidence showing a significant improvement in word accuracy from the reduction in homophonic words in a pronunciation dictionary, we suggest that phonemes are the current optimal class labels for lipreading. We have also illustrated the noticeable performance gain by changing visual representation from DCT to Eigenlips. The best word accuracy in this work is $48.74\%$ on SD and $42.97\%$ on SI achieved with the DNN-HMM phoneme unit recognizer trained on Eigenlip features. However, the disadvantage of Eigenlip feature is a learned linear mapping that needs to be trained. Conventional systems have shown speaker independence to be a challenge, here with a novel DNN-HMM architecture, we have reduced the effect between these arrangements. We speculate that the success of the DNN is likely to do its ability to better model the effects of co-articulation which is a well known bugbear of human and machine lip-readers.

ArXiv

--- abstract: 'Some phase space transport properties for a conservative bouncer model are studied. The dynamics of the model is described by using a two-dimensional measure preserving mapping for the variables velocity and time. The system is characterized by a control parameter $\epsilon$ and experiences a transition from integrable ($\epsilon=0$) to non integrable ($\epsilon\ne 0$). For small values of $\epsilon$, the phase space shows a mixed structure where periodic islands, chaotic seas and invariant tori coexist. As the parameter $\epsilon$ increases and reaches a critical value $\epsilon_c$ all invariant tori are destroyed and the chaotic sea spreads over the phase space leading the particle to diffuse in velocity and experience Fermi acceleration (unlimited energy growth). During the dynamics the particle can be temporarily trapped near periodic and stable regions. We use the finite time Lyapunov exponent to visualize this effect. The survival probability was used to obtain some of the transport properties in the phase space. For large $\epsilon$, the survival probability decays exponentially when it turns into a slower decay as the control parameter $\epsilon$ is reduced. The slower decay is related to trapping dynamics, slowing the Fermi Acceleration, i.e., unbounded growth of the velocity' author: - 'André L. P. Livorati$^1$, Tiago Kroetz$^{2}$, Carl P. Dettmann$^{3,4}$, Iberê Luiz Caldas$^{1}$ and Edson D. Leonel$^{4,5}$' title: 'Stickiness in a bouncer model: A slowing mechanism for Fermi acceleration' --- Introduction {#sec1} ============ It is known that Hamiltonian systems are typical non-ergodic and non-integrable [@ref1]. The phase space of such systems is divided into regions with regular and chaotic dynamics. These dynamical regions are connected by a layer, where regular or irregular motion, can or cannot mix, depending upon on the number of degrees of freedom of the system, as well properties of the limiting surface itself. Such a division leads to the stickiness phenomenon [@ref2; @ref3] which is manifested through the fact that a phase trajectory in a chaotic region passing near enough a Kolmogorov-Arnold-Moser (KAM) island, evolves there almost regularly during a time that may be very long. However, when an orbit resides in a chaotic region far from the set of KAM regions, it moves chaotically in the sense that two nearby initial conditions apart from each other exponentially as the time evolves. Therefore the stickiness of phase trajectories has a crucial influence on the transport properties of Hamiltonian systems, and its relation to physical systems is one of the most important open problems of nonlinear dynamics [@ref4; @ref4a]. Applications of stickiness can be found in astronomy [@ref5], fluid mechanics [@ref6], Levy flights [@ref7], also in biology [@ref8], in plasma physics [@ref9; @ref10] and many others. One of the main consequences of the influence of orbits in sticky regime is observed in the transport properties along the phase space. Therefore it may give rise to the following question: May sticky orbits influence the Fermi acceleration phenomenon? Fermi acceleration (FA) was introduced by the first time in 1949 by Enrico Fermi [@ref11] as an attempt to explain the possible origin of the high energies of the cosmic rays. Fermi claimed that the charged cosmic particles could acquire energy from the moving magnetic fields present in the cosmos. His original idea generated a prototype model which exhibits unlimited energy growth and is called the bouncer model. The model consists of a free particle (making allusions to the cosmic particles) which is falling under influence of a constant gravitational field $g$ (a mechanism to inject the particle back to the collision zone) and suffering collisions with a heavy and time-periodic moving wall (denoting the magnetic fields). The model is characterized by a control parameter $\epsilon$ and has a transition from integrability $\epsilon=0$ to non integrability $\epsilon\ne 0$. A mixed structure of the phase space is observed for lower values of $\epsilon$ and strong chaotic properties are present in the regime of large values of the parameter, say $\epsilon>\epsilon_c$ where at $\epsilon_c$ the system experiences a transition from local to globally chaotic regime (destruction of invariant spanning curves). In this paper we revisit the bouncer model seeking to understand and describe some transport properties along the phase space particularly focusing on the dynamics of sticky orbits. The model is described by a two dimensional, nonlinear and measure preserving mapping for the variables velocity of the particle and time at the collision with the moving wall. As the parameter $\epsilon$ is increased, the number of islands in the phase space decreases. For the regime of high nonlinearity $\epsilon\gg 1$, almost no islands are observed. The temporarily trapping dynamics due to the sticky regions are more often observed in the regime of small $\epsilon$ where a mixed structure of the phase space is present. We use the finite time Lyapunov exponent spectrum of the orbits and a statistical analysis of escape rates to investigate the influence of the stickiness in dynamics of an ensemble of non interacting particles. We therefore conclude that the stickiness present in the system acts as a slowing mechanism for FA. The paper is organized as follows: In Sec. \[sec2\] the mapping that describes the dynamics of the model is obtained. In Sec.\[sec3\], the numerical results are present which include the calculation of the finite time Lyapunov exponent and escape rates for the velocity as a function of $\epsilon$. Finally, the conclusions and final remarks are drawn in Sec.\[sec4\]. The model, the mapping and chaotic properties {#sec2} ============================================= We discuss in this section the procedures used to construct the mapping that describes the dynamics of the system. The model consists of a classical particle of mass $m$ which is moving in the vertical direction under the influence of a constant gravitational field $g$. It also suffers elastic collisions with a periodically moving wall whose position is given by $y(t)=\varepsilon \cos(wt)$, where $w$ is the frequency and $\varepsilon$ is the amplitude of oscillation respectively. The dynamics of the system is made by the use of a two dimensional, nonlinear and measure preserving mapping for the variables velocity of the particle $v$ and time $t$ immediately after a $n^{th}$ collision of the particle with the moving wall. During the dynamics, two distinct kinds of collisions may be observed: (i) multiple collisions of the particle with the moving wall – those happening before the particle leaves the collision zone (the collision zone is defined as the region $y\in[-\varepsilon,\varepsilon]$) – or; (ii) a single collision of the particle with the moving wall (causing the particle to leave the collision zone). Before writing the equations of the mapping, it is important to mention there are an excessive number of control parameters, 3 in total, namely $\varepsilon$, $g$ and $w$. We may define dimensionless and more convenient variables as: $V_n=v_n w/g$, $\epsilon=\varepsilon w^2/g$ and measure the time in terms of the number of oscillations of the moving wall $\phi_n=w t_n$. We assume that at the instant $\phi\in[0,2\pi]$ the position of the particle is $y_p(\phi_n)=\epsilon\cos(\phi_n)$ with initial velocity $V_n>0$, which lead us to obtain the following expression for the mapping $$T_c:\left\{\begin{array}{ll} V_{n+1}=-{V_n^*}+{\phi_c}-2\epsilon\sin(\phi_{n+1})\\ \phi_{n+1}=[\phi_n+\Delta T_n]~~{\rm mod (2\pi)}\\ \end{array} \right., \label{eq1}$$ where the index $c$ stands for the complete version of the model (the one which takes into account the movement of the moving wall) and the expressions for $V_n^*$ and $\Delta T_n$ depend on what kind of collision happens. For case (i), i.e. the multiple collisions, the expressions are $V_n^*=V_n$ and $\Delta T_n=\phi_c$ where $\phi_c$ is obtained from the condition that matches the same position for the particle and the moving wall. It leads to the following transcendental equation that must be solved numerically $$G(\phi_c)=\epsilon\cos(\phi_n+\phi_c)-\epsilon\cos(\phi_n)-V_n\phi_c+{{ 1}\over{2}}\phi_c^2~. \label{eq2}$$ If the particle leaves the collision zone case (ii) applies. The expressions are $V_n^*=-\sqrt{V_n^2+2\epsilon(\cos(\phi_n)-1)}$ and $\Delta T_n=\phi_u+\phi_d+\phi_c$ with $\phi_u=V_n$ denoting the time spent by the particle in the upward direction up to reaching the null velocity, $\phi_d=\sqrt{V_n^2+2\epsilon(\cos(\phi_n)-1)}$ corresponds to the time that the particle spends from the place where it had zero velocity up to the entrance of the collision zone at $\epsilon$. Finally the term $\phi_c$ has to be obtained numerically from the equation $F(\phi_c)=0$ where $$F(\phi_c)=\epsilon\cos(\phi_n+\phi_u+\phi_d+\phi_c)-\epsilon-V_n^* \phi_c+{{1}\over{2}}\phi_c^2~. \label{eq3}$$ {width="17cm" height="12.0cm"} The extended phase space for the whole version of the model considers four variables namely: (1) $x_w$ denoting the position of the moving wall; (2) $V_p$ corresponding to the velocity of the particle; (3) $E_p$ which is the energy of the particle and (4) the time $t$. The canonical pairs however are: position and velocity $(x_w,V_p)$ and; energy and time $(E_p,t)$. As the way the mapping was constructed, the variables used are not canonical ones therefore the determinant of the Jacobian matrix is $${\rm Det~J}=\left[{{V_n+\epsilon\sin(\phi_n)}\over{V_{n+1} +\epsilon\sin(\phi_{n+1})}}\right]~, \label{eq4}$$ which is clearly different from unity as it should be if the canonical pair was considered. However we may say that it preserves the following measure in the phase space $d\mu=[V+\epsilon\sin(\phi)]dVd\phi$. A common version which is also present in the literature is the so called simplified version. It was proposed many years ago [@ref17] as an attempt to keep the essence of the problem but at the same time allow numerical computations to be realized in a reasonable time when computers were far slower. Also it could reduce the complexity of the equations at a level that analytical calculations could be obtained. It assumes that the wall is fixed – so that the calculation of the time between collision does not evolve numerical solution of transcendental equations –, but at the instant of the collision, the particle suffers an exchange of energy and momentum as if the wall were moving. In this version, the extended phase does not consider more the position of the moving wall, because by definition it is fixed, causing the canonical pair to be the velocity and time. The mapping is then written as $$T_s:\left\{\begin{array}{ll} V_{n+1}= | V_n-2\epsilon\sin(\phi_{n+1})|\\ \phi_{n+1}=\phi_n+2V_n~~{\rm mod (2\pi)}\\ \end{array} \right., \label{eq5}$$ where the modulus function is introduced to avoid the particle to move beyond the wall. After a collision, if the particle has a negative velocity, we re-inject it back with the same velocity. For the simplified version and given the variables describing the dynamics are the canonical pair, the determinant of the Jacobian is given by ${\rm Det~J}=\pm 1$. The simplified version of the model also allow us to make a connection with the so called standard mapping. Defining $I_n=2V_n$, $K=4\epsilon$ and $\theta_n=\phi_{n+1}+\pi$ the simplified version is written as the standard mapping. The variation of the control parameter $\epsilon$ leads the dynamics to experience a transition from locally to globally chaotic dynamics as similarly observed in the standard mapping [@ref33]. Indeed for $\epsilon<\epsilon_c\approx0.2429$ the phase space has invariant spanning curves (also called invariant tori) and unlimited energy growth, which characterizes FA, is not observed. As the parameter $\epsilon$ is increased, the fixed points become unstable and bifurcate for $\epsilon>1$ ($K>4$). The period-1 fixed points are obtained solving the two equations simultaneously $V_{n+1}=V_n=V^*$ and $\phi_{n+1}=\phi_n=\phi^*$ and are given by $$\begin{aligned} V^*&=&\pi l~~~,~~~ l=0,1,2,\ldots\label{eq7a}\\ \phi^*&=&\arcsin \left({\pi m \over 2\epsilon} \right)~~~,~~~m=0,1,2,\ldots \label{eq7}\end{aligned}$$ Thus, there are windows of periodicity for the period one fixed points which depend on $\epsilon$. The linear stability for these fixed points are given by $$(2\pi)^2(p-1)^2<16\epsilon^2<(2\pi)^2(p-1)^2+4~, \label{eq8}$$ where $p=l-m$. Figure \[fig2\] shows the structure of the phase space for the complete version of the bouncer as a function of the control parameter $\epsilon$. The accuracy used to solve numerically both $F$ and $G$ was $10^{-12}$ using the bisection method. As $\epsilon$ is increased the stable regions (mainly marked by periodic fixed points) reduce leading the phase space to have large unstable regions. The regions of sticky are more often observed for smaller values of $\epsilon$ due to the existence of many islands in the phase space as compared to large values of $\epsilon$. Analyzing Fig. \[fig2\] we see that the phase space has a repeating structure in $\pi$ in the velocity axis. Thus, let us plot the phase space taking the $mod(\pi)$ for velocity. Such a plot is useful for observing the evolution of the fixed points and the possible trappings caused by sticky orbits. The control parameters used to construct Fig. \[fig2\] were: (a) and (e) $\epsilon=0.40$; (b) and (f) $\epsilon=0.60$; (c) and (g) $\epsilon=0.80$; and (d) and (h) $\epsilon=1.20$. For each figure a set of $100$ different initial conditions were evolved in time until $10^5$ collisions with the moving wall. The initial velocity was chosen such that its minimum value was higher than the stable region in $V\in[0,2\epsilon]$. ![(a) Plot of the FTLE for an initial condition chosen in the chaotic sea. (b) shows the evolution of the same initial condition of (a) for a plot of velocity against the number of collisions. (c) the zoom-in window of the previously selected area of (b) showing these trapping orbits in the phase space coordinates $(V,\phi)$.[]{data-label="fig3"}](Fig2.eps){width="1.0\linewidth"} Numerical Results {#sec3} ================= This section is divided in two parts. In the first one we discuss the results for the Lyapunov exponent obtained at finite time while in the second we present our discussions and show results for orbits that survive longer the dynamics after being trapped by some sticky regions. Lyapunov exponents ------------------ Let us start discussing our results for the positive Lyapunov exponent for chaotic components of the phase space. The Lyapunov exponent has been widely used to quantify the average expansion or contraction rate for a small volume of initial conditions. If the Lyapunov exponent is positive, the orbit is said to be chaotic leading to an exponential separation of two nearby initial conditions. On the other hand, a non positive Lyapunov exponent indicates regularity and the dynamics can be in principle periodic or quasi-periodic. The Lyapunov exponents are defined as follows [@ref35] (see for example [@ref36] for applications in higher dimensional systems): $$\lambda_j = \lim_{n\rightarrow\infty} {1\over{n}} \ln |{\Lambda_j^n}|, ~~~j=1,2~, \label{eq12}$$ where $\Lambda_j^n$, are the eigenvalues of the matrix $M={\prod_{i=1}^n} J(V_i,\phi_i)$ and $J_i$ is the Jacobian matrix evaluated over the orbit. ![(Color online). Plot of the FTLE distributions for several values of the parameter $\epsilon$. One sees two distinct peaks, a larger one representing the mean value of the Lyapunov exponent, and the secondary one, is due to orbits in stickiness regime. As $\epsilon$ increases, the magnitude of the secondary peak decreases indicating that for higher values of $\epsilon$, less sticky orbits are observed. The control parameters used were: (a) $\epsilon=0.3$; (b) $\epsilon=0.35$; (c) $\epsilon=0.45$; (d) $\epsilon=0.55$; (e) $\epsilon=0.70$; (f) $\epsilon=0.80$; (g) $\epsilon=1.0$; (h) $\epsilon=1.2$.[]{data-label="fig4"}](Fig3.eps){width="1.0\linewidth"} {width="16cm" height="12.0cm"} {width="17cm" height="12.0cm"} In the dynamics of the bouncer model, chaotic and regular motion can coexist in the phase space, which introduces large variations and local instability along a reference chaotic trajectory. Such variations, are related to alternations between different motions, in a qualitative way of saying, as well as chaotic and quasi-regular motions. In order to characterize such peculiar variation dynamics, we used the Finite-Time Lyapunov Exponent (FTLE) [@ref27]. Once the trappings caused by orbits in stickiness regime happen just for a finite time, this technique is useful to quantify the trapping effects. It was shown [@ref27] that when the FTLE distributions present small values, it is related to existence of long-lived jets from a two-dimensional model for fluid mixing and transport. This can be understood, in a dynamic point of view as stickiness trajectories in the phase space. Figure \[fig3\](a) shows the evolution of the FTLE, for an initial condition chosen in the chaotic sea, for $\epsilon=0.4$. One sees a very irregular behavior along the time, alternating average contractions and repulsions, leading to and average value as $\bar{\lambda}=0.3078(1)$. In Fig. \[fig3\](b) it is shown the evolution of the same initial condition of Fig. \[fig3\](a) however plotted the velocity as a function of the number of collisions. It is clear in Fig. \[fig3\](b) the successive trappings along the orbit, and how they “slow down” the energy growth, that characterizes the FA. Also, we set a zoom-in window in Fig. \[fig3\](b) and plot the corresponding orbit in the phase space portrait $(V,\phi)$, in order to identify some of these stickiness orbits in Fig. \[fig3\](c). To optimize the window of time to be used in the FTLE calculations, we have considered different lengths in several simulations. After some comparisons of the results we come up, based in fluctuations of the Lyapunov exponents, to a finite time of $100$ collisions that was then used to study the distribution of FTLE. It is known in the literature [@ref27]that the FTLE distribution has a Gaussian shape, where the large peak can be interpreted as the mean value of the Lyapunov exponent. If the system presents any periodic or quasi-periodic motion, besides chaos in its dynamics, the FTLE distribution can have a secondary peak in the region of very low value of the Lyapunov exponent. Such secondary peak is interpreted as sticky orbits along the dynamics evolution [@ref36; @ref27] responsible for trapping the dynamics. The distribution for several FTLE are shown in Fig. \[fig4\] for different control parameter $\epsilon$ as labeled in the figure. We can see from Fig. \[fig4\] that the secondary peak of the FTLE distribution is more evident for small values of $\epsilon$. Just to have a glance of the influence of the second peak in the distribution represents up about $20\%$ of the whole distribution of Fig. \[fig4\](b). The fraction of the distribution of the FTLE for the secondary peak decreases as $\epsilon$ is increased. Such a result is expected because for higher values of $\epsilon$ less islands in the phase space are observed as previously shown in Fig. \[fig2\]. Survival Probability and Escape Rates ------------------------------------- In this section we discuss results for orbits that survive until reaching a pre-defined velocity at which they are assumed to escape. To do that we consider the existence of a hole in the velocity coordinate of the phase space. If the particle reaches such a velocity or higher, its dynamics is stopped and a new initial condition is started. The introduction of the hole allow us to study transport properties as well as characterize, through statistical analysis of survival probability and time-correlation decays, the influence of sticky orbits along the dynamics of the model [@ref37; @ref38; @ref39]. To study the transport properties, we set a grid of initial conditions equally distributed along the velocity and phase. Indeed a grid of $500\times500$ initial conditions with $V_0\in[\epsilon,30]$ and $\phi_0\in[0,2\pi]$ were considered. Then each initial condition was evolved in time up to the limit of $10^5$ collisions with the moving wall or until a hole placed in the velocity axis at $V_{\rm hole}=30$ is reached. Figure \[fig5\] shows a plot of the initial conditions evolved until $10^5$ collisions with the moving wall or up to the particle reaching the hole. The color ranging from red (fast escape) to blue (long time dynamics) denotes the time (plotted in logarithmic scale) the particle spends until reaching the escape velocity. White regions denote that the particle never escaped. The control parameters used to construct the figures were: (a) $\epsilon=0.4$; (b) $\epsilon=0.6$; (c)$\epsilon=0.8$ and; (d) $\epsilon=1.2$. We see from Fig. \[fig5\](a) where $\epsilon=0.40$, that low initial velocities spend large time accumulating energy until reach the hole at $V=30$. Additionally one sees many stability islands where the orbits can get temporally trapped and been released after a while. These temporally trappings are caused by sticky regions. Such dynamical regimes can be visualized by the dark regions marked by blue color in Figs. \[fig5\](b,c) whose control parameters are respectively $\epsilon=0.6$ and $\epsilon=0.8$. When the control parameter $\epsilon$ is raised, the particles reach the hole faster as we can see from Figs. \[fig5\](b,c,d). In particular for Fig. \[fig5\](c) one sees that the first stability island disappeared. The stability regions are getting smaller and smaller as the control parameter $\epsilon$ raises and from Fig. \[fig5\](d) they appear to be very small for $\epsilon=1.2$. However even for a control parameter where the stability islands are small, we see that the sticky orbits are still present and indeed are marked by the dark blue color in the plot. The statistics of the cumulative recurrence time distribution which is obtained from the integration of the frequency histogram distribution for the escape can also be obtained. To do that we consider now that the escaping velocity is set as $V_{\rm hole}=100$ although any other velocity could be considered. Their cumulative recurrence time distribution is also called survival probability and is obtained as $$P_{surv}={1 \over N} \sum_{j=1}^N N_{rec}(n)~, \label{eq13}$$ where, the summation is taken along an ensemble of $N=10^6$ different initial conditions. The term $N_{rec}(n)$ indicates the number of initial conditions that do not escape through the hole at $V_{\rm hole}=100$ (i.e. recur), until a collision $n$. The ensemble of initial conditions was set for a constant velocity as $V_0=2\pi$ while $10^6$ phase were distributed evenly in $\phi_0\in[2.8,3.2]$. $\epsilon$ $\epsilon-\epsilon_c$ $-\zeta$ $-\gamma$ ------------ ----------------------- --------------- ------------ -- -- $1.40$ $1.1557025$ $1.404(6)E-3$ $-$ $1.30$ $1.057352$ $1.036(3)E-3$ $-$ $1.20$ $0.957025$ $7.219(5)E-4$ $-$ $1.10$ $0.857025$ $4.675(3)E-4$ $2.92(1)$ $1.00$ $0.757025$ $3.430(7)E-4$ $2.18(1)$ $0.90$ $0.657025$ $2.739(2)E-4$ $1.95(1)$ $0.80$ $0.557025$ $2.105(2)E-4$ $1.625(9)$ $0.70$ $0.457025$ $1.218(4)E-4$ $1.73(3)$ $0.60$ $0.357025$ $5.260(1)E-5$ $2.16(2)$ $0.575$ $0.332025$ $4.387(9)E-5$ $1.79(3)$ $0.55$ $0.307025$ $3.71(7)E-5$ $1.52(1)$ $0.525$ $0.282025$ $3.101(7)E-5$ $1.70(1)$ $0.50$ $0.257025$ $2.463(5)E-5$ $1.91(9)$ $0.475$ $0.232025$ $2.280(1)E-5$ $1.29(1)$ $0.45$ $0.207025$ $1.408(3)E-5$ $1.71(1)$ $0.425$ $0.182025$ $7.654(3)E-6$ $1.45(2)$ $0.40$ $0.157025$ $5.73(9)E-6$ $1.90(5)$ $0.375$ $0.132025$ $3.25(3)E-6$ $1.90(2)$ $0.35$ $0.107025$ $1.536(4)E-6$ $1.84(1)$ : Exponents obtained from numerical fitting for the curves of $P_{surv}$ for different values of $\epsilon$.[]{data-label="Tab1"} It is known in the literature that if a system has fully chaotic behavior the curves of $P_{surv}$ have an exponential decay [@ref40]. However, when a mixed dynamics is observed in the phase space, the curves of $P_{surv}$ may present different behaviors that may include: (i) a power law decay [@ref41] or; (ii) a stretched exponential decay [@ref42]. For the bouncer model which has a mixed phase space the curves of $P_{surv}$ may present either behaviors, depending on the parameter $\epsilon$ and the set of initial conditions, as shown in Fig. \[fig6\]. We see a transition in the behavior of the curves of $P_{surv}$ as the parameter $\epsilon$ is decreased. For large values of $\epsilon$ as for example $\epsilon=1.4$ and $\epsilon=1.3$, the phase space has quite few islands and the chaotic sea is dominant over the dynamics. It is therefore expected an exponential decay in the curves of $P_{surv}$, as shown in Figs. \[fig6\](a,b). As the parameter $\epsilon$ is getting smaller, more and more stability islands appear in the phase space leading to the appearance of more and more sticky regions. With these stable regions around in the phase space, a change in the behavior of the curves of $P_{surv}$ is expected. For values of $\epsilon<1$, we may observe a combination of decays in the curves of $P_{surv}$. Firstly the curves exhibit an exponential decay and suddenly they change to a slow decay that we observed to be described as a power law which marks the presence of orbits in stickiness regime [@ref41]. Considering the curves of the survival probability shown in Fig. \[fig6\], a numerical fitting can be made therefore according to: (i) the exponential decay is given as $P_{surv}(n)\propto \exp(n\zeta)$ while; (ii) the power law decay is described by $P_{surv}(n)\propto n^{\gamma}$ where $\zeta$ and $\gamma$ are respectively the exponents for exponential and power law time decays. Table \[Tab1\] shows the set of exponents for different values of the control parameter $\epsilon$. ![(Color online). Plot of $-\zeta$ and $-\gamma$ as a function of $\epsilon-\epsilon_c$.[]{data-label="fig7"}](Fig6.eps){width="1.0\linewidth"} ![(Color online). Plot of $\overline{V}$ as function of $n$ for: (a) $\epsilon=0.8$ and (b)$\epsilon=0.525$. One can see two distinct growth exponents for Regular Fermi Acceleration and Sticky Fermi Acceleration. Such difference can be undesrtood as sticky orbits acting as a slowing mehanism for FA.[]{data-label="fig8"}](Fig7.eps){width="1.0\linewidth"} We see that as the parameter $\epsilon$ decreases the exponential decay of the curves of $P_{surv}$ also suffer a change. The exponent $\zeta$ decreases too as $\epsilon$ decreases, a result which is quite expected given the periodic regions of the phase space are getting larger and larger. Figure \[fig7\](a) shows the behavior of the exponent $\zeta$ as a function of $(\epsilon-\epsilon_c)$. Looking at Fig. \[fig7\](a) we see that the exponent $\zeta$ can be described by a power law of the type $-\zeta\propto(\epsilon-\epsilon_c)^z$ and that the slope of the power law is given by $z=2.719(4)$. The exponent $\gamma$ however does not show the mathematical beauty as observed for the exponent $\zeta$. The slower decay observed in the curves of the survival probability is indeed due to sticky regions present in the phase space. For our simulations, most of the slower decay was characterized as a power law. Indeed in the literature, it is known that the power law decay, for such cumulative recurrence time distribution for other dynamical systems [@ref43; @ref44] which includes also billiards systems [@ref41; @ref45; @ref46; @ref47; @ref48] is set in a range of $-\gamma \in [1.5,2.5]$ and that our results match this range. We stress however that the total understanding and this behavior is still an open problem and extensive theoretical and numerical simulations, are required to describe its behavior properly. Let us now address specifically the assumption that stickness may affect the phenomenon of Fermi acceleration. Indeed the trapping dynamics of the particles around stable regions makes the unlimited energy growth slower than the usual. For a large set of initial conditions that lead the dynamics of the particle to present diffusion in the velocity, the average velocity $\bar{V}$ is described by $\bar{V}\propto \sqrt{n}$. However we expect the initial conditions that spend large time trapped in sticky regions lead the slope of growth to be smaller than $1/2$. This is indeed true and figure \[fig8\] confirms this assumption. The curves shown in bullets in both Fig. \[fig8\](a,b) are named as Regular Fermi Acceleration (RFA) and were obtained for evolution of the initial conditions which produce a fast decay in the survival probability (those along the exponential decay in Fig. \[fig6\]) and as expected, an exponent of $\cong 0.5$ was obtained. On the other hand, the curves plotted as squares show the evolution of initial conditions chosen in the very final tail of the power law decay shown in Fig. \[fig6\] and are called as Sticky Fermi Acceleration (SFA). Power law fitting furnish slopes $0.398(7)$ for (a) and $0.400(1)$ for (b). These curves indeed give support for our claim that sticky regions slow down the Fermi acceleration. Final Remarks and Conclusions {#sec4} ============================= The dynamics of the bouncer model was investigated by using a two dimensional measure preserving mapping controlled by a single control parameter $\epsilon$. For $\epsilon=0$ the system is integrable while it is non integrable for $\epsilon\ne 0$. As soon as $\epsilon$ increases, the periodic regions of the phase space reduce given rise to chaotic dynamics. Indeed for $\epsilon>\epsilon_c$ invariant tori are not observed in the phase space while periodic regions are still observed. The influence of sticky regions also reduces with the increase of $\epsilon$. Our numerical investigation of the FTLE spectrum distribution give support that trapping dynamics is often observed in the phase space and is confirmed by the secondary peaks of the FTLE distribution. The survival probability is characterized by two decaying regimes: (1) for strong chaotic dynamics, the decay is given by an exponential type while (2) it changes to a slower decay marked by a power law type when mixed dynamics is present in the phase space. Finally, according to the results shown in Fig.\[fig8\], we see that when a strong regime of stickiness is present in the system, it acts as a slowing mechanism for FA. As with the survival probability, it would interesting to investigate whether the stickiness associated with mixed phase space in general models leads to a universal “slowing exponen”. ALPL acknowledges CNPq for financial support. CPD thanks Pró Reitoria de Pesquisa - PROPe/UNESP and DEMAC for hospitality during his stay in Brazil. ILC thanks CNPq and FAPESP. EDL acknowledges CNPq, FAPESP and FUNDUNESP, Brazilian agencies. This research was supported by resources supplied by the Center for Scientific Computing (NCC/GridUNESP) of the São Paulo State University (UNESP). [9]{} L. Markus, K. R. Meyer, “Mem. Am. Math. Soc." **144**, 1 (1978). A. Perry, S. Wiggins, Physica D, **71**, 102 (1994). G. M. Zaslasvsky, “Hamiltonian Chaos and Fractional Dynamics", Oxford University Press, New York (2008). L.A.Bunimovich, Nonlinearity, **21**, T13 (2008). L. A. Bunimovich and L. V. Vela-Arevalo, Chaos, [**22**]{}, 026103, 2012. G. Contopoulos e M. Harsoula, Celest. Mech. Dyn. Astr., [**107**]{}, 77 (2010). T. H. Solomon, E. R. Weeks, H. L. Swinney, Phys. Rev. Lett., [**71**]{}, 3975, (1993). M. F. Shlesinger, G. M. Zaslasvsky e J. Klafter, Nature, [**363**]{}, 31 (1993). T. T ́l, A. de Moura, C. Grebogi e G. K ́rolyi, Phys. Rep., [**413**]{}, 91 (2005). D. del-Castillo-Negrete, Phys. Fluids, [**10**]{}, 576 (1998). D. del-Castillo-Negrete, B. A. Carreras, and V. E. Lynch, Phys. Rev. Lett., [**94**]{}, 065003 (2005). E. Fermi. Phys. Rev., [**75**]{}, 1169 (1949). A.J. Lichtenberg, M.A. Lieberman, R.H. Cohen, Physica D, [**1**]{} p.291 (1980). A. J. Lichtenberg e M. A. Lieberman, Regular and Chaotic Dynamics, Appl. Math. Sci., Spring Verlag, New York, 1992. J. P. Eckmann e D. Ruelle, Rev. Mod. Phys. [**57**]{}, 617, 1985. C. Manchein, J. Rosa and M. W. Beims, Physica D, [**238**]{}, 1688, 2009. J. D. Szezech Jr., S. R. Lopes and R. L. Viana, Phys. Lett. A, [**335**]{}, 394, 2005. C. P. Dettmann and O. Georgiou, Physica D, [**238**]{}, 2395, 2009. C. P. Dettmann and T. B. Howard, Physica D, [**238**]{}, 2404, 2009. C. P. Dettmann and O. Georgiou, J. Phys. A, [**44**]{}, 195102, 2011. C. P. Dettmann and E. D. Leonel, Physica D, [**241**]{}, 403, 2012. E. G. Altmann, A. E. Motter, H. Krantz, Phys. Rev. E, [**73**]{}, 026207, 2006. E. D. Leonel and C. P. Dettmann, Phys. Lett. A, [**376**]{}, 1669, 2012. D. R. Costa, C. P. Dettmann and E. D. Leonel, Phys. Rev. E, [**83**]{}, 066211, 2011. Y. Zou, M. Thiel, M. C. Romano and J. Kurths, Chaos, [**17**]{}, 043101, 2007. J. D. Szezech, I. L. Caldas, S. R. Lopes, R. L. Viana and J. P. Morrison, Chaos, [**19**]{}, 043108, 2009. E. G. Altmann, A. E. Motter, H. Krantz, Chaos, [**15**]{}, 033105 (2005). F. Lenz, C. Petri, F. K. Diakonos, P. Schmelcher, Phys. Rev. E, [**82**]{}, 016206, 2010. M. S. Custódio and M. W. Beims, Phys. Rev. E, [**83**]{}, 056201, 2011

ArXiv

--- abstract: | We study the problem of constructing a linear sketch of minimum dimension that allows approximation of a given real-valued function $f \colon \ftwo^n \rightarrow \mathbb R$ with small expected squared error. We develop a general theory of linear sketching for such functions through which we analyze their dimension for most commonly studied types of valuation functions: additive, budget-additive, coverage, $\alpha$-Lipschitz submodular and matroid rank functions. This gives a characterization of how many bits of information have to be stored about the input $x$ so that one can compute $f$ under additive updates to its coordinates. Our results are tight in most cases and we also give extensions to the distributional version of the problem where the input $x \in \ftwo^n$ is generated uniformly at random. Using known connections with dynamic streaming algorithms, both upper and lower bounds on dimension obtained in our work extend to the space complexity of algorithms evaluating $f(x)$ under long sequences of additive updates to the input $x$ presented as a stream. Similar results hold for simultaneous communication in a distributed setting. author: - 'Grigory Yaroslavtsev[^1]' - 'Samson Zhou[^2]' bibliography: - 'linsketch.bib' title: 'Approximate $\ftwo$-Sketching of Valuation Functions' --- Acknowledgments {#acknowledgments .unnumbered} =============== We would like to thank Swagato Sanyal for multiple discussions leading to this paper, including the proof of Theorem \[thm:approx-f2-sketch-uniform\] and Nikolai Karpov for his contributions to Section \[sec:rank-2\]. We would also like to thank Amit Chakrabarti, Qin Zhang and anonymous reviewers for their comments. [^1]: Indiana University, Bloomington, IN., USA & The Alan Turing Institute, London, UK. `gyarosla@iu.edu` [^2]: Indiana University, Bloomington, IN., USA. `samsonzhou@gmail.com`

ArXiv

--- abstract: 'We study a non-Hermitian Rice-Mele model without breaking time-reversal symmetry, with the non-Hermiticity arising from imbalanced hopping rates. The Berry connection, Berry curvature and Chern number are introduced in the context of biorthonormal inner product. It is shown that for a bulk system, although the Berry connection can be complex numbers, the Chern number is still quantized, as topological invariant. For an opened chain system, the mid-gap edge modes are obtained exactly, obeying the bulk-edge correspondence. Furthermore, we also introduce a local current in the context of biorthonormal inner product to measure the pumping charge generated by a cyclic adiabatic evolution. Analytical analysis and numerical simulation of the time evolution of the mid-gap states show that the pumping charge can be a dynamical topological invariant in correspondence with the Chern number. It indicates that the geometric concepts for Hermitian topological insulator can be extended to the non-Hermitian regime.' author: - 'R. Wang' - 'X. Z. Zhang' - 'Z. Song' title: 'Dynamical topological invariant for non-Hermitian Rice-Mele model' --- Introduction ============ Nowadays, non-Hermitian Hamiltonian is no longer a forbidden regime in quantum mechanics since the discovery that a certain class of non-Hermitian Hamiltonians could exhibit the entirely real spectra [@Bender; @Bender1]. In non-Hermitian quantum mechanics, the reality of the spectrum, unitary evolution and probability conservation still hold if the Dirac inner product is replaced by biorthonormal inner product. The origin of the reality of the spectrum of a non-Hermitian Hamiltonian is the pseudo-Hermiticity of the Hamiltonian operator [@Ali1]. Such kinds of Hamiltonians possess a particular symmetry, i.e., it commutes with the combined operator $\mathcal{% PT}$, but not necessarily with $\mathcal{P}$ and $\mathcal{T}$ separately. Here $\mathcal{P}$ is a unitary operator, such as parity, translation, rotation operators etc., while $\mathcal{T}$ is an anti-unitary operator, such as time-reversal operator. The combined symmetry is said to be unbroken if every eigenstate of the Hamiltonian is $\mathcal{PT}$-symmetric; then, the entire spectrum is real, while is said to be spontaneously broken if some eigenstates of the Hamiltonian are not the eigenstates of the combined symmetry. However, even within the unbroken symmetric region, many quantities, such as local particle probability, geometric phase of an adiabatic evolution of an eigenstate, etc., exhibit anomalous behavior in comparison with that in a Hermitian system. For instance, the biorthonormal expectation value of a local particle number operator and the geometric phase can be complex [@Bender2; @F.G; @D.P]. A natural question is that to what extent a non-Hermitian system could inherit the property of a Hermitian one by introducing biorthonormal inner product. In this paper, we investigate a non-Hermitian RM model, the Hermitian counterpart of which is a prototype system for topological matter. We focus on the region without breaking time reversal symmetry, which supports fully real spectrum. We investigate the topology of the degeneracy point by introducing the concepts of Berry connection, Berry curvature and Chern number in the context of biorthonormal inner product. In contrast to a Hermitian system, the Berry connection can be complex numbers. However, the Chern number is still quantized, as topological invariant to characterize the feature of energy band. For an opened chain system, the mid-gap edge modes are obtained exactly, that are identical to the ones in a Hermitian system. The bulk-edge correspondence still hold in the corresponding non-Hermitian SSH chain. Furthermore, we also introduce a local current in the context of biorthonormal inner product to measure the pumping charge generated by a cyclic adiabatic evolution, which is also defined by biorthonormal probability. Analytical analysis and numerical simulation of the time evolution of the mid-gap states show that the pumping charge can be a dynamical topological invariant in correspondence with the Chern number. Our results indicate that the geometric concepts for Hermitian topological insulator can be extended to the non-Hermitian regime. This paper is organized as follows. In Section \[Model Hamiltonian\], we present the model Hamiltonian and the solutions. In Section \[Geometric quantities\], we introduce the concepts of Berry connection, Berry curvature, and Chern number for a non-Hermitian system. In Section \[Chern number\], we calculate the Chern number in the concrete system. In Section [Edge-mode operators]{}, we study the edge modes for the open chain system. Section \[Pumping charge\] devotes to the dynamical signature of topological feature of the degeneracy point. Finally, we give a summary and discussion in Section \[Summary\]. Model Hamiltonian {#Model Hamiltonian} ================= We start our investigation by considering a non-Hermitian model with imbalanced hopping$$H=\sum_{l=1}^{2N}[\kappa _{l,l+1}a_{l}^{\dagger }a_{l+1}+\kappa _{l+1,l}a_{l+1}^{\dagger }a_{l}-V\left( -1\right) ^{l}a_{l}^{\dagger }a_{l}],$$where $\kappa _{l,l+1}=\frac{1+\left( -1\right) ^{l}\delta }{2}\lambda ^{\left( -1\right) ^{l+1}}$, $\kappa _{l+1,l}=\frac{1+\left( -1\right) ^{l}\delta }{2}\lambda ^{\left( -1\right) ^{l}}$. The spinless fermions satisfy the periodic boundary condition $a_{l}\equiv a_{l+2N}$. In this paper, we focus on the non-Hermitian system, $\lambda \neq 1$ and $>0$. It is a bipartite lattice, i.e., it has two sublattices $A$, $B$ such that each site on lattice $A$ has its nearest neighbors on sublattice $B$, and vice versa.** **The system is a variant of Hermitian SSH model by introducing the imbalanced hopping and staggered real potentials. The original Hermitian system at half-filling, proposed by Su, Schrieffer, and Heeger to model polyacetylene [@Su; @Schrieffer], is the prototype of a topologically nontrivial band insulator with a symmetry protected topological phase [@Ryu; @Wen]. In recent years, it has been attracted much attention and extensive studies have been demonstrated [Xiao,Hasan,Delplace,ChenS1,ChenS2,LS PRA]{}. The non-Hermiticy arises from asymmetry factor $\lambda $, which has been proposed to be realized in experiment** **[@S]. We note that the Hamiltonian preserves time-reversal $(\mathcal{T})$ symmetry. It has been shown that such type of system has entirely full real spectrum [@ZXZ1]. We introduce the Fourier transformations in two sub-lattices$$a_{l}=\frac{1}{\sqrt{N}}\sum_{k}e^{ikj}\left\{ \begin{array}{cc} \beta _{k}, & l=2j \\ \alpha _{k}, & l=2j-1% \end{array}% \right. ,$$where $j=1,2,...,N$, $k=2m\pi /N$, $m=0,1,2,...,N-1$. Spinless fermionic operators in $k$ space $\alpha _{k},$ $\beta _{k}$ are$$\left\{ \begin{array}{cc} \beta _{k}=\frac{1}{\sqrt{N}}\sum\limits_{j}e^{-ikj}a_{l}, & l=2j \\ \alpha _{k}=\frac{1}{\sqrt{N}}\sum\limits_{j}e^{-ikj}a_{l}, & l=2j-1% \end{array}% \right. .$$This transformation block diagonalizes the Hamiltonian due to its translational symmetry, i.e.,$$H=\sum_{k\in \lbrack 0,2\pi )}H_{k}=\sum_{k\in \lbrack 0,2\pi )}\psi _{k}^{\dagger }h_{k}\psi _{k},$$satisfying $\left[ H_{k},H_{k^{\prime }}\right] =0$. Here $H$ is rewritten in the Nambu representation with the basis$$\psi _{k}=\left( \begin{array}{c} \alpha _{k} \\ \beta _{k}% \end{array}% \right) ,$$and $h_{k}$ is a $2\times 2$ matrix$$h_{k}=\left( \begin{array}{cc} V & \lambda \gamma _{-k} \\ \lambda ^{-1}\gamma _{k} & -V% \end{array}% \right) , \label{h_k}$$where $\gamma _{k}=(\gamma _{-k})^{\dagger }$ $=\frac{1}{2}[\left( 1-\delta \right) +\left( 1+\delta \right) e^{ik}]$. The eigenvalues of $h_{k}$ are $% |\varphi _{\rho }^{k}\rangle $ $\left( \rho =\pm \right) $ with eigenvalues$$\varepsilon _{\rho }^{k}=\rho \sqrt{|\gamma _{k}|^{2}+V^{2}}. \label{spectrum}$$In the case of nonzero $\lambda $, we have all real eigenvalue. Let us denote the eigenvectors of a non-Hermitian Hamiltonian as$$\begin{aligned} h_{k}|\varphi _{\rho }^{k}\rangle &=&\varepsilon _{\rho }^{k}|\varphi _{\rho }^{k}\rangle , \\ h_{k}^{\dagger }|\eta _{\rho }^{k}\rangle &=&\varepsilon _{\rho }^{k}|\eta _{\rho }^{k}\rangle , \notag\end{aligned}$$where the explicit expression of $|\varphi _{\rho }^{k}\rangle $ and $|\eta _{\rho }^{k}\rangle $ is$$|\varphi _{\rho }^{k}\rangle =\frac{1}{\Omega _{\rho }}\left( \begin{array}{c} V+\varepsilon _{\rho }^{k} \\ \frac{\gamma _{k}}{\lambda }% \end{array}% \right) ,$$$$\left\vert \eta _{\rho }^{k}\right\rangle =\frac{1}{\Omega _{\rho }}\left( \begin{array}{c} V+\varepsilon _{\rho }^{k} \\ \lambda \gamma _{k}% \end{array}% \right) ,$$where the normalization factors $\Omega _{\rho }=\sqrt{\left( V+\varepsilon _{\rho }^{k}\right) ^{2}+|\gamma _{k}|^{2}}$. It is ready to check that biorthogonal bases $\left\{ |\varphi _{\rho }^{k}\rangle ,|\eta _{\rho }^{k}\rangle \right\} $ $\left( \rho =\pm \right) $ obey the biorthogonal and completeness conditions$$\langle \eta _{\rho ^{\prime }}^{k^{\prime }}|\varphi _{\rho }^{k}\rangle =\delta _{kk^{\prime }}\delta _{\rho \rho ^{\prime }},\sum_{\rho ,k}|\varphi _{\rho }^{k}\rangle \langle \varphi _{\rho }^{k}|=1. \label{bior}$$There are two Bloch bands from the eigenvalues of $h_{k}$, indexed by $\rho =\pm $. The band touching points occur at $k_{c}$ when$$\sqrt{|\gamma _{k_{c}}|^{2}+V^{2}}=0.$$The solutions of above equation indicates the degeneracy point$$\delta =0,V=0,$$at $k_{c}=\pi $. The energy band structure is illustrated Fig. \[fig1\]. We would like to stress that the origin is a degeneracy point rather than an exceptional point, although the system is a non-Hermitian system. The exceptional point appears only at the case with $\lambda =0$ and $\infty $, which is beyond our interest of this paper. Geometric quantities {#Geometric quantities} ==================== The topological property of the degeneracy point for a Hermitian RM model is well established. It can be regarded as a monopole in the parameter space [@Xiao]. We will investigate what happens when the Hamiltonian is non-Hermitian. Before we focus on the concrete model, we first present a general formalism for a non-Hermitian system. In this section, we will develop a parallel theory for non-Hermitian case within the unbroken symmetry region. Without loss of generality, we neglect the detail of the model and consider a non-Hermitian spin-$1/2$ system in an external magnetic field, which can be described by the following Hamiltonian$$h(k,q)=\mathbf{B}(k,q)\cdot \mathbf{\sigma },$$where $\mathbf{B}(k,q)=\left( B_{x},B_{y},B_{z},\right) $ is a complex magnetic field ($B_{x},B_{y}$ are complex, and $B_{z}$ is real) and $% \sigma =\left( \sigma _{x},\sigma _{y},\sigma _{z}\right) $ is Pauli matrices$$\sigma _{x}=\left( \begin{array}{cc} 0 & 1 \\ 1 & 0% \end{array}% \right) ,\sigma _{y}=\left( \begin{array}{cc} 0 & -i \\ i & 0% \end{array}% \right) ,\sigma _{z}=\left( \begin{array}{cc} 1 & 0 \\ 0 & -1% \end{array}% \right) .$$The Hamiltonian is periodic with real $q$ and $k$, i.e., $% h(k,q)=h(k+k_{0},q)$ $=h(k,q+q_{0})$. Taking$$\left\{ \begin{array}{c} B_{x}=B\cos \phi \sin \theta \\ B_{y}=B\sin \phi \sin \theta \\ B_{z}=B\cos \theta \\ \left\vert \mathbf{B}\right\vert =B=\sqrt{\left( B_{x}\right) ^{2}+\left( B_{y}\right) ^{2}+\left( B_{z}\right) ^{2}}% \end{array}% \right. ,$$with$$\tan \phi =B_{y}/B_{x},\cos \theta =B_{z}/B,$$we rewrite the Hamiltonian as the form$$h=B\left( \begin{array}{cc} \cos \theta & \sin \theta e^{-i\phi } \\ \sin \theta e^{i\phi } & -\cos \theta% \end{array}% \right) .$$We note that $\theta $ is real, while $\phi $ is complex. In parallel, for the Hermitian conjugation counterpart$$h^{\dagger }=B\left( \begin{array}{cc} \cos \theta & \sin \theta e^{-i\phi ^{\ast }} \\ \sin \theta e^{i\phi ^{\ast }} & -\cos \theta% \end{array}% \right) ,$$we have$$\begin{aligned} h|\varphi _{\rho }\rangle &=&\rho B|\varphi _{\rho }\rangle , \\ h^{\dagger }|\eta _{\rho }\rangle &=&\rho B|\eta _{\rho }\rangle , \notag\end{aligned}$$where $|\varphi _{\rho }\rangle $ and $|\eta _{\rho }\rangle $** **$% \left( \rho =\pm \right) $ is the instantaneous eigenvectors of $H$ and $% H^{\dagger }$, respectively. Here the explicit expressions of these eigenvectors are$$\left\vert \varphi _{+}\right\rangle =\left( \begin{array}{c} \cos \frac{\theta }{2} \\ \sin \frac{\theta }{2}e^{i\phi }% \end{array}% \right) ,\left\vert \varphi _{-}\right\rangle =\left( \begin{array}{c} -\sin \frac{\theta }{2} \\ \cos \frac{\theta }{2}e^{i\phi }% \end{array}% \right) ,$$and$$\left\vert \eta _{+}\right\rangle =\left( \begin{array}{c} \cos \frac{\theta }{2} \\ \sin \frac{\theta }{2}e^{i\phi ^{\ast }}% \end{array}% \right) ,\left\vert \eta _{-}\right\rangle =\left( \begin{array}{c} -\sin \frac{\theta }{2} \\ \cos \frac{\theta }{2}e^{i\phi ^{\ast }}% \end{array}% \right) ,$$which obey the biorthonormal relations$$\langle \eta _{\rho }|\varphi _{\rho ^{\prime }}\rangle =\langle \varphi _{\rho ^{\prime }}|\eta _{\rho }\rangle =\delta _{\rho \rho ^{\prime }}.$$We define the Berry connection $A_{\sigma }^{\rho }$ $(\sigma =k,q)$ and the Berry curvature $\Omega _{kq}^{\rho }$ in the context of biorthonormal inner product$$A_{\sigma }^{\rho }=-i\langle \eta _{\rho }|\nabla _{\sigma }|\varphi _{\rho }\rangle ,\Omega _{kq}^{\rho }=\partial _{k}A_{q}^{\rho }-\partial _{q}A_{k}^{\rho }, \label{A and Omega}$$Similar to the Hermitian case, $A_{\sigma }^{\rho }$ is gauge dependent while $\Omega _{k\rho }^{\rho }$ is not. Accordingly Chern number can be defined as$$c_{\rho }=\frac{1}{2\pi }\int_{k}^{k+k_{0}}\int_{q}^{q+q_{0}}\Omega _{kq}^{\rho }dkdq. \label{chern1}$$For well defined $A_{k}^{\rho }$ and $A_{q}^{\rho }$, the Chern number can be rewritten as the form$$c_{\rho }=\frac{1}{2\pi }\oint_{\partial D}\mathcal{\mathbf{A}}^{\rho }\cdot d\mathbf{r,} \label{chern2}$$where** **$\mathcal{\mathbf{A}}^{\rho }=A_{k}^{\rho }\hat{k}% +A_{q}^{\rho }\hat{q}$**,** $r=k\hat{k}+q\hat{q}$**.** Here $\hat{q% }$ and $\hat{k}$ are unit vectors, and $\partial D$ (boundary of the domain $D$, which covers the square of $k_{0}\times q_{0}$) is the path of the integral.** **We concern whether $c_{\rho }$ is still quantized for real $\left\vert \mathbf{B}\right\vert $ in the non-Hermitian system as a topological invariant. We will investigate these issues based on a concrete model. Chern number {#Chern number} ============ In this section, we will calculate the explicit expressions of the geometric quantities for the non-Hermitian RM model. The Bloch Hamiltonian is periodic through the periodic functions $V(q)=V(q+2\pi )$ and $\delta (q)=\delta (q+2\pi )$. When $q$ sweeps over a period, the system experiences a loop in the $V\delta $ plane. For this concrete model, we have the explicit form of magnetic field$$\left\{ \begin{array}{c} B_{x}=\frac{1}{2}\left( \lambda \gamma _{-k}+\lambda ^{-1}\gamma _{k}\right) , \\ B_{y}=\frac{i}{2}\left( \lambda \gamma _{-k}-\lambda ^{-1}\gamma _{k}\right) , \\ B_{z}=V,% \end{array}% \right.$$and$$B=\sqrt{|\gamma _{k}|^{2}+V^{2}}.$$Obviously, field $\mathbf{B}$ is periodic vector, $\mathbf{B}(q,k)=\mathbf{B}% (q+2\pi ,k)$ $=\mathbf{B}(q,k+2\pi )$. In the following we only focus on the lower band, neglecting the band index for the geometric quantity. Direct derivations show that$$\mathcal{\mathbf{A}}=-i\langle \eta ^{k}|\partial _{q}|\varphi ^{k}\rangle \hat{q}-i\langle \eta ^{k}|\partial _{k}|\varphi ^{k}\rangle \hat{k}$$with$$\begin{aligned} \langle \eta |\partial _{\sigma }|\varphi \rangle &=&\frac{1}{2}[\frac{1}{2}% \sin \theta \left( \partial _{\sigma }\theta \right) \left( 1-e^{2i\text{% \textrm{Re}}\phi }\right) \notag \\ &&+i\left( 1+\cos \theta \right) \left( \partial _{\sigma }\phi \right) e^{2i% \text{\textrm{Re}}\phi }],\end{aligned}$$where $\hat{q}$ and $\hat{k}$ are unit vectors for parameters $\sigma =q$, $% k$ respectively, and the angles are$$\begin{aligned} \phi &=&\arctan [\frac{i\left( \lambda \gamma _{-k}-\lambda ^{-1}\gamma _{k}\right) }{\left( \lambda \gamma _{-k}+\lambda ^{-1}\gamma _{k}\right) }], \label{phi} \\ \theta &=&\arccos (\frac{V}{\sqrt{|\gamma _{k}|^{2}+V^{2}}}). \label{theta}\end{aligned}$$We note that $\theta $ is real, while $\phi $ is complex. In contrast to a Hermitian case, Berry connection $\mathcal{\mathbf{A}}$ is complex as expected, which accords with the violation of the conservation of Dirac probability. Apparently, the Berry curvature can be directly obtained by the definition in Eq. (\[A and Omega\]), and then the Chern number. However, it is impossible to get all the Berry curvatures for a given loop through a single expression of $\mathcal{\mathbf{A}}$, since the wave functions $% \left\vert \varphi \right\rangle $ and $\left\vert \eta \right\rangle $ are not smooth and single valued everywhere. For instance, when taking $\theta =0$, (or $B_{y}=B_{x}=0,$ and $B_{z}>0$) we have$$\left\vert \varphi \right\rangle =\left( \begin{array}{c} 0 \\ e^{i\phi }% \end{array}% \right) ,\left\vert \eta \right\rangle =\left( \begin{array}{c} 0 \\ e^{i\phi ^{\ast }}% \end{array}% \right) .$$where $\varphi $ is not well-defined, or indefinite. So $\left\vert \varphi \right\rangle $ and $\left\vert \eta \right\rangle $ have singularity at $% \mathbf{B}=\left( 0,0,B_{z}\right) $ with $B_{z}>0$. We can choose another gauge by multiplying $\left\vert \varphi \right\rangle $ and $\left\vert \eta \right\rangle $ by $e^{-i\mathrm{Re}\phi }$. Then the wave function is smooth and single valued everywhere expect at the south pole, i.e., $\varphi $ is not well-defined, or indefinite at $\theta =\pi $. For a given loop, one should choose one or two specific gauges, or expressions of $\left\vert \varphi \right\rangle $ and $\left\vert \eta \right\rangle $ in order to calculate the Chern number. If a set of proper chosen wave functions $% \left\{ \left\vert \varphi \right\rangle ,\left\vert \eta \right\rangle \right\} $ are well-defined for a given loop, the Chern number is zero. If a loop requires two different gauges, the Chern number is nonzero. Obviously, the difference between the Berry connections $\mathcal{\mathbf{A}} $ in two gauges is $\nabla _{\mathbf{R}}\mathrm{Re}\phi $, and thus Chern number are definitely real. We choose four typical loops, involving: (a) quadrants I and II $(V>0)$; (b) quadrants III and IV $(V<0)$; (c) one of four quadrants, or quadrants I and IV $(\delta >0)$, or quadrants II and III $(\delta <0)$, and (d) quadrants I, II, III, and IV (enclosing the origin). For the cases (a-c), one can take the gauge as$$\begin{aligned} \text{(a)} &:&\left\vert \varphi \right\rangle ,\left\vert \eta \right\rangle ; \notag \\ \text{(b)} &:&e^{i\mathrm{Re}\phi }\left\vert \varphi \right\rangle ,e^{i% \mathrm{Re}\phi }\left\vert \eta \right\rangle ; \notag \\ \text{(c)} &:&\left\vert \varphi \right\rangle ,\left\vert \eta \right\rangle ; \notag \\ &&\text{or }e^{i\mathrm{Re}\phi }\left\vert \varphi \right\rangle ,e^{i% \mathrm{Re}\phi }\left\vert \eta \right\rangle ,\end{aligned}$$while one need two gauges for (d) as$$\begin{aligned} \text{(dI)} &:&\left\vert \varphi \right\rangle ,\left\vert \eta \right\rangle ,\text{ }\left( V>0\right) ; \notag \\ \text{(dII)} &:&e^{i\mathrm{Re}\phi }\left\vert \varphi \right\rangle ,e^{i% \mathrm{Re}\phi }\left\vert \eta \right\rangle ,\text{ }\left( V<0\right) .\end{aligned}$$We conclude from this analysis that the Chern number is nonzero when the loop encloses the origin. Explicitly, with the Eq. (\[chern2\]), we have$$\begin{aligned} c &=&\frac{1}{2\pi }(\oint_{\partial D_{\text{I}}}\mathcal{\mathbf{A}}_{% \text{I}}+\oint_{\partial D_{\text{II}}}\mathcal{\mathbf{A}}_{\text{II}% })\cdot d\mathbf{r} \notag \\ &=&\frac{1}{2\pi }\oint_{\partial D_{\text{I}}}\left( \mathcal{\mathbf{A}}_{% \text{I}}-\mathcal{\mathbf{A}}_{\text{II}}\right) \cdot d\mathbf{r} \notag \\ &\neq &0, \label{chern3}\end{aligned}$$where the subindex of $\mathcal{A}_{\text{I}}$ and $\mathcal{A}_{\text{II}}$** **distinguish $\left( \left\vert \varphi \right\rangle ,\left\vert \eta \right\rangle \right) $ from $\left( e^{i\mathrm{Re}\phi }\left\vert \varphi \right\rangle ,e^{i\mathrm{Re}\phi }\left\vert \eta \right\rangle \right) $ respectively, when we define the Berry connection. To demonstrate the result, we consider a loop in the form$$\left\{ \begin{array}{c} \delta =\delta _{c}+r\cos q \\ V=V_{c}+r\sin q% \end{array}% \right. \label{cir loop}$$which is a circle in $V\delta $ plane with radius $r$, centered at $\left( \delta _{c},V_{c}\right) $. The Chern number can be obtained as $$c=\left\{ \begin{array}{cc} 0, & \delta _{c}^{2}+V_{c}^{2}>r^{2} \\ 1, & \delta _{c}^{2}+V_{c}^{2}<r^{2}% \end{array}% \right. , \label{c cir}$$ i.e., Chern number is zero if the loop does not encircle the degeneracy point, while is nonzero if encircle the degeneracy point. In addition, if we take $q\rightarrow -q$, we will get $c\rightarrow -c$, which means that one gets opposite Chern number for the same loop but with opposite direction. We schematically illustrate this loop in Fig. \[fig3\](a). It shows that the degeneracy point still has topological feature in a non-Hermitian system even though the Berry connection may be complex. The underlying mechanism of the quantization of the Chern number for a non-Hermitian system is that the single-valuedness of the wave function is always true no matter the system is Hermitian or not. Edge-mode operators {#Edge-mode operators} =================== In a Hermitian SSH model, the degenerate zero modes take the role of topological invariant for opened chain. In this section, we investigate the similar feature for non-Hermitian RM model. Considering the RM model with an open boundary condition, the Hamiltonians read$$\begin{aligned} &&H_{\mathrm{CH}}=H-M, \notag \\ &&M=\kappa _{2N,1}a_{2N}^{\dagger }a_{1}+\kappa _{1,2N}a_{1}^{\dagger }a_{2N},\end{aligned}$$which represents the original system with broken the coupling across two sites $(2N,1)$. We introduce a pair of edge-mode operators $\overline{A}_{% \mathrm{L,R}}$ in the infinite $N$ limit, which are defined as$$\begin{aligned} \overline{A}_{\text{L}} &=&\frac{1}{\sqrt{\Omega }}\sum\limits_{j=1}^{N}% \left( \frac{\delta -1}{\delta +1}\right) ^{N-j}a_{2j}^{\dagger }, \\ \overline{A}_{\text{R}} &=&\frac{1}{\sqrt{\Omega }}\sum\limits_{j=1}^{N}% \left( \frac{\delta -1}{\delta +1}\right) ^{j-1}a_{2j-1}^{\dagger },\end{aligned}$$where $\Omega =\{1-\left[ \left( \delta -1\right) /\left( \delta +1\right) % \right] ^{2N}\}/\{1-\left[ \left( \delta -1\right) /\left( \delta +1\right) % \right] ^{2}\}$. It is easy to check that$$\begin{aligned} \lbrack \overline{A}_{\text{L}},H_{\mathrm{CH}}] &=&V\overline{A}_{\text{L}}, \\ \lbrack \overline{A}_{\text{R}},H_{\mathrm{CH}}] &=&-V\overline{A}_{\text{R}% },\end{aligned}$$which ensues that $$\begin{aligned} H_{\mathrm{CH}}\overline{A}_{\text{L}}\left\vert \text{Vac}\right\rangle &=&-V\overline{A}_{\text{L}}\left\vert \text{Vac}\right\rangle , \\ H_{\mathrm{CH}}\overline{A}_{\text{R}}\left\vert \text{Vac}\right\rangle &=&V% \overline{A}_{\text{R}}\left\vert \text{Vac}\right\rangle , \\ H_{\mathrm{CH}}\overline{A}_{\text{L}}\overline{A}_{\text{R}}\left\vert \text{Vac}\right\rangle &=&0\times \overline{A}_{\text{L}}\overline{A}_{% \text{R}}\left\vert \text{Vac}\right\rangle .\end{aligned}$$Here $\left\vert \text{Vac}\right\rangle $ is the vacuum state of fermion operator, i.e., $a_{l}\left\vert \text{Vac}\right\rangle =0$. Obviously, states $\overline{A}_{\text{L,R}}\left\vert \text{Vac}\right\rangle $ and $% \overline{A}_{\text{L}}\overline{A}_{\text{R}}\left\vert \text{Vac}% \right\rangle $ are the eigenstates of $H_{\mathrm{CH}}$ with eigen energies $\mp V$ and $0$, respectively. In parallel, we can also define the biorthogonal conjugation operators$$\begin{aligned} A_{\text{L}} &=&\frac{1}{\sqrt{\Omega }}\sum\limits_{j=1}^{N}\left( \frac{% \delta -1}{\delta +1}\right) ^{N-j}a_{2j}, \\ A_{\text{R}} &=&\frac{1}{\sqrt{\Omega }}\sum\limits_{j=1}^{N}\left( \frac{% \delta -1}{\delta +1}\right) ^{j-1}a_{2j-1},\end{aligned}$$which satisfy the canonical commutation relations $$\{A_{\mu },\overline{A}_{\nu }\}=\delta _{\mu \nu },\{A_{\mu },A_{\nu }\}=\{% \overline{A}_{\mu },\overline{A}_{\nu }\}=0,$$with the indices $\mu $, $\nu =$ L, R. Similarly, we have$$\lbrack A_{\text{L}}^{\dag },H_{\mathrm{CH}}^{\dag }]=VA_{\text{L}}^{\dag },[A_{\text{R}}^{\dag },H_{\mathrm{CH}}^{\dag }]=-VA_{\text{R}}^{\dag },$$which indicate that states $A_{\text{L,R}}^{\dag }\left\vert \text{Vac}% \right\rangle $ and $A_{\text{L}}^{\dag }A_{\text{R}}^{\dag }\left\vert \text{Vac}\right\rangle $ are the eigenstates of $H_{\mathrm{CH}}^{\dag }$with eigen energies $\mp V$ and $0$, respectively. A surprising fact is that $A_{\mu }^{\dag }=\overline{A}_{\mu }$, which does not hold true in general since $H_{\mathrm{CH}}\neq H_{\mathrm{CH}}^{\dag }$. It is due to the special eigenstates of a particular model. This feature allows us to treat the edge modes in the framework of Hermitian regime. The biorthonormal and Dirac probabilities of the edge modes are the same and can be expressed as$$\mathcal{P}_{\mu }(l)=\left\langle \text{Vac}\right\vert A_{\mu }a_{l}^{\dagger }a_{l}A_{\mu }^{\dag }\left\vert \text{Vac}\right\rangle .$$or explicit form $$\mathcal{P}_{\mathrm{L}}(l)=\left\{ \begin{array}{cc} \frac{1}{\Omega }\left( \frac{\delta -1}{\delta +1}\right) ^{2N-l}, & l=2j \\ 0, & l=2j-1% \end{array}% \right. ,$$ and $$\mathcal{P}_{\mathrm{R}}(l)=\left\{ \begin{array}{cc} 0, & l=2j \\ \frac{1}{\Omega }\left( \frac{\delta -1}{\delta +1}\right) ^{l-1}, & l=2j-1% \end{array}% \right. ,$$ which obey the normalization condition $\sum_{l}\mathcal{P}_{\mu }(l)=1$. We note that the edge modes are independent of $\lambda $ and $V$, identical to that for the standard SSH chain ($V=0$ and $\lambda =1$). In contrast to the standard SSH chain, the eigenvalues of edged modes can be nonzero. The profiles of the edge modes are illustrated schematically in Fig. \[fig2\]. Based on the above analysis, it turns out that the bulk system exhibits the similar topological feature when taking $V=0$. In Hermitian systems, the existence of edge modes is intimately related to the bulk topological quantum numbers, which is referred as the bulk-edge correspondence relations [@Thouless; @Kane; @Zhang; @Lu]. We are interested in the generalization of the bulk-edge correspondence to this non-Hermitian system. Previous works show that when sufficiently weak non-Hermiticity is introduced to topological insulator models, the edge modes can retain some of their original characteristics [@Esaki; @Hu]. Similarly, we can get the same conclusion for the case with $V=0$, if we define the Zak phase in the framework of biorthonormal inner product, demonstrating the bulk-edge correspondence. ![Numerical simulations of particle transport across two ends of the chain current obtained by exact diagonalization, driven by the time-dependent parameter $V=1-2\protect\omega t$, $t\in \left[ 0,T=\protect% \omega ^{-1}\right]$. Plots of the current (a1-d1) and charge accumulation as function of time (a2-d2) for several typical speed $\protect\omega $, indicated in the figures. The total charge transfer during the interval $[0,% \protect\omega ^{-1}]$ is also indicated in each figure. We note that the result accords with our prediction well when the speed is slow enough in (a1, b1). As $\protect\omega $ increases, the deviation becomes larger and larger. In the case of (a4, b4), the accumulated charge becomes positive, which is qualitatively different from the adiabatic process. The parameters of the system are $N=10$, $\protect\delta =0.5$, $\protect\kappa =0.05$, and $\protect\lambda =1.5$.[]{data-label="fig4"}](fig4_1.eps){width="4.6cm"} ![Numerical simulations of particle transport across two ends of the chain current obtained by exact diagonalization, driven by the time-dependent parameter $V=1-2\protect\omega t$, $t\in \left[ 0,T=\protect% \omega ^{-1}\right]$. Plots of the current (a1-d1) and charge accumulation as function of time (a2-d2) for several typical speed $\protect\omega $, indicated in the figures. The total charge transfer during the interval $[0,% \protect\omega ^{-1}]$ is also indicated in each figure. We note that the result accords with our prediction well when the speed is slow enough in (a1, b1). As $\protect\omega $ increases, the deviation becomes larger and larger. In the case of (a4, b4), the accumulated charge becomes positive, which is qualitatively different from the adiabatic process. The parameters of the system are $N=10$, $\protect\delta =0.5$, $\protect\kappa =0.05$, and $\protect\lambda =1.5$.[]{data-label="fig4"}](fig4_2.eps){width="4.6cm"} ![Numerical simulations of particle transport across two ends of the chain current obtained by exact diagonalization, driven by the time-dependent parameter $V=1-2\protect\omega t$, $t\in \left[ 0,T=\protect% \omega ^{-1}\right]$. Plots of the current (a1-d1) and charge accumulation as function of time (a2-d2) for several typical speed $\protect\omega $, indicated in the figures. The total charge transfer during the interval $[0,% \protect\omega ^{-1}]$ is also indicated in each figure. We note that the result accords with our prediction well when the speed is slow enough in (a1, b1). As $\protect\omega $ increases, the deviation becomes larger and larger. In the case of (a4, b4), the accumulated charge becomes positive, which is qualitatively different from the adiabatic process. The parameters of the system are $N=10$, $\protect\delta =0.5$, $\protect\kappa =0.05$, and $\protect\lambda =1.5$.[]{data-label="fig4"}](fig4_3.eps){width="4.6cm"} ![Numerical simulations of particle transport across two ends of the chain current obtained by exact diagonalization, driven by the time-dependent parameter $V=1-2\protect\omega t$, $t\in \left[ 0,T=\protect% \omega ^{-1}\right]$. Plots of the current (a1-d1) and charge accumulation as function of time (a2-d2) for several typical speed $\protect\omega $, indicated in the figures. The total charge transfer during the interval $[0,% \protect\omega ^{-1}]$ is also indicated in each figure. We note that the result accords with our prediction well when the speed is slow enough in (a1, b1). As $\protect\omega $ increases, the deviation becomes larger and larger. In the case of (a4, b4), the accumulated charge becomes positive, which is qualitatively different from the adiabatic process. The parameters of the system are $N=10$, $\protect\delta =0.5$, $\protect\kappa =0.05$, and $\protect\lambda =1.5$.[]{data-label="fig4"}](fig4_4.eps){width="4.6cm"} ![Numerical simulations of particle transport across two ends of the chain current obtained by exact diagonalization, driven by the time-dependent parameter $V=1-2\protect\omega t$, $t\in \left[ 0,T=\protect% \omega ^{-1}\right]$. Plots of the current (a1-d1) and charge accumulation as function of time (a2-d2) for several typical speed $\protect\omega $, indicated in the figures. The total charge transfer during the interval $[0,% \protect\omega ^{-1}]$ is also indicated in each figure. We note that the result accords with our prediction well when the speed is slow enough in (a1, b1). As $\protect\omega $ increases, the deviation becomes larger and larger. In the case of (a4, b4), the accumulated charge becomes positive, which is qualitatively different from the adiabatic process. The parameters of the system are $N=10$, $\protect\delta =0.5$, $\protect\kappa =0.05$, and $\protect\lambda =1.5$.[]{data-label="fig4"}](fig4_5.eps){width="4.6cm"} ![Numerical simulations of particle transport across two ends of the chain current obtained by exact diagonalization, driven by the time-dependent parameter $V=1-2\protect\omega t$, $t\in \left[ 0,T=\protect% \omega ^{-1}\right]$. Plots of the current (a1-d1) and charge accumulation as function of time (a2-d2) for several typical speed $\protect\omega $, indicated in the figures. The total charge transfer during the interval $[0,% \protect\omega ^{-1}]$ is also indicated in each figure. We note that the result accords with our prediction well when the speed is slow enough in (a1, b1). As $\protect\omega $ increases, the deviation becomes larger and larger. In the case of (a4, b4), the accumulated charge becomes positive, which is qualitatively different from the adiabatic process. The parameters of the system are $N=10$, $\protect\delta =0.5$, $\protect\kappa =0.05$, and $\protect\lambda =1.5$.[]{data-label="fig4"}](fig4_6.eps){width="4.6cm"} ![Numerical simulations of particle transport across two ends of the chain current obtained by exact diagonalization, driven by the time-dependent parameter $V=1-2\protect\omega t$, $t\in \left[ 0,T=\protect% \omega ^{-1}\right]$. Plots of the current (a1-d1) and charge accumulation as function of time (a2-d2) for several typical speed $\protect\omega $, indicated in the figures. The total charge transfer during the interval $[0,% \protect\omega ^{-1}]$ is also indicated in each figure. We note that the result accords with our prediction well when the speed is slow enough in (a1, b1). As $\protect\omega $ increases, the deviation becomes larger and larger. In the case of (a4, b4), the accumulated charge becomes positive, which is qualitatively different from the adiabatic process. The parameters of the system are $N=10$, $\protect\delta =0.5$, $\protect\kappa =0.05$, and $\protect\lambda =1.5$.[]{data-label="fig4"}](fig4_7.eps){width="4.6cm"} ![Numerical simulations of particle transport across two ends of the chain current obtained by exact diagonalization, driven by the time-dependent parameter $V=1-2\protect\omega t$, $t\in \left[ 0,T=\protect% \omega ^{-1}\right]$. Plots of the current (a1-d1) and charge accumulation as function of time (a2-d2) for several typical speed $\protect\omega $, indicated in the figures. The total charge transfer during the interval $[0,% \protect\omega ^{-1}]$ is also indicated in each figure. We note that the result accords with our prediction well when the speed is slow enough in (a1, b1). As $\protect\omega $ increases, the deviation becomes larger and larger. In the case of (a4, b4), the accumulated charge becomes positive, which is qualitatively different from the adiabatic process. The parameters of the system are $N=10$, $\protect\delta =0.5$, $\protect\kappa =0.05$, and $\protect\lambda =1.5$.[]{data-label="fig4"}](fig4_8.eps){width="4.6cm"} Pumping charge {#Pumping charge} ============== In a Hermitian system, the physical meaning of Chern number is well known. In a Hermitian RM model, it has been shown that the adiabatic particle transport over a time period takes the form of the Chern number and it is quantized [@Xiao]. The pumped charge counts the net number of degenerate point enclosed by the loop. For example, consider a time-dependent Hamiltonian, which varies adiabatically along the circle in Eq. (\[cir loop\]). After a period of time, the particle transport for the half-filled ground state equals to the Chern number of the loop in Eq. (\[c cir\]). In the above, we have shown that a non-Hermitian system can exhibit the similar topological feature as that in a Hermitian system. The Chern number as the topological invariant defined in the context of biorthonormal inner product. In parallel, such a Chern number should have the similar physical meaning. Actually, one can rewrite Eq. (\[chern2\]) in the form$$\begin{aligned} c &=&\oint (\frac{\partial \mathcal{Z}}{\partial \delta }d\delta +\frac{% \partial \mathcal{Z}}{\partial V}dV) \notag \\ &=&\int_{0}^{T}(\frac{\partial \mathcal{Z}}{\partial \delta }\dot{\delta}+% \frac{\partial \mathcal{Z}}{\partial V}\dot{V})dt,\end{aligned}$$where $\mathcal{Z}$ is Zak phase$$\mathcal{Z}=\frac{i}{2\pi }\int_{0}^{2\pi }\langle \eta ^{k}|\partial _{k}|\varphi ^{k}\rangle dk,$$defined in the context of biorthonormal inner product. Here $V$ and $\delta $ is periodic function of time $t$ and the sub-index is neglected for simplicity. Furthermore, we can find out the physical meaning of the Chern number by the relation $$c=\int_{0}^{T}\mathcal{J}(t)dt,$$where$$\mathcal{J}=\frac{i}{2\pi }\int_{0}^{2\pi }[(\partial _{t}\langle \eta ^{k}|)\partial _{k}|\varphi ^{k}\rangle -(\partial _{k}\langle \eta ^{k}|)\partial _{t}|\varphi ^{k}\rangle ]dk$$is the adiabatic current of biorthonormal version. Then $c$ is pumped charge of all channel $k$ driven by the time-dependent Hamiltonian varying in a period. This conclusion is obtained for a model Hamiltonian with translational symmetry. Next, we consider this issue based on the edge states for the chain system.** **Based on the exact expression of the edge states, we have known that only $\delta $ can change the distribution of the probability. Specifically, for a given $V>0$, if we vary $\delta $ from $% \delta _{0}>0$ to $-\delta _{0}$ adiabatically, the state $\overline{A}_{% \text{L}}\left\vert \text{Vac}\right\rangle $ will evolves to $\overline{A}% _{\text{R}}\left\vert \text{Vac}\right\rangle $. Then particle will transport adiabatically from the rightmost of the chain to the leftmost. Remarkably, one can get the same result from an alternative way. We consider a system with additional weak tunneling $\kappa $ between two ends. The corresponding Hamiltonian has the form$$H_{\kappa }=H_{\mathrm{CH}}+\kappa M, \label{H_cappa}$$which is schematically illustrated in Fig. \[fig2\] and the structure of spectrum of $H_{\kappa }$ is plotted in Fig. \[fig2\](a). We note that the weak tunneling $\kappa $ hybridize two edge $\overline{A}_{\text{L}% }\left\vert \text{Vac}\right\rangle $ and to $\overline{A}_{\text{R}% }\left\vert \text{Vac}\right\rangle $ states when $\delta >0$ and $V$around $0$. For a given $\delta >0$, if we vary $V$ from $V_{0}>0$ to $% -V_{0}$ adiabatically, the state $\overline{A}_{\text{L}}\left\vert \text{% Vac}\right\rangle $ will evolves to $\overline{A}_{\text{R}}\left\vert \text{Vac}\right\rangle $. Then a particle will transport adiabatically form the rightmost of the chain to the leftmost. Fig. \[fig2\](b) and (d), illustrate this point. During this process, particle should pass through the connection of two ends rather than the bulk region of the chain. This transport can be detect by watching the current across two ends.** **The accumulated charge of the current should be integer.** ** Inspired by these results we expect that the hidden topology behind the non-Hermitian model can be unveiled by the pumping charge with the respect to the mid-gap edge mode rather than all the energy levels. As illustrated in Fig. \[fig3\](a), the pumping charge is the sum of all the pumping charges obtained by an infinite rectangular loops. The pumping charge of a rectangular loop is simply determined by the positions of two vertical sides. Therefore, we can conclude that the pumping charge of an edge state for an arbitrary loop equals to the corresponding Chern number of the loop. To characterize the charge transfer quantitatively, we developed the concept of the biorthonormal current in a non-Hermitian tight-binding model. In the following, only single-particle case is concerned. We begin with the rate of change of the biorthonormal probability $\rho _{l}$ at an arbitrary site $l$ for given eigenstate $\left( |\varphi \left( t\right) \rangle ,\left\vert \eta \left( t\right) \right\rangle \right) $ at instant $t$, which can be expressed as$$\begin{aligned} \frac{d\rho _{l}}{dt} &=&\frac{d\langle \eta \left( t\right) |l\rangle \langle l|\varphi \left( t\right) \rangle }{dt} \notag \\ &=&\frac{1}{i}\langle \eta \left( t\right) |\left[ \left\vert l\right\rangle \left\langle l\right\vert ,H\right] |\varphi \left( t\right) \rangle\end{aligned}$$where $\left\vert l\right\rangle =a_{l}^{\dag }\left\vert \text{Vac}% \right\rangle $. For the concerned Hamiltonian we have$$\frac{d\rho _{l}}{dt}=j_{l}-j_{l-1},$$where$$j_{l}=\frac{1}{i}\langle \eta \left( t\right) |(\kappa _{l,l+1}\left\vert l\right\rangle \left\langle l+1\right\vert -\kappa _{l+1,l}|l+1\rangle \langle l|)|\varphi \left( t\right) \rangle .$$The quantity $j_{l}$ refers to the biorthonormal current across sites $l$and $l+1$. The accumulated charge passing the site $l$ during the period $T$ is$$Q_{l}=\int_{0}^{T}\left( j_{l}-j_{l-1}\right) dt.$$We consider the case by taking $V=V_{0}\left( 1-2\omega t\right) $ with $% \omega \ll 1$, and $$|\varphi \left( 0\right) \rangle =\overline{A}_{\text{L}}\left\vert \text{Vac% }\right\rangle \text{, }\left\vert \eta \left( 0\right) \right\rangle =A_{% \text{L}}^{\dag }\left\vert \text{Vac}\right\rangle .$$According to the analysis above, if $t$ varies from $0$ to $T=\omega ^{-1}$, $Q_{l}$ should be $1$ or $-1$, which is consistent with the direction of the loop. To examine how the scheme works in practice, we simulate the quasi-adiabatic process by computing the time evolution numerically for finite system. In principle, for a given initial eigenstate $\left\vert \psi \left( 0\right) \right\rangle $, the time evolved state under a Hamiltonian $% H_{\kappa }\left( t\right) $ is$$\left\vert \Phi \left( t\right) \right\rangle =\mathcal{T}\{\exp (-i\int_{0}^{t}H_{\kappa }\left( t\right) \mathrm{d}t)\left\vert \psi \left( 0\right) \right\rangle \},$$where $\mathcal{T}$ is the time-ordered operator. In low speed limit $\omega \rightarrow 0$, we have$$f\left( t\right) =\left\vert \langle \overline{\psi }\left( t\right) \left\vert \Phi \left( t\right) \right\rangle \right\vert \rightarrow 1,$$where $\left\langle \overline{\psi }\left( t\right) \right\vert $ is the corresponding instantaneous eigenstate of $H_{\kappa }^{\dagger }\left( t\right) $. The computation is performed by using a uniform mesh in the time discretization for the time-dependent Hamiltonian $H_{\mathrm{CH}}\left( t\right) $. In order to demonstrate a quasi-adiabatic process, we keep $% f\left( t\right) >0.9985$ during the whole process by taking sufficient small $\omega $ (see the case (a1, b1) in Fig. \[fig4\]). Fig. \[fig4\] plots the simulations of particle current and the corresponding total probability for several typical cases, in order to see to what extent the process can be regarded as a quasi-adiabatic one. It shows that the obtained dynamical quantities are in close agreement with the expected Chern number. Summary {#Summary} ======= In summary, we have analyzed a one-dimensional non-Hermitian RM model that exhibits the similar topological features of a Hermitian one within the time-reversal symmetry-unbroken regions, in which the Berry connection, Berry curvature, Chern number, current, and pumped charge, etc., are defined in the context of biorthonormal inner product. We also examined the dynamical signature for topological invariant, which is pumped charge driven by the parameters. The underlying mechanism of our finding is that if a non-Hermitian system is pseudo-Hermitian and the Hermitian counterpart of its is topological within the unbroken symmetric region, such a non-Hermitian system inherits the same topological characterization of the counterpart. We acknowledge the support of the CNSF (Grant No. 11374163). [99]{} C. M. Bender and S. Boettcher, Phys. Rev. Lett. **80**, 5243 (1998). C. M. Bender, D. C. Brody and H. F. Jones, Phys. Rev. Lett. **89**, 270401 (2002). A. Mostafazadeh, J. Math. Phys. **43**, 205 (2002). C. M. Bender, Making sense of non-Hermitian Hamiltonians. Rep. Prog. Phys. **70**, 947–1018 (2007). F. G. Scholtz, H. B. Geyer and F. J. W. Hahne, Quasi-Hermitian operators in quantum mechanics and the variational principle, Ann. Phys. (NY) **213**, 74–101 (1992). D. P. Musumbu, H. B. Geyer and W. D. Heiss, J. Phys. **40**, 75 (2007). W. P. Su, J. R. Schrieffer, and A. J. Heeger, Phys. Rev. Lett. **42**, 1698 (1979). J. R. Schrieffer, *The Lesson of Quantum Theory* (North Holland, Amsterdam, 1986). Shinsei Ryu and Yasuhiro Hatsugai, Phys. Rev. Lett. **89**, 077002 (2002). Xiao-Gang Wen, Phys. Rev. B **85**, 085103 (2012). D. Xiao, M. C. Chang, and Q. Niu, Rev. Mod. Phys. **82**, 1959 (2010). M. Z. Hasan and C. L. Kane, Rev. Mod. Phys. **82**, 3045 (2010); X. -L. Qi and S. -C. Zhang, ibid. **83**, 1057 (2011). P. Delplace, D. Ullmo, and G. Montambaux Phys. Rev. B **84**, 195452 (2011). Linhu Li, Zhihao Xu, and Shu Chen, Phys. Rev. B **89**, 085111 (2014). Linhu Li and Shu Chen, Phys. Rev. B **92**, 085118 (2015). S. Lin, and Z. Song, Phys. Rev. A **96**, 052121 (2017). S. Longhi, Phys. Rev. A **95**, 062122 (2017). X. Z. Zhang and Z. Song, Ann. Physics **339,** 109–121 (2013). D. J. Thouless, *Topological Quantum Numbers in Nonrelativistic Physics* (World Scientific, Singapore, 1998). M. Z. Hasan and C. L. Kane, Rev. Mod. Phys. **82,** 3045 (2010). X.-L. Qi and S.-C. Zhang, Rev. Mod. Phys. **83,** 1057 (2011). L. Lu, J. D. Joannopoulos, and M. Soljačić, Nat. Photonics **8,** 821 (2014). K. Esaki, M. Sato, K. Hasebe, and M. Kohmoto, Phys. Rev. B **84,** 205128 (2011). Y. C. Hu and T. L. Hughes, Phys. Rev. B **84,** 153101 (2011).

ArXiv

--- abstract: 'Let $r, s \ge 2$ be integers. Suppose that the number of blue $r$-cliques in a red/blue coloring of the edges of the complete graph $K_n$ is known and fixed. What is the largest possible number of red $s$-cliques under this assumption? The well known Kruskal-Katona theorem answers this question for $r=2$ or $s=2$. Using the shifting technique from extremal set theory together with some analytical arguments, we resolve this problem in general and prove that in the extremal coloring either the blue edges or the red edges form a clique.' author: - 'Hao Huang[^1]' - 'Nati Linial[^2]' - 'Humberto Naves[^3]' - 'Yuval Peled[^4]' - 'Benny Sudakov[^5]' title: On the densities of cliques and independent sets in graphs --- Introduction {#section_introduction} ============ As usual we denote by $K_s$ the complete graph on $s$ vertices and by $\overline{K}_s$ its complement, the edgeless graph on $s$ vertices. By the celebrated Ramsey’s theorem, for every two integers $r, s$ every sufficiently large graph must contain $K_r$ or $\overline{K}_s$. Turán’s theorem can be viewed as a quantitative version of the case $s=2$. Namely, it shows that among all $\overline{K}_r$-free $n$-vertex graphs, the graph with the least number of $K_2$ (edges) is a disjoint union of $r-1$ cliques of nearly equal size. More generally, one can ask the following question. Fix two graphs $H_1$ and $H_2$, and suppose that we know the number of induced copies of $H_1$ in an $n$-vertex graph $G$. What is the maximum (or minimum) number of induced copies of $H_2$ in $G$? In its full generality, this problem seems currently out of reach, but some special cases already have important implications in combinatorics, as well as other branches of mathematics and computer science. To state these classical results, we introduce some notation. Adjacency between vertices $u$ and $v$ is denoted by $u \sim v$, and the neighbor set of $v$ is denoted by $N(v)$. If necessary, we add a subscript $G$ to indicate the relevant graph. The collection of induced copies of a $k$-vertex graph $H$ in an $n$-vertex graph $G$ is denoted by ${\textup{Ind}}(H; G)$, i.e. $${\textup{Ind}}(H; G) := \{X \subseteq V(G): G[X] \simeq H\}$$ and the *induced $H$-density* is defined as$$d(H; G) := \frac{|{\textup{Ind}}(H; G)|}{\binom{n}{k}}.$$ In this language, Turán’s theorem says that if $d(K_r;G)=0$ then $d(K_2; G)\le 1-\frac{1}{r-1}$ and this bound is tight. For a general graph $H$, Erdős and Stone [@erdos-stone] determined $\max d(K_2; G)$ when $d(H;G)=0$ and showed that the answer depends only on the chromatic number of $H$. Zykov [@zykov] extended Turán’s theorem in a different direction. Given integers $2 \le r<s$, he proved that if $d(K_s;G)=0$ then $d(K_r; G) \le \frac{(s-1) \cdots (s-r)}{(s-1)^r}$. The balanced complete $(s-1)$-partite graphs show that this bound is also tight. For fixed integers $r<s$, the Kruskal-Katona theorem [@katona; @kruskal] states that if $d(K_r; G)=\alpha$ then $d(K_s; G) \le \alpha^{s/r}$. Again, the bound is tight and is attained when $G$ is a clique on some subset of the vertices. On the other hand, the problem of [*minimizing*]{} $d(K_s; G)$ under the same assumption is much more difficult. Even the case $r=2$ and $s=3$ has remained unsolved for many years until it was recently answered by Razborov [@razborov] using his newly-developed flag algebra method. Subsequently, Nikiforov [@nikiforov] and Reiher [@reiher] applied complicated analytical techniques to solve the cases $(r,s)=(2,4)$, and ($r=2$, arbitrary $s$), respectively. In this paper, we study the following natural analogue of the Kruskal-Katona theorem. Given $d(\overline{K}_r; G)$, how large can $d(K_s; G)$ be? For integers $a \ge b > 0$ we let $Q_{a,b}$ be the $a$-vertex graph whose edge set is a clique on some $b$ vertices. The complement of this graph is denoted by $\overline Q_{a,b}$. Let $\mathcal{Q}_a$ denote the family of all graphs $Q_{a,b}$ and its complement $\overline Q_{a,b}$ for $0 < b \le a$. Note that for $r=2$ or $s=2$, the Kruskal-Katona theorem implies that the extremal graph comes from $\mathcal{Q}_n$. Our first theorem shows that a similar statement holds for all $r$ and $s$. \[maintheorem\] Let $r, s \ge 2$ be integers and suppose that $d(\overline{K}_r; G) \geq p$ where $G$ is an $n$-vertex graph and $0 \le p \le 1$. Let $q$ be the unique root of $q^r+rq^{r-1}(1-q)=p$ in $[0,1]$. Then $d(K_s;G) \le M_{r,s,p} + o(1)$, where $$M_{r,s,p} := \max \{(1-p^{1/r})^s + sp^{1/r}(1-p^{1/r})^{s-1}, (1-q)^s\}.$$ Namely, given $d(\overline{K}_r; G)$, the maximum of $d(K_s; G)$ (up to $\pm o_n(1)$) is attained in one of two graphs, (or both), one of the form $Q_{n,t}$ and another $\overline Q_{n,t'}$. We obtain as well a [*stability version*]{} of Theorem \[maintheorem\]. Two $n$-vertex graphs $H$ and $G$ are *$\epsilon$-close* if it is possible to obtain $H$ from $G$ by adding or deleting at most $\epsilon n^2$ edges. As the next theorem shows, every near-extremal graph $G$ for Theorem \[maintheorem\] is $\epsilon$-close to a specific member of $\mathcal{Q}_n$. \[stabilitytheorem\] Let $r, s \ge 2$ be integers and let $p \in [0,1]$. For every $\epsilon > 0$, there exists $\delta > 0$ and an integer $N$ such that every $n$-vertex graph $G$ with $n > N$ satisfying $d(\overline{K}_r;G) \ge p$ and $|d(K_s;G) - M_{r,s,p}| \le \delta$, is $\epsilon$-close to some graph in $\mathcal{Q}_n$. ![Illustration for the case $r=s=3$. The green curve is $(d(\overline K_3;Q_{n,\theta n}), d(K_3;Q_{n,\theta n}))$ for $\theta\in{[0,1]}$, and the red curve defined the same with $\overline Q_{n,\theta n}$. The maximum between the curves is the extremal function in Theorem \[maintheorem\]. The intersection of the curves represents the solution of the max-min problem in Theorem \[thm:max\_min\]](fig1.png){width="50.00000%"} Rather than talking about an $n$-vertex graph and its complement, we can consider a two-edge-coloring of $K_n$. A quantitative version of Ramsey Theorem asks for the minimum number of monochromatic $s$-cliques over all such colorings. Goodman [@goodman] showed that for $r=s=3$, the optimal answer is essentially given by a random two-coloring of $E(K_n)$. In other words, $\min_G d(K_3; G) + d(\overline{K}_3; G) = 1/4-o(1)$. Erdős [@erdos-false] conjectured that the same random coloring also minimizes $d(K_r; G) + d(\overline{K}_r; G)$ for all $r$, but this was refuted by Thomason [@thomason] for all $r \ge 4$. A simple consequence of Goodman’s inequality is that $\min_G \max \{d(K_3; G), d(\overline{K}_3; G)\}=1/8$. The following construction by Franek and Rödl [@franek-rodl] shows that the analogous statement for $r \ge 4$ is again false. Let $H$ be a graph with vertex set $[2]^{13}$, the collection of all $8192$ binary vectors of length $13$. Two vertices are adjacent if the Hamming distance between the corresponding binary vectors is a number in $\{1, 4, 5, 8, 9, 11\}$. Let $G$ be obtained from $H$ by replacing each vertex with a clique of size $n$, and every edge with a complete bipartite graph. The number of $K_4$ and $\overline{K}_4$ in $G$ can be easily expressed in terms of the parameters of $H$ (see [@franek-rodl]), for large enough $n$ one can show that $d(K_4; G)<0.99\cdot \frac{1}{64}$ and $d(\overline{K}_4; G) <0.993\cdot \frac{1}{64}$. While the min-max question remains at present very poorly understood, we succeeded to completely answer the max-min version of this problem. \[thm:max\_min\] $$\max_G \min \{d(K_r; G), d(\overline{K}_r; G)\} = \rho^r +o(1),$$ where $\rho$ is the unique root in $[0,1]$ of the equation $\rho^r = (1-\rho)^r + r\rho(1-\rho)^{r-1}$. This theorem follows easily from Theorem \[maintheorem\]. Moreover, using Theorem \[stabilitytheorem\], we can also show that for every $\epsilon>0$ there is a $\delta>0$ such that every $n$-vertex graph $G$ with $\min \{d(K_r; G), d(\overline{K}_r; G)\} > \rho^r-\delta$ is $\epsilon$-close to a clique of size $\rho n$ or to the complement of this graph. Here we prove these theorems using the method of shifting. In the next section we describe this well-known and useful technique in extremal set theory. Using shifting, we show how to reduce the problem to *threshold graphs*. Section \[section\_main\] contains the proof of our main result for threshold graphs and section \[section\_stability\] contains the proof of the stability result. In Section \[section\_shift\] we sketch a second proof for the case $r=s$, based on a different representation of threshold graphs. We make a number of comments on the analogous problems for hypergraphs in Section \[section\_hyper\]. We finish this paper with some concluding remarks and open problems. Shifting {#section_shifting} ======== [*Shifting*]{} is one of the most important and widely-used tools in extremal set theory. This method allows one to reduce many extremal problems to more structured instances which are usually easier to analyze. Our treatment is rather shallow and we refer the reader to Frankl’s survey article [@frankl-survey] for a fuller account. Let $\mathcal{F}$ be a family of subsets of a finite set $V$, and let $u, v$ be two distinct elements of $V$. We define the *$(u, v)$-shift map* $S_{u\to v}$ as follows: for every $F \in \mathcal{F}$, let $$S_{u\to v}(F, \mathcal{F}) := \left\{\begin{array}{ll} (F \cup \{v\})\setminus \{u\} & \text{if } u \in F, v\not \in F \text{ and } (F \cup \{v\})\setminus \{u\} \not\in \mathcal{F}, \\ F & \text {otherwise.} \end{array}\right.$$ We define the $(u,v)$-shift of $\mathcal{F}$, to be the following family of subsets of $V$: $S_{u\to v}(\mathcal{F}) := \{ S_{u\to v}(F, \mathcal{F}) : F \in \mathcal{F}\}$. We observe that $|S_{u\to v}(\mathcal{F})|=|\mathcal{F}|$. In this context, one may think of $\mathcal{F}$ as a hypergraph over $V$. When all sets in $\mathcal F$ have cardinality $2$ this is a graph with vertex set $V$. As the next lemma shows, shifting of graph does not reduce the number of $l$-cliques in it for every $l$. Recall that ${\textup{Ind}}(K_l;G)$ denotes the collection of all cliques of size $l$ in $G$. \[lemma\_increase\] For every integer $l>0$, every graph $G$, and every $u\neq v \in V(G)$ there holds $$S_{u\to v}({\textup{Ind}}(K_l;G))\subseteq {\textup{Ind}}(K_l;S_{u\to v}(G)).$$ Let $A=S_{u\to v}(B,G)$, where $B$ is an $l$-clique in $G$. First, consider the cases when $u\notin B$ or both $u, v\in B$ or $B\setminus\{u\}\cup\{v\}$ is also a clique in $G$. Then $A=B$ and we need to show that $B$ remains a clique after shifting. Which edge in $B$ can be lost by shifting? It must be some edge $uw$ in $B$ that gets replaced by the non-edge $vw$ (otherwise we can not shift $uw$). Note that $vw$ is not in $B$, since $B$ is a clique. Hence $u, w\in B$ and $v\not\in B$. But then $B\setminus\{u\}\cup\{v\}$ is not a clique, contrary to our assumption. In the remaining case when $u\in B$, $v\notin B$ and $B\setminus\{u\}\cup\{v\}$ is not a clique in $G$, we need to show that $A=B\setminus\{u\}\cup\{v\}$ is a clique after shifting $S_{u\to v}(G)$. Every pair of vertices in $A\setminus\{v\}$ belongs to $B$ and the edge they span is not affected by the shifting. So consider $v\not = w\in A$. If $vw \in E(G)$, this edge remains after shifting. If, however, $vw \notin E(G)$, note that $uw \in E(G)$ since both vertices belong to the clique $B$. In this case $vw=S_{u\to v}(uw, G)$ and the claim is proved. Since shifting edges from $u$ to $v$ is equivalent to shifting non-edges from $v$ to $u$, it is immediate that $S_{u\to v}({\textup{Ind}}(\overline{K_l};G))\subseteq {\textup{Ind}}(\overline{K_l};S_{u\to v}(G))$. Therefore we obtain the following corollary. \[corollary\_increase\] Let $G$ be a graph, let $H = S_{u\to v}(G)$ and let $l$ be a positive integer. Then $$d(K_l; H) \ge d(K_l; G) \textrm{~~and~~~} d(\overline{K}_l; H) \ge d(\overline{K}_l; G).$$ We say that vertex $u$ *dominates* vertex $v$ if $S_{v\to u}(\mathcal{F})=\mathcal{F}$. In the case when $ \cal F$ is a set of edges of $G$, this implies that every $w\not = u$ which is adjacent to $v$ is also adjacent to $u$. If $V=[n]$, we say that a family $\mathcal{F}$ is *shifted* if $i$ dominates $j$ for every $i<j$. Every family can be made shifted by repeated applications of shifting operations $S_{j\to i}$ with $i<j$. To see this note that a shifting operation that changes $\mathcal F$ reduces the following non-negative potential function $\sum_{A\in\mathcal{F}}\sum_{i\in A}i$. As Corollary \[corollary\_increase\] shows, it suffices to prove Theorem \[maintheorem\] for shifted graphs. In Section \[section\_main\] we use the notion of *threshold graphs*. There are several equivalent ways to define threshold graph (see [@chvatal-hammer]), and we adopt the following definition. \[definition\_threshold\] We say that $G=(V,E)$ is a threshold graph if there is an ordering of $V$ so that every vertex is adjacent to either all or none of the preceding vertices. \[lemma\_threshold\] A graph is shifted if and only if it is a threshold graph. Let $G$ be a shifted graph. We may assume that $V=[n]$, and $i$ dominates $j$ in $G$ for every $i<j$. Consider the following order of vertices, $$...,\;3,\;N_G(2)\backslash N_G(3),\;2,\;N_G(1)\backslash N_G(2),\;1,\; V\backslash N_G(1)\;,$$ where the vertices inside the sets that appear here are ordered arbitrarily. We claim that this order satisfies Definition \[definition\_threshold\]. First, every vertex $v\notin N_G(1)$ is isolated. Indeed, if $u\sim v$, then necessarily $v\sim 1$, since $1$ dominates $u$. Therefore, vertex $1$ and its non-neighbors satisfy the condition in the definition. The proof that $G$ is threshold proceeds by induction applied to $G[N_G(1)]$. Conversely, let $G$ be a threshold graph. Let $v_1, v_2, \ldots, v_n$ be an ordering of $V$ as in Definition \[definition\_threshold\]. We say that a vertex is good (resp. bad) if it is adjacent to all (none) of its preceding vertices. Consider two vertices $v_i$ and $v_j$. It is straightforward to show that $v_i$ dominates $v_j$ if either (1) $v_i$ is good and $v_j$ is bad, (2) they are both good and $i>j$ or (3) they are both bad and $i<j$. Therefore we can reorder the vertices by first placing the good vertices in reverse order followed by the bad vertices in the regular order. This new ordering demonstrates that $G$ is shifted. Main result {#section_main} =========== In this section, we prove Theorem \[maintheorem\]. It will be convenient to reformulate the theorem, in a way that is analogous to the Kruskal-Katona theorem. \[maintheorem\_unrestricted\] Let $r, s \ge 3$ be integers and let $a,b > 0$ be real numbers. The maximum (up to $\pm o_n(1)$) of the function $f(G):= \min\{a\cdot d(K_s; G), b \cdot d(\overline{K}_r; G)\}$ over all $n$-vertex graphs is attained in one of two graphs, (or both), one of the form $Q_{n,t}$ and another $\overline Q_{n,t'}$. In particular, $f(G) \le \max \{a \cdot \alpha^{s}, b \cdot \beta^{r}\} + o(1)$, where $\alpha$ is the unique root in $[0,1]$ of $a \cdot \alpha^s = b \cdot [(1-\alpha)^r + r \alpha (1-\alpha)^{r-1}]$ and $\beta$ is the unique root in $[0,1]$ of $b \cdot \beta^r = a \cdot [(1-\beta)^s + s \beta (1-\beta)^{s-1}]$. We turn to show how to deduce Theorem \[maintheorem\] from Theorem \[maintheorem\_unrestricted\]. We assume that $r,s \ge 3$, since the other cases follow from Kruskal-Katona theorem. Let $M$ be the maximum of $d(K_s;G)$ over all graphs $G$ on $n$ vertices with $d(\overline{K}_r;G) \geq p$. Fix such an extremal $G$ with $d(\overline{K}_r;G) = p'\geq p$ and $d(K_s;G) = M$. Now apply Theorem \[maintheorem\_unrestricted\] with $a = p$ and $b = M$ and the same $n$, $r$ and $s$. The extremal graph $G'$ that Theorem \[maintheorem\_unrestricted\] yields, satisfies $$f(G') \ge f(G) = \min\{a\cdot d(K_s;G), b\cdot d(\overline{K}_r;G)\} = p \cdot M,$$ hence $d(K_s;G') \ge M$ and $d(\overline{K}_r;G')\ge p$. Therefore, the same $G'$ is extremal for Theorem \[maintheorem\] as well and we know that the maximum in this theorem is achieved asymptotically by a graph of $\mathcal Q_n$. Note that we can always assume that in the extremal graph $d(\overline{K}_r;G')=p$ since otherwise we can add edges to $G'$ without decreasing $d(K_s;G')$ until $d(\overline{K}_r;G')=p$ is obtained. Therefore the maximum is attained either by a graph of the form $\overline Q_{n,p^{1/r}n}$ or by $Q_{n,(1-q)n}$, where $q^r+rq^{r-1}(1-q)=p$. This implies that asymptotically the maximum in Theorem \[maintheorem\] is indeed $$M_{r,s,p} = \max \{(1-p^{1/r})^s + sp^{1/r}(1-p^{1/r})^{s-1}, (1-q)^s\}.$$ By Corollary \[corollary\_increase\] and Lemma \[lemma\_threshold\], $f(G)$ is maximized by a threshold graph. We turn to prove Theorem \[maintheorem\_unrestricted\] for threshold graphs. Let $G$ be a threshold graph on an ordered vertex set $V$, as in Definition \[definition\_threshold\]. There exists an integer $k>0$, and a partition $A_1,\ldots,A_{2k}$ of $V$ such that 1. If $v\in A_i$ and $u\in A_j$ for $i<j$, then $v<u$. 2. Every vertex in $A_{2i-1}$ (respectively $A_{2i}$) is adjacent to all (none) of its preceding vertices. Let $x_i=\frac{|A_{2i-1}|}{|V|}$ and $y_i=\frac{|A_{2i}|}{|V|}$. Clearly $\sum_{i=1}^k (x_i + y_i) = 1$. Up to a negligible error-term, $$\begin{aligned} d(K_s;G)=p(\mathbf{x},\mathbf{y}) &:= \left(\sum_{i=1}^{k} x_i\right)^s + s \cdot \sum_{i = 1}^{k-1} \left[y_i\cdot \left(\sum_{j=i + 1}^k x_j\right)^{s-1}\right], \\ d(\overline K_r;G)=q(\mathbf{x},\mathbf{y}) &:= \left(\sum_{i=1}^{k} y_i\right)^r + r \cdot \sum_{i = 1}^{k} \left[x_i\cdot \left(\sum_{j=i}^k y_j\right)^{r-1}\right].\end{aligned}$$ Where $\mathbf{x} = (x_1,x_2,\ldots, x_k)$ and $\mathbf{y} = (y_1,y_2,\ldots, y_k)$. Occasionally, $p$ will be denoted by $p_s$ and $q$ by $q_r$ to specify the parameter of these functions. Our problem can therefore be reformulated as follows. For given integers $k\ge 2$, $r,s\ge 3$ and real $a,b>0$, let $W_k \subseteq \mathbb{R}^{2k}$ be the set $$W_k := \left\{(x_1,x_2,\ldots, x_k,y_1,y_2,\ldots,y_k)\in \mathbb{R}^{2k} : x_i,y_i \ge 0 \text{ for all $i$ and } \sum_{i=1}^k (x_i + y_i) = 1\right\}.$$ Let $p, q : W_k \to \mathbb{R}$ be the two homogeneous polynomials defined above, We are interested in maximizing the real function $$\varphi(\mathbf{x},\mathbf{y}) := \min \{a \cdot p(\mathbf{x},\mathbf{y}), b \cdot q(\mathbf{x},\mathbf{y})\}.$$ This problem is well defined since $W_k$ is compact and $\varphi$ is continuous. We say that $(\mathbf x,\mathbf y)\in W_k$ is *non-degenerate* if the set of zeros in the sequence $(y_1, x_2, y_2,\ldots, x_k, y_k)$, with $x_1$ omitted, forms a suffix. If $(\mathbf x,\mathbf y)\in W_k$ is degenerate, then there is a non-degenerate $(\mathbf x',\mathbf y')\in W_k$ with $\varphi(\mathbf{x},\mathbf{y})=\varphi(\mathbf{x}',\mathbf{y}')$. Indeed, if $y_i=0$ and $x_{i+1}\ne 0$ for some $1\le i <k$, let $(\mathbf x',\mathbf y')\in W_{k-1}$ be defined by $$\mathbf x' = (x_1,\ldots,x_{i-1},x_i+x_{i+1},x_{i+2},\ldots,x_k)$$ $$\mathbf y' = (y_1,\ldots,y_{i-1},y_{i+1},\ldots,y_k)$$ It is easy to verify that $p(\mathbf x,\mathbf y)=p(\mathbf x',\mathbf y')$ and $q(\mathbf x,\mathbf y)=q(\mathbf x',\mathbf y')$. By induction on $k$, we assume that $(\mathbf x',\mathbf y')$ is non-degenerate, and by padding $\mathbf x'$ and $\mathbf y'$ with a zero, the claim is proved. The case $x_i=0$ and $y_i\ne 0$ is proved similarly. In particular, $\varphi$ has a non-degenerate maximum in $W_k$. Our purpose is to show that the original problem is optimized by graphs from $\mathcal{Q}_n$. This translates to the claim that a non-degenerate $(\mathbf{x},\mathbf{y})$ that maximizes $\varphi$ is supported only on either $x_1,y_1$ or $y_1,x_2$, which corresponds to either a clique $Q_{n,t}$ or a complement of a clique $\overline{Q}_{n,t}$, respectively. \[lemma\_mainLemma\] Let $(\mathbf{x},\mathbf{y}) \in W_k$ be a non-degenerate maximum of $\varphi$. If $x_1>0$, then for every $i\ge 2$, $x_i=y_i=0$. On the other hand, if $x_1=0$ then $y_i=0$ for every $i\ge 2$, and $x_i=0$ for every $i\ge 3$. We note first that the second part of the lemma is implied by the first part. Define $\mathbf{x}'$ by $$x'_i := \left\{\begin{array}{ll} x_{i+1} & \text{ if } i < k, \\ 0 & \text{ if } i = k. \end{array}\right.$$ Clearly, if $x_1=0$, then $p_s(\mathbf{x,y})=q_s(\mathbf{y,x'})$, $q_r(\mathbf{x,y})=p_r(\mathbf{y,x'})$, and $$\varphi'(\mathbf{y},\mathbf{x'}) := \min \{b \cdot p_r(\mathbf{y},\mathbf{x'}), a \cdot q_s(\mathbf{y},\mathbf{x'})\}=\varphi(\mathbf{x},\mathbf{y}).$$ Since $\varphi$ attains its maximum when $x_1=0$, maximizing it is equivalent to maximizing $\varphi'(\mathbf{y,x'})$. Since $(\mathbf{x},\mathbf{y})$ is non-degenerate, $y_1>0$, and applying the first part of Lemma \[lemma\_mainLemma\] for $\varphi '(\mathbf y,\mathbf x ')$ finishes the proof, by obtaining that for every $i\geq 2$, $y_i=x'_i=0$. The first part of Lemma \[lemma\_mainLemma\] is proved in the following lemmas. We successively show that $x_3=0$, then $y_2=0$ and finally $x_2=0$. Here is a local condition that maximum points of $\varphi$ satisfy. \[lemma\_equality\] If $\varphi$ takes its maximum at a non-degenerate $(\mathbf{x}, \mathbf{y}) \in W_k$, then $a \cdot p(\mathbf{x}, \mathbf{y}) = b\cdot q(\mathbf{x}, \mathbf{y})$. Note that $0<y_1<1$, since $(\mathbf{x},\mathbf{y})\in W$ is non-degenerate. We consider two perturbations of the input, one of which increases $p(\mathbf{x},\mathbf{y})$, and the other increases $q(\mathbf{x},\mathbf{y})$. Consequently, if $a\cdot p(\mathbf{x},\mathbf{y}) \ne b\cdot q(\mathbf{x},\mathbf{y})$, by applying the appropriate perturbation, we increase the smallest between $a\cdot p(\mathbf{x},\mathbf{y})$ and $b\cdot q(\mathbf{x},\mathbf{y})$, thus increasing $\min\{a\cdot p(\mathbf{x},\mathbf{y}), b\cdot q(\mathbf{x},\mathbf{y})\}$, contrary to the maximality assumption. To define the perturbation that increases $p$, let $\mathbf{x'}=\mathbf{x}+t\mathbf{e_1}$ and $\mathbf{y'}=\mathbf{y}-t\mathbf{e_1}$, where $0<t<y_1$, and $\mathbf{e_1}$ is the first unit vector in $\mathbb{R}^k$. Then, $(\mathbf{x'},\mathbf{y'}) \in W$ and $$\frac{\partial p(\mathbf{x'},\mathbf{y'})}{\partial t}= s \left(t+\sum_{i=1}^{k} x_i\right)^{s-1} - s \cdot \left(\sum_{j=2}^k x_j\right)^{s-1} > 0$$ as claimed. In order to increase $q$, consider two cases. If $x_1=0$, let $\mathbf{x'}=\mathbf{x}-t\mathbf{e_2}$ and $\mathbf{y'}=\mathbf{y}+t\mathbf{e_1}$, where $0<t<x_2$. Then, $(\mathbf{x'},\mathbf{y'}) \in W$ and $$\frac{\partial q(\mathbf{x'},\mathbf{y'})}{\partial t}= r \left(t+\sum_{i=1}^{k} y_i\right)^{r-1} - r \cdot \left(\sum_{j=k}^n y_j\right)^{r-1} > 0.$$ If $x_1>0$, we let $\mathbf{x'}=\mathbf{x}-t\mathbf{e_1}$ and $\mathbf{y'}=\mathbf{y}+t\mathbf{e_1}$, where $0<t<x_1$. Then, $$\frac{\partial q(\mathbf{x'},\mathbf{y'})}{\partial t}= r(x_1-t)(r-1)\left(t+\sum_{i=1}^{k} y_i\right)^{r-2}> 0.$$ \[lemma\_small0\] If $(\mathbf{x},\mathbf{y}) \in W_k$ is a non-degenerate maximum of $\varphi$ with $x_1>0$, then $x_3=0$. Suppose, that $x_3 > 0$ and let $1 \le l \le m \le k$. Then $$\begin{aligned} \frac{\partial p}{\partial x_l} &= s \cdot \left(\sum_{i=1}^{k} x_i\right)^{s-1} + s(s-1) \cdot \sum_{i = 1}^{l-1} \left[y_i\cdot \left(\sum_{j=i + 1}^k x_j\right)^{s-2}\right], \\ \frac{\partial q}{\partial x_l} &= r \cdot \left(\sum_{j=l}^k y_j\right)^{r-1},\end{aligned}$$ and $$\begin{aligned} \frac{\partial^2 p}{\partial x_l \partial x_m} &= s(s-1)\cdot \left(\sum_{i=1}^{k} x_i\right)^{s-2} + s(s-1)(s-2) \cdot \sum_{i = 1}^{l-1} \left[y_i\cdot \left(\sum_{j=i + 1}^k x_j\right)^{s-3}\right], \\ \frac{\partial^2 q}{\partial x_l \partial x_m} &\equiv 0.\end{aligned}$$ Clearly $\frac{\partial^2 p}{\partial x_l \partial x_m} = \frac{\partial^2 p}{\partial x_l^2}$, for $l \le m$. We define two matrices $\mathbf{A}$ and $\mathbf{B}$ as following. $$\mathbf{A} = \begin{bmatrix} 1 & 1 & 1 \\ \frac{\partial p}{\partial x_1} & \frac{\partial p}{\partial x_2} & \frac{\partial p}{\partial x_3} \\ \frac{\partial q}{\partial x_1} & \frac{\partial q}{\partial x_2} & \frac{\partial q}{\partial x_3} \\ \end{bmatrix}, \quad \mathbf{B} = \begin{bmatrix} \frac{\partial^2 p}{\partial x_1^2} & \frac{\partial^2 p}{\partial x_1\partial x_2} & \frac{\partial^2 p}{\partial x_1\partial x_3} \\ \frac{\partial^2 p}{\partial x_1\partial x_2} & \frac{\partial^2 p}{\partial x_2^2} & \frac{\partial^2 p}{\partial x_2\partial x_3} \\ \frac{\partial^2 p}{\partial x_1\partial x_3} & \frac{\partial^2 p}{\partial x_2\partial x_3} & \frac{\partial^2 p}{\partial x_3^2} \\ \end{bmatrix}=\begin{bmatrix} \frac{\partial^2 p}{\partial x_1^2} & \frac{\partial^2 p}{\partial x_1^2} & \frac{\partial^2 p}{\partial x_1^2} \\ \frac{\partial^2 p}{\partial x_1^2} & \frac{\partial^2 p}{\partial x_2^2} & \frac{\partial^2 p}{\partial x_2^2} \\ \frac{\partial^2 p}{\partial x_1^2} & \frac{\partial^2 p}{\partial x_2^2} & \frac{\partial^2 p}{\partial x_3^2} \\ \end{bmatrix}.$$ It is easy to see that if $(\mathbf{x},\mathbf{y})$ is non-degenerate with $x_3>0$, then $\frac{\partial^2 p}{\partial x_3^2}>\frac{\partial^2 p}{\partial x_2^2}>\frac{\partial^2 p}{\partial x_1^2}>0$. This implies that $\mathbf{B}$ is positive definite. For a vector $\mathbf{v}\in\mathbb{R}^3$ and $\epsilon>0$, we define $\mathbf{x'}$ by $$x_i'= \left\{\begin{array}{ll} x_i + \epsilon v_i & \text{ if } i \le 3, \\ x_i & \text{ if } i > 3, \\ \end{array}\right.$$ If $\mathbf{A}$ is invertible, let $\mathbf{v}$ be the (unique) vector for which $$\mathbf{A} \cdot \mathbf{v}^{T} = \begin{bmatrix} 0 \\ 1 \\ 1 \\ \end{bmatrix}.$$ In particular $\sum_i x_i' = \sum_i x_i$. For $\epsilon$ sufficiently small, $$p(\mathbf{x}', \mathbf{y}) = p(\mathbf{x},\mathbf{y}) + \epsilon + O(\epsilon^2)>p(\mathbf{x}, \mathbf{y})$$$$q(\mathbf{x}', \mathbf{y}) = q(\mathbf{x}, \mathbf{y}) + \epsilon> q(\mathbf{x}, \mathbf{y})$$ contrary to the maximality of $(\mathbf{x},\mathbf{y})$. If $\mathbf{A}$ is singular, pick some $\mathbf{v} \neq 0$ with $\mathbf{A} \cdot \mathbf{v}^{T} = \mathbf{0}$. Again $\sum_i x_i' = \sum_i x_i$. Since $\mathbf{B}$ is positive definite, for a sufficiently small $\epsilon$, $$p(\mathbf{x}', \mathbf{y}) = p(\mathbf{x},\mathbf{y}) + \frac{\epsilon^2}{2}\cdot \mathbf{v} \cdot \mathbf{B} \cdot \mathbf{v}^{T} + O(\epsilon^3)>p(\mathbf{x},\mathbf{y})$$$$q(\mathbf{x}', \mathbf{y}) = q(\mathbf{x}, \mathbf{y}),$$ Contradicting Lemma \[lemma\_equality\]. \[lemma\_small1\] If $(\mathbf{x},\mathbf{y}) \in W_k$ is a non-degenerate maximum of $\varphi$ with $x_1> 0$, then $y_2=0$. By Lemma \[lemma\_small0\] we may assume that $x_i=y_i=0$ for all $i \ge 3$. Suppose, towards contradiction, that $y_2\ne 0$. Let $$\mathbf{M} = \begin{bmatrix} a_1 & a_2 \\ b_1 & b_2 \end{bmatrix},$$ where $$\begin{aligned} a_1 = \frac{\partial p}{\partial x_1} - \frac{\partial p}{\partial x_2} &= -s(s-1)\cdot y_1 \cdot x_2^{s-2},\quad & b_1 = \frac{\partial q}{\partial x_1} - \frac{\partial q}{\partial x_2} &= r\cdot ((y_1 + y_2)^{r-1} - y_2^{r-1}), \\ a_2 = \frac{\partial p}{\partial y_1} - \frac{\partial p}{\partial y_2} &= s\cdot x_2^{s-1}, \quad & b_2 = \frac{\partial q}{\partial y_1} - \frac{\partial q}{\partial y_2} &= -r(r-1) \cdot x_2\cdot y_2^{r-2}, \\\end{aligned}$$ If $\mbox{rank}(\mathbf{M})=2$, then there is a vector $\mathbf{v}=\left(v_1 \atop v_2\right)$ such that $\mathbf{M}\cdot \mathbf{v}= \left(1 \atop 1\right)$. Define $x'_1=x_1+\epsilon v_1, x'_2=x_2-\epsilon v_1$ and $y'_1=y_1+\epsilon v_2, y'_2=y_2-\epsilon v_2$. Then $x'_1+x'_2+y'_1+y'_2=1$ and for sufficiently small $\epsilon>0$ $$\begin{aligned} p(\mathbf{x}', \mathbf{y}') &= & p(\mathbf{x},\mathbf{y})+\epsilon\Big(\frac{\partial p}{\partial x_1}v_1-\frac{\partial p}{\partial x_2}v_1+ \frac{\partial p}{\partial y_1}v_2-\frac{\partial p}{\partial y_2}v_2 \Big)+ O(\epsilon^2)\\ &= &p(\mathbf{x},\mathbf{y})+\epsilon\big(a_1v_1+a_2v_2\big)+ O(\epsilon^2)=p(\mathbf{x},\mathbf{y})+\epsilon+ O(\epsilon^2)>p(\mathbf{x},\mathbf{y}).\end{aligned}$$ Similarly $q(\mathbf{x}', \mathbf{y}')=q(\mathbf{x},\mathbf{y})+\epsilon+ O(\epsilon^2)>q(\mathbf{x},\mathbf{y})$. Thus $(\mathbf{x},\mathbf{y})$ cannot be a maximum of $\varphi$. Hence, $\mbox{rank}(\mathbf{M})\le 1$, and in particular $$\det \begin{bmatrix} a_1 & b_1 \\ a_2 & b_2 \end{bmatrix} = 0,$$ which implies that $$0 = x_2^{s-1}y_2^{r-1}\left((r-1)(s-1)\frac{y_1}{y_2} - \left(\frac{y_1}{y_2}+1\right)^{r-1} + 1\right), \\$$ The function $$g(\alpha) = (r-1)(s-1)\alpha-(\alpha + 1)^{r-1}+1$$ is strictly concave for $\alpha > 0$ and vanishes at $0$. Since $\alpha=0$ is not a maximum of $g$, the equation $g\left(\frac{y_1}{y_2}\right)=0$ determines $\frac{y_1}{y_2}$ uniquely. Denote $\alpha=\frac{y_1}{y_2}$, and consider the following change of variables. $$\begin{aligned} x_1' &= x_1 + \frac{1}{1 + (r-1)(s-1)\alpha}\cdot x_2,\quad & x_2' &= \frac{(r-1)(s-1)\alpha}{1 + (r-1)(s-1)\alpha} \cdot x_2\\ y_1' &= y_1 + y_2 = (\alpha + 1) y_2,\quad & y_2' &= 0\end{aligned}$$ Clearly, $x_1' + x_2' = x_1 + x_2$ and $y_1' + y_2' = y_1 + y_2$. Moreover, $$\begin{aligned} q(\mathbf{x}', \mathbf{y}') &= (y_1')^{r} + r\cdot x_1' \cdot (y_1')^{r-1}\\ &= (y_1 + y_2)^r + r\cdot x_1 \cdot (y_1 + y_2)^{r-1} + \frac{r\cdot x_2 \cdot (y_1 + y_2)^{r-1}}{1 + (r-1)(s-1)\alpha} \\ &= (y_1 + y_2)^r + r\cdot x_1 \cdot (y_1 + y_2)^{r-1} + \frac{r\cdot (1+\alpha)^{r-1} \cdot x_2 \cdot y_2^{r-1}}{(1 + \alpha)^{r-1}} = q(\mathbf{x}, \mathbf{y}) \\ p(\mathbf{x}', \mathbf{y}') &= (x_1' + x_2')^{s} + s \cdot y_1' \cdot (x_2')^{s-1} \\ &= (x_1 + x_2)^s + s\cdot (\alpha + 1)\cdot \left(\frac{(r-1)(s-1)\alpha}{1 + (r-1)(s-1)\alpha}\right)^{s-1}\cdot y_2 \cdot x_2^{s-1} \\ &> (x_1 + x_2)^s + s\cdot \alpha \cdot y_2 \cdot x_2^{s-1} = p(\mathbf{x}, \mathbf{y}),\end{aligned}$$ Where the last inequality is a consequence of Lemma \[lemma\_ineq\] below. This contradicts Lemma \[lemma\_equality\]. \[lemma\_ineq\] Let $r,s \ge 3$ be integers. Let $\alpha > 0$ be the unique positive root of $$(\alpha + 1)^{r-1} - 1 = (r-1)(s-1)\alpha.$$ Then $$\left(1 + \frac{1}{(r-1)(s-1)\alpha}\right)^{s-1} < 1 + \frac{1}{\alpha}.$$ First, we show that $(r-1) \alpha > 1$. Let $t = (r-1)\alpha$ and assume, by contradiction, that $t \le 1$. For $0 < t\le 1$, we have $e^t < 1 + 2t$. On the other hand, $e \ge (1+\alpha)^{1/\alpha}$, implying $e^t \ge \left( 1 + \alpha\right)^{t / \alpha} = (1 + \alpha)^{r-1}$. Thus we have $2t > (1+\alpha)^{r-1}-1 = (r-1)(s-1)\alpha = (s-1)t$, which implies $2 > s-1$, a contradiction. Therefore $(r-1)\alpha > 1$. Also, since $1+x<e^x$ for all $x>0$, we have that $\big(1 + \frac{1}{(r-1)(s-1)\alpha}\big)^{s-1} < e^{\frac{1}{(r-1)\alpha}}$. So it suffices to show that $e^{\frac{1}{(r-1)\alpha}} \le 1 + \frac{1}{\alpha}$. But since $(r-1)\alpha > 1$, we have $$\left(1 + \frac{1}{\alpha}\right)^{(r-1)\alpha} > 1 + \frac{(r-1)\alpha}{\alpha} = r \ge 3 > e,$$ which finishes the proof of the lemma. \[lemma\_small2\] If $(\mathbf{x},\mathbf{y}) \in W_k$ is a non-degenerate maximum of $\varphi$ with $x_1>0$, then $x_2=0$. This proof is very similar to the proof of Lemma \[lemma\_small1\]. Now $x_1, x_2, y_1 > 0$ and $x_1 + x_2 + y_1 = 1$. Also $$\begin{aligned} p(\mathbf{x},\mathbf{y}) &= (x_1 + x_2)^s + s \cdot y_1 \cdot x_2^{s-1}, \\ q(\mathbf{x},\mathbf{y}) &= y_1^r + r\cdot x_1 \cdot y_1^{r-1}.\end{aligned}$$ Let $$\mathbf{M} = \begin{bmatrix} a_1 & a_2 \\ b_1 & b_2 \end{bmatrix},$$ where $$\begin{aligned} a_1 = \frac{\partial p}{\partial x_1} - \frac{\partial p}{\partial x_2} &= -s(s-1) \cdot y_1 \cdot x_2^{s-2}, \quad & b_1 = \frac{\partial q}{\partial x_1} - \frac{\partial q}{\partial x_2} &= r\cdot y_1^{r-1}, \\ a_2 = \frac{\partial p}{\partial y_1} - \frac{\partial p}{\partial x_1} &= -s\cdot ((x_1+x_2)^{s-1} - x_2^{s-1}), \quad & b_2 = \frac{\partial q}{\partial y_1} - \frac{\partial q}{\partial x_1} &= r(r-1)\cdot x_1\cdot y_1^{r-2}, \\\end{aligned}$$ If $\mathbf{M}$ is nonsingular, then there is a vector $\mathbf{v}=\left(v_1 \atop v_2\right)$ such that $\mathbf{M}\cdot \mathbf{v}= \left(1 \atop 1\right)$. Define $x'_1=x_1+\epsilon (v_1-v_2), x'_2=x_2-\epsilon v_1$ and $y'_1=y_1+\epsilon v_2$. Then $x'_1+x'_2+y'_1=1$ and for sufficiently small $\epsilon>0$ $$\begin{aligned} p(\mathbf{x}', \mathbf{y}') &=& p(\mathbf{x},\mathbf{y})+\epsilon\Big(\frac{\partial p}{\partial x_1}(v_1-v_2)-\frac{\partial p}{\partial x_2}v_1+ \frac{\partial p}{\partial y_1}v_2\Big)+ O(\epsilon^2)\\ &=& p(\mathbf{x},\mathbf{y})+\epsilon\big(a_1v_1+a_2v_2\big)+ O(\epsilon^2)=p(\mathbf{x},\mathbf{y})+\epsilon+ O(\epsilon^2)>p(\mathbf{x},\mathbf{y}).\end{aligned}$$ Similarly $q(\mathbf{x}', \mathbf{y}')=q(\mathbf{x},\mathbf{y})+\epsilon+ O(\epsilon^2)>q(\mathbf{x},\mathbf{y})$ and therefore $(\mathbf{x},\mathbf{y})$ cannot be a maximum of $\varphi$. Hence, $$\det \begin{bmatrix} a_1 & b_1 \\ a_2 & b_2 \end{bmatrix} = 0,$$ which implies $$0 = y_1^{r-1}x_2^{s-1}\left((r-1)\cdot (s-1) \cdot\frac{x_1}{x_2} - \left(\frac{x_1}{x_2}+1\right)^{s-1} + 1\right).$$ Let $\gamma = \frac{x_1}{x_2} > 0$. Then $1+(r-1)(s-1)\gamma-(1+\gamma)^{s-1}=0$ and concavity of the left hand side shows that $\gamma$ is determined uniquely by this equation. Now make the following substitution: $$\begin{aligned} x_1' &= 0\\ x_2' &= x_1 + x_2 = (1 + \gamma) \cdot x_2\\ y_1' &= \frac{1}{1 + (r-1)(s-1)\gamma}\cdot y_1\\ y_2' &= \frac{(r-1)(s-1)\gamma}{1 + (r-1)(s-1)\gamma} \cdot y_1\end{aligned}$$ Clearly $x_1' + x_2' = x_1 + x_2$ and $y_1' + y_2' = y_1$. Since $(1+\gamma)^{s-1}=1+(r-1)(s-1)\gamma$, we have $$\begin{aligned} p(\mathbf{x}', \mathbf{y}') &= (x_2')^{s} + s\cdot y_1' \cdot (x_2')^{s-1}\\ &= (x_1 + x_2)^s + s\cdot y_1\cdot x_2^{s-1} = p(\mathbf{x},\mathbf{y})\\ q(\mathbf{x}', \mathbf{y}') &= (y_1' + y_2')^{r} + r\cdot x_2' \cdot (y_2')^{r-1} \\ &= y_1^r + r\cdot \frac{(1+\gamma)}{\gamma}\cdot\left(\frac{(r-1)(s-1)\gamma}{1 + (r-1)(s-1)\gamma}\right)^{r-1}\cdot x_1 \cdot y_1^{r-1} \\ &> y_1^r + r\cdot x_1 \cdot y_1^{r-1} = q(\mathbf{x}, \mathbf{y}),\end{aligned}$$ Where the last inequality follows from Lemma \[lemma\_ineq\], with $r$ and $s$ switched. Again, this contradicts Lemma \[lemma\_equality\]. By combining Lemmas \[lemma\_equality\] – \[lemma\_small2\], we obtain a proof of Lemma \[lemma\_mainLemma\], which states that the maximum of $\varphi$ is attained by a non-degenerate $(\mathbf{x},\mathbf{y})$ supported only on either $x_1,y_1$ or $y_1,x_2$. In the first case, let $x_1=\alpha$ and $y_1=1-\alpha$. Then by Lemma \[lemma\_equality\], $a\cdot p(\mathbf{x},\mathbf{y})=a\cdot \alpha^s=b\cdot q(\mathbf{x},\mathbf{y})=b\big[(1-\alpha)^r+r\alpha(1-\alpha)^{r-1}\big]$ and $\varphi(\mathbf{x},\mathbf{y})=a \cdot \alpha^s$. In the second case, let $y_1=\beta$ and $x_2=1-\beta$. Then $b\cdot q(\mathbf{x},\mathbf{y})=b\cdot \beta^r=a \cdot p(\mathbf{x},\mathbf{y})=a\big[(1-\beta)^s+s(1-\beta)^{s-1}\big]$ and $\varphi(\mathbf{x},\mathbf{y})=b \cdot \beta^r$. This shows that the maximum of $\varphi$ is $\max \{a \cdot \alpha^{s}, b \cdot \beta^{r}\}$ with $\alpha, \beta$ satisfying the above equations. In terms of the original graph, this proves that $\varphi$ is maximized by a graph of the form $Q_{n,t}$ or $\overline Q_{n,t}$, respectively. In particular, our problem has at most two extremal configurations (in some cases a clique and the complement of a clique can give the same value of $\varphi$). Stability analysis {#section_stability} ================== In this section we discuss the proof of Theorem \[stabilitytheorem\]. In essentially the same way that Theorem \[maintheorem\_unrestricted\] implies Theorem \[maintheorem\], this theorem follows from a stability version of Theorem \[maintheorem\_unrestricted\]: \[stabilitytheorem\_unrestricted\] Let $r, s \ge 3$ be integers and let $a,b > 0$ be real. For every $\epsilon > 0$, there exists $\delta > 0$ and an integer $N$ such that every $n$-vertex $G$ with $n > N$ for which $$f(G) \ge \max \{a \cdot \alpha^{s}, b \cdot \beta^{r}\} - \delta$$ is $\epsilon$-close to some graph in $\mathcal{Q}_n$. Here $f, \alpha$ and $\beta$ are as in Theorem \[maintheorem\_unrestricted\]. If $G$ is a threshold graph, the claim follows easily from Lemma \[lemma\_mainLemma\]. Since $G$ is a threshold graph, $f(G) = \varphi(\mathbf{x},\mathbf{y})+o(1)$ for some $(\mathbf{x},\mathbf{y})\in W_k$ and some integer $k$. As this lemma shows, the continuous function $\varphi$ attains its maximum on the compact set $W_k$ at most twice, and this in points that correspond to graphs from $\mathcal{Q}_n$. Since $f(G)$ is $\delta$-close to the maximum, it follows that $(\mathbf{x},\mathbf{y})$ must be $\epsilon'$-close to at least one of the two optimal points in $W_k$. This, in turn implies $\epsilon$-proximity of the corresponding graphs. For the general case, we use the stability version of the Kruskal-Katona theorem due to Keevash [@keevash]. Suppose $G$ is a large graph such that $f(G) \ge \max \{a \cdot \alpha^{s}, b \cdot \beta^{r}\} - \delta$. Let $G_1$ be the shifted graph obtained from $G$. Thus $G_1$ is a threshold graph with the same edge density as $G$, and $f(G_1) \ge f(G)$ by Corollary \[corollary\_increase\]. Pick a small $\epsilon' > 0$. We just saw that for $\delta$ sufficiently small, $G_1$ is $\epsilon'$-close to $G_{max} \in \mathcal{Q}_n$. As we know, either $G_{max}= Q_{n,t}$ or $G_{max}= \bar Q_{n,t}$ for some $0 < t \le n$. We deal with the former case, and the second case can be done similarly. Now $|d(K_2;G) - d(K_2;G_{max})| \le \epsilon'$, since $G$ and $G_1$ have the same edge density. Moreover, $d(K_s; G) \ge d(K_s;G_{max}) -\delta/a$, because $f(G) \ge f(G_{max}) - \delta$. Since $G_{max}$ is a clique, it satisfies the Kruskal-Katona inequality with equality. Consequently $G$ has nearly the maximum possible $K_s$-density for a given number of edges. By choosing $\epsilon'$ and $\delta$ small enough and applying Keevash’s stability version of Kruskal-Katona inequality, we conclude that $G$ and $G_{max}$ are $\epsilon$-close. Second proof {#section_shift} ============ In this section we briefly present the main ingredients for an alternative approach to Theorem \[maintheorem\]. We restrict ourselves to the case $r=s$. This proof reduces the problem to a question in the calculus of variations. Such calculations occur often in the context of shifted graphs. Let $G$ be a shifted graph with vertex set $[n]$ with the standard order. Then, there is some $n \ge i \ge 1$ such that $A=\{1,...,i\}$ spans a clique, whereas $B=\{i+1,...,n\}$ spans an independent set. In addition, there is some non-increasing function $F:A\rightarrow B$ such that for every $j \in A$ the highest index neighbor of $j$ in $B$ is $F(j)$, and all vertices of $B$ up to index $F(j)$ are connected to $j$. Let $x$ be the relative size of $A$ and $1-x$ the relative size of $B$. In this case we can express (up to a negligible error term) $$\begin{aligned} d(\overline{K}_k; G)&=&{n \choose k}^{-1}\left[{(1-x)n \choose k} + \sum_{1 \leq j \leq xn} {n-F(j) \choose k-1} \right]=(1-x)^k+\frac{k}{n}\sum_{1 \leq j \leq xn}\left(\frac{n-F(j)}{n}\right)^{k-1}\\ &=& (1-x)^k+kx(1-x)^{k-1}\sum_{1 \leq j \leq xn} \frac{1}{nx}\left(1-\frac{F(j)-xn}{(1-x)n}\right)^{k-1}.\end{aligned}$$ Let $f$ be a non-increasing function $f:{[0,1]}\rightarrow{[0,1]}$ such that $f(t)=\frac{F(j)-xn}{(1-x)n}$ for every $\frac{j-1}{xn} \leq t \leq \frac{j}{xn}$ (Think of $f$ as a relative version of $F$ both on its domain with respect to $A$ and its codomain with respect to $B$). Then we can express $d(\overline{K}_k; G)$ in terms of $x$ and $f$ $$d(\overline{K}_k; G)=(1-x)^k+kx(1-x)^{k-1}\int_{0}^{1}(1-f(t))^{k-1}dt={d(\overline{K_k};G_{x,f})}.$$ Similarly one can show that $$d(K_k; G)=x^k+kx^{k-1}(1-x)\int_{0}^{1}(k-1)t^{k-2}f(t)dt={d(K_k;G_{x,f})}.$$ Note that in this notation, $x=\theta$, $f=0$ (resp. $x=1-\theta$, $f=1$) corresponds to $Q_{n, \theta \cdot n}$, (resp. $\overline Q_{n, \theta \cdot n}$). To prove Theorem \[maintheorem\] for the case $r=s=k$, we show that assuming ${d(K_k;G_{x,f})}\geq \alpha$, the maximum of ${d(\overline{K_k};G_{x,f})}$ is attained for either $f=0$ or $f=1$. For this purpose, we prove upper bounds on the integrals. \[lemma\_integrals\] If $f:{[0,1]}\rightarrow{[0,1]}$ is a non-increasing function, then $$\int_{0}^{1}(1-f(t))^{k-1}dt\leq\max\left\{ 1-\left(\int_{0}^{1}(k-1)t^{k-2}f(t)dt\right)^{\frac{1}{k-1}},\left(1-\int_{0}^{1}(k-1)t^{k-2}f(t)dt\right)^{k-1}\right\}.$$ The bounds in Lemma \[lemma\_integrals\] are tight. Equality with the first term holds for $f$ that takes only the values $1$ and $0$, and equality with the second term occurs for $f$ a constant function. Proving Theorem \[maintheorem\] for such functions is done using rather standard (if somehow tedious) calculations. Lemma \[lemma\_integrals\] itself is reduced to the following lemma through a simple affine transformation and normalization. What non-decreasing function in ${[0,1]}$ minimizes the inner product with a given monomial? \[lemma\_reduced\] Let $g:{[0,1]}\rightarrow[0,B]$ be a non-decreasing function with $B\geq 1$ and $\|g\|_{k-1}=1$. Then $$\langle (k-1)t^{k-2},g \rangle=\int_{0}^{1}(k-1)t^{k-2}g(t)dt\geq\min\left\{B\left(1-\left(1-\frac{1}{B^{k-1}}\right)^{k-1}\right),1\right\}.$$ Equality with the first term holds for $$g(t)=\left\{ \begin{matrix} 0 & t<1-\frac{1}{B^{k-1}}\\ B & t\geq 1-\frac{1}{B^{k-1}} \end{matrix} \right.$$ The second equality holds for $g=1$. We omit the proof which is based on standard calculations and convexity arguments. Shifting in hypergraphs {#section_hyper} ======================= In this section, we will discuss a possible extension of Lemma \[lemma\_increase\] to hypergraphs. Consider two set systems $\mathcal{F}_1$ and $\mathcal{F}_2$ with vertex sets $V_1$ and $V_2$ respectively. A (not necessarily induced) *labeled copy of $\mathcal{F}_1$ in $\mathcal{F}_2$* is an injection $I:V_1 \to V_2$ such that $I(F) \in \mathcal{F}_2$ for every $F \in \mathcal{F}_1$. We denote by ${\textup{Cop}}(\mathcal{F}_1;\mathcal{F}_2)$ the set of all labeled copies of $\mathcal{F}_1$ in $\mathcal{F}_2$ and let $$t(\mathcal{F}_1; \mathcal{F}_2):= |{\textup{Cop}}(\mathcal{F}_1;\mathcal{F}_2)|.$$ Recall that a vertex $u$ *dominates* vertex $v$ if $S_{v\to u}(\mathcal{F})=\mathcal{F}$. If either $u$ dominates $v$ or $v$ dominates $u$ in a family $\mathcal{F}$, we call the pair $\{u, v\}$ *stable* in $\mathcal{F}$. If every pair is stable in $\mathcal{F}$, then we call $\mathcal{F}$ a *stable set system*. \[theorem\_increase\] Let $\mathcal{H}$ be a stable set system and let $\mathcal{F}$ be a set system. For every two vertices $u, v$ of $\mathcal{F}$ there holds $$t(\mathcal{H}; S_{u\to v}(\mathcal{F})) \ge t(\mathcal{H}; \mathcal{F}).$$ Let $G$ be an arbitrary graph and let $H$ be a threshold graph $H$. Then $$t(H; S_{u\to v}(G)) \ge t(H; G),$$ for every two vertices $u, v$ of $G$. We define a new shifting operator $\tilde{S}_{u\to v}$ for sets of labeled copies. First, for every $u,v\in V$, and a labeled copy $I:U\to V$, define $I_{u\leftrightarrow v}: U\to V$ by $$I_{u\leftrightarrow v}(w) = \left\{\begin{array}{ll} I(w) & \text{ if } I(w) \ne u, v,\\ v & \text{ if } I(w) = u, \\ u & \text{ if } I(w) = v \end{array}\right.$$ For $\mathcal{I}$ a set of labeled copies, $I\in\mathcal{I}$, we let $$\tilde{S}_{u\to v}(I, \mathcal{I}) = \left\{\begin{array}{ll} I_{u\leftrightarrow v} & \text{if } I_{u\leftrightarrow v}\not\in \mathcal{I} \text{ and } \textup{Im}(I) \cap \{u,v\} = \{u\},\\ I_{u\leftrightarrow v} & \text{if } I_{u\leftrightarrow v}\not\in \mathcal{I}, \{u,v\}\subset \textup{Im}(I), \text{ and } I^{-1}(u) \text{ dominates } I^{-1}(v) \text{ in } \mathcal{H},\\ I & \text{otherwise}. \end{array}\right.$$ Finally, let $\tilde{S}_{u\to v}(\mathcal{I}):= \{\tilde{S}_{u\to v}(I,\mathcal{I}) : I \in \mathcal{I}\}$. Clearly, $|\tilde{S}_{u\to v}(\mathcal{I})| = |\mathcal{I}|$, and we prove that $$\tilde{S}_{u\to v}({\textup{Cop}}(\mathcal{H};\mathcal{F}))\subseteq {\textup{Cop}}(\mathcal{H};S_{u\to v}(\mathcal{F}))$$ thereby proving that $t(\mathcal{H};S_{u\to v}(\mathcal{F})) \ge t(\mathcal{H}; \mathcal{F})$. As often in shifting, the proof is done by careful case analysis which is omitted. Concluding remarks {#section_concluding} ================== In this paper, we studied the relation between the densities of cliques and independent sets in a graph. We showed that if the density of independent sets of size $r$ is fixed, the maximum density of $s$-cliques is achieved when the graph itself is either a clique on a subset of the vertices, or a complement of a clique. On the other hand, the problem of minimizing the clique density seems much harder and has quite different extremal graphs for various values of $r$ and $s$ (at least when $\alpha=0$, see [@das-et-al; @pikhurko]). Given that $d(\overline{K}_r; G) = \alpha$ for some integer $r \ge 2$ and real $\alpha \in [0,1]$, which graphs minimize $d(K_s; G)$? In particular, when $\alpha=0$ we ask for the least possible density of $s$-cliques in graphs with independence number $r-1$. This is a fifty-year-old question of Erdős, which is still widely open. Das et al [@das-et-al], and independently Pikhurko [@pikhurko], solved this problem for certain values of $r$ and $s$. It would be interesting if one could describe how the extremal graph changes as $\alpha$ goes from $0$ to $1$ in these cases. As mentioned in the introduction, the problem of minimizing $d(K_s; G)$ in graphs with fixed density of $r$-cliques for $r<s$ is also open and so far solved only when $r=2$.\ [**Note added in proof.**]{} After writing this paper, we learned that P. Frankl, M. Kato, G. Katona and N. Tokushige [@frankl-kato-katona-tokushige] independently considered the same problem and obtained similar results when $r=s$. [**Acknowledgment.**]{} We would like to thank the anonymous referee for valuable comments and suggestions which improve the presentation of the paper. [99]{} M. H. Albert, M. D. Atkinson, C. C. Handley, D. A. Holton and W. Stromquist, , (2002), \#R5. V. Chvátal and P. Hammer, , (1977), 145–162. S. Das, H. Huang, J. Ma, H. Naves, and B. Sudakov, , (2013), 344–373. P. Erdős, , (1962), 459–464. P. Erdős and A. Stone, , (1946), 1087–1091. P. Frankl, , **123** (1987), 81–110. P. Frankl, M. Kato, G. Katona, and N. Tokushige, , , **103** (2013), 415–427. A. Goodman, , (1959), 778–783. G. Katona, , in Theory of Graphs, Akadémia Kiadó, Budapest (1968), 187–207. P. Keevash, , (2008), 1685–1703. J. Kruskal, , Mathematical Optimization Techniques, Univ. of California Press (1963), 251–278. V. Nikiforov, , (2011), 1599–1618. O. Pikhurko and E. R. Vaughan, , (2013), 910–934. A. Razborov, , , [**17**]{} (2008), 603–618. C. Reiher, , manuscript. A. Thomason, , (1989), 246–255. P. Turán, , , [**48**]{} (1941), 436–452. A. Zykov, , 66 (1949), 163–188. F. Franek and V. Rödl, , (1993), 199–203. [^1]: School of Mathematics, Institute for Advanced Study, Princeton 08540. Email: [huanghao@math.ias.edu]{}. Research supported in part by NSF grant DMS-1128155. [^2]: School of Computer Science and engineering, The Hebrew University of Jerusalem, Jerusalem 91904, Israel. Email: [nati@cs.huji.ac.il]{}. Research supported in part by the Israel Science Foundation and by a USA-Israel BSF grant. [^3]: Department of Mathematics, UCLA, Los Angeles, CA 90095. Email: [hnaves@math.ucla.edu]{}. [^4]: School of Computer Science and engineering, The Hebrew University of Jerusalem, Jerusalem 91904, Israel. Email: [yuvalp@cs.huji.ac.il]{} [^5]: Department of Mathematics, ETH, 8092 Zurich, Switzerland and Department of Mathematics, UCLA, Los Angeles, CA 90095. Email: bsudakov@math.ucla.edu. Research supported in part by SNSF grant 200021-149111 and by a USA-Israel BSF grant.

ArXiv

--- abstract: 'Data exhibiting heavy-tails in one or more dimensions is often studied using the [framework]{} of regular variation. In a multivariate setting this [requires identifying]{} specific forms of dependence in the data; this means [identifying that the]{} data tends to concentrate along particular directions and does not cover the full space. This is observed in various data sets [from]{} finance, insurance, network traffic, social networks, etc. In this paper we discuss the notions of full and strong asymptotic dependence for bivariate data [along with the idea of hidden regular variation in these cases]{}. [In a risk analysis setting, this leads to improved risk estimation accuracy when regular methods provide a zero estimate of risk]{}. Analyses of both real and simulated data sets illustrate concepts [of generation and detection of such models]{}.' address: - | Engineering Systems and Design,\ Singapore University of Technology and Design,\ Singapore 487372\ - | School of ORIE,\ Cornell University,\ Ithaca, NY 14853 USA\ author: - - title: Hidden Regular Variation under Full and Strong Asymptotic Dependence --- Introduction ============ Data that may be modeled by distributions having heavy tails appear in many contexts, for example, hydrology ([@anderson:meerschaert:1998]), finance ([@smith:2003]), insurance ([@embrechts:kluppelberg:mikosch:1997]), Internet traffic ([@crovella:bestavros:taqqu:1998]), social networks and random graphs ([@durrett:2010; @bollobas:borgs:chayes:riordan:2003; @resnick:samorodnitsky:towsley:davis:willis:wan:2016; @resnick:samorodnitsky:2015]) and risk management ([@das:embrechts:fasen:2013; @ibragimov:jaffee:walden:2011]). Empirical evidence often indicates heavy-tailed marginal distributions and the dependence structure between the various components must be discerned. We focus here on the case where components are strongly dependent. The purpose of this paper is twofold. First, the paper encourages a definition of strong asymptotic [or extremal]{} dependence that means the limit measure of regular variation concentrates on a cone smaller than the full state space. Thus, directions where multivariate data from such a model [are]{} found fall in a restricted set. Secondly, the paper shows that strong asymptotic dependence is a tractable case for the applicability of [*hidden regular variation*]{}. Hidden regular variation (HRV) [@resnick:2002a; @resnick:2008; @resnickbook:2007; @das:mitra:resnick:2013; @das:resnick:2015; @lindskog:resnick:roy:2014] is often considered for multivariate data exhibiting heavy tails when [ *asymptotic independence*]{} is present. Asymptotic independence in a bivariate data set of positive values implies that both coordinates cannot be large simultaneously and therefore the multivariate regular variation (MRV) limit measure concentrates on the co-ordinate axes. To improve risk estimation, one then seeks HRV on the non-negative orthant after removing the two axes. However, hidden regular variation is applicable whenever the limit measure of regular variation in standard scale concentrates on a cone which is smaller than the entire state space, and [is]{} not restricted to the case of asymptotic independence; see for details. If the limit measure of regular variation concentrates on a relatively small cone, a risk calculation of a region in the complement of the support of the limit measure will yield an answer of [zero]{} and HRV has the potential to produce positive estimates of such risks. We distinguish two related cases: 1. *Full asymptotic dependence:* the MRV limit measure concentrates in standard scale on a single diagonal ray. 2. [*Strong asymptotic dependence:* the MRV]{} limit measure in standard scale concentrates on a relatively small cone about the diagonal. This case is illustrated by analyzing out- and in-degree for Facebook wall posts and returns of Chevron vs Exxon. The variables in these examples are highly dependent, but they are not fully asymptotically dependent. In our experience, it is much easier to find examples of strong asymptotic dependence compared with full asymptotic dependence. We [review and adapt]{} general model generation and detection techniques based on the generalized polar coordinate transform; see [@resnickbook:2007 p. 198], [@das:mitra:resnick:2013; @das:resnick:2015; @lindskog:resnick:roy:2014]). The model generation [methods]{} produce tractable models and the methods are illustrated in Sections \[ex:sim1\] and \[ex:sim2\]. The detection methods show when regularly varying models are consistent with data. We apply the detection methods to the data examples of strong asymptotic dependence. The mathematical framework for the study of multivariate heavy tails is regular variation of measures. The theory is flexible when given for closed subcones of metric spaces [@lindskog:resnick:roy:2014]; we specialize to subcones of $\mathbb{R}_+^2$ and $\mathbb{R}^{2}$ where statistical results are most readily exhibited. Statistical extensions to higher dimensions are possible [and require more sophisticated graphics.]{} We list needed notation in Section \[subsec:notation\] for reference. The definitions of multivariate regular variation (MRV) and hidden regular variation (HRV) are reviewed in Section \[subsec:regVarMod\] where general concepts are adapted for subcones in two dimensions. Sections \[subsec:polar\] and \[subsec:MRVHRV\] give equivalent formulations in polar co-ordinates and discuss the particular cases of strong [and full]{} dependence. Section \[subsec:Hillish\] gives techniques for detecting when data is consistent with a model exhibiting MRV and HRV. These techniques rely on the fact that under broad conditions, if a vector $\bX$ has a multivariate regularly varying distribution on a cone $\mathbb{C}$, then under a [*generalized polar coordinate transformation*]{} (see ), the transformed vector satisfies a conditional extreme value (CEV) model for which detection techniques exist from [@das:resnick:2011b]. [This methodology adds to the toolbox of ]{} one dimensional techniques such as checking if one dimensional marginal distributions are heavy tailed or checking whether one dimensional functions of the data vector such as the maximum and the minimum component are heavy tailed. See [@resnickbook:2007 p. 326], [@resnick:2002a]. In Section \[sec:dataAnal\], we analyze real and simulated data and show that our estimation and detection techniques [produce results consistent with presence of both MRV with strong dependence and HRV.]{} Section \[sec:conc\] presents concluding comments. [Background on Regular Variation of Measures]{} {#sec:basics} =============================================== We provide a brief review of the mathematical setup for multivariate regularly varying measures with the notion of ${\mathbb{M}}$-convergence. More detail is found in [@hult:lindskog:2006a; @hult:lindskog:2006b; @das:mitra:resnick:2013; @das:resnick:2015; @lindskog:resnick:roy:2014]. The notions of hidden regular variation (HRV) and regular variation expressed by polar coordinate transforms are discussed in Sections \[subsec:regVarMod\] and \[subsec:polar\] with emphasis on cases of the *strong asymptotic dependence*. Finally in Section \[subsec:Hillish\] we discuss detection of HRV using the Hillish estimator. Basic notation. {#subsec:notation} --------------- A summary of some notation and concepts are provided here. [For this paper, we restrict to dimension $d=2$ unless otherwise specified.]{} We use bold letters to denote vectors, with capital letters for random vectors and small letters for non-random vectors, e.g., $\by=(y_1,y_2)\in {\mathbb{R}}^2$. We also define $\bzero=(0,0)$ and $\binfty=(\infty,\infty)$. Vector operations are always understood component-wise, e.g., for vectors $\bx$ and $\by$, $\bx{\leqslant}\by$ means $x_i{\leqslant}y_i$ for $i=1, 2$. Some additional notation follows with explanations that are amplified in subsequent sections. Detailed discussions are in the references. $$\begin{array}{llll} \RV_\beta & \text{Regularly varying functions with index $\beta>0$; that is, functions $f:\mathbb{R}_+\mapsto \mathbb{R}_+$}\\ {}& \text{satisfying $\lim_{t\to\infty}f(tx)/f(t)=x^\beta,$ for $x>0.$ We can and do assume such functions }\\ &\text{are continuous and strictly increasing. See \cite{bingham:goldie:teugels:1989, resnickbook:2008, dehaan:ferreira:2006}.}\\[2mm] {\mathbb{E}}& \mathbb{R}_+^2 \setminus \{\bzero\} \text{ or }\mathbb{R}^2 \setminus \{\bzero\}.\\[2mm] [\diag] & \{(x,x):x{\geqslant}0\}.\\[2mm] [\smallwedge] & {\{\bx\in{\mathbb{R}}_+^2: a_lx_1 {\leqslant}x_2 {\leqslant}a_u x_1\} }\;\;\text{for some $0<a_l<a_u<\infty$.}\\[2mm] \nu_\alpha (\cdot) & \text{The Pareto measure on $(0,\infty)$ given by $\nu_\alpha(x,\infty)=x^{-\alpha},x>0$.}\\[2mm] \MRV & \text{Multivariate regular variation; for this paper, it means regular variation on ${\mathbb{E}}$}.\\[2mm] \HRV & \text{Hidden regular variation; for this paper, it means a second regular variation after}\\ {}& \text{removal of a cone as well as $\bzero$.}\\[2mm] {\mathbb{M}}(\mathbb{C}\setminus \mathbb{C}_0) & \text{The set of all non-zero measures on $\mathbb{C}\setminus \mathbb{C}_0$ which are finite on subsets bounded}\\ &\text{away from the \emph{forbidden zone} $\mathbb{C}_0$, {a closed cone removed from the state space.}}\\[2mm] \mathcal{C}(\mathbb{C}\setminus \mathbb{C}_0) & \text{Continuous, bounded, positive functions on $\mathbb{C}\setminus \mathbb{C}_0$ whose supports are bounded}\\&\text{away from the \emph{forbidden zone} $\mathbb{C}_0$. Without loss of generality (\cite{lindskog:resnick:roy:2014}), we may assume}\\&\text{the functions are uniformly continuous.}\\[2mm] \mu_n\to \mu & \text{Convergence in ${\mathbb{M}}(\mathbb{C}\setminus \mathbb{C}_0)$ means $\mu_n (f) \to \mu(f)$ for all } f \in \mathcal{C}(\mathbb{C}\setminus \mathbb{C}_0). \text{ See \cite{hult:lindskog:2006a, das:mitra:resnick:2013, lindskog:resnick:roy:2014}}\\ {}& \text{and Definition \ref{dfn:mconv}.}\\[2mm] d(\bx,\by) & \text{Metric in ${\mathbb{R}}^2$, usually the $L_2$ distance $d(\bx,\by)=\bigl((x_1-y_1)^2+(x_2-y_2)^2\bigr)^{1/2}.$}\\[2mm] \text{diamond plot} & \text{Mapping of thresholded data onto the $L_1$ unit sphere } \bx \mapsto \bigl(\frac{x_1}{|x_1|+|x_2|},\frac{x_2}{|x_1|+|x_2|}\bigr) \\[2mm] d(\bx,\mathbb{C}) & \inf\limits_{\by\in\mathbb{C}} d(\bx,\by) \text{ for } \bx \in {\mathbb{E}}\text{ and } \bC\subset {\mathbb{E}}.\\[2.5mm] \aleph_{\mathbb{C}} & \{\bx : d(\bx,\mathbb{C} )=1\}. \text{ For instance: } \aleph_{\bzero} = \{\bx \in {\mathbb{E}}: d(\bx, \{\bzero \})=1\}, \\&{} \aleph_{[\diag]} = \{\bx \in {\mathbb{E}}: d(\bx,[\diag])=1\} \text{ and } \aleph_{[\smallwedge] } = \{\bx \in {\mathbb{E}}: d(\bx,[\smallwedge])=1\}. \\[2mm] \end{array}$$ $$\begin{array}{llll} \gpolar & \text{{Generalized} polar co-ordinate transformation relative to the deleted forbidden}\\ &\text{zone $\mathbb{C}_0$.} \quad\gpolar(\bx)=\left(d(\bx, \mathbb{C}_0), \bx/d(\bx,\mathbb{C}_0)\right). \text{ See \cite{lindskog:resnick:roy:2014, das:mitra:resnick:2013}.}\\[2mm] \bX {\protect\mathpalette{\protect\independenT}{\perp}}\bY & \text{The random elements $\bX$ and $\bY$ are independent.} \end{array}$$ Regularly varying distributions on cones. {#subsec:regVarMod} ----------------------------------------- We review material from [@hult:lindskog:2006a; @das:mitra:resnick:2013; @lindskog:resnick:roy:2014] describing MRV and HRV specialized to two dimensions. The convergence concept used for defining regular variation is ${\mathbb{M}}$-convergence which is slightly different from vague convergence traditionally used. Reasons for preferring ${\mathbb{M}}$-convergence are discussed in [@lindskog:resnick:roy:2014; @das:mitra:resnick:2013]. ### Forbidden zones. {#subsub:forbidden} Consider $\mathbb{R}_+^2 $ [or $\mathbb{R}^2 $]{} as a metric space with Euclidean metric $d(\bx,\by)$. A subset $\mathbb{C}$ is a [*cone*]{} if it is closed under positive scalar multiplication: if $\bx \in \mathbb{C}$ then $c\,\bx \in \mathbb{C}$ for $c>0$. A framework for discussing regularly varying measures is ${\mathbb{M}}$-convergence ([@lindskog:resnick:roy:2014; @das:mitra:resnick:2013]) on a closed cone $\mathbb{C}\subset \mathbb{R}_+^2 $ [or ${\mathbb{R}}^{2}$]{} with a closed cone $\mathbb{C}_0 \subset \mathbb{C}$ deleted. Call the deleted cone $\mathbb{C}_0 $ the [*forbidden zone*]{}. [Here are some cases of interest for this paper; see Figure \[fig:diag\_wedge\].]{} 1. Suppose $\mathbb{C} = \mathbb{R}_+^2$ and $\mathbb{C}_0 =\{\bzero\}$. Then ${\mathbb{E}}:=\mathbb{C} \setminus \mathbb{C}_0 = \mathbb{R}_+^2 \setminus \{\bzero\}$ is the space for defining ${\mathbb{M}}$-convergence appropriate for regular variation of distributions of positive random vectors. The forbidden zone is the origin $\{\bzero\}.$ 2. Suppose $\mathbb{C} = \mathbb{R}^2$ and $\mathbb{C}_0 =\{\bzero\}$. Then ${\mathbb{E}}:=\mathbb{C} \setminus \mathbb{C}_0 = \mathbb{R}^2 \setminus \{\bzero\}$ is the space appropriate for regular variation of distributions of pairs of real valued random variables such as those representing financial returns. The forbidden zone is still the origin $\{\bzero\}.$ 3. Suppose $\mathbb{C} = \mathbb{R}_+^2$ and $\mathbb{C}_0 =\{(x,x):x{\geqslant}0\}=:[\diag]$. Then $\mathbb{C} \setminus \mathbb{C}_0= {\mathbb{R}}_+^2 \setminus [\diag]$, the first quadrant without its diagonal, is the right space for defining ${\mathbb{M}}$-convergence appropriate for HRV when asymptotic full dependence is present. The forbidden zone is the diagonal; see Figure \[fig:diag\_wedge\]. 4. A related example is $\mathbb{C}={\mathbb{R}}^{2}$ and $\mathbb{C}_0 =\{(x,x):x\in {\mathbb{R}}\}$ and we seek regular variation on $\mathbb{C} \setminus \mathbb{C}_0= {\mathbb{R}}^{2} \setminus \{(x,x):x\in {\mathbb{R}}\}$. 5. Suppose $\mathbb{C} = \mathbb{R}_+^2$ and for $0<\theta_l<\theta_u<1$, and $$\label{e:parCone} 0<a_l=\theta_u^{-1}-1 <a_u=\theta_l^{-1}-1 <\infty,$$ $$\begin{aligned} \mathbb{C}_0 =&\{\bx\in {\mathbb{R}}_+^2: 0<\theta_l {\leqslant}\frac{x_1}{x_1+x_2} {\leqslant}\theta_u<1\} \label{eq:polar} \\ =& \{ \bx \in {\mathbb{R}}_+^2: a_lx_1 {\leqslant}x_2 {\leqslant}a_u x_1\}=:[\smallwedge],\label{eq:peewee}\end{aligned}$$ where $[\smallwedge]$ is a pizza slice removed from the first quadrant. We then seek hidden regular variation on ${\mathbb{R}}_+^2\setminus [\smallwedge] $; see Figure \[fig:diag\_wedge\]. When $a_l=a_u=1$ (or equivalently $\theta_l=\theta_u=1/2$), then $[\smallwedge]$ reduces to $[\diag]$. Note we [may]{} parameterize $[\smallwedge ]$ in two ways, [one using the slopes $a_l,a_u$ and one using angles $\theta_l,\theta_u$]{}. In we use the traditional $L_1$ polar coordinate transform $\polar $ from ${\mathbb{R}}^2\setminus \{\bzero\}\mapsto (0,\infty)\times [-1,1]$ given by $$\polar: \bx \mapsto \Bigl(|x_1|+|x_2|, \frac{x_1 }{|x_1|+|x_2 |}\Bigr)=(r,\theta).$$ to express $[\smallwedge]$ in polar coordinates as ${\mathbb{R}}_{+}\times[\theta_l,\theta_u]$. In we use the slopes $a_l,a_u$ of the boundary lines of $[\smallwedge]$. In this paper we give particular attention to $[\smallwedge ]$ because - when the limit measure of regular variation concentrates on $[\smallwedge ]\subsetneq {\mathbb{R}}_+^2$ we have a tractable notion of [*strong asymptotic dependence*]{}; and - data examples of strong asymptotic dependence seem to be far more common than for the case of full asymptotic dependence. Of course, other types of forbidden zones are possible and to date most attention has been directed to removing axes when asymptotic independence is present. ### Regular variation of measures. Let ${\mathbb{M}}(\mathbb{C} \setminus \mathbb{C}_0)$ be the set of Borel measures on $\mathbb{C} \setminus \mathbb{C}_0$ which are finite on sets bounded away from the forbidden zone $\mathbb{C}_0$ ([@das:mitra:resnick:2013; @hult:lindskog:2006a; @lindskog:resnick:roy:2014]). We think of sets bounded away from the forbidden zone $\mathbb{C}_0$ as [*tail regions*]{}. ${\mathbb{M}}$-convergence is the basis for the definition of multivariate regular variation: \[dfn:mconv\] For $\mu_n, \mu \in {\mathbb{M}}(\mathbb{C} \setminus \mathbb{C}_0)$ we say $\mu_n \to \mu$ in ${\mathbb{M}}(\mathbb{C} \setminus \mathbb{C}_0)$ if $\int f\mathrm d \mu_n \to \int f \mathrm d \mu$ for all $f\in \mathcal{C}( \mathbb{C} \setminus \mathbb{C}_0)$. \[dfn:regvarwithmconv\] A random vector $\bZ{\geqslant}\bzero$ is regularly varying on $\mathbb{C} \setminus \mathbb{C}_0$ with index $\alpha>0$ if there exists $b(t) \in \RV_{1/\alpha}$, called the [*scaling function*]{}, and a measure $\nu(\cdot) \in {\mathbb{M}}(\mathbb{C} \setminus \mathbb{C}_0)$, called the [*limit or tail measure*]{}, [such that]{} as $t \to\infty$, $$\label{eq:RegVarMeas} t\,{\mathbb{P}}[ \bZ/b(t) \in \cdot \,] \to \nu(\cdot), \quad \text{ in } {\mathbb{M}}(\mathbb{C} \setminus \mathbb{C}_0).$$ We write $\bZ \in \MRV (\alpha, b(t), \nu, \mathbb{C} \setminus \mathbb{C}_0)$ to emphasize that regular variation depends on an index $\alpha$, scaling function $b \in \RV_{1/\alpha}$, limit measure $\nu$, and state space $ \mathbb{C} \setminus \mathbb{C}_0.$ Since $b(t) \in \RV_{1/\alpha}$, the limit measure $\nu(\cdot) $ has a scaling property, $$\label{eq:limMeasScales} \nu(c \,\cdot) =c^{-\alpha} \nu(\cdot),\qquad c>0.$$ Suppose $\mathbb{C}=\mathbb{R}_+^2$, $\mathbb{C}_0=\{\bzero\}$. We distinguish between different forms of dependence and identify them as follows: 1. If $\nu(\cdot)$ satisfies $\nu((0,\infty)^2)=0$ so that $\nu$ concentrates on the axes, then $\bZ$ possesses [*asymptotic independence*]{}; see [@resnickbook:2007; @dehaan:ferreira:2006; @resnickbook:2008]. 2. \[item:2\] If $\nu(\cdot)$ concentrates on $[\diag]$ then $\bZ$ has [*full asymptotic dependence*]{}. 3. \[item:3\] If $\nu(\cdot)$ concentrates on a narrow wedge as in , then $\bZ$ has [*strong asymptotic dependence*]{}. An analogous classification can be made for the case $\mathbb{C}=\mathbb{R}^2.$ [**Diamond plot:**]{} When [ doing empirical analyses for cases \[item:2\] or \[item:3\], it is convenient and informative to map points $$\bx \mapsto \Bigl(\frac{x_1}{|x_1|+|x_2},\frac{x_2}{|x_1|+|x_2|}\Bigr) =\btheta =(\theta_1,\theta_2)$$ onto the $L_1$ unit sphere or diamond, perhaps after thresholding data according to [the]{} $L_1$ norm. We call the resulting plot the [*diamond plot*]{}. Observing how points cluster on the $L_1$ unit sphere provides a visualization of dependence.]{} Regular variation and the polar coordinate transformation. {#subsec:polar} ---------------------------------------------------------- When the forbidden zone is the origin, it is useful to rephrase regular variation of measures using the polar coordinate transformation. Theoretically, we may choose any norm $\|\cdot\|$ and the polar coordinate transform maps $\bx$ into the unit sphere determined by the chosen norm: $\bx \mapsto \bigl(\|\bx\|, {\bx}/{\|\bx\|}\bigr)$. The limit measure expressed in polar coordinates is a product measure and this provides a way to construct regularly varying measures and is useful for inference. (See [@resnickbook:2007 p. 168 ff, 173 ff].) When the forbidden zone is a more general cone than just the origin, the polar coordinate transform no longer brings benefits and the limit angular measure expressed in these coordinates may be infinite. To get a limit measure expressed as a product [where the analogue of the angular measure is a probability measure,]{} one may transform and using [*generalized*]{} polar coordinates ([@lindskog:resnick:roy:2014; @das:mitra:resnick:2013]). We can define the generalized polar coordinate transform for general cones of the form $\mathbb{C}\setminus\mathbb{C}_0$ and an [associated metric $d(\cdot,\cdot)$ satisfying $d(c\bx,c\by)=cd(\bx,\by)$ for scalars $c>0$. The metric $d(\cdot, \cdot)$ that we use in practice is the usual $L_2$ Euclidean metric [but note that the $L_1$ norm is used for]{} visualizations using the diamond plot.]{} ### Generalized polar coordinates. Define $\gpolar: \mathbb{C}\setminus \mathbb{C}_0 \mapsto (0,\infty)\times \aleph_{\mathbb{C}_0}$ by $$\label{eq:defgpolar} \gpolar (\bx) =\left(d(\bx,\mathbb{C}_0) , \frac{\bx} {d(\bx,\mathbb{C}_0)}\right).$$ Consequently, the inverse $\gpolar^{\leftarrow}:(0,\infty)\times \aleph_{{\bC}_0}\mapsto \bC\setminus\bC_{0}$ of the $\gpolar$ function is $$\label{eq:defgpolarinv} \gpolar^{\leftarrow} (r, \theta) =r\theta.$$ The transformation $\gpolar$ depends on the forbidden zone $\bC_0$ and [the choice of metric]{}. In practice, the metric $d(\cdot,\cdot)$ is taken to be the usual $L_2$ Euclidean distance. This is practical and customary but not obligatory. ### Generalized unit sphere. [When transforming from Cartesian to polar coordinates, a central role is played by the unit sphere $\aleph_0:=\{\bx \neq \bzero:\|\bx\|=1\}.$ The comparable set when using generalized polar coordinates with respect to the forbidden zone $\mathbb{C}_0$ is ]{} $\aleph_{\mathbb{C}_0}=\{\bx \in \mathbb{C} \setminus \mathbb{C}_0: d(\bx, \mathbb{C}_0)=1\},$ the locus of points at distance 1 from the deleted forbidden zone $\mathbb{C}_0$. We then have an equivalent form of and , namely, $$\label{eq:limMeasPolar} t{\mathbb{P}}\Bigl[\gpolar \left(\frac{\bZ}{b(t)}\right) \in \;\cdot \;\Bigr] = t{\mathbb{P}}\Bigl[ \Bigl( \frac{d(\bZ, \bC_{0})}{b(t)}, \frac{\bZ}{d(\bZ,\bC_0)}\Bigr) \in \cdot \,\Bigr] \to (\nu_{\alpha} \times S_0)(\cdot) \, = (\nu\circ\gpolar^{\leftarrow})(\cdot)\,,$$ in ${\mathbb{M}}\left((0,\infty) \times \aleph_{\mathbb{C}_0} \right)$ where $\nu_{\alpha} (x, \infty) =x^{-\alpha}, \,x>0,\,\alpha>0$ and $S_0(\cdot) $ is a probability measure on $\aleph_{\mathbb{C}_0} $ ([@das:mitra:resnick:2013; @lindskog:resnick:roy:2014]), [provided $b(t)$ is appropriately chosen]{}. [Note that $\aleph_{\mathbb{C}_0}$ depends on the choice of $d(\cdot,\cdot)$, and the limit in is a product measure.]{} ### Examples of unit spheres. Figure \[fig:diag\_wedge\] shows different shapes of $\aleph_{{\bC_{0}}}$ [for $L_2$ distance]{} and different choices of $\bC_{{0}}$ where $\bC= {\mathbb{R}}_{+}^{2}$. ![Left: ${\mathbb{R}}_{+}^2 \setminus \{\bzero\}$ and distance is $L_2$. Center: ${\mathbb{R}}_{+}^2 \setminus [{\diag}]$ and distance is $L_2$. Right: ${\mathbb{R}}_{+}^2 \setminus [\smallwedge]$ and distance is $L_2$. The dotted lines represent $\aleph_{\mathbb{C}_0}$ which is the locus of points at distance one from $\bC_0$.[]{data-label="fig:diag_wedge"}](Cones){width="6in"} (i) For $\mathbb{R}_{{+}}^2 \setminus \{\bzero\}$, where the forbidden zone is $\{\bzero\}$, we have $\aleph_{\bzero}=\{\bx \in {\mathbb{R}}_{{+}}^2: d(\bx, \{\bzero\})=1\}$. (ii) If we delete the forbidden zone $\mathbb{C}_0=[\diag]=\{(x,x):x \in \mathbb{R}_{{+}}\}$ from ${\mathbb{R}}_{{+}}^2$, the appropriate unit sphere with respect to $L_2$ distance is $$\label{eq:alephdiag} \aleph_{[\diag]}:=\aleph_{>[\diag]} \cup \aleph_{<[\diag]} =\{(u,u+\sqrt 2):{u {\geqslant}0} \} \bigcup \{(u,u-\sqrt 2):{u {\geqslant}\sqrt 2}\} ,$$ the lines of slope 1, above and below the diagonal, which are at distance 1 from the diagonal. For $ \bx \notin [\diag]$, $$\label{eq:dist(diag)} d\bigl(\bx , [\diag]\bigr)=|x_1-x_2|/\sqrt{2}.$$ (iii) When the forbidden zone is $[\smallwedge]\subset {\mathbb{R}}_+^2$, we have $$\begin{aligned} \aleph_{[\smallwedge]}=&\aleph_{>[\smallwedge]}\cup \aleph_{<[\smallwedge]}\nonumber \\ =&\{(u,a_uu+\sqrt{1+a_u^2}):u{\geqslant}0\} \bigcup \{(u,a_lu-\sqrt{1+a_l^2}):u{\geqslant}a_l^{-1}\sqrt{1+a_l^2}\},\label{eq:alephsmallwedge} \end{aligned}$$ which are the lines parallel to the two rays defining $[\smallwedge]$ [at a distance of 1 from $[\smallwedge]$]{}. When $a_l=a_u=1$, $\aleph_{[\smallwedge]}$ reduces to $\aleph_{[\diag]}$. For the distance to the forbidden zone from a point $\bx \notin [\smallwedge]$, we have $$\begin{aligned} d\bigl(\bx, [\smallwedge]\bigr)=& \frac{|x_2-a_ux_1|}{\sqrt{1+a_u^2}}, && \text{ if } x_2>a_ux_1,\label{eq:dist(>wedge)}\\ d\bigl(\bx, [\smallwedge]\bigr)=& \frac{|x_2-a_lx_1|}{\sqrt{1+a_l^2}}, && \text{ if }x_2<a_lx_1,\label{eq:dist(<wedge)}\end{aligned}$$ which reduces to if $a_l=a_u=1$. Obvious changes apply when the state space is ${\mathbb{R}}^2$. MRV and HRV under strong asymptotic dependence. {#subsec:MRVHRV} =============================================== Definitions. {#subsub:gen} ------------ Consider simultaneous existence of regular variation on both the big cone ${\mathbb{R}}_+^2\setminus \{\bzero\}$ and a smaller cone ${\mathbb{R}}_+^2\setminus \mathbb{C}_0$, where $\mathbb{C}_0$ is either $[\diag]$ or $[\smallwedge]$. We provide equivalent polar-coordinate conditions for this simultaneous existence. \[dfn:MRVHRV\] The vector $ \bZ$ is regularly varying on $ {\mathbb{R}}_+^2\setminus \{\bzero\}$ and has [*hidden regular variation*]{} on ${\mathbb{R}}_+^2\setminus \mathbb{C}_0$ if there exist $0<\alpha {\leqslant}\alpha_0$, scaling functions $b(t) \in \RV_{1/\alpha}$ and $b_0(t) \in \RV_{1/\alpha_0}$ with $b(t)/b_0(t) \to \infty$ and limit measures $\nu,\, \nu_0$ such that $$\label{eq:mrv+hrv} \bZ\in \MRV(\alpha, b(t), \nu, {\mathbb{R}}_+^2\setminus \{\bzero\}) \cap \MRV(\alpha_0, b_0(t), \nu_0, {\mathbb{R}}_+^2\setminus \mathbb{C}_0).$$ Unpacking the notation we obtain the two regular variation limits $$\begin{aligned} t{\mathbb{P}}[\bZ / b(t) \in \cdot \,]\to & \,\nu (\cdot) &&\quad \text{ in }{\mathbb{M}}({\mathbb{R}}_+^2\setminus \{\bzero\}), \label{eq:regVarE}\\ t{\mathbb{P}}[\bZ / b_0(t) \in \cdot \,]\to & \,\nu_0 (\cdot) &&\quad \text{ in } {\mathbb{M}}({\mathbb{R}}_+^2\setminus \mathbb{C}_0).\label{eq:regVarE0} \intertext{ Using polar coordinates, \eqref{eq:regVarE} can be written as} t{\mathbb{P}}\left[\left(\|\bZ \|/ b(t), \bZ/\|\bZ\|\right) \in \cdot \,\right]\to & \, \nu_\alpha \times S (\cdot) &&\quad \text{ in } {\mathbb{M}}((0,\infty)\times \aleph_{\bzero }), \label{eq:regVarEPolar}\end{aligned}$$ where $S$ is a probability measure on $\aleph_{\bzero}$. [Similarly when removing $\mathbb{C}_0$ from the state space, generalized polar coordinates allow re-writing]{} as $$\label{e:delC0} t{\mathbb{P}}\Bigl[ \Bigl( \frac{d(\bZ, \bC_{0})}{b_0(t)}, \frac{\bZ}{d(\bZ,\bC_0)}\Bigr) \in \cdot \,\Bigr] \to (\nu_{\alpha_{0}} \times S_0)(\cdot)$$ in ${\mathbb{M}}\left((0,\infty) \times \aleph_{\mathbb{C}_0} \right)$ where $\nu_{\alpha_{0}} (x, \infty) =x^{-\alpha_{0}}, \,x>0,\,\alpha_{0}>0$ and $S_0(\cdot) $ is a probability measure on $\aleph_{\mathbb{C}_0}$. Regular variation when deleting $[\smallwedge]$. ------------------------------------------------ [Focus on the special case where the forbidden zone is $[\smallwedge]$. Since $[\diag ]$ is a particular case of $[\smallwedge]$, we do not treat $[\diag]$ separately. Recall the notation in and the two parameterizations of $[\smallwedge ]$ given in , and . [The]{} distance of points to $[\smallwedge]$ is given in and .]{} When $\mathbb{C}_0=[\smallwedge]$, becomes two statements. With $x>0$ and $\Lambda \subset \aleph_{[\smallwedge]}$ we have $$\begin{aligned} t{\mathbb{P}}\left[ \frac{Z_2-a_uZ_1}{{b_{0}(t)}\sqrt{1+a_u^2}}>x, \frac{\sqrt{1+a_u^2}\bZ}{Z_2-a_uZ_1} \in \Lambda \right] \to & x^{-\alpha_0} S_0(\Lambda) &&\quad \text{ in } {\mathbb{M}}\left((0,\infty)\times \aleph_{>[\smallwedge]}\right), \label{eq:wedge1st}\\ t{\mathbb{P}}\left[ \frac{a_lZ_1-Z_2}{{b_{0}(t)}\sqrt{1+a_l^2}}>x, \frac{\sqrt{1+a_l^2}\bZ}{a_lZ_1-Z_2} \in \Lambda \right] \to & x^{-\alpha_0} S_0(\Lambda) &&\quad \text{ in } {\mathbb{M}}\left((0,\infty)\times \aleph_{<[\smallwedge]}\right) \label{eq:wedge2nd}.\end{aligned}$$ Modifying variables leads to the two simpler statements. For $x>0$, $$\begin{aligned} t{\mathbb{P}}\left[ \frac{Z_2-a_uZ_1}{b_0(t)} >x, \frac{Z_2}{Z_1} {\leqslant}y\right] \to & (1+a_u^{2})^{\frac{\alpha_0}{2}} x^{-\alpha_0} S_0 \Biggl\{\Bigl(v,a_uv+\sqrt{1+a_u^2}\Bigr): v{\geqslant}\frac{\sqrt{1+a_u^2}}{y-a_u}\Biggr\},\; y>a_u, \label{eq:>smallwedge}\\ t{\mathbb{P}}\left[ \frac{a_lZ_1-Z_2}{b_0(t)} >x, \frac{Z_1}{Z_2} {\leqslant}y\right] \to & (1+a_l^{2})^{\frac{\alpha_0}{2}} x^{-\alpha_0} S_0\Bigl\{ \bigl( v+\sqrt{1+a_l^2}/a_l,v\bigr): v{\geqslant}\frac{\sqrt{1+a_l^2}}{a_ly-1}\Bigr\},\; y>\frac{1}{a_l}. \label{eq:<smallwedge}\end{aligned}$$ \[rem:genius\] Thus a necessary condition for non-trivial regular variation on ${\mathbb{R}}_+^2\setminus [\smallwedge]$ is that both $(Z_2-a_uZ_1)_+$ and $(a_lZ_1-Z_2)_+$ be regularly varying with index $\alpha_0 {\geqslant}\alpha$. This fact suggests the exploratory diagnostic of testing whether these 1 dimensional variables have power laws with the same index. If so, one can continue to explore with the Hillish statistic and associated plot; see Section \[subsec:Hillish\]. However, there is nothing to prevent the possibility that regular variation exists on the region above $[\smallwedge]$ but that tails are of lower order below $[\smallwedge]$. This would happen for instance if $S_{0}(\aleph_{<[\smallwedge]}) =0$ but $S_{0}(\aleph_{>[\smallwedge]})>0$. If this happened, one could search for another higher index or thinner tailed regular variation on $<[\smallwedge].$ \[rem:altgenius\] [[Analogous statements to the limits , hold true when $\bC={\mathbb{R}}^{2}$ and $[\diag]$ and $[\smallwedge]$ are their appropriate equivalents in ${\mathbb{R}}^{2}$ to get.]{} ]{} ### Restrictions on the choice of $[\smallwedge]$. In this paper, we assume that $1< a_{l}{\leqslant}1{\leqslant}a_{u}<\infty$ where $[\smallwedge] = \{\bx\in {\mathbb{R}}_{+}^{2}: a_{l}x_{1}{\leqslant}x_{2}{\leqslant}a_{l}x_{1}\}$. This choice is partly governed by the fact that it is easier for us to deal with data portraying tail equivalence with $ \lim_{t\to\infty} {{\mathbb{P}}(Z_{1}>t)}/{{\mathbb{P}}(Z_{2}>t)}=1.$ Now, when $a_{l}=a_{u}$, we get $[\smallwedge]=[\diag]$, which means under a model of full asymptotic dependence the only limit measure that we allow is restricted to $[\diag]$. If we assume $ \bZ\in \MRV(\alpha, b(t), \nu, {\mathbb{R}}_+^2\setminus \{\bzero\}) $ and $\nu$ is supported on $[\diag]$ then using , this clearly implies, that $$\begin{aligned} \lim_{t\to\infty} \frac{{\mathbb{P}}(Z_{1}>t)}{{\mathbb{P}}(Z_{2}>t)} & = \lim_{t\to\infty} \frac{{\mathbb{P}}(Z_{1}>b(t))}{{\mathbb{P}}(Z_{2}>b(t))}\\ & = \lim_{t\to\infty} \frac{t\,{\mathbb{P}}\left(\bZ/b(t) \in (1,\infty)\times[0,\infty)\right)}{t\,{\mathbb{P}}\left(\bZ/b(t) \in [0,\infty)\times(1,\infty)\right)}\\ & = \frac{\nu\left( (1,\infty)\times[0,\infty)\right)}{\nu\left( [0,\infty)\times(1,\infty)\right)}\\ & = \frac{\nu\left( (x_{1},x_{2}): x_{1}>1, (x_{1},x_{2})\in [\diag]\right)}{\nu\left( (x_{1},x_{2}): x_{2}>1, (x_{1},x_{2})\in [\diag]\right)} =1, \end{aligned}$$ where the last line is a consequence of $\nu$ being concentrated on $[\diag]$. Thus, not only are $Z_{1}, Z_{2}$ tail equivalent, but in fact $ \lim_{t\to\infty} {{\mathbb{P}}(Z_{1}>t)}/{{\mathbb{P}}(Z_{2}>t)}=1.$ The following lemma shows that that we cannot choose a $[\smallwedge]$ which does not contain $[\diag]$ if we want to guarantee $ \lim_{t\to\infty} {{\mathbb{P}}(Z_{1}>t)}/{{\mathbb{P}}(Z_{2}>t)}=1.$ If $\bZ\in \MRV(\alpha, b(t), \nu, {\mathbb{R}}_+^2\setminus \{\bzero\})$ where $\nu$ is supported on $[\smallwedge]=\{\bx\in {\mathbb{R}}_{+}^{2}: a_{l}x_{1}{\leqslant}x_{2}{\leqslant}a_{l}x_{1}\}$ and $ \lim_{t\to\infty} {{\mathbb{P}}(Z_{1}>t)}/{{\mathbb{P}}(Z_{2}>t)}=1$ then $0<a_{l} {\leqslant}1{\leqslant}a_{u}<\infty$. [To get a contradiction,]{} suppose we have $1<a_{l}{\leqslant}a_{u}<\infty.$ [(A similar contradiction is obtained if we assume $0<a_{l}{\leqslant}a_{u}<1.$) ]{} Assume, $$\begin{aligned} 1= \ \lim_{t\to\infty} \frac{{\mathbb{P}}(Z_{1}>t)}{{\mathbb{P}}(Z_{2}>t)} & = \lim_{t\to\infty} \frac{t\,{\mathbb{P}}\left(\bZ/b(t) \in (1,\infty)\times[0,\infty)\right)}{t\,{\mathbb{P}}\left(\bZ/b(t) \in [0,\infty)\times(1,\infty)\right)} = \frac{\nu\left( (1,\infty)\times[0,\infty)\right)}{\nu\left( [0,\infty)\times(1,\infty)\right)}.\end{aligned}$$ Hence $\nu\left( (1,\infty)\times[0,\infty)\right)=\nu\left( [0,\infty)\times(1,\infty)\right)$. We know that $\nu$ is supported on $[\smallwedge]$. So, $$\begin{aligned} \nu\left( (1,\infty)\times[0,\infty)\right) & = \nu ((x_{1},x_{2})\in {\mathbb{R}}_{+}^{2}: x_{1}>1, (x_{1},x_{2}) \in [\smallwedge] )&&{}\\ & = \nu \left((x_{1},x_{2})\in {\mathbb{R}}_{+}^{2}: x_{1}>1, a_{l}{\leqslant}\frac{x_{2}}{x_{1}}{\leqslant}a_{u} \right) &&\\ & < \nu \left((x_{1},x_{2})\in {\mathbb{R}}_{+}^{2}: x_{1}>{\frac {1}{a_{l}}}, a_{l}{\leqslant}\frac{x_{2}}{x_{1}}{\leqslant}a_{u} \right)&&(\text{since $a_l>1$}) \\ & < \nu \left((x_{1},x_{2})\in {\mathbb{R}}_{+}^{2}: x_{2}>1, a_{l}{\leqslant}\frac{x_{2}}{x_{1}}{\leqslant}a_{u} \right)&&\text{(since $x_2{\geqslant}a_lx_1$)}\\ & =\nu\left( [0,\infty)\times(1,\infty)\right),&&{}\end{aligned}$$ which is a contradiction. Regular variation when deleting $[\smallwedge]$ expressed in traditional polar coordinates. ------------------------------------------------------------------------------------------- Regular variation on ${\mathbb{R}}_+^2 \setminus [\smallwedge]$ expressed using generalized polar coordinates in , can also be written in terms of the traditional polar coordinates $\bz\mapsto \bigl(r,(\theta,1-\theta)\bigr)$ where $z_1=r\theta$ and $z_2=r(1-\theta)$ and $r=z_1+z_2.$ Using capital letters for random variables, the left most probability in becomes for $x>0,\,y>a_u$, $$\begin{aligned} t{\mathbb{P}}\Bigl[ \frac{Z_2-a_uZ_1}{b_0(t)} >x, &\frac{Z_2}{Z_1} {\leqslant}y \Bigr] = t{\mathbb{P}}\left[ \frac{R(1-\Theta (1+a_u))}{b_0(t)}>x, \frac{(1-\Theta)}{\Theta} {\leqslant}y \right] \\ \intertext{and using \eqref{e:parCone} this is} =&t{\mathbb{P}}\left[ \frac{R(1-\theta_l^{-1}\Theta )}{b_0(t)}>x, \Theta^{-1} {\leqslant}y+1 \right] =t{\mathbb{P}}\left[ \frac{R(1-\theta_l^{-1}\Theta )}{b_0(t)}>x, \Theta{\geqslant}\frac{1}{1+y}\right]\end{aligned}$$ Set $s=1/(1+y)$ where $y>a_u$ and thus $s<\theta_l$, $$t{\mathbb{P}}\left[ \frac{R(1-\theta_l^{-1}\Theta )}{b_0(t)}>x, \Theta>s\right] \to (1+a_u^{2})^{\frac{\alpha_0}{2}} x^{-\alpha_0} S_0 \Biggl\{\Bigl(v,a_uv+\sqrt{1+a_u^2}\Bigr): v{\geqslant}\frac{\sqrt{1+a_u^2}}{s^{-1}-\theta_l^{-1}}\Biggr\}.$$ An analogous expression holds for . So if regular variation with index $\alpha$ holds on ${\mathbb{R}}_+^2\setminus \{\bzero\}$, $R$ is a random variable with regularly varying distribution tail with index $\alpha$ and multiplying $R$ by $1-\theta_l^{-1}\Theta$ produces a variable with a lighter tail having index $\alpha_0$. The forbidden zone of HRV and the limit measure on [${\mathbb{E}}$]{}. ---------------------------------------------------------------------- Suppose there are two regular variation properties that hold for a vector $\bZ{\geqslant}0$ so that holds with $b(t)/b_0(t)\to\infty$. If $\mathbb{C}_0=[\smallwedge]$ with [$0< a_l {\leqslant}1 {\leqslant}a_u <\infty$]{}; then $\nu$, the limit measure on ${\mathbb{R}}_+^2\setminus \{\bzero\}$, must concentrate on the forbidden zone $[\smallwedge]$ used to define the second regular variation. (Cf. [@resnickbook:2007 p. 324-5].) This means that when detecting MRV, if the limit measure is typical of strong asymptotic dependence and concentrates on $[\smallwedge]$, we are encouraged to look for additional regular variation regimes on ${\mathbb{R}}_+^2\setminus [\smallwedge]$. Hence we get the following result. \[prop:forbid\] Suppose $$\label{e:doublewhammy} \bZ\in \MRV(\alpha, b(t), \nu, {\mathbb{R}}_+^2\setminus \{\bzero\}) \cap \MRV(\alpha_0, b_0(t), \nu_0, {\mathbb{R}}_+^2\setminus [\smallwedge])$$ with $b(t)/b_0(t)\to \infty$ and $0 {<} a_l{\leqslant}1{\leqslant}a_u<\infty.$ Then $\nu$, the limit measure on ${\mathbb{R}}_+^2\setminus \{\bzero\}$, concentrates on $[\smallwedge]$. To see this, consider the region above the ray $y=a_ux$. Then for $\delta>0$, as $t\to\infty$, referring to , $$\begin{aligned} t{\mathbb{P}}\Bigl[ \frac{Z_2-a_uZ_1}{b(t)} >\delta \Bigr] =&t{\mathbb{P}}\Bigl[ \frac{Z_2-a_uZ_1}{b_0(t)} > \frac{b(t)}{b_0(t)}\delta\Bigr] \to 0,\end{aligned}$$ since $b(t)/b_0(t) \to \infty$ and $S_0(\cdot)$ is a probability measure. Similarly, for the region below the ray $y=a_lx$, $$\begin{aligned} t{\mathbb{P}}\Bigl[ \frac{a_lZ_1 -Z_2}{b(t)} >\delta \Bigr] =&t{\mathbb{P}}\Bigl[ \frac{a_lZ_1-Z_2}{b_0(t)} > \frac{b(t)}{b_0(t)}\delta \Bigr] \to 0,\end{aligned}$$ from . Thus $\nu$ places no mass outside $[\smallwedge]$. [Clearly, a result analogous to Proposition \[prop:forbid\] holds where $\bC={\mathbb{R}}^{2}$ and $\bC_{0}$ is the appropriate equivalent of $[\diag]$ and $[\smallwedge]$ on ${\mathbb{R}}^{2}$.]{} How HRV on ${\mathbb{R}}^2\setminus [\smallwedge]$ can improve risk estimates. {#subsec:riskestimate} ------------------------------------------------------------------------------ Suppose $I_1$ and $I_2$ are financial instruments that have positive risks $Z_1$ and $Z_2$ per unit of investment where $\bZ=(Z_1,Z_2)$ satisfies . Suppose we buy one unit of $I_2$ and sell $2a_l$ units of $I_1$. The risk of this portfolio is $Z_2-2a_lZ_1$ and we have two asymptotic regimes that can be used to estimate the probability the risk is large. If we use MRV with scale function $b(t)$ then for large $x>0$, $$\begin{aligned} {\mathbb{P}}[Z_2-2a_u Z_1>x]=&{\mathbb{P}}\Bigl[\Bigl(\frac{Z_1}{b(t)},\frac{Z_2}{b(t)} \Bigr) \in \{(v,w)\in {\mathbb{R}}_+^2: w-2a_u v>x/b(t)\} \Bigr]\\ \approx & \frac{1}{t} \nu \{(v,w)\in {\mathbb{R}}_+^2: w-2a_u v >x/b(t)\} =0\\ \intertext{since the required region is outside the support $[\smallwedge]$ of the measure $\nu$. Is the risk really $0$ or did we use the wrong asymptotic approximation? If we use HRV with scale function $b_0(t)$, then we get a non-zero limit:} {\mathbb{P}}[Z_2-2a_u Z_1>x]=&{\mathbb{P}}\Bigl[\Bigl(\frac{Z_1}{b_0(t)},\frac{Z_2}{b_0(t)} \Bigr) \in \{(v,w)\in {\mathbb{R}}_+^2: w-2a_u v>x/b_0(t)\} \Bigr]\\ \approx & \frac{1}{t} \nu_0 \{(v,w)\in {\mathbb{R}}_+^2: w-2a_u v >x/b_0(t)\}.\\ \intertext{Now switch to generalized polar coordinates with $(v,w)=r(\mu_1,\mu_2) $ and $\mu_2=a_u\mu_1$ and the risk calculation is with respect to the product measure $\nu_{\alpha_0} \times S_0\bigl(d(\mu_1,\mu_2)\bigr)$ and } {\mathbb{P}}[Z_2-2a_u Z_1>x]\approx &\frac 1t \iint_{\{(r,(\mu_1,\mu_2)): r\mu_2-2a_u r\mu_1>x/b_0(t)\}} \alpha_0 r^{-\alpha_0 -1} S_0\bigl(\mathrm d(\mu_1,\mu_2) \bigr)\\ =& \frac 1t \Bigl(\frac{x}{b_0(t)}\Bigr)^{-{\alpha_0}} \int_{\{(\mu_1,\mu_2):\mu_2-2a_u\mu_1>0\}} (\mu_2-2a_u\mu_1)^{\alpha_0} S_0\bigl(\mathrm d(\mu_1,\mu_2)\bigr).\end{aligned}$$ Of course, in practice $S_0, \alpha_0, b_0$ must be replaced by estimators and $t$ is replaced by $n/k$ where $n$ is the sample size of observations and $k$ is the number of observations used in estimation. [An example where we carry out these calculations is given for simulated data in Section \[subsubsec:rare\].]{} [Exploring for HRV with the Hillish estimator]{}. {#subsec:Hillish} ------------------------------------------------- The Hillish estimator was designed for detection of the CEV model [@heffernan:resnick:2005; @heffernan:tawn:2004; @heffernan:resnick:2007; @lindskog:resnick:roy:2014; @das:mitra:resnick:2013; @das:resnick:2011b; @das:resnick:2011] and extended to detecting hidden regular variation in [@das:resnick:2015]. The generalized polar coordinate transform converts Cartesian coordinates in the definition of regular variation into coordinates satifying the CEV model. In this paper we show that the Hillish technique can detect HRV when the cone removed from ${\mathbb{R}}^{2}_{+}$ or ${\mathbb{R}}^{2}$ is $[\smallwedge]$. [The Hillish procedure is described below. First we define a conditional extreme value model.]{} ### The CEV model. {#subsub:cev} Suppose the [random variables]{} $(\xi,\eta) $ [form a random element of]{} ${\mathbb{R}}_+\times {\mathbb{R}}$ and there exists a regularly varying function $b(t)\to\infty$ and a non-null measure [$\mu \in {\mathbb{M}}((0,\infty)\times {\mathbb{R}})$]{} and $$\label{eqn:CEV} t {\mathbb{P}}\Bigl[\Bigl(\frac{\xi}{{b}(t)},{\eta}\Bigr) \in \; \cdot\;\Bigr] \to \mu(\cdot),\qquad \text{ in }{\mathbb{M}}((0,\infty)\times {\mathbb{R}}).$$ [Note that is of this form where only the first component $\xi=d(\bZ,\bC_0)$ is scaled. See also and . ]{} Additionally assume that - $\mu((r,\infty] \times \cdot \,)$ is a non-degenerate measure for any fixed $r>0$, and, - $H(\cdot ):=\mu((1,\infty)\times \cdot \,)$ is a probability distribution. Then $(\xi,\eta)$ satisfies a *conditional extreme value model* and we write $(\xi,\eta) \in \CEV({b},\mu)$. Note and are of the form given in . [Hence the HRV statements are equivalent to the appropriate transforms of the variables following a CEV model.]{} ### The Hillish procedure. {#subsub:hillish} Now suppose $(\xi_i,\eta_i); 1{\leqslant}i {\leqslant}n$ are iid replicates of $(\xi,\eta)$. Define $$\begin{array}{llll} \xi_{(1)} {\geqslant}\ldots {\geqslant}\xi_{(n)} & \text{The decreasing order statistics of $\xi_1,\ldots,\xi_n$.}\\[1mm] \eta_i^*, ~ 1 {\leqslant}i {\leqslant}n & \text{The $\eta$-variable corresponding to $\xi_{(i)}$, also called}\\[1mm] & \text{ the concomitant of $\xi_{(i)}$.}\\ [1mm] N_{i}^k= \sum\limits_{l=i}^k \bone_{\{\eta_l^* {\leqslant}\eta_i^*\}}& \text{Rank of $\eta^*_i$ among $\eta_1^*,\ldots,\eta_k^{*}$. We write $N_i=N_i^k$.}\\ [1mm] \eta_{1:k}^* {\leqslant}\eta_{2:k}^* {\leqslant}\ldots {\leqslant}\eta_{k:k}^* & \text{The increasing order statistics of $\eta_1^*,\ldots,\eta_k^*$.}\\[1mm] \end{array}$$ [By analogy with the Hill estimator and the Hill plot, ]{} the [*Hillish statistic*]{} is defined for $1{\leqslant}k{\leqslant}n$ as $$\begin{aligned} \label{def:Hillish} \Hillish_{k,n}=\Hillish_{k,n}(\bbxi,\bbeta) := \frac{1}{k} \sum\limits_{j=1}^{k} \log \frac{k}{j} \log \frac{k}{N_{j}^k} \end{aligned}$$ According to [@das:resnick:2011b Propositions 2.2 and 2.3], if $(\xi,\eta)\in \CEV(b,\mu) $ then [there exists a limit $I_\mu$ and ]{} $\Hillish_{k,n}\stackrel{P}{\to} I_\mu$ and moreover, $\mu $ is a product measure iff both $$\begin{aligned} \label{eq:onetwo} \Hillish_{k,n}(\bbxi,\bbeta) {\stackrel{\lower0.2ex\hbox{$\scriptscriptstyle \it{P} $}}{\rightarrow}}1 \quad\text{and} \quad \Hillish_{k,n}(\bbxi,-\bbeta) {\stackrel{\lower0.2ex\hbox{$\scriptscriptstyle \it{P} $}}{\rightarrow}}1, \end{aligned}$$ [as $k\to \infty, n\to \infty, n/k\to \infty$]{}. Note the limits in , are all product measures. [Hence the diagnostic for detecting regular variation with a specified forbidden zone for the random vector $\bZ$ is to plot the Hillish statistic of $\gpolar (\bZ)$.]{} [We emphasize that if holds, we have empirical behavior [*consistent*]{} with the presence of regular variation but this does not prove existence of regular variation. When the Hillish technique fails because the plots do not hug the line at height 1, we can reject a hypothesis of regular variation. ]{} Data Analysis with Simulated Data {#sec:dataAnal} ================================= Before moving to real data, we test our analysis techniques on two simulated data sets to see how well they perform in Section \[ex:sim1\] and \[ex:sim2\]. Further, we discuss MRV and HRV properties and their detection. [In Section \[subsec:data\],]{} we analyze two real bivariate data sets both of which exhibit heavy-tailed margins and strong asymptotic dependence. Example 1: Full asymptotic dependence. {#ex:sim1} -------------------------------------- Suppose $Z_{1}\sim$ Pareto(1.5) and $Z_{2}\sim$ Pareto(2.5) and independent of each other. Let $B_{1}, B_{2}$ be iid Bernoulli (0.5) random variables also independent of $Z_{1}$ and $Z_{2}$. Now define the vector $\bX=(X_{1},X_{2})$ as $$\begin{aligned} X_{1}&=B_{1}Z_{1}+(1-B_{1})Z_{2},\\ X_{2}&=B_{1}Z_{1}+B_{2}(1-B_{1})(1.5Z_{2})+(1-B_{2})(1-B_{1})(0.5Z_{2})\end{aligned}$$ ![Example \[ex:sim1\]: (Left) Scatter plot of 10,000 data points. (Center and right) Hill plots for tail parameters of the marginal distribution[s of $X_1$ (center) and $X_2$ (right).]{}[The horizonal lines are at height $1.5.$]{}[]{data-label="fig:Margins_sim"}](Sim1_scatter.pdf){width="6in"} By construction $${ \bX= \begin{cases} (Z_1,Z_1),& \text{ with probability }P[B_1=1]=\frac 12,\\ (Z_2,1.5Z_2) ,& \text{ with probability }P[B_1=0,B_2=1]=\frac 14,\\ (Z_2,0.5 Z_2), & \text{ with probability }P[B_1=0,B_2=0]=\frac 14,\\ \end{cases} }$$ so that $\bX=(X_{1},X_{2})$ lies on $y=x$ with probability 0.5. With probability 0.25 each it is either on the line $y=0.5x$ or on $y={1.5}x$. In this model, $\bX$ is MRV with parameter $\alpha=1.5$ and it has full dependence on the diagonal $[\diag]:= \{(x,y)\in{\mathbb{R}}_{+}^{2}: y=x\}$. On the other hand, on ${\mathbb{R}}^{2}_{+} \setminus [\diag]$, $$\begin{aligned} t\,{\mathbb{P}}\left(\frac{d((X_{1},X_{2}), [\diag])}{t^{1/2.5}/2\sqrt{2}} > x, \frac{(X_{1},X_{2})}{\sqrt{2}d((X_{1},X_{2}), [\diag])} = \ba \right) & = t\,{\mathbb{P}}\left(\frac{(2|X_{1}-X_{2}|}{t^{1/2.5}} > x, \frac{(X_{1},X_{2})}{|X_{1}-X_{2}|} = \ba \right)\\ & \to x^{{-2.5}} \times \frac 14 = \frac 12 \nu_{{2.5}}(x,\infty)S_{{0}}(\{{\sqrt 2 \ba }\}).\end{aligned}$$ where $\ba=(2,1)$ or $(2,3)$ [ and $S_0(\sqrt 2 \ba)=0.5$]{}. Hence we have HRV on ${\mathbb{R}}_+^{2}\setminus[\diag]$ with tail parameter $\alpha_{0}=2.5$. ![Example \[ex:sim1\]: [Histogram]{} of $\theta_1$ (left), diamond plot (center), and Hill plot for tail estimate of $|X_{1}-X_{2}|$ (right). [The horizontal line is at height 2.5]{}.[]{data-label="fig:diamond_sim1"}](Sim1_density.pdf){width="6in"} We generate $n=10,000$ iid samples from this data set. The scatter plot in Figure \[fig:Margins\_sim\] shows the dependence structure of $\bX$ along with Hill plots of $X_{1},X_2$ which supports the premise that $\alpha=1.5$. ![Example \[ex:sim1\]: Hillish plots for $(\xi,\eta)$ (left) and $(\xi,-\eta)$ (right) where $(\xi,\eta)=(|X_1-X_2|,X_1/|X_1-X_2|)$.[]{data-label="fig:hillish_sim1"}](Sim1_Hillish.pdf){width="4.5in"} To understand the dependence structure of the variables $\bX$, we [make]{} [the diamond plot, the]{} transformation from ${\mathbb{R}}^2\mapsto \aleph_{\bzero} \subset {\mathbb{R}}^2$ onto the $L_1$ unit sphere represented by the diamond $\{(\theta_1,\theta_2):|\theta_1|+|\theta_u|=1\}$ We do the mapping at various thresholds determined by $k$, the number of order statistics of the norms $\|\bx\|=|x_1|+|x_2|$. In Figure \[fig:diamond\_sim1\], the diamond plot and histogram of the angular measure is shown for $k=100$. Clearly the data is concentrated at $x=y$. The Hill plot of the quantity $|X_1-X_2|$ supports the fact that data was generated with hidden tail parameter $\alpha_{0}=2.5$. Finally, we look at the Hillish statistic for $(\xi,\eta)=(|X_1-X_2|,X_1/|X_1-X_2|)$ after removing $[\diag]$. The Hillish plot in Figure \[fig:hillish\_sim1\] is convincingly stable and close to 1, supporting the presence of HRV as expected from the generation procedure here. Example 2: Strong asymptotic dependence. {#ex:sim2} ---------------------------------------- [In this example, we simulate data from a model with strong asymptotic dependence. We also exhibit estimation of rare probabilities and conduct a sensitivity analysis when the support of the regular variation for the the first level, and hence that for HRV is not correctly identified.]{} ![Example \[ex:sim2\]: (Left) Scatter plot of 30,000 data points. (Center and right) Hill plots for tail parameters of the margin[al distributions ]{} of $X_1,X_2$. [The horizontal lines are at height $1.5$.]{}[]{data-label="fig:Margins_sim2"}](Sim2_scatter.pdf){width="6.5in"} Suppose $R_{1}\sim$ Pareto(1.5) and $R_{2}\sim$ Pareto(2.5) and independent of each other. Let $\Theta_{1}\sim \Unif[0.4,0.6]$, $\Theta_{2}\sim \Unif([0,1]\setminus[0.4,0.6)]$, $B\sim$ Bernoulli (0.5) random variables. Assume the random variables are all independent. Now define the vector $\bX=(X_{1},X_{2})$ as $$\begin{aligned} X_{1}&=BR_{1}\Theta_{1} + (1-B)R_{2}\Theta_{2},\\ X_{2}&=BR_{1}(1-\Theta_{1}) + (1-B)R_{2}(1-\Theta_{2}).\end{aligned}$$ By construction, $\bX$ is MRV on ${\mathbb{R}}_{+}^{2}\setminus\{\bzero\}$ with tail parameter $\alpha=1.5$. [Corresponding to $(\theta_l,\theta_u)=(0.4,0.6)$, we have by that $(a_l,a_u)=(0.67,1.5)$ and therefore]{} $$[\smallwedge] := \{(x,y)\in {\mathbb{R}}_{+}^{2}: 0.67 x {\leqslant}y {\leqslant}1.5 x\}.$$ [This gives hidden regular variation with tail parameter $\alpha_{0}=2.5$ on ${\mathbb{R}}_{+}^{2}\setminus[\smallwedge]$.]{} ![Example \[ex:sim2\]: Diamond plot, histogram and Hill plot for tail estimate of $d(\bX,[\smallwedge])$. [The horizontal line in the Hill plot is at height 2.5.]{}[]{data-label="fig:diamond_sim2"}](Sim2_density.pdf){width="6.5in"} ![[Example \[ex:sim2\]: Hillish plots for (i)$(\xi_{1},\eta_{1}),$ (ii)$(\xi_{1},-\eta_{1})$, (iii)$(\xi_{2},\eta_{2}),$ (iv)$(\xi_{2},-\eta_{2})$ respectively where $(\xi_{1},\eta_{1})$ and $(\xi_{2},\eta_{2})$ are obtained by using and on $\bX=(X_{1},X_{2})$.]{} []{data-label="fig:hillish_sim2"}](Sim2_Hillish.pdf){width="6.5in"} ![Example \[ex:sim2\]: Boxplots for $\frac{\bar{p}_{1}(1)}{p_{1}(1)},\frac{\hat{p}_{1}(1)}{p_{1}(1)},\frac{\bar{p}_{1}(4)}{p_{1}(1)},\frac{\hat{p}_{1}(4)}{p_{1}(4)}$ (top) and $\frac{\bar{p}_{2}(1)}{p_{2}(1)},\frac{\hat{p}_{2}(1)}{p_{2}(1)},\frac{\bar{p}_{2}(4)}{p_{2}(4)},\frac{\hat{p}_{2}(4)}{p_{2}(4)}$ (bottom) .[]{data-label="fig:estimates_sim2"}](Sim2_estimates.pdf){width="6.5in"} We generate $n=30,000$ iid samples from this data set. The scatter plot in Figure \[fig:Margins\_sim2\] shows the dependence structure of $\bX$ along with Hill plots of $X_{1},X_2$ which supports the premise that $\alpha=1.5$. To understand the dependence structure of $\bX$, we [graph]{} the diamond plot as used in the previous example. We do the mapping at various thresholds determined by $k$, the number of order statistics of the norms $|x|+|y|$. In Figure \[fig:diamond\_sim2\], the histogram of angles and the diamond plot are shown for $k=100$ and shows the angles are Uniform in \[0.4, 0.6\] for high values of $|x|+|y|$. The Hill plot of the quantity $d(\bX,[\smallwedge])$ supports the fact that data was generated with hidden tail parameter $\alpha_{0}=2.5$. Finally, we look at a Hillish statistic for [$(\xi_{1},\eta_{1})$ and $(\xi_{2},\eta_{2})$ respectively which are obtained by using and on $\bX=(X_{1},X_{2})$ after removing $[\smallwedge]$.]{} The Hillish plots in Figure \[fig:hillish\_sim2\] are again convincingly stable and close to 1 [and detect the hidden regular variation.]{} ### Probabilities of rare sets for this example. {#subsubsec:rare} Now to further illustrate our methods, we compute ${\mathbb{P}}(X_{2}-2X_{1}>x)$ and ${\mathbb{P}}(X_{2}-3X_{1}>x)$. Without resorting to hidden regular variation we have $\bX$ is MRV on ${\mathbb{R}}_{+}^{2}\setminus\{\bzero\}$ with tail parameter $\alpha=1.5$ and the limit measure concentrates on $$[\smallwedge] := \{\bx\in {\mathbb{R}}_{+}^{2}: a_{l} x_1 {\leqslant}x_2 {\leqslant}a_{u} x_1\} = \{\bx \in {\mathbb{R}}_{+}^{2}: 0.67 x_1 {\leqslant}x_2 {\leqslant}1.5 x_1\}.$$ Hence with [the usual regular variation]{} techniques we would estimate both $${\mathbb{P}}(X_{2}-2X_{1}>x) \approx 0 \quad \text{and} \quad {\mathbb{P}}(X_{2}-3X_{1}>x) \approx 0.$$ But for this example we can compute the exact answer without resorting to asymptotic approximations and we get, $$\begin{aligned} p_{1}(x):=& {\mathbb{P}}(X_{2}-2X_{1}>x) = \frac 12 {\mathbb{P}}(R_{1}(1-\Theta_{1}) {-} 2R_{1}\Theta_{1} > x) + \frac 12 {\mathbb{P}}(R_{2}(1-\Theta_{2}) {-} 2R_{2}\Theta_{2} > x)\nonumber \\ \intertext{and because $3\Theta_1>1$, this is} =& \frac 12 {\mathbb{P}}(R_{2}(1-3\Theta_{2}) > x) = \frac 5{84} x^{{-2.5}}. \label{e:p1x} \intertext{ Similarly we can compute} p_{2}(x):=&{\mathbb{P}}(X_{2}-{3}X_{1}>x) = \frac 5{112} x^{{-2.5}}.\label{e:p2x} \end{aligned}$$ ![[Example \[ex:sim2\]: Hill plots for estimating $\alpha_{0}$ (known to be 2.5 in the model) when the support set of MRV is incorrectly specified as $(a_{l}, a_{u})= (0.5,2)$ (left), $(a_{l}, a_{u})= (0.9,1.11)$ (middle), $(a_{l}, a_{u})= (1,1)$ (right) respectively. The horizontal line in the Hill plot is at height 2.5.]{}[]{data-label="fig:sim2_alpha0_err"}](Sim2_alpha0_err.pdf){width="6in"} ![[Example \[ex:sim2\]: Hillish plots for (i)$(\xi_{1},\eta_{1}),$ (ii)$(\xi_{1},-\eta_{1})$, (iii)$(\xi_{2},\eta_{2}),$ (iv)$(\xi_{2},-\eta_{2})$ respectively where $(\xi_{1},\eta_{1})$ and $(\xi_{2},\eta_{2})$ are obtained by using and on $\bX=(X_{1},X_{2})$ and the support of MRV is (incorrectly) identified by $(a_{l},a_{u})=(0.9,1.11)$.]{}[]{data-label="fig:sim2_hillish_err"}](Sim2_Hillish_err.pdf){width="6in"} ![Example \[ex:sim2\]: Boxplots for $\frac{\bar{p}_{1}(1)}{p_{1}(1)},\frac{\hat{p}_{1}(1)}{p_{1}(1)},\frac{\bar{p}_{1}(4)}{p_{1}(4)},\frac{\hat{p}_{1}(4)}{p_{1}(4)}$ (top) and $\frac{\bar{p}_{2}(1)}{p_{2}(1)},\frac{\hat{p}_{2}(1)}{p_{2}(1)},\frac{\bar{p}_{2}(4)}{p_{2}(4)},\frac{\hat{p}_{2}(4)}{p_{2}(4)}$ (bottom) with (incorrect) identification of support where $a_l=0.5, a_{u}=2$.[]{data-label="fig:estimates_sim2_alt1"}](Sim2_estimates_052.pdf){width="6in"} ![Example \[ex:sim2\]: Boxplots for $\frac{\bar{p}_{1}(1)}{p_{1}(1)},\frac{\hat{p}_{1}(1)}{p_{1}(1)},\frac{\bar{p}_{1}(4)}{p_{1}(1)},\frac{\hat{p}_{1}(4)}{p_{1}(4)}$ (top) and $\frac{\bar{p}_{2}(1)}{p_{2}(1)},\frac{\hat{p}_{2}(1)}{p_{2}(1)},\frac{\bar{p}_{2}(4)}{p_{2}(4)},\frac{\hat{p}_{2}(4)}{p_{2}(4)}$ (bottom) with (incorrect) identification of support where $a_l=0.9, a_{u}=1.1$.[]{data-label="fig:estimates_sim2_alt2"}](Sim2_estimates_9111.pdf){width="6in"} [If we pretend we do not know the exact answers provided by , , can we give]{} better estimates than 0 using asymptotic methods on our simulated data set? Under hidden regular variation after removing $\bC_{0}=[\smallwedge]$, we know that holds for $\bX$ and hence $$\label{e:delC000} \frac nk{\mathbb{P}}\Bigl[ \frac{d({\bX}, \bC_{0})}{b_0(\frac nk)} > x, \frac{\bX}{d(\bX,\bC_0)}\Bigr) \in \Lambda \,\Bigr] \to x^{{-\alpha_{0}}} \times S_0(\Lambda)$$ as $n\to\infty$, $k\to \infty$, $k/n\to0$, where $S_0(\cdot) $ is a probability measure and $\Lambda\subset\aleph_{\mathbb{C}_0}$. From Section \[subsec:riskestimate\], we have for $x>0$, as $n\to \infty$, $k\to\infty$, $k/n \to 0$, $$\begin{aligned} \label{estimate} \frac nk p_1(b_0(\frac nk) x)= \frac nk {\mathbb{P}}[X_{2}-2X_{1}>b_0(\frac nk )x] \to x^{-{\alpha_0}} \int_{\{(\mu_1,\mu_2):\mu_2-2\mu_1>0\}} (\mu_2-2\mu_1)^{\alpha_0} S_0\bigl(d(\mu_1,\mu_2)\bigr), \end{aligned}$$ with a similar limiting expression for $p_2(x)$. This suggests we need to estimate $\alpha_0$, $b_0(n/k)$, $S_0(\cdot)$ and of course the wedge $\mathbb{C}_0$. For the wedge, we estimate $\hat a_{l}, \hat a_{u}$ using the $5th$ and $95th$ percentile of the range of $X_{1}/(X_{1}+X_{2})$ for 100 highest values of $X_{1}+X_{2}$ for each simulation. In replace $x$ by 1 and $\Lambda$ by $\aleph_{\mathbb{C}_0}$ and then we estimate $b_0(n/k)$ with the $k$th largest value of $d(\bX_i,\mathbb{C}_0) $ corresponding to $\bX's $ outside $\mathbb{C}_0. $ Alternatively, if we fix $b_0(n/k)$ then, we obtain the appropriate $k$ largest value of $d(\bX_i,\mathbb{C}_0) $ corresponding to $\bX's $ outside $\mathbb{C}_0.$ to be used. For $S_0(\cdot)$ we modify the argument leading to [@resnickbook:2007 Eq. 9.47, p. 313]. Corresponding to estimates $\widehat{\mathbb{C}}_0$, $\hat b_0(n/k)$, and choice of $k$, we enumerate the gpolar-transformed points outside $\widehat{\mathbb{C}}_0$, corresponding to the $k$ largest values of $d(\bX_i,\widehat{\mathbb{C}}_0)$ as $\{(r_{i},\mu_{1i},\mu_{2i}): 1{\leqslant}i {\leqslant}k\}$. Then using these points we estimate $S_0 (\cdot)$ with the empirical distribution as $$\hat S_0(\cdot)=\frac 1k \sum_{i=1}^k \epsilon_{(\mu_{1i},\mu_{2i})} (\cdot).$$ This leads to the risk estimates, $$\begin{aligned} \hat{p}_{1}(x) =& {x^{-{\hat \alpha_0}} } \frac kn (\hat b_0(n/k))^{\hat{\alpha_{0}}} \frac 1k \sum_{i=1}^{k} (\mu_{2i}-2\mu_{1i})^{\hat{\alpha_{0}}}\bone_{\{\mu_{2i}>2\mu_{1i}\}} , \label{newestimate}\\ \hat{p}_{2}(x)=& {x^{-{\hat \alpha_0}} } \frac kn (\hat b_{0}(n/k))^{\hat{\alpha_{0}}} \frac 1k \sum_{i=1}^{k} (\mu_{2i}-{3}\mu_{1i})^{\hat{\alpha_{0}}}\bone_{\{\mu_{2i}>2\mu_{1i}\}}.\label{newestimate2} \end{aligned}$$ Since $\alpha_{0}, a_{l}, a_{u}$ are known for this simulation example, we may compare $\hat p_i(x)$ with $\bar p_i(x)$ estimated using the three known values. We carry out comparisons using $x=1$ and $x=4$. We conduct simulations with $n=10,000$ and use a value of $k$ corresponding to $b_0(n/k)=2$. We compute $\hat{p}_{1}(1),\hat{p}_{1}(4),\bar{p}_{1}(1),\bar{p}_{1}(4),$ for 100 iterations and create box plots of $\frac{\hat{p}_{1}(1)}{p_{1}(1)},$ $\frac{\hat{p}_{1}(4)}{p_{1}({4})}$, $\frac{\bar{p}_{1}({1})}{p_{1}(1)}$, $\frac{\bar{p}_{1}(4)}{p_{1}(4)}$ and we do the same for $p_{2}(1)$ and $p_{2}(4).$ From Figure \[fig:estimates\_sim2\], clearly the estimates perform pretty well since the ratios of the estimates to the real values are very close to 1. Clearly, when $\alpha_{0}$ is estimated the error bounds become larger, but still perform reasonably. Note that the quantities we compute have low probabilities: $$p_{1}(1)=0.{0}595,\quad p_{2}(1)=0.044, \quad p_{1}(4)=0.002, \quad p_{2}(4)=0.0014.$$ To summarize: This estimation procedure can be used to calculate risk probabilities in the presence of hidden regular variation when the primary regular variation gives a zero risk estimate. ### Sensitivity analysis in this example Clearly, the probability estimation procedure we discussed hinges on our ability to appropriately estimate the support set of regular variation at the first level, given by $$[\smallwedge] := \{\bx \in {\mathbb{R}}_{+}^{2}: a_{l} x_{1} {\leqslant}x_{2} {\leqslant}a_{u} x_{1}\}.$$ An inaccurate estimation of the support leads to an improper estimation of $\alpha_{0}$ and hence also the probabilities of rare events. Under the current Example \[ex:sim2\], we conduct a sensitivity analysis of our estimation procedures by choosing a support set which is different from the one that is specified by the model. In the example, the support set is identified by $(a_{l}^{*},a_{u}^{*})=(0.67,1.5)$. Recall that we have 30,000 data points from this model. First we estimate $\alpha_{0}$ under an improper specification of $a_{l}$ and $a_{u}$. This is estimated using a Hill plot of points comprised of $(X_{2}-a_{u}X_{1})/\sqrt{1+a_{u}^{2}}$ for $X_{2}-a_{u}X_{1}>0$ and $(a_{l}X_{1}-X_{2})/\sqrt{1+a_{l}^{2}}$ for $a_{l}X_{1}-a_{u}X_{2}>0$ for different choices of $(a_{l}, a_{u})$. In Figure \[fig:sim2\_alpha0\_err\], we provide Hill plots for estimating $\alpha_{0}$ by using $(a_{l}, a_{u})= (0.5,2), (a_{l}, a_{u})= (0.9,1.11), (a_{l}, a_{u})= (1,1)$ respectively. Comparing the plots with the one in Figure \[fig:diamond\_sim2\], the estimates clearly move away from the actual value $\alpha_{0}=2.5$ as the size of the support decreases. When we take $(a_{l},a_{u}) = (0.5,2)$ the data points used to estimate $\alpha_{0}$ are a subset of the points used to estimate $\alpha_{0}$ when $(a_{l}^{*},a_{u}^{*})=(0.67,1.5)$, and thus are regularly varying with parameter $\alpha_{0}=2.5$; hence the Hill plot clearly hugs the horizontal line at $y=2.5$. As the size of the support set decreases we see the Hill estimates become lower than $2.5$ and moves towards $\alpha=1.5$. We also observe in Figure \[fig:sim2\_hillish\_err\] that the Hillish plots are not that close to the horizontal line at height 1 when the support is not correctly identified; in this case $(a_{l},a_{u})=(0.9,1.11)$. In comparison, Figure \[fig:hillish\_sim2\] clearly shows that the Hillish plots are close to 1, when the support is correctly specified. Finally we estimate probabilities $p_{1}(x)={\mathbb{P}}(X_{2}-2X_{1}>x)$ and $p_{2}(x)={\mathbb{P}}(X_{2}-3X_{1}>x)$ when the support sets are identified incorrectly. Figure \[fig:estimates\_sim2\_alt1\] corresponds to boxplots of $\frac{\bar{p}_{1}(1)}{p_{1}(1)},\frac{\hat{p}_{1}(1)}{p_{1}(1)},\frac{\bar{p}_{4}(4)}{p_{1}(1)},\frac{\hat{p}_{1}(4)}{p_{1}(4)}$ and $\frac{\bar{p}_{2}(1)}{p_{2}(1)},\frac{\hat{p}_{2}(1)}{p_{2}(1)},\frac{\bar{p}_{2}(4)}{p_{2}(4)},\frac{\hat{p}_{2}(4)}{p_{2}(4)}$ where $\hat{p}_{i}, i=1,2$ uses the estimators given in , with $\alpha_{0}$ computed using $(a_{l},a_{u})=(0.5,2)$ and $\bar{p}_{i}, i=1,2$ uses $\alpha_{0}=2.5, a_{l}=0.5,a_{u}=2$. Observe that the boxplots are quite close to 1, since $\alpha_{0}$ is estimated well. On the other hand in Figure \[fig:estimates\_sim2\_alt2\], the similar boxplots are done for $(a_{l},a_{u})=(0.9,1.11)$. In this case, $\alpha_{0}$ is not that well-estimated and hence the boxplots are clearly away from 1. In both cases, 100 replications of data sets with 10,000 data points in each were used to create the boxplots. In conclusion we can see that an incorrect identification of the support of regular variation can often lead to incorrect estimates. Although if the identified support of hidden regular variation is a bit smaller than the correct one (which means that the identified support of MRV is larger than the correct one), then the estimates are still quite accurate. Examples of strong asymptotic dependence with real data. {#subsec:data} ======================================================== We now analyze two real data sets: (i) facebook wall posts and (ii) returns from Exxon and Chevron. Facebook wall posts {#subsec:facebk} ------------------- ![Scatter plot of node-wise out-degree and in-degree of Facebook wallpost graph.[]{data-label="fig:fbwallscatter"}](Fbwall_scatter.pdf){width=".4\textwidth"} The Facebook wall posts data was downloaded from <http://konect.uni-koblenz.de/networks/facebook-wosn-wall> and has been analyzed in [@viswanath:2009]. Conversion of edge data to node-indexed in- and out-degree counts was done using the R-package [*igraph*]{} [@csardi:nepusz:2006]. The data is a directed network representing posts by Facebook users to other users’ walls. Nodes are users and a directed edge represents one post from the user to the user whose wall is receiving the post. There are 46,952 users and 876,993 edges. We focus on out- and in-degree [indexed by the nodes as $\{(Z_{1,i},Z_{2,i}): 1{\leqslant}i {\leqslant}46952\}$]{}. Of course this data is not the result of iid replication but is rather node-indexed; however, for reasons still being investigated, conventional tools of heavy tail analysis seem quite effective on node-indexed network data. The scatter-plot of (out,in)-degrees in Figure \[fig:fbwallscatter\] shows the expected strong asymptotic dependence between out- and in-degrees. ![(Left) Marginal estimation of tail indices by QQ plot slope estimation and (right) altHill plotting for out- and in-degree. Horizontal lines are at height 2.8.[]{data-label="fig:MarginalAnal"}](Fbwall_qq_altHill.pdf){width="6.5in"} The plots in Figure \[fig:MarginalAnal\] give the estimation of distribution tail indices for out- and in-degree. The slope estimator based on QQ-plots ([@beirlant:vynckier:teugels:1996; @kratz:resnick:1996; @resnickbook:2007]) gives approximately $\alpha=2.8$ for both out- and in-degree. Note this estimate is for the tail of the cumulative distribution functions and not, as is customary in network science, the index of the power law of the mass functions. The Hill estimator is ineffective and we have provided altHill plots ([@resnickbook:2007; @resnick:starica:1997a; @drees:dehaan:resnick:2000]). {width="5in"} \[fig:1stQuadrant\_facebook\] To get more information about the dependence structure, we construct a diamond plot using thresholding corresponding to the 200 largest $L_1$ norms of (out,in). The scatter plot in Figure \[fig:fbwallscatter\] is less clear than for simulated data and shows points dispersed from the main cluster about the diagonal and so the estimates of the support of $\btheta$ in the diamond plot are not as evident as in Figure \[fig:1stQuadrant\_facebook\]. [We estimate the support interval ]{} using the interquartile range and obtain $[.4479, .5305]=[\theta_l, \theta_u]$. This corresponds to slopes $(a_l,a_u)=(\theta_u^{-1}-1,\theta_l^{-1}-1)=(.885,1.23).$ We also include a boxplot of the values of $\theta_1$ corresponding to the 200 largest values of the $L_1$ norm of (out,in). [l]{}[0.35]{} {width=".35\textwidth"} \[fig:minusWedge\] Having determined $[\smallwedge]$, we remove it from the first quadrant and use the remaining points, illustrated in Figure \[fig:minusWedge\], to seek hidden regular variation. Preliminary diagnostics use equations and corresponding to points above and below $[\smallwedge]$ to estimate $\alpha_0$, the index of hidden regular variation. There are 12,089 points above $[\smallwedge]$ in the region we refer to as $[>\smallwedge]=\{(x,y): y>(1.23)x>0\}$ and 24,687 below in the region $[<\smallwedge]=\{(x,y): 0<y<(0.885)x\}.$ ![Facebook above $[\smallwedge]$: QQ estimate is 3.2 for $k=100$ and altHill is 3.3; the red line is at height 3.3.[]{data-label="fig:facebk0alpha0"}](Fbwall_qq_alt_hidden.pdf){width="3.5in"} For the region $[<\smallwedge]$, a combination of QQ and altHill plotting gives a stable region for various values of $k$, the number of upper order statistics, between 100-500 and a value of $\hat \alpha_0=2.8$, the tail index of $(0.885)Z_1-Z_2$ from , which is not measurably different from $\hat \alpha =2.8$ found for the marginal distributions of (out,in). This raises doubts about the presence of HRV in the region $[<\smallwedge]$. For the region $[>\smallwedge]$ we use as data $(Z_2-(1.23)Z_1)_+$ and estimate $\hat \alpha_0\approx 3.2$. The QQ-estimate is 3.2 and altHill gives about 3.3; both estimates are greater than $\alpha=2.8$ so there is evidence of the existence of HRV in the region above $[\smallwedge]$. The QQ and altHill plots for the region $[>\smallwedge]$ are given in Figure \[fig:facebk0alpha0\]. The evidence for HRV is further strengthened by excellent Hillish plots described in Section \[subsec:Hillish\] applied to the generalized polar coordinates $(Z_2-(1.23)Z_1, Z_2/Z_1)$ as descibed in . Both Hillish plots in Figure \[fig:fb\_gpolar\] hug the horizontal line at height 1. ![Hillish plots for generalized polar coordinates of points in $[>\smallwedge]$.[]{data-label="fig:fb_gpolar"}](Hillishgpolar.jpeg){height="2.5in"} The diamond plot in Figure \[fig:1stQuadrant\_facebook\] shows the presence of points on the line $x_1+x_2=1$ with small values of $\theta$, corresponding to two dimensional points in $[<\smallwedge]$. Previously the estimation of the range of the angular measure of the primary regular variation discounted these points. However, the estimation of the tail index of the distance to $[\smallwedge]$ being $2.8$, the same as the marginal distribution indices of (out,in), suggests an alternate model which lumps together $[\smallwedge]\cup[<\smallwedge]$ as the region of concentration for the limit measure $\nu(\cdot)$ of the primary regular variation in . So our alternate model is regular variation on ${\mathbb{R}}_+^2 \setminus \{\bzero\}$ with index $2.8$ and limit measure which concentrates on $\{\bx \in {\mathbb{R}}_+^2\setminus \{\bzero\}: x_2/x_1<1.23\}$ and hidden regular variation on ${\mathbb{R}}_+^2\setminus ([\smallwedge]\cup [<\smallwedge])$ with index 3.3. Exxon and Chevron returns. {#subsec:oil} -------------------------- ![Stock prices and scatterplot of Chevron and Exxon returns.[]{data-label="fig:scatter"}](oilprice.pdf "fig:"){width="3.in"} ![Stock prices and scatterplot of Chevron and Exxon returns.[]{data-label="fig:scatter"}](returns.pdf "fig:"){width="3.in"} For this example of financial returns, the state space is ${\mathbb{R}}^2\setminus \{\bzero\}$ and for the HRV property we could try deleting $[\smallwedge_+]\subset {\mathbb{R}}_+^2$ in the first quadrant and $[\smallwedge_-] \subset (-\infty,0)^2$ in the third quadrant. For illustration, we concentrate on deleting only a wedge from the first quadrant and seeking HRV with points above the upper boundary of $[\smallwedge_+]$. This is done partly because there is no guarantee that HRV will hold globally. The data consists of closing daily prices of Exxon (XOM) and Chevron (CVX) from January 2, 1998 to August 9, 2013. For each variable we calculate daily returns for each company called (exxonr, chevronr). The length of the return vector is 3925. One expects strong dependence from two big companies engaged in similar economic activities and this is shown in the raw scatter plot of the variables in Figure \[fig:scatter\]. The four tails of the variables ($\pm$exxonr, $\pm$chevronr) are quite similar. Based on analyses (not shown) using the QQ estimator, Hill and altHill plots, (eg. [@kratz:resnick:1996], [@resnickbook:2007 p. 101, 366]) we estimate [l]{}[.45]{} {width="3in"} the marginal tail indices $\alpha =2.7$ in all four cases. Since the tails are estimated to have the same $\alpha$, we did not attempt to standardize the variables to $\alpha=1$ as is often done by either the power method or the ranks transform. To understand the dependence structure of the variables (exxonr,chevronr), we [make]{} a diamond plot of the data. We do the mapping after thresholding the data at various values determined by $k$, the number of order statistics of the norms $|x_1|+|x_2|$. This is the two-tail empirical equivalent to using the $L_1$ norm. After experimenting with thresholds, we settled on $k=200$ which in the first quadrant produced a range of $\theta_1=x_1/(x_1+x_2)$ equal to $(.095,.649).$ [We finalized our estimate of the support of the limit angular measure]{}, by using the 10% and 90% quantiles of the values of $\theta_1$ as $(.393,.589)$. This corresponds to slope estimates for $[\smallwedge]$ of $(\hat a_l,\hat a_u)= (.698,1.545)$. The strong asymptotic dependence among the marginals is evident from the diamond plot and histogram of $\theta_1$ in Figure . There is little evidence that a large positive change in one variable is accompanied by a large negative change in the other as shown by the lack of points in the second and fourth quadrants in Figure \[fig:oil\_diamond\]. There is no visual evidence supporting the hypothesis of full asymptotic dependence. [r]{}[.5]{} {width=".45\textwidth"} Remark \[rem:genius\] suggests verifying the necessary condition for HRV on ${\mathbb{R}}^2\setminus [\smallwedge_+]$ by computing the tail index of what is essentially the distance of a point to $[\smallwedge_+]$. We seek evidence of regular variation on ${\mathbb{R}}^2\setminus [\smallwedge_+]$ by using points $\bx$ of the return sample that satisfy, 1. $x_2>0$ (points above the horizontal axis); 2. $x_2-(1.545)x_1>0$ (points in the first or second quadrant above the ray $x_2=1.545x_1,\,x_1>0$) 3. $x_1+1.545x_2>0$ (points in the first or second quadrant in the region bounded by the ray $x_2=1.545x_1,\,x_1>0$ and the ray perpendicular to this ray emanating from the origin into the third quadrant). Points to the left of this perpendicular would be closer to $[\smallwedge_-]$ rather than $[\smallwedge_+]$ and are excluded. There are 706 points satisfying the three conditions; we call these points [*oilReturns2013Gr*]{}. These are plotted in Figure \[fig:niceOilypts\]. The angle between the two rays of biggest slope is 90 degrees. We estimate the tail index $\alpha_0$ of the distance of points in [ *oilReturns2013Gr*]{} to the boundary of $[\smallwedge_+]$ to be greater than $\alpha =2.7$ using altHill, Hill and QQ plots. This corresponds to estimating the tail index of $Z_2-a_uZ_1$ as in . The plots are given next. ![AltHill, Hill and QQ plots to estimate $\alpha_0$. The red horizontal lines are drawn at height $\alpha=2.7$.[]{data-label="fig:csReturnsGr"}](altHilloilReturns2013GrNew.pdf "fig:"){width="2.in" height="1.5in"} ![AltHill, Hill and QQ plots to estimate $\alpha_0$. The red horizontal lines are drawn at height $\alpha=2.7$.[]{data-label="fig:csReturnsGr"}](HilloilReturns2013GrNew.pdf "fig:"){width="2.in" height="1.5in"} ![AltHill, Hill and QQ plots to estimate $\alpha_0$. The red horizontal lines are drawn at height $\alpha=2.7$.[]{data-label="fig:csReturnsGr"}](parfitoilReturns2013GrNew.pdf "fig:"){width="2.in" height="1.5in"} ![Hillish plots for generalized polar coordinates of points in [*oilReturns2013Gr*]{}.[]{data-label="fig:HillisPosRet"}](HillishOilreturns2013GrNew.pdf){width=".6\textwidth"} More evidence for existence of HRV in the region corresponding to [ *oilReturns2013Gr*]{} is provided by converting the data points in this region using the generalized polar coordinates suggested by . This produces the Hillish plots given in Figure \[fig:HillisPosRet\]. Both Figures \[fig:csReturnsGr\] and \[fig:HillisPosRet\] are encouraging. Conclusions {#sec:conc} =========== Whenever the limit measure of multivariate regular variation concentrates on a cone smaller than the full state space, there is the potential for seeking hidden regular variation. This idea has been most often applied to the case of asymptotic independence where the limit measure concentrates on the axes. Here we have shown the idea is also applicable when the limit measure concentrates on the diagonal or a narrow cone such as $[\smallwedge]$. Without hidden regular variation, asymptotic independence causes analysts to miss risk contagion. Analogously, when the limit measure concentrates on the diagonal, analysis would estimate the probability of a risk region $\{(x,y): y-x>4\}$ to be zero when in fact, hidden regular variation would yield a small but non-zero probability. [Our data analyses show potential for such estimation in strongly dependent data with heavy-tailed marginal distributions.]{} Without doubt, much work remains to be done on implementation. Both our network data which is node based and our returns data is nothing like independent replicated data. [Also, our methods for estimating the support of the angular measure $S(\cdot)$ are primitive at best. Higher dimensional examples present increased visualization and estimation difficulties. ]{} None-the-less, we believe the worked out examples are useful and illustrate practical cases. Other examples exist and in particular we have analyzed Microsoft vs Dell returns with results similar to those found in Section \[subsec:oil\]. Acknowledgements ================ We acknowledge with thanks the contribution in fall 2014 of Amy Willis who skilfully analyzed many financial data sets seeking examples of asymptotic full and strong asymptotic dependence. We are also grateful to Paul Embrechts who read an early draft and had many useful and encouraging comments. Two referees made many helpful and insightful comments. B. Das was supported by MOE-2013-T2-1-158 and IDG31300110. B. Das also acknowledges hospitality from Cornell University during visits in June 2015 and January 2016. S. Resnick was supported by Army MURI grant W911NF-12-1-0385 to Cornell University. [100]{} P.L. Anderson and M.M. Meerschaert. , 34(9):2271–2280, 1998. J. Beirlant, P. Vynckier, and J. Teugels. Tail index estimation, [P]{}areto quantile plots, and regression diagnostics. , 91(436):1659–1667, 1996. N. H. Bingham, C. M. Goldie, and J. L. Teugels. , volume 27 of [*Encyclopedia of Mathematics and its Applications*]{}. Cambridge University Press, Cambridge, 1989. B. Bollobás, C. Borgs, J. Chayes, and O. Riordan. Directed scale-free graphs. In [*Proceedings of the Fourteenth Annual ACM-SIAM Symposium on Discrete Algorithms (Baltimore, 2003)*]{}, pages 132–139, New York, 2003. ACM. M. Crovella, A. Bestavros, and M.S. Taqqu. Heavy-tailed probability distributions in the world wide web. In M.S. Taqqu R. Adler, R. Feldman, editor, [*A Practical Guide to Heavy Tails: Statistical Techniques for Analysing Heavy Tailed Distributions*]{}. Birkh[ä]{}user, Boston, 1999. G. Csardi and T. Nepusz. The igraph software package for complex network research. , 1695(5):1–9, 2006. B. Das, P. Embrechts, and V. Fasen. Four theorems and a financial crisis. , 54(6):701–716, 2013. B. Das, A. Mitra, and S.I. Resnick. Living on the multidimensional edge: seeking hidden risks using regular variation. , 45(1):139–163, 2013. B. Das and S.I. Resnick. Conditioning on an extreme component: Model consistency with regular variation on cones. , 17(1):226–252, 2011. B. Das and S.I. Resnick. Detecting a conditional extreme value model. , 14(1):29–61, 2011. B. Das and S.I. Resnick. Models with hidden regular variation: generation and detection. , 5:195–238 (electronic), 2015. <http://www.i-journals.org/ssy/viewarticle.php?id=141>. L. de Haan and A. Ferreira. . Springer-Verlag, New York, 2006. H. Drees, L. [de Haan]{}, and S.I. Resnick. How to make a [H]{}ill plot. , 28(1):254–274, 2000. R.T. Durrett. . Cambridge Series in Statistical and Probabilistic Mathematics. Cambridge University Press, Cambridge, 2010. P. Embrechts, C. Klüppelberg, and T. Mikosch. . Springer-Verlag, Berlin, 1997. J.E. Heffernan and S.I. Resnick. Hidden regular variation and the rank transform. , 37(2):393–414, 2005. J.E. Heffernan and S.I. Resnick. Limit laws for random vectors with an extreme component. , 17(2):537–571, 2007. J.E. Heffernan and J.A. Tawn. A conditional approach for multivariate extreme values (with discussion). , 66(3):497–546, 2004. H. Hult and F. Lindskog. On regular variation for infinitely divisible random vectors and additive processes. , 38(1):134–148, 2006. H. Hult and F. Lindskog. Regular variation for measures on metric spaces. , 80(94):121–140, 2006. R. Ibragimov, D. Jaffee, and J. Walden. Diversification disasters. , 99(2):333–348, 2011. M. Kratz and S.I. Resnick. The qq–estimator and heavy tails. , 12:699–724, 1996. F. Lindskog, S.I. Resnick, and J. Roy. Regularly varying measures on metric spaces: hidden regular variation and hidden jumps. , 11:270–314, 2014. S.I. Resnick. Hidden regular variation, second order regular variation and asymptotic independence. , 5(4):303–336, 2002. S.I. Resnick. . Springer Series in Operations Research and Financial Engineering. Springer-Verlag, New York, 2007. S.I. Resnick. . Springer Series in Operations Research and Financial Engineering. Springer, New York, 2008. Reprint of the 1987 original. S.I. Resnick. Multivariate regular variation on cones: application to extreme values, hidden regular variation and conditioned limit laws. , 80:269–298, 2008. S.I. [Resnick]{} and G. [Samorodnitsky]{}. Tauberian theory for multivariate regularly varying distributions with application to preferential attachment networks. , 18(3):349–367, 2015. S.I. Resnick and C. Stărică. Smoothing the [H]{}ill estimator. , 29:271–293, 1997. G. [Samorodnitsky]{}, S. [Resnick]{}, D. [Towsley]{}, R. [Davis]{}, A. [Willis]{}, and P. [Wan]{}. . , 53(1):146–161, March 2016. <http://arxiv.org/pdf/1405.4882.pdf>. R.L. Smith. Statistics of extremes, with applications in environment, insurance and finance. In B. Finkenstadt and H. Rootzén, editors, [*SemStat: Seminaire Europeen de Statistique, Exteme Values in Finance, Telecommunications, and the Environment*]{}, pages 1–78. Chapman-Hall, London, 2003. B. Viswanath, A. Mislove, M. Cha, and K.P. Gummadi. On the evolution of user interaction in facebook. In [*Proceedings of the 2nd ACM SIGCOMM Workshop on Social Networks (WOSN’09)*]{}, August 2009.

ArXiv

--- abstract: 'Here we report small-angle neutron scattering (SANS) measurements and theoretical modeling of U$_3$Al$_2$Ge$_3$. Analysis of the SANS data reveals a phase transition to sinusoidally modulated magnetic order, at $T_{\mathrm{N}}=63$ K to be second order, and a first order phase transition to ferromagnetic order at $T_{\mathrm{c}}=48$ K. Within the sinusoidally modulated magnetic phase ($T_{\mathrm{c}} < T < T_{\mathrm{N}}$), we uncover a dramatic change, by a factor of three, in the ordering wave-vector as a function of temperature. These observations all indicate that U$_3$Al$_2$Ge$_3$ is a close realization of the three-dimensional Axial Next-Nearest-Neighbor Ising model, a prototypical framework for describing commensurate to incommensurate phase transitions in frustrated magnets.' author: - 'David M. Fobes' - 'Shi-Zeng Lin' - 'Nirmal J. Ghimire' - 'Eric D. Bauer' - 'Joe D. Thompson' - Markus Bleuel - 'Lisa M. DeBeer-Schmitt' - Marc Janoschek title: 'Realization of the Axial Next-Nearest-Neighbor Ising model in U$_3$Al$_2$Ge$_3$' --- Introduction ============ The Axial Next-Nearest-Neighbor Ising (ANNNI) model is a historical framework which has successfully described frustrated magnetism in a variety of materials, and in particular, commensurate to incommensurate phase transitions.[@bak_commensurate_1982; @selke_annni_1988] The three-dimensional ANNNI model, describing the competing nearest neighbor (NN) interaction $J_1$, and the next nearest neighbor (NNN) interaction $J_2$, between Ising moments in one direction and simple ferromagnetic (FM) interactions between Ising moments in the same plane, was first proposed by R. J. Elliott in 1961,[@PhysRev.124.346] and has been studied extensively ever since. Subsequently, the ANNNI model was generalized to lower dimensions, and to $XY$ or Heisenberg spins.[@selke_annni_1988; @yamashita_1998; @harris_1990] This model and its variants have been shown to be relevant to a broad class of systems, including alloys, magnets, ferroelectrics and adsorbates.[@bak_commensurate_1982; @selke_annni_1988] For the case of magnetic materials, compounds containing rare-earth or actinide ions with $4f$ and $5f$ electrons, respectively, are expected to feature ANNNI physics due to the presence of Ruderman-Kittel-Kasuya-Yosida (RKKY) interactions, which can be approximated in many cases by the competing NN and NNN exchange integrals.[@Coleman2010] $f$-electron materials, where the Fermi-surface-topology-determined RKKY interaction also competes with crystal field effects and spin orbit coupling, contain rich phase diagrams with a multitude of magnetically ordered states. For example, in the presence of a quasi-nested Fermi surface, the RKKY interaction becomes maximal at a nonzero wave vector $\mathbf{Q}$, resulting in a magnetic spiral ground state, with a period $2\pi/Q$, typically incommensurate with the underlying chemical lattice.[@jensen_1991; @das_2015; @fobes_jpcm_2017] With the addition of uniaxial magnetic anisotropy, the magnetic spiral can become distorted, resulting in higher harmonic wave vectors, and potential quasi-continuous changes to the ordering wave vector $\mathbf{Q}$ as a function of temperature, arising due to the competition of RKKY interactions and thermal fluctuations.[@zaliznyak_1995] Although the ANNNI model provides a theoretical framework to describe this succession of long-period incommensurate phases as function of temperature, it was found that close fulfillment of its predictions for the temperature dependence of $\mathbf{Q}$ is rare. Maybe the best known realization of the ANNNI model in $f$-electron materials is the semimetal CeSb,[@fischer_1978; @meier_1978; @pokrovskii_1982] with the caveat that additional further-neighbor interactions were found to be important.[@meier_1978; @pokrovskii_1982] In this work, we demonstrate that the tetragonal $f$-electron compound U$_3$Al$_2$Ge$_3$ is a close realization of the three-dimensional ANNNI model. Hallmarks of the ANNNI model include (i) a magnetic phase which features a temperature-dependent magnetic ordering wave vector, owing to a competition between the NN and NNN interactions and magnetic fluctuations, in the vicinity of the phase transition to the paramagnetic (PM) phase, between which lies a second order phase transition, and (ii) a FM phase at low temperature that is entered via a first order transition, at which the magnetic ordering wave vector features a logarithmic singularity.[@selke_annni_1988] A previous neutron powder diffraction study suggests that U$_3$Al$_2$Ge$_3$ features several of these ingredients, including a low temperature FM phase below $T_{\mathrm{c}}=48~\mathrm{K}$, an incommensurately sinusoidally modulated magnetic state ($T_{\mathrm{c}} < T < T_{\mathrm{N}}$, with $T_{\mathrm{N}}=63~\mathrm{K}$) with ordering wave vector $\mathbf{Q}=[0,0,\delta]$, and a second-order transition between said phase and a PM phase, making this material a potential candidate for ANNNI realization.[@rogl_magnetic_1999] However, to confirm if U$_3$Al$_2$Ge$_3$ is indeed well-characterized by the ANNNI model several key aspects need to be clarified. (a) Although the incommensurate magnetic phase was reported to be sinusoidally modulated, symmetry, in principle, also allows for spiral magnetic order, which would not agree with the ANNNI model. Notably, powder diffraction can frequently not unambiguously distinguish between a sinusoidal and spiral magnetic order. (b) Previous to our study, no temperature dependence of the magnetic ordering wave vector has been reported. (c) The nature of the transition between the FM and incommensurate phases remains unknown. Here, to address these issues, we have carried out a detailed neutron scattering study on a single crystal of U$_3$Al$_2$Ge$_3$ in combination with theoretical modeling based on the ANNNI model. Using small-angle neutron scattering (SANS), a powerful technique for accurately studying long-wavelength magnetic structures, we have confirmed that $\mathbf{Q}=[0,0,\delta]$ indeed shows a pronounced temperature dependence, and determined the nature of the magnetic phase transitions. Our results clearly reveal a second order phase transition from paramagnetic to sinusoidally modulated Ising moments at $T_{\mathrm{N}}=63$ K, followed by a first-order phase transition at $T_{\mathrm{c}}=48$ K to a FM state. Through theoretical modeling we are able to ascribe these experimental observations to the ANNNI framework; both the measured and calculated phase diagram can be understood in terms of the 3D ANNNI lattice model, with frustrated interactions in the direction of modulation, given by the ratio $J_2/J_1=-0.2815$, where $J_1$ is the nearest neighbor ferromagnetic interaction, $J_2$ is the next nearest neighbor antiferromagnetic interaction,[@selke_annni_1988] and the ratio $J_2/J_1$ is given by the $Q$ vector at $T_{\mathrm{N}}$, *i.e.* $\cos(\delta c/ 2\pi)=-J_1/4J_2$, in which $c$ is the $c$-axis lattice parameter. Furthermore, at the incommensurate-to-FM phase transition we confirm the presence of a logarithmic singularity in the temperature dependence of $\mathbf{Q}$, an essential feature of the three-dimensional ANNNI model, as stated above. Therefore, we find that U$_3$Al$_2$Ge$_3$ is a close realization of ANNNI model, offering an important playground for investigating its rich physics. The manuscript is organized in the following way: In Section \[model\] we describe the theoretical model that motivated our experimental study. In Sections \[synthesis\] and \[results\] we report the sample synthesis and characterization, and our SANS results, respectively. Finally, in Section \[discussion\] we will discuss and summarize our combined theoretical and experimental results. Model ===== To motivate this experimental study we develop the following model accounting for the experimentally observed magnetic phase diagram [@rogl_magnetic_1999], making several experimental predictions. The neutron powder diffraction results[@rogl_magnetic_1999] determined that in U$_3$Al$_2$Ge$_3$ the magnetic moments lying on the Uranium $8c$ and $2a_1$ symmetry sites (but not the $2a_2$ sites), with magnetic moments $\mu_U=2.37\ \mu_B$ and $\mu_U=2.12\ \mu_B$, respectively, align parallel to the $a$-axis due to strong easy-axis anisotropy, modulate along the $c$-axis, and order ferromagnetically in the $ab$-plane, allowing us to reduce the problem to one dimension. Due to the small ordering wave vector, we expand the magnetic free energy in terms of the magnetization density $\mathbf{M}$ and ordering wave vector $\mathbf{Q}$ in the continuum limit, $$\begin{aligned} \label{eq1} \begin{split} {\cal F}({\mathbf{M}}) = \frac{\alpha }{2}\mathbf{M}^2 + \frac{\beta }{4}\mathbf{M}^4 + \frac{\delta }{6}\mathbf{M}^6 - \frac{\mu }{2}{(\partial_z {\mathbf{M}})^2} \\+ \frac{\eta }{2}{({\partial_z ^2}{\mathbf{M}})^2} + \frac{\gamma }{2}\mathbf{M}^2{(\partial_z {\mathbf{M}})^2} - \frac{{{A_2}}}{2}M_x^2, \end{split}\end{aligned}$$ where $A_2$ is the easy-axis anisotropy, and $\mu$, $\eta$ and $\gamma$ are associated with the stiffness of the magnetic modulation. The competing interactions $J_1$ and $J_2$ along the $c$-axis in the ANNNI lattice model are captured by the $\mu>0$ and $\eta>0$ terms. Approaching the Néel temperature $T_{\mathrm{N}}$ from the paramagnetic phase, we can neglect the quartic and higher order terms of $\mathbf{M}$. To take advantage of the anisotropy, the system stabilizes a state with sinusoidal moment modulation, $M_x\propto\sin(Q r)$ and $M_y=M_z=0$. We obtain the optimal $Q$ by minimizing $\mathcal{F}$ with respect to $Q$, which yields $Q_o=\sqrt{\mu/2\eta}$ at $T_{\mathrm{N}}$. Here $T_{\mathrm{N}}$ is determined by the condition that the coefficient of the quadratic term in $M_x$ vanishes $$\begin{aligned} {\alpha } - {{{A_2}}} - \frac{{{\mu ^2}}}{{4\eta }} = 0,\end{aligned}$$ which is enhanced due to the presence of the easy-axis anisotropy $A_2$. To fulfill the condition for second order phase transition between incommensurate and paramagnetic phase at $T_{\mathrm{N}}$, we assume $\beta>0$. Lowering the temperature results in an increase of the magnetic moment. Higher-order terms and easy-axis anisotropy both tend to distort the simple sinusoidal modulation. The wave is squared up as moments are forced to align along the easy-axis, creating harmonics in the wave vectors. To take the advantage of the anisotropy, the moments lie in the easy-axis direction. The magnetic state can be described by an elliptic function $$\begin{aligned} \label{eq3} {M_x} = \Delta\ \mathrm{sn}({z}/{\xi },k),\ \ M_y=M_z=0,\end{aligned}$$ where $\mathrm{sn}(z)$ is the Jacobi elliptic function. The meanings of $\xi$, $k$, $\Delta$ become clear in the limit $k\rightarrow 0$, $\mathrm{sn}(z, k=0)=\sin(z)$: $k$ describes the deviation from a perfect sinusoidal modulation, thus reflecting the importance of harmonics for the magnetic modulation, $\xi$ is a length scale appearing in the period of the magnetic modulation, and $\Delta$ is the amplitude of the modulation. When the coefficients in Eq. satisfy a certain relation (shown below), the magnetic state can be found exactly using Eq. . For generic parameters, we can use variational calculations using Eq. to determine $\xi$, $k$ and $\Delta$. ![(color online) Temperature dependence of (a) free energy $\mathcal{F}$, and (b) wave vector $Q$ for model in Eq. . Near the commensurate-incommensurate transition, $Q$ follows Eq. . The magnetic order sets in at $\tau=1/4$. We have used $\beta\gamma/\delta\mu=-10/3$ in the calculations. []{data-label="f3"}](fig1.pdf){width="1\columnwidth"} Let us first construct the exact solution.[@buzdin_1997] We note that $M_x$ in Eq. solves the following equations $$\begin{aligned} \label{eq4} {\xi ^2}\partial _z^2{M}_x + (1 + {k^2}){M}_x - 2\frac{{{k^2}}}{{{\Delta ^2}}}M_x^3 = 0,\end{aligned}$$ $$\begin{aligned} \label{eq5} {\xi ^2}{({\partial _z}{M_x})^2} + (1 + {k^2})M_x^2 - \frac{{{k^2}M_x^4}}{{{\Delta ^2}}} = {\Delta ^2}.\end{aligned}$$ Differentiating Eq. with respect to $z$ twice, we obtain $$\begin{aligned} \label{eq6} \begin{split} \partial _z^4{M_x} + \frac{{(1 + {k^2})}}{{{\xi ^2}}}\partial _z^2{M_x} + \frac{{{12k^4}}}{{{\Delta ^4}{\xi ^4}}}M_x^5 - \frac{{6{k^2}(1 + {k^2})}}{{{\Delta ^2}{\xi ^4}}}M_x^3 \\ - 12\frac{{{k^2}}}{{{\Delta ^2}{\xi ^2}}}[{M_x}{({\partial _z}{M_x})^2} + M_x^2\partial _z^2{M_x}] = 0. \end{split}\end{aligned}$$ Multiplying Eq. by $M_x^2$ and Eq. by $M_x$ and then adding them together, we have $$\begin{aligned} \label{eq7} \begin{split} \frac{{{k^2}}}{{{\Delta ^2}{\xi ^2}}}\left[M_x^2\partial _z^2{M_x} + {M_x}{({\partial _z}{M_x})^2}\right] + \frac{{2{k^2}(1 + {k^2})}}{{{\Delta ^2}{\xi ^4}}}M_x^3 \\ - \frac{{{3k^4}}}{{{\Delta ^4}{\xi ^4}}}M_x^5 - \frac{{{k^2}}}{{{\xi ^4}}}{M_x} = 0. \end{split}\end{aligned}$$ We multiply Eq. by a factor $A$ and Eq. by a factor $B(1+k^2)/\xi^4$ and add to Eq. . We then compare the resulting equation to Eq. , from which we obtain the equations for coefficients $$\begin{aligned} \label{eq8} \begin{split} (12 - A)\frac{{{k^2}}}{{{\Delta ^2}{\xi ^2}}} = \frac{\gamma }{\eta }, \\ (B + 1)\frac{{(1 + {k^2})}}{{{\xi ^2}}} = \frac{\mu }{\eta }, \\ \frac{{(12 - 3A){k^4}}}{{{\Delta ^4}{\xi ^4}}} = \frac{\delta }{\eta }, \\ \frac{{{k^2}(1 + {k^2})(2A - 6 - 2B)}}{{{\Delta ^2}{\xi ^4}}} = \frac{\beta }{\eta },\\ \frac{{{{(1 + {k^2})}^2}B - A{k^2}}}{{{\xi ^4}}} = \frac{{\alpha '}}{\eta }, \end{split}\end{aligned}$$ with $\alpha'=\alpha-A_2$. In principle we can determine $A$, $B$, $k$, $\Delta$ and $\xi$ from Eq. ; however, the solutions are not guaranteed to exist for arbitrarily coefficients of $\mathcal{F}$, as can been seen from the equations with $\gamma$ and $\delta$. This generally implies that Eq. is not an exact solution. We consider the case of a particular set of coefficients, such that Eq. is the exact solution, to demonstrate explicitly the dependence of $M_x$ on temperature. In any case, Eq. should be a good variation ansatz to describe the spatial profile of $M_x$. At $T_{\mathrm{N}}$ corresponding to $\alpha'=\mu^2/4\eta$, the modulation of $M_x$ should be a sinusoidal wave. We have $\Delta=0$ and $k=0$ while $\Delta_k=\Delta/k=\mathrm{constant}$, in the limit $k\rightarrow 0$. We obtain $B=1$, $\xi^2=2\eta/\mu$, $\Delta_k=-3\beta/2\delta$. As in the usual Ginzburg-Landau theory, we assume that the temperature dependence can be modeled in $\alpha'$ by introducing $\tau = \alpha '/(\frac{{{\mu ^2}}}{\eta })$, and other coefficients do not depend on $T$. Here $\tau$ is the reduced temperature. Upon a change in temperature, $k$ and $B$ change continuously, but both $A$ and $\Delta_k\xi$ are independent of $T$. We have $$B(\tau ,k) = - \frac{{ - 1 + 2\tau + \sqrt {1 - 4[1 + (12 + \frac{{3\beta \gamma }}{{\delta \mu }}){k^2}{{(1 + {k^2})}^{ - 2}}]\tau } }}{{2\tau }}.$$ $$\xi = \sqrt {(B + 1)\frac{{(1 + {k^2})\eta }}{\mu }},\ \ {\Delta _k} = \sqrt {\frac{-3\beta\eta}{\delta(B+1)(1+k^2)}}$$ The corresponding free energy density $$\begin{aligned} \label{eq9} \frac{{{{\cal F}}}}{{{{\cal F}_0}}} = \frac{{ - {k^4}\left[10{I_2} - ({k^2} + 1)(11 + B){I_4} + 14{k^2}{I_6}\right]}}{{{{[({k^2} + 1)(B + 1)]}^3}}},\end{aligned}$$ where $$\begin{aligned} \mathcal{F}_0=\frac{{-3\beta\mu^2 }}{{2\delta {\eta}}},\ \ \ {I_2} = \frac{1}{{{k^2}}}\left[1 - \frac{{E(k)}}{{K(k)}}\right],\end{aligned}$$ $$\begin{aligned} {I_4} = \frac{{2 + {k^2}}}{{3{k^4}}} - \frac{{2(1 + {k^2})}}{{3{k^4}}}\frac{{E(k)}}{{K(k)}},\end{aligned}$$ $$\begin{aligned} {I_6} = \frac{{4({k^2} + 1)}}{{5{k^2}}}{I_4} - \frac{3}{{5{k^4}}}\left[1 - \frac{{E(k)}}{{K(k)}}\right],\end{aligned}$$ with $E(k)$ and $K(k)$ being the complete elliptic integral of the second and first kind, respectively. We have used the convention $E(k)\equiv\int_0^{\pi/2}\sqrt{1-k^2\sin^2\theta}d\theta$. To determine the optimal $k$, we numerically minimize the free energy density $\mathcal{F}$, given by Eq. , with respect to $k$, and then obtain other parameters from $k$. We then obtain $\xi$. The period of the sinusoidal wave is given by the expression $$\begin{aligned} \label{eq14} {\lambda _I} = 4\xi K(k). \end{aligned}$$ Here $\lambda_I$ diverges when $k\rightarrow 1$, corresponding to the FM state. Near the phase transition between sinusoidal wave and FM at $\tau_2$, we can expand $K(k)$ near $k=1$, $$\begin{aligned} \label{eq15} {\lambda _I}=\xi(\tau_2)\left[8\ln 2-2 \ln(1-k^2)\right],\end{aligned}$$ Close to $\tau_2$ from above, we expand $1-k^2=\alpha_2(\tau-\tau_2)^{\beta_2}$ and then obtain the logarithmic temperature dependence of the ordering wave vector $$\begin{aligned} \label{eq15cc} Q=2\pi/\lambda_I\propto -1/\ln(\tau-\tau_2),\end{aligned}$$ near the commensurate-incommensurate (sinusoidal modulation-FM) transition. [@bak_commensurate_1982; @selke_annni_1988] The free energy $\mathcal{F}$ and $Q$ as a function of temperature $\tau$ are presented in Fig. \[f3\]. Upon lowering the temperature from $T_{\mathrm{N}}$, the system first becomes an Ising density wave with $Q$ varying continuously with $T$, becoming FM at low temperature, with the free energy density $$\begin{aligned} \frac{{{{\cal F}_{FM}}}}{{{{\cal F}_0}}} = - \frac{1}{{18}}\left[(1 + \sqrt {1 - 6\tau } )\left(\frac{1}{3} - 2\tau \right) - \tau \right].\end{aligned}$$ The transition from the paramagnetic phase into the sinusoidally-modulated phase is of second order by construction, following the experiments shown below and Ref. , and the transition from the sinusoidally-modulated phase to the FM is shown to be first order, which can be seen from the slope of $\mathcal{F}$ in Fig. \[f3\] (a). As we will show, these results agree qualitatively with results of the experimental analysis of the SANS data (cf. Fig. \[f2\](b)). Synthesis and Characterization {#synthesis} ============================== To confirm the accuracy and predictions of our model, a single crystal of U$_3$Al$_2$Ge$_3$ (tetragonal structure $I4$ (No. 79 in the International Tables for Crystallography), $a=7.769~\AA$, $c=11.036~\AA$, cf. Fig. \[f1\](a), derivative of ordered antitype-Cr$_5$B$_3$) was prepared by the Czochralski technique in a tri-arc furnace with a continuously purified Argon atmosphere ($< 10^{-12}~\mathrm{ppm}~\mathrm{O}_2$). The sample was characterized using X-ray Laue backscattering and via magnetic susceptibility measurements performed in a Quantum Design Magnetic Property Measurement System (MPMS). As shown in Fig. \[f0\], magnetic susceptibility data, both taken on warming after zero-field cooling, with no changes between runs, reveal typical ferromagnetic behavior with $T_{\mathrm{c}}=48~\mathrm{K}$, with a fully saturated susceptibility below 30 K, The susceptibility also exhibits an additional anomaly at 63 K corresponding to the onset temperature of the incommensurate magnetic state.[@rogl_magnetic_1999] For the small-angle neutron scattering (SANS) measurements, a 1 g piece of the single crystal was orientated such that \[100\] was along the beam and \[001\] was in the scattering plane, allowing to access the magnetic ordering wave vector $\mathbf{Q} = [0,0,\delta]$. SANS measurements were performed at the NG7-SANS beam line at the NIST Center for Neutron Research and the GP-SANS beam line at the High-Flux Isotope Reactor (HFIR) at Oak Ridge National Laboratory. SANS data were collected by rocking the sample $\pm10^{\circ}$ about the vertical axis, with an incident neutron wavelength of $\lambda=6~\AA$. ![(Color Online) Magnetic susceptibility $\chi$ of a single crystal of U$_3$Al$_2$Ge$_3$ taken with $H=10$ Oe. Vertical lines indicate second- and first-order transitions at $T_{\mathrm{N}}=63~\mathrm{K}$ and $T_{\mathrm{c}}=48~\mathrm{K}$, respectively. Inset: Zoomed region of magnetic susceptibility $\chi$ showing the second-order transition at $T_{\mathrm{N}}=63~\mathrm{K}$. Different symbols represent separate runs on the same sample. []{data-label="f0"}](fig2.pdf){width="1\columnwidth"} ![(Color Online) (a) U$_3$Al$_2$Ge$_3$ tetragonal crystal structure and the magnetic configuration in the ferromagnetic phase, with the three U site symmetries individually labeled ($8c$, $2a_1$, $2a_2$). (b) Low-$q$ scattering in the ferromagnetic phase ($T=40$ K), fitted to Porod’s Law $I \propto q^{-4}$, illustrating the expected behavior for FM domains. (c) Polarized SANS cross-sections with the incident beam polarized parallel ($\sigma_+$) and antiparallel ($\sigma_-$) to the momentum transfer $q$, and their difference, taken at NG7-SANS. As explained in the text, the difference should only be non-zero in the case of a spiral magnetic structure. (d) Radial cuts of SANS data showing the evolution of the ordering wave vector as a function of temperature taken at GP-SANS. Data in (b) and (c) are normalized to monitor counts, and data in (d) are scaled and vertically shifted for clarity. []{data-label="f1"}](fig3.pdf){width="0.99\columnwidth"} Experimental Results {#results} ==================== In the following section we present the results of the SANS measurements. All SANS data shown were obtained by angle integrating monitor-normalized SANS detector intensities. In order to obtain the magnetic intensity as a function of momentum transfer $q$, the magnetic scattering was azimuthally averaged in the FM phase, whereas in the incommensurate phase radial cuts through the two magnetic satellite positions at $\mathbf{Q}=[0,0,\delta]$ were performed. Starting in the low-temperature FM phase, $T < 48$ K we only observe strong scattering near the direct beam and no characteristic magnetic ordering wave vectors, confirming the FM state. Furthermore, scattering near the direct beam approximately follows Porod’s Law $I\propto S q^{-4}$,[@porod_1951a; @porod_1951b] as shown in Fig. \[f1\](b), which indicates the presence of large ferromagnetic domains with specific surface area $S = 0.251(7)~\mathrm{nm}^{2}$, from least-squares fitting. Increasing temperature above 48 K we observe the emergence of strong magnetic scattering at $\mathbf{Q}=[0,0,\delta]$, indicating the entrance of the incommensurate phase. Although the original neutron powder diffraction study indicated this phase to be a sinusoidally modulated magnetic phase, from representational analysis of the crystallographic lattice structure determined in Ref. , as shown in Fig. \[f1\](a), and the magnetic ordering wave vector $\mathbf{Q}=[0,0,\delta]$, both helical and sinusoidal magnetic ordering are allowed by symmetry. To confirm the magnetic structure, we therefore performed spin-polarized SANS measurements at the NG7-SANS instrument at $T=56$ K, with the polarization pointed along the \[001\] direction, and the detector positioned 4 m from the sample. In the case where the incident polarization is aligned along $q$, *i.e.* $P_0 \parallel q$, the magnetic cross-section is given by $$\begin{aligned} \label{magcrosssec} {\sigma_{\pm}}&={\vert\bm{M}_{\perp}\vert^2\pm P_{0}C}. $$ Here $\bm{M}_{\perp}$ is the magnetic interaction vector defined as $\mathbf{M_{\perp}}=\hat{\mathbf{Q}}\times(\mathbf{\rho}(\mathbf{Q})\times\hat{\mathbf{Q}})$, where $\bf{\rho}(\mathbf{Q})=-2\mu_B\int\mathbf{\rho}(\mathbf{r})\exp(i\bf{Q}\cdot\mathbf{r})d\mathbf{r}$ is the Fourier transform of the magnetization density $\mathbf{\rho}(\bf{r})$ of the investigated sample, and $\hat{\mathbf{Q}}$ is a unit vector parallel to the scattering vector $\mathbf{Q}$. The associated coordinate frame is defined to have $x$ parallel to $\mathbf{Q}$, $z$ perpendicular to the scattering plane and $y$ completing the right-handed set. We note that the term $C=2\Im(M_{\perp y}^\ast\cdot M_{\perp z})$ is only non-zero for magnetic structures that display chirality, such as magnetic spirals, and is therefore denoted as the chiral term.[@janoschek_2007] Because, $(\sigma_{+} - \sigma_{-}) / P_0 = 2C$, we would expect that for a *single* peak the difference between the cross section obtained with incident neutrons polarized parallel ($\sigma_{+}$) and antiparallel ($\sigma_{-}$) with respect to $q$ would give rise to a non-zero chiral term $C$, only if the magnetic structure had spiral order. This is contrary to our observation, where such a subtraction results in no residual intensity, within the error bars (cf. Fig. \[f1\](c)), confirming that the magnetic structure is sinusoidal, as surmised originally in Ref. . Further increase in temperature reveals a pronounced change, by a factor of three, in the magnetic ordering wave vector, as illustrated by the select SANS data for $49~\mathrm{K} \leq T \leq 67~\mathrm{K}$ shown in Fig. \[f1\](d), which have been scaled and shifted to illustrate the drastic change in the magnetic ordering wave vector as a function of temperature. In Fig. \[f2\] we display the results of analysis of the angle-integrated SANS data for each temperature, taken at the GP-SANS instrument. For the temperature-dependent analysis we use data measured in two configurations: (1) with the detector at 7 m from the sample, which provides a momentum range of $0.1 \lesssim q \lesssim 0.8~\mathrm{nm}^{-1}$, and (2) with the detector 17 m from the sample, which provides a momentum range of $0.025 \lesssim q \lesssim 0.3~\mathrm{nm}^{-1}$. In the 7 m configuration, we can clearly see that the propagation vector $Q$ changes abruptly as a function of temperature at $T_{\mathrm{c}}\sim48$ K, while at 63 K the $Q$ saturates, as shown by the circle symbols in Fig. \[f2\](b). The 17 m detector configuration, due to its lower momentum range which allows us to observe scattering that would have been lost in the direct beam signal in the 7 m configuration, reveals that there is indeed a quasi-discontinuous change in the wave vector at $T_{\mathrm{c}}$. It is noteworthy that the intensity continues to increase slightly above the transition. This situation may arise in the case where the data were taken upon warming, as we have done here; since the transition from the FM to the modulated phase is first order, we may expect some hysteresis, *i.e.* it is possible for “droplets” of the FM phase to persist above $T_{\mathrm{c}}$.[@PhysRevB.87.134407] This would therefore result in the appearance of the satellite peaks associated with the modulated phase immediately above $T_{\mathrm{c}}$, but with reduced integrated intensities. As the FM droplets shrink, the intensity of the satellite peaks will gradually increase, reaching a maximum at the temperature where the entire system is within the modulated phase, here $T\sim50$ K. It should also be noted here that in the 17 m configuration, the ordering wave vector of the sinusoidal phase lies at the limits of the detector range, resulting in substantial error in the intensity, as illustrated by the error bars in Fig. \[f2\](b). Furthermore, we find that the wave vector is well described by the logarithmic function $-1/\ln(T-T_{\mathrm{c}})$ (cf. Eq. \[eq15cc\]), as illustrated by the dashed line in Fig. \[f2\](b). This function directly originates from the ANNNI models, as we discussed above in Section \[model\]. This quasi-discontinuity in $Q$ indicates a first-order transition between the sinusoidal and FM phases, supporting the assertion of a first-order transition predicted by our model. In contrast, above $T_{\mathrm{N}}=63$ K we observe no discrete peaks on the detector, but instead a broad ring of scattering which gradually decreases in intensity up to $\sim$67 K, above which it disappears below the background completely. This behavior, along with the magnetization data in Fig. \[f0\], strongly supports a second-order transition between the paramagnetic (PM) and sinusoidal phase. . Different symbols represent unique configurations: blue circles were obtained using a 7m detector distance, the orange triangles using a 17m detector distance. Error bars represent the errors obtained from least-squares fitting. The red dashed line in panel (b) represents a fit to the logarithmic singularity expression, $Q \propto -1/\ln(T-T_{\mathrm{c}})$. []{data-label="f2"}](fig4.pdf){width="0.99\columnwidth"} Discussion and summary {#discussion} ====================== It is found that the lattice constants change in the sinusoidally modulated phase, indicating the existence of magneto-elastic coupling.[@rogl_magnetic_1999] Therefore, the measured change of $Q$ by neutron scattering has two contributions, with one being from the lattice expansion/shrinkage, and the other from the magnetic competing interaction. The former contribution is negligible compared to the latter one according to the experiments.[@rogl_magnetic_1999] One key feature of the ANNNI framework on a lattice model is the appearance of the devil’s staircase, where $Q$ varies quasi-continuously with temperature. In certain temperature window, $Q$ is fixed. Outside the temperature window, $Q$ then jumps to another $Q$ through the proliferation of magnetic solitons. [@PhysRevB.21.5297] This behavior is not captured by the model in the continuum limit in Eq. . For $Q=0.76~\mathrm{nm}^{-1}$, the $Q$ steps are extremely narrow. In experiments with finite resolution, $Q$ varies nearly continuously with temperature, as depicted in Fig. \[f2\]. True realizations of the ANNNI model in magnets is rare, particularly in metals. In fact many of the potential examples of ANNNI-like materials require notable modification to the ANNNI model to explain observed behavior. The best known example is semimetallic CeSb,[@fischer_1978] the magnetization of which features the famous “devil’s staircase”, discrete sharp steps in the magnetization within the incommensurate phase.[@bak_commensurate_1982] However, the ANNNI model notably does not capture additional experimentally observed Bragg reflections in CeSb, requiring a modification of the ANNNI model.[@meier_1978; @pokrovskii_1982] A later study of CeBi also required modification of the ANNNI model, *i.e.* an additional competing exchange coupling.[@uimin_1982] More recently, the phase diagram of metallic TmB$_4$, which features a fractional plateau in magnetization, has been described in terms of the ANNNI model, but again, two additional exchange terms were required as modification of the model.[@wierschem_2015] Several other modern examples of potential ANNNI compounds do exist, however. In U(Ru$_{1-x}$Rh$_x$)$_2$Si$_2$, powder neutron diffraction observed multiple $Q$ states as a function of temperature, suggested by the authors to perhaps be related to the ANNNI model.[@kawarazaki_1994] A recent study has reported a semimetallic compound closely related to CeSb, CeSbSe which was found to feature discrete steps in magnetization and resistivity, which have been suggested to arise due to underlying ANNNI type magnetic interactions and the associated existence of a devil’s staircase.[@PhysRevB.96.014421] The heavy fermion metal CeRhIn$_5$, which exhibits incommensurate helical and elliptical magnetic ordering that transition to a commensurate order with magnetic field,[@raymond_2007] has been suggested as a candidate.[@das_2015; @fobes_2017] In summary, we have studied the magnetic order in $f$-electron compound U$_3$Al$_2$Ge$_3$, finding it to be well described by the three-dimensional Axial Next-Nearest Neighbor Ising model. Our experimental results determine the transition between the paramagnetic and sinusoidal phases to be second order, and the lower transition between the sinusoidal and ferromagnetic phases to be first order, which arises naturally from our model. Furthermore, we have confirmed the logarithmic singularity in the sinusoidal ordering wave vector at the boundary to the FM phase, an essential feature of the ANNNI model. These results demonstrate that U$_3$Al$_2$Ge$_3$ represents a prototypical material for investigating ANNNI physics. Acknowledgements ================ Research at LANL was supported by the LANL Directed Research and Development program (neutron scattering) and the U.S. Department of Energy, Office of Basic Energy Sciences, Division of Materials Sciences and Engineering under the project ‘Complex Electronic Materials’ (material synthesis and characterization). A portion of this research used resources at the High Flux Isotope Reactor, a DOE Office of Science User Facility operated by the Oak Ridge National Laboratory. We acknowledge the support of the National Institute of Standards and Technology, U.S. Department of Commerce, in providing the neutron research facilities used in this work. Access to the polarized NG7SANS was provided by the Center for High Resolution Neutron Scattering, a partnership between the National Institute of Standards and Technology and the National Science Foundation under Agreement No. DMR-1508249. We thank Cedric Gagnon and Jeff Krzywon for technical support during the NG7SANS measurements. Data reduction and analysis was performed with the use of the GRASP and NeutronPy neutron scattering analysis software packages. [26]{}ifxundefined \[1\][ ifx[\#1]{} ]{}ifnum \[1\][ \#1firstoftwo secondoftwo ]{}ifx \[1\][ \#1firstoftwo secondoftwo ]{}““\#1””@noop \[0\][secondoftwo]{}sanitize@url \[0\][‘\ 12‘\$12 ‘&12‘\#12‘12‘\_12‘%12]{}@startlink\[1\]@endlink\[0\]@bib@innerbibempty [****,  ()](\doibase 10.1088/0034-4885/45/6/001) [****,  ()](\doibase 10.1016/0370-1573(88)90140-8) [****, ()](\doibase 10.1103/PhysRev.124.346) [****,  ()](\doibase 10.1143/JPSJ.67.198) [****,  ()](\doibase 10.1063/1.345838) [****,  ()](\doibase 10.1007/s10909-010-0213-4) @noop [**]{} (, ) [****,  ()](\doibase 10.1103/PhysRevLett.113.246403) [****,  ()](\doibase 10.1088/1361-648X/aa6696) @noop [****,  ()]{} [****,  ()](http://stacks.iop.org/0022-3719/11/i=2/a=018) [****,  ()](http://stacks.iop.org/0022-3719/11/i=6/a=023) @noop [****,  ()]{} [****,  ()](\doibase 10.1016/S0304-8853(98)00370-9) [****,  ()](\doibase 10.1016/S0375-9601(96)00894-8) @noop [****,  ()]{} @noop [****,  ()]{} [****,  ()](\doibase 10.1016/j.physb.2007.02.074) [****,  ()](\doibase 10.1103/PhysRevB.87.134407) [****, ()](\doibase 10.1103/PhysRevB.21.5297) [****,  ()](\doibase 10.1051/jphyslet:019820043018066500) [****,  ()](\doibase 10.1103/PhysRevB.92.214433) [****,  ()](\doibase 10.1143/JPSJ.63.716) [****,  ()](\doibase 10.1103/PhysRevB.96.014421) [****,  ()](http://stacks.iop.org/0953-8984/19/i=24/a=242204) @noop [ ()]{}

ArXiv

--- abstract: 'Hindman’s Theorem says that every finite coloring of the positive natural numbers has a monochromatic set of finite sums. Ramsey algebras, recently introduced, are structures that satisfy an analogue of Hindman’s Theorem. It is an open problem posed by Carlson whether every Ramsey algebra has an idempotent ultrafilter. This paper developes a general framework to study idempotent ultrafilters. Under certain countable setting, the main result roughly says that every nondegenerate Ramsey algebra has a nonprincipal idempotent ultrafilter in some nontrivial countable field of sets. This amounts to a positive result that addresses Carlson’s question in some way.' address: - 'The Ohio State University, Columbus, OH 43210 United States' - | School of Mathematical Sciences\ Universiti Sains Malaysia\ 11800 USM, Malaysia author: - Wen Chean Teh title: Ramsey Algebras and the Existence of Idempotent Ultrafilters --- introduction ============ The set of natural numbers $\{0,1,2,\dotsc\}$ is denoted by $\omega$. Suppose $\langle x_i \rangle_{i \in \omega}$ is a sequence of natural numbers. Let $\operatorname{FS}(\langle x_i \rangle_{i \in \omega} )$ denote the set $\{\, \sum_{i \in F} x_i \mid F \in\mathcal{P}_f(\omega)\backslash \{\emptyset\}\,\}$, where $\mathcal{P}_f(\omega)$ is the set of all finite subsets of $\omega$. Hindman’s Theorem [@nH74] says that for every finite partition of the set of positive natural numbers $\mathbb{N}=X_0\cupdot X_1\cupdot \dotsb \cupdot X_N$, there exists a sequence $\langle x_i \rangle_{i \in \omega}$ of positive natural numbers such that $\operatorname{FS}(\langle x_i \rangle_{i \in \omega} )\subseteq X_j$ for some $0\leq j \leq N$. A Ramsey algebra is an algebraic structure which possesses the property analogous to that possesed by the semigroup $(\omega, +)$ as in Hindman’s Theorem. The formal definition of Ramsey algebras was suggested by Carlson and recently introduced by this author [@wcT13; @wcT13a]. The notion of Ramsey algebras is motivated by the study of Ramsey spaces, introduced by Carlson [@tC88] in 1988. He showed that some space of infinite sequences of variable words over a finite alphabet, endowed with an analogous Ellentuck topology is a Ramsey space. This result, known as Carlson’s Theorem in [@HS12 XVIII, §4], implies many earlier Ramsey theoretic results including Hindman’s Theorem, Ellentuck’s Theorem, the dual Ellentuck Theorem [@CS84], the Galvin-Prikry Theorem [@GP73] and the Hales-Jewett Theorem [@HJ63]. Due to his abstract version of Ellentuck’s Theorem [@eE74], this space is Ramsey because the corresponding algebra of variable words is a Ramsey algebra. This connection between Ramsey algebras and Ramsey spaces was addressed in [@wcT13]. In 1975 Galvin and Glazer (see [@wC77] or [@nH79]) gave a simple proof of Hindman’s Theorem by showing the existence of idempotent ultrafilters. These are exactly the idempotent elements of the semigroup $(\beta\mathbb{N}, +)$, where $+$ is the extension of addition on $\mathbb{N}$ to $\beta\mathbb{N}$, the Stone-Čech compactification of $\mathbb{N}$. As $(\beta\mathbb{N}, +)$ is a compact right topological semigroup, by the Ellis-Numakura Lemma, it has an idempotent element. Later in 1987, under Martin’s Axiom, Hindman [@nH87] showed the existence of a strongly summable ultrafilter, that is, an idempotent ultrafilter $U$ with the stronger property that it is generated by sets of finite sums. Blass and Hindman [@BH87] later showed that their existence is independent of $\mathsf{ZFC}$. Carlson’s algebras of variable words are interesting Ramsey algebras that are not semigroups. The collection of operations in each of these algebras is finite but can be arbitrarily large depending on the size of the underlying finite alphabet. The nice interplay among the operations in a Ramsey algebra of variable words may provide insight into the problem of formation of new Ramsey algebras from semigroups or other elementary Ramsey algebras, which remains wide open. A key feature in Carlson’s proof is the construction of certain ultrafilters idempotent for every operation in the corresponding algebra. These ultrafilters in turn allow the construction of homogeneous sequences, generalizing Galvin-Glazer proof of Hindman’s Theorem. Furthermore, assuming Martin’s Axiom, this author [@wcT13b] has shown that every nondegenerate Ramsey algebra has nonprincipal strongly reductible ultrafilters, that is, analogues of strongly summable ultrafilters. This led Carlson to propose the following open problem. Can the existence of idempotent ultrafilters for a Ramsey algebra be proven in $\mathsf{ZFC}$? This paper is motivated by the work of reverse mathematicians. In this area, the logical strength of true statements that can be formalized using the language of second order arithmetic are determined. To do that, particularly, field of sets and ultrafilters have to be encoded as sets of natural numbers. Hence, countability is a major consideration. Therefore, a general framework is proposed so that the notion of idempotent ultrafilters can be dealt with in this countable setting. Our main result indicates that a corresponding positive assertion pertaining to Carlson’s problem may be provable under the theory of second order arithmetic. Preliminaries ============= To us an *algebra* is a pair $(A, \mathcal{F})$, where $A$ is a nonempty set and $\mathcal{F}$ is a collection of operations on $A$, none of which is nullary. “Subalgebra" will have its natural meaning. Tuples are defined inductively using ordered pairs. The $1$-tuple $(x)$ is defined to be $x$. The $2$-tuple $(x,y)$ is the ordered pair of $x$ and $y$. If $n \geq 2$, then $(x_0,\dotsc,x_n)$ is the ordered pair $((x_0,\dotsc,x_{n-1}),x_n)$. The set of infinite and finite sequences in $A$ are denoted by ${^\omega}\!A$ and ${^{<\omega}}\!A$ respectively. Suppose $\vec{a}$ is an infinite sequence $\langle a_0, a_1,a_2,\dotsc\rangle$. For each $n \geq 1$, let $\vec{a}\!\upharpoonright \! n$ denote the initial segment of $\vec{a}$ of length $n$, namely $\langle a_0, a_1,\dotsc, a_{n-1}\rangle$. For every $n \in \omega$, the cut-off sequence $\langle a_n, a_{n+1},a_{n+2},\dotsc\rangle$ is denoted by $\vec{a}-n$. If $\vec{b}$ is a finite sequence $\langle b_0, b_1,\dotsc,b_{n-1}\rangle$, then $\vert \vec{b}\vert$ is the length of $\vec{b}$, and the concatenation $\vec{b}\ast \vec{a}$ of $\vec{b}$ and $\vec{a}$ is $\langle b_0, b_1,\dotsc,b_{n-1}, a_0, a_1,a_2,\dotsc\rangle$. Suppose $f$ is an $n$-ary operation on a set $A$ and $\vec{a}$ is a finite sequence in $A$. We will write $f(\vec{a})$ for $f(\vec{a}(0), \dotsc,\vec{a}(n-1))$ for notational convenience, where it is implicitly assumed that the length of $\vec{a}$ is $n$. Similarly, if $\bar{x}$—a symbol with a bar over it indicate a list—is a list of variables, we may write $f(\bar{x})$ and it is understood that the list $\bar{x}$ consists of $n$ variables. A *field of sets* over a set $S$ is a nonempty collection $\mathcal{A}$ of subsets of $S$ that is closed under finite unions, finite intersections and complementation (relative to $S$). For our purposes, we will always assume that every field of sets contains all the singletons. Suppose $\mathcal{A}$ is a field of sets over $S$. An *ultrafilter* $U$ on $\mathcal{A}$ is a subset of $\mathcal{A}$ closed under intersections such that $S \in U$, $\emptyset \notin U$ and for every $X \in \mathcal{A}$, either $X\in U$ or $ X^c\in U$. Note that if $\mathcal{A}$ is the power set of $S$, an ultrafilter on $\mathcal{A}$ is the same as what is typically called an ultrafilter on $S$. The set of ultrafilters on $S$ is denoted by $\beta S$. The *principal ultrafilter* $U$ on $\mathcal{A}$ *generated* by an element $a$ in $S$ is the ultrafilter $\{\,X \in \mathcal{A}\mid a \in X\,\}$. An ultrafilter $U$ on $\mathcal{A}$ is *nonprincipal* if[f]{} it is not principal. Note that an ultrafilter $U$ on $\mathcal{A}$ over $S$ is nonprincipal if and only if every $X\in U$ is infinite. As an example, consider the collection $\mathcal{A}$ of all finite and cofinite subsets of a set $S$. Clearly, $\mathcal{A}$ is a field of sets over $S$. Additionally, the collection of all cofinite subsets of $S$ is the unique nonprincipal ultrafilter on $\mathcal{A}$. Suppose $\{A_i\}_{i \in I}$ is an indexed collection of distinct sets. An $n$-ary operation on $\{A_i\}_{i \in I}$ is a function with domain $A_{i_1}\times \dotsb \times A_{i_n}$ for some $i_1,\dotsc, i_n\in I$ and codomain $A_j$ for some $j\in I$. Suppose $\mathcal{F}$ is a collection of operations on $\{ A_i \}_{i \in I}$. The structure $(\{ A_i \}_{i \in I}, \mathcal{F})$ is known as a *heterogeneous algebra* in the literature (see [@BL70]). Suppose $B_i \subseteq A_i$ for each $i\in I$. We say that $\{ B_i \}_{i \in I}$ is closed under $\mathcal{F}$ if[f]{} for every $f\colon A_{i_1}\times \dotsb \times A_{i_n}\rightarrow A_{j}$ in $\mathcal{F}$, we have $f(x_1, \dotsc,x_n)\in B_j$ whenever $(x_1, \dotsc, x_n)\in B_{i_1}\times \dotsb \times B_{i_n}$. The following lemma should be nothing new to the universal algebraists, logicians and perhaps mathematicians in general, especially for the case where $I$ is a singleton. Nevertheless, to our knowledge, the result in this form does not appear in the literature. It will be needed later and so we state it without proof. \[2111\] Suppose $(\{ A_i \}_{i \in I}, \mathcal{F})$ is a heterogeneous algebra, where $\mathcal{F}$ is countable. If $X_i$ is a countable subset of $A_i$ for each $i \in I$, then there exists a countable superset $B_i$ of $X_i$ for every $i \in I$ such that $\{ B_i \}_{i \in I}$ is closed under $\mathcal{F}$. Ramsey Algebras and Idempotent Ultrafilters {#2706b} =========================================== A few terminologies are needed before we can introduce the notion of Ramsey algebra. The following is a special type of composition of operations, to the author’s knowledge, introduced by Carlson in [@tC88]. In fact, Carlson’s definition is more general because it is defined for any heterogenoeus algebra. \[0601a\] Suppose $(A, \mathcal{F})$ is an algebra. An operation $f \colon A^m\rightarrow A$ is an *orderly composition* of $\mathcal{F}$ if[f]{} there exist $g, h_1, \dotsc, h_n \in \mathcal{F}$ such that $f( \bar{x}_1, \dotsc, \bar{x}_n)=g(h_1(\bar{x}_1), \dotsc, h_n(\bar{x}_n))$. We say that $\mathcal{F}$ is *closed under orderly composition* if[f]{} $f \in \mathcal{F}$ whenever $f$ is an orderly composition of $\mathcal{F}$. The collection of *orderly terms* over $\mathcal{F}$ is the smallest collection of operations on $A$ that includes $\mathcal{F}$, contains the identity function on $A$ and is closed under orderly composition. \[0524c\] Suppose $(A, \mathcal{F})$ is an algebra and $\vec{a}, \vec{b}$ are infinite sequences in $A$. We say that $\vec{a}$ is a *reduction* of $\vec{b}$ with respect to $\mathcal{F}$, and write $\vec{a} \leq_{\mathcal{F}} \vec{b}$ if[f]{} there are finite sequences $\vec{b}_n$ and orderly terms $f_n$ over $\mathcal{F}$ for all $n \in \omega$ such that $\vec{b}_0 \ast \vec{b}_1 \ast \vec{b}_2 \ast \dotsb$ is a subsequence of $\vec{b}$ and $\vec{a}(n)=f_n(\vec{b}_n)$ for all $n \in \omega$. The following is an analogous definition of reduction defined for finite sequences. Suppose $(A, \mathcal{F})$ is an algebra and $\vec{a}, \vec{b}$ are finite sequences in $A$. We say that $\vec{a}$ is a *reduction* of $\vec{b}$ with respect to $\mathcal{F}$, and write $\vec{a} \unlhd_{\mathcal{F}} \vec{b}$ if[f]{} there are finite sequences $\vec{b}_n$ and orderly terms $f_n$ over $\mathcal{F}$ for $n< \vert \vec{a}\vert$ such that $\vec{b}_0 \ast \vec{b}_1 \ast \dotsb\ast \vec{b}_{\vert \vec{a}\vert -1}$ is a subsequence of $\vec{b}$ and $\vec{a}= \langle f_0(\vec{b}_0), f_1(\vec{b}_1), \dotsc,f_{\vert \vec{a}\vert -1}(\vec{b}_{\vert \vec{a}\vert -1})\rangle$. It is easy to check that $\leq_{\mathcal{F}}$ is a pre-partial ordering on ${^\omega}\!A$ and that $\unlhd_{\mathcal{F}}$ is a pre-partial ordering on ${^{<\omega}}\!A$. Suppose $(A, \mathcal{F})$ is an algebra and $\vec{b}$ is an infinite sequence in $A$. An element $a$ of $A$ is a *finite reduction* of $\vec{b}$ with respect to $\mathcal{F}$ if[f]{} $a$ is equal to $f(\vec{b}_0)$ for some orderly term $f$ over $\mathcal{F}$ and some finite subsequence $\vec{b}_0$ of $\vec{b}$. The set of all finite reductions of $\vec{b}$ with respect to $\mathcal{F}$ is denoted by $\operatorname{FR}_{\mathcal{F}}(\vec{b})$. Our definitions of $\leq_{\mathcal{F}}$ and $\unlhd_{\mathcal{F}}$ are equivalent to a special case of the one given in [@tC88], where the collection of operations contains all projections. Our choice is driven by Hindman’s Theorem; under our definition, $\operatorname{FR}_{\{+\}}(\vec{b})=\operatorname{FS}(\vec{b})$, where $\vec{b} \in {^\omega}{\mathbb{N}}$. Note that if $\vec{a} \leq_{\mathcal{F}} \vec{b}$, then $\operatorname{FR}_{\mathcal{F}}(\vec{a})\subseteq \operatorname{FR}_{\mathcal{F}}(\vec{b})$. Suppose $(A, \mathcal{F})$ is an algebra. We say that $(A, \mathcal{F})$ is a *Ramsey algebra* if[f]{} for every $\vec{a}\in {^\omega}\!A$ and $X \subseteq A$, there exists $\vec{b} \leq_{\mathcal{F}} \vec{a}$ such that $\operatorname{FR}_{\mathcal{F}}(\vec{b})$ is either contained in or disjoint from $X$. It is a consequence of Hindman’s Theorem that every semigroup is a Ramsey algebra (see [@HS12 V, §2]). Suppose $(A, \mathcal{F})$ is an algebra such that for every $\vec{a} \in {^\omega}\!A$, there exists $\vec{b} \leq_{\mathcal{F}} \vec{a}$ such that $\vert \operatorname{FR}_{\mathcal{F}}(\vec{b}) \vert=1$. Then $(A, \mathcal{F})$ is trivially Ramsey, and we say that it is a *degenerate Ramsey algebra*. \[130613b\] Suppose $(A, \mathcal{F})$ is a nondegenerate Ramsey algebra. Then there exists $\vec{a}\in {^\omega}\! A$ such that $\operatorname{FR}_{\mathcal{F}}(\vec{b})$ is infinite whenever $\vec{b}\leq_{\mathcal{F}} \vec{a}$. For the remaining of this section, some definitions and known results that will not be needed are briefly presented for the sake of comparison with our work in the new framework. \[1207g\] Assume $U$ is an ultrafilter on $A$ and $f \colon A \rightarrow B$. Let $f_{\ast}(U)$ be the ultrafilter on $B$ defined by $\{\,X \subseteq B \mid f^{-1}[X] \in U \,\}$. \[0923a\] Assume $U$ and $V$ are ultrafilters on sets $A$ and $B$ respectively. $U \times V$ is the ultrafilter on $A\times B$ defined by $$\{\,X \subseteq A\times B \mid \{\, a \in A \mid \{\,b\in B \mid (a,b) \in X \,\}\in V \,\}\in U\,\}$$ Suppose $U_i$ is an ultrafilter on $A_i$ for each $i=1, \dotsc,n$ and $f \colon A_1 \times \dotsb \times A_n \rightarrow B$. We will write $f(U_1, \dotsc, U_n)$ for $f(U_1\times \dotsb \times U_n)$. Suppose $\mathcal{F}$ is a collection of operations on a set $A$. An ultrafilter $U$ on $A$ is said to be *idempotent* for $\mathcal{F}$ if[f]{} $f_*(U, \dotsc, U)=U$ whenever $f \in \mathcal{F}$. \[1007a\] Suppose $\mathcal{F}$ is a collection of operations on a set $A$. If $U$ is an ultrafilter on $A$ idempotent for $\mathcal{F}$, then it is idempotent for the collection of orderly terms over $\mathcal{F}$. Suppose $(A,f)$ is a semigroup. Then $(\beta A, f_\ast)$ is a compact right topological semigroup. By the Ellis-Nakamura Lemma, there is an ultrafilter $U$ on $A$ idempotent for $f$. An extensive exposition to the theory of compact right topological semigroup of this form can be found in [@HS12]. The following is a generalization of strongly summable ultrafilter. \[0207a\] Suppose $\mathcal{F}$ is a collection of operations on a set $A$ and $U$ is an ultrafilter on $A$. We say that $U$ is *strongly reductible* for $\mathcal{F}$ if[f]{} for every $X \in U$, there exists $\vec{a} \in {^\omega}\!A$ such that $\operatorname{FR}_{\mathcal{F}}(\vec{a}) \subseteq X$ and $\operatorname{FR}_{\mathcal{F}}(\vec{a}-n) \in U$ for all $n \in \omega$. Every ultrafilter strongly reductible for $\mathcal{F}$ is necessarily idempotent for $\mathcal{F}$. \[0707b\] Assume Martin’s Axiom. Suppose $(\omega, \mathcal{F})$ is a nondegenerate Ramsey algebra. Then there exists a nonprincipal strongly reductible ultrafilter. framework ========= In this section, a framework will be proposed to study idempotent ultrafilters in some general setting. We begin by generalizing Definition \[1207g\] and \[0923a\] to field of sets. Later, under the new framework, results analogous to Theorem \[1007a\] and \[0707b\] will be presented. \[1006a\] Suppose $\mathcal{A}$ is a field of sets over a set $S$ and $U$ is an ultrafilter on $\mathcal{A}$. Suppose $T$ is a set and $f \colon S\rightarrow T$. If $\mathcal{B}$ is a field of sets over $T$ such that $f^{-1}[X]\in \mathcal{A}$ for all $X\in \mathcal{B}$, then the collection $\{\, X\in \mathcal{B}\mid f^{-1}[X]\in U\,\}$ is an ultrafilter on $\mathcal{B}$. Straightforward. Consider the special case where $\mathcal{A}$ and $\mathcal{B}$ are the power sets $\mathcal{P}(S)$ and $\mathcal{P}(T)$. For every function $f \colon S\rightarrow T$, it is automatic that $f^{-1}[X]\in \mathcal{P}(S)$ for all $X\in \mathcal{P}(T)$. In this case, the ultrafilter $\{\, X\in \mathcal{P}(T)\mid f^{-1}[X]\in U\,\}$ equals $f_*(U)$. \[0915a\] Suppose $\mathcal{A}$ is a field of sets over $S$ and $\mathcal{B}$ is a field of sets over $T$. Suppose $U$ and $V$ are ultrafilters on $\mathcal{A}$ and $\mathcal{B}$ respectively. Suppose $\mathcal{C}$ is a field of sets over $S\times T$ such that 1. $\{\,t\in T \mid (s,t) \in X \,\}\in \mathcal{B}$ whenever $X \in \mathcal{C}$ and $s \in S$; 2. $\{\,s\in S \mid \{\,t\in T \mid (s,t) \in X \,\} \in V\,\} \in \mathcal{A}$ whenever $X \in \mathcal{C}$. Then $ \{\,X \in \mathcal{C} \mid \{\,s\in S \mid \{\,t\in T \mid (s,t) \in X \,\} \in V\,\} \in U\,\}$ is an ultrafilter on $\mathcal{C}$. Straightforward. The first condition ensures that $\{\,s\in S \mid \{\,t\in T \mid (s,t) \in X^c \,\} \in V\,\} =\{\,s\in S \mid \{\,t\in T \mid (s,t) \in X \,\} \in V\,\}^c$ whenever $X \in \mathcal{C}$. \[1117a\] Suppose $\mathcal{A}, \mathcal{B}, \mathcal{C}, S, T, U,V$ are as stated in Theorem \[0915a\]. Let $U \times_{ \mathcal{C}} V$ denote the ultrafilter on $\mathcal{C}$ given by $\{\,X \in \mathcal{C} \mid \{\,s\in S \mid \{\,t\in T \mid (s,t) \in X \,\} \in V\,\} \in U\,\}$. Every subalgebra of a Ramsey algebra is Ramsey. In fact, it is easy to show that assuming the collection of operations is countable, an algebra is Ramsey if and only if every countable subalgebra is Ramsey. This is due to the fact that the subalgebra generated by the terms of a given infinite sequence is countable. Furthermore, any ultrafilter idempotent for a subalgebra of a given algebra can be extended naturally to an ultrafilter idempotent for the algebra. Therefore, for our purposes, without loss of generality, we may assume the underlying set is always $\omega$. Henceforth, we will focus on fields of sets over $\omega$ or Cartesian power of $\omega$. For the sake of readability, we assume the variables $a,b, a_0, a_1, x_0,x_1$ and et cetera run through $\omega$ whenever it is applicable; for example, $\{ \,( a_1,\dotsc, a_{n-1}) \mid \{\, a_n \mid (a_1, \dotsc, a_n) \in X \,\}$ stands for $\{ \,( a_1,\dotsc, a_{n-1}) \in \omega^{n-1} \mid \{\, a_n \in \omega \mid (a_1, \dotsc, a_n) \in X \,\}$. \[1010a\] Suppose $\mathfrak{A}=\{ \mathcal{A}_n \}_{n\geq 1}$ is an indexed collection such that $\mathcal{A}_n$ is a field of sets over $\omega^n$ for every $n\geq 1$. We say that $\mathfrak{A}$ is *admissible* if[f]{} for every $n \geq 2 $ and $X \in \mathcal{A}_{n}$ 1. $ \{\,(a_2, \dotsc,a_{n}, a_1) \mid (a_1, \dotsc,a_{n})\in X \,\} \in \mathcal{A}_{n}$; 2. $ \{\,(a_2, \dotsc,a_{n}) \mid (a_1, \dotsc,a_{n})\in X \,\} \in \mathcal{A}_{n-1}$ for all $a_1 \in \omega$. \[1510a\] Suppose $\mathfrak{A}=\{ \mathcal{A}_n \}_{n\geq 1}$ is admissible. If $n \geq 2 $ and $X \in \mathcal{A}_{n}$, then $\square(k)$ holds for every $1 \leq k \leq n-1$, where $\square(k)$ stands for the statement: $$\{\,(a_{n-k+1}, \dotsc,a_{n}) \mid (a_1, \dotsc,a_{n})\in X \,\} \in \mathcal{A}_k \text{ for every } a_1, \dotsc, a_{n-k} \in \omega.$$ Suppose $n \geq 2 $ and $X \in \mathcal{A}_{n}$. We will prove by inverse induction on $k$ that $\square(k)$ holds for every $1\leq k \leq n-1$. By admissibility, $\square(n-1)$ holds. Assume $\square(k)$ holds and $k>1$. Fix $a_1, \dotsc, a_{n-k+1} \in \omega$. Let $Y$ be the set $\{\,(a_{n-k+2}, \dotsc, a_{n}) \mid (a_1, \dotsc,a_{n-k+1}, a_{n-k+2}, \dotsc, a_n)\in X \,\}$ and let $Z$ be the set $\{\,(b,a_{n-k+2}, \dotsc,a_{n}) \mid (a_1,\dotsc, a_{n-k}, b, a_{n-k+2}, \dotsc,a_{n})\in X \,\}$. By the induction hypothesis, $Z$ is in $\mathcal{A}_k$. Since $Y$ is equal to $\{\,(a_{n-k+2}, \dotsc,a_{n}) \mid (a_{n-k+1}, a_{n-k+2}, \dotsc,a_{n}) \in Z \,\}$, by admissibility, it follows that $Y\in \mathcal{A}_{k-1}$, showing that $\square(k-1)$ holds. A result stronger than Proposition \[1510a\] can be proven. For example, assuming $\{ \mathcal{A}_n \}_{n\geq 1}$ is admissible, it can be shown that $ \{\,(a_3, a_5,a_2) \mid (a_1, a_2,a_3,a_4,a_5)\in X \,\} \in \mathcal{A}_3$ for every $a_1,a_4\in \omega$ and $X\in \mathcal{A}_5$. It is not done because it is unnecessary and notationally cumbersome. Suppose $\mathfrak{A}=\{ \mathcal{A}_n \}_{n\geq 1}$ is admissible. An *$\mathfrak{A}$-ultrafilter* is an ultrafilter $U$ on $\mathcal{A}_1$ such that for every $n \geq 1$ and $X\in \mathcal{A}_{n+1} $, $$\{ \,( a_1,\dotsc, a_{n}) \mid \{\, a_{n+1} \mid (a_1, \dotsc, a_{n+1}) \in X \,\}\in U\,\}\in \mathcal{A}_n.$$ The set of all $\mathfrak{A}$-ultrafilters is denoted by $\beta\mathfrak{A}$. Note that by Proposition \[1510a\], $\{\,a_{n+1} \mid (a_1, \dotsc,a_{n+1})\in X \,\} \in \mathcal{A}_1$ whenever $X\in \mathcal{A}_{n+1} $ and $ a_1, \dotsc, a_{n}\in \omega$. \[1912a\] Suppose $\mathcal{A}_n$ is the set of finite and cofinite subsets of $\omega^n$ for each $n \in \omega$. Then $\mathfrak{A}=\{ \mathcal{A}_n \}_{n\geq 1}$ is admissible. All principal ultrafilters and the unique nonprincipal ultrafilter on $\mathcal{A}_1$ can easily be verified to be $\mathfrak{A}$-ultrafilters. \[2312b\] Suppose $\mathfrak{A}=\{\mathcal{A}_{n}\}_{n \geq 1}$ is admissible. Then every principal ultrafilter on $\mathcal{A}_1$ is an $\mathfrak{A}$-ultrafilter. Suppose $U$ is a principal ultrafilter on $\mathcal{A}_1$ generated by some $c\in \omega$. Suppose $n \geq 1 $ and $X \in \mathcal{A}_{n+1}$. For every $a_1, \dotsc, a_{n} \in \omega$, by the definition of a principal ultrafilter, $\{\, a_{n+1} \mid (a_1, \dotsc, a_{n+1}) \in X \,\}\in U$ if and only if $(a_1, \dotsc, a_{n}, c) \in X$. Hence, to see that $U$ is an $\mathfrak{A}$-ultrafilter, it suffices to show that $$Y:=\{\,(a_1, \dotsc,a_n) \mid (a_1, \dotsc, a_{n}, c) \in X \,\} \in \mathcal{A}_n.$$ Using admissibility repeatedly, $Z:=\{\,(c,a_1, \dotsc,a_n) \mid (a_1, \dotsc, a_{n}, c) \in X \,\} \in \mathcal{A}_{n+1}$. Hence, $Y$ is in $\mathcal{A}_n$ because $Y$ is equal to $\{\,(a_1, \dotsc,a_n) \mid (c,a_1, \dotsc, a_{n}) \in Z \,\}$. Theorem \[0915a\] justifies the following definition. Suppose $\mathfrak{A}=\{ \mathcal{A}_n \}_{n\geq 1}$ is admissible and $U_i$ is an $\mathfrak{A}$-ultrafilter for every $i\geq 1$. Define $(U_1)_{\mathfrak{A}}$ to be $U_1$. For each $n \geq 2$, define $(U_1 \otimes \dotsb \otimes U_n)_{\mathfrak{A}}$ inductively to be the ultrafilter $(U_1 \otimes \dotsb \otimes U_{n-1})_{\mathfrak{A}} \times_{\mathcal{A}_n } U_n $ on $\mathcal{A}_n$. For simplicity, we will write $U_1 \otimes \dotsb \otimes U_n$ for $(U_1 \otimes \dotsb \otimes U_n)_{\mathfrak{A}}$ and $U^n$ for $(\underbrace{U \otimes \dotsb \otimes U}_{n \text{ times}})_{\mathfrak{A}}$ whenever $\mathfrak{A}$ is understood. To be clear, $U_1 \otimes U_2= \{\, X\in \mathcal{A}_2 \mid \{ \, a_1 \mid \{\, a_2 \mid (a_1, a_2) \in X \,\}\in U_2\,\} \in U_1 \,\}$. Assuming $U_1 \otimes \dotsb \otimes U_{n-1}$ is an ultrafilter on $\mathcal{A}_{n-1}$, let us justify that $U_1 \otimes \dotsb \otimes U_n$ is an ultrafilter on $\mathcal{A}_n$. We apply Theorem \[0915a\] with $\mathcal{A}=\mathcal{A}_{n-1}$, $S=\omega^{n-1}$, $\mathcal{B}=\mathcal{A}_1$, $T=\omega$, $U=U_1 \otimes \dotsb \otimes U_{n-1}$, $V=U_n$ and $\mathcal{C}=\mathcal{A}_n$. Recall $((a_1, \dotsc, a_{n-1}), a_n)$ is the same as $(a_1, \dotsc, a_n)$. Hence, the two conditions in Theorem \[0915a\] translate into: 1. $\{\,a_n \mid (a_1, \dotsc,a_{n})\in X \,\} \in \mathcal{A}_1$ whenever $X\in \mathcal{A}_n $ and $ a_1, \dotsc, a_{n-1}\in \omega$; 2. $\{ \,( a_1,\dotsc, a_{n-1}) \mid \{\, a_{n} \mid (a_1, \dotsc, a_{n}) \in X \,\}\in U\,\}\in \mathcal{A}_{n-1}$ whenever $X\in \mathcal{A}_n$. By Proposition \[1510a\], the first condition is satisfied because $\mathfrak{A}$ is admissible. The second condition is satisfied because $U$ is an $\mathfrak{A}$-ultrafilter. Therefore, $\{\, X\in \mathcal{A}_n \mid \{\, ( a_1,\dotsc, a_{n-1}) \mid \{\, a_n \mid (a_1, \dotsc, a_n) \in X\, \}\in U_n\,\}\in U_1 \otimes \dotsb \otimes U_{n-1}\, \}$ is an ultrafilter on $\mathcal{A}_n$, and that is our $U_1 \otimes \dotsb \otimes U_n$. \[0913a\] Suppose $\mathfrak{A}=\{ \mathcal{A}_n \}_{n \geq 1}$ is admissible and $U_1, \dotsc,U_n$ are $\mathfrak{A}$-ultrafilters, where $n \geq 2$. Then for every $X \in \mathcal{A}_n$ and $1 \leq k \leq n-1$, $$W:=\{(a_1, \dotsc,a_{n-k}) \mid \{(a_{n-k+1}, \dotsc,a_n) \mid (a_1, \dotsc,a_n)\in X \} \in U_{n-k+1} \otimes \dotsb \otimes U_n\}$$ is in $\mathcal{A}_{n-k}$ and $X \in U_1 \otimes \dotsb \otimes U_n$ if and only if $W \in U_1 \otimes \dotsb \otimes U_{n-k}$. (Note that by Proposition \[1510a\], $ \{\,(a_{n-k+1}, \dotsc,a_n) \mid (a_1, \dotsc,a_n)\in X \,\} \in \mathcal{A}_k$ whenever $a_1, \dotsc, a_{n-k} \in \omega$.) We proceed by induction on $n$. The base step $n=2$ is immediate by the definition of $\mathfrak{A}$-ultrafilter and $U_1 \otimes U_2$. Assume $n>2$. Suppose $X \in \mathcal{A}_n$, $1 \leq k \leq n-1$ and $W$ is as stated in the lemma. If $k=1$, then the conclusion follows similarly. Now, assume $k>2$. Since $U_n$ is an $\mathfrak{A}$-ultrafilter, $Z:=\{\,(a_1, \dotsc,a_{n-1}) \mid \{\,a_n\mid (a_1, \dotsc,a_n)\in X \,\} \in U_n\,\}$ is in $\mathcal{A}_{n-1}$ and $X \in U_1 \otimes \dotsb \otimes U_n$ if and only if $Z \in U_1 \otimes \dotsb \otimes U_{n-1}$. Now, $W$ is equal to $\{(a_1, \dotsc,a_{n-k}) \mid \{\,(a_{n-k+1}, \dotsc,a_{n-1})\mid \{\,a_n \mid (a_1, \dotsc,a_n)\in X \,\} \in U_n \,\}\in U_{n-k+1} \otimes \dotsb \otimes U_{n-1}\,\}$, which in turn is equal to $\{\,(a_1, \dotsc,a_{n-k}) \mid \{\,(a_{n-k+1},\dotsc a_{n-1}) \mid (a_1, \dotsc,a_{n-1})\in Z\,\} \in U_{n-k+1} \otimes \dotsb \otimes U_{n-1}\,\}$. By the induction hypothesis, $W$ is in $\mathcal{A}_{n-k}$ and $Z \in U_1 \otimes \dotsb \otimes U_{n-1}$ if and only if $W \in U_1 \otimes \dotsb \otimes U_{n-k}$. Therefore, $X \in U_1 \otimes \dotsb \otimes U_n$ if and only if $W \in U_1 \otimes \dotsb \otimes U_{n-k}$. \[1510b\] Suppose $\mathfrak{A}=\{ \mathcal{A}_n \}_{n\geq 1}$ is admissible. An $m$-ary operation $f$ on $\omega$ is an *$\mathfrak{A}$-operation* if[f]{} for every $n \geq 1$ and $X \in\mathcal{A}_n$, $$\{\,( a_1, \dotsc, a_{n+m-1} ) \mid ( a_1,\dotsc, a_{n-1}, f(a_{n}, \dotsc,,a_{n+m-1}) ) \in X\,\} \in \mathcal{A}_{n+m-1}.$$ In particular, $f^{-1}[X] \in \mathcal{A}_m$ for all $X \in\mathcal{A}_1$. \[0911a\] Suppose $\mathfrak{A}=\{ \mathcal{A}_n \}_{n\geq 1}$ is admissible. Suppose $f$ is an $m$-ary $\mathfrak{A}$-operation. If $U_1, \dotsc, U_m$ are $\mathfrak{A}$-ultrafilters, then the collection $\{\, X\in \mathcal{A}_1 \mid f^{-1}[X]\in U_1\otimes \dotsb\otimes U_m\,\}$ is an $\mathfrak{A}$-ultrafilter. Let $\mathscr{C}$ be $\{\, X\in \mathcal{A}_1 \mid f^{-1}[X]\in U_1\otimes \dotsb\otimes U_m\,\}$. Since $f^{-1}[X] \in \mathcal{A}_m$ for all $X \in\mathcal{A}_1$ and $U_1\otimes \dotsb\otimes U_m$ is an ultrafilter on $\mathcal{A}_m$, by Proposition \[1006a\], $\mathscr{C}$ is an ultrafilter on $\mathcal{A}_1$. Fix $n \geq 1$ and $X \in \mathcal{A}_{n+1}$. We need to show that $W:=\{ \,( a_1,\dotsc, a_{n}) \mid \{\, a_{n+1} \mid (a_1, \dotsc, a_{n+1}) \in X \,\}\in \mathscr{C}\,\}\in \mathcal{A}_{n}$. For every $a_1, \dotsc, a_{n}\in \omega$, it is easy to verify that $\{\, a_{n+1} \mid (a_1, \dotsc, a_{n+1}) \in X \,\}\in \mathscr{C}$ if and only if $\{\, (a_{n+1}, \dotsc,a_{n+m}) \mid (a_1, \dotsc, a_{n}, f(a_{n+1}, \dotsc,a_{n+m}) ) \in X\,\} \in U_1\otimes \dotsb\otimes U_{m} $. Since $f$ is an $\mathfrak{A}$-operation, there exists $Z \in \mathcal{A}_{n+m}$ such that $(a_1, \dotsc,a_{n+m}) \in Z$ if and only if $(a_1, \dotsc, a_{n}, f(a_{n+1}, \dotsc,a_{n+m}) ) \in X$. Therefore, $W$ is equal to $\{ \,( a_1,\dotsc, a_{n}) \mid \{\, (a_{n+1}, \dotsc,a_{n+m}) \mid (a_1, \dotsc,a_{n+m}) \in Z\,\} \in U_1\otimes \dotsb\otimes U_m\,\}$. By Proposition \[0913a\], $W$ is in $\mathcal{A}_{n}$. Theorem \[0911a\] justifies the following definition. Suppose $\mathfrak{A}=\{ \mathcal{A}_n \}_{n\geq 1}$ is admissible and $f$ is an $m$-ary $\mathfrak{A}$-operation. Define $f_*^{\mathfrak{A}}$ to be the $m$-ary operation on $\beta\mathfrak{A}$ given by $$f_*^{\mathfrak{A}}(U_1, \dotsc, U_m)=\{\, X\in \mathcal{A}_1 \mid f^{-1}[X]\in U_1\otimes \dotsb\otimes U_m\,\}.$$ In fact, $f_*^{\mathfrak{A}}$ is an operation on $\beta\mathfrak{A}$ that extends $f$, when we identify the natural numbers with the principal ultrafilters on $\mathcal{A}_1$. Let $\mathfrak{A}$ be $\{\mathcal{P}(\omega^n)\}_{n\geq 1}$. Then $\mathfrak{A}$ is trivially admissible. If $f$ is any operation on $\omega$, then $f$ is immediately an $\mathfrak{A}$-operation. In fact, $(\beta\mathfrak{A},f_*^{\mathfrak{A}} )$ is equal to $(\beta\omega, f_*) $. \[3112a\] Suppose $\mathfrak{A}=\{ \mathcal{A}_n \}_{n\geq 1}$ and $\mathfrak{B}=\{ \mathcal{B}_n \}_{n\geq 1}$ are admissible. For every $n \geq 1$, assume $\mathcal{A}_n\subseteq \mathcal{B}_n$, $U_n$ is an $\mathfrak{A}$-ultrafilter and $V_n$ is a $\mathcal{B}$-ultrafilter, where $U_n=V_n\cap \mathcal{A}_1$. Then for every $m \geq 1$, we have $$(U_1\otimes \dotsb \otimes U_m)_{\mathfrak{A}} = (V_1\otimes \dotsb \otimes V_m)_{\mathfrak{B}} \cap \mathcal{A}_m.$$ Furthermore, if $f$ is an $m$-ary $\mathfrak{A}$-operation as well as a $\mathfrak{B}$-operation, then $f_*^{\mathfrak{A}}(U_1, \dotsc, U_m )= f_*^{\mathfrak{B}}(V_1, \dotsc, V_m )\cap \mathcal{A}_1$. We prove the first part by induction on $m$. The base step holds trivially. For the inductive step, first, suppose $X\in \mathcal{A}_{m+1}$. Let $W_X^{\mathfrak{A}}$ denote $\{\, (a_1, \dotsc, a_m) \mid \{\, a_{m+1} \mid (a_1, \dotsc, a_{m+1})\in X \,\}\in U_{m+1}\,\}$ and $W_X^{\mathfrak{B}}$ denote $\{\, (a_1, \dotsc, a_m) \mid \{\, a_{m+1} \mid (a_1, \dotsc, a_{m+1})\in X \,\}\in V_{m+1}\,\}$. Note that $W_X^{\mathfrak{A}}=W_X^{\mathfrak{B}}$ because $U_{m+1}=V_{m+1}\cap \mathcal{A}_1$. Additionally, since $U_{m+1}$ is an $\mathfrak{A}$-ultrafilter, $W_X^{\mathfrak{A}}\in \mathcal{A}_m$. Therefore, $$\begin{aligned} &(V_1\otimes \dotsb \otimes V_{m+1})_{\mathfrak{B}} \cap \mathcal{A}_{m+1}\\ &=\{ X\in \mathcal{B}_{m+1} \mid W_X^{\mathfrak{B}} \in (V_1\otimes \dotsb \otimes V_{m})_{\mathfrak{B}} \}\cap \mathcal{A}_{m+1}\\ &=\{\, X\in \mathcal{A}_{m+1} \mid W_X^{\mathfrak{A}} \in (V_1\otimes \dotsb \otimes V_{m})_{\mathfrak{B}} \,\}\\ &=\{\, X\in \mathcal{A}_{m+1} \mid W_X^{\mathfrak{A}} \in (U_1\otimes \dotsb \otimes U_{m})_{\mathfrak{A}} \,\} \qquad \text{(by the induction hypothesis)}\\ &=(U_1\otimes \dotsb \otimes U_{m+1})_{\mathfrak{A}}\end{aligned}$$ For the second part, $f_*^{\mathfrak{B}}(V_1, \dotsc, V_m )\cap \mathcal{A}_1=\{\, X\in \mathcal{A}_1\mid f^{-1}[X]\in (V_1\otimes \dotsb \otimes V_m)_{\mathfrak{B}}\,\}=\{\, X\in \mathcal{A}_1\mid f^{-1}[X]\in (U_1\otimes \dotsb \otimes U_m)_{\mathfrak{A}}\,\}= f_*^{\mathfrak{A}}(U_1, \dotsc, U_m )$. The second equality holds because of the first part and $f^{-1}[X]\in \mathcal{A}_m$ for all $X\in \mathcal{A}_1$. \[2312a\] Suppose $\mathfrak{A}=\{ \mathcal{A}_n \}_{n\geq 1}$ is admissible. If $(\omega,f)$ is a semigroup and $f$ is an $\mathfrak{A}$-operation, then $(\beta\mathfrak{A}, f_\ast^{\mathfrak{A}})$ is a semigroup. Let $\mathfrak{B}$ equal to $\{\mathcal{P}(\omega^n)\}_{n\geq 1}$. Then $(\beta\mathfrak{B}, f_\ast^{\mathfrak{B}})$, being equal to $(\beta\omega, f_*)$, is a semigroup. Suppose $U_1, U_2,U_3\in \beta\mathfrak{A}$. Let $V_1, V_2,V_3$ be any ultrafilter on $\mathcal{P}(\omega)$ that extends $U_1,U_2,U_3$ respectively. By Proposition \[3112a\], $f_*^{\mathfrak{A}}(U_1, U_2)=f_*^{\mathfrak{B}}(V_1, V_2)\cap \mathcal{A}_1$ and $f_*^{\mathfrak{A}}(U_2, U_3)=f_*^{\mathfrak{B}}(V_2, V_3)\cap\mathcal{A}_1$. Therefore, $f_*^{\mathfrak{A}}(f_*^{\mathfrak{A}}(U_1, U_2),U_3) = f_*^{\mathfrak{B}}(f_*^{\mathfrak{B}}(V_1, V_2),V_3)\cap \mathcal{A}_1=f_*^{\mathfrak{B}} (V_1,f_*^{\mathfrak{B}}(V_2, V_3) ) \cap \mathcal{A}_1 =f_*^{\mathfrak{A}} (U_1,f_*^{\mathfrak{A}}(U_2, U_3)) $. The first and the third equality follow from Proposition \[3112a\]. \[1120a\] Suppose $\mathfrak{A}=\{ \mathcal{A}_n \}_{n\geq 1}$ is admissible and $U$ is an $\mathfrak{A}$-ultrafilter. Suppose $\mathcal{F}$ is a collection of $\mathfrak{A}$-operations. We say that $U$ is *idempotent* for $\mathcal{F}$ with respect to $\mathfrak{A}$ if[f]{} $f_*^{\mathfrak{A}}(U,\dotsc,U)=U$ whenever $f \in \mathcal{F}$. \[2706e\] Suppose $\mathcal{A}_n$ is the set of finite and cofinite subsets of $\omega^n$ for each $n \in \omega$. Suppose $f$ is any binary operation on $\omega$ such that $f^{-1}[\{a\}]$ is finite for all $a\in \omega$. Then $f$ is an $\mathfrak{A}$-operation. Furthermore, the unique nonprincipal $\mathfrak{A}$-ultrafilter is idempotent for $f$ (or $\{f\}$ precisely) with respect to $\mathfrak{A}$. However, $(\omega,f)$ need not be a Ramsey algebra and ultrafilters on $\mathcal{P}(\omega)$ idempotent for $f$ may not exist. Suppose $\mathfrak{A}=\{ \mathcal{A}_n \}_{n\geq 1}$ is admissible and $f$ is a binary associative $\mathfrak{A}$-operation. Take any ultrafilter $V$ on $\omega$ such that $f_*(V,V)=V$. Let $U$ be $V\cap \mathcal{A}_1$. If $U$ is an $\mathfrak{A}$-ultrafilter, by Proposition \[3112a\], $U$ is idempotent for $f$ with respect to $\mathfrak{A}$. Equivalently, if an $\mathfrak{A}$-ultrafilter $U$ can be extended to an ultrafilter on $\omega$ idempotent for $f$, then $U$ is idempotent for $f$ with respect to $\mathfrak{A}$. However, Example \[2706e\] suggests that this may not be possible, even if $U$ is idempotent for $f$ with respect to $\mathfrak{A}$. Therefore, although $(\beta\mathfrak{A}, f_\ast^{\mathfrak{A}})$ is a semigroup, it is conceivable that an $\mathfrak{A}$-ultrafilter idempotent for $f$ with respect to $\mathfrak{A}$ may not exist. If that is true, one can further analyze the admissible $\mathfrak{A}$ such that the existence of such idempotent ultrafilters is guaranteed. Nevertheless, our framework is sufficient and minimal in some sense to allow the construction of a homegeneous sequence for every set in an idempotent ultrafilter. Suppose $\mathfrak{A}=\{ \mathcal{A}_n \}_{n\geq 1}$ is admissible and $f$ is a binary associative $\mathfrak{A}$-operation. If $U$ is an $\mathfrak{A}$-ultrafilter idempotent for $f$ with respect to $\mathfrak{A}$, then for every $X\in U$, there exists $\vec{a}\in {^\omega}\omega$ such that $\operatorname{FR}_{\mathcal{F}}(\vec{a})\subseteq X$. The standard Galvin’s method of constructing homogeneous sequences, for example, as in Theorem 5.8 of [@HS12] can be carried out. Existence of Idempotent Ultrafilters ==================================== In this section, we will exbibit the existence of idempotent ultrafilters for every nondegenerate Ramsey algebra in some countable sense. To do this, countability is imposed on the general framework introduced in the previous section. Apart from the reason from the point of reverse mathematics, the countability assumption is actually crucial in our construction. We start off with some lemmas. \[1120b\] Suppose $\mathcal{F}$ is a countable collection of operations on $\omega$. Suppose $S_n$ is a countable collection of subsets of $\omega^n$ for every $n \geq 1$. Then there exists an admissible $\mathfrak{A}=\{ \mathcal{A}_n \}_{n\geq 1}$ such that $\mathcal{F}$ is a collection of $\mathfrak{A}$-operations and $\mathcal{A}_n$ is a countable superset of $S_n$ for every $n\geq 1$. For each $n\geq 1$, let $\cap_n$, $\cup_n$ and $\operatorname{comp}_n$ denote the respective operations on $\mathcal{P}(\omega^n)$ defined by $\cap_n(X,Y)=X\cap Y$, $\cup_n(X,Y)=X\cup Y$, $\operatorname{comp}_n(X)=\omega^n\backslash X$. For $n \geq 2$, the operation $\operatorname{cyc}_n$ on $\mathcal{P}(\omega^n)$ is defined by $\operatorname{cyc}_n(X)= \{\,(a_2, \dotsc,a_n, a_1) \mid (a_1, \dotsc,a_n)\in X \,\}$. For each $n\geq 2$ and $c\in \omega$, let $\operatorname{fib}_n^c \colon \mathcal{P}(\omega^{n}) \rightarrow \mathcal{P}(\omega^{n-1})$ be defined by $\operatorname{fib}_n^c(X)= \{\,(a_1, \dotsc,a_{n-1}) \mid (c,a_1, \dotsc,a_{n-1})\in X \,\}$. For every $f\in \mathcal{F}$ and $n\geq 1$, say $f$ is $m$-ary, let $\operatorname{pre}_n^f \colon \mathcal{P}(\omega^n) \rightarrow \mathcal{P}(\omega^{n+m-1})$ be defined by $\operatorname{pre}_n^f(X)= \{\,( a_1, \dotsc, a_{n+m-1} ) \mid ( a_1,\dotsc, a_{n-1}, f(a_{n}, \dotsc,,a_{n+m-1}) ) \in X\,\}$. Suppose $\mathscr{H}=\{\, \cup_n, \cap_n, \operatorname{comp}_n \mid n\geq 1\,\} \bigcup \{\, \operatorname{cyc}_n, \operatorname{fib}_n^c \mid n\geq 2, c\in \omega\,\} \bigcup \{\, \operatorname{pre}_n^f \mid f\in \mathcal{F}, n\geq 1\,\} $. Then $\mathscr{H}$ is a countable collection of operations on $\{ \mathcal{P}(\omega^n) \}_{n\geq 1}$. By Theorem \[2111\], choose $\mathfrak{A}=\{ \mathcal{A}_n \}_{n\geq 1}$ such that $\mathcal{A}_n$ is a countable superset of $S_n$ for every $n\geq 1$ and that $\mathfrak{A}$ is closed under $\mathscr{H}$. For every $n \geq 1$, since $\mathfrak{A}$ is closed under $\{\cap_n,\cup_n,\operatorname{comp}_n \mid n \geq 1\}$, $\mathcal{A}_n$ is a field of sets over $\omega^n$. Since $\mathfrak{A}$ is closed under $\{ \,\operatorname{cyc}_n, \operatorname{adm}_n^c \mid n \geq 2, c\in \omega\,\}$, it is admissible. For every $f\in \mathcal{F}$, since $\mathfrak{A}$ is closed under $\{\, \operatorname{pre}_n^f \mid n\geq 1\,\} $, it follows that $f$ is an $\mathfrak{A}$-operation. \[2311a\] Suppose $\mathfrak{A}_i=\{\mathcal{A}_{n,i}\}_{n \geq 1}$ is admissible for every $i$ in an index set $I$. Suppose $\mathcal{F}$ is a collection of operations on $\omega$, each of which is an $\mathfrak{A}_i$-operation for every $i\in I$. For every $n \geq 1$, take $\mathcal{A}_n= \bigcup_{i \in I} \mathcal{A}_{n,i}$. Then $\mathfrak{A}= \{\mathcal{A}_n\}_{n\geq 1}$ is admissible and $\mathcal{F}$ is a collection of $\mathfrak{A}$-operations. Straightforward. Definition \[0207a\] is modified very slightly to the following. It serves us well as shown by the following theorem. \[1119e\] Suppose $\mathcal{F}$ is a collection of operations on $\omega$ and $\mathcal{A}$ is a field of sets over $\omega$. An ultrafilter $U$ on $\mathcal{A}$ is *strongly reductible* for $\mathcal{F}$ if[f]{} for every $X \in U$, there exists $\vec{a} \in {^\omega}\omega$ such that $\operatorname{FR}_{\mathcal{F}}(\vec{a}) \subseteq X$ and $\operatorname{FR}_{\mathcal{F}}(\vec{a}-i) \in U$ for all $i \in \omega$. \[1207f\] Suppose $\mathfrak{A}=\{ \mathcal{A}_n \}_{n\geq 1}$ is admissible and $U$ is an $\mathfrak{A}$-ultrafilter. Suppose $\mathcal{F}$ is a collection of $\mathfrak{A}$-operations and $U$ is strongly reductible for $\mathcal{F}$. Then $U$ is idempotent for $\mathcal{F}$ with respect to $\mathfrak{A}$. Suppose $f$ is an $m$-ary operation in $\mathcal{F}$. Since $f^{\mathfrak{A}}_*(U^m)$ is an ultrafilter on $\mathcal{A}_1$, to see that $U$ is idempotent for $f$ with respect to $\mathfrak{A}$, it suffices to show that $X\in f^{\mathfrak{A}}_*(U^m)$ for every $X\in U$. Suppose $X\in U$. Choose $\vec{b} \in {^\omega}\omega$ such that $\operatorname{FR}_{\mathcal{F}}(\vec{b}) \subseteq X$ and $\operatorname{FR}_{\mathcal{F}}(\vec{b}-i) \in U$ for all $i \in \omega$. We need to show that $f^{-1}[X] \in U^m$ and this follows from the case $k=m$ in the following claim. For each $1 \leq k \leq m$ and $a_1, \dotsc, a_{m-k}\in \omega$ such that $\langle a_1, \dotsc, a_{m-k}\rangle\unlhd_{\mathcal{F}} \vec{b}\!\upharpoonright\! N$ for some $N\in\omega$, it follows that $\{\, (a_{m-k+1}, \dotsc, a_m) \mid f(a_1, \dotsc, a_m) \in X \,\}\in U^{k}$. The claim is proved by induction on $k$. First, note that $\{\, (a_{m-k+1}, \dotsc, a_m) \mid f(a_1, \dotsc, a_m) \in X \,\}\in \mathcal{A}_k$ for every $1 \leq k \leq m$ and $a_1, \dotsc, a_{m-k}\in \omega$ by $f^{-1}[X]\in \mathcal{A}_m$ and Proposition \[1510a\]. Consider the base case $k=1$. Suppose $\langle a_1, \dotsc, a_{m-1}\rangle\unlhd_{\mathcal{F}} \vec{b} \!\upharpoonright \! N$ for some $N \in \omega$. By the definition of orderly term, $f(a_1, \dotsc, a_m) \in \operatorname{FR}_{\mathcal{F}}(\vec{b})$ for all $a_m\in \operatorname{FR}_{\mathcal{F}}(\vec{b}-N)$. Thus $\operatorname{FR}_{\mathcal{F}}(\vec{b}-N)\subseteq \{\, a_m \mid f(a_1, \dotsc, a_m) \in X \,\}$. Since $\operatorname{FR}_{\mathcal{F}}(\vec{b}-N)$ is in $ U$, so is $\{\, a_m \mid f(a_1, \dotsc, a_m) \in X \,\}$. Now, assume $1<k \leq m$. Suppose $\langle a_1, \dotsc, a_{m-k}\rangle\unlhd_{\mathcal{F}} \vec{b} \!\upharpoonright \! N$ for some $N \in \omega$. By Lemma \[0913a\], it suffices to show that $\{\, a_{m-k+1} \mid \{\, (a_{m-k+2}, \dotsc, a_m) \mid f(a_1, \dotsc, a_m) \in X \,\}\in U^{k-1} \,\}\in U$. This will follow if $\operatorname{FR}_{\mathcal{F}}(\vec{b}-N) \subseteq \{\, a_{m-k+1} \mid \{\, (a_{m-k+2}, \dotsc, a_m) \mid f(a_1, \dotsc, a_m) \in X \,\}\in U^{k-1} \,\}$. Suppose $a_{m-k+1} \in \operatorname{FR}_{\mathcal{F}}(\vec{b}-N)$. Then $\langle a_1, \dotsc, a_{m-k+1}\rangle$ is a reduction of some initial segment of $\vec{b}$ with respect to $\mathcal{F}$. By the induction hypothesis, $\{\, (a_{m-k+2}, \dotsc, a_m) \mid f(a_1, \dotsc, a_m) \in X \,\}\in U^{k-1}$. In the case of addition, the next theorem is known as the Iterated Hindman Theorem, which is equivalent to Hindman’s Theorem in $\mathsf{ZFC}$. Before that, we state a technical crucial lemma. [@wcT13b] \[0729\] Suppose $\langle \vec{a}_n \rangle_{n \in \omega}$ is a sequence in ${^\omega}\!\omega$ such that $\vec{a}_{n+1} \leq_{\mathcal{F}} \vec{a}_n$ for all $n \in \omega$. Then there exists $\vec{b}\in {^\omega}\!\omega$ such that $\vec{b}-n \leq_{\mathcal{F}} \vec{a}_n$ for all $n \in \omega$. \[1211a\] Suppose $(\omega, \mathcal{F})$ is a Ramsey algebra. Then for every $\vec{a} \in {^\omega}\omega$ and every indexed collection $\{A_n\}_{n \in \omega}$ of subsets of $\omega$, there exists $\vec{b} \leq_{\mathcal{F}}\vec{a}$ such that $\operatorname{FR}_{\mathcal{F}}(\vec{b}-n)$ is either contained in or disjoint from $A_n$ whenever $n \in \omega$. Suppose $\vec{a} \in {^\omega}\omega$ and $\{A_n\}_{n \in \omega}$ is an indexed collection of subsets of $\omega$. Since the algebra is Ramsey, for every $n \in \omega$, we can choose $\vec{a}_n \in {^\omega}\omega$ inductively with $\vec{a}_0 \leq_{\mathcal{F}} \vec{a}$ such that $\vec{a}_{n+1} \leq_{\mathcal{F}} \vec{a}_n$ and $\operatorname{FR}_{\mathcal{F}}(\vec{a}_n)$ is either contained in or disjoint from $A_n$ for each $n \in \omega$. By Lemma \[0729\], choose $\vec{b}\in {^\omega}\omega$ such that $\vec{b}-n \leq_{\mathcal{F}} \vec{a}_n$ for all $n \in \omega$. By transitivity, $\vec{b} =\vec{b}-0 \leq_{\mathcal{F}} \vec{a}$. The conclusion follows since $\operatorname{FR}_{\mathcal{F}}(\vec{b}-n) \subseteq \operatorname{FR}_{\mathcal{F}}(\vec{a}_n)$ for all $n \in \omega$. \[0805a\] Suppose $(\omega, \mathcal{F})$ is a Ramsey algebra and $\mathcal{A}$ is a countable field of sets over $\omega$. For every $\vec{a} \in {^\omega}\omega$, there exists $\vec{b} \leq_{\mathcal{F}} \vec{a}$ such that $\{\,X \in \mathcal{A} \mid \operatorname{FR}_{\mathcal{F}}(\vec{b}-n) \subseteq X \text{ \textnormal{for some} } n \in \omega\,\}$ is an ultrafilter on $\mathcal{A}$. Fix an enumeration $A_0, A_1, A_2, \dotsc$ of the sets in $\mathcal{A}$. Suppose $\vec{a} \in {^\omega}\omega$. By Theorem \[1211a\], choose $\vec{b} \leq_{\mathcal{F}}\vec{a}$ such that $\operatorname{FR}_{\mathcal{F}}(\vec{b}-n)$ is either contained in or disjoint from $A_n$ for each $n \in \omega$. We claim that $U:=\{\,X \in \mathcal{A} \mid \operatorname{FR}_{\mathcal{F}}(\vec{b}-n) \subseteq X \text{ for some } n \in \omega\,\}$ is an ultrafilter on $\mathcal{A}$. Clearly, $\omega\in U$ and $\emptyset \notin U$. Suppose $A, B \in U$, say $\operatorname{FR}_{\mathcal{F}}(\vec{b}-n) \subseteq A$ and $\operatorname{FR}_{\mathcal{F}}(\vec{b}-m) \subseteq B$ for some $n,m \in \omega$. Then $A\cap B \in \mathcal{A}$ and $\operatorname{FR}_{\mathcal{F}}(\vec{b}-\max\{n,m\}) \subseteq A\cap B$, implying that $A\cap B\in U$. Meanwhile, by the choice of $\vec{b}$, either $A_n\in U$ or ${A_n}^{\!\!c} \in U$ for each $n \in \omega$. Hence, $U$ is an ultrafilter on $\mathcal{A}$. \[0905b\] Assume $(\omega, \mathcal{F})$ is a nondegenerate Ramsey algebra and $\mathcal{F}$ is countable. Suppose $S_n$ is a countable collection of subsets of $\omega^n$ for every $n \geq 1$. Then there exists an admissible $\mathfrak{A}=\{ \mathcal{A}_n \}_{n\geq 1}$ such that 1. $\mathcal{A}_n$ is countable superset of $S_n$ for every $n \geq 1$; 2. $\mathcal{F}$ is a collection of $\mathfrak{A}$-operations; 3. there exists a nonprincipal $\mathfrak{A}$-ultrafilter $U$ strongly reductible for $\mathcal{F}$. First of all, by Theorem \[130613b\], fix a sequence $\vec{a}$ such that $\operatorname{FR}_{\mathcal{F}}(\vec{b})$ is infinite whenever $\vec{b}\leq_{\mathcal{F}} \vec{a}$. For every $m \in \omega$, we will construct inductively an admissible $\mathfrak{A}_m=\{\mathcal{A}_{n,m}\}_{n \geq 1}$, an ultrafilter $U_m$ on $\mathcal{A}_{1,m}$ and a sequence $\vec{a}_m$ of natural numbers. By Lemma \[1120b\], choose an admissible $\mathfrak{A}_0=\{\mathcal{A}_{n,0}\}_{n \geq 1}$ such that $\mathcal{A}_{n,0}$ is countable superset of $S_n$ for every $n \geq 1$ and $\mathcal{F}$ is a collection of $\mathfrak{A}_0$-operations. Since $(\omega, \mathcal{F})$ is a Ramsey algebra, by Lemma \[0805a\], choose $\vec{a}_0\leq_{\mathcal{F}}\vec{a}$ such that $U_0=\{\,X \in \mathcal{A}_{1,0} \mid \operatorname{FR}_{\mathcal{F}}(\vec{a}_0-i) \subseteq X \text{ for some } i \in \omega\,\}$ is an ultrafilter on $\mathcal{A}_{1,0}$. At the inductive step, by Lemma \[1120b\], choose an admissible $\mathfrak{A}_{m+1}=\{\mathcal{A}_{n,m+1}\}_{n \geq 1}$ such that $\mathcal{A}_{n,m+1}$ is countable superset of $\mathcal{A}_{n,m}$ for every $n \geq 1$, that $\mathcal{F}$ is a collection of $\mathfrak{A}_{m+1}$-operations and additionally - $\operatorname{FR}_{\mathcal{F}}(\vec{a}_m -i)\in \mathcal{A}_{1,m+1}$ for every $ i \in \omega$; - for every $n \geq 1$, we have $\{ \,( a_1,\dotsc, a_n) \mid \{\, a_{n+1} \mid (a_1, \dotsc, a_{n+1}) \in X \,\}\in U_{m}\,\}\in \mathcal{A}_{n,m+1}$ for all $X \in \mathcal{A}_{n+1,m}$. Now, by Lemma \[0805a\], choose $\vec{a}_{m+1}\leq_{\mathcal{F}}\vec{a}_m$ such that $U_{m+1}=\{\,X \in \mathcal{A}_{1,m+1} \mid \operatorname{FR}_{\mathcal{F}}(\vec{a}_{m+1}-i) \subseteq X \text{ for some } i \in \omega\,\}$ is an ultrafilter on $\mathcal{A}_{1,m+1}$. For every $n \geq 1$, take $\mathcal{A}_n= \bigcup_{m\in \omega} \mathcal{A}_{n,m}$. Take $\mathfrak{A}$ to be $\{\mathcal{A}_n\}_{n\geq 1}$. Clearly, $\mathcal{A}_n$ is countable for every $n \geq 1$ since it is a countable union of countable sets. By our construction, $S_n\subseteq \mathcal{A}_{n,0}\subseteq \mathcal{A}_n$ for every $n \geq 1$. By Lemma \[2311a\], $\mathfrak{A}$ is admissible and $\mathcal{F}$ is a collection of $\mathfrak{A}$-operations. Now, take $U=\bigcup_{m \in \omega} U_m$. We claim that $\operatorname{FR}_{\mathcal{F}}(\vec{a}_m-i)$ is infinite and is in $U$ for every $m,i \in \omega$. To see this, suppose $m,i \in \omega$. By transitivity, $\vec{a}_m-i\leq_{\mathcal{F}} \vec{a}$. By our choice of $\vec{a}$, we know $\operatorname{FR}_{\mathcal{F}}(\vec{a}_m-i)$ is infinite. Since $\vec{a}_{m+1}\leq_{\mathcal{F}} \vec{a}_m$, we have $\operatorname{FR}_{\mathcal{F}}(\vec{a}_{m+1}-i)\subseteq \operatorname{FR}_{\mathcal{F}}(\vec{a}_m-i)$. Thus $\operatorname{FR}_{\mathcal{F}}(\vec{a}_m-i) \in U_{m+1}\subseteq U$ as $\operatorname{FR}_{\mathcal{F}}(\vec{a}_m-i) \in \mathcal{A}_{1,m+1}$. Therefore, $\operatorname{FR}_{\mathcal{F}}(\vec{a}_m-i)\in U$. Because $U_0\subseteq U_1\subseteq U_2\, \dotsb $, it follows easily that $U$ is an ultrafilter on $\mathcal{A}_1$. To see that $U$ is an $\mathfrak{A}$-ultrafilter, suppose $n \geq 1$ and $X\in \mathcal{A}_{n+1}$. Then $X\in \mathcal{A}_{n+1,m}$ for some $m \in \omega$. Since $\mathfrak{A}_m$ is admissible, $\{\, a_{n+1} \mid (a_1, \dotsc, a_{n+1}) \in X \,\}\in \mathcal{A}_{1,m}$ for every $a_1, a_2, \dotsc, a_{n}\in \omega$. Therefore, $\{ \,( a_1,\dotsc, a_{n}) \mid \{\, a_{n+1} \mid (a_1, \dotsc, a_{n+1}) \in X \,\}\in U\,\} =\{ \,( a_1,\dotsc, a_{n}) \mid \{\, a_{n+1} \mid (a_1, \dotsc, a_{n+1}) \in X \,\}\in U_m\,\}\in \mathcal{A}_{n,m+1}\subseteq \mathcal{A}_{n}$. Finally, suppose $X\in U$. Then $X\in U_m$ for some $m\in \omega$. Hence, $\operatorname{FR}_{\mathcal{F}}(\vec{a}_{m}-i) \subseteq X$ for some $i \in \omega$. By our claim, $X$ is infinite and $\operatorname{FR}_{\mathcal{F}}(\vec{a}_m-j)\in U$ for each $j \geq i$. Therefore, $U$ is nonprincipal and strongly reductible. Extension of Idempotentness to Orderly Terms ============================================ In this section, we will show that the analogue of Theorem \[1007a\] holds in our framework. Suppose $\mathfrak{A}=\{ \mathcal{A}_n \}_{n\geq 1}$ is admissible. If $X\subseteq \omega^n$ for some $n\in \omega$, whose value is implicitly implied by the context, then we write $X\overset{.}{\in} \mathfrak{A}$ to mean $X\in \mathcal{A}_n$. This is done out of convenience as seen in the following lemma. \[1207c\] Suppose $\mathfrak{A}=\{ \mathcal{A}_n \}_{n\geq 1}$ is admissible. Suppose $m \geq 1 $ and $h_1, \dotsc, h_m$ are $\mathfrak{A}$-operations. Then for every $n \geq 1$ and $X\in \mathcal{A}_{n+m-1}$, $$\{\, (a_1, \dotsc,a_{n-1}, \bar{x}_1, \dotsc, \bar{x}_m) \mid (a_1, \dotsc,a_{n-1}, h_1(\bar{x}_1), \dotsc, h_m(\bar{x}_m) ) \in X\,\}\overset{.}{\in} \mathfrak{A}$$ The proof proceeds by induction on $m$. The base case $m=1$ holds since $h_1$ is an $\mathfrak{A}$-operation. Assume $m >1$. Suppose $n \geq 1$ and $X\in \mathcal{A}_{n+m-1}$. We need to show that $W\overset{.}{\in} \mathfrak{A}$, where $W$ is as stated in the lemma. By the induction hypothesis, $ Y_1:=\{\,(a_1, \dotsc,a_{n}, \bar{x}_2, \dotsc, \bar{x}_m) \mid (a_1, \dotsc,a_{n}, h_2(\bar{x}_2), \dotsc, h_m(\bar{x}_m)) \in X\,\}\overset{.}{\in} \mathfrak{A}$. By applying admissibility repeatedly, $Y_2 :=\{\, ( \bar{x}_2, \dotsc, \bar{x}_m, a_1, \dotsc,a_{n}) \mid ( a_1, \dotsc,a_{n}, \bar{x}_2, \dotsc, \bar{x}_m )\in Y_1\,\} \overset{.}{\in} \mathfrak{A}$. Since $h_1$ is an $\mathfrak{A}$-operation, it follows that $Y_3 :=\{\, ( \bar{x}_2, \dotsc, \bar{x}_m, a_1, \dotsc,a_{n-1}, \bar{x}_1 ) \mid ( \bar{x}_2, \dotsc, \bar{x}_m, a_1,\dotsc,a_{n-1},h_1(\bar{x}_1) )\in Y_2\,\} \overset{.}{\in} \mathfrak{A}$. Finally, $\{\, ( a_1, \dotsc,a_{n-1},\bar{x}_1 , \bar{x}_2, \dotsc, \bar{x}_m) \mid (\bar{x}_2, \dotsc, \bar{x}_m, a_1, \dotsc,a_{n-1},\bar{x}_1 )\in Y_3\,\} \overset{.}{\in} \mathfrak{A}$. It is easy to verify that this last set is equal to $W$. \[1207d\] Suppose $\mathfrak{A}=\{ \mathcal{A}_n \}_{n\geq 1}$ is admissible. If $\mathcal{F}$ is a collection of $\mathfrak{A}$-operations, then every orderly term over $\mathcal{F}$ is an $\mathfrak{A}$-operation. The proof goes by induction on the generation of the orderly terms over $\mathcal{F}$. The identity function on $\omega$ trivially is an $\mathfrak{A}$-operation while every operation in $\mathcal{F}$ is an $\mathfrak{A}$-operation by the assumption. Suppose that $f( \bar{x}_1, \dotsc, \bar{x}_m)=g(h_1(\bar{x}_1), \dotsc, h_m(\bar{x}_m))$ for some orderly terms $g, h_1, \dotsc, h_m$ over $\mathcal{F}$. By the induction hypothesis, $g, h_1, \dotsc, h_m$ are $\mathfrak{A}$-operations. Suppose $n \geq 1$ and $X\in \mathcal{A}_n$. We want $W:=\{\,( a_1, \dotsc,a_{n-1}, \bar{x}_1, \dotsc, \bar{x}_m) \mid ( a_1, \dotsc,a_{n-1}, f( \bar{x}_1, \dotsc, \bar{x}_m) ) \in X\,\}\overset{.}{\in} \mathfrak{A}$. Since $g$ is an $\mathfrak{A}$-operation, it follows that $Y:= \{\, (a_1, \dotsc,a_{n-1}, b_1, \dotsc, b_m ) \mid ( a_1, \dotsc,a_{n-1}, g(b_1, \dotsc, b_m) ) \in X\,\}\in \mathcal{A}_{n+m-1}$. Therefore, Lemma \[1207c\] implies that $\{\,( a_1, \dotsc,a_{n-1}, \bar{x}_1, \dotsc, \bar{x}_m) \mid (a_1,\dotsc,a_{n-1}, h_1(\bar{x}_1), \dotsc, h_m(\bar{x}_m)) \in Y\,\}\overset{.}{\in} \mathfrak{A}$. This last set is equal to $W$. We introduce some additional notations for the sake of brevity in the next lemma. Suppose $\mathfrak{A}$ is admissible and $\bar{U}_i$ is a list of $\mathfrak{A}$-ultrafilters $U_{1, i}, \dotsc, U_{p_i,i}$ for each $1\leq i\leq n$. Let $\otimes(\bar{U}_1, \dotsc, \bar{U}_n)$ denote $U_{1,1} \otimes \dotsb \otimes U_{p_1,1} \otimes \dotsb \otimes U_{1,n} \otimes \dotsb \otimes U_{p_n,n}$. \[2606c\] Suppose $\mathfrak{A}=\{ \mathcal{A}_n \}_{n\geq 1}$ is admissible. Suppose $m \geq 1 $ and $h_1, \dotsc, h_m$ are $\mathfrak{A}$-operations. Then $X \in {h_1}^{\mathfrak{A}}_*(\bar{U}_1)\otimes \dotsc\otimes {h_m}^{\mathfrak{A}}_*(\bar{U}_m)$ if and only if $\{\, (\bar{x}_1, \dotsc, \bar{x}_m) \mid (h_1(\bar{x}_1), \dotsc, h_m(\bar{x}_m))\in X \,\} \in \otimes(\bar{U}_1, \dotsc, \bar{U}_m)$ for each $X \in \mathcal{A}_m$ and lists $\bar{U}_1, \dotsc, \bar{U}_m$ of $\mathfrak{A}$-ultrafilters of the correct length. The proof goes by induction on $m$. The base case $m=1$ holds by the definition of ${h_1}^{\mathfrak{A}}_*$. Assume $m >1$. Suppose $X \in \mathcal{A}_m$ and $\bar{U}_1, \dotsc, \bar{U}_m$ are lists of $\mathfrak{A}$-ultrafilters of the correct length. Let $Z$ be $\{\, (\bar{x}_1, \dotsc, \bar{x}_m ) \mid(h_1(\bar{x}_1), \dotsc, h_m(\bar{x}_m) ) \in X\,\}$. By Lemma \[1207c\], $Z\overset{.}{\in} \mathfrak{A}$. By definition, $X \in {h_1}^{\mathfrak{A}}_*(\bar{U}_1)\otimes \dotsc\otimes {h_m}^{\mathfrak{A}}_*(\bar{U}_m)$ if and only if $\{\, (b_1, \dotsc, b_{m-1}) \mid \{\, b_m \mid (b_1, \dots, b_m) \in X \,\} \in {h_m}^{\mathfrak{A}}_*(\bar{U}_m)\,\} \in {h_1}^{\mathfrak{A}}_*(\bar{U}_1)\otimes \dotsc\otimes {h_{m-1}}^{\mathfrak{A}}_*(\bar{U}_{m-1})$. By the induction hypothesis, this holds if and only if $\{\, (\bar{x}_1, \dotsc, \bar{x}_{m-1}) \mid \{\, b_m \mid (h_1(\bar{x}_1), \dotsc, h_{m-1}(\bar{x}_{m-1}), b_m)\in X \,\} \in {h_m}^{\mathfrak{A}}_*(\bar{U}_m)\,\} \in$ $\otimes(\bar{U}_1, \dotsc, \bar{U}_{m-1})$. In turn, this holds if and only if $\{\, (\bar{x}_1, \dotsc, \bar{x}_{m-1}) \mid \{\, (\bar{x}_m) \mid (\bar{x}_1, \dotsc, \bar{x}_m) \in Z\,\} \in \bar{U}_m\,\} \in \otimes(\bar{U}_1, \dotsc, \bar{U}_{m-1})$. By Lemma \[0913a\], this holds if and only if $Z \in \otimes(\bar{U}_1, \dotsc, \bar{U}_m)$. \[1208h\] Suppose $\mathfrak{A}=\{ \mathcal{A}_n \}_{n\geq 1}$ is admissible and $\mathcal{F}$ is a collection of $\mathfrak{A}$-operations. If $U$ is an $\mathfrak{A}$-ultrafilter idempotent for $\mathcal{F}$ with respect to $\mathfrak{A}$, then $U$ is idempotent for the collection of orderly terms over $\mathcal{F}$ with respect to $\mathfrak{A}$. By Lemma \[1207d\], every orderly term $f$ over $\mathcal{F}$ is indeed an $\mathfrak{A}$-operation The proof goes by induction on the generation of the orderly terms over $\mathcal{F}$. If $I$ is the identity function on $\omega$, then $I^{\mathfrak{A}}_*$ is the identity function on $\beta\mathfrak{A}$ and hence $I^{\mathfrak{A}}_*(U)=U$. If $f \in \mathcal{F}$, then $U$ is idempotent for $f$ with respect to $\mathfrak{A}$ by the assumption. Suppose that $f( \bar{x}_1, \dotsc, \bar{x}_m)=g(h_1(\bar{x}_1), \dotsc, h_m(\bar{x}_m))$ for some orderly terms $g, h_1, \dotsc, h_m$ over $\mathcal{F}$. By the induction hypothesis, $g^{\mathfrak{A}}_*(U,\dotsc,U)=U$ and ${h_i}^{\mathfrak{A}}_*(U,\dotsc,U)=U$ for each $1\leq i\leq m$. It suffices to show that $f^{\mathfrak{A}}_*( U,\dotsc, U)=g^{\mathfrak{A}}_*({h_1}^{\mathfrak{A}}_*(U,\dotsc, U ), \dotsc, {h_m}^{\mathfrak{A}}_*(U,\dotsc, U))$. Suppose $X \in \mathcal{A}_1$. By Lemma \[2606c\], $g^{-1}[X] \in {h_1}^{\mathfrak{A}}_*(U,\dotsc, U)\otimes \dotsc\otimes {h_m}^{\mathfrak{A}}_*(U,\dotsc, U)$ if and only if $\{\, (\bar{x}_1, \dotsc, \bar{x}_m) \mid (h_1(\bar{x}_1), \dotsc, h_m(\bar{x}_n))\in g^{-1}[X] \,\} \in U\otimes \dotsb\otimes U$ and so if and only if $f^{-1}[X] \in U\otimes \dotsb\otimes U$. Open problems ============= Our introduced framework opens up a wide range of possible research problems. Here, only some problems that interest us and are most relevant to this paper are mentioned. Suppose $S_n$ is a countable collection of subsets of $\omega^n$ for every $n \geq 1$. The construction in the proof of Theorem \[0905b\] can be modified to give us a nontrivial admissible $\mathfrak{A}=\{ \mathcal{A}_n \}_{n\geq 1}$ such that $\mathcal{A}_n$ is a countable superset of $S_n$ for every $n\geq 1$ and there exists a nonprincipal $\mathfrak{A}$-ultrafilter. In view of Example \[1912a\], it is natural to ask whether this construction can be refined further to ensure that every ultrafilter on $\mathcal{A}_1$ is an $\mathfrak{A}$-ultrafilter. On the other hand, whether there exists an admissible $\mathfrak{A}$ such that only the principal ultrafilters are $\mathfrak{A}$-ultrafilters seems to be interesting and not obvious. Suppose $\mathfrak{A}$ is admissible. If $f$ is a binary associative $\mathfrak{A}$-operation, an ultrafilter idempotent for $f$ with respect to $\mathfrak{A}$ need not exist. On the other hand, Example \[2706e\] shows that such ultrafilters may exist even if the algebra is not Ramsey. These unexpected phenomena may be explained by the Ramseyness of the algebra from the point of view of $\mathfrak{A}$. Hence, one can try to formulate the notion of $\mathfrak{A}$-Ramsey algebra. With the right formulation, one might prove that whenever $(\omega,f)$ is an $\mathfrak{A}$-Ramsey algebra, there exists an $\mathfrak{A}$-ultrafilter idempotent for $f$ with respect to $\mathfrak{A}$, at least in some countable setting. Finally, we emphasize that Carlson’s original problem of the existence of idempotent ultrafilters for every Ramsey algebra is the ultimate goal. The work initiated here is the starting point for continuation work towards that direction as well as motivating others to do so. Acknowledgements {#acknowledgements .unnumbered} ================ This paper grows partially out of the author’s thesis submitted as a partial fulfilment for the award of PhD to the Ohio State University under the supervision of Timothy Carlson. The author would like to thank Carlson for introducing the concept of Ramsey algebras and for his great mentorship. The open problem regarding the existence of idempotent ultrafilters for every Ramsey algebra was conveyed by him during the supervision. Conflict of Interests {#conflict-of-interests .unnumbered} ===================== The author declares that there is no conflict of interests regarding the publication of this article. G. Birkhoff and J.D. Lipson, Heterogeneous algebras, J. Combinatorial Theory 8 (1970), 115–133. A. Blass and N. Hindman, On strongly summable ultrafilters and union ultrafilters, Trans. Amer. Math. Soc. 304 (1987), no. 1, 83–97. T.J. Carlson, Some unifying principles in Ramsey theory, Discrete Mathematics 68 (1988), 117–169. T.J. Carlson and S. G. Simpson, A dual form of Ramsey’s theorem, Adv. in Math. 53 (1984), 265-290. W. Comfort, Ultrafilters - some old and some new results, Bull. Amer. Math. Soc. 83 (1977), 417-455. E. Ellentuck, A new proof that analytic sets are Ramsey, J. Symb. Logic 39 (1974), 163-165. F. Galvin and K. Prickry, Borel sets and Ramsey’s theorem, J. Symb. Logic 38 (1973), 193-198. A.W. Hales and R.I. Jewett, Regularity and positional games, Trans. Amer. Math. Soc. 124 (1966), 360-367. N. Hindman, Finite sums from sequences within cells of a partition of $\mathbb{N}$, J. Combin. Theory (Series A) 17 (1974), 1-11. N. Hindman, Ultrafilters and combinatorial number theory, in Number Theory Carbondale, M. Nathanson ed., Lecture Notes in Math. 751 (1979), 119-184. N. Hindman, Summable ultrafilters and finite sums, in Logic and Combinatorics, S. Simpson ed., Contemp. Math. 65 (1987), 263-274. N. Hindman and D. Strauss, Algebra in the Stone-Čech Compactification: Theory and Applications, Second Edition, Walter de Gruyter, Berlin (2012). W.C. Teh, Ramsey algebras, arxiv:1403.5831v2 (to be updated to v3). W.C. Teh, Ramsey algebras and formal orderly terms, Notre Dame J. Form. Log. (to appear). W.C. Teh, Ramsey algebras and strongly reductible ultrafilters, Bull. Malays. Math. Sci. Soc. 37(4) (2014), 931-938.

ArXiv

--- abstract: | We study the impact of the convective terms on the global solvability or finite time blow up of solutions of dissipative PDEs. We consider the model examples of 1D Burger’s type equations, convective Cahn-Hilliard equation, generalized Kuramoto-Sivashinsky equation and KdV type equations, we establish the following common scenario: adding sufficiently strong (in comparison with the destabilizing nonlinearity) convective terms to equation prevents the solutions from blowing up in finite time and makes the considered system globally well-posed and dissipative and for weak enough convective terms the finite time blow up may occur similarly to the case when the equation does not involve convective term. This kind of result has been previously known for the case of Burger’s type equations and has been strongly based on maximum principle. In contrast to this, our results are based on the weighted energy estimates which do not require the maximum principle for the considered problem. address: - '${}^3$ University of Surrey, Department of Mathematics, Guildford, GU2 7XH, United Kingdom, s.zelik@surrey.ac.uk.' - '${}^{2,1}$ Department of mathematics, Ko[ç]{} University, Rumelifeneri Yolu, Sariyer, Istanbul, Turkey, vkalantarov@ku.edu.tr, bilbilgin@ku.edu.tr' author: - 'Bilgesu Bilgen ${}^1$, Varga Kalantarov ${}^2$ and Sergey Zelik${}^2$' title: Preventing blow up by convective terms in dissipative PDEs --- [^1] Introduction ============ It is well-known that the solutions of nonlinear evolutionary PDEs may blow up in a finite time. The most studied is the case of a semilinear heat equation, for instance, all positive solutions of the problem $$\label{0.heat} \partial_t u=\partial^2_x u+u^2,\ \ u\big|_{x=\pm1}=0,\ \ u\big|_{t=0}=u_0$$ blow up in finite time, see e.g., [@QS] and references therein, see also [@AlKoSv; @Ba; @BeTi; @Gal; @KaLa; @Le1; @Le2; @LePa; @Po] for the analogous results for more complicated equations. It is also known that the presence of convective terms may prevent blow up and makes the problem globally well-posed. In particular, as shown in [@ChLeSa] and [@LePaSaSt], the solutions of the problem $$\label{0.heat-bur} \partial_t u+\varepsilon u\partial_xu =\partial^2_x u+u^2,\ \ u\big|_{x=\pm1}=0,\ \ u\big|_{t=0}=u_0$$ which differs from by the presence of the convective term $\eb u\partial_xu$ becomes globally well-posed if $\eb\ne0$. However, the proof of this fact given there is strongly based on the maximum principle and the situation becomes essentially less clear for more complicated equations where the impact of the convective terms is a bit controversial. Indeed, for example, in the case of Navier-Stokes equations, the convective/inertial form $(u,\nabla_x)u$ is the only nonlinearity in the system and the only source of instability and possible blow up in the 3D case. On the other hand, if we consider the associated vorticity equation $$\label{0.vort} \partial_t\omega+(u,\nabla_x)\omega=\Delta_x \omega+(\omega,\nabla_x)u$$ and replace (very roughly) the term $\nabla_x u$ by $\omega$ in the vorticity stretching term and $u$ by $\omega$ in the vorticity transport term, we end up with the model equations similar to . This analogy can be made more precise, namely, as shown in [@fursikov], the quadratic nonlinearity $B(\omega):=(\omega,\nabla_x)u-(u,\nabla_x)\omega$ can be split into the sum of two non-local operators $B(\omega)=B_\tau(\omega)+B_n(\omega)$ such that the tangential component $B_\tau(\omega)$ satifies $$(B_\tau(\omega),\omega)=0$$ and can be interpreted as a generalized convective term and the equation $$\label{0.vort-norm} \partial_t\omega=\Delta_x \omega +B_n(\omega),$$ where only the normal component of the nonlinearity is presented possesses solutions blowing up in finite time. Thus, the Millennium problem concerning the Navier-Stokes equations can be reduced to the question whether or not the “convective” term $B_\tau(\omega)$ is strong enough to prevent blow up in the normal parabolic system . Unfortunately, the above question seems out of reach of the modern methods, so the main aim of the present paper is to analyze the impact of the convective terms for simpler equations where more or less complete answer is available despite the absence of the maximum principle. As the model examples, we consider the generalized Kuramoto-Sivashinsky equation, the so-called convective Cahn-Hilliard equation and the Korteveg de Vries type equations on a finite interval with Dirichlet boundary conditions. As we will see below, despite the fact that these equations have essentially different structure, they can be treated in the unified way and the obtained results look very similar to the case of Burgers type equations studied before. Namely, if the convective term is strong enough (in comparison with the destabilizing non-linearity), it prevents blow up of solutions and makes the considered problem globally well-posed and dissipative and if it is not sufficiently strong, the blow up may occur in the convective equations as well. We believe that this picture has a general nature and may be useful in the study of more complicated cases including the 3D Navier-Stokes equations. The paper is organized as follows. In Section \[s1\] we revisit the case of 1D Burger’s type equations: $$\label{0.eq1} \partial_tu+\partial_x(|u|^{p+1})=\partial_x^2u+f(u)+g,\ \ u\big|_{x=\pm1}=0,\ \ u\big|_{t=0}=u_0,$$ where the exponent $p>0$, the external force $g\in L^2(-1,1)$ and the destabilizing nonlinearity is growing not faster than $|u|^{q+1}$ for some other exponent $q>0$, namely, $f\in C^1(\R)$ and $$\label{0.f} |f'(u)|\le C(1+|u|^q).$$ The typical nonlinearities have the form $f(u)=u|u|^q$ and $f(u)=|u|^q$ although other choices are also allowed. Note also that the convective term of the form $\partial_x(u|u|^p)$ is also allowed here and can be treated even in a simpler way. As we have already mentioned, equation has been studied in [@LePaSaSt] and [@ChLeSa], see also [@SoWe1; @SoWe2; @Te1; @Te2] and more or less complete answer on the blow up/global solvability question has been obtained: blow up is prevented if $p\ge q$ and occurs at least for some solutions of this equation if $p<q$ (e.g., for the non-linearity $f(u)=|u|^{q+1}$). In the present paper, we present an alternative/simplified method of proving this result which based on the weighted energy estimates (multiplication of the equation on $ue^{-Lx}$ or $u_+e^{-Lx}-u_-e^{Lx}$ where $L$ is a properly chosen parameter) rather than maximum principle and which will be used throughout of the paper. In addition, in Section \[s1\], we apply this method to some model 2D Burger’s type system without the maximum principle. In section \[s2\], we study the 1D forth order parabolic equations with convective terms. The first considered example is the generalized Kuramoto-Sivashinsky equation: $$\label{0.eq2} \partial_t u+\partial_x^4 u-\lambda\partial^2_x u+\partial_x(u|u|^p)=f(u)+g,\ u\big|_{x=\pm1}=\partial_x^2u\big|_{x=\pm1}=0,\ u\big|_{t=0}=u_0,$$ where $\lambda\in\R$ and the parameters $p,q$, the external force $g$ and the nonlinearity $f$ are the same as in the case of Burger’s equation. The proved result for this equation is also almost the same as for the Burger’s equation: the blow up is prevented if $p\ge q$ and remains possible if $p<q$. The only difference is that since we do not have not $L^s$-estimates for $s\ne2$, we have to require extra restriction $p\le6$ to obtain the global well-posedness and regularity in the case $p\ge q$ (actually we do not know whether or not this extra restriction is essential). The second considered example is the convective Cahn-Hilliard equation: $$\label{0.eq3} \partial_t u+\partial_x^2(\partial^2_x u+f(u))+\partial_x(u|u|^p)=f(u)+g,\ u\big|_{x=\pm1}=\partial_x^2u\big|_{x=\pm1}=0,\ u\big|_{t=0}=u_0,$$ where the parameters $p,q$, the external force $g$ and the nonlinearity $f$ are the same as in the case of Burger’s equation. This equation looks more complicated since the destabilizing nonlinearity $\partial_x^2f(u)$ is stronger due to the presence of the second derivative. The particular case of this equation with the convective term $\partial_x(u^2)$ and periodic boundary conditions has been studied in our previous paper [@EdKaZe], see also [@EdKa] and references therein. In particular, the following partial results on global solvability/blow up have been obtained there: blow up is prevented if $q<\frac49$ and is occurred at least for some solutions if $q\ge2$. So, the behavior of solutions remains unclear if $\frac49\le q<2$. The results proved in this paper using the different method of weighted estimates are slightly better, but still not optimal, namely, the global well-posedness is verified if $q\le p/2$ (and $p\le6$) and the blow up is proved if $q>p+1$. Thus, the behavior of solutions is not clear if $p/2< q\le p+1$. Finally, in Section \[s3\], we study the 3rd order KdV type equation: $$\label{0.eq4} \partial_t u+\partial_x^3u=\partial_x(u|u|^p)+f(u)+g,\ \ u\big|_{x=\pm1}=\partial_xu\big|_{x=1}=0,\ u\big|_{t=0}=u_0,$$ where the parameters $p,q$, the external force $g$ and the nonlinearity $f$ are the same as in the case of Burger’s equation. It worth emphasizing that, in contrast to the case of KdV equations on the whole line or with periodic boundary conditions where the corresponding equations are conservative and even completely integrable, they are [*dissipative*]{} if the Dirichlet boundary conditions are posed. Although the corresponding problem is not parabolic, it possesses the smoothing property on a finite time interval and the linear part generates a $C^\infty$-semigroup, see e.g., [@KdV] and references therein. Moreover, the analytic properties of this equation is similar to Burger’s type equations (of course, up to the maximum principle). This essentially simplifies the proof of local well-posedness in comparison to the conservative case. As we show, similar to the Burger’s type equations, the blow up is prevented if $p\ge q$ (under the extra restriction $p\le2$ which is posed in order to prove the global well-posedness in the phase space $L^2(-1,1)$) and in the case $p<q$ blow up remains possible and is occurred at least for some solutions in the case where $f(u)=|u|^{q+1}$. Burger’s type equations {#s1} ======================= The aim of this section is to illustrate the impact of convective terms to reaction-diffusion equations and systems. We start with the simplest model example of 1D Burgers equation: $$\label{1.eq1} \partial_t u+\partial_x(u^2)=\partial_x^2 u+k u^2+g,\ \ u\big|_{x=\pm1}=0, \ \ u\big|_{t=0}=u_0,$$ considered on the interval $x\in(-1,1)$ and endowed by the Dirichlet boundary conditions. It is well-known that without the convective term $\partial_x(u^2)$ this equation possesses solutions blowing up in finite time if $k\ne0$. Moreover, as it is shown in [@LePaSaSt] (see also references therein), the presence of the convective term prevents the solutions to blow up and the global existence of solutions holds for any $k\in\mathbb R$. We present below an alternative/simplified proof of this fact which can be extended to much wider class of equations. Namely, the following result holds. \[Th1.bur\] Let $k\in\R$ and $g\in L^2(-1,1)$. Then, for any $u_0\in L^2(-1,1)$, equation possesses a unique solution $u(t)$ defined globally in time $t\ge0$ and this solution satisfies the following dissipative estimate: $$\label{1.estdiss-bur} \|u(t)\|_{L^2}^2\le C\|u_0\|_{L^2}^2e^{-\alpha t}+C(\|g\|^2_{L^2}+1),$$ where the positive constants $C$ and $\alpha$ are independent of $u_0$ and $t$. The local well-posedness of problem is straightforward, so we concentrate only on the derivation of dissipative estimate . To this end, we multiply equation by by $u_+(t)e^{-Lx}-u_-(t)e^{Lx}$ where $u_+=\max\{u,0\}$, $u_-=u-u_+$ and $L$ is sufficiently large positive constant which will be specified below. Then, after the standard transformations, we end up with $$\begin{gathered} \label{1.huge} \frac12\frac d{dt}\(\|u_+\|^2_{L^2_{e^{-Lx}}}+\|u_-\|^2_{L^2_{e^{Lx}}}\)+\|\Nx u_+\|^2_{L^2_{e^{-Lx}}}+\|\Nx u_-\|^2_{L^2_{e^{Lx}}}=\\= \frac12L^2\(\|u_+\|^2_{L^2_{e^{-Lx}}}+\|u_-\|^2_{L^2_{e^{Lx}}}\)+\\+(k- L)\(\|u_+\|^{3}_{L^{3}_{e^{-Lx}}}+\|u_-\|^{3}_{L^{3}_{e^{Lx}}}\)+(g,u_+(t)e^{-Lx}-u_-(t)e^{Lx}),\end{gathered}$$ where we denote by $L^p_\varphi$ the weighted Lebesgue space $L^p$ with the norm $$\|u\|^p_{L^p_\varphi}:=\int_{-1}^1|u(x)|^p\varphi(x)\,dx.$$ Crucial for our purposes is the fact that the weighted norms $\|u\|_{L^p_{e^{Lx}}}$ are equivalent to the non-weighted ones when the underlying domain is bounded. Fixing $L:=k+1$ in and using the Young inequality, we have $$\frac12\frac d{dt}\(\|u_+\|^2_{L^2_{e^{-Lx}}}+\|u_-\|^2_{L^2_{e^{Lx}}}\)+ \(\|u_+\|^2_{L^2_{e^{-Lx}}}+\|u_-\|^2_{L^2_{e^{Lx}}}\)\le C(\|g\|^2_{L^2}+1).$$ Applying the Gronwall inequality to this relation, we end up with the desired estimate and finish the proof of the theorem. The above result can be easily generalized to the problem $$\label{1.eq2} \partial_t u+\partial_x(|u|^{p+1})=\partial_x^2 u+f(u)+g,\ \ u\big|_{t=0}=u_0,$$ where the nonlinear function $f$ satisfies with the restriction $q\le p$. Indeed, repeating word by word the derivation of estimate , we end up with the same $L^2$-dissipative estimate for the solutions of equation . However, when the exponent $q$ is large enough, the only dissipative estimate in $L^2$ is not sufficient to prevent blow up in higher norms and verify the global existence and dissipativity of [*smooth*]{} solutions. To this end, we need the dissipative control of $L^s$-norm for some $s\ge s(p)<\infty$ (actually, it is not difficult to show that $s(p)=p$). Namely, the following result holds. \[Th1.bur1\] Let $q\le p$, $p>0$ and $u(t)$ be a sufficiently regular solution of . Then, for every $s\ge2$, the following estimate holds: $$\label{1.est-ls} \|u(t)\|_{L^s}^s+\int_{t}^{t+1}\|\partial_x(u^{s/2}(\tau))\|^2_{L^2}\,d\tau\le C_s\|u_0\|^s_{L^s}e^{-\alpha t}+C_s(\|g\|_{L^2}^s+1),$$ where $\alpha>0$ and the constant $C_s$ depends on $s$, but is independent of $u_0$ and $t$. We multiply equation by $u_+(t)^{s-1}e^{-Lx}-u_-(t)^{s-1}e^{Lx}$ and integrate over the interval $(-1,1)$. Then, similarly to , but using in order to estimate the term with the non-linearity $f$, we have $$\begin{gathered} \label{1.huge1} \frac1s\frac d{dt}\(\|u_+\|^s_{L^s_{e^{-Lx}}}+\|u_-\|^s_{L^s_{e^{Lx}}}\)+ \frac{4(s-1)}{s^2}\((|\partial_x(u_+^{s/2})|^2,e^{-Lx})+(|\partial_x(u_-^{s/2})|^2,e^{Lx})\)\le\\\le \frac{L^2}{s}\(\|u_+\|^{s}_{L^{s}_{e^{-Lx}}}+\|u_-\|^{s}_{L^s_{e^{Lx}}}\)+ (g,u_+^{s-1}e^{-Lx}-u_-^{s-1}e^{Lx})+\\+ C\(\|u_+\|^{s+q}_{L^{s+q}_{e^{-Lx}}}+\|u_-\|^{s+q-}_{L^{s+q}_{e^{Lx}}}+1\)- L\(\|u_+\|^{s+p}_{L^{s+p}_{e^{-Lx}}}+\|u_-\|^{s+p}_{L^{s+p}_{e^{Lx}}}\).\end{gathered}$$ Using the embedding $H^1\subset L^\infty$ (for the 1D case), we estimate the term containing the external forces $g$ as follows: $$\begin{gathered} \label{1.ext} |(g,u_+^{s-1}e^{-Lx}-u_-^{s-1}e^{Lx})|\le C_L\|g\|_{L^2}\|u\|_{L^\infty}^{s-1}\le\\\le C_L\|g\|^s_{L^2}+ \frac{2(s-1)}{s^2}\((|\partial_x(u_+^{s/2})|^2,e^{-Lx})+(|\partial_x(u_-^{s/2})|^2,e^{Lx})\).\end{gathered}$$ Using now the Young inequality and together with the facts that $p\ge q$ and $p>1$ and fixing $L>C$, we end up with the differential inequality $$\begin{gathered} \frac d{dt}\(\|u_+\|^s_{L^s_{e^{-Lx}}}+\|u_-\|^s_{L^s_{e^{Lx}}}\)+ \frac{2(s-1)}{s}\((|\partial_x(u_+^{s/2})|^2,e^{-Lx})+(|\partial_x(u_-^{s/2})|^2,e^{Lx})\)+\\+\alpha \(\|u_+\|^s_{L^s_{e^{-Lx}}}+\|u_-\|^s_{L^s_{e^{Lx}}}\)\le C_s(\|g\|^s_{L^s}+1).\end{gathered}$$ The Gronwall inequality applied to this relation gives the desired dissipative estimate and finishes the proof of the theorem. \[Cor1.well\] Under the assumptions of Theorem \[Th1.bur1\] problem is globally well-posed and dissipative in, say, $H^1_0(-1,1)$. Indeed, the local well-posedness of this problem in $H^1_0$ is straightforward and the global solvability follows in a standard way from the dissipative estimate with $s\ge p-1$ and the parabolic smoothing property, see e.g., [@LaSoUr], so we left details to the reader. Let us now consider the case $q>p$. In this case, as it is shown in the next theorem, the convective term is not strong enough to prevent finite time blow up of solutions. \[Th1.blowup\] Let $q>p$, $q>1$ and $k\ne0$. Then, there exists a smooth initial data $u_0$ such that the corresponding solution $u(t)$ of equation with $f(u)=|u|^{q+1}$ blows up in finite time. Without loss of generality we may assume that $k>0$ (otherwise, it is enough to replace $u$ by $-u$. Let us introduce a weight function $\varphi(x)=(\eb^2-x^2)^n$, where $\eb>0$ and $n>0$ is a sufficiently large number which will be specified below, multiply equation by $\varphi$ and integrate over $x\in(-\eb,\eb)$. Then, after the direct calculations, we get $$\label{1.dif} \frac d{dt}\int_{-\eb}^\eb u\varphi\,dx=\int_{-\eb}^\eb|u|^{q+1}\varphi\,dx-\int_{-\eb}^\eb|u|^{p+1}\varphi'\,dx+\int_{-\eb}^\eb u\varphi''\,dx+\int_{-\eb}^\eb g\varphi\,dx.$$ Since $q>p$, by the Hölder inequality, we have $$\Big|\int_{-\eb}^\eb|u|^{p+1}\varphi'\,dx\Big|\le\frac 14\int_{-\eb}^\eb|u|^{q+1}\varphi\,dx+ C\int_{-\eb}^\eb|\varphi'|^{\frac {q+1}{q-p}}\varphi^{\frac{-1-p}{q-p}}\,dx.$$ Note that, for all $n\ge \frac{q+1}{q-p}$, the second integrand in the right hand side has no singularities at $x=\pm\eb$ and can be majorated by a constant. Analogously, $$\label{1.2d} \Big|\int_{-\eb}^\eb u\varphi''\,dx\Big|\le \frac 14\int_{-\eb}^\eb|u|^{q+1}\varphi\,dx+C\int_{-\eb}^\eb|\varphi''|^{\frac {q+1}{q}}\varphi^{\frac{-1}{q}}\,dx$$ and the last integral in the right-hand side can be majorated by a constant if $n\ge\frac{2(q+1)}{q}$. Finally, the Jensen inequality gives $$\Big|\int_{-\eb}^\eb u\varphi\,dx\Big|^{q+1}\le C\int_{-\eb}^\eb|u|^{q+1}\varphi\,dx$$ and inserting the obtained estimates into the right-hand side of , we get $$\frac d{dt}\(\int_{-\eb}^\eb u\varphi\,dx\)\ge \alpha\Big|\int_{-\eb}^\eb u\varphi\,dx\Big|^{q+1}-C$$ for some positive $\alpha$ and $C$ which are independent of $u$. The last inequality shows that the solution $u(t)$ indeed blows up in a finite time if $$\int_{-\eb}^\eb u_0(x)\varphi(x)\,dx>\frac {C^{1/(q+1)}}{\alpha^{1/(q+1)}}$$ and the theorem is proved. \[Rem1.bound\] Note that the results on preventing blow up by convective terms, see theorems \[Th1.bur\] and \[Th1.bur1\] strongly depend on boundary conditions. Indeed, if we take, say, Neumann boundary conditions for equation , the effect will disappear since the spatially homogeneous solutions will blow up no matter how strong the convective term is. On the other hand, the result of Theorem \[Th1.blowup\] is based on the interior estimates and will hold no matter what the boundary conditions are. Mention also that the proof of theorems \[Th1.bur\] and \[Th1.bur1\] becomes [*essentially*]{} simpler if the convective term has the form $\partial_{x}(u|u|^{p})$. Indeed, in this case, we need not to use the functions $u_+$ and $u_-$ and may multiply the equation by $u(t)e^{-Lx}$. Then only the (weighted) energetic arguments are used and this allows us to apply the technique to more general equations (e.g., fourth or third order equations where the maximum principle does not hold). We consider these cases in the next sections. We conclude this section by considering the model 2D Burgers type equations where the maximum principle does not work any more, but the above arguments will allow us to establish the blow up preventing by convection. Namely, let us consider a system $$\label{1.eq3} \partial_t u+\operatorname{div}(u\otimes u)=\Delta_x u+f(u)+g,\ \ u\big|_{\partial\Omega}=0$$ where $\Omega$ is a bounded smooth domain of $\R^2$ and $u=(u_1,u_2)$ is an unknown vector field. The convective term here $$\operatorname{div}(u\otimes u):=\(\begin{matrix} \partial_{x_1}(u_1^2) &\partial_{x_2}(u_1u_2)\\ \partial_{x_1}(u_1u_2) &\partial_{x_2}(u_2^2)\end{matrix}\)$$ is the standard convective term for the Navier-Stokes system and it is clearly non-monotone, so the methods based on the maximum principle will not work at least directly. We also assume that $g\in L^2(\Omega)$ and the non-linearity $f\in C^1(\R^2,\R^2)$ has the form $$\label{f.cond} f(u)=\(\begin{matrix} k_1u_1^2\\k_2u_2^2\end{matrix}\)+\bar f(u),$$ where $|\bar f'(u)|\le C(1+|u|^{1-\eb})$ for some $\eb>0$ and $k_1,k_2\in\R$. The next theorem is the analogue of Theorem \[Th1.bur\] for this problem. \[Th1.bur3\] Let the nonlinearity $f$ satisfy and let $u(t)$ be a sufficiently regular solution of equation . Then, the following estimate holds: $$\label{1.est-l1} \|u(t)\|_{L^1}+\int_{t}^{t+1}\|u(s)\|^2_{L^2}\,ds\le C\|u_0\|_{L^1}e^{-\alpha t}+C(\|g\|_{L^2}+1),$$ where the positive constants $C$ and $\alpha$ are independent of $u_0$ and $t$. We multiply the first and the second equation of by $\operatorname{sgn}((u_1)_+)e^{-Lx_1}-\operatorname{sgn}((u_1)_+)e^{Lx_1}$ and $\operatorname{sgn}((u_2)_+)e^{-Lx_2}-\operatorname{sgn}((u_2)_+)e^{Lx_2}$ respectively, take a sum and integrate over $x\in\Omega$. Then, thanks to the Kato inequality (see, e.g., [@HiSi]) we have $$\int_{\Omega} \Delta u(x)\operatorname{sgn} u_+(x)\,dx\le 0.$$ Employing this inequality, after the straightforward transformations, we end up with $$\begin{gathered} \frac d{dt}\(\|(u_1)_+\|_{L^1_{e^{-Lx_1}}}+\|(u_1)_-\|_{L^1_{e^{Lx_1}}}+ \|(u_2)_+\|_{L^1_{e^{-Lx_2}}}+\|(u_2)_-\|_{L^1_{e^{Lx_2}}}\)+\\+ (L-|k_1|-|k_2|)\(\|(u_1)_+\|_{L^2_{e^{-Lx_1}}}^2+\|(u_1)_-\|_{L^2_{e^{Lx_1}}}^2+ \|(u_2)_+\|_{L^2_{e^{-Lx_2}}}^2+\|(u_2)_-\|_{L^2_{e^{Lx_2}}}^2\)\le\\\le L^2\(\|(u_1)_+\|_{L^1_{e^{-Lx_1}}}+\|(u_1)_-\|_{L^1_{e^{Lx_1}}}+ \|(u_2)_+\|_{L^1_{e^{-Lx_2}}}+\|(u_2)_-\|_{L^1_{e^{Lx_2}}}\)+\\+C(1+\|g\|_{L^2}+\|u\|_{L^{2-\eb}}^{2-\eb}).\end{gathered}$$ Fixing $L>|k_1|+|k_2|$ and applying Young and Gronwall inequality to this relation, we end up with the desired estimate and finish the proof of the theorem. \[Rem1.blowup\] It is not difficult to see that without the convective term, the solutions of may blow up in finite time. Also, if we take the nonlinearity $f$ growing faster than quadratically, we may construct the blow up solutions arguing as in Theorem \[Th1.blowup\]. Note, however, that the dissipative estimate is too weak in order to verify that the higher norms of solutions do not blow up in finite time. To overcome this obstacle, we need either to obtain stronger dissipative estimate (which we actually do not know how to do at the moment) or somehow regularize equations . For instance, it can be done in the spirit of the $\alpha$-models for the Navier-Stokes equations (see [@FHT]), namely, let us consider the following regularized system: $$\label{1.eq4} \begin{cases} \partial_t u_1+\partial_{x_1}(u_1^2)+\partial_{x_2}(u_1v_2)=\Delta u_1+k_1 u_1^2+\bar f_1(u)+g_1,\\ \partial_t u_2+\partial_{x_2}(u_2^2)+\partial_{x_1}(u_2v_1)=\Delta u_2+k_2 u_2^2+\bar f_2(u)+g_2,\\ v=(1-\alpha\Delta)^{-1}u,\ \ u,v\big|_{\partial\Omega}=0,\ \ u\big|_{t=0}=u_0, \end{cases}$$ where $\alpha>0$ is a regularization parameter. Then, on the one hand, repeating word by word the proof of Theorem \[Th1.bur3\], we see that the solution satisfies the dissipative estimate uniformly with respect to $\alpha$. On the other hand, the dissipative estimates in higher norms can be obtained from using the standard bootstrapping arguments. Indeed, since the key convective nonlinearity in is similar to the Navier-Stokes one, it is enough to deduce the dissipative $L^2$ estimate which is done in the following corollary. \[Cor1.bur-l2\] Under the above assumptions, problem is globally well-posed in $L^2(\Omega)$ and the following estimate holds: $$\label{1.est-bur4} \|u(t)\|^2_{L^2}+\int_t^{t+1}\|\nabla_x u(s)\|^2_{L^2}\,ds\le Q(\|u_0\|_{L^2})e^{-\gamma t}+Q(\|g\|_{L^2}),$$ where $\gamma>0$ and the monotone function $Q$ is independent of $u_0$ and $t$. Multiplying equations by $u$ and integrate over $x\in\Omega$ after the integration by parts, we get $$\label{1.est-dif1} \frac12\frac d{dt}\|u\|^2_{L^2}+\|\nabla_x u\|^2_{L^2}=\frac12\((u_1^2,\partial_{x_2}v_2)+(u_2^2,\partial_{x_1}v_1)\)+(f(u),u)+(g,u).$$ The first terms on the right-hand side of can be estimated using the embedding $H^2\subset C$ and the maximal regularity of the Laplace operator: $$\frac12\(|(u_1^2,\partial_{x_2}v_2)|+|(u_2^2,\partial_{x_1}v_1)\)|\le\|u\|^2_{L^2}\|\nabla v\|_{L^\infty}\le C\|u\|^2_{L^2}\|\nabla u\|_{L^2}\le \frac14\|\nabla u\|^2+C\|u\|^2_{L^2}.$$ Analogously using that $f(u)$ has at most quadratic growth together with the interpolation inequality, we get $$|(f(u),u)|\le C(\|u\|^3_{L^3}+1)\le C(1+\|u\|_{L^2}^2\|\nabla u\|_{L^2})\le \frac14\|\nabla u\|^2_{L^2}+C(\|u\|^4_{L^2}+1).$$ Thus, reads $$\label{1.est-dif2} \frac d{dt}\|u\|^2_{L^2}+\|\nabla_x u\|^2_{L^2}\le C\|u\|^4_{L^2}+C(\|g\|^2_{L^2}+1).$$ Inequality together with the dissipative estimate are enough to verify the desired estimate and finish the proof of the corollary. Indeed, for $t\in[0,1]$, we get by applying the Gronwall inequality to and using that the norm $\int_0^1\|u(s)\|^2_{L^2}\,ds$ is under the control. To obtain the estimate for $t\ge1$, we use the smoothing property $$\|u(1)\|^2_{L^2}\le Q(\|u(0)\|_{L^1})+Q(\|g\|_{L^2})$$ which follows again from by multiplying it by $t$ and applying the Gronwall inequality. From this estimate, we derive the estimate $$\|u(t+1)\|^2_{L^2}\le Q(\|u(t)\|_{L^1})+Q(\|g\|_{L^2})$$ which together with gives the desired dissipative estimate for $t\ge1$. Thus, the corollary is proved. Fourth order equations with convective terms {#s2} ============================================ In this section, we apply the method of weighted estimates to some classes of fourth order parabolic equations with convective terms. We start with the following generalized Kuramoto-Sivashinsky equation in $\Omega=(-1,1)$: $$\label{2.eq1} \partial_t u+\partial_x^4u+\lambda\partial_{x}^2u+\partial_x(u|u|^p)=f(u)+g,\ \ u\big|_{x=\pm1}=\partial_x^2 u\big|_{x=\pm1}=0,\ \ u\big|_{t=0}=u_0,$$ where $\lambda\in \R$ and $p>0$ are some parameters, $g\in L^2(\Omega)$ is a given external force and $f\in C^1(\R,\R)$ is a given nonlinearity satisfying the growth restriction $$\label{2.f} |f'(s)|\le C(1+|s|^q), \ \ \forall s\in \R,$$ for some $q>0$. The next result gives the analogue of Theorem \[Th1.bur\] for this equation. \[Th2.KS\] Let the nonlinearity $f$ satisfy assumption for some $q\le p$ and $g\in L^2(\Omega)$. Then, any sufficiently regular solution $u(t)$ of problem possesses the following estimate: $$\label{2.est-dis-KS} \|u(t)\|^2_{L^2}+\int_t^{t+1}\left(\|\partial_{x}^2 u(s)\|^2_{L^2}+\|u(s)\|^{p+2}_{L^{p+2}}\right)\,ds\le C\|u_0\|_{L^2}^2e^{-\alpha t}+C(\|g\|^2_{L^2}+1),$$ where the positive constants $\alpha$ and $C$ are independent of $t$, $u_0$ and $g$. We multiply equation by $ue^{-Lx}$, integrate by parts and use that $$(\partial_x(u|u|^p),ue^{-Lx})=\frac{p+1}{p+2}(\partial_x(|u|^{p+2}), e^{-Lx})=\frac{p+1}{p+2}L(|u|^{p+2},e^{-Lx}).$$ Then, we get $$\begin{gathered} \label{2.equal} \frac12\frac d{dt}\|u\|^2_{L^2_{e^{-Lx}}}+\|\partial_{x}^2u\|_{L^2_{e^{-Lx}}}^2+\frac{L^4+\lambda L^2}2\|u\|^2_{L^2_{e^{-Lx}}}+\\+\frac{p+1}{p+2}L(|u|^{p+2},e^{-Lx})=(L^2+\lambda)\|\partial_x u\|^2_{L^2_{e^{-Lx}}}+(f(u),ue^{-Lx})+(g,ue^{-Lx}).\end{gathered}$$ Using assumption on the nonlinearity $f$ together with the fact that $p\ge q$, we may fix $L$ in such way that the nonlinear term on the right-hand side of can be estimated by the convective term on the left-hand side. Using also the interpolation inequality between $H^2$ and $L^2$ and the fact that $p>0$, we finally arrive at $$\label{2.36} \frac d{dt}\|u\|^2_{L^2_{e^{-Lx}}}+\|u\|^2_{L^2_{e^{-Lx}}}+\alpha(\|\partial_x^2 u\|^2_{L^2}+\|u\|^{p+2}_{L^{p+2}})\le C_L(\|g\|^2_{L^2}+1)$$ and the Gronwall inequality applied to this relation gives the desired estimate and finishes the proof of the theorem. Note that, analogously to the previous section, estimate is not strong enough to prevent blow up in finite time for the higher norms of the solution $u(t)$ for large exponents $p$. However, for sufficiently small values of $p$ it is sufficient to verify the global existence of smooth solutions by the standard bootstrapping arguments. Since $H^1\subset L^\infty$ in the 1D case, it is sufficient to obtain the dissipative estimate in the $H^1$-norm which is done in the following corollary. \[Cor2.KS-h1\] Let the assumptions of Theorem \[Th2.KS\] hold and let, in addition $p\le6$. Then, problem is globally well-posed in $H^1_0$ and the following estimate holds: $$\label{2.est-KS-h1} \|u(t)\|_{H^1}\le Q(\|u_0\|_{H^1})e^{-\alpha t}+Q(\|g\|^2_{L^2}),$$ where the positive constant $\alpha$ and a monotone function $Q$ are independent of $u$ and $t$. We multiply equation by $\partial_x^2u$ and integrate over $x$. This gives $$\label{2.h1-est} \frac12\frac d{dt}\|\partial_x u\|^2_{L^2}+\|\partial_x^3 u\|^2_{L^2}\le\lambda\|\partial_x u\|^2_{L^2}+(p+1)(|u|^p\partial_xu,\partial_x^2u)-(f(u),\partial_x^2u)-(g,\partial_x^2u).$$ We assume for simplicity that $p\ge1$ and estimate below only the most complicated second term in the right-hand side of (the other terms are of lower order and are simpler to estimate). To this end, we integrate it by parts once more and use the interpolation inequalities $$\|u\|_{L^\infty}\le C\|u\|_{L^2}^{3/4}\|\partial_x^2u\|^{1/4}_{L^2},\ \ \|\partial_x u\|_{L^\infty}\le \|u\|_{L^2}^{1/4}\|\partial_{x}^2u\|_{L^2}^{3/4}$$ and obtain $$\begin{gathered} |(p+1)(|u|^p\partial_x u,\partial_x^2u)|\le C(|u|^{p-1},|\partial_x u|^3)\le C\|u\|_{L^\infty}^{p-1}\|\partial_x u\|_{L^\infty}\|\partial_x u\|^2_{L^2}\le\\\le C\|u\|_{L^2}^{\frac{3p-2}4}\|\partial_x^2 u\|_{L^2}^{\frac{p+2}4}\|\partial_x u\|^2_{L^2}.\end{gathered}$$ Thus, in order to control this term, we need the function $t\to\|u(t)\|_{L^2}^{\frac{3p-2}4}\|\partial_x^2 u(t)\|_{L^2}^{\frac{p+2}4}$ to be integrable. According to estimate it will be indeed the case if $p\le6$ and the corollary is proved. Let us consider now the case where $q>p$. In this case, the solution $u(t)$ may blow up in finite time at least for some initial data $u_0$ and nonliearity $f(u)$. We demonstrate it on the following example: $$\label{2.eq2} \partial_t u+\partial_x^4u+\lambda\partial_{x}^2u+\partial_x(u|u|^p)=|u|^{q+1}+g,\ \ u\big|_{x=\pm1}=\partial_x^2 u\big|_{x=\pm1}=0,\ \ u\big|_{t=0}=u_0.$$ \[Th2.blowup-KS\] Let $q>p$, $g\in L^2(\Omega)$ and $q>0$. Then, there exists smooth initial data $u_0$ such that the corresponding solution $u(t)$ of problem blows up in finite time. The proof of this fact is based on the multiplication of equation by $\varphi(x):=(\eb^2-x^2)^n$ and repeats word by word the proof of theorem \[Th1.blowup\]. Indeed, the only new term is $(u,\varphi'''')$ which appear after the integration by parts in the term containing fourth derivative of $u$. This term can be estimated similar to if $n>\frac{4(q+1)}{q}$. So, leaving the details to the reader, we finish the proof of the theorem. We now switch to the so-called convective Cahn-Hilliard equation which gives another model example of fourth order 1D equations where the convective terms may prevent blow up. Namely, let us consider the following problem in $\Omega=(-1,1)$: $$\label{2.eq3} \partial_{t} u+\partial_x^2\(\partial_x^2 u+f(u)\)+\partial_x(u|u|^p)=g,\ \ u\big|_{x=\pm1}=\partial_x^2u\big|_{x=\pm1}=0,\ \ u\big|_{t=0}=u_0,$$ where the nonlinearity $f$ satisfies assumptions and $g\in L^2(\Omega)$. In this case, the destabilizing nonlinearity $\partial_x^2f(u)$ is stronger than in the previous examples and we need stronger convective term in order to compensate it. The next theorem is the analogue of Theorem \[Th2.KS\] for this equation. \[Th2.KH\] Let $g\in L^2(\Omega)$ and the nonlinearity $f$ satisfy . Assume also that $p\ge2q$. Then, for any sufficiently smooth solution $u(t)$ of equation , the following dissipative estimate holds: $$\label{2.dis-KH} \|u(t)\|^2_{L^2}+\int_t^{t+1}\|\partial_{x}^2 u(s)\|^2_{L^2}+\|u(s)\|^{p+2}_{L^{p+2}}\,ds\le C\|u_0\|_{L^2}^2e^{-\alpha t}+C(\|g\|^2_{L^2}+1),$$ where the positive constants $\alpha$ and $C$ are independent of $u_0$, $t$ and $g$. As before, we multiply equation by $ue^{-Lx}$ and integrate over $x$. Then, analogously to , we get $$\begin{gathered} \label{2.equal-KH} \frac12\frac d{dt}\|u\|^2_{L^2_{e^{-Lx}}}+\|\partial_{x}^2u\|_{L^2_{e^{-Lx}}}^2+ \frac{L^4}2\|u\|^2_{L^2_{e^{-Lx}}}+\frac{p+1}{p+2}L(|u|^{p+2},e^{-Lx})=L^2\|\partial_x u\|^2_{L^2_{e^{-Lx}}}+\\+(f(u),\partial_x^2u e^{-Lx})+L^2(f(u)u+\Phi(u),e^{-Lx})+(g,ue^{-Lx}),\end{gathered}$$ where $\Phi(u):=\int_0^uf(v)\,dv$. Using now assumption , we see that $$|(f(u),\partial_x^2u e^{-Lx})|+|L^2(f(u)u+\Phi(u),e^{-Lx})|\le C\|u\|^{2q+2}_{L^{2q+2}_{e^{-Lx}}}+\frac12\|\partial_x^2 u\|^2_{L^2_{e^{-Lx}}}+CL^2\|u\|^{q+2}_{L^{q+2}_{e^{-Lx}}}.$$ Since $p\ge2q$, we may fix $L$ large enough to absorb the right-hand side of the last estimate by the convective term in the left-hand side of and end up (with the help of Young inequality) with estimate . This together with the Gronwall inequality give the desired estimate and finish the proof of the theorem. As in the case of Kuramoto-Sivashinsky equation, estimate is not strong enough to prevent blow up in higher norms if $p$ is large and some further restrictions on the exponents $p$ and $q$ are necessary for that. As in the case of Kuramoto-Sivashinsky equation, it is sufficient to verify the dissipative estimate in $H^1_0$-norm and the control of the higher norms will be obtained from it by the bootstrapping arguments. This estimate is given in the next corollary. \[Cor2.KH-dis-h1\] Let $g\in L^2(\Omega)$, the nonlinearity $f$ satisfy and, in addition, $p\le6$. Then, for any $u_0\in H^1_0(\Omega)$ the problem is globally well-posed in $H^1$ and estimate holds. Analogously to the proof of Corollary \[Cor2.KS-h1\], we multiply equation by $\partial_x^2 u$ and integrate over $\Omega$. Then, we only need to estimate two terms containing nonlinearities. The convective nonlinearity $\partial_x(u|u|^p)$ is estimated in the proof of Corollary \[Cor2.KS-h1\] and it is under the control if $p\le6$. The second nonlinearity $\partial_{x}^2f(u)$ can be estimated using assumption , integration by parts and the interpolation inequlaity as follows $$\begin{gathered} |(\partial_x^2 f(u),\partial_x^2 u)|=|(f'(u)\partial_x u,\partial_x^3u)|\le C(\|u\|^{2q}_{L^\infty}+1)\|\partial_xu\|^2_{L^2}+\\+\frac12\|\partial_x^3u\|^2_{L^2}\le C(\|u\|_{L^2}^{3q/2}\|\partial_x^2u\|^{q/2}_{L^2}+1)\|\partial_xu\|^2_{L^2}+ \frac12\|\partial_x^3u\|^2_{L^2}.\end{gathered}$$ Thus, keeping in mind the dissipative estimate , we see that this term is also under the control if $q\le4$. This condition is automatically satisfied since due to the assumption $q\le p/2\le6$ and the corollary is proved. To conclude this section, we give the analogue of Theorem \[Th2.blowup-KS\] for the case of the Cahn-Hilliard equation. As in the previous case, we consider the problem $$\label{2.eq4} \partial_tu+\partial_x^2(\partial_x^2u+|u|^{q+1})+\partial_x(u|u|^p)=g,\ \ u\big|_{x=\pm1}=\partial_x^2u\big|_{x=\pm1}=0,\ \ u\big|_{t=0}=u_0.$$ \[Th2.blowup-KH\] Let $g\in L^2(\Omega)$ and $p<q-1$. Then, there exist smooth initial data $u_0$ such that the corresponding solution $u(t)$ of equation blows up in finite time. We multiply equation by the function $\varphi(x):=(1-x^2)(x^4-14x^2+61)$ and integrate over $x\in(-1,1)$. Then, integrating by parts and using the boundary conditions, we have $$\label{2.blow} \frac d{dt}(u,\varphi)+(|u|^{q+1},\varphi'')+(u,\varphi'''')=(u|u|^p,\varphi')+(g,\varphi).$$ Using also the obvious fact that $$\varphi''(x)=-30(1-x^2)(5-x^2),\ \ \varphi''''(x)=360(1-x^2),$$ we see that the linear term in can be absorbed by the term containing $|u|^{q+1}$ and we end up with $$\label{2.blow1} \frac d{dt}(u,\varphi)\ge\alpha(|u|^{q+1},\varphi)-C(|u|^{p+1},1)-C$$ for some positive constants $\alpha$ and $C$. Thus, we only need to estimate the second term in the right-hand side of . We will do this by Hölder inequality as follows: $$(|u|^{p+1},1)=(|u|^{p+1}\varphi^{\frac1{q+1}},\varphi^{-\frac1{q+1}})\le \(|u|^{q+1},\varphi\)^{\frac{p+1}{q+1}}\(\varphi^{-\frac{1}{q-p}},1\)^{\frac{q-p}{q+1}}.$$ Since $p<q-1$, the last integral in the right-hand side is finite and applying the Young inequality, we finally arrive at $$\frac d{dt}(u,\varphi)\ge \beta|(u,\varphi)|^{q+1}-C$$ for some positive $\beta$ and $C$ which finishes the proof of the theorem. \[Rem2.bad\] In contrast to the case of Kuramoto-Sivashinsky equation and reaction diffusion equations considered before, the obtained results are not complete. In particular, the behavior of solutions remain unclear if $$\frac p2< q\le p+1.$$ The similar equation with periodic boundary conditions has been studied in [@EdKaZe] using different methods. In particular, the examples of blowing up solutions for the case $q=2$ and $p=1$ were given there. The analogous method works for our case as well, but it gives the condition $q\ge 2p$ for the blow up which is weaker than our result if $p>1$. KdV type equations {#s3} ================== In this section, we apply the above developed methods to study the generalized KdV equation equipped with Dirichlet boundary conditions: $$\label{3.eq1} \partial_t u+\partial_x^3u=\partial_x(u|u|^p)+f(u)+g,\ \ u\big|_{x=-1}=u\big|_{x=1}=\partial_x u\big|_{x=1}=0,\ \ u\big|_{t=0}=u_0,$$ where the nonlinearity $f$ satisfies assumptions . We first note that, in contrast to the case of periodic boundary conditions, the KdV type equations with Dirichlet boundary conditions are not conservative and the corresponding linear equation possesses a smoothing property on a finite interval and even generates a $C^\infty$-semigroup, see [@KdV] for the details. By this reason, the local solvability properties of these equations are close to the parabolic ones, so we will concentrate only on the derivation of the proper a priori estimates. The next theorem is the analogue of Theorem \[Th1.bur\] for this equation. Let $g\in L^2(\Omega)$, $p>0$ and $p\ge q$. Then, any sufficiently regular solution of possesses the following estimate: $$\label{3.dis-l2} \|u(t)\|^2_{L^2}+\int_t^{t+1}\|\partial_x u(s)\|^2_{L^2}\,ds\le C\|u_0\|^2_{L^2}e^{-\alpha t}+C(\|g\|^2_{L^2}+1),$$ where the positive constants $C$ and $\alpha$ are independent of $u_0$, $g$ and $t$. As before, we multiply the equation by $ue^{Lx}$ and integrate over $\Omega$. Then, using the obvious integration by parts $$\begin{gathered} (\partial_x^3 u,ue^{Lx})=-(\partial_x^2 u,e^{Lx}\partial_x u)-L(\partial_x^2 u,ue^{Lx})= \frac L2(|\partial_x u|^2,e^{Lx})+\\+L(|\partial_x^2 u|^2,e^{Lx})+L^2(\partial_x u,ue^{Lx})+e^{-L}|\partial_x u(-1)|^2=\\=\frac32 L(|\partial_x u|^2,e^{Lx})-\frac{L^3}{2}(|u|^2,e^{Lx})+e^{-L}|\partial_x u(-1)|^2\end{gathered}$$ and using assumptions on the nonlinearity $f$, we end up with $$\begin{gathered} \label{1.est-dif} \frac12\frac d{dt}\|u\|^2_{L^2_{e^{Lx}}}+\frac{3L}2\|\partial_x u\|^2_{L^2_{e^{Lx}}}+\frac{p+1}{p+2}L\|u\|^{p+2}_{L^{p+2}_{e^{Lx}}}\le\\\le C(\|u\|^{q+2}_{L^{q+2}_{e^{Lx}}}+1+\|g\|^2_{L^2_{e^{Lx}}})+C(L^3+1)\|u\|^2_{L^2_{e^{Lx}}}.\end{gathered}$$ Since $p>0$ and $p\ge q$, we may fix $L$ large enough that the terms containing $u$ on the right-hand side will be absorbed by the $L^{p+2}$-norm in the left-hand side. Applying then the Gronwall inequality to the obtained relation, we get the desired estimate and finish the proof of the theorem. As before, the dissipative estimate give the global well-posedness and smoothness of solutions of if $p$ is small enough. \[Co3.smooth\] Let $g\in L^2(\Omega)$, $0<p\le 2$ and $q\le p$. Then, problem is globally well-posed in $L^2(\Omega)$. Moreover, $u(t)\in H^3$ for all $t>0$ and the following estimate holds: $$\label{3.h3-dis} \|u(t)\|_{H^3}^2\le \frac{1+t^3}{t^3}\(Q(\|u_0\|_{L^2})e^{-\alpha t}+Q(\|g\|_{L^2})\),$$ for some positive $\alpha$ and monotone function $Q$. We give below only the formal derivation of the corresponding estimates which can be justified in a straightforward way. We start with the uniqueness. Let $u_1(t)$ and $u_2(t)$ be two slutions of and let $v=u_1-u_2$. Then, this function solves $$\label{3.dif} \partial_t v+\partial_x^3 v=\partial_x(u_1|u_1|^p-u_2|u_2|^p)+f(u_1)-f(u_2).$$ Multiplying this equation by $ve^{Lt}$ using assumptions and arguing as before, we end up with $$\label{3.dif1} \frac12\frac d{dt}\|v\|^2_{L^2_{e^{Lx}}}+\alpha \|\partial_x v\|^2_{L^2_{e^{Lx}}}+\alpha(|u_1|^p+|u_2|^p,v^2e^{Lx})\le C(\|u_1\|^{2p}_{L^\infty}+\|u_2\|^{2p}_{L^\infty}+1)\|v\|^2_{L^2_{e^{Lx}}}$$ for some positive $C$ and $\alpha$ which are independent of $u_1$ and $u_2$. Using the interpolation inequality $\|u\|_{L^\infty}^2\le \|u\|_{L^2}\|\partial_x u\|_{L^2}$ together with the dissipative estimate and assumption $p\le2$, we see that the $L^1$-norm in time of $\|u_i\|^{2p}_{L^\infty}$ is under the control and the Gronwall inequality applied to gives $$\label{3.dif2} \|v(t)\|^2_{L^2}\le C\|v(0)\|^2_{L^2}e^{Kt},$$ where the constants $C$ and $K$ may depend on the $L^2$-norms of $u_i$. This proves the uniqueness. To verify the smoothing property , we differentiate equation in time and denote $w=\partial_t u$. Then this function solves $$\label{3.dif4} \partial_t w+\partial_x^3 w=(p+1)\partial_x(|u|^pw)+f'(u)w.$$ Multiplying this equation by $we^{Lx}$ and arguing analogously, we end up with $$\label{3.dif5} \frac12\frac d{dt}\|w\|^2_{L^2_{e^{Lx}}}+\alpha \|\partial_x w\|^2_{L^2_{e^{Lx}}}+\alpha(|u|^p,w^2e^{Lx})\le C(\|u\|^{2p}_{L^\infty}+1)\|w\|^2_{L^2_{e^{Lx}}}.$$ Applying the Gronvall inequality to this relation, we get $$\label{3.dif6} \|\Dt u(t)\|^2_{L^2}\le Q(\|u_0\|^2_{L^2})e^{K(\|u_0\|_{L^2})t}\|\Dt u(0)\|^2_{L^2}$$ for some monotone functions $Q$ and $K$. Using the obvious estimate $$\|\Dt u(0)\|_{L^2}\le Q(\|u_0\|_{H^3}),$$ we see that the $L^2$-norm of $\partial_t u(t)$ is under the control. Let us derive the opposite estimate. To this end, we multiply equation by $\partial_x^3u$ and use the Hölder inequality to obtain $$\label{3.l2h3} \|\partial_x^3u(t)\|^2_{L^2}\le C\|\partial_t u(t)\|^2_{L^2}+C(\|u(t)\|^{4}_{L^\infty}(\|\partial_x u(t)\|^2_{L^2}+1)+\|g\|^2_{L^2}+1).$$ Using the obvious interpolation, we get $$\|u\|^{4}_{L^\infty}\|\partial_x u\|^2_{L^2}\le C\|u\|^{10/3}_{L^2}\|\partial_x^3 u\|_{L^2}^{2/3}\|u\|_{L^2}\|\partial_x^3u\|_{L^2}=C\|u\|_{L^2}^{13/3}\|\partial_x^3u\|_{L^2}^{5/3}$$ and, therefore, $$\label{3.dif7} \|\partial_x^3 u(t)\|^2_{L^2}\le C(1+\|\partial_t u(t)\|_{L^2}^2+\|g\|^2_{L^2}+\|u(t)\|^{26}_{L^2}).$$ Combining estimates and , we get the $H^3$-control of the solution $u(t)$: $$\label{3.dif8} \|u(t)\|_{H^3}\le Q(\|u_0\|_{H^3}+\|g\|_{L^2})e^{K(\|u_0\|_{L^2})t}.$$ However, this estimate is still not dissipative. To obtain the dissipative estimate, we need the smoothing property for $\partial_t u$. To this end, we note that $$\|\partial_t u(t)\|_{H^{-2}}^2\le C(\|\partial_x u(t)\|_{L^2}^2+\|u|u|^p\|_{H^{-1}}^2+\|f(u)\|_{H^{-2}}^2+\|g\|^2_{L^2})$$ and $$|(u|u|^p,\varphi)|\le(\|u\|^{3}_{L^3}+1)\|\varphi\|_{L^\infty}\le (\|u\|^3_{L^3}+1)\|\varphi\|_{H^1_0}.$$ Therefore, $$\|u|u|^p\|_{H^{-1}}^2\le C(\|u\|^6_{L^3}+1)\le C(\|u\|^4_{L^2}\|\partial_x u\|^2_{L^2}+1).$$ Estimate for the lower order term $f(u)$ can be obtained analogously and we arrive at $$\label{3.eq9} \|\Dt u(t)\|^2_{H^{-2}}\le C\((\|u\|^4_{L^2}+1)\|\partial_x u(t)\|_{L^2}^2+1+\|g\|^2_{L^2}\).$$ We are now ready to finish the proof of the smoothing property. Indeed, we multiply by $t$ and use the interpolation $$\label{3.dif-infty} 3t^{2}\|v(t)\|^2_{L^2_{e^{Lx}}}\le C\|v(t)\|_{H^{-2}}^{2/3}(t^{3/2}\|v(t)\|_{H^1})^{4/3}\le \frac\alpha2t^3\|\partial_x v(t)\|^2_{L^2_{e^{Lx}}}+C\|\partial_t u(t)\|^2_{H^{-2}}.$$ Then, combining the last estimate with we get $$\begin{gathered} \label{3.dif10} \frac d{dt}(t^3\|v\|^2_{L^2_{e^{Lx}}})+\alpha t^3\|\partial_x v\|^2_{L^2_{e^{Lx}}}\le Ct^3(\|u\|^{2}_{L^2}\|\partial_x u\|^2_{L^2}+1)\|v\|^2_{L^2_{e^{Lx}}}+\\+C\((\|u\|^4_{L^2}+1)\|\partial_x u(t)\|_{L^2}^2+1+\|g\|^2_{L^2}\).\end{gathered}$$ Applying the Gronwall inequality to this relation and using the dissipative estimate , we finally arrive at $$\label{3.dis11} t^3\|\partial_t u(t)\|^2_{L^2}\le Q(\|u_0\|_{L^2})+Q(\|g\|_{L^2}),\ \ t\in(0,1]$$ for some monotone function $Q$. In particular, fixing $t=1$ in this inequality, we have $$\label{3.dis12} \|\partial_t u(t+1)\|_{L^2}\le Q(\|u(t)\|_{L^2})+Q(\|g\|_{L^2}).$$ It only remains to note that estimates and together with and the dissipative estimate give the desired estimate and finish the proof of the corollary. Finally, we show that the blow up in finite time of smooth solutions of becomes possible if the condition $p\ge q$ is violated. To this end, we consider the following equation: $$\label{3.eq2} \partial_t u+\partial_x^3 u=\partial_x(u|u|^p)+|u|^{q+1}+g, \ \ u\big|_{x=-1}=u\big|_{x=1}=\partial_x u\big|_{x=1}=0,\ \ u\big|_{t=0}=u_0.$$ Then, the following result holds. \[Th3.blow\] Let $g\in L^2(\Omega)$, $q>0$ and $p<q$. Then, there exists somooth initial data $u_0$ such that the corresponding solution $u(t)$ of blows up in finite time. Indeed, the proof of this result is based on the multiplication of by $\varphi(x):=(\eb^2-x^2)^n$ for sufficiently large $n$ and $\eb>0$ and repeats word by word the proof of Theorem \[Th2.blowup-KS\], so we leave the details to the reader. \[Rem3.difference\] We remind that the result of Corollary \[Co3.smooth\] on the global existence of smooth solutions contains a restriction $p\le2$ (even in the case where $f(u)\equiv 0$) which looks unusual and too strict in comparison with the classical theory of KdV equations on the whole line or on the circle. Indeed, in the defocusing case of a generalized KdV equation: $$\label{3.df} \partial_t u+\partial_x^3 u=\partial_x(u|u|^p)$$ which is considered in this section, the corresponding Cauchy problem or periodic boundary value problem is globally well-posed in $H^1$ for any $p\ge0$, see [@BoSa; @Tao] and references therein. This is related with the following energy conservation law for problem : $$\label{3.conservation} \partial_t\(\frac12|\partial_x u|^2+\frac1{p+2}|u|^{p+2}\)-\partial_x\(\partial_tu\partial_xu+\frac12(\partial_{x}^2u-u|u|^p)^2\)=0.$$ Moreover, in the self-focusing case $$\partial_tu+\partial_x^3u+\partial_x(u|u|^p)=0,$$ where the energy is not sign-definite, the critical exponent is $p=4$, namely, if $p<4$, the global existence of smooth solutions can be established and for $p=4$ the blow up of higher norms is occurred at least for the Caucgy problem with properly chosen initial data, see [@MaMe] and references therein. Unfortunately, identity is not very helpful for the case of Dirichlet boundary conditions since its integration over $x$ leads to the extra term $\frac12|\partial_x^2u|^2\big|_{x=-1}^{x=1}$ which is out of control. Thus, we are unable to use the energy identity and this, in turn, leads to the restriction $p\le2$. Actually, we do not know whether or not this restriction is essential. \[Rem3.rev\] One more essential difference with the conservative case is the solvability of the generalized KdV backward in time. In the case of Cauchy problem or periodic boundary value problem, the corresponding problem is [*reversible*]{} in time due to the invariance with respect to the transformation $t\to-t$, $x\to-x$. Thus, the global solvability backward in time follows from the global solvability forward in time. That is clearly not the case for the Dirichlet boundary conditions already due to the smoothing property established in Corollary \[Co3.smooth\]. Moreover, it can be shown that the only smooth solution of with Dirichlet boundary conditions defined for all $t\in\R$ is $u\equiv0$. Indeed, the above mentioned transformation does not change the equation , but changes the boundary conditions and leads to the problem $$\label{3.back} \partial_tu+\partial_x^3u=\partial_x(u|u|^p),\ \ u\big|_{x=-1}=\partial_xu\big|_{x=-1}=u\big|_{x=1}=0.$$ Multiplying this equation by $(1-x)u$ and integrating over $x$, we have $$\frac12\frac d{dt}\|u\|^2_{L^2_{1-x}}=\frac32\|\partial_xu\|^2_{L^2}+\frac{p+1}{p+2}\|u\|^{p+2}_{L^{p+2}}\ge C\|u\|_{L^2_{1-x}}^{p+2}$$ which gives the desired result. [9]{} A. Alshin, M. Korpusov, A. Sveshnikov, [*Blow up in Nonlinear Sobolev Type Equations*]{}, De Gruyter Series in Nonlinear Analysis and Applications, 2011. J. Ball, [*Remarks on blow-up and nonexistence theorems for nonlinear evolution equations*]{}, Quart. J. Math. Oxford Ser. (2) [**28**]{}, no. 112, (1977), 473–486. H. Bellout, S. Benachour and E. Titi, [*Finite-time singularity versus global regularity for hyper-viscous Hamilton-Jacobi-like equations.*]{} Nonlinearity [**16**]{} (2003), 1967–1989. J. Bona and J. Saut, [*Dispersive blow up of solutions of generalized Korteweg-de Vries equations.*]{} J. Differential Equations [**103**]{} (1993), 3–57. J. Bona, S. Sun and B, Zhang, [*A Nonhomogeneous Boundary-Value Problem for the Korteweg–de Vries Equation Posed on a Finite Domain*]{}, Comm. in PDEs, Vol 28, No. 7-8, (2003), 1391–1436, T-F. Chen, H. Levine, P. Sacks, [*Analysis of a convective reaction-diffusion equation*]{}, Nonlinear Anal. [**12**]{}, (1988), 1349–1370. A. Eden, V. Kalantarov, [*3D convective Cahn-Hilliard equation*]{}, Commun. Pure Appl. Anal. [**6**]{}, (2007), 1075–1086. A. Eden, V. Kalantarov, S. Zelik, [*Global solvability and blow up for the convective Cahn-Hilliard equations with concave potentials*]{}, J. Math. Phys. [**54**]{}, 041502, (2013), 1–12. C. Foias, D. Holm, E. Titi, [*The Navier-Stokes-alpha model of fluid turbulence*]{}, Physica D, vol. 152–153, (2001), 505–519. A. Fursikov, [*On the Normal Semilinear Parabolic Equations Corresponding to 3D Navier-Stokes System*]{}, in System Modeling and Optimization, IFIP Advances in Communication and Information Technology, Vol. 391, (2013), 338–347. V. Galaktionov, E. Mitidieri and S. Pohozhaev, [*Blow up for Higher Order Parabolic, Hyperbolic and Schrödinger Equations*]{}, CRC Press, Boca Raton-London-New-York, 2015. P. Hislop and I. Sigal, [*Introduction to Spectral Theory*]{}, Springer, 1996. V. Kalantarov, [*Global behaviour of the solutions of some fourth order nonlinear equations*]{}, Zap. Nauchn. Semin. LOMI [**163**]{}, (1987), 66–75 (in Russian). V. Kalantarov, O. Ladyzhenskaya, [*The occurrence of collapse for quasilinear equations of parabolic and hyperbolic type*]{}, J. Soviet Math. [**10**]{}, (1978), 53–70. O. Ladyzhenskaya, V. Solonnikov, N. Uraltseva, [*Linear and Quasilinear Equations of Parabolic Types*]{}, American Mathematical Society, Providence, RI, 1968. N. Larkin, [*Modified KdV equation with a source term in a bounded domain*]{}. Math. Methods Appl. Sci. [**29**]{}, no. 7, (2006), 751–765. H. Levine, [*Some nonexistence and instability theorems for solutions of formally parabolic equations of the form $Pu_t = Au + \mathcal{F}(u)$*]{}, Arch. Ration. Mech. Anal. [**51**]{}, (1973), 371–386. H. Levine, [*Stability and Instability for solutions of Burgers’ equation with a semilinear boundary condition*]{}, SIAM J. Math. Anal. [**19**]{}, (1988), 312–336. H. Levine and L. Payne, [*Nonexistence theorems for the heat equations with nonlinear boundary conditions and for porous medium equation backward in time*]{}, J. Diff. Eq. [**16**]{}, (1974), 319–334. H. Levine, L. Payne, P. Sacks, B. Straughan, [*Analysis of a convective reaction-diffusion equation II*]{}, SIAM J. Math. Anal. [**20**]{}, (1989), 133–147. Y. Martel and F. Merle, [*Blow up in finite time and dynamics of blow up solutions for the $L^2$-critical generalized KdV equation*]{}. J. Amer. Math. Soc. [**15**]{} (2002), 617–664. S. Pokhozhaev, [*On the blow-up of solutions of the Kuramoto-Sivashinsky equation*]{}, Sb. Math. [**199**]{}, (2008), 1355–1365. S. Pokhozhaev, [*On the nonexistence of global solutions of the Korteweg-de Vries equation*]{}. J. Math. Sci. (N. Y.) [**190**]{}, (2013), 147–156. P. Quittner and Ph. Souplet, [*Superlinear parabolic problems. Blow up, global existence and steady states*]{}, Birkhauser advanced texts, Basel-Boston-Berlin, 2007. P. Souplet and F. Weissler, [*Poincare’s inequality and global solutions of a nonlinear parabolic equation*]{}, Ann. Inst. H. Poincare, Anal. Non Lineaire [**16**]{}, (1999), 335–371. P. Souplet and F. Weissler, [*Self-similar subsolutions and blowup for nonlinear parabolic equations*]{}, J. Math. Anal. and Appl., [**212**]{}, (1997), 60–74. T. Tao, [*Nonlinear dispersive equations. Local and global analysis.*]{} American Mathematical Society, Providence, RI, 2006 A. Tersenov, [*The preventive effect of the convection and of the diffusion in the blow-up phenomenon for parabolic equations*]{}, Ann. Inst. H. Poincare, Anal. Non Lineaire [**21**]{}, (2004), 533–541. A. Tersenov, [*A condition guaranteeing the abscence of the blow-up phenomenon for the generalized Burgers equation*]{}, Nonlinear Anal. [**75**]{}, (2012), 5119–5122. E. Yushkov, [*Blowup in Korteweg-de Vries-type systems.*]{} Theoret. and Math. Phys. [**173**]{} (2012), 1498–1506. [^1]: This work is partially supported by the grant 14-41-00044 of RSF

ArXiv

plus 1pt [EUROPEAN ORGANIZATION FOR NUCLEAR RESEARCH ]{} CERN-EP/2002-043\ 18 June 2002 [**Search for Charged Excited Leptons\ in [$\rm e^+e^-$]{} Collisions at [$\sqrt{s}$]{} = 183-209 GeV** ]{} [The OPAL Collaboration ]{} [Abstract]{} A search for charged excited leptons decaying into a lepton and a photon has been performed using approximately 680 $\rm pb^{-1}$ of  collision data collected by the OPAL detector at LEP at [centre-of-mass]{} energies between 183 GeV and 209 GeV. No evidence for their existence was found. Upper limits on the product of the cross-section and the branching fraction are inferred. Using results from the search for singly produced excited leptons, upper limits on the ratio of the excited lepton coupling constant to the compositeness scale are calculated. From pair production searches, 95% confidence level lower limits on the masses of excited electrons, muons and taus are determined to be [103.2]{} GeV. [(To be submitted to Physics Letters B) ]{} [The OPAL Collaboration ]{} [ G.Abbiendi$^{ 2}$, C.Ainsley$^{ 5}$, P.F.kesson$^{ 3}$, G.Alexander$^{ 22}$, J.Allison$^{ 16}$, P.Amaral$^{ 9}$, G.Anagnostou$^{ 1}$, K.J.Anderson$^{ 9}$, S.Arcelli$^{ 2}$, S.Asai$^{ 23}$, D.Axen$^{ 27}$, G.Azuelos$^{ 18, a}$, I.Bailey$^{ 26}$, E.Barberio$^{ 8}$, R.J.Barlow$^{ 16}$, R.J.Batley$^{ 5}$, P.Bechtle$^{ 25}$, T.Behnke$^{ 25}$, K.W.Bell$^{ 20}$, P.J.Bell$^{ 1}$, G.Bella$^{ 22}$, A.Bellerive$^{ 6}$, G.Benelli$^{ 4}$, S.Bethke$^{ 32}$, O.Biebel$^{ 32}$, I.J.Bloodworth$^{ 1}$, O.Boeriu$^{ 10}$, P.Bock$^{ 11}$, D.Bonacorsi$^{ 2}$, M.Boutemeur$^{ 31}$, S.Braibant$^{ 8}$, L.Brigliadori$^{ 2}$, R.M.Brown$^{ 20}$, K.Buesser$^{ 25}$, H.J.Burckhart$^{ 8}$, J.Cammin$^{ 3}$, S.Campana$^{ 4}$, R.K.Carnegie$^{ 6}$, B.Caron$^{ 28}$, A.A.Carter$^{ 13}$, J.R.Carter$^{ 5}$, C.Y.Chang$^{ 17}$, D.G.Charlton$^{ 1, b}$, I.Cohen$^{ 22}$, A.Csilling$^{ 8, g}$, M.Cuffiani$^{ 2}$, S.Dado$^{ 21}$, G.M.Dallavalle$^{ 2}$, S.Dallison$^{ 16}$, A.De Roeck$^{ 8}$, E.A.De Wolf$^{ 8}$, K.Desch$^{ 25}$, M.Donkers$^{ 6}$, J.Dubbert$^{ 31}$, E.Duchovni$^{ 24}$, G.Duckeck$^{ 31}$, I.P.Duerdoth$^{ 16}$, E.Elfgren$^{ 18}$, E.Etzion$^{ 22}$, F.Fabbri$^{ 2}$, L.Feld$^{ 10}$, P.Ferrari$^{ 12}$, F.Fiedler$^{ 31}$, I.Fleck$^{ 10}$, M.Ford$^{ 5}$, A.Frey$^{ 8}$, A.Fürtjes$^{ 8}$, P.Gagnon$^{ 12}$, J.W.Gary$^{ 4}$, G.Gaycken$^{ 25}$, C.Geich-Gimbel$^{ 3}$, G.Giacomelli$^{ 2}$, P.Giacomelli$^{ 2}$, M.Giunta$^{ 4}$, J.Goldberg$^{ 21}$, E.Gross$^{ 24}$, J.Grunhaus$^{ 22}$, M.Gruwé$^{ 8}$, P.O.Günther$^{ 3}$, A.Gupta$^{ 9}$, C.Hajdu$^{ 29}$, M.Hamann$^{ 25}$, G.G.Hanson$^{ 4}$, K.Harder$^{ 25}$, A.Harel$^{ 21}$, M.Harin-Dirac$^{ 4}$, M.Hauschild$^{ 8}$, J.Hauschildt$^{ 25}$, C.M.Hawkes$^{ 1}$, R.Hawkings$^{ 8}$, R.J.Hemingway$^{ 6}$, C.Hensel$^{ 25}$, G.Herten$^{ 10}$, R.D.Heuer$^{ 25}$, J.C.Hill$^{ 5}$, K.Hoffman$^{ 9}$, R.J.Homer$^{ 1}$, D.Horváth$^{ 29, c}$, R.Howard$^{ 27}$, P.Hüntemeyer$^{ 25}$, P.Igo-Kemenes$^{ 11}$, K.Ishii$^{ 23}$, H.Jeremie$^{ 18}$, P.Jovanovic$^{ 1}$, T.R.Junk$^{ 6}$, N.Kanaya$^{ 26}$, J.Kanzaki$^{ 23}$, G.Karapetian$^{ 18}$, D.Karlen$^{ 6}$, V.Kartvelishvili$^{ 16}$, K.Kawagoe$^{ 23}$, T.Kawamoto$^{ 23}$, R.K.Keeler$^{ 26}$, R.G.Kellogg$^{ 17}$, B.W.Kennedy$^{ 20}$, D.H.Kim$^{ 19}$, K.Klein$^{ 11}$, A.Klier$^{ 24}$, S.Kluth$^{ 32}$, T.Kobayashi$^{ 23}$, M.Kobel$^{ 3}$, T.P.Kokott$^{ 3}$, S.Komamiya$^{ 23}$, L.Kormos$^{ 26}$, R.V.Kowalewski$^{ 26}$, T.Krämer$^{ 25}$, T.Kress$^{ 4}$, P.Krieger$^{ 6, l}$, J.von Krogh$^{ 11}$, D.Krop$^{ 12}$, M.Kupper$^{ 24}$, P.Kyberd$^{ 13}$, G.D.Lafferty$^{ 16}$, H.Landsman$^{ 21}$, D.Lanske$^{ 14}$, J.G.Layter$^{ 4}$, A.Leins$^{ 31}$, D.Lellouch$^{ 24}$, J.Letts$^{ 12}$, L.Levinson$^{ 24}$, J.Lillich$^{ 10}$, S.L.Lloyd$^{ 13}$, F.K.Loebinger$^{ 16}$, J.Lu$^{ 27}$, J.Ludwig$^{ 10}$, A.Macpherson$^{ 28, i}$, W.Mader$^{ 3}$, S.Marcellini$^{ 2}$, T.E.Marchant$^{ 16}$, A.J.Martin$^{ 13}$, J.P.Martin$^{ 18}$, G.Masetti$^{ 2}$, T.Mashimo$^{ 23}$, P.Mättig$^{ m}$, W.J.McDonald$^{ 28}$, J.McKenna$^{ 27}$, T.J.McMahon$^{ 1}$, R.A.McPherson$^{ 26}$, F.Meijers$^{ 8}$, P.Mendez-Lorenzo$^{ 31}$, W.Menges$^{ 25}$, F.S.Merritt$^{ 9}$, H.Mes$^{ 6, a}$, A.Michelini$^{ 2}$, S.Mihara$^{ 23}$, G.Mikenberg$^{ 24}$, D.J.Miller$^{ 15}$, S.Moed$^{ 21}$, W.Mohr$^{ 10}$, T.Mori$^{ 23}$, A.Mutter$^{ 10}$, K.Nagai$^{ 13}$, I.Nakamura$^{ 23}$, H.A.Neal$^{ 33}$, R.Nisius$^{ 8}$, S.W.O’Neale$^{ 1}$, A.Oh$^{ 8}$, A.Okpara$^{ 11}$, M.J.Oreglia$^{ 9}$, S.Orito$^{ 23}$, C.Pahl$^{ 32}$, G.Pásztor$^{ 8, g}$, J.R.Pater$^{ 16}$, G.N.Patrick$^{ 20}$, J.E.Pilcher$^{ 9}$, J.Pinfold$^{ 28}$, D.E.Plane$^{ 8}$, B.Poli$^{ 2}$, J.Polok$^{ 8}$, O.Pooth$^{ 14}$, M.Przybycień$^{ 8, j}$, A.Quadt$^{ 3}$, K.Rabbertz$^{ 8}$, C.Rembser$^{ 8}$, P.Renkel$^{ 24}$, H.Rick$^{ 4}$, J.M.Roney$^{ 26}$, S.Rosati$^{ 3}$, Y.Rozen$^{ 21}$, K.Runge$^{ 10}$, D.R.Rust$^{ 12}$, K.Sachs$^{ 6}$, T.Saeki$^{ 23}$, O.Sahr$^{ 31}$, E.K.G.Sarkisyan$^{ 8, j}$, A.D.Schaile$^{ 31}$, O.Schaile$^{ 31}$, P.Scharff-Hansen$^{ 8}$, J.Schieck$^{ 32}$, T.Schoerner-Sadenius$^{ 8}$, M.Schröder$^{ 8}$, M.Schumacher$^{ 3}$, C.Schwick$^{ 8}$, W.G.Scott$^{ 20}$, R.Seuster$^{ 14, f}$, T.G.Shears$^{ 8, h}$, B.C.Shen$^{ 4}$, C.H.Shepherd-Themistocleous$^{ 5}$, P.Sherwood$^{ 15}$, G.Siroli$^{ 2}$, A.Skuja$^{ 17}$, A.M.Smith$^{ 8}$, R.Sobie$^{ 26}$, S.Söldner-Rembold$^{ 10, d}$, S.Spagnolo$^{ 20}$, F.Spano$^{ 9}$, A.Stahl$^{ 3}$, K.Stephens$^{ 16}$, D.Strom$^{ 19}$, R.Ströhmer$^{ 31}$, S.Tarem$^{ 21}$, M.Tasevsky$^{ 8}$, R.J.Taylor$^{ 15}$, R.Teuscher$^{ 9}$, M.A.Thomson$^{ 5}$, E.Torrence$^{ 19}$, D.Toya$^{ 23}$, P.Tran$^{ 4}$, T.Trefzger$^{ 31}$, A.Tricoli$^{ 2}$, I.Trigger$^{ 8}$, Z.Trócsányi$^{ 30, e}$, E.Tsur$^{ 22}$, M.F.Turner-Watson$^{ 1}$, I.Ueda$^{ 23}$, B.Ujvári$^{ 30, e}$, B.Vachon$^{ 26}$, C.F.Vollmer$^{ 31}$, P.Vannerem$^{ 10}$, M.Verzocchi$^{ 17}$, H.Voss$^{ 8}$, J.Vossebeld$^{ 8}$, D.Waller$^{ 6}$, C.P.Ward$^{ 5}$, D.R.Ward$^{ 5}$, P.M.Watkins$^{ 1}$, A.T.Watson$^{ 1}$, N.K.Watson$^{ 1}$, P.S.Wells$^{ 8}$, T.Wengler$^{ 8}$, N.Wermes$^{ 3}$, D.Wetterling$^{ 11}$ G.W.Wilson$^{ 16, k}$, J.A.Wilson$^{ 1}$, G.Wolf$^{ 24}$, T.R.Wyatt$^{ 16}$, S.Yamashita$^{ 23}$, V.Zacek$^{ 18}$, D.Zer-Zion$^{ 4}$, L.Zivkovic$^{ 24}$ ]{} $^{ 1}$School of Physics and Astronomy, University of Birmingham, Birmingham B15 2TT, UK $^{ 2}$Dipartimento di Fisica dell’ Università di Bologna and INFN, I-40126 Bologna, Italy $^{ 3}$Physikalisches Institut, Universität Bonn, D-53115 Bonn, Germany $^{ 4}$Department of Physics, University of California, Riverside CA 92521, USA $^{ 5}$Cavendish Laboratory, Cambridge CB3 0HE, UK $^{ 6}$Ottawa-Carleton Institute for Physics, Department of Physics, Carleton University, Ottawa, Ontario K1S 5B6, Canada $^{ 8}$CERN, European Organisation for Nuclear Research, CH-1211 Geneva 23, Switzerland $^{ 9}$Enrico Fermi Institute and Department of Physics, University of Chicago, Chicago IL 60637, USA $^{ 10}$Fakultät für Physik, Albert-Ludwigs-Universität Freiburg, D-79104 Freiburg, Germany $^{ 11}$Physikalisches Institut, Universität Heidelberg, D-69120 Heidelberg, Germany $^{ 12}$Indiana University, Department of Physics, Swain Hall West 117, Bloomington IN 47405, USA $^{ 13}$Queen Mary and Westfield College, University of London, London E1 4NS, UK $^{ 14}$Technische Hochschule Aachen, III Physikalisches Institut, Sommerfeldstrasse 26-28, D-52056 Aachen, Germany $^{ 15}$University College London, London WC1E 6BT, UK $^{ 16}$Department of Physics, Schuster Laboratory, The University, Manchester M13 9PL, UK $^{ 17}$Department of Physics, University of Maryland, College Park, MD 20742, USA $^{ 18}$Laboratoire de Physique Nucléaire, Université de Montréal,$\;$ Montréal,$\;$Québec$\;$H3C$\;$3J7,$\;$Canada $^{ 19}$University of Oregon, Department of Physics, Eugene OR 97403, USA $^{ 20}$CLRC Rutherford Appleton Laboratory, Chilton, Didcot, Oxfordshire OX11 0QX, UK $^{ 21}$Department of Physics, Technion-Israel Institute of Technology, Haifa 32000, Israel $^{ 22}$Department of Physics and Astronomy, Tel Aviv University, Tel Aviv 69978, Israel $^{ 23}$International Centre for Elementary Particle Physics and Department of Physics, University of Tokyo, Tokyo 113-0033, and Kobe University, Kobe 657-8501, Japan $^{ 24}$Particle Physics Department, Weizmann Institute of Science, Rehovot 76100, Israel $^{ 25}$Universität Hamburg/DESY, Institut für Experimentalphysik, Notkestrasse 85, D-22607 Hamburg, Germany $^{ 26}$University of Victoria, Department of Physics, P O Box 3055, Victoria BC V8W 3P6, Canada $^{ 27}$University of British Columbia, Department of Physics, Vancouver BC V6T 1Z1, Canada $^{ 28}$University of Alberta, Department of Physics, Edmonton AB T6G 2J1, Canada $^{ 29}$Research Institute for Particle and Nuclear Physics, H-1525 Budapest, P O Box 49, Hungary $^{ 30}$Institute of Nuclear Research, H-4001 Debrecen, P O Box 51, Hungary $^{ 31}$Ludwig-Maximilians-Universität München, Sektion Physik, Am Coulombwall 1, D-85748 Garching, Germany $^{ 32}$Max-Planck-Institute für Physik, Föhringer Ring 6, D-80805 München, Germany $^{ 33}$Yale University, Department of Physics, New Haven, CT 06520, USA $^{ a}$ and at TRIUMF, Vancouver, Canada V6T 2A3 $^{ b}$ and Royal Society University Research Fellow $^{ c}$ and Institute of Nuclear Research, Debrecen, Hungary $^{ d}$ and Heisenberg Fellow $^{ e}$ and Department of Experimental Physics, Lajos Kossuth University, Debrecen, Hungary $^{ f}$ and MPI München $^{ g}$ and Research Institute for Particle and Nuclear Physics, Budapest, Hungary $^{ h}$ now at University of Liverpool, Dept of Physics, Liverpool L69 3BX, UK $^{ i}$ and CERN, EP Div, 1211 Geneva 23 $^{ j}$ and Universitaire Instelling Antwerpen, Physics Department, B-2610 Antwerpen, Belgium $^{ k}$ now at University of Kansas, Dept of Physics and Astronomy, Lawrence, KS 66045, USA $^{ l}$ now at University of Toronto, Dept of Physics, Toronto, Canada $^{ m}$ current address Bergische Universität, Wuppertal, Germany Introduction {#section:intro} ============ Models in which fermions have substructure attempt to explain, among other things, the well-ordered pattern of fermion generations observed in nature. The existence of excited states of the Standard Model fermions would be a natural consequence of fermion compositeness. Excited leptons could be produced in  collisions and are expected to decay via the emission of a gauge boson (,  or ) [@bib:old_theory_paper]. This paper presents results from a search for excited electrons (), muons () and tau leptons () decaying electromagnetically, using data collected by the OPAL experiment at LEP. Results presented in this paper are obtained using a larger data sample, as well as a significantly improved analysis, compared to our previous publications [@bib:ex_opal_results]. Searches for excited leptons have also been performed by other LEP collaborations [@bib:lep] and by the HERA experiments in electron-proton collisions [@bib:hera]. At LEP, excited leptons could be produced in pairs or in association with a Standard Model lepton. Both single and pair production of excited leptons proceed through [*s*]{}-channel photon and  diagrams. In addition, [*t*]{}-channel photon and  exchange diagrams contribute to the single and pair production of excited electrons. The [*t*]{}-channel contribution, although expected to be negligible for pair production, causes excited electrons to be singly produced predominantly in the forward region with the recoiling electron outside the detector acceptance. Thus, in addition to final states containing two leptons and one or two photons (, ), a separate search for the single production of excited electrons with an undetected electron () is also performed. The results presented in this paper are interpreted in the context of the framework described in [@bib:ex_theory1; @bib:ex_theory2]. In this phenomenological model, the interaction between excited leptons and a gauge boson ($\ell^*\ell^*V$), which largely determines the cross-section for pair production of excited leptons, is vector-like. The single production cross-section and branching fractions of excited leptons are determined by the strength of the $\ell\ell^*\rm V$ coupling. This interaction can be described by the following SU(2)$\times$U(1) gauge invariant effective Lagrangian [@bib:ex_theory1; @bib:ex_theory2] $${\cal L}_{\ell\ell^* V} = \frac{1}{2 \Lambda} \bar{\ell}^*\sigma^{\mu\nu} \left[g f \frac{ \mbox{\boldmath $\tau$} }{2} \mbox{\boldmath {\rm\bf W}}_{\mu\nu} + g^\prime f^\prime \frac{Y}{2} B_{\mu\nu} \right] \ell_{\rm L} + {\rm hermitian~conjugate},$$ where $\sigma^{\mu\nu}$ is the covariant bilinear tensor, denotes the Pauli matrices, $Y$ is the weak hypercharge, ${\bf W_{\mu\nu}}$ and $B_{\mu\nu}$ represent the Standard Model gauge field tensors and the couplings $g,g^\prime$ are the SU(2) and U(1) coupling constants of the Standard Model. The compositeness scale is set by the parameter $\Lambda$ which has units of energy. Finally, the strength of the $\ell\ell^*\rm V$ coupling is governed by the constants $f$ and $f^\prime$. These constants can be interpreted as weight factors associated with the different gauge groups. The values of $f$ and $f^\prime$ dictate the relative branching fractions of excited leptons to each gauge boson. The branching fraction of electromagnetically decaying excited charged leptons is significant for most values of $f$ and $f^\prime$ except in the case where $f=-f^\prime$ which entirely forbids this particular decay. As a result of the clean characteristic signatures expected, the photon decay constitutes one of the most sensitive channels for the search for excited leptons, even for values of $f$ and $f^\prime$ for which other decay modes dominate. To reduce the number of free parameters it is customary to assume either a relation between $f$ and $f^\prime$ or set one coupling to zero. For easy comparison with previously published results, limits calculated in this paper correspond to the coupling choice $f=f^\prime$. This assignment is a natural choice which forbids excited neutrinos from decaying electromagnetically. For this particular coupling choice, the electromagnetic branching fraction of charged excited leptons drops smoothly from 100% for masses below the  and  mass thresholds to about 30% for masses in excess of 200 GeV. Data and Simulated Event Samples ================================ The data analysed were collected by the OPAL detector [@bib:opal] at [centre-of-mass]{} energies ranging from 183 GeV to 209 GeV during the LEP runs in the years 1997 to 2000. The search for excited leptons is based on a total of 684.4 $\rm pb^{-1}$ of data for which all relevant detector components were fully operational. For the purpose of accurately interpreting the results in terms of limits on excited lepton masses and couplings, the data are divided into 16 [centre-of-mass]{} energy bins analysed separately. The energy range, luminosity weighted mean [centre-of-mass]{} energy and integrated luminosity of each bin are summarised in Table \[table:bin\]. The uncertainty on the measured beam energy is approximately 25 MeV [@bib:LEP_energy] and is correlated between [centre-of-mass]{} energy bins. In addition to the high energy data, approximately 10 $\rm pb^{-1}$ of calibration data collected in 1997-2000 at a [centre-of-mass]{} energy near the  mass were used to study the detector response. [|c|c|r|]{} bin range & $<\sqrt{s}>$ &\ & &\ 178.00 - 186.00 & 182.7 & 63.8\ 186.00 - 190.40 & 188.6 & 183.2\ 190.40 - 194.00 & 191.6 & 29.3\ 194.00 - 198.00 & 195.5 & 76.5\ 198.00 - 201.00 & 199.5 & 76.9\ 201.00 - 203.75 & 201.9 & 44.5\ 203.75 - 204.25 & 203.9 & 1.5\ 204.25 - 204.75 & 204.6 & 9.7\ 204.75 - 205.25 & 205.1 & 60.0\ 205.25 - 205.75 & 205.4 & 3.6\ 205.75 - 206.25 & 206.1 & 14.3\ 206.25 - 206.75 & 206.5 & 107.3\ 206.75 - 207.25 & 206.9 & 5.7\ 207.25 - 207.75 & 207.5 & 0.5\ 207.75 - 208.25 & 208.0 & 7.2\ $>$ 208.25 & 208.3 & 0.5\ & 684.4\ Distributions of kinematic variables and selection efficiencies for excited leptons were modelled using samples of simulated events obtained using the EXOTIC [@bib:exotic] Monte Carlo event generator. The matrix elements [@bib:ex_theory1; @bib:exotic_matrix] implemented in EXOTIC include all the spin correlations in the production and decay of excited leptons. The Standard Model processes at different [centre-of-mass]{} energies were simulated using a variety of Monte Carlo event generators. Bhabha events were simulated using the BHWIDE [@bib:bhwide] and TEEGG [@bib:teegg] generators, muon and tau pair events using both KORALZ [@bib:koralz] and KK2F [@bib:kk2f], $\rm e^+e^-\rightarrow\rm q\bar{\rm q}(\gamma)$ events using PYTHIA [@bib:pythia] and KK2F, four fermion processes using KORALW [@bib:koralw] and grc4f [@bib:grc4f], di-photon production using RADCOR [@bib:radcor], and two-photon events ($\rm e^+e^-$ $\rightarrow$ $e^+e^-\gamma\gamma$ $\rightarrow$ $e^+e^-f\bar{f}$) using VERMASEREN [@bib:vermaseren], PHOJET [@bib:phojet] and HERWIG [@bib:herwig]. Each of the simulated event samples was processed through the OPAL detector simulation program [@bib:gopal] and analysed in the same way as data. Event Selection and Kinematic Fits {#section:selection} ================================== Events are reconstructed from tracks and energy clusters defined by requirements similar to those described in [@bib:tracks]. The background from multihadronic events is substantially reduced by requiring at least one but no more than six tracks in an event. Furthermore, the ratio of the number of good tracks, as defined in [@bib:tracks], to the total number of tracks reconstructed in the detector must be greater than 0.2 in order to reduce background from beam-gas and beam-wall collisions. Cosmic ray events are suppressed using information from the time-of-flight counters and the central tracking chamber [@bib:cosmic]. Tracks and energy clusters in an event are grouped into jets using a cone algorithm [@bib:cone] with a cone half-angle of 0.25 radians and minimum jet energy of 2.5 GeV. The parameters defining a jet were chosen to maximise the signal efficiency over the broadest possible range of excited lepton masses and [centre-of-mass]{} energies. Events are required to contain between two and four jets. Jets are classified as leptons or photons using the criteria described below applied in the same order as given in the text. Photon candidates must have a minimum energy deposited in the electromagnetic calorimeter equivalent to 5% of the beam energy. A photon jet must either contain no tracks or be identified as a photon conversion using a neural network technique [@bib:idncon_nn8]. Jets in which the most energetic track has a neural network output greater than 0.9 and the energy deposited in the hadronic calorimeter is less than 10% of the beam energy are defined to be photon conversions. All photon candidates must lie within $|\cos\theta| < 0.9$ to avoid poorly modelled regions of the detector [^1]. The energy and direction of each photon candidate are determined from the energy and position of the energy cluster in the electromagnetic calorimeter. Muon candidates are jets containing exactly one track with associated hits in the muon detectors or hits in the hadronic calorimeter consistent with the particle being a muon [@bib:ww]. Muons, unlike electrons and photons, do not shower while traversing additional material present in the forward region of the detector. Muon candidates are thus allowed to lie within a larger angular acceptance of $|\cos\theta|<0.95$. The direction of each muon candidate is given by the polar and azimuthal angles of the track and the momentum is calculated from the track curvature and polar angle. A jet is identified as an electron if it contains exactly one track satisfying one of the following two requirements: the ratio of the electromagnetic energy to the track momentum ([*E/p*]{}) lies between 0.8 and 1.4 or the track has an output greater than 0.9 from a neural network developed to identify electrons [@bib:idncon_nn8]. Electron candidates are also required to lie within $|\cos\theta|<0.9$ to avoid poorly modelled regions of the detector. The energy of each electron candidate is taken to be the energy deposited in the electromagnetic calorimeter while the direction is given by the polar and azimuthal angles of the track. Finally, unidentified jets containing at least one track and lying within $|\cos\theta|<0.95$ are considered to be tau candidates. Jets in the region $0.90<|\cos\theta|<0.95$ which would satisfy the electron or photon requirements are discarded from the sample of tau candidates. Jets identified as tau candidates are mostly hadronically decaying taus. Tau leptons decaying leptonically are tagged as electrons or muons by the criteria described above. The polar and azimuthal angles of tau candidates are given by the axis of the jet, corrected for double-counting of tracks and energy clusters [@bib:MT]. The different selections used to identify the final states of interest are described in the following sections and the results are summarised in Table \[table:results\]. Selection of  Final States {#section:llgg} -------------------------- Events containing two lepton candidates of the same flavour and two identified photons are considered as candidate events for the pair production of excited leptons. In addition, events containing two leptons of different flavours and two photons are considered as excited tau candidates. In order to reduce the background from Standard Model processes, additional selection criteria are applied to the different types of candidate events. The quantity  is defined to be the sum of the energy of the particles considered for a given event final state, divided by the [centre-of-mass]{} energy. This quantity is required to exceed 0.8 for  and  candidates and 0.4 for  candidates. This criterion reduces the background from two-photon events. It also decreases the contamination from  events in the  sample. Figure \[fig:rvis\](a-c) shows the  distributions obtained using the entire data set for each type of candidate event. The observed discrepancy at small values of  in the sample corresponds to a region where the background is dominated by two-photon events and does not affect the analysis as the events of interest lie in a region of  that is well modelled. A similar mis-modelling is present in distributions of $\ell\ell\gamma$ and $\rm e\gamma$ candidate events. The remaining background in the  and  samples comes almost entirely from  and  events with additional photons. The background in the  sample consists mostly of  events with more than one radiated photon, and a small fraction of  events. Selection of  Final States {#section:llg} -------------------------- Events containing two lepton candidates of the same flavour and at least one identified photon are considered as candidate events for the single production of excited leptons. In addition, events with two leptons of different flavours and at least one photon are considered as excited tau candidates. If more than one photon is identified in the event, the most energetic photon is chosen and the other photon is ignored. Events selected as candidates for the pair production of excited leptons are also considered as single production candidates. To reduce the background from two-photon events the quantity  must be greater than 0.8 for  and candidates, and greater than 0.4 for  candidates. The distributions of each type of candidate event are shown in Figure \[fig:rvis\](d-f) for data from all the [centre-of-mass]{} energies combined. The dominant  background in the  final state is reduced by requiring that the angle between the most energetic electron and photon ($\theta_{e\gamma}$) be greater than $90^0$. The $\cos\theta_{\rm e\gamma}$ distribution obtained using data from all [centre-of-mass]{} energies combined is shown in Figure \[fig:other\](a). Background from both  and  events in the  sample is reduced by requiring the total energy deposited in the electromagnetic calorimeter to be between 20% and 80% of the [centre-of-mass]{} energy. Finally, the polar angle of the missing momentum vector for the particles considered in the  final state must lie within $|\cos\theta_{\rm miss}| < 0.9$. This requirement reduces the contamination from  events. Figures \[fig:other\](b,c) show both the total electromagnetic energy and $|\cos\theta_{\rm miss}|$ distributions of  events before applying each cut. After this selection, the remaining background in the  and   samples consists almost entirely of  and  events with an additional photon. The background in the  sample consists mostly of  events with one radiated photon and a small fraction of radiative ,  and  events. Selection of $\rm e\gamma$ Final State -------------------------------------- A separate selection for events with one electron and one photon was developed to increase the efficiency of the search for singly produced excited electrons where one electron travels in the forward region outside the detector acceptance. Candidate events are required to contain at least one photon and at least one electron candidate. Additional jets, if present, are ignored. Since the  and  final states are combined to calculate a limit on the single production of excited electrons, it is important to ensure that events are not double-counted. All events that are selected by the set of general requirements discussed at the beginning of Section \[section:selection\], but that fail the selection are considered as possible  candidates. To reduce the two-photon background, the quantity  must satisfy  $> 0.4$. The angle between the electron and photon ($\theta_{e\gamma}$) is also required to be greater than $90^0$. Further reduction of the dominant  background is achieved by requiring that the measured polar angle of the photon satisfies $|\cos\theta_{\gamma}| < 0.8$ and by rejecting events where the photon is identified as a conversion. Figures \[fig:other\](d-f) show the , $\cos\theta_{e\gamma}$ and $|\cos\theta_{\gamma}|$ distributions of  events obtained using the entire data set with cuts applied in the order described above. The irreducible background consists almost entirely of   events with one radiated photon. Kinematic Fits {#section:kinfit} -------------- The existence of excited leptons would reveal itself as an excess in the total number of observed events, appearing as a peak in the reconstructed $\ell\gamma$ invariant mass distributions. Kinematic fits are used to improve the reconstructed mass resolution of the selected events and also further reduce the background thereby increasing the sensitivity of the analysis to excited leptons. The kinematic variables used as input to the fit are the energy and direction ([*E*]{}, $\theta$, $\phi$) of each identified jet in an event. The energy of tau candidates is left as a free parameter and the direction is taken to be the jet axis. A kinematic fit also requires as input the error on each measured variable. Estimates of the uncertainties on the energy and direction for the different types of leptons and for photons are obtained from studies of di-lepton events in data recorded at [centre-of-mass]{} energies near and greater than the  mass. The error estimates are parameterised as functions of the jet energy and polar angle. The uncertainty on the jet energy is typically 2 GeV for electrons and photons, and about 5 GeV for muon candidates. The uncertainty on the jet polar angle is about 2 mrad for electrons and muons, 4 mrad for photons, and 7 mrad for tau candidates. Finally, the azimuthal angle of electron, muon, photon and tau candidates is typically known to 0.4 mrad, 0.4 mrad, 3.5 mrad and 7 mrad respectively. The kinematic fit enforces conservation of energy and momentum while taking into account the beam energy spread as measured by the LEP energy working group [@bib:LEP_energy]. This last constraint is necessary since the expected mass resolution for excited leptons is of the same order as the [centre-of-mass]{} energy spread, which is measured to be approximately 250 MeV. Slightly different kinematic fits are applied for single and pair production candidate events. In addition to energy and momentum conservation, the kinematic fits for pair produced excited lepton candidates also require the invariant masses of the two lepton-photon pairs in the event to be equal. There are two possible lepton-photon pairings in each event, and for each pairing an additional fit is performed assuming the presence of an undetected initial state radiation photon along the beam axis. Thus for each pair production candidate, four kinematic fits are performed. Similarly, two kinematic fits are applied to singly produced excited lepton candidate events. In the first case, only the two leptons and one photon are included in the fit. In the second case, the fit is performed assuming the presence of an initial state radiation photon along the beam axis. Finally, a single kinematic fit is performed for $\rm e\gamma$ events. The fit assumes the presence of an undetected electron along the beam axis. For a given final state, events are rejected if every kinematic fit attempted has a probability less than 0.001. When more than one successful kinematic fit is obtained for an event, the fit performed without the presence of an initial state radiation photon is chosen if the fit probability is greater than 0.001. For pair production candidates where two fits without initial state radiation are performed, the lepton-photon pairing corresponding to the fit with the highest probability in excess of 0.001 is chosen. Otherwise, results from the fit with an initial state radiation photon are retained where, for pair production candidates, the lepton-photon pairing corresponding to the fit with the highest probability is chosen. This requirement on the kinematic fit probabilities reduces the number of selected events by more than 70% for  final states and by about 10% for final states compatible with the single production of an excited lepton. Results from the chosen kinematic fit for each event are used to calculate $\ell\gamma$ invariant masses. Using the procedure outlined above, the correct lepton-photon pairing for pair produced excited leptons is chosen more than 98% of the time as determined using simulated signal events. For  events, two lepton-photon combinations are possible, both of which are included in the analysis. Mass resolutions of approximately 0.2-0.4 GeV for excited electrons and muons and 0.7-2.0 GeV for excited taus are obtained using results from the kinematic fits. The natural decay width of excited leptons for couplings not excluded by previous searches is constrained to be much smaller than these mass resolutions and is thus neglected. Figure \[fig:mlg\] shows the invariant mass distributions of selected  and  events. There are two entries per  event, corresponding to the two possible lepton-photon pairings. These distributions are obtained by combining data from all the [centre-of-mass]{}energies considered. In both single and pair production event samples, no mass peaks are observed in the data. Results ======= The numbers of events observed in the data and the corresponding numbers of background events expected from Standard Model processes are shown in Table \[table:results\]. Selected candidate events for the pair production of excited leptons are listed in Table \[table:pair\_candidates\]. Typical selection efficiencies for the pair production of excited leptons vary from about 35% to 55%. The efficiency for the single production of excited muons is 70% and approximately constant over the entire kinematically allowed range of masses. Near the kinematic limit for the single production of excited taus, the efficiency rapidly drops from 53% down to approximately 20% since the recoiling tau has low energy and thus often fails the initial set of selection criteria. For singly produced excited electrons, the efficiencies of the  and selections depend strongly on the mixture of $s$-channel and $t$-channel components. The sum of the  and  efficiencies is typically between 50% and 70%. The evaluation of the systematic uncertainties on the selection efficiencies and background estimates is discussed in the following section. No excess of data indicating the existence of excited leptons is found in either the single or pair production search. [|l|r|r|r|r|r|r|r|]{} & & & $\mu\mu\gamma$ & $\rm \tau\tau\gamma$ & $\rm ee\gamma\gamma$ & $\mu\mu\gamma\gamma$ & $\tau\tau\gamma\gamma$\ Data & 1172 & 1123 & 212 & 248 & 3 & 3 & 7\ Background & 1283 & 1229 & 239 & 260 & 4.0 & 2.0 & 8.0\ Background statistical errors & 11 & 14 & 2 & 2 & 0.6 & 0.2 & 0.6\ Background systematic errors & 161 & 159 & 28 & 40 & 1.2 & 0.7 & 6.3\ \ \ ISR modelling & 141 & 135 & 26 & 29 & 0.3 & 0.1 & 0.6\ Error estimate of fit variables & 68 & 57 & 10 & 24 & 1.1 & 0.6 & 6.2\ Jet classification & 36 & 27 & 3 & 12 & 0.2 & 0.1 & 0.5\ Energy and angular resolution & 9 & 16 & 2 & 2 & 0.0 & 0.3 & 0.6\ Modelling of selection variables & 6 & 54 & 2 & 8 & 0.3 & 0.1 & 0.3\ [|c|c||c|c||c|c|]{} & &\ ------------------------------------------------------------------------ & $m_{\rm e\gamma}$ & & $m_{\rm \mu\gamma}$ & & $m_{\rm \tau\gamma}$\ (GeV) & (GeV) & (GeV) & (GeV) & (GeV) & (GeV)\ 188.7 & 39.3 & 188.7 & 44.5 & 188.6 & 70.9\ 199.6 & 80.1 & 205.1 & 28.0 & 189.0 & 52.7\ 201.6 & 92.7 & 206.2 & 34.0 & 199.6 & 30.1\ & 199.7 & 76.2\ & 204.7 & 39.1\ & 204.8 & 38.6\ & 205.1 & 88.6\ Systematic Uncertainties ------------------------ The following sources of systematic uncertainties on the signal efficiencies and background estimates were investigated. These are described in order of importance. Uncertainties in the modelling of initial state photon radiation (ISR) in di-lepton events affect the background estimates. They are assessed by comparing background expectations from the KORALZ and KK2F event generators for the processes $\rm e^+e^-\rightarrow\mu^+\mu^-$ and $\rm e^+e^-\rightarrow\tau^+\tau^-$. The Monte Carlo program KK2F, used in this analysis to estimate the background contributions from $\mu^+\mu^-$ and $\tau^+\tau^-$ events, has the most complete description of initial state radiation including second-order subleading corrections and the exact matrix elements for two hard photons [@bib:ceex]. The relative variations in background expectations between the two Monte Carlo generators are assigned as systematic uncertainties representing the effect of missing higher orders. These are found to be 11% for final states compatible with the single production of excited muons and taus, and 7% for  and  events. The BHWIDE and TEEGG event generators, used to simulate the background from radiative  events, have a precision for radiative corrections similar to the KORALZ program. The background estimates for events expected from the production of excited electrons are thus assigned an uncertainty of 7% for the  final state and 11% for both  and events. This uncertainty is significantly larger than the error for inclusive electron pair production cited in [@bib:lepmc]. For the purpose of calculating limits on the product of the cross-section and the branching fraction, it is necessary to be able to calculate the efficiency and mass resolution of signal events at arbitrary excited lepton masses and [centre-of-mass]{} energies. For each final state, the selection efficiencies and mass resolutions are parameterised as a function of the excited lepton mass scaled by the [centre-of-mass]{} energy ($m_{\ell^*}/\sqrt{s}$). The systematic uncertainties associated with the interpolation of efficiencies and mass resolutions were estimated by calculating the root-mean-square spread between simulated signal event samples and the parameterisation functions. Uncertainties on the fit variable error estimates are evaluated by varying the errors on each variable independently. The errors are varied by an amount representing one standard deviation as calculated from the uncertainties on the energy and angular parameterisation. Background estimates for final states containing two leptons and two photons are particularly sensitive to changes in the errors due to the additional constraint in the kinematic fit requiring events to have equal reconstructed lepton-photon invariant masses. Also, the smaller sample of tau pair events used to parameterise the errors on the tau direction results in larger statistical uncertainties on the error parameterisation which in turn dictate the larger variations used to estimate the systematic error contributions. The jet classification into leptons or photons contributes to the overall systematic uncertainty through the modelling of the lepton and photon identification efficiencies. Using di-lepton and di-photon events recorded at [centre-of-mass]{} energies equal to and greater than the  mass, the systematic uncertainty associated with each set of lepton and photon requirements was evaluated by comparing the identification efficiencies obtained from data and simulated events. Relative errors of 1% for electron and muon, and 2% for the tau and photon classifications are assigned. Systematic uncertainties associated with each final state were determined by adding linearly contributions from identical jet classifications and adding in quadrature contributions from different types of leptons and photons. The resulting uncertainties on the signal efficiencies, shown in Table \[table:syst\_eff\], are fully correlated with the corresponding errors on the background estimates presented in Table \[table:results\]. The systematic uncertainty associated with the energy scale, energy resolution and angular resolution of the leptons and photons was evaluated by modifying each parameter independently in Monte Carlo simulated events. Comparisons between data and simulated distributions of di-lepton events recorded at different [centre-of-mass]{} energies were used to determine the size of these variations. The energy (momentum) of electron and photon (muon) candidates was shifted by 0.3%. The energy and angular resolutions of jets were smeared by the maximum values for which simulated events were compatible with the distribution of data within one standard deviation. Variations in the energy scale result in negligible changes in efficiencies and background. Contributions to the systematic uncertainty of each final state from individual changes in the energy and angular resolution are added in quadrature. The systematic uncertainty due to Monte Carlo modelling of the event selection variables was estimated by varying each selection cut independently and measuring the corresponding changes in the overall signal efficiencies and background estimates. The difference between the mean value of the data and background expectation for each selection variable determined the range of variation of each cut. Systematic uncertainties varying between 0.5% and 6.3% are assigned to the different background estimates. Contributions to the systematic error on the signal efficiencies are shown in Table \[table:syst\_eff\]. Lastly, the uncertainty on the integrated luminosity measurements (0.2%) is considerably smaller than the systematic effects already described and is therefore neglected. Summaries of the systematic effects on the background expectations and signal efficiencies are presented in Tables \[table:results\] and \[table:syst\_eff\], respectively. These systematic uncertainties are included in the calculation of limits as described in the following section. ---------------------------------- ---------------- --------------- ---------------- ------------------------ ---------------------- ---------------------- ------------------------ $\rm ee\gamma$ $\rm e\gamma$ $\mu\mu\gamma$ $\rm $\rm ee\gamma\gamma$ $\mu\mu\gamma\gamma$ $\tau\tau\gamma\gamma$ \tau\tau\gamma$ Resolution interpolation 18.6 12.5 20.7 7.1 23.5 18.1 12.9 Efficiency interpolation 8.6 3.0 2.3 5.3 4.5 3.4 4.4 Error estimate of fit variables 5.0 5.0 3.0 5.0 3.0 2.0 6.0 Jet classification 2.8 2.2 1.4 4.5 4.5 4.5 5.7 Energy and angular resolution 0.9 1.1 0.8 1.1 1.8 0.6 0.9 Modelling of selection variables 0.0 1.5 0.1 1.6 0.4 0.4 0.8 Total 21.3 14.1 21.1 11.3 24.6 19.1 16.0 ---------------------------------- ---------------- --------------- ---------------- ------------------------ ---------------------- ---------------------- ------------------------ : \[table:syst\_eff\] Relative systematic uncertainties on the signal efficiencies for each final state considered. Limit Calculations ------------------ Limits on the product of the cross-section and the electromagnetic branching fraction of excited leptons are obtained from both pair and single production searches. The numbers of data and expected background events at each [centre-of-mass]{} energy are binned as a function of the reconstructed invariant mass. For selected  candidates, both possible $\ell\gamma$ invariant masses are used. Due to the excellent mass resolution, the double-counting of  events does not affect the limits calculated. Each mass bin at a given [centre-of-mass]{} energy is treated as an independent counting experiment. For the purpose of calculating limits, the signal invariant mass is assumed to be well described by a Gaussian distribution centred at the test mass value and with a width equal to the expected mass resolution. The validity of this assumption is verified with Monte Carlo simulation of signal events at different masses and [centre-of-mass]{} energies. Efficiency corrections due to non-Gaussian tails in the invariant mass distributions are applied to the signal expectation. These correction factors, signal efficiencies and mass resolutions are all parameterised as a function of the excited lepton mass scaled by the [centre-of-mass]{} energy, $m_{\ell^*}/\sqrt{s}$. The efficiency correction factors are constant over the entire kinematically allowed range and vary from approximately 0.7 to 0.85 depending on the event final state. Efficiencies and mass resolutions are well-described by polynomial functions of various degrees. The efficiencies for the single production of excited leptons are calculated with the production and decay angular distributions corresponding to $f=f^\prime$. The assignment $f=f^\prime$ particularly affects the relative fraction of excited electron events in the  and  selections. For a given test mass, the Gaussian distributions describing the invariant mass of signal events at each [centre-of-mass]{} energy considered are normalised to the expected excited lepton cross-section at the highest [centre-of-mass]{}energy, thereby taking into account the energy dependence of the cross-section. A likelihood ratio method [@bib:likelihood_ratio] is used to compute the 95% confidence level upper limit on the number of signal events produced in the entire data set ($\rm N_{95}$). Systematic uncertainties on the signal efficiency and background expectation are incorporated by fluctuating, over many iterations, the background expectation and signal efficiency according to their respective systematic uncertainties. The final limits are determined from the average of all the $\rm N_{95}$ values obtained at each iteration. Systematic errors on the background estimates are treated as being fully correlated. The systematic uncertainties on the signal efficiencies due to the jet classification are also fully correlated with the corresponding errors on the background estimates and are treated as such in the limit calculations. Limits on the product of the cross-section and the branching fraction are scaled to  = [208.3]{} GeV assuming the cross-section evolution as a function of [centre-of-mass]{} energy expected for excited leptons. The upper limits on the single production of excited muons and tau leptons do not depend on the coupling assignment of $f$ and $f^\prime$. The excited electron selection efficiencies, however, depend on the relative magnitude of the [*s*]{}-channel and [*t*]{}-channel diagrams. For comparison with previously published results, the limits on excited electrons presented here assume $f=f^\prime$. Figures \[fig:limits\](a,b) show the 95% confidence level upper limits on the product of the cross-section at  = [208.3]{} GeV and the branching fraction obtained from the search for singly and pair produced excited leptons. The theoretical calculation [@bib:ex_theory1] of the product of the pair production cross-section at  = [208.3]{} GeV and the branching fraction squared is overlayed on Figure \[fig:limits\](b). As part of this calculation, the electromagnetic branching fraction is calculated assuming $f=f^\prime$. The 95% confidence level lower mass limits on excited leptons correspond to the mass at which the cross-section times branching fraction limit curves cross the theoretical expectation. Lower mass limits of $m_{e^*} >$ [103.2]{} GeV, $m_{\mu^*} >$ [103.2]{} GeV and $m_{\tau^*} >$ [103.2]{} GeV are obtained. Although systematic errors are incorporated into the limit calculations, an additional uncertainty on the mass limits arises from the finite width of the [centre-of-mass]{} energy bins considered. The 0.5 GeV [centre-of-mass]{} energy bin width near the kinematic limit corresponds to an uncertainty of 0.1 GeV on the mass limits. Limits on the product of the cross-section and the electromagnetic branching fraction of singly produced excited leptons are used to constrain parameters of the model introduced in Section \[section:intro\]. Since the cross-section for the single production of excited leptons is proportional to $(f/\Lambda)^2$, limits on the ratio of the coupling to the compositeness scale as a function of excited lepton mass are calculated using $$\left[\frac{(f/\Lambda)}{(1\ \rm TeV^{-1})}\right]_{\rm 95\% CL} = \sqrt{\frac{\rm N_{95}}{\rm N_{\rm exp}}}\:\:\: ,$$ where $\rm N_{\rm exp}$ is the number of expected signal events assuming $f/\Lambda = 1~\rm TeV^{-1}$. Figure \[fig:limits\](c) shows these limits for each type of excited lepton. The $f/\Lambda$ limit for excited electrons is approximately an order of magnitude better than for muons and taus due to the enhancement of the cross-section coming from the [*t*]{}-channel contribution. Conclusion ========== A search for electromagnetically decaying charged excited leptons was performed using 684.4 $\rm pb^{-1}$ of data collected by the OPAL detector at  = 183-209 GeV. No evidence was found for the existence of excited leptons. Upper limits on the product of the cross-section and the branching fraction were calculated. From pair production searches, 95% confidence level lower limits on the mass of excited leptons are determined to be $m_{\ell^*}$ $>$ [103.2]{} GeV for $\ell = \rm e,\mu, \tau$. From the results of the search for singly produced excited leptons, limits were calculated on the ratio of the coupling constant to the compositeness scale ($f/\Lambda$) as a function of excited lepton mass. The results are currently the most stringent constraints on the existence of excited leptons and therefore represent a significant improvement on limits previously published [@bib:ex_opal_results; @bib:lep; @bib:hera]. [**Acknowledgements**]{}\ We particularly wish to thank the SL Division for the efficient operation of the LEP accelerator at all energies and for their close cooperation with our experimental group. In addition to the support staff at our own institutions we are pleased to acknowledge the\ Department of Energy, USA,\ National Science Foundation, USA,\ Particle Physics and Astronomy Research Council, UK,\ Natural Sciences and Engineering Research Council, Canada,\ Israel Science Foundation, administered by the Israel Academy of Science and Humanities,\ Benoziyo Center for High Energy Physics,\ Japanese Ministry of Education, Culture, Sports, Science and Technology (MEXT) and a grant under the MEXT International Science Research Program,\ Japanese Society for the Promotion of Science (JSPS),\ German Israeli Bi-national Science Foundation (GIF),\ Bundesministerium für Bildung und Forschung, Germany,\ National Research Council of Canada,\ Hungarian Foundation for Scientific Research, OTKA T-029328, and T-038240,\ Fund for Scientific Research, Flanders, F.W.O.-Vlaanderen, Belgium.\ [99]{} F.E. Low, [Phys. Rev. Lett. **14** (1965) 238]{}. OPAL Collaboration, G. Abbiendi [[*et al.*]{}]{}, [Eur. Phys. J. [**C14**]{} (2000) 73]{};\ OPAL Collaboration, K. Ackerstaff [[*et al.*]{}]{}, [Eur. Phys. J. [**C1**]{} (1998) 45]{};\ OPAL Collaboration, K. Ackerstaff [[*et al.*]{}]{}, [Phys. Lett. [**B391**]{} (1997) 197]{};\ OPAL Collaboration, G. Alexander [[*et al.*]{}]{}, [Phys. Lett. [**B386**]{} (1996) 463]{};\ OPAL Collaboration, M.Z. Akrawy [[*et al.*]{}]{}, [Phys. Lett. [**B244**]{} (1990) 135]{}. L3 Collaboration, M. Acciarri [[*et al.*]{}]{}, [Phys. Lett. [**B502**]{} (2001) 37]{};\ DELPHI Collaboration, P. Abreu [[*et al.*]{}]{}, [Eur. Phys. J. [**C8**]{} (1999) 41]{};\ ALEPH Collaboration, R. Barate [[*et al.*]{}]{}, [Eur. Phys. J. [**C4**]{} (1998) 571]{}. H1 Collaboration, C. Adloff [[*et al.*]{}]{}, [Phys. Lett. [**B525**]{} (2002) 9]{};\ H1 Collaboration, C. Adloff [[*et al.*]{}]{}, [Eur. Phys. J. [**C17**]{} (2000) 567]{};\ ZEUS Collaboration, S. Chekanov, [[*et al.*]{}]{}, Submitted to Phys. Lett. [**B**]{} \[hep-ex/0109018\];\ ZEUS Collaboration, J. Breitweg [[*et al.*]{}]{}, [Z. Phys. [**C76**]{} (1997) 631]{};\ ZEUS Collaboration, M. Derrick [[*et al.*]{}]{}, [Z. Phys. [**C65**]{} (1995) 627]{}. F. Boudjema, A. Djouadi and J.L. Kneur, [Z. Phys. [**C57**]{} (1993) 425]{}. K. Hagiwara, D. Zeppenfeld and S. Komamiya, [Z. Phys. [**C29**]{} (1985) 115]{}. OPAL Collaboration, S. Anderson [[*et al.*]{}]{}, [Nucl. Instr. Meth. [**A403**]{} (1998) 326]{};\ B.E. Anderson [[*et al.*]{}]{}, IEEE Trans. Nucl. Sci. [**41**]{} (1994) 845;\ OPAL Collaboration, K. Ahmet [[*et al.*]{}]{}, [Nucl. Instr. Meth. [**A305**]{} (1991) 275]{}. LEP Energy Working Group, A. Blondel [[*et al.*]{}]{}, [Eur. Phys. J. [**C11**]{} (1999) 573]{}. R. Tafirout and G. Azuelos, [Comp. Phys. Comm. [**126**]{} (2000) 244]{}. A. Djouadi, [Z. Phys. [**C63**]{} (1994) 317]{}. S. Jadach, W. P[ł]{}aczek and B.F.L. Ward, [Phys. Lett. [**B390**]{} (1997) 298]{}. D. Karlen, [Nucl. Phys. [**B289**]{} (1987) 23]{}. S. Jadach, B.F.L. Ward and Z. Was, [Comp. Phys. Comm. [**79**]{} (1994) 503]{}. S. Jadach, B.F.L. Ward and Z. Was, [Comp. Phys. Comm. [**130**]{} (2000) 260]{};\ S. Jadach, B.F. Ward and Z. Was, [Phys. Lett. [**B449**]{} (1999) 97]{}. T. Sjöstrand, [Comp. Phys. Comm. [**82**]{} (1994) 74]{}. S. Jadach, W. P[ł]{}aczek, M. Skrzypek, B.F.L. Ward and Z. Was, [Comp. Phys. Comm. [**119**]{} (1999) 272]{}. J. Fujimoto [[*et al.*]{}]{}, [Comp. Phys. Comm. [**100**]{} (1996) 128]{}. F.A. Berends, R. Kleiss, [Nucl. Phys. [**B186**]{} (1981) 22]{}. R. Bhattacharya, J. Smith and G. Grammer, [Phys. Rev. [**D15**]{} (1977) 3267]{};\ J. Smith, J.A.M. Vermaseren and G. Grammer, [Phys. Rev. [**D15**]{} (1977) 3280]{}. R. Engel and J. Ranft, [Phys. Rev. [**D54**]{} (1996) 4244]{}. G. Marchesini [[*et al.*]{}]{}, [Comp. Phys. Comm. [**67**]{} (1992) 465]{}. J. Allison [[*et al.*]{}]{}, [Nucl. Instr. Meth. [**A317**]{} (1992) 47]{}. OPAL Collaboration, G. Alexander [[*et al.*]{}]{}., [Z. Phys. [**C72**]{} (1996) 191]{}. OPAL Collaboration, G. Alexander [[*et al.*]{}]{}, [Z. Phys. [**C52**]{} (1991) 175]{}. OPAL Collaboration, R. Akers [[*et al.*]{}]{}, [Z. Phys. [**C63**]{} (1994) 197]{}. OPAL Collaboration, G. Abbiendi [[*et al.*]{}]{}, [Eur. Phys. J. [**C16**]{} (2000) 41]{};\ OPAL Collaboration, G. Abbiendi [[*et al.*]{}]{}, [Eur. Phys. J. [**C8**]{} (1999) 217]{};\ OPAL Collaboration, G. Alexander [[*et al.*]{}]{}, [Z. Phys. [**C70**]{} (1996) 357]{}. OPAL Collaboration, K. Ackerstaff [[*et al.*]{}]{}, [Phys. Lett. [**B389**]{} (1996) 416]{}. OPAL Collaboration, K. Ackerstaff [[*et al.*]{}]{}, [Eur. Phys. J. [**C2**]{} (1998) 213]{}. S. Jadach, B. F. L. Ward and Z. Was, [Phys. Rev. [**D63**]{} (2001) 113009]{}. S. Jadach, W. P[ł]{}aczek and B.F.L. Ward in “Two-Fermion Production in Electron-Positron Collisions”, edited by M. Kobel and Z. Was, CERN-2000-009-D \[hep-ph/0007180\]. T. Junk, [Nucl. Instr. Meth. [**A434**]{} (1999) 435]{}. [^1]: The OPAL coordinate system is defined to be right-handed, with the z-axis pointing along the electron beam direction and the x-axis pointing toward the centre of the LEP ring. Thus the polar angle $\theta$ used in this paper refers to the angle with respect to the electron beam direction and the azimuthal angle $\phi$, the angle measured with respect to the x-axis.

ArXiv

[****]{}\ \ [*National Centre for Physics, Quaid-i-Azam University Campus, Islamabad, Pakistan.*]{}\ [M. Jamil Aslam]{}\ [*Physics Department, Quaid-i-Azam University, Islamabad, Pakistan.*]{}\ [**Abstract**]{}\ It is shown that for the effective Lagrangian with factorization ansatz considered here, the two body hadronic decay $\mathcal{B}_{b}(\frac{1}{2}^+) \to \mathcal{B}(\frac{1}{2}^{+},\; \frac{3}{2}^{+}) + V$, for $\mathcal{B}_{b}(\frac{1}{2}^{+})$ belonging to the representation $\bar{3}$, only allowed decay channel is $\mathcal{B}_{b}(\frac{1}{2}^+) \to \mathcal{B}(\frac{1}{2}^{+})+ V$, where $\mathcal{B}(\frac{1}{2}^{+})$ belongs to the representation $8$ of $SU(3)$. However, for $\mathcal{B}_{b}(\frac{1}{2}^{+})$ belonging to the sextet representation $6$, the allowed decay channels are $\mathcal{B}_{b}(\frac{1}{2}^+) \to \mathcal{B}(\frac{1}{2}^{+},\; \frac{3}{2}^{+}) + V$, where $\mathcal{B}(\frac{1}{2}^{+})$ and $\mathcal{B}(\frac{3}{2}^{+})$ belongs to the octet representation $8^{\prime}$ and the decuplet $10$ of $SU(3)$, respectively. The decay channel $\mathcal{B}_{b}(\frac{1}{2}^+) \to \mathcal{B}(\frac{1}{2}^{+}) + V$ is analyzed in detail. The decay rate ($\Gamma$) and the asymmetry parameters $\alpha\;, \alpha^{\prime}\;, \beta\;, \gamma$ and $\gamma^{\prime}$ are expressed in terms of four amplitudes. In particular for the decay $\Lambda_b \to \Lambda + J/\psi$ it is shown that within the factorization framework, using heavy quark spin symmetry, the decay rate and the asymmetry parameters can be expressed in terms of two form factors $F_1$ and $F_{2}/F_{1}$, which are to be evaluated in some model. By using the values of these form factors calculated in a quark model, the branching ratio and the asymmetry parameters $\alpha$ and $\alpha^{\prime}$ are calculated numerically. For other heavy quarks belonging to the triplet and sextet representation, the results can be easily obtained by using $SU(3)$ symmetry and phase space factor. Finally, the decay $\Omega_{b}^{-} \to \Omega^{-} + J/\psi$ is analyzed within the factorization framework. It is shown that the asymmetry parameter $\alpha$ in this particular decay is zero. The branching ratio obtained in the first approximation is compared with the experimental value. INTRODUCTION ============ Heavy flavor physics is of topical interest. New data for decays of $b-$hadrons will be coming out from the LHCb. In 2013 the LHCb Collaboration has performed an angular analysis of the decay $\Lambda_b \to \Lambda + J/\Psi$ where $\Lambda_{b}$’s are produced in proton-proton $(pp)$ collisions at the centre of mass energy $\sqrt{s} = 7$ TeV at the LHC [@LHCb-Collaboration]. By fitting several asymmetry parameters in the cascade decay distribution of $\Lambda_{b}\to \Lambda (\to p \pi^{-}) + J/\Psi(\to \ell^{+}\ell^{-})$, the collaboration has reported the relative magnitude of helicity amplitudes in $\Lambda_{b}\to \Lambda+J/\Psi$ decay and also the transverse polarization of $\Lambda_b$ relative to the production plane. Theoretically, the nonleptonic decay $\Lambda_b \to \Lambda +J/\Psi$ is quite attractive because only factorizable tree diagram contributes to the decay and there is no contribution due to $W-$exchange diagrams [@2]. In the $b-$baryon sector, the decay $\Lambda_b \to \Lambda +J/\Psi$ has been studied theoretically in quark model by using the factorization hypothesis [@3; @4; @5; @6; @7; @8; @9; @10; @11] and the results of some of these calculations have been compared to the new experimental results by the LHCb Collaboration. The results of the branching fraction of $\Lambda_b \to \Lambda +J/\Psi$ decay given in the PDG $\mathcal{B}r(\Lambda_b \to \Lambda +J/\Psi)\times \mathcal{B}r(b\to\Lambda_{b}^{0})= (5.8\pm0.8) \times 10^{-5}$ [@17] is deduced from the measurements by the CDF [@13] and the D$0$ collaborations [@14]. The result of branching fraction from the LHCb is still missing for this decay. In the present study, we give a general formalism for $\mathcal{B}_{b}(\frac{1}{2}^+) \to \mathcal{B}(\frac{1}{2}^{+}) +V$, especially with $V=J/\psi$. Using this formalism, we analyze $\Lambda_b \to \Lambda +J/\Psi$ decay in detail. Heavy baryons with $J^{P}=\frac{1}{2}^{+}$, belong to either representation $\bar{3}$ or the sextet $6$, whereas $J^{P}=\frac{3}{2}^{+}$ belongs only to the the sextet representation of the $SU(3)$[@15]: $$\begin{aligned} \bar{3}: \ \ \ \ \ \ \ \ \ \ A_{ij}&=&\frac{1}{\sqrt{2}}(q_i q_j - q_j q_i)Q\chi_{M A}\;, \notag \\ 6: \ \ \ \ \ \ \ \ \ \ S_{ij}&=&\frac{1}{\sqrt{2}}(q_i q_j + q_j q_i)Q\chi_{M S}\;, \notag \\ 6: \ \ \ \ \ \ \ \ \ \ S^{*}_{ij}&=&\frac{1}{\sqrt{2}}(q_i q_j + q_j q_i)Q\chi_{S} \;, \label{1}\end{aligned}$$ where $q_{i}\; , q_j$ are $u\; , d\; , s$; $Q = b$ or $c$ and $\chi$’s are the spin wave functions [@15]. In Eq. (\[1\]) $A_{ij}\;, S_{ij}$ and $S^{*}_{ij}$ correspond to $J^{P} = \frac{1}{2}^{+}$ and $J^{P}=\frac{3}{2}^{+}$, respectively. The triplet of heavy baryons are $$(A_{12}\;, A_{13}\;, A_{23}) : (\Lambda_{b\;,\; c}^{0\;,\; +}\;, \Xi_{b\;,\; c}^{0\;,\; +}\;, \Xi_{b\;,\;c}^{-\;,\; 0} )\;, \label{2}$$ whereas the sextet are $$\begin{aligned} (S_{11}\;, S_{12}\;, S_{22}) &:& (\sqrt{2}\Sigma_{b\;,\;c}^{+\;, ++}\;,\;\Sigma_{b\;,\; c}^{0\;, \;+}\;, \sqrt{2}\Sigma_{b\;,\;c}^{-\;,\; 0} )\;, \notag \\ (S_{13}\;, S_{23}) &:& (\Xi^{\prime 0\;,\; +}_{b\;,\;c}\;, \Xi^{\prime -\;,\; 0}_{b\;,\;c})\;, \notag \\ S_{33}&:& \sqrt{2}\Omega_{b\;,\; c}^{-\;,\; 0}\label{3}.\end{aligned}$$ In the Standard Model (SM) two body hadronic decays of heavy flavor mesons and baryons are analyzed in terms of the effective Lagrangian or Hamiltonian. Here, we take the Hamiltonian $$H_{eff} = V_{cb}V^{*}_{cs}[a_1(\bar{s}c)_{V-A}(\bar{c}b)_{V-A}+a_2(\bar{c}c)_{V-A}(\bar{s}b)_{V-A}]\;, \label{4}$$ where $a_1 = C_{1}+\zeta C_{2}$ and $a_2 = C_{2}+\zeta C_{1}$ , with $\zeta$ being the parameter for the possible number of colors. In terms of the diagrams, $a_{1}$ and $a_{2}$ correspond to the contribution from tree and color suppressed tree diagrams, respectively. In the factorization ansatz, for the tree diagram and color suppressed tree diagram, the relevant matrix elements are $\langle \mathcal{B}_c|(\bar{c}b)_{V-A}|\mathcal{B}_b\rangle$ and $\langle \mathcal{B}_s|(\bar{s}b)_{V-A}|\mathcal{B}_b\rangle$, respectively. First, one can notice that that $\bar{c}b$ is $SU(3)$ singlet, whereas $\bar{s}b$ is $SU(3)$ triplet. Now $$\begin{aligned} 3 \times \bar{3} &=& 8 +1\;, \notag \\ 3 \times 6 &=& 10 +8^{\prime}\;. \label{5}\end{aligned}$$ Hence the possible decay modes for $\mathcal{B}_b(\frac{1}{2}^{+})$, for the first term in Eq. (\[4\]) are $$\begin{aligned} \bar{3}: \; \;\;\ \ (\Lambda_{b}^{0}\;, \Xi_{b}^{0}\;, \Xi_{b}^{-} ) &\to & (\Lambda_{c}^{+}\; , \Xi_{c}^{+}\; , \Xi_{c}^{0})(D_s^{-})^{\ast}\; , \notag \\ 6: \; \;\;\ (\Sigma_{b}^{+}\; ,\; \Sigma_{b}^{0}\; ,\; \Sigma_{b}^{-}) &\to & (\Sigma_{c}^{++}\; ,\; \Sigma^{+}_{c}\; ,\; \Sigma_{c}^{0})^{*}(D_{s}^{-})^{*} \; ,\notag \\ (\Xi_{b}^{\prime 0}\; , \;\Xi_{b}^{\prime -}) &\to & (\Xi^{+}_{c}\; ' \;\Xi^{0}_{c})^{*}(D_{s}^{-})^{*} \; ,\notag \\ \Omega_{b}^{-} &\to & \Omega_{c}^{0}(\Omega^{*0}_{c}) (D_{s}^{-})^{*} \; .\label{6}\end{aligned}$$ Some of these decays given in Eq. (\[6\]) have been studied in ref. [@19]. The main focus of the present study is the heavy to light decays of $b$-baryons. For the color suppressed tree diagram, as noted in Eq. (\[5\]), for $\mathcal{B}_{b}(\frac{1}{2}^{+})$ belonging to the representation $\bar{3}$ the possible decay mode is $$\mathcal{B}_{b}(\frac{1}{2}^{+}) \to \mathcal{B}({\frac{1}{2}}^{+}) J/\Psi \; ,\label{7}$$ with $B({\frac{1}{2}}^{+})$ belongs to the octet representation $8$ of $SU(3)$. However, for $\mathcal{B}_{b}(\frac{1}{2}^{+})$ belonging to the sextet representation, we have two possible decay modes: $$\begin{aligned} \mathcal{B}_{b}(\frac{1}{2}^{+}) &\to & \mathcal{B}({\frac{1}{2}}^{+}) J/\Psi \; ,\notag \\ \mathcal{B}_{b}(\frac{1}{2}^{+}) &\to & \mathcal{B}({\frac{3}{2}}^{+}) J/\Psi \; .\label{8}\end{aligned}$$ For this case, $\mathcal{B}(\frac{1}{2}^{+})$ and $\mathcal{B}(\frac{3}{2}^{+})$ belong to the octet representation $8^{\prime}$ and decuplet representation $10$ of $SU(3)$, respectively. For the decay $\mathcal{B}_{b}(\frac{1}{2}^{+}) \to \mathcal{B}({\frac{1}{2}}^{+}) J/\Psi $, the decay channels are $$\begin{aligned} \bar{3}: \;\;\; (\Lambda_{b},\; \Xi_{b}^{0},\; \Xi_{b}^{-}) &\to & (\Lambda,\; \Xi^{0},\; \Xi^{-}) J/\Psi \; ,\notag \\ 6:\;\;\;(\Sigma_{b}^{0},\; \Sigma_{b}^{-},\; \Sigma_{b}^{+}) &\to & (\Sigma^{0},\; \Sigma^{-},\; \Sigma^{+}) J/\Psi \; , \notag \\ (\Xi_{b}^{0\prime},\; \Xi_{b}^{-\prime}) &\to & (\Xi^{0},\; \Xi^{-}) J/\Psi \; ,\label{9}\end{aligned}$$ where $\Lambda\;, \Xi^{0}\;, \Xi^{-}$ are members of the octet representation $8$ and $\Sigma^{0},\; \Sigma^{-},\;\Sigma^{+},\;\Xi^{0},\;\Xi^{-}$ are members of the octet representation $8^{\prime}$. This study focus on the analysis of $\mathcal{B}_{b}(\frac{1}{2}^{+}) \to \mathcal{B}({\frac{1}{2}}^{+}) V$ decays. For the decay $\mathcal{B}_{b}(\frac{1}{2}^{+}) \to \mathcal{B}({\frac{3}{2}}^{+}) J/\Psi$, where $\mathcal{B}_{b}(\frac{1}{2}^{+})$ belong the representation $6$, the decay channels are $$\begin{aligned} (\Sigma_{b}^{0},\; \Sigma_{b}^{-},\; \Sigma_{b}^{+}) &\to & (\Sigma^{* 0},\; \Sigma^{* -},\; \Sigma^{* +}) J/\Psi \; , \notag \\ (\Xi_{b}^{0\prime},\; \Xi_{b}^{-\prime}) &\to & (\Xi^{* 0},\; \Xi^{* -}) J/\Psi \; ,\label{9a}\\ \Omega^{-}_{b} &\to & \Omega^{-} J/\Psi \; .\notag \end{aligned}$$ where the last decay is most interesting in this category. Hadronic Weak Decay of Baryon $\mathcal{B}_{b}(\frac{1}{2}^{+}) \to \mathcal{B}({\frac{1}{2}}^{+}) V$: A General Formalism ========================================================================================================================== For the decay $$\mathcal{B}_{b}(\frac{1}{2}^{+})(p) \to \mathcal{B}({\frac{1}{2}}^{+})(p^\prime)+ V(k,\epsilon)\label{11}$$ where $p = p^{\prime}+k$ and $k \cdot \epsilon = 0$, the Lorentz structure of the $T-$ matrix is given by $$T=\frac{1}{(2\pi)^{9/2}}\sqrt{\frac{m m^{\prime}}{2 p_0 p^{\prime}_0 k_0}}\bar{u}(p^{\prime})[\gamma \cdot \epsilon (A(s)+B(s)\gamma_{5})+ i \epsilon^{\mu}\sigma_{\mu \nu}k^{\nu}(C(s)+D(s)\gamma_{5})]u(p)\; .\label{12}$$ In Eq. (\[12\]) the amplitudes $A,\;B,\;C$ and $D$ are the function of the square of momentum transfer, i.e., $s = (p-p^{\prime})^2$. In the rest frame of baryon $\mathcal{B}_{b}$ $$\begin{aligned} m &=& p^{\prime}_0+k_0 \; ,\notag \\ \vec{p}^{\; \prime} &=&- \vec{k} = -|\vec{k}|\vec{n}\; . \label{13}\end{aligned}$$ In this particular frame, one can write $$T = \chi^{\dagger}_{f} M \chi_{i} \label{14}$$ where $$M = \frac{1}{(2\pi)^{9/2}}\frac{1}{\sqrt{2k_0}}[i f_1\vec{\sigma}\cdot (\vec{n}\times \vec{\epsilon})+g_1\vec{\sigma}\cdot \vec{\epsilon}+f_2 \vec{n}\cdot \vec{\epsilon}+g_2 (\vec{n}\cdot \vec{\epsilon})(\vec{\sigma}\cdot \vec{n})].\label{15}$$ with $\vec{\sigma}$ are the Pauli matrices. The amplitudes $f_{1,\; 2}$, $g_{1,\; 2}$ and $h$ can be written in terms of $A\;, B\;, C\;, D$: $$\begin{aligned} f_1 & = & \frac{|\vec{k}|}{\sqrt{2p^{\prime}_0(p^{\prime}_0 + m^{\prime})}}[A(s)-C(s)(m + m^{\prime})] \; ,\label{16a} \\ g_1 & = & -\frac{1}{\sqrt{2p^{\prime}_0(p^{\prime}_0 + m^{\prime})}}[B(s)(p^{\prime}_0 + m^{\prime})+D(s)(k_{0}(m+ m^{\prime})-m^2_{V})]\; ,\label{16b} \\ f_2 & = & \frac{1}{\sqrt{2p^{\prime}_0(p^{\prime}_0 + m^{\prime})}}\frac{|\vec{k}|}{k_0}[A(s)(m+m^{\prime})-C(s)m_{V}^2]\; ,\label{16c} \\ g_2 & = & \frac{1}{\sqrt{2p^{\prime}_0(p^{\prime}_0 + m^{\prime})}}\frac{|\vec{k}|^2}{k_0}[-B(s)+D(s)(m + m^{\prime})]\; ,\label{16d} \\ h &=& g_1 + g_2 = \frac{-1}{\sqrt{2p^{\prime}_0(p^{\prime}_0 + m^{\prime})}}\frac{1}{k_0}[B(s)((m+m^{\prime})k_0 - m_{V}^2)+D(s)m_{V}^{2}(p^{\prime}_0 + m^{\prime})]\; .\label{18}\end{aligned}$$ Under space reflection $\vec{\sigma} \to \vec{\sigma}$, $\vec{n} \to -\vec{n}$ and $\vec{\epsilon} \to -\vec{\epsilon}$, thus $f_1$ and $f_2$ are the parity conserving i.e., $p-$wave amplitudes whereas $g_1$ and $g_2$ are the parity violating $s-$wave amplitudes. We also note that for the transverse polarization of $V$ meson, only $f_1$ and $g_1$ are relevant, whereas, for the longitudinal polarization the relevant amplitudes are $f_2$ and $h$. The decay width of the above mode is given by $$d\Gamma = (2\pi)^{7}\delta^4(p-p^{\prime}-k)\big[\frac{1}{2}Tr(MM^{\dagger}\big]d^{3}p^{\prime}d^{3}k\label{19}$$ which gives $$\Gamma = \frac{|\vec{k}|p_{0}^{\prime}}{2\pi m}\big[2(|f_1)|^2+|g_1|^2)+\frac{k_0^{2}}{m_{V}^2}(f_2)|^2+|h|^2)\big]\label{20}$$ The first term on the left hand side of Eq. (\[20\]) corresponds to the transverse polarization and the second term to the longitudinal one. Let $\vec{S}$ and $\vec{s}$ be the polarizations (spins) of $\mathcal{B}_{b}$ and $\mathcal{B}$, respectively. The decay probability in terms of these polarization vectors is given by $$dW = (2\pi)^{7} \delta^{4}(p-p^{\prime}-k)\frac{1}{2}Tr\big[(1+\vec{\sigma}\cdot \vec{s})M(1+\vec{\sigma}\cdot \vec{S})M^{\dagger}\big]d^{3}p^{\prime}d^{3}k\; . \label{21}$$ Hence, the transition rate is: $$\frac{dW}{\Gamma} = \frac{d\Omega_{S}d\Omega_{s}}{(4\pi)^2}[1+\alpha \vec{S}\cdot \vec{n} + \alpha^{\prime}\vec{s}\cdot \vec{n} + \beta \vec{s}\cdot(\vec{S} \times \vec{n}) + ((\vec{s}\cdot \vec{n}) (\vec{S}\cdot \vec{n}))(-1+\gamma^{\prime}) + \gamma \vec{s} \cdot (\vec{n}\times (\vec{S} \times \vec{n}))]\; ,\label{22}$$ where $$\begin{aligned} \alpha & = & 2 Re[-2f_{1}^{*}g_1+(\frac{k_0}{m_V})^{2}f_2^{*}h]p_{0}^{\prime} \frac{|\vec{k|}}{2\pi m \Gamma}\; , \label{23} \\ \alpha^{\prime} & = & 2 Re[2f_{1}^{*}g_1+(\frac{k_0}{m_V})^{2}f_2^{*}h]p_{0}^{\prime} \frac{|\vec{k|}}{2\pi m \Gamma}\; , \label{24} \\ \beta & = & 2 Im[f_{2}^{*}h]p_{0}^{\prime} \frac{|\vec{k|}}{2\pi m \Gamma}\; , \label{25} \\ \gamma & = & (\frac{k_0}{m_V})^{2}[|f_2|^2 - |h|^2]p_{0}^{\prime} \frac{|\vec{k|}}{2\pi m \Gamma}\; , \label{26} \\ \gamma^{\prime} & = & (\frac{k_0}{m_V})^{2}[|f_2|^2 + |h|^2]p_{0}^{\prime} \frac{|\vec{k|}}{2\pi m \Gamma}\; . \label{27} \end{aligned}$$ Following comments are in order. For the transverse polarization, the asymmetry parameters are $$\alpha = -\frac{4Re[f^{*}_{1}g_{1}]p^{\prime}_0 k_{0}}{2\pi m \Gamma} = -\alpha^{\prime}\; , \label{28}$$ whereas in case of the longitudinal polarization $$\alpha = \big(\frac{k_0}{m_V}\big)^{2}\frac{2Re[f^{*}_{2}h]p^{\prime}_0 k_{0}}{2\pi m \Gamma} = \alpha^{\prime}. \label{29}$$ It is clear from Eqs. (\[25\], \[26\]) and Eq. (\[27\]), that $\beta$, $\gamma$ and $\gamma^{\prime}$ are non-zero only for the longitudinal polarization. For the longitudinal polarization, we get exactly the same result as that in the non-leptonic decay of $\mathcal{B}$ baryon, when $V$ is replaced by pseudo-scalar meson $P$ [@16]. Factorization: Baryon Form Factors ================================== In the factorization framework, the effective Hamiltonian for the decay $\mathcal{B}_{b}(\frac{1}{2}^{+})\to \mathcal{B}(\frac{1}{2}^{+})+J/\psi$ is given by $$H_{eff} = \frac{G_{F}}{\sqrt{2}} V_{cb}V^{*}_{cs}a_2 \langle 0 |\bar{c}\gamma^{\mu} (1-\gamma_5) c | J/\Psi\rangle \langle \mathcal{B} |\bar{s}\gamma_{\mu}(1-\gamma_5)b | \mathcal{B}_b \rangle \; . \label{30}$$ The relevant matrix elements are $$\begin{aligned} \langle 0 |\bar{c}\gamma^{\mu} (1-\gamma_5) c | J/\Psi\rangle & = &(\frac{1}{2\pi})^{3/2}\frac{1}{\sqrt{2k_0}}F_{J/\Psi}m_{J/\Psi} \epsilon^{\mu}\; ,\label{30a} \\ \langle \mathcal{B} |\bar{s}\gamma_{\mu}(1-\gamma_5)b | \mathcal{B}_b \rangle & = & (\frac{1}{2\pi})^{3}\sqrt{\frac{m m^{\prime}}{p_0 p^{\prime}_0}} \bar{u}(p^{\prime})[(g_V(k^2)-g_A(k^2)\gamma_{5})\gamma_{\mu}\notag \\ && - i (f_V(k^2) + h_A(k^2)\gamma_5)\sigma_{\mu \nu}k^{\nu}-(h_V(k^2) - f_A(k^2)\gamma_{5})k_{\mu}u(p) \label{31}]\end{aligned}$$ where $f_V(k^2)\;, g_V(k^2), f_A(k^2)\;, g_A(k^2)\;, h_V(k^2)$ and $h_A(k^2)$ are the form factors. Now using Eqs. (\[30a\]) and (\[31\]) in Eq. (\[30\]), the $T-$matrix can be written as $$T=\frac{1}{(2\pi)^{9/2}}\sqrt{\frac{m m^{\prime}}{2 k_0 p_0 p^{\prime}_0 k_0}}\frac{G_{F}}{\sqrt{2}} V_{cb}V^{*}_{cs}a_2 F_{J/\Psi}m_{J/\Psi} \bar{u}(p^{\prime})[\gamma \cdot \epsilon (g_V(k^2)+g_A(k^2)\gamma_{5})- i \epsilon^{\mu}\sigma_{\mu \nu}k^{\nu}(f_V(k^2)+h_A(k^2)\gamma_{5})]u(p). \label{T-matrix}$$ Hence, comparing Eq. (\[12\]) and Eq. (\[T-matrix\]), one gets $$\begin{aligned} A & = & G^{\prime}m_{J/\psi} F_{J/\psi} g_V(k^2)\;,\notag \\ B & = & G^{\prime}m_{J/\psi} F_{J/\psi} g_A(k^2)\;, \notag \\ C & = & - G^{\prime}m_{J/\psi} F_{J/\psi} f_V(k^2)\;,\notag \\ D & = & - G^{\prime}m_{J/\psi} F_{J/\psi} h_A(k^2)\;, \label{32}\end{aligned}$$ where $$G^{\prime} =V_{cb}V^{*}_{cs}\frac{G_{F}}{\sqrt{2}}(C_2+\zeta C_1)\;.$$ The short distance QCD effects are taken care of in the Wilson Coefficients $C_1$ and $C_2$. The long distance interactions are shifted to the form factors $g_V$, $g_A$, $f_V$ and $h_A$ which are needed to be evaluated in some model. Using the heavy quark spin symmetry, one can relate the different form factors [@Mannel:1990vg] for which there are two choices: $$\begin{aligned} (i):\ \ \ \ \ \ \ \ \ \ g_V(k^2) = g_A(k^2) = F_{1}(k^2)+\frac{m_b}{m}F_{2}(k^2)\;, \notag \\ \ \ \ \ \ \ \ \ \ \ \ \ f_V(k^2) = h_A(k^2) = \frac{1}{m}F_2(k^2) \notag \\ (ii):\ \ \ \ \ \ \ \ \ \ g_V(k^2) = - g_A(k^2) = F_{1}(k^2)+\frac{m_b}{m}F_{2}(k^2)\;, \notag \\ \ \ \ \ \ \ \ \ \ \ \ \ f_V(k^2) = - h_A(k^2) = \frac{1}{m}F_2(k^2)\;, \label{33}\end{aligned}$$ where $m = m_{\Lambda_b}$ and $m_b$ is the mass of $b-$quark which in this work is taken to be to be $4.65$GeV. Thus in terms of the form factors $F_{1}(k^2)$ and $F_2(k^2)$, we can write $$\begin{aligned} A & = & G^{\prime}m_{J/\psi} F_{J/\psi} F_{1}(k^2)\bigg(1+\frac{m_b}{m}\frac{F_{2}(k^2)}{F_{1}(k^2)}\bigg) = \pm B\;, \notag \\ C & = & - G^{\prime}m_{J/\psi} F_{J/\psi} F_1(k^2)\frac{1}{m}\frac{F_2(k^2)}{F_1(k^2)} = \pm D.\label{34}\end{aligned}$$ where $\pm$ sign in Eq. (\[34\]) corresponds to the choices $(i)$ and $(ii)$, respectively. We need the form factors at $k^2 = m^2_{J/\psi}$: $$F_1( m^2_{J/\psi}) \equiv F_1 \; , \frac{F_2( m^2_{J/\psi})}{F_1( m^2_{J/\psi})} \equiv \frac{F_2}{F_1}\;. \label{35}$$ From Eqs. (\[16a\] - \[16d\]) by using Eq. (\[34\]) and Eq. (\[35\]), we can express the amplitudes $f_1$, $g_1$, $f_2$ and $h$ in terms of form factors $F_1$ and $F_2/F_1$ as $$\begin{aligned} f_1 & = & R \frac{|\vec{k}|}{\sqrt{2p^{\prime}_0(p^{\prime}_0 + m^{\prime})}}F_1\big[1+(m_b+(m + m^{\prime}))\frac{1}{m}\frac{F_2}{F_1}\big]\;, \label{36a} \\ g_1 & = & R \frac{p^{\prime}_0 + m^{\prime}}{\sqrt{2p^{\prime}_0(p^{\prime}_0 + m^{\prime})}}F_1\big[\mp1+\big(\frac{\mp m_b(p^{\prime}_0 + m^{\prime})\pm k_{0}(m+m^{\prime})-m^2_{J/\Psi}}{(p^{\prime}_0 + m^{\prime})}\big)\frac{1}{m}\frac{F_2}{F_1}\big]\;, \label{36b} \\ f_2 & = &R \frac{|\vec{k}|}{k_0}\frac{m+m^{\prime}}{\sqrt{2p^{\prime}_0(p^{\prime}_0 + m^{\prime})}}F_1\big[1+\frac{m_b(m+m^{\prime})+m^2_{J/\Psi}}{m+m^{\prime}}\frac{1}{m}\frac{F_2}{F_1}\big]\;, \label{36c} \\ h & = &R \frac{(m + m^{\prime})k_{0}-m^2_{J/\Psi}}{\sqrt{2p^{\prime}_0(p^{\prime}_0 + m^{\prime})}}F_1\big[\mp 1+\big(\frac{\mp m_b(k_{0}(m+m^{\prime}-m^2_{J/\Psi})\pm m^2_{J/\Psi}(p^{\prime}_{0}+m^{\prime}))}{(k_{0}(m+m^{\prime})-m^{2}_{J/\Psi})}\big)\frac{1}{m}\frac{F_2}{F_1}\big]\;, \label{36d}\end{aligned}$$ with $R=G^{\prime}m_{J/\Psi}F_{J/\Psi}$ which is a dimensionless parameter. We now consider the decay $\Lambda_b \to \Lambda + J/\psi$ which is of experimental interest. In order to calculate this decay, various models to evaluate the form factors have been considered in the literature [@9; @10; @11]. In reference [@6] form factors were evaluated in a quark model, and their values are $F_1 \approx -0.219$ and $F_2/F_1 \approx 0.169$. After putting $F_2/F_1 \approx 0.169$ and other input parameters in Eqs. (\[36a\], \[36b\], \[36c\]) and Eq. (\[36d\]), the numerical values of the amplitudes are given in Table \[Table-I\]. These results can be extended for other baryons, by using physical masses for relevant parameters and $SU(3)$ symmetry. Amplitudes Numerical Values ------------ ---------------------- $f_1$ $R\;F_1 (0.644)$ $g_1$ $R\;F_1(\mp 0.880) $ $f_2$ $R\;F_1 (1.075)$ $h$ $R\;F_1 (\mp 1.197)$ Making use of the values of amplitudes outlines in Table \[Table-I\] ,the value of branching ratio for $\Lambda_b \to \Lambda + J/\psi$ decay is obtained to be (c.f. Eq. (\[20\])) $$\mathcal{B}_r \approx 1.18 \times 10^{-2} (C_2 + \zeta C_1)^2\;, \label{37}$$ where $\zeta = \frac{1}{N_c}$, where $N_c$ is the effective number of colors. As noted in [@6], there are two regime, viz $N_{c}<1/3$ (Eq. (46)) and large $N_c$ limit. Using the values of Wilson coefficients $C_{2} = -0.257$, $C_{1} = 1.009$ [@21] and for different values of $\zeta$ that correspond to the large $N_c$ limit, the values are given in Table \[large-NC\]. One can see that for $\zeta = 0$, our results of branching ratio is compared with the $8.9\times 10^{-4}$ that is obtained in ref. [@11]. $\Lambda_b \to \Lambda + J/\psi$ $\zeta = 0$ $\zeta = 0.01$ $\zeta = 0.05$ ---------------------------------- --------------------------- --------------------------- ------------------------ -- -- -- --    $Br$       $7.8\times 10^{-4}$       $6.1\times 10^{-4}$       $5.0\times 10^{-4}$ \[large-NC\] Similarly, for the value of $\zeta$ that correspond to small $N_c$ limit, the values of branching ratios are given in Table \[small-NC\]. The experimental value of the branching ratio [@17] is $$\mathcal{B}_{r}(\Lambda_b \to \Lambda J/\psi)\times \mathcal{B}_{r}(b \to \Lambda^{0}_{b}) = (5.8 \pm 0.8)\times 10^{-5}. \nonumber$$ Using $\mathcal{B}(b \to \text{baryon}) \approx 9.29 \times 10^{-3}$, the experimental value of branching ratio for $\Lambda_{b}\to \Lambda J/\psi$ is $\mathcal{B}_{r} = (6.2 \pm 0.8)\times 10^{-4}$ and it is comparable to our value $6.1\times 10^{-4}$ when $\zeta = 0.48$ as well as for $\zeta = 0.01$. $\Lambda_b \to \Lambda + J/\psi$ $\zeta = 1/3$ $\zeta = 0.40$ $\zeta = 0.45$ $\zeta = 0.48$ $\zeta = 0.50$ ---------------------------------- ---------------------------- --------------------------- --------------------------- --------------------------- ------------------------ --    $Br$       $0.74\times 10^{-4}$       $2.6\times 10^{-4}$       $4.6\times 10^{-4}$       $6.1\times 10^{-4}$       $7.2\times 10^{-4}$ \[small-NC\] The values of asymmetry parameters for $\Lambda_b \to \Lambda + J/\psi$ decay are obtained from Eqs. (\[28\]) and (\[29\]) and these are $$\begin{aligned} \alpha \approx \mp 0.19 \; \; \; ,\; \; \alpha_{T} \approx \pm 0.39 \; \; \; , \; \; \alpha_{L} \approx \mp 0.58 \notag \\ \alpha^{\prime} \approx \mp 0.98 \; \; \; , \; \; \alpha^{\prime}_{T} \approx \mp 0.39 \; \; \; , \; \; \alpha^{\prime}_{L} \approx \mp 0.58 \label{39}\end{aligned}$$ The experimental value of the asymmetry parameter $\alpha = 0.18 \pm 0.13$. With our choice $(i)$ of the form factors given in Eq. (\[33\]), the value of asymmetry $\alpha = -0.19$ is comparable to the values obtained in refs. [@3; @4; @5; @6; @7; @8; @9]. However, for choice $(ii)$ of the form factors, the value of asymmetry parameter $\alpha = 0.19$ is comparable to the experimental value $\alpha = 0.18 \pm 0.13$. We have discussed $\Lambda_b \to \Lambda + J/\Psi$ decay in detail and with this in hand, for heavy baryon belonging to the representation $\bar{3}$ and $6$, the branching ratio can be easily obtained by using $SU(3)$ symmetry, taking into account the phase space for each baryon decay. For the decays $\mathcal{B}_{b}(\frac{1}{2}^{+}) \to \mathcal{B}({\frac{1}{2}}^{+}) J/\Psi$, $SU(3)$ gives the relation $$\bar{3}:\ \ \ \ \ \ (\Xi_{b}^{-}\;, \Xi_{b}^{0}\;, \Lambda_b) \to (\Xi^{-}\;, \Xi^{0}\;, \Lambda)J/\Psi: \ \ \ \ (1\;, 1\;, \sqrt{2/3})\;.\label{39a}$$ for $\mathcal{B}_{b}(\frac{1}{2}^+)$ belong to representation $\bar{3}$ and $\mathcal{B}(\frac{1}{2}^+)$ belonging to the octet representation. In case of $\mathcal{B}_{b}(\frac{1}{2}^{+})$ belonging to the sextet representation and $\mathcal{B}(\frac{1}{2}^{+})$ belonging to the representation $8^{\prime}$, $SU(3)$ gives $$\begin{aligned} (\Sigma_{b}^{+}\;, \Sigma_{b}^{0}\; \Sigma_{b}^{-}) &\to & (\Sigma^{+}\;, \Sigma^{0}\;, \Sigma^{-})J/\Psi : \ \ \ \ \ \sqrt{2}(-1\;, 1\;, 1)\;, \notag \\ (\Xi^{\prime -}_{b}\;, \Xi^{\prime 0}_{b}) & \to & (\Xi^{-}\;, \Xi^{0})J/\Psi: \ \ \ \ (1\;, 1)\;.\label{39b}\end{aligned}$$ The Decay $\Omega_{b} \to \Omega^{-} + J/\Psi$ =============================================== In the factorization ansatz, corresponding to the effective Hamiltonian given in Eq. (\[4\]) the matrix element for $\Omega_{b} \to \Omega^{-} + J/\Psi$ decay is $$\mathcal{M} = \frac{G_F}{\sqrt{2}}V_{cb}V^{*}_{cs} (C_2 +\zeta C_1)\langle 0|\bar{c}\gamma^{\mu}(1-\gamma^{5})c|J/\Psi\rangle \langle\Omega^{-}|\bar{s}\gamma_{\mu}(1-\gamma^{5})b|\Omega_{b}^{-}\rangle. \label{41}$$ We can write $$\langle\Omega^{-}|\bar{s}\gamma_{\mu}(1-\gamma^{5})b|\Omega_{b}^{-}\rangle =\frac{1}{(2\pi)^3}\sqrt{\frac{mm^{\prime}}{p_0 p^{\prime}_0}}[(F_1^{V}- \gamma^{5}F_1^{A})(\bar{u}_{\mu}(p^{\prime})u(p))+. . . ]\;.\label{42}$$ where dots denote the contribution from other form factors which are suppressed by a factor of $\frac{1}{m_{\Omega_b}}$ compared to $F_1^{V}$ and $F_1^{A}$ and hence will be neglected. From Eq. (\[41\]) and Eq. (\[42\]) along with Eq. (\[33\]), we get $$|\mathcal{M}|^2 = G^{\prime}F^{2}_{J/\Psi}m^{2}_{J/\Psi}\epsilon^{\mu}\epsilon^{\nu}[u_{\nu}(p^{\prime})\bar{u}_{\mu}(p^{\prime})(F^{V}_{1}-\gamma^{5}F^{A}_1)u(p)\bar{u}(p)(F^{*\; V}_{1}+\gamma^{5}F^{*\; A}_1)]\;, \label{43}$$ where $G^{\prime} = \frac{G_F}{\sqrt{2}}V_{cb}V^{*}_{cs}(C_2+\zeta C_1)$. Now $$\begin{aligned} \sum_{\text{Polarization}}\epsilon^{\mu}(k)\epsilon^{\nu}(k)&=&\big(-\eta^{\mu\; \nu}+\frac{k^{\mu}k^{\nu}}{m^2_{J/\Psi}}\big)\;,\notag \\ \sum_{\text{Spin}}u_{\nu}(p^{\prime})\bar{u}_{\mu}(p^{\prime}) &=& =-\frac{\gamma \cdot p^{\prime} +m^{\prime}}{2m}\big[\eta_{\nu\; \mu}-\gamma_{\nu}\gamma_{\mu}+\frac{i}{3m^{\prime}}(\gamma_{\nu}p^{\prime}_{\mu}-p^{\prime}_{\nu}\gamma_{\mu})-\frac{2}{3m^{\prime\; 2}}p^{\prime}_{\nu}p^{\prime}_{\mu}\big]\;,\notag \\ \bar{\sum_{\text{spin}}}u(p)\bar{u}(p) &=&\frac{1}{2}\frac{\gamma \cdot p+m}{2m}\;.\label{44}\end{aligned}$$ Using above equations, the decay rate is given by $$\Gamma = \frac{1}{2\pi m}|\vec{k}|(G^{\prime}F_{J/\Psi}m_{J/\Psi})^2\big(1+\frac{1}{3}\frac{m^2}{m^{\prime\; 2}}\frac{|\vec{k}|^2}{m^2_{J/\Psi}}\big)\big[|F^{V}_{1}|^2(p^{\prime}_0+m^{\prime})+|F^{A}_{1}|^2)(p^{\prime}_{0}-m^{\prime})\big](C_2+\zeta C_1)^{2}\;.\label{45}$$ In particular, for $\Omega^{-}_{b} \to \Omega^{-} + J/\Psi$, we have $m = m_{\Omega_{b}}$ and $m^{\prime} = m_{\Omega}$. Now $$\begin{aligned} |\Omega^{-} \rangle&=& \frac{1}{\sqrt{3}}(s^{\uparrow}s^{\uparrow}s^{\downarrow}+s^{\uparrow}s^{\downarrow}s^{\uparrow}+s^{\downarrow}s^{\uparrow}s^{\uparrow}\rangle\nonumber\\ |\Omega_{b}^{-} \rangle&=& -\frac{1}{\sqrt{6}}|s^{\uparrow}s^{\downarrow}b^{\uparrow}+s^{\downarrow}s^{\uparrow}b^{\uparrow}-2s^{\uparrow}s^{\uparrow}b^{\downarrow}\rangle\nonumber\\end{aligned}$$ In NRQM, relevant operators to $\mathcal{O}(v^2/c^2)$ are (for details see [@FD-2017]) $\beta$ and $\beta \sigma_{i}$ with $i = z$. Using $\beta|b\rangle = |s\rangle$, we have $$\beta |\Omega_{b}^{-},\frac{1}{2}\rangle = -\frac{1}{\sqrt{6}}|(s^{\uparrow}s^{\downarrow}s^{\uparrow}+s^{\downarrow}s^{\uparrow}s^{\uparrow}-2s^{\uparrow}s^{\uparrow}s^{\downarrow})\rangle$$ and $$\beta\sigma_{z} |\Omega_{b}^{-},\frac{1}{2}\rangle = -\frac{1}{\sqrt{6}}|(s^{\uparrow}s^{\downarrow}s^{\uparrow}+s^{\downarrow}s^{\uparrow}s^{\uparrow}+2s^{\uparrow}s^{\uparrow}s^{\downarrow})\rangle.$$ This gives $F^{V}_{1}=0$ and $F^{A}_{1}(0) = -\frac{2\sqrt{2}}{3}$ [@18]. Thus $$\Gamma (\Omega^{-}_{b} \to \Omega^{-}+J/\Psi) \approx 1.76\times 10^{-14} |F^A_{1}|^2 (C_2+\zeta C_1)^2 \text{GeV}\;.\label{46}$$ Hence the branching ratio $$Br(\Omega^{-}_{b} \to \Omega^{-} + J/\Psi) = \frac{\tau_{\Omega_b}}{\hbar}\Gamma (\Omega^{-}_{b} \to \Omega^{-} + J/\Psi) = 2.94 \times 10^{-2}|F^A_{1}|^2 (C_2+\zeta C_1)^2\label{47}$$ where $F^{A}_{1} = F^{A}_{1}(m^2_{J/\Psi})$ and $\tau_{\Omega_b}$ is the decay time of $\Omega_{b}$. Now using $$F^{A}_{1} = \frac{1}{m_{J/\Psi}}\frac{m_{b}m_{s}}{m_{b}+m_{s}}F^{A}_{1}(0)\approx \frac{m_{s}}{m_{J/\Psi}}F^{A}_{1}(0) \approx 0.152\;, \label{48}$$ The form factor $F^{A}_{1}$ at $m_{J/\psi}$ is expected to be smaller than $F^{A}_{1}(0)$. For this purpose, we have introduced a dimensionless phenomenological factor $(\frac{1}{m_{J/\psi}})(\frac{m_{b}m_{s}}{m_{b}+m_{s}})$, where the second factor is the reduced mass of the constituents of $\Omega^{-}_{b}$. Using $F_{1}^{A}\approx 0.152$, the branching ratio is $$\mathcal{B}_r \approx 6.8 \times 10^{-4} (C_2+\zeta C_1)^2.$$ Corresponding to the different values of $\zeta$, the value of branching ratio is given in the Table \[Omega-value\]. $\Omega_b \to \Omega + J/\psi$ $\zeta = 0$ $\zeta = 0.01$ $\zeta = 0.05$ $\zeta = 1/3$ $\zeta = 0.40$ $\zeta = 0.44$ -------------------------------- --------------------------- --------------------------- --------------------------- --------------------------- --------------------------- ------------------------ --    $Br$       $4.5\times 10^{-5}$       $4.1\times 10^{-5}$       $2.9\times 10^{-5}$       $0.8\times 10^{-5}$       $1.8\times 10^{-5}$       $3.0\times 10^{-5}$ \[Omega-value\] Experimental $\mathcal{B}r(\Omega^{-}_{b} \to \Omega^{-} + J/\Psi)\times \mathcal{B}r(b \to \Omega_{b}) = (2.9^{+1.1}_{-0.8})\times 10^{-6}$ with $\mathcal{B}(b \to \text{baryon}) \approx 9.29 \times 10^{-3}$ [@17] gives $$\mathcal{B}r(\Omega^{-}_{b} \to \Omega^{-}J/\Psi) =(3.12^{+1.1}_{-0.8})\times 10^{-5}\;. \label{50}$$ Finally, in this model, the asymmetry parameter is $$\alpha(\Omega^{-}_{b} \to \Omega^{-}J/\Psi) = 0\;.\label{51}$$ To conclude: using the effective Lagrangian together with factorization ans[ä]{}tz the two body hadronic decay $\mathcal{B}_{b}(\frac{1}{2}^+) \to \mathcal{B}(\frac{1}{2}^{+},\; \frac{3}{2}^{+}) + V$ is calculated. In case of the $\mathcal{B}_{b}(\frac{1}{2}^{+})$ belonging to the representation $\bar{3}$, the only allowed decay channel is $\mathcal{B}_{b}(\frac{1}{2}^+) \to \mathcal{B}(\frac{1}{2}^{+})+ V$, where $\mathcal{B}(\frac{1}{2}^{+})$ belong to the representation $8$ of $SU(3)$. However, if $\mathcal{B}_{b}(\frac{1}{2}^{+})$ belongs to the sextet representation $6$, the allowed decay channels are $\mathcal{B}_{b}(\frac{1}{2}^+) \to \mathcal{B}(\frac{1}{2}^{+},\; \frac{3}{2}^{+}) + V$ where $\mathcal{B}(\frac{1}{2}^{+})$ and $\mathcal{B}(\frac{3}{2}^{+})$ belong to the octet representation $8^{\prime}$ and the decuplet $10$ of $SU(3)$, respectively. We have analyazed the decay channel $\mathcal{B}_{b}(\frac{1}{2}^+) \to \mathcal{B}(\frac{1}{2}^{+}) + V$ in detail, where the decay rate $\Gamma$ and the asymmetry parameters $\alpha\;, \alpha^{\prime}\;, \beta\;, \gamma$ and $\gamma^{\prime}$ are expressed in terms of four amplitudes. These amplitudes are written in terms of the transverse and the longitudinal polarization of $V$. This general formalism is then applied to the decay $\Lambda_b \to \Lambda J/\psi$. It is shown that within the factorization framework, using heavy quark spin symmetry, the decay rate and asymmetry parameters can be expressed in terms of two form factors $F_1$ and $F_{2}/F_{1}$, which being the non-perturbative quantities needed to be evaluated in some model. Here, by taking the values of these form factors calculated in the quark model [@6] the branching ratio and asymmetry parameters $\alpha$ and $\alpha^{\prime}$ are obtained numerically. By taking the color factor $\zeta = 0.01$ or $\zeta = 0.48$ , the branching ratio for the decay $\Lambda_b \to \Lambda + J/\psi$ is matchable to the corresponding experimental value. Having worked out $\Lambda_b \to \Lambda + J/\psi$ decay, this formalism can be easily applied to other heavy quarks belonging to triplet and the sextet representation, by using $SU(3)$ symmetry and the phase space factor. Finally, the decay $\Omega_{b}^{-} \to \Omega^{-} + J/\psi$ is analyzed within the factorization framework. It is shown that the asymmetry parameter $\alpha$ in this particular decay is zero. The branching ratio obtained in the first approximation is compared with the experimental value. [99]{} R. Aaij et al. (LHCb Collaboration), Phys. Lett. [**B**]{} 724, 27 (2013) \[arXiv:1302.5578 \[hep-ex\]\]. A. K. Leibovich, Z. Ligeti, I. W. Stewart and M. B. Wise, Phys. Lett. B [**586**]{}, 337 (2004) \[hep-ph/0312319\]. H. -Y. Cheng and B. Tseng, Phys. Rev. D [**53**]{}, 1457 (1996) \[Erratum-ibid. D 55, 1697 (1997)\] \[hep-ph/9502391\]. H. -Y. Cheng, Phys. Rev. D [**56**]{}, 2799 (1997) \[hep-ph/9612223\]. M. A. Ivanov, J. G. Korner, V. E. Lyubovitskij and A. G. Rusetsky, Phys. Rev. D [**57**]{}, 5632 (1998) \[hep-ph/9709372\]. Fayyazuddin and Riazuddin, Phys. Rev. D [**58**]{}, 014016 (1998) \[hep-ph/9802326\]. R. Mohanta, A. K. Giri, M. P. Khanna, M. Ishida, S. Ishida and M. Oda, Prog. Theor. Phys. [**101**]{}, 959 (1999) \[hep-ph/9904324\]. C. -H. Chou, H. -H. Shih, S. -C. Lee and H. -n. Li, Phys. Rev. D [**65**]{}, 074030 (2002) \[hep-ph/0112145\]. Z. -T. Wei, H. -W. Ke and X. -Q. Li, Phys. Rev. D [**80**]{}, 094016 (2009) \[arXiv:0909.0100 \[hep-ph\]\]. L. Mott and W. Roberts, Int. J. Mod. Phys. A [**27**]{}, 1250016 (2012) \[arXiv:1108.6129 \[nucl-th\]\]. T. Gutsche, M. A. Ivanov, J. G. Korner, V. E. Lyubovitskij and P. Santorelli, Phys. Rev. D [**88**]{} (2013) no.11, 114018 doi:10.1103/PhysRevD.88.114018 \[arXiv:1309.7879 \[hep-ph\]\]. J. Beringer et al. (Particle Data Group), Phys. Rev. D [**86**]{}, 010001 (2012) and 2013 partial update for the 2014 edition. F. Abe et al. (CDF Collaboration), Phys. Rev. D [**55**]{}, 1142 (1997). V. M. Abazov et al. (D0 Collaboration), Phys. Rev. D [**84**]{}, 031102 (2011) \[arXiv:1105.0690 \[hep-ex\]\]. See for example: Fayyazuddin and Riazuddin, A Modern Introduction to Particle Physics, 3rd Edition, Chapter 8 (sec. 8.4), World Scientific Singapore, 2011. F. Hussain, J. G. Korner, M. Kramer and G. Thompsan, Z. Phys. C [**51**]{} (1991) 321. Reference [@15], Chapter 10, section 10.51. T. Mannel, W. Roberts and Z. Ryzak, Nucl. Phys. B [**355**]{}, 38 (1991). C. Patrignani *et al.* (Particle Data Group), Chin. Phys. [**C40**]{} (2016) 100001. Fayyazuddin, Two Body Hadronic Decays $\Lambda_{b}(\frac{1}{2}^{+})\rightarrow B^{\ast}(\frac{3}{2}^{+})+P$ in a quark model, Accepted for publication in Phys. Rev. D. Fayyazuddin and Riazuddin, Phys. Rev. D[**55**]{}, 255 (1997). A. Buras *it al*, Nucl. Phys. [**B370**]{} (1991) 38; A. Ali and C. Greub, Phys. Rev. [**D 57**]{} (1998) 2996. W. Altmannshofer, P. Ball, A. Bharucha, A. J. Buras, D. M. Straub and M. Wick, JHEP [**0901**]{} (2009) 019 \[arXiv:0811.1214 \[hep-ph\]\].

ArXiv

--- abstract: 'Motivated by the improving sensitivity, $R$, of experiments on $\mu~Ti \rightarrow e~Ti$ and the enhanced Higgs nucleon interaction, we study this lepton number violating process induced by Higgs exchange. Taking the possible sensitivity, $R \simeq 10^{-16}$, we obtain the constraint on the Higgs-muon-electron vertex, $\kappa_{\mu e}$, to be less than $2.4\times10^{-7}$ if the masses of the Higgs scalar and $W$ gauge boson are the same. $\kappa_{\mu e}$ is also calculated for some models.' address: | TRIUMF, 4004 Wesbrook Mall\ Vancouver, B.C., V6T 2A3, Canada author: - 'Daniel Ng and John N. Ng' title: | Can $\mu$–$e$ Conversion in Nuclei be a Good Probe for\ Lepton-Number Violating Higgs Couplings ? --- In this paper we report on a study of direct muon-electron conversion in nuclei as a probe of new physics represented by an effective $\mu-e-S$ vertex where $S$ is a neutral scalar particle. For example, the scalar $S$ can be the Higgs scalars in a one doublet model with an extended fermion sector, or linear combinations of scalar particles in some extended Higgs models such as the supersymmetric standard model. Previous discussion of $\mu-e$ conversion concentrated mostly on the effects of virtual photon and $Z$–boson exchanges [@shanker]. Effects of an extra $Z$–boson has also been considered recently [@bernabeu]. Scalar and pseudoscalar effects were outlined in Ref. [@shanker] where the details of the nuclear effects were emphasized. If the Higgs coupling to nucleon is taken to be proportional to the current masses of the $u$– and $d$–quarks, then the effect would be very small. Here, we treat the scalar-nucleon-nucleon via the approach of Shifman, et. al. [@shifman], which increase the coupling strength to that of $\frac{2}{27}m_N$ where $m_N$ is the mass of the nucleon. This is approximately one order of magnitude enhancement over the use of the current quark masses. A second enhancement of the $S-N-N$ coupling can arise in extended Higgs model where it is multiplied by ratios of vacuum expectation values of scalars fields. In supersymmetry, the ratio $\tan\beta \geq 10$ is certainly acceptable. The third factor comes from the fact that scalar exchange in $\mu-e$ is coherent [^1] over the nuclei [@goodman]. In this respect, it is similar to photon and Z exchange. On the experimental side, we are encouraged by the on going experiment at PSI [@psi] of $\mu ~ Ti \rightarrow e ~ Ti$ which will achieve a sensitivity $R(Ti)=\Gamma({\mu Ti \rightarrow e Ti})/\Gamma({\mu Ti \rightarrow \nu_\mu Ti})\simeq 3\times10^{-14}$ and prospects of lowering this limit to the level of $10^{-15}-10^{-16}$ being considered at INS Moscow [@melc] and TRIUMF [@triumf]. This motivated us to reexamine $\mu-e$ conversion and focus on it as a probe of the non-standard $\mu-e-S$ vertex. Comparison with $\mu \rightarrow e~\gamma$ and/or $\mu \rightarrow 3e$ where they appliy are also given. At the quark level, the effective interaction Lagrangian induced by an exchange of a scalar $S$ is given by $${\cal L}_q(S) = \frac{G_F}{\sqrt2}~\frac{m_W^2}{m_S^2}~ \bar e~\left[~\kappa_{\mu e}~\left(1+\gamma_5\right)~+~ ~\kappa'_{\mu e}~\left(1-\gamma_5\right)\right]~\mu \sum_{\scriptstyle q=all} \frac{m_q}{m_W}~\lambda_q~\bar q q \ ,$$ where $\kappa_{\mu e}$ and $\kappa'_{\mu e}$ are coefficient of the effective $\mu-e-S$ vertex and $G_F$ is the Fermi coupling constant. $m_S$ and $m_W$ are the masses for the scalar $S$ and the standard W gauge boson whereas $m_q$’s are the current quark masses and the sum is taken over for all quark flavors of a given model. For extended Higgs models, $\lambda_q$ is not equal to unity as in the standard model. In particular, in the two Higgs doublet extension of the standard model with natural flavor conservation, we have $\lambda_{up} = \cot\beta$ and $\lambda_{down} = \tan\beta$. These correspond to the linear combination given by $S=-\sin\beta \sqrt2 Re \phi_1^0 + \cos\beta \sqrt2 Re \phi_2^0$, where $\phi_1^0$ and $\phi_2^0$ are the neutral components of the Higgs doublets that provide masses for $down$– and $up$–type quarks separately. In order to see how the discussed factors enter into the study of lepton number violation in general and the $\mu-e-S$ vertex in particular, we first study the cases where the scalar $S$ couples to quark like that of the standard model Higgs boson, $H$. The effects of Higgs mixing will be illustrated by the minimal supersymmetric model with lepton number violation added in. To compute the interaction at the nucleon level, we follow the procedure suggested by Shifman et. al. [@shifman]. Including the effects of the strange and heavy quark contributions [@cheng; @gasser], we obtain the effective Lagrangian for $\mu-e$ conversion in nuclei as follows, $${\cal L}_N(S) = \frac{G_F}{\sqrt2}~\frac{m_W^2}{m_S^2}~ \bar e\left[~\kappa_{\mu e}~\left(1+\gamma_5\right)~+~ ~\kappa'_{\mu e}~\left(1-\gamma_5\right)\right]~\mu ~\frac{\tilde m_N}{m_W}~\bar\Psi_N\Psi_N \ ,$$ and $$\label{mn} {\tilde m_N}= \frac{2}{27} n_h m_N + \left(1+\frac{y}{2}\frac{m_s}{\bar m} \right) \left(1- \frac{2}{27} n_h \right) \sigma_{\pi N} \ ,$$ where $\Psi_N$ is the nucleon wave function. $n_h$ is the number of heavy quarks other than $u$, $d$ and $s$. $y$ is the strange content in the nucleon and $\sigma_{\pi N}$ is nucleon matrix element of the $\sigma$ term in the chiral Lagrangian. $m_N$ is the nucleon mass, $\bar m = (m_u+m_d)/2$, and we take ${m_s}/{\bar m} \simeq 25$. The quantity $\tilde m_N$ conveniently expresses the heavy quark effects in Higgs-nucleon-nucleon coupling. Its value depends on $y$ and $\sigma_{\pi N}$, where $(y,\sigma_{\pi N}) = (0,0),~(0.47,60{\rm MeV})$ and $(0.22,45{\rm MeV})$ are used in Refs. [@shifman], [@cheng] and [@gasser] respectively. Particularly, $\tilde m_N = 350 \rm MeV$ for $(y,\sigma_{\pi N})=(0.22,45{\rm MeV})$ for $n_h=3$. Hence, $\tilde m_N$ is almost two orders of magnitude bigger than the current quark masses of $u$ and $d$. This a a larger enhancement factor than originally anticipated as discussed in the introduction. Using the standard procedure [@feinberg], we obtain the conversion rate of $\mu~N \rightarrow e~N$ as follows, $$\begin{aligned} \Gamma(\mu N \rightarrow e N)= &&\frac{G_F^2}{2}\left(\frac{\tilde m_N}{m_W}\right)^2 \frac{\alpha^3 m_\mu^5 Z_{eff}^4}{\pi^2 Z}A^2 |F(q^2)|^2 \frac{m_W^4}{m_S^4} \nonumber \\ && \int \left[ |\kappa_{\mu e}|^2 \frac{\left(1-{\bf s_{\mu}} \cdot {\bf \hat p_e} \right)}{2}+ |\kappa'_{\mu e}|^2 \frac{\left(1+{\bf s_{\mu}} \cdot {\bf \hat p_e} \right)}{2}\right] d \cos\theta\end{aligned}$$ where $A$ and $Z$($Z_{eff}$) are the nucleon and (effective) atomic numbers. $F(q^2)$ is the nucleon form factor. $\bf s_\mu$ and $\hat p_e$ are the muon spin and the direction of the outgoing electron. Particularly, $Z_{eff}=17.6$ [@pla] and $F(q^2=-m_{\mu}^2)=0.54$ [@frois] for $^{22}_{48}Ti$. Using the muon capture rate in $Ti$, $\Gamma({\mu Ti \rightarrow \nu_\mu Ti})= 2.590\pm 10^6 sec^{-1}$ [@suzuki], we obtain $$\label{const} \left(|\kappa_{\mu e}|^2+|\kappa'_{\mu e}|^2\right)^{1/2} \leq 2.4\times10^{-7} \left(\frac{0.5 {\rm GeV}}{\tilde m_N}\right) \left(\frac{R}{10^{-16}}\right)^{1/2}\frac{m_S^2}{m_W^2} \ .$$ This is the model independent constraint on $\kappa_{\mu e}$ and $\kappa'_{\mu e}$. In Ref. [@shanker], the author obtained the constraint on $\left(|\kappa_{\mu e}|^2+|\kappa'_{\mu e}|^2\right)^{1/2}\frac{\tilde m_N}{m_W}\frac{m_W^2}{m_S^2} \leq 10^{-6}$ for sulphur. If we take the current quark mass approach, namely $\tilde m_N = (m_u+m_d)/2 = 5 \rm MeV$, it yields $\left(|\kappa_{\mu e}|^2+|\kappa'_{\mu e}|^2\right)^{1/2} \leq 1.6\times 10^{-2}$ assuming $m_S=m_W$. Even with the improved sensitivity of two orders of magnitude, the constraint is no better than $10^{-5}$ if the current quark masses are used. Obviously, the improved calculation of Eq. (\[mn\]) gives a much better limit as evident from Eq. (\[const\]). Encouraged by the enhancement of the Higgs nucleon interaction, we study three examples to see when this lepton number violating Higgs interaction be important for the muon–electron conversion in nuclei. [ 1. ]{} In the standard model, the Yukawa interactions of Higgs $H$ and leptons are flavor diagonal, and is given by $$-\frac{g}{2m_W}~H~\left[m_e \bar e e + m_\mu \bar \mu \mu + m_\tau \bar \tau \tau \right] \ .$$ When we include exotic leptons which mix with the ordinary leptons, there will be lepton flavor changing Higgs interactions. In the lowest order, the coefficient of the $\mu-e-H$ vertex is given by $$\kappa_{\mu e} \simeq \frac{1}{m_W}\left( m_e U_{e\mu} + m_\mu U_{\mu e} + m_\tau U_{\tau e}U_{\tau \mu} \right)\ ,$$ where $U_{\alpha\beta}$ is the mixing in the charged lepton sector induced by exotic leptons. Using the constraint in Eq. (\[const\]), we obtain $U_{e\mu} \leq 0.04$, $U_{\mu e} \leq 2\times 10^{-4}$ and $U_{\tau e}U_{\tau \mu}\leq 1\times 10^{-5}$. From $\mu \rightarrow 3 e$ and $\tau \rightarrow 3 \ell $ [^2] by tree level exchanges of $Z$ gauge boson, the constraints are $U_{\mu e} \leq 3\times10^{-6}$ and $U_{\tau e}U_{\tau \mu}\leq 1.6\times10^{-3}$ [@london]. Hence, the future $\mu-e$ conversion experiments can improve the constraint on the $\tau-\mu$ and $\tau-e$ mixings by two orders. In the following two examples, we consider models with lepton flavor conservation at tree level. Hence both the $\mu-e-S$ and $\mu-e-\gamma$ vertices are induced at one-loop level. [ 2. ]{} When the standard model is extended to include the 4th generation, a right-handed neutrino is necessary to provide the mass for the 4th neutrino which must be heavier than $45 {\rm GeV}$ from the LEP experiments, leading to three massless and one massive neutrinos. In this model, the scalar $S$ is the standard model Higgs, $H$. Since the vertex $\mu-e-H$ is induced by the $V-A$ current, hence $\kappa'_{\mu e} =0$; whereas $\kappa_{\mu e}$ [@grzadkowski] is given by $$\kappa_{\mu e} = \frac{g^2}{16\pi^2}U_{\mu 4}U_{e 4}\frac{m_\mu}{m_W} \left[ \frac{3}{4}x+\frac{m_H^2}{m_W^2}\left( \frac{3x-x^2}{8(x-1)^2} +\frac{x^3-2x^2}{4(x-1)^3}\ln x \right) \right] \ ,$$ where $x=m_{\nu_4}^2/m_w^2$ and $U_{\alpha\beta}$ is the CKM matrix in the lepton sector of four flavors. For the decay $\mu \rightarrow e~\gamma$, the decay rate is given by $$\Gamma(\mu \rightarrow e \gamma)=\frac{\alpha^3m_\mu}{64\pi\sin^4\theta_W} \frac{m_\mu^2}{m_W^2}|U_{\mu 4}U_{e 4}I(x)|^2$$ and $$I(x)=\frac{-x+5x^2+2x^3}{4(1-x)^3}+\frac{3x^3}{2(1-x)^4}\ln x \ .$$ where $U_{\mu 4}U_{e 4} \leq 3\times10^{-3}$ for $m_{\nu_4} \geq 45 \rm GeV$ are obtained [@acker]. However, $\kappa_{\mu e}$ is suppressed by ${m_\mu}/{m_W}$. Unless $m_{\nu_4}$ is greater than $2$ TeV, the processes $\mu \rightarrow e~\gamma$ is more important to probe the lepton number violation mechanism. [ 3. ]{} In the minimal supersymmetric extension of the standard model (MSSM), the lepton number processes can be induced through the slepton mixing. In analogy to the Yukawa interactions, there exist soft-breaking terms, $A Re\phi^0_1 {\tilde e_L}^\ast {\tilde \mu_R} + h.c.$. Therefore the mass matrix in the basis $\{{\tilde e_L},{\tilde \mu_R}\}$ is given by $$\pmatrix { \tilde m_e^2 & A v_1 \cr A v_1 & \tilde m_{\mu}^2 \cr} \ .$$ yielding the decay rate for $\mu \rightarrow e~\gamma$ to be $$\Gamma(\mu \rightarrow e \gamma)=\frac{\alpha^3 m_\mu}{256\pi^2\sin^4\theta_W} \sin^22\theta |\Gamma_{\mu e \gamma}|^2 \ ,$$ where $$\Gamma_{\mu e \gamma}=\sum_{\scriptstyle i=1}^{\scriptstyle 4} \tan\theta_W N_{1i}\left(N_{2i}+N_{1i}\tan\theta_W\right) \frac{m_\mu}{m_{\chi_i}} \left[ x_i F(x_i) - y_i F(y_i) \right] \ ,$$ with $$F(x)=-\frac{1+x}{2(x-1)^2}+\frac{x\ln{x}}{(x-1)^3} \ .$$ $\sin\theta$ and $N_{ij}$ are the scalar and neutralino mixing parameters. $x_i=m_{\chi_i}^2/m_1^2$ and $y_i=m_{\chi_i}^2/m_2^2$, where $m_{\chi_i}$ and $m_{1,2}$ are the neutralino and slepton masses. The $\mu-e$ conversion in nuclei is induced [@comment] by the the vertex, $\sqrt{2}Re\phi^0_1 {\tilde e_L}^\ast {\tilde \mu_R}$, leading to the coefficient of the effective vertex $\mu-e-\sqrt{2}Re\phi^0_1$, $$\begin{aligned} \kappa_{\mu e} = &&\frac{g}{32\pi^2} \sum_{\scriptstyle i=1}^{\scriptstyle 4} \tan\theta_W N_{1i}\left(N_{2i}+N_{1i}\tan\theta_W\right) \frac{A}{\sqrt{2} m_{\chi_i}} \nonumber \\ && \left\{ \sin^22\theta \left[ x_i G(x_i)+ y_i G(y_i) \right] + 2\cos^22\theta \left[x_i H(x_i,m_2^2/m_1^2)\right] \right\}\ ,\end{aligned}$$ where $$\begin{aligned} G(x)=&&\frac{1}{1-x}+\frac{x\ln{x}}{(1-x)^2} \ ,\\ H(x,r)=&&\frac{x\ln{x}}{(x-1)(x-r)}+\frac{r\ln{r}}{(r-1)(r-x)} \ ,\end{aligned}$$ and $$\label{nucmass} \frac{\tilde m_N}{m_S^2} = \left[\cos\beta\frac{m_A^2+m_Z^2\sin^2{2\beta}}{m_A^2m_Z^2\cos^2{2\beta}} -\sin\beta \frac{\tan{2\beta}}{m_A^2} \right] \tilde m_{N_1} +\left[\cos\beta\frac{\tan{2\beta}}{m_A^2} -\sin\beta\frac{1}{m_A^2}\right] \tilde m_{N_2}\ ,$$ where the effective nucleon mass induced by interacting with $\cos\beta\sqrt{2}Re\phi^0_1+\sin\beta\sqrt{2}Re\phi^0_2$ and $-\sin\beta\sqrt{2}Re\phi^0_1+\cos\beta\sqrt{2}Re\phi^0_2$ are given by $$\begin{aligned} \tilde m_{N_1}&=&\frac{2}{9}m_N+\frac{7}{9} \left(1+\frac{y}{2}\frac{m_s}{\bar m}\right)\sigma_{\pi N} \ ,\\ m_{N_2}&=& -\frac{2}{27}\left(\tan\beta-2\cot\beta\right)m_N -\left(\frac{4}{27}\cot\beta+\frac{25}{27}\tan\beta\right) \left(1+\frac{y}{2}\frac{m_s}{\bar m}\right)\sigma_{\pi N}\ .\end{aligned}$$ The square brackets in Eq. (\[nucmass\]) are the effective Higgs propagators. In table \[table\], we tabulate the branching ratio for $\mu \rightarrow e~\gamma$ and $\mu~Ti \rightarrow e~Ti$ in MSSM for different values of $A$ and $tan\beta$. We take $(y,\sigma_{\pi N}) = (0.22,45\rm GeV)$. For a large $\tan\beta$, $v_1=\sqrt{v_1^2+v_2^2}\cos\beta$ is small. thus the Higgs interaction would be at least as important as $\mu \rightarrow e~\gamma$. Particularly, for $A=500{\rm GeV}$, $\tan\beta=50$ and an intermediate mass scalar $m_A=250\rm GeV$, the process $\mu~Ti \rightarrow e~Ti$ is about $4$ times below the present experimental limit [@ahmad]; whereas the branching ratio for the process $\mu \rightarrow e~\gamma$ is $20$ times below the present experimental values [@bolton]. This is especially relevant when the sensitivity of the former is improved by two orders of magnitude; whereas we do not foresee a similar improvement in the $\mu \rightarrow e~\gamma$ measurement. In conclusion, we have considered the $\mu-e$ conversion in nuclei induced by Higgs exchange for three popular models. This process would be negligible if the Higgs nucleon coupling is taken to be proportional to the current quark masses. Here, we have shown how the Higgs nucleon interaction is enhanced by using the approach first employed by Shifman, et. al., and this yields $\kappa_{\mu e} \leq 2.4\times10^{-7}$. $\kappa_{\mu e}$ in a model of 4th generation lepton is small because it is suppressed by the muon mass. On the other hand, with the existence of the soft breaking terms in the MSSM, the Higgs induced $\mu-e$ conversion is at least as important as $\mu \rightarrow e~\gamma$. The process will be more important for a larger $\tan\beta$ as the rate increases as the square of this parameter. Furthermore, we have shown that $\mu-e$ conversion can be a sensitive probe to scalar particles in the mass range of hundreds of GeV even when the lepton-number violation is an one-loop effect. The minimal supersymmetric standard model is used as an illustrative example. This work was supported in part by the Natural Science and Engineering Council of Canada. $A$(GeV) $\mu \rightarrow e \gamma~^{\rm a}$ $R(\mu Ti\rightarrow e Ti)~^{\rm b}$ ---------- ------------------------------------- -------------------------------------- $500$ $4.7(0.2)\times10^{-11}$ $0.05(1.0)\times10^{-12}$ $250$ $11(0.5)\times10^{-12}$ $0.13(2.5)\times10^{-13}$ $50$ $46(1.9)\times10^{-14}$ $0.05(1.0)\times10^{-14}$ : The branching ratio for $\mu \rightarrow e~\gamma$ and $\mu~Ti \rightarrow e~Ti$ in MSSM. we take $\tan\beta=10(50)$, $m_A = 250 \rm GeV$ and $\tilde m_{e,\mu} = 5 \rm TeV$. For the gaugino masses, we take $2M_1=M_2=\mu=250 \rm GeV$. \[table\] O. Shanker, Phys. Rev. D[**20**]{}, 1608 (1979). J. Bernabeu, et. al., Preprint No. FTUV/93-24. M.A. Shifman, A.I. Vainstein and V.I. Zakharov, Phys. Lett. B[**78**]{}, 443 (1978). M.W. Goodman and E. Witten, Phys. Rev D[**31**]{}, 3059 (1985). A. Badertscher, et. al., SINDRUM II collaboration, preprint PSI-PR-90-41 (1990), invited talk given at 14th Europhysics Conf. on Nuclear Physics, Bratislava, Czechoslovakia, Oct 22-26, 1990. V. S. Abadjev, et. al., MELC collaboration, preprint INS/MOSCOW (1992). J. M. Poutissou, private communication. T.P. Cheng, Phys. Rev. D[**38**]{}, 2869 (1988). J. Gasser, H. Leutwyler and M.E. Sainio, Phys. Lett. B[**253**]{}, 252 (1991). G. Feinberg and S. Weinberg, Phys. Rev. Lett. [**3**]{}, 111, 244(E) (1959); W.J. Warciano and A.I. Sanda, ibid., [**78**]{}, 1512 (1977). K.W. Ford and J.G. Wills, Nucl. Phys. [**35**]{}, 295 (1962); R. Pla and J. Bernabeu, An. Fis. [**67**]{}, 455 (1971). B. Dreher, et. al., Nucl. Phys. A[**235**]{}, 219 (1974); B. Frois and C.N. Papanicolas, Ann. Rev. Nucl. Sci. [**37**]{}, 133 (1987). T. Suzuki, D.F. Measday and J.P. Roalsvig, Phys. Rev. C[**35**]{}, 2212 (1987). P. Langacker and D. London, Phys. Rev. D[**38**]{}, 886 (1988). B. Grzadkowski and P. Krawczyk, Zeit. Phys. C[**18**]{}, 43 (1983). We found that there is a sign typo in their Eq. (4). A. Acker and S. Pakvasa, Mod. Phys. Lett. A[**7**]{}, 1219 (1991). We neglect the effective $\mu-e-\sqrt{2}Re\phi^0_{1,2}$ vertices induced by attaching $\sqrt{2}Re\phi^0_{1,2}$ to the neutralino line because they are negligible if the constraint from $\mu \rightarrow e~\gamma$ is imposed. R.D. Bolton, et. al., Phys. Rev. D[**38**]{}, 2077 (1988). S. Ahmad, et. al., Phys. Rev. D[**38**]{}, 2101 (1988). [^1]: Such is not the case for pseudoscalar and axial vector exchanges and henceforth we shall ignore them [^2]: Note that when there are tree level lepton flavor changing interactions, the processes $\mu \rightarrow e~\gamma$ induced at one-loop level are less important than $\mu \rightarrow 3 e$ and $\tau \rightarrow 3 \ell$.

ArXiv

--- author: - Kenton Lee - Shimi Salant - Tom Kwiatkowksi - Ankur Parikh - Dipanjan Das - Jonathan Berant bibliography: - 'main.bib' title: Learning Recurrent Span Representations for Extractive Question Answering ---

ArXiv

--- author: - 'Pablo Groisman[^1]  and Matthieu Jonckheere[^2]' bibliography: - 'biblio2.bib' title: 'Front propagation and quasi-stationary distributions: the same selection principle?' --- [***Keywords***]{}: Selection principle, Quasi-stationary distributions, Branching Brownian Motion with selection, Traveling waves. [***AMS 2000 subject classification numbers***]{}: 60J70, 60J80, 60G50, 60G51. Introduction ============ A selection mechanism in front propagation can be thought of as follows: a certain phenomenology is described through an equation that admits an infinite number of traveling-wave solutions, but there is only one which has a physical meaning, the one with minimal velocity. Under mild assumptions on initial conditions, the solution converges to this minimal-velocity traveling wave. The most remarkable example of this fact is the celebrated F-KPP equation(for Fisher, Kolmogorov-Petrovskii-Piskunov) $$\label{KPP} \begin{array}{l} \displaystyle \frac{\partial v}{\partial t}= \displaystyle \frac12 \frac{\partial^2 v}{\partial x^2} + r f(v), \quad x \in \R,\, t>0,\\ \\ \displaystyle v(0,x)= \displaystyle v_0(x), \quad x \in \R. \end{array}$$ Assume for simplicity that $f$ has the form $f(s)=s^2 -s$, but this can be generalized up to some extent. We also restrict ourselves to initial data $v_0$ that are distribution functions of probability measures in $\R$. The equation was introduced in 1937 [@Fisher; @KPP] as a model for the evolution of a genetic trait, and since then has been widely studied (in fact more than two thousand works refer to one of these papers). Both Fisher and Kolmogorov, Petrovskii and Piskunov proved independently that this equation admits an infinite number of traveling wave solutions of the form $v(t,x) = w_c(x-ct)$ that travel at velocity $c$. This fact is somehow unexpected from the modeling point of view. In words of Fisher [@Fisher p. 359] > “Common sense would, I think, lead us to believe that, though the velocity of advance might be temporarily enhanced by this method, yet ultimately, the velocity of advance would adjust itself so as to be the same irrespective of the initial conditions. If this is so, this equation must omit some essential element of the problem, and it is indeed clear that while a coefficient of diffusion may represent the biological conditions adequately in places where large numbers of individuals of both types are available, it cannot do so at the extreme front and back of the advancing wave, where the numbers of the mutant and the parent gene respectively are small, and where their distribution must be largely sporadic”. Fisher proposed a way to overcome this difficulty, related to the probabilistic representation given later on by McKean [@MK], weaving links between solutions to and Branching Brownian Motion. The general principle behind is that microscopic effects should be taken into account to properly describe the physical phenomena. With a similar point of view in mind, Brunet, Derrida and coauthors [@BDMM; @BDMM2; @BD2; @BD] started in the nineties a study of the effect of microscopic noise in front propagation for equation and related models, which resulted in a huge number of works that study the change in the behavior of the front when microscopic effects are taken into account. These works include both numerical and heuristic arguments [@BDMM; @BDMM2; @BD2; @BD; @KL] as well as rigorous proofs [@BG; @BBS; @DR; @M2; @M]. Before that, Bramson et.al [@Betal] gave the first rigorous proof of a microscopic model for that has a unique velocity for every initial condition. They also prove that these velocities converge in the macroscopic scale to the minimum velocity of , and call this fact a [*microscopic selection principle*]{}, as opposed to the macroscopic selection principle stated above, that holds for solutions of the hydrodynamic equation. The theory of quasi-stationary distributions (QSD) has their own counterpart. It is a typical situation that there is an infinite number of quasi-stationary distributions, but the [*Yaglom limit*]{} (the limit of the conditioned evolution of the process started from a deterministic initial condition) selects the minimal one, i.e. the one with minimal expected time of absorption. Up to our knowledge, despite of a shared feeling that similar principles do occur in the context of QSD and of traveling waves, this relation has never been stated precisely. The purpose of this note is to show that they are two faces of the same coin. We first explain this link through the example of Brownian motion. Then we show how to extend these results to more general Lévy processes. The paper is organized as follows. In Section \[sec:macro\], we introduce traveling waves and QSDs as macroscopic models. We focus in particular on the KPP equation and the links between its traveling waves and the QSDs of a drifting Brownian motion. In Section \[sec:micro\], we introduce particle systems enlightening the selection principles observed for the macroscopic models. Finally in Section \[sec:Levy\] we export these observations to more general models. We consider general Lévy processes under suitable assumptions and analyze selection principles in this context. Macroscopic models {#sec:macro} ================== We elaborate on the two macroscopic models we study: front propagation and QSD. Front propagation in the KPP ---------------------------- Since the seminal papers [@Fisher; @KPP], equation has received a huge amount of attention for several reasons. Among them, it is one of the simplest models explaining several phenomena that are expected to be universal. For instance, it admits a continuum of traveling wave solutions that can be parametrized by their velocity $c$. More precisely, for each $c \in [\sqrt{2r}, +\infty)$ there exists a function $w_c \colon \R \to [0,1]$ such that $$v(t,x)=w_c(x-ct)$$ is a solution to . For $c< \sqrt{2r}$, there is no traveling wave solution, [@AW; @KPP]. Hence $c^*=\sqrt{2r}$ represents the minimal velocity and $w_{c^*}$ the minimal traveling wave. Moreover, if $v_0$ verifies for some $0<b < \sqrt{2r}$ $$\lim_{x\to \infty} e^{bx}(1-v_0(x)) = a >0,$$ then $$\label{domain.tw} \lim_{t \to \infty} v(t,x+ct)=w_c(x), \quad \mbox{for }c=r/b + \frac12 b,$$ see [@MK; @MK2]. If the initial measure has compact support (or fast enough decay at infinity), the solution converges to the minimal traveling wave and the domain of attraction and velocity of each traveling wave is determined by the tail of the initial distribution [@AW; @MK; @U]. A smooth traveling wave solution of that travels at velocity $c$ is a solution to $$\label{TWKPP} \frac12 w'' + cw + r(w^2 -w)= 0.$$ The behavior at infinity of these traveling waves is given by $$1- w_c(x) \sim \begin{cases} c_1e^{-bx} & c> \sqrt{2r}\\ c_2xe^{-x\sqrt{2r}} & c= \sqrt{2r}. \end{cases}$$ This behavior is determined by the linearization of at $w=1$, i.e. the solution of $$\label{TWKPPLin} \frac12 w'' + cw' + rw = 0.$$ See [@SH; @U]. We come back to this equation when dealing with quasi-stationary distributions. Quasi-stationary distributions {#qsd} ------------------------------ Quasi-stationary distributions have been extensively studied since the pioneering work of Kolmogorov (1938), Yaglom (1947) and Sevastyanov (1951) on the behavior of Galton-Watson processes. The beginning of this theory and an important part of the research in the area has been motivated by models on genetics and population biology, where the notion of quasi-stationarity is completely natural to describe the behavior of populations that are expected to get extinct, conditioned on the event that extinction has not yet occurred, on large time scales. Being more precise, consider a Markov process $Z=(Z_t,\, t\ge 0)$, killed at some state or region that we call $0$. The absorption time is defined by $\tau=\inf\{t>0 \colon Z_t \in 0\}$. The conditioned evolution at time $t$ is defined by $$\mu_t^\gamma(\cdot):= \P_\gamma(Z_t\in \cdot|\tau>t).$$ Here $\gamma$ denotes the initial distribution of the process. A probability measure $\nu$ is said to be a quasi-stationary distribution (QSD) if $\mu_t^\nu = \nu$ for all $t\ge 0$. For Markov chains in finite state spaces, the existence and uniqueness of QSDs as well as the convergence of the conditioned evolution to this unique QSD for every initial measure follows from Perron-Frobenius theory. The situation is more delicate for unbounded spaces as there can be $0$, $1$ or an infinite number of QSD. Among those distributions, the [*minimal*]{} QSD is the one that minimizes $\E_\nu(\tau)$. The [*Yaglom limit*]{} is a probability measure $\nu$ defined by $$\nu:= \lim_{t \to \infty} \mu_t^{\delta_x},$$ if it exists and does not depend on $x$. It is known that if the Yaglom limit exists, then it is a QSD. A general principle is that the Yaglom limit [*selects*]{} the minimal QSD, i.e. the Yaglom limit is the QSD with minimal mean absorption time. This fact has been proved for a wide class of processes that include birth and death process, Galton-Watson processes, random walks, Brownian motion, more general Lévy processes, etc. It can also be proved for $R$-positive processes by means of the theory of $R$-positive matrices [@SVJ]. To give a flavor of the results that hold in this situation, consider a discrete time Markov chain in $\N$ that it is absorbed at $0$. Denote $p=(p(i,j),\,\, i, j \in \N)$ its transition matrix so that $p$ is sub-stochastic. We use $p^{(n)}$ for the $n$-th power of $p$. We say that $p$ is $R$-positive if one (and hence both) of the following equivalent statements hold 1. For some $i$ and $j$, the sequence $R^np^{(n)}(i,j)$ tends to a finite non-zero limit as $n\to \infty$. 2. There exist non-negative, non-zero eigenvectors $\nu=(\nu(k))_{k\in\N}, \, \beta=(\beta(k))_{k\in\N}$ associated to the eigenvalue $1/R$ such that $\sum_{k=1}^\infty \nu(k) \beta(k) < \infty$. In 1966, Seneta and Vere-Jones proved the following theorem Assume that the matrix $p$ is $R$-positive, then the conditioned evolution converges to $\nu$ as $n\to \infty$ if one of the following conditions hold 1. The left eigenvector $\nu$ satisfies $\sum \nu(i) <\infty$ and the initial distribution $\mu$ is dominated (pointwise) by a multiple of $\nu$. 2. The right eigenvector $\beta$ is bounded away from zero and $\sum \mu(j)\beta(j)<\infty$ \[SenetaVere-Jones\] Observe that in both situations we have that $\nu$ is the minimal QSD and the Yaglom limit. Also every initial distribution with tail light enough is in the domain of attraction of $\nu$. So, in the $R$-positive case, the situation is pretty clear. These results can be applied for instance to the Galton-Watson process. In that case, a detailed study of the domain of attraction of the other QSDs (which are parametrized by an interval) is given in [@RVJ] where it can be seen that the limiting conditional distribution is given by the tail of the initial distribution. Unfortunately, on the one hand $R$-positivity is a property difficult to check and on the other hand, there is a lot of interesting processes that are not $R$-positive. For example, a birth and death process with constant drift towards the origin, has a continuum of QSDs and initial distributions with light tails are attracted by the minimal QSD, which can be computed explicitly [@FMP], but this process is not $R$-positive and hence Theorem \[SenetaVere-Jones\] does not apply. The presence of an infinite number of quasi-stationary distributions is something anomalous from the modeling point of view, in the sense that no physical nor biological meaning has been attributed to them. The reason for their presence here and in the front propagation context is similar: when studying for instance population or genes dynamics through the conditioned evolution of a Markov process, we are implicitly considering an infinite population and microscopic effects are lost. So, as Fisher suggests, in order to avoid the undesirable infinite number of QSD, we should take into account microscopic effects. A natural way to do this is by means of interacting particle systems. We discuss this in Section \[particle.systems\]. #### Brownian Motion with drift. Quasi-stationary ditributions for Brownian Motion with constant drift towards the origin are studied in [@MPSM; @MSM]. We briefly review here some of the results of these papers and refer to them for the details. For $c>0$ we consider a one-dimensional Brownian Motion $X=(X_t)_{t\ge0}$ with drift $-c$ defined by $X_t=B_t - ct$. Here $B_t$ is a one dimensional Wiener process defined in the standard Wiener space. We use $\P_x$ for the probability defined in this space such that $B_t$ is Brownian Motion started at $x$ and $\E_x$ for expectation respect to $\P_x$. Define the hitting time of zero, when the process is started at $x>0$ by $\tau_x(c)=\inf \{ t >0 \colon X_t=0\}$ and denote with $P_t^c$ the submarkovian semigroup defined by $$\label{qsd.semigroup} P_t^cf(x)=\E_x(f(X_t) \one_{\{\tau_x(c)>t\} }).$$ In this case, differentiating and after some manipulation it can be seen that the conditioned evolution $\mu P_t$ has a density $u(t,\cdot)$ for every $t>0$ and verifies $$\label{cond.ev.BM} \begin{array}{rcl} \displaystyle \frac{\partial u}{\partial t}(t,x) &= & \displaystyle\frac{1}{2} \frac{\partial^2 u}{\partial^2 x}(t,x) + c \frac{\partial u}{\partial x}(t,x) + \frac12\frac{\partial u}{\partial x}(t,0)u(t,x), \quad t>0, x>0,\\ u(t,0)&=& u(t,+\infty)=0, \quad t>0,\\ \end{array}$$ Recall now that a probability measure $\nu$ in $\R_+$ is a QSD if $$\P_\nu(X_t \in \cdot|X_t>0) = \nu(\cdot).$$ It is easy to check that if $\nu$ is a QSD, the hitting time of zero, started with $\nu$ is an exponential variable of parameter $r$ and hence $\nu$ is a QSD if and only if there exists $r>0$ such that $$\nu P_t^c = e^{-r t}\nu, \quad \mbox{for any} \quad t>0.$$ Differentiating and using the semigroup property we get that $\nu$ is a QSD if and only if $$\int (\frac12 f'' - cf')\,d\nu = -r \int f\, d\nu, \quad \mbox{for all } f \in C_0^\infty(\R_+).$$ Integrating by parts we get that the density $w$ of $\nu$ must verify $$\label{qsd.brownian} \frac12 w'' + cw' + r w = 0.$$ Solutions to this equation with initial condition $w(0)=0$ are given by $$w(x)= \begin{cases} me^{-c x} \sin(\sqrt{c^2- 2r}x)& r > \frac{c^2}2,\\ mx e^{- c x} & r = \frac{c^2}2,\\ me^{-c x} \sinh(\sqrt{c^2- 2r}x)& r < \frac{c^2}2.\\ \end{cases}$$ Observe that $w$ defines an integrable density function if and only if $0<r\le c^2/2$ (or equivalently, $c\ge \sqrt{2r}$). One can thus parametrize the set of QSDs by their eigenvalues $r$, $\{\nu_r \colon 0<r\le c^2/2\}$. For each $r$, the distribution function of $\nu_r$, $v(x)=\int_0^x w(y)\,dy$ is a monotone solution of with boundary conditions $$\label{qsd.brownian.bc} v(0)=0,\qquad v(+\infty)=1,$$ which is the same equation but in a different domain. The following theorem characterizes the domain of attraction of each QSD. Let $\gamma$ be a probability measure on $(0,+\infty)$ with density $\rho$. If $$\label{tail.qsd} \lim_{x \nearrow \infty} -\frac{1}{x} \log \rho(x) = b < c,$$ then $\lim_t \mu^\gamma_t = \nu_{r(b)}$, where $r(b)=cb - b^2/2$. Observe that this last equation is equivalent to $c=r/b + \frac12 b$, which should be compared with . Also remark that if holds then $$\lim_{x\nearrow \infty} -\frac{1}{x} \log \mu([x,+\infty) = b.$$ We come back to equation later, shedding light on the links between QSD and traveling waves. Finally let us mention that a similar result holds for the discrete-space analog, i.e. birth and death processes with constant drift towards the origin [@FMP2; @FMP] Particle systems {#sec:micro} ================ \[particle.systems\] In this section we introduce two particle systems. The first one is known as Branching Brownian Motion (BBM) with selection of the $N$ right-most particles ($N-$BBM). As a consequence of the link between BBM and F-KPP that we describe below, this process can be thought of as a microscopic version of F-KPP. The second one is called Fleming-Viot and was introduced by Burdzy, Ingemar, Holyst and March [@BHIM], in the context of Brownian Motion in a $d$-dimensional bounded domain. It is a slight variation of the original one introduced by Fleming and Viot [@FV]. The first interpretation of this process as a microscopic version of a conditioned evolution is due to Ferrari and Maric [@FM]. BBM and F-KPP equation ---------------------- One-dimensional supercritical Branching Brownian Motion is a well-understood object. Particles diffuse following standard Brownian Motion started at the origin and branch at rate 1 according to an offspring distribution that we assume for simplicity to be $\delta_2$. When a particle branches, it has two children and then dies. As already underlined, its connection with the F-KPP equation and traveling waves was pointed out by McKean in the seminal paper [@MK]. Denote with $N_t$ the number of particles alive at time $t\ge 0$ and $\xi_t(1) \le \dots \le \xi_t(N_t)$ the position of the particles enumerated from left to right. McKean’s representation formula states that if $0\le v_0(x) \le 1$ and we start the process with one particle at $0$ (i.e. $N(0)=1$, $\xi_0(1)=0$), then $$v(t,x):= \E\left(\prod_{i=1}^{N_t}v_0(\xi_t(i)+x)\right)$$ is the solution of . Of special interest is the case where the initial condition is the Heaviside function $v_0 = \one\{[0,+\infty)\}$ since in this case $$v(t,x)=\P(\xi_t(1)+x>0)=\P( \xi_t(N_t) <x).$$ This identity as well as various martingales obtained as functionals of this process have been widely exploited to obtain the precise behavior of solutions of , using analytic as well as probabilistic tools [@B; @B2; @HHK; @SH; @MK; @U]. $N-$BBM and Durrett-Remenik equation ------------------------------------ Consider now a variant of BBM where the $N$ right-most particles are selected. In other words, each time a particle branches, the left-most one is killed, keeping the total number of particles constant. This process was introduced by Brunet and Derrida [@BD2; @BD] as part of a family of models of branching-selection particle systems to study the effect of microscopic noise in front propagation. By means of numerical simulations and heuristic arguments, they conjectured that the linear speed of $N-$BBM differs from the speed of standard BBM by $(\log N)^{-2}$ and in a series of papers with coauthors they study various statistics of the process [@BDMM; @BDMM2; @BD2; @BD]. Bérard and Gouéré proved this shift in the velocity for a similar process. We refer to the work of Maillard [@M2] for the most detailed study of this process. Durret and Remenik [@DR] considered a slightly different process in the class of Brunet and Derrida: $N-$BRW. The system starts with $N$ particles. Each particle gives rise to a child at rate one. The position of the child of a particle at $x\in\R$ is $x+y$, where $y$ is chosen according to a probability distribution with density $\rho$, which is assumed symmetric and with finite expectation. After each birth, the $N+1$ particles are sorted and the left-most one is deleted, in order to keep always $N$ particles. They prove that the empirical measure of this system converges to a deterministic probability measure $\nu_t$ for every $t$, which is absolutely continuous with density $u(t,\cdot)$, a solution of the following free-boundary problem $$\label{Durrett-Remenik} \begin{array}{rcl} \mbox{{\em Find }}(\gamma,u) \mbox{{\em such that}}& \\ \\ \displaystyle \frac{\partial u}{\partial t}(t,x) & = & \displaystyle \int_{-\infty}^\infty u(t,y)\rho(x - y) \, dy \quad \forall x > \gamma(t),\\ \displaystyle \int_{\gamma(t)} ^\infty u(t,y) \, dy & = &1, \quad u(t,x)=0, \quad \forall x \le \gamma(t),\\ u(0,x) & = &u_0(x).\\ \end{array}$$ They also find all the traveling wave solutions for this equation. Just as for the BBM, there exists a minimal velocity $c^*\in\R$ such that for $c\ge c^*$ there is a unique traveling wave solution with speed $c$ and no traveling wave solution with speed $c$ for $c < c^*$. The value $c^*$ and the behavior at infinity of the traveling waves can be computed explicitly in terms of the Laplace transform of the random walk. In Section \[sec:Levy\] we show that these traveling waves correspond to QSDs of drifted random walks. It follows from renewal arguments that for each $N$, the process seen from the left-most particle is ergodic, which in turn implies the existence of a velocity $v_N$ at which the empirical measure travels for each $N$. Durrett and Remenik prove that these velocities are increasing and converge to $c^*$ as $N$ goes to infinity. We can interpret this fact as a [*weak selection principle*]{}: the microscopic system has a unique velocity for each $N$ (as opposed to the limiting equation) and the velocities converge to the minimal velocity of the macroscopic equation. The word “weak” here refers to the fact that only convergence of the velocities is proved, but not convergence of the empirical measures in equilibrium. In view of these results, the same theorem is expected to hold for a $N-$BBM that branches at rate $r$. In this case the limiting equation is conjectured to be given by $$\label{Durrett-Remenik.BM} \begin{array}{rcl} \mbox{{\em Find }}(\gamma,u) \mbox{{\em such that}}& \\ \\ \displaystyle \frac{\partial u}{\partial t}(t,x) & = & \displaystyle \frac12\frac{\partial^2u}{\partial^2x}(t,x) + r u(t,x) \quad \forall x > \gamma(t),\\ \displaystyle \int_{\gamma(t)} ^\infty u(t,y) \, dy & = &1, \quad u(t,x)=0, \quad \forall x \le \gamma(t),\\ u(0,x) & = &u_0(x).\\ \end{array}$$ The empirical measures in equilibrium are also expected to converge to the minimal traveling wave. More precisely, Both N-BBM and N-BRW are ergodic, with (unique) invariant measure $\lambda^N$ and the empirical measure distributed according to $\lambda^N$ converges to the delta measure supported on the minimal quasi-stationary distribution. #### Traveling waves. Let us look at the traveling wave solutions $u(t,x)=w(x-ct)$ of . Plugging-in in we see that they must verify $$\label{TWDR} \frac12 w'' + cw' + rw = 0, \quad w(0)=0, \quad \int_0^\infty w(y)\, dy=1.$$ Which is exactly $\eqref{qsd.brownian}$. Note nevertheless that in the parameter $r$ is part of the data of the problem (the branching rate) and $c$ is part of the unknown (the velocity), while in the situation is reversed: $c$ is data (the drift) and $r$ unknown (the absorption rate under the QSD). However, we have the following relation $$\begin{array}{ccc} \begin{array}{c} c \mbox{ is a minimal velocity for } r\\ \mbox{in \eqref{TWDR}} \end{array} & \iff \begin{array}{c} r \mbox{ is a maximal absorption rate for } c\\ \mbox{in \eqref{qsd.brownian}} \end{array} \end{array}$$ Observe also that $1/r$ is the mean absorption time for the QSD associated to $r$ and hence, if $r$ is maximal, the associated QSD is minimal. So the minimal QSD for Brownian Motion in $\R_+$ and the minimal velocity traveling wave of are one and the same. They are given by $$u_{c^*(r)}(x)=u_{r^*(c)}= 2r^*xe^{-\sqrt{2r^*}} = (c^*)^2 x e^{- c^* x},$$ which is the one with fastest decay at infinity. Again, the distribution function $v$ of $u$ is a monotone solution to the same problem but with boundary conditions given by $v(0)=0$, $v(+\infty)=1$. It is worth noting that although the solutions to and are not the same since they are defined in different domains, there is a natural way to identify them (and also with solutions of ). Given positive constants $c$ and $r$, there is a solution $w$ of if and only if there is solution $\tilde w$ of . Moreover, we have $$\lim_{x\to \infty} \frac{w(x)}{\tilde w(x)}=1.$$ In this case, the proof of this statement is immediate since solutions of and are explicit and the relation among solutions of and is very well understood [@SH]. We conjecture that the same situation holds in much more generality. Fleming-Viot and QSD -------------------- The Fleming-Viot process can be thought of as a microscopic version of conditioned evolutions. Its dynamics are built with a continuous time Markov process $Z=(Z_t, \, t\ge0)$ taking values in the metric space $\Lambda \cup \{0\}$, that we call the [*driving process*]{}. We assume that $0$ is absorbing in the sense that $$\P(Z_t =0|Z_0 = 0) = 1, \qquad \forall t\ge 0.$$ We use $\tau$ for the absorption time $$\tau=\inf\{t>0 \colon Z_t \notin \Lambda \}.$$ As, before, we use $P_t$ for the submarkovian semigroup defined by $$P_t f(x) = \E_x(f(Z_t)\one\{\tau>t\}).$$ For a given $N\ge 2$, the Fleming-Viot process is an interacting particle system with $N$ particles. We use $\xi_t=(\xi_t(1),\dots,\xi_t(N)) \in \Lambda^N$ to denote the state of the process, $\xi_t(i)$ denotes the position of particle $i$ at time $t$. Each particle evolves according to $Z$ and independently of the others unless it hits $0$, at which time, it chooses one of the $N-1$ particles in $\Lambda$ uniformly and takes its position. The guenine definition of this process is not obvious and in fact is not true in general. It can be easily constructed for processes with bounded jumps to $0$, but is much more delicate for diffusions in bounded domains [@BBF; @GK] and it does not hold for diffusions with a strong drift close to the boundary of $\Lambda$, [@BBP]. Here we are also interested in the empirical measure of the process $$\label{empirical.measure} \mu^N_t = \frac1 N \sum_{i=1}^N \delta_{\xi_t(i)}.$$ Its evolution is mimicking the conditioned evolution: the mass lost from $\Lambda$, is redistributed in $\Lambda$ proportionally to the mass at each state. Hence, as $N$ goes to infinity, we expect to have a deterministic limit given by the conditioned evolution of the driving process $Z$, i.e. $$\mu^N_t(A) \to \P(Z_t \in A | \tau >t) \qquad (N\to\infty).$$ This is proved in $\cite{V}$ by the Martingale method in great generality. See also [@GJ] for a proof based on sub and super-solutions and correlations inequalities. A much more subtle question is the ergodicity of the process for fixed $N$ and the behavior of these invariant measures as $N\to \infty$. As a general principle it is expected that If the driving process $Z$ has a Yaglom limit $\nu$, then the Fleming-Viot process driven by $Z$ is ergodic, with (unique) invariant measure $\lambda^N$ and the empirical measures distributed according to $\lambda^N$ converge to $\nu$. We refer to [@GJ] for an extended discussion on this issue. This conjecture has been proved for subcritical Galton-Watson processes, where a continuum of QSDs arises [@AFGJ]. We have again here a [*microscopic selection principle*]{}: whereas there exists an infinite number of QSDs, when microscopic effects are taken into account (through the dynamics of the Fleming-Viot process), there is a unique stationary distribution for the empirical measure, which selects asymptotically the minimal QSD of the macroscopic model. When the driving process is a one dimensional Brownian motion with drift $-c$ towards the origin as in Section \[qsd\], the proof of the whole picture remains open, but the ergodicity of FV for fixed $N$ has been recently proved [@AT; @AK]. So, from [@V Theorem 2.1] we have that for every $t>0$, $\mu^N_t$ converges as $N\to \infty$ to a measure $\mu_t$ with density $u(t,\cdot)$ satisfying (\[cond.ev.BM\]). The open problem is to prove a similar statement in equilibrium. Observe that $u$ is a stationary solution of if and only if it solves for some $r>0$. Hence, although equations and are pretty different, stationary solutions to coincide with traveling waves of . Summing up ---------- 1. The link between $N-$BBM and Fleming Viot, in the Brownian Motion case is clear. Both processes evolve according to $N$ independent Brownian Motions and branch into two particles. At branching times, the left-most particle is eliminated (selection) to keep the population size constant. The difference is that while $N-$BBM branches at a constant rate $N r$, Fleming-Viot branches each time a particle hits 0. This explains why in the limiting equation for $N-$BBM the branching rate is data and the velocity is determined by the system while in the hydrodynamic equation for Fleming-Viot the velocity is data and the branching rate is determined by the system. 2. The empirical measure of $N-$BBM is expected to converge in finite time intervals to the solution of . This is supported by the results of [@DR] where the same result is proved for random walks. 3. The empirical measure of Fleming-Viot driven by Brownian Motion converges in finite time intervals to the solution of . 4. Both $N$-BBM seen from the left-most particle and FV are ergodic and their empirical measure in equilibrium is expected to converge to the deterministic measure given by the minimal solution of . Note though that while for $N-$BBM $r$ is data and minimality refers to $c$, for Fleming-Viot $c$ is data and minimality refers to $1/r$ (microscopic selection principle). 5. $u(t,x)=w(x-ct)$ is a traveling wave solution of if and only if $w$ is the density of a QSD for Brownian Motion with drift $-c$ and eigenvalue $-r$. 6. $c$ is minimal for $r$ (in ) if and only if $1/r$ is minimal for $c$. So, we can talk of a “minimal solution of ”, which is both a minimal QSD and a minimal velocity traveling wave. 7. The microscopic selection principle is conjectured to hold in both cases, with the same limit, but a proof is still unavailable. Traveling waves and QSD for Lévy processes {#sec:Levy} ========================================== Let $Z=(Z_t, \, t\ge 0)$ be a Lévy process with values in $\R$, defined on a filtered space $(\Omega, \F, (\F_t),\P)$ and Laplace exponent $\psi: \mathbb R \to \mathbb R $ defined by $$\E(e^{\theta Z_t})= e^{\psi(\theta ) t}$$ such that $$\psi(\theta)= b\theta+ \sigma^2 \frac{\theta^2}{2} + g(\theta),$$ where $b\in\R$, $\sigma>0$ (which ensures that $Z$ is non-lattice) and $g$ is defined in terms of the jump measure $\Pi$ supported in $\R\setminus\{0\}$ by $$g(\theta)=\int_x (e^{\theta x}- 1 -\theta x \one_{\{|x|<1\}}) \Pi(dx), \qquad \int_\R (1 \wedge x^2) \Pi(dx) < \infty.$$ Let $\theta^\star=\sup\{\theta\colon |\psi(\theta)|<\infty\}$ and recall that $\psi$ is strictly convex on $(0,\theta^\star)$ and by monotonicity $\psi(\theta^\star)=\psi(\theta^\star-)$ and $\psi'(\theta^\star)=\psi'(\theta^\star-)$ are well defined as well as the right derivative at zero $\psi'(0)=\psi'(0+)=\E(Z_1)$, that we assume to be zero. We also assume that $\theta^\star>0$. In this case we can relate $\psi$ to the characteristic function $\Psi(\lambda)=-\log\E(e^{{\rm i}\lambda Z_1})$ by $\psi(\theta)=-\Psi(-{\rm i}\theta)$ for $0\le \theta < \theta^\star$. This centered Lévy process plays the role of Brownian Motion in the previous sections. The generator of $Z$ applied to a function $f\in C_0^2$, the class of compactly supported functions with continuous second derivatives, gives $$\L f(x)=\frac12 \sigma^2 f''(x) + bf'(x) + \int_\R (f(x+y) - f(x) - yf'(x)\one{\{|y|\le 1\}} ) \Pi(dy).$$ The adjoint of $\L$ is also well defined in $C_0^2$ and has the form $$\L^*f(x)=\frac12 \sigma^2 f''(x) - bf'(x) + \int_\R (f(x-y) - f(x) + yf'(x)\one{\{|y|\le 1\}} ) \Pi(dy).$$ Now, for $c>0$ we consider the [*drifted process*]{} $Z^c$ given by $$Z^c_t = Z_t - ct$$ It is immediate to see that the Laplace exponent of $Z^c$ is given by $\psi_c(\theta)=\psi(\theta) - c \theta$ for $\theta \in [0,\theta^\star]$, that $C_0^2$ is contained in the domain of the generator $\L_c$ of $Z^c$, and that $\L_c f= \L f - cf'$. Recall that the forward Kolmogorov equation for $Z$ is given by $${d \over dt} E^x(f(Z_t))= \L f(x),$$ while the forward Kolmogorov (or Fokker-Plank) equation for the density $u$ (which exists since $\sigma>0$) is given by $${d \over dt} u(t,x)= \L^*u (t,\cdot)(x).$$ As in the Brownian case, we consider - A branching Lévy process (BLP) $(N_t, (\xi_t(1),\dots, \xi_t(N_t)))$ driven by $\L$. - A branching Lévy process with selection of the $N$ rightmost particles ($N-$BLP), also driven by $\L$. - A Fleming-Viot process driven by $\L_c$ (FV). We focus on the last two processes. For a detailed account on BLP, we refer to [@Ky]. Let us just mention that the KPP equation can be generalized in this context to: $$\label{KPP.L} \begin{array}{l} \displaystyle \frac{\partial v}{\partial t}= \L v + r f(v), \quad x \in \R,\, t>0,\\ \\ \displaystyle v(0,x)= \displaystyle v_0(x), \quad x \in \R. \end{array}$$ A characterization of the traveling waves as well as sufficient conditions of existence are then provided in [@Ky]. For $N-$BLP we expect (but a proof is lacking) that the empirical measure converges to a deterministic measure whose density is the solution of the generalized Durrett-Remenik equation $$\label{DR.Levy} \begin{array}{rcl} \mbox{{\em Find }}(\gamma,u)\,\, \mbox{{\em such that}}& \\ \\ \displaystyle \frac{\partial u}{\partial t}(t,x) &=& \displaystyle \L^*u(t,x) + r u(t,x), \quad x>\gamma(t),\\ \displaystyle \int_{\gamma(t)}^\infty u(t,y)\, dy &=& 1,\quad u(t,x)=0, \quad x\le \gamma(t),\\ u(0,x)&=&u_0(x), \qquad x\ge 0. \end{array}$$ Existence and uniqueness of solutions to this problem have to be examined, but we believe that the proof presented in [@DR] can be extended to this context. We show below the existence of traveling wave solutions for this equation under mild conditions on $\L$ based on the existence of QSDs. Concerning FV, it is known [@V] that the empirical measure converges to the deterministic process given by the conditioned evolution of the process, which has a density for all times and verifies $$\label{cond.ev.Levy} \begin{array}{rcl} \displaystyle \frac{\partial u}{\partial t}(t,x) &= & \displaystyle\L^* u(t,x) + c \frac{\partial u}{\partial x}(t,x) - u(t,x)\int_\R\L^*u(t,y,)dy \quad t>0, x>0,\\ u(t,0)&=& u(t,+\infty)=0, \quad t>0,\\ \end{array}$$ Traveling waves vs. QSDs ------------------------ We now establish the link between traveling waves and QSD. \[qsd.iff.tw\] The following statements are equivalent: - The probability measure $\nu$ with density $w$ is a QSD for $Z^c$ with eigenvalue $-r$ , - $u(t,x)=w(x-ct)$ is a traveling wave solution with speed $c$ for the free-boundary problem , with parameter $r$. Denote $\langle f,g \rangle= \int f(x)g(x) dx$. A QSD $\nu$ for $Z^c$ with eigenvalue $-r$ is a solution of the equation $$\langle \nu \L_c + r \nu , f \rangle= 0, \forall f \in C_0^2.$$ Using that $ \L_c^* f= \L^* f + c f' $ and writing that $\nu$ has density $w$, we obtain that $$\L^* w + c w' + r w=0,$$ which in turn is clearly equivalent to $w$ being a traveling wave solution with speed $c$ for . Minimal QSD and minimal velocities ---------------------------------- We now study the relation between minimal traveling waves and minimal QSD. Using known results on QSDs for Lévy processes, we can describe the traveling waves for the corresponding set of equations. Since we rely on results of Kyprianou and Palmowski [@KP1], we assume that $Z$ is non-lattice and consider the following two classes. The Lévy process $Z$ belongs to class [C1]{} if there exists $0<\theta_0<\theta^\star$ such that $\psi'(\theta_0)>0$ and the process $(Z,\P^{\theta_0})$ is in the domain of attraction of a stable law with index $1 < \alpha \le 2$, where the probability $\P^{\theta}$ is given by $$\left. \frac{\d\P^\theta_z}{\d\P_z}\right|_{\F_t} = e^{\theta (Z_t - z) - \psi(\theta)t}.$$ The process $Z$ is in class [C2]{} if $- \infty< \psi'(\theta^\star)<0$, and the function $x \to \Pi_{\theta^\star} \big([x,\infty) \big)$ is regularly varying at infinity with index $-\beta < -2$, where $\Pi_{\theta}(dx)= e^{\theta x} \Pi(dx)$. Recall that we are also assuming $\psi'(0)=0$ and hence there exists a critical $c^*$, possibly infinity, such that for $c\le c^*$, $\psi_c$ attains its negative infimum at a point $\theta_c\le \theta^\star$ with $\psi_c'(\theta_c)=0$ and for $c>c^*$, the negative infimum is attained at $\theta^\star$. In this case we write $\theta_c=\theta^\star$. Since $\psi_c(\theta)= \psi(\theta) -c \theta$, observe that $e^{\theta (Z^c_t - z) - \psi_c(\theta)t} = e^{\theta (Z_t - z) - \psi(\theta)t}$. Hence, if $Z$ is in class [C1]{}, then for $c\le c^*$, $Z^c$ is in class A in the sense of Bertoin and Doney [@BD; @KP1]. Similarly if $Z$ is in class [C2]{} and $c>c^*$, then $Z^c$ is in class B in the sense of Bertoin and Doney. The union of class A and B represents a very large family of Lévy processes including for instance Brownian motion and spectrally negative (positive) processes, as well as many others. The following theorem is proved in [@KP1]. \[teo.KP\] Assume $c\le c^*$ and $Z$ is in class [C1]{} or $c>c^*$ and $Z$ is in class [C2]{}. Then the Yaglom limit of $Z^c$ exists and is given by $$\nu(dx)=\theta_0 \kappa_{\theta_c}(0,\theta_c) e^{-\theta_c x}V_{\theta_c}(x)\, dx.$$ Here $\kappa_\theta$ and $V_\theta$ are respectively the Laplace exponent of the ascending ladder process and the renewal function of the ladder heights process corresponding to $(Z^c,\P^\theta)$. The proof is based on a careful control of the asymptotics of the process as $t \to \infty$ that in particular yields $$\label{behavior.hitting.time} \P_x(\tau >t) \sim H(x,\theta_c)\ell(t)e^{\psi_c(\theta_c)t}.$$ Here $H$ is a function that depends on the characteristic exponent of the process and $\ell$ is a function regularly varying at infinity. This allows us to prove the following The probability measure $\nu$ defined in Theorem \[teo.KP\] is the minimal QSD. Recall that under a QSD, the hitting time of zero is exponentially distributed. Since $\nu$ is the Yaglom limit, for any bounded function $f$ $$\int f d \nu = \lim_{t \to \infty} \frac{\E_x(f(X_t), \tau >t)}{\P_x(\tau>t)}.$$ Applying this to $f(y)= \P_y(\tau>s)$ and using Corollary 4 in [@KP1], one obtains that $$\P_\nu(\tau>s) = \lim_{t \to \infty} \frac{\P_x(\tau >t+s)}{\P_x(\tau>t)} = \exp(\psi_c(\theta_c) s).$$ Hence the parameter of $\nu$ equals $-\psi_c(\theta_c)$. If $\nu$ is not minimal, there exists another QSD $\tilde \nu$, with parameter $\tilde r >- \psi_c(\theta_c)$. Then $\E_{\tilde \nu}(e^{-\psi_c(\theta_c) \tau}) < \infty$ and as a consequence there exists $x >0$ such that $\E_x(e^{-\psi_c (\theta_c) \tau}) < \infty,$ but this contradicts . Let $\C$ be the subset of $\mathbb R^2$ such that $(c,r) \in \C$ if and only if there exists a QSD $\nu$ for $Z^c$ with eigenvalue $-r$. Proposition \[qsd.iff.tw\] states that this set coincides with the set of pairs $(c,r)$ such that there exists a traveling wave for with velocity $c$. We say that $r$ is maximal for $c$ if $r=\max \{r' \colon (c,r') \in \C\}$. In the same way, $c$ is minimal for $r$ if $c=\min \{c' \colon (c',r) \in \C\}$. Under the same hypotheses of Theorem \[teo.KP\] we have that $c$ is minimal for $r$ if and only if $r$ is maximal for $c$. For each $c>0$, the maximal absorption rate is given by $r(c)=-\psi_c(\theta_c)$. The proof follows by observing that the function $c \mapsto \psi_c(\theta_c)$ is strictly decreasing and continuous. Conclusions and general conjectures =================================== We emphasized the direct links between quasi stationary distributions QSDs for space invariant Markov processes in $\R$ and traveling waves of Durrett-Remenik equation, that are closely related to the generalized F-KPP equation. We proved also that minimal QSDs correspond to minimal velocity traveling waves. As a general fact both macroscopic and microscopic selection principles are expected to hold for QSDs and for traveling waves. They have been proved for a series of models. This suggests that the selection principles in front propagation and in QSDs are one and the same. For random walks in $\R$, the microscopic selection principle is an open problem in both cases (traveling waves and QSD). [**Acknowledgments.**]{} We would like to thank UBACyT 20020090100208, ANPCyT PICT No. 2008-0315, CONICET PIP 2010-0142 and 2009-0613 and MATHAMSUD’s project “Stochastic structure of large interacting systems" for financial support. [^1]: Departamento de Matemática, Fac. Cs. Exactas y Naturales, Universidad de Buenos Aires and IMAS-CONICET. [pgroisma@dm.uba.ar]{}, [http://mate.dm.uba.ar/$\sim$pgroisma.]{} [^2]: IMAS-CONICET. [mjonckhe@dm.uba.ar]{}, [http://matthieujonckheere.blogspot.com]{}.

ArXiv

--- abstract: 'The dynamics of a mosquito population depends heavily on climatic variables such as temperature and precipitation. Since climate change models predict that global warming will impact on the frequency and intensity of rainfall, it is important to understand how these variables affect the mosquito populations. We present a model of the dynamics of a [*Culex quinquefasciatus*]{} mosquito population that incorporates the effect of rainfall and use it to study the influence of the number of rainy days and the mean monthly precipitation on the maximum yearly abundance of mosquitoes $M_{max}$. Additionally, using a fracturing process, we investigate the influence of the variability in daily rainfall on $M_{max}$. We find that, given a constant value of monthly precipitation, there is an optimum number of rainy days for which $M_{max}$ is a maximum. On the other hand, we show that increasing daily rainfall variability reduces the dependence of $M_{max}$ on the number of rainy days, leading also to a higher abundance of mosquitoes for the case of low mean monthly precipitation. Finally, we explore the effect of the rainfall in the months preceding the wettest season, and we obtain that a regimen with high precipitations throughout the year and a higher variability tends to advance slightly the time at which the peak mosquito abundance occurs, but could significantly change the total mosquito abundance in a year.' address: - 'Facultad de Matemática, Astronomía, Física y Computación, Universidad Nacional de Córdoba, Instituto de Física Enrique Gaviola, CONICET, Ciudad Universitaria, 5000 Córdoba, Argentina' - 'Instituto de Investigaciones Biológicas y Tecnológicas–CONICET–Universidad Nacional de Córdoba, Córdoba, Argentina' - 'Laboratorio de Arbovirus–Instituto de Virología “Dr. J. M. Vanella”–Facultad de Ciencias Médicas–Universidad Nacional de Córdoba, Córdoba, Argentina' author: - 'L.D. Valdez' - 'G.J. Sibona' - 'L.A. Diaz' - 'M.S. Contigiani' - 'C.A. Condat' bibliography: - 'bib.bib' title: 'Effects of rainfall on [*Culex*]{} mosquito population dynamics' --- Culex ,rainfall ,arbovirus ,mosquito abundance ,mathematical modeling Introduction ============ Mosquito-transmitted flaviviruses are an increasing health threat. In particular, members of [*Culex*]{} species (Diptera: Culicidae) such as [*Cx. quinquefasciatus*]{} and [*Cx. interfor*]{} are responsible for transmitting the West Nile and St. Louis encephalitis (SLEV) viruses to humans and domestic animals ([@beltran2015evidencia; @diaz2008eared; @diaz2016activity; @gubler2002global; @gubler2007flaviviruses; @lumsden1958st]). For instance, the SLEV is endemic in Argentina, where the principal vector is postulated to be [ *Cx. quinquefasciatus*]{} ([@diaz2013transmission]). In the last decades, mosquito-borne diseases have emerged and re-emerged as a result of multiple factors such as increasing urbanization, international travel, and climate change ([@harrigan2014continental; @kilpatrick2011globalization]). The development of mathematical models is essential to quantify the effect of each of these factors on the dynamics of the mosquito population, and to determine the most effective strategies to control the epidemic outbreaks transmitted by mosquito vectors ([@And_01; @ewing2016modelling; @lord2001simulation; @lord2001simulation0; @marini2016role]). Multiple studies have shown that the life cycle of the [ *Cx. quinquefasciatus*]{} is closely related to temperature ([@almiron1996winter; @ciota2014effect; @gunay2011narrow; @loetti2011development; @strickman1988rate]). [@strickman1988rate] demonstrated that its reproductive activity increases with temperature, and [@almiron1996winter] showed that this species can only live in environments with a temperature above 10$^{\circ}$C. Other studies have found that temperature has a strong influence on the development and survival of both adult and immature mosquitoes ([@ciota2014effect; @gunay2011narrow; @loetti2011development]). In turn, it was observed that [ *Cx. quinquefasciatus*]{} does not enter diapause, but it may undergo quiescence or remain gonoactive in protected (indoor or underground) habitats ([@almiron1996winter; @nelms2013overwintering]). Urbanization therefore helps [*quinquefasciatus*]{} populations survive the mild winters of temperate regions. Similarly, rainfall is an important climatological variable to predict the abundance of [*Culex*]{} mosquitoes, since its copiousness and distribution determine the production and size of mosquito breeding sites. [@reisen2008impact] studied the changes in the [*Cx. tarsalis*]{} population in California and found that, in most regions, it is positively correlated with an increase in total precipitation. However, these authors also found that in some places of the driest region of California, the correlation between these two variables was negative. On the other hand, [@olson1983correlation] showed that a very large rainfall is not always accompanied by proportionately large increases in the abundance of [ *Cx. tritaeniorhynchus*]{} and [*Cx. gelidus*]{}. In consequence, these results indicate that there exists a nonlinear relationship between rainfall and [*Culex*]{} abundance, which should be modeled in order to predict mosquito abundance. Additionally, since the climatological projections suggest that global warming will alter the frequency and intensity of rainfall, it is crucial to understand how different rainfall patterns will affect mosquito populations. In this paper we develop a dynamic model of the [ *Cx. quinquefasciatus*]{} population, adapting a fracturing procedure ([@finley2014exploring]) to describe the rainfall distribution. We use a system of compartmental ordinary differential equations that describe the immature and adult mosquito populations, in which we introduce the influence of temperature and rainfall on the reproduction rate. To study the influence of different rainfall patterns, we use a synthetic time series of rainfall based on the amount of rainfall per month and the monthly number of rainy days. We find that, for a given constant value of monthly precipitation, there is an optimum number of rainy days for which the maximum $M_{max}$ in the mosquito population is highest. On the other hand, we also study the variability of daily rainfall intensity through a fracturing process ([@finley2014exploring]), which allows us to study homogeneous and heterogeneous rainfall regimes, including those characterized by heavy rain events. We show that increasing daily rainfall variability reduces the dependence of $M_{max}$ on the number of rainy days, leading also to a higher abundance of mosquitoes for the case of low mean monthly precipitation. Finally, we explore the effect of different winter precipitation regimes on the mosquito abundance in the summer season, obtaining that a higher variability tends to advance slightly the peak time of mosquito abundance. Interestingly, we predict that the accumulated abundance of mosquitoes will decrease in a regime with high variability in the rainfall intensity. The boundary of the [*Cx. quinquefasciatus*]{} habitat in South America runs across central Argentina. This region is thus expected to exhibit intense changes in mosquito populations due to the undergoing climatic change; this is likely to have a strong impact on flavivirus prevalence. For this reason, we use the climatic data for the city of Córdoba to calibrate our model in the period 2008-2009. The paper is organized as follows: in Sec. \[Sec.modd\] we present the model of the dynamics of mosquito population and in Secs.\[SecConst\] and \[SecVarRai\] we explain the methods for generating two different synthetic time series. Then in Sec. \[secResul\] we show our results and finally we present our conclusions in Sec. \[sec.Discussion\]. Methods ======= The model of mosquito abundance {#Sec.modd} ------------------------------- In this section, we construct a compartmental, ordinary differential equation model for the mosquito abundance. We consider that the total vector population is stage-structured with an immature class consisting of all aquatic stages, and a mature or adult class . We assume that these population groups are restricted only to female mosquitoes as the reproductive sex. In our model, the total birth rate, $i.e.$ the total number of new immature female mosquitoes per unit of time, is proportional to the number of adult mosquito females and to $\beta_L\lambda(t) \theta(t)$, where $\beta_L$ corresponds to the reproduction rate in optimal conditions of temperature and water availability, and $\lambda(t)$ and $\theta(t)$ are normalized factors describing the influence of rainfall and temperature on the total birth rate, respectively. In Sec. \[Sec.lambThe\] we will explain how we construct these factors. In addition, we assume that the total birth rate is also regulated by a carrying capacity effect that depends on the occupation of the available immature habitats. Therefore we propose that the immature population growth is logistic-like with a carrying capacity $K_L$. Additionally, immature individuals either go to the mature class with rate $m_L$ or die at a rate $\mu_L$. We stress that the ratio $1/m_L$ gives the average development time from immature mosquito to adult. Finally, adult mosquitoes die with a mortality rate $\mu_M$. For simplicity, we assume that $m_L$, $\mu_L$ and $\mu_M$ do not depend on temperature or rainfall. With these definitions, we propose the following dynamic mass-balance equations for the abundance of immature mosquitoes, $L(t)$, and adult mosquitoes, $M(t)$, $$\begin{aligned} L(t+\Delta t)&=& L(t)+\Delta t \left[ \beta_L \theta(t) \lambda(t) M(t)\left(1-\frac{L(t)}{K_L}\right)-m_L L(t)-\mu_L L(t)\right],\label{eq.larv}\end{aligned}$$ and $$\begin{aligned} M(t+\Delta t)&=& M(t)+\Delta t \left[ m_L L(t)-\mu_M M(t)\right],\label{eq.adult}\end{aligned}$$ where $\Delta t$ is the time step size. Here we use $\Delta t=0.1$ \[days\]. Equation (\[eq.larv\]) can be easily derived by assuming that the fraction of the carrying capacity corresponding to female immature mosquitoes is the same as the fraction of females in the immature population. Table \[tab.Trans\] summarizes the different state variables and parameters used in this paper. Quantity Definition Value Refs. ----------- -------------------------------------------------------------------------------------------------------------------------------------------- ------- --------------------------- $L$ number of immature female mosquitoes — $M$ number of adult female mosquitoes — $\beta_L$ birth rate of immature female mosquito per female adult mosquito in optimal conditions of temperature and water availability (days$^{-1}$) 13.5 fitted (see \[AppCalib\]) $\theta$ effect of the temperature on the birth rate of mosquitoes — $\lambda$ effect of the water availability on the birth rate of mosquitoes — $m_L$ rate at which immature mosquitoes develop into adults (days$^{-1}$) 0.098 [@loetti2011development] $\mu_L$ immature mortality rate (days$^{-1}$) 0.03 [@loetti2011development] $\mu_M$ mosquito mortality rate (days$^{-1}$) 0.078 [@david2012bionomics] $K_L$ carrying capacity of the immature female population 16.2 fitted (see \[AppCalib\]) $H_{max}$ maximum daily amount of accumulated rainwater \[mm\] 9.86 fitted (see \[AppCalib\]) $H_{min}$ minimum daily amount of accumulated rainwater \[mm\] 0.067 fitted (see \[AppCalib\]) $H$ accumulated amount of rainwater \[mm\] — $R$ daily rainfall \[mm\] — $E$ daily evapotranspiration \[mm\] — $T$ average daily temperature \[$^{\circ}$C\] — $Hum$ daily relative humidity — $P_{max}$ total rainfall in the wettest month \[mm\] — $P_{min}$ total rainfall in the driest month \[mm\] — $D_{max}$ total number of rainy days in the wettest month \[days\] — $D_{min}$ total number of rainy days in the driest month \[days\] — : The variables and parameters for Eqs. (\[eq.larv\])-(\[eq.ArtD\]).[]{data-label="tab.Trans"} ### Effect of temperature and rainfall on the total birth rate {#Sec.lambThe} We add the effect of the temperature through a temperature factor $\theta(t)$. Several studies have shown that [*Cx. quinquefasciatus*]{} can breed only at temperatures above 10$^{\circ}$C ([@almiron1996winter; @ribeiro2004thermal]) and that the number of egg rafts collected per day is closely correlated with temperature ([@strickman1988rate]). Therefore we propose that the factor $\theta$(t) is a piecewise linear function, $$\begin{aligned} \label{cpattern} \theta(t)=\left\{% \begin{array}{ll} \frac{T(t)-T_{A}}{T_{B}-T_{A}} &\;\;\;\; \text{if}\;\;\;\; T_{A}\leqslant T(t) \leqslant T_{B} \\ 1 & \;\;\;\; \text{if}\;\;\;\;T(t)>T_{B}\\ 0 &\;\;\;\; \text{if}\;\;\;\; T(t)<T_{A}, \end{array}% \right.\end{aligned}$$ where $T(t)$ is the average daily temperature and $T_{A}=10^{\circ}$C corresponds to a minimum temperature below which the net birth rate vanishes. The choice of the function $\theta(t)$ is not unique: for instance, a power law may be used instead ([@ewing2016modelling]). Here we assume that the positive effect of the temperature on the birth rate saturates at $T_B$, since it was observed by [@oda1980effects] that the number of egg rafts per female does not significantly change between 21$^{\circ}$C and 30$^{\circ}$C. On the other hand, mosquito reproduction is also triggered by rainfalls since these increase the number of breeding sites, such as temporary ground pools. In order to introduce the effect of rainfall on the mosquito birth rate, we compute the accumulated amount of rainwater $H$, whose variation is given by the total daily rainfall $R(t)$ minus the evapotranspiration $E(t)$ ([@gong2011climate]), $$\begin{aligned} \label{eq.HRE} H(t+1)=H(t)+ \left[ R(t)-E(t) \right].\end{aligned}$$ The function $H(t)$ is a convenient measure of the quantity of water available for breeding sites. It should represent the average level of puddles, ponds, drains, small streams, and underground sources such as waste water channels in urban environments. As we will explain below, equation (\[eq.HRE\]) is applied on a daily time scale. The term of evapotranspiration is estimated using the Ivanov model ([@romanenko1961computation; @valipour2014application]), which is based on the mean temperature and the relative humidity \[$Hum(t)$\], and it is given by $$\begin{aligned} \label{eq.EvapEq} E(t)=6.10^{-5}(25+T(t))^2(100-Hum(t)).\end{aligned}$$ Note that the evapotranspiration $E(t)$ is a monotonically increasing function of temperature and a decreasing function of humidity. In particular, $E(t)$ vanishes for $Hum=100$%. In addition, we assume that the level of accumulated water $H$ varies only between a minimum ($H_{min}$) and a maximum ($H_{max}$) boundary level, $i.e.$, $$\begin{aligned} \label{cpattern} H(t+1)=\left\{% \begin{array}{ll} H_{min} &\;\;\;\; \text{if}\;\;\;\; H(t)+ \left[ R(t)-E(t) \right]\leqslant H_{min} \\ H_{max} & \;\;\;\; \text{if}\;\;\;\;H(t)+\left[ R(t)-E(t) \right]\geqslant H_{max}\\ H(t)+ \left[ R(t)-E(t) \right] &\;\;\;\; \text{otherwise}. \\ \end{array}% \right.\end{aligned}$$ Here, $H_{min}$ represents a minimum amount of water that is always available for mosquito breeding, for instance in permanent streams or in the drainage system, while $H_{max}$ is a level of water above which the breeding sites overflow ([@karl2014spatial]). Finally, the factor $\lambda(t)$ that takes into account the effect of rainfall on the birth rate (see Eq. (\[eq.larv\])), is the normalization of $H(t)$: $$\label{eq.lamFin} \lambda(t)=\frac{H(t)}{H_{max}}.$$ Note that the minimum value of $\lambda(t)$ is $H_{min}/H_{max}$. Although we take the integration time step to be $\Delta t=0.1$ days, the data of temperature, rainfall and humidity are available only on a daily time scale. Therefore, for all the integration time steps within a day “$d$” (i.e. $d\leq t<d+1$), we set $\lambda(t)$ and $\theta(t)$ to have the values computed through Eqs. (\[cpattern\]) and (\[eq.EvapEq\]) using the corresponding meteorological data for day “$d$”. Model of synthetic rainfall using the monthly number of rainy days and the monthly precipitation {#SecConst} ------------------------------------------------------------------------------------------------ The amount of average rainfall $P$ and the number of days $D$ with rain per month are two parameters commonly used to characterize the long-term precipitation trend ([@madsen2009update; @owusu2013changing; @zhai2005trends]). In this section, we show how they can be used to construct a synthetic rainfall time series. The average monthly rainfall is assumed to follow a sinusoidal function, $$\begin{aligned} \label{eq.ArtP} P(m)= \frac{P_{max}-P_{min}}{2}\cos\left(\frac{2\pi}{12}(m-m_0)\right)+\frac{P_{max}+P_{min}}{2},\end{aligned}$$ where $m=1,...,12$ represents the month (with $m=1$ for January and $m=12$ for December), $m_0$ corresponds to the month of maximum rainfall, $P_{max}$ is the total precipitation of the wettest month $m_0$, and $P_{min}$ is the total precipitation of the driest month, corresponding to $m = m_0 + 6$. Note that for a higher value of either $P_{max}$ or $P_{min}$ there is an increase in the annual amount of precipitation but, while a higher $P_{max}$ enhances the precipitation difference between the rainy and dry seasons, a higher $P_{min}$ reduces this difference. Similarly, we propose that the number of rainy days is given by $$\begin{aligned} \label{eq.ArtD} D(m)= \frac{D_{max}-D_{min}}{2}\cos\left(\frac{2\pi}{12}(m-m_0)\right)+\frac{D_{max}+D_{min}}{2},\end{aligned}$$ where $D_{max}$ and $D_{min}$ correspond to the number of rainy days in the months labeled by $m_0$ and $m_0 + 6$, respectively. Choosing $m_0=2$, this distribution would be suitable for the city of Córdoba. In Fig. \[fig.Esq1\](a) and (b) we show a schematic of the parameters used in Eqs. (\[eq.ArtP\]) and (\[eq.ArtD\]). [Fig1.pdf]{} (40,55)[(a)]{} (40,20)[(b)]{} (76,55)[(c)]{} From Eqs. (\[eq.ArtP\]) and (\[eq.ArtD\]) we construct the amount of daily rainfall $R(t)$ (see Eq. (\[eq.HRE\])), placing the rainy days in each month at random and assuming that the amount of rain specified in Eq. (\[eq.ArtP\]) is equally distributed over these days. Then we calculate the factor $\lambda(t)$ and integrate Eqs. (\[eq.larv\]) and (\[eq.adult\]). For simplicity we use in these equations the values of humidity and temperature obtained from meteorological data. It is important to note that, by construction, $R(t)$ is a stochastic time series since the rainy days for each month are chosen at random; therefore our results for the effect of the synthetic series $R(t)$ on the mosquito abundance must be averaged over a large number (we take $10^4$) of realizations. Using this model of time series of rainfall, we measure the highest peak of mosquito abundance $M_{max}$ and the time $\tau_{max}$ at which this peak is reached (see Fig. \[fig.Esq1\](c)). In the following section, we explain how to introduce variability on the amount of daily rainfalls. Model of synthetic rainfalls with variable daily intensity of precipitation {#SecVarRai} --------------------------------------------------------------------------- In general, the daily rainfall intensity can range from drizzles with less than $1$ mm to torrential downpours exceeding 200 mm ([@li2013dry; @lei2008effect]). Various distributions, such as Weibull ([@suhaila2007fitting]), lognormal ([@cho2004comparison]) and generalized Pareto ([@deidda2010multiple]), have been proposed to model the precipitation amount. While the above-mentioned distributions could be used to generate a sequence of rainfalls with variable or heterogeneous intensity, the disadvantage of this approach is that the total monthly rainfall is also a stochastic variable. In order to isolate the effect of the heterogeneity of the rainfalls, we will use a fracturing or fragmentation process (FT), which allows us to maintain the total monthly rainfall constant. This method is related to a cascading procedure that was used by physicists to study the fragmentation of brittle material ([@hernandez2003two]). Recently the FT process was also applied to obtain empirical distributions with a finite tail ([@finley2014exploring]). From a geometrical point of view ([@borgos2000partitioning; @finley2014exploring]), this process performs a sequential breakage of a segment or interval of length $\ell$ to obtain $D$ subintervals with variable length $\widetilde{\ell}$, which can be used to decompose the total monthly rainfall into daily rainfalls with heterogeneous intensity. In this representation, $\ell$ and $\widetilde{\ell}$ stand for the total amount of rainfall in a month and in a day, respectively. In \[AppFT\] we explain the steps of the FT process in detail. This method depends on a parameter $\alpha \in [0,1]$ which controls the heterogeneity of the segment length. In particular: - $\alpha=0$ corresponds to the case where an interval is split into two subintervals of equal length $\widetilde{\ell}=\ell/2$, - $\alpha=1$ corresponds to the special case where two “intervals” are generated, one of length zero and the other of length $\widetilde{\ell}=\ell$. In Fig. \[fig.alff\] we show how $\alpha$ controls the shape of the fragment length distribution $\mathcal{P}(\widetilde{\ell})$, using $\ell=150$ mm and $D=10$. Note that the resulting length of each subinterval corresponds to the intensity of rainfall in one day. [Fig2.pdf]{} (85,50) For small values of $\alpha$ the distribution is concentrated around the mean value close to $\ell/D$, while for intermediate values of $\alpha$, the distribution of lengths has a longer tail. Finally, for high values of $\alpha$, $\mathcal{P}(\widetilde{\ell})$ has a peak near $\ell$ which depicts a regime where most of the corresponding month total rainfall is confined to one day. Interestingly, we also note that in this case the rainfall distribution has a region in which it decays as a power law. In order to model the temporal variation of precipitation $R(t)$ (see Sec. \[Sec.lambThe\]), we apply a fracturing process (FT) for each month, partitioning an interval whose length is the amount of monthly precipitation given by Eq. (\[eq.ArtP\]) and where the number of subintervals (the number of rainy days) is given by Eq. (\[eq.ArtD\]). See \[AppFT\] for further details on the construction of $R(t)$. Results {#secResul} ======= Calibration ----------- We calibrate our model of mosquito abundance using a Metropolis-Hastings algorithm (see \[AppCalib\]) with the number of female [*Culex quinquefasciatus*]{} mosquitoes collected in Córdoba city ($31^\circ$24$'$30$''$ S, $64^\circ$11$'$02$''$ W, Córdoba province, Argentina) every two weeks from January 2008 to December 2009 (see  [@batallanthesis; @batallan2015st] for details on the data and their sources). The climate of Córdoba is temperate with dry winters and hot rainy summers. The mean annual temperature ranges between $16^\circ$C-$17^\circ$C and the mean annual rainfall is 800 mm ([@jarsun2003caracterizacion]). The temperature \[$T(t)$\], relative humidity \[$Hum(t)$\], and rainfall \[$R(t)$\] data for Córdoba were obtained from the website [@Exxon_01]. We calibrate the following parameters: $\beta_L$, $H_{max}$, $H_{min}$, and $K_L$. In \[Sec.Sensit\] we perform a sensitivity analysis of these calibrated parameters. As initial conditions of Eqs. (\[eq.larv\]) and (\[eq.adult\]), we set $M(t)=L(t)=20$ . To attenuate the effect of these initial conditions, the integration of the equations starts 12 months before we implement the fitting and study our model. [Fig3.pdf]{} (80,50)[(a)]{} [Fig4.pdf]{} (15,50)[(b)]{} Fig. \[fig.Ajust\](a) shows the fit of our model to the data, where the abundance $M(t)$ is given as the number of female mosquitoes per night per trap. We observe that the mosquito population, which is assumed to be proportional to $M(t)$, increases in summer as expected. Although there are no daily abundance data to compare with, we remark that our model predicts day-to-day changes in the abundance of adult mosquitoes $M(t)$ due to fluctuations in temperature, humidity, and rainfall. In Fig. \[fig.Ajust\](b) we plot the evolution of the immature population of mosquitoes $L(t)$ which shows that temperature and precipitation lead to more abrupt fluctuations in this group than in the compartment of adult mosquitoes. This was to be expected, since these climatic variables are directly introduced into the equation of the population of immature mosquitoes (see Eq. (\[eq.larv\])). In turn, we find that after a rainfall event there is a large increase in the abundance of this group which frequently approaches the carrying capacity, leading to a subsequent population decline due to competition among immature mosquitoes ([@roberts2010larval; @suleman1982effects]). In the following section we study how different rainfall patterns affect the mosquito population dynamics. Effect of different rainfall regimes on $M_{max}$ ------------------------------------------------- The proposed model of synthetic rainfall allows us to explore the evolution of mosquito abundance under possible scenarios in which the weather becomes, for instance, rainier, or with persistent drought conditions. In this Section we discuss how different rain regimes influence the mosquito population when it is at its highest (see Fig. \[fig.Esq1\](c)). To do this, we first assume that the total rainfall $P(m)$ (see Eq. (\[eq.ArtP\])) is equally distributed in $D(m)$ days (see Eq. (\[eq.ArtD\])), the remainder of the days in the month being rainless. In Fig. \[fig.PhaDiag\](a) we consider the weather data (temperature and humidity) of the austral summer season 2008-2009, to show the influence of $P_{max}$ and $D_{max}$ on the highest peak $M_{max}$ of mosquito abundance, with fixed values of the parameters $P_{min} = 10$ mm and $D_{min} =1$. [Fig5.pdf]{} (70,67)[[**(a)**]{}]{} [Fig6.pdf]{} (79,75)[[**(b)**]{}]{} From Fig. \[fig.PhaDiag\](a), we note that, for a fixed number of rainy days $D_{max}$, the highest peak of mosquito abundance increases as $P_{max}$ grows since, as expected, a greater amount of water promotes breeding sites for mosquitoes. Similarly, it can be seen from Figs. \[fig.PhaDiag\](a) and (b) that for $P_{max} \approx 190$ mm, $M_{max}$ is an increasing function with $D_{max}$, because a higher frequency of rainfall events provides more opportunities for mosquitoes to breed. The opposite happens for the lowest level of precipitation ($P_{max} \approx 10$ mm) as there is very little daily rainfall in this regime and the accumulated water evaporates quickly. Interestingly, we find that at moderate precipitation levels $M_{max}$ has a maximum at an intermediate value of the number of rainy days. If the rainfall is equally distributed over a few days, the mosquito population will increase with more rainy days, but, beyond certain point, the rain becomes too thin to maintain all breeding sites active and the mosquito population must decrease. However, as it was mentioned above, the intensity of daily rainfalls is usually far from uniform ([@Exxon_01]), and recent studies suggest that the amount of mosquitoes depend on the distribution of precipitation ([@cheng2016climate; @wang2016stage; @bomblies2012modeling; @bomblies2008hydrology]). Therefore, in the following we study how heterogeneity in the daily rainfall affects the dynamics of mosquito abundance. In figure \[fig.AltAlpha\], using the weather data of the austral summer season 2008-2009 (temperature and humidity), we show the peak mosquito abundance for different values of $P_{max}$, $D_{max}$, and $\alpha$, as obtained from the Eqs. (\[eq.larv\])-(\[eq.lamFin\]). For simplicity, we use the same value of $\alpha$ for each month. [Fig7.pdf]{} (90,73)[(a)]{} [Fig8.pdf]{} (90,75)[(b)]{} [Fig9.pdf]{} (20,60)[(c)]{} [Fig10.pdf]{} (20,60)[(d)]{} From Figs. \[fig.AltAlpha\](a), (b) and (d), we note that, similarly to the case of constant rainfall intensity, the abundance $M_{max}$ is a decreasing (increasing) function with $D_{max}$ for a low (high) amount of monthly precipitation $P_{max}$. Furthermore, Figs.\[fig.AltAlpha\](b)-(d) show that a higher variability in the intensity of rainfalls reduces the dependence of $M_{max}$ with $D_{max}$ with respect to the case of homogeneous rainfall. In particular, for $P_{max}=10$ mm (see Fig.\[fig.AltAlpha\](b)), we note that $M_{max}$ is higher in the heterogeneous case ($\alpha>0$) than in the homogeneous one, since a greater variability tends to confine most of the total monthly precipitation to a few days, in which there is a higher level of water (see Eq. (\[cpattern\])) available for the immature mosquito population. In contrast, for $P_{max}=190$ mm (see Fig.\[fig.AltAlpha\](d)) the heterogeneity diminishes the abundance $M_{max}$ with respect to the homogeneous one. In this case, even if a higher amount of precipitation in a few days would increase the rate of immature mosquito birth, the effect of the intense rain days is limited by the threshold $H_{max}$ because the impact of precipitation saturates for a precipitation higher than $H_{max}$ ($i.e.$, $\lambda(t)=1$, see Eqs. (\[cpattern\]) and (\[eq.lamFin\])). Moreover, there are more rainy days with low precipitation in this regime which further reduces the growth of the mosquito population. As a consequence, the heterogeneity in the intensity of rainfall implies that the monthly total precipitation is a more relevant variable for the prediction of the maximum abundance of mosquitoes than the number of rainy days. Another aspect of relevance for the prediction of vector-borne diseases is the effect of the rainfall in the dry-season (winter) on the future abundance of mosquitoes in summer. To study this relationship, we measure the maximum abundance of mosquitoes $M_{max}$ and the timing $\tau_{max}$ at which the peak of mosquito abundance is reached (see Fig. \[fig.Esq1\](c)), for different values of $P_{min}$ and $D_{min}$, which correspond to the parameters that control the intensity and frequency of rainfall in the driest month, respectively. Here, we keep fixed the parameters $P_{max} =150$ mm and $D_{max}=10$, and assume that February is always the wettest month of the year ($m_0=2$, see Eq. (\[eq.ArtP\])). [Fig11.pdf]{} (20,67)[(a)]{} [Fig12.pdf]{} (20,71)[(b)]{} [Fig13.pdf]{} (20,26)[(c)]{} [Fig14.pdf]{} (20,26)[(d)]{} [Fig15.pdf]{} (20,26)[(e)]{} [Fig16.pdf]{} (20,26)[(f)]{} We note from Figs. \[fig.Winter\](a) and (c) that, in the case of a homogeneous distribution of rainfalls, the time of peak $\tau_{max}$ moves forward for high values of $P_{min}$ and $D_{min}$, because in this case the rainfalls are regular and abundant throughout the year, which favors mosquito breeding. However, for the explored values of $P_{min}$ and $D_{min}$, the position of this peak is in late February or March, $i.e.$, just after the wettest month of the year. On the other hand, in a scenario of heterogeneous rainfalls with $\alpha=0.9$ (see Figs. \[fig.Winter\](b) and (d)), $\tau_{max}$ is moved forward by only approximately 10 days with respect to the homogeneous intensity case. Correspondingly, for constant values of $\alpha$ (0.1, 0.5 and 0.9) the changes in $P_{min}$ and $D_{min}$ only affect $M_{max}$ by less than 5% (not shown here). Therefore these results suggest that the peak of abundance of mosquitoes and $\tau_{max}$ are mainly determined by summer weather conditions and the carrying capacity of the system and not by the intensity and distribution of precipitation throughout the year. Despite the weak effect of $P_{min}$ on $M_{max}$, Figs. \[fig.Winter\] (e) and (f) show, as expected, that an increasing value of $P_{min}$ could have a remarkable effect on the accumulated abundance of mosquitoes (one measure of which is the time integral of $M(t)$ over the period of interest), since from $P_{min}=10$ mm to $P_{min}=130$ mm, it could increase by more than 40%. However, for the case of a fixed value of $P_{min}$ and higher values of $\alpha$ we obtain that the accumulated abundance of mosquitoes diminishes down to a 50% of the value for a homogeneous rain distribution. Consequently, the heterogeneity could help to attenuate the enhancement of the mosquito population. Therefore, these findings suggest that in order to predict the total annual abundance, it is not only necessary to take into account the overall amount of rainfall throughout the year but also the heterogeneity in daily rainfall intensity. Discussion {#sec.Discussion} ========== Since projections of climate change ([@nunez2009regional]) suggest that for the late twenty-first century in regions of South America, such as Córdoba province, the pattern of precipitation will change towards a regime with rainier autumns and an increase in extreme events, it is crucial to study how this variation would affect the mosquito abundance. In this paper we studied the effects of the total intensity, number of rainy days and heterogeneity of rainfall on the mosquito population. We found that for a regime with a low total rainfall, the abundance of mosquitoes is a decreasing function with the number of rainy days, while for a high total rainfall regime it is an increasing function of this number. Interestingly, for an intermediate precipitation regime, we found that there is a halfway number $D_{max}$ of rainy days for which $M_{max}$ is optimized. Since $P_{max}$ is fixed, fewer rainy days would imply dry intervals, leading to a lessening of the mosquito abundance. If the number of rainy days exceeds the optimal value of $D_{max}$, a considerable fraction of the rainwater resulting from the typically meager rainfall would disappear due to evapotranspiration, again leading to a reduction of the mosquito abundance. In order to study the effect of the heterogeneity in the daily rainfall, we used a fracturing process that keeps constant the total amount of monthly precipitation. We observed that a higher heterogeneity reduces the dependence of $M_{max}$ on the number of rainy days. However, an increasing variability favors the mosquito production in the low rainfall regime, while the opposite behavior takes place in the case of high precipitation $P_{max}$. Therefore, if climatic models predict the intensification of storms, but not an increase in the total amount of precipitation, our model predicts that the enhancement of mosquito abundance would be more significant in semiarid areas than in humid climates. Finally we study the effect of an increasing amount of rainfall in the dry season on the mosquito abundance dynamics, obtaining that high precipitation throughout the year does not significantly alter the maximum abundance or the time at which this peak occurs, but it could notably increase the accumulated abundance of mosquitoes. However, we also observed that a regime with a higher variability of rainfall intensity could reduce this increase. While our model captures multiple relationships between rainfall and mosquito population, additional extensions could be considered. For instance, there is evidence that rainfalls reduce the immature population in the short term due to flushing of breeding sites ([@gardner2012weather; @strickman1988rate]) and affect the bacterial concentration used as food by mosquito larvae ([@chaves2011weather]); therefore, it would be interesting to study the relevance of these effects on the dynamic of mosquito population. We think that our findings could be used as support and reference guidance for the assessment of the influence of different rainfall regimes on the mosquito population dynamics, using the weather data for any specific region. Such an assessment would impact positively on our ability to make predictions for the spread of various possible arboviruses. It is also known that rainfall could have a substantial effect on insecticide residence times ([@allan2009environmental]). The model presented here can be used to optimize the efficacy of mosquito control campaigns, using temperature and rainfall data to select the best times for the application of population reduction procedures. Calibration {#AppCalib} =========== The Metropolis-Hastings (MH) algorithm is a stochastic optimization tool for fitting statistical models to data that has been used in cosmology ([@christensen2001bayesian; @christensen2003metropolis; @lewis2002cosmological]), epidemiology ([@merler2015spatiotemporal]), and in the study of mosquito population dynamics ([@marini2016role]). This algorithm allows us to estimate the unknown values of some parameters $\Theta$ ($\Theta$ represents either a single parameter or a parameter set), by means of a stochastic search in the parameter space that generates a sequence or chain $\Theta^{(i)}$, where $i$ represents the step number of the MH algorithm ([@bonamente2013statistics]). Each value of this chain is sampled from a proposal distribution and accepted with a probability $\sigma$ defined by an acceptance function, which depends on the likelihood function of the observations. In our model we estimate the parameters $\beta_L$, $H_{max}$, $H_{min}$ and $K_L$ using a MH algorithm and the data for adult mosquito abundance from Córdoba city in the period 2008-2009 ([@batallanthesis; @batallan2015st]). We propose that the likelihood of the observations is given by $$\begin{aligned} L=\prod_{j=1}^{n}p(x_j(H_{max},H_{min},K_L,\beta_L);k_j),\end{aligned}$$ where $n$ is the number of data points and $p(x_j(H_{max},H_{min},K_L,\beta_L);k_j)$ is the probability to observe the abundance $k_j$ of mosquitoes obtained from the data. Here we assume that $p(\cdot)$ follows a Poisson distribution whose mean $x(H_{max},H_{min},K_L,\beta_L)$ is the number of adult mosquitoes predicted by Eqs. (\[eq.larv\])-(\[eq.lamFin\]). The MH algorithm implemented in this paper has the following steps: - Step 1: Initialize the starting value of the parameters $\Theta^{(i=0)}$, using a uniform distribution in order to avoid favoring any initial value. - Step 2: Generate a new sample of the parameters, $\Theta^{New}$ starting from a proposal distribution that indicates a candidate for the next sample value. To ensure that the new values of the parameters are positive, we use as a proposal distribution a log-normal density which has a mean equal to the logarithm of the current value parameter and constant variance $\delta$. The value of this variance is chosen in order to guarantee an acceptance rate between 10% and 30% in the burn-in period. - Step 3: accept the new candidate $\Theta^{New}$ with probability $\sigma$: $$\begin{aligned} \sigma &=& min\bigg\{1,\frac{L(\Theta^{New})}{L(\Theta^{(i)})}\bigg\}. \end{aligned}$$ - Step 4: repeat steps 2 and 3 until convergence is reached. We perform $2.10^6$ iterations and check convergence by visual inspection of the chain $\Theta^{(i)}$. In order to construct the posterior distribution of the parameters, we discard the first $10^5$ iterations as a burn-in and we only keep every $20^{th}$ sampled value of the remaining iterations to reduce autocorrelation within successive samples. Finally, the values of the parameters that we will use in our model are the averages of the medians of the posterior distributions obtained from 5 different initial conditions (see step one of the MH algorithm). Fig. \[fig.AppMH\] shows the posterior distribution obtained for the parameters $\beta_L$, $H_{max}$, $H_{min}$ and $K_{L}$. We note that all of these distributions are unimodal, except for $H_{max}$. Although in this paper we set $H_{max}=9.86$ mm, since it is the average value of the median obtained from the MH algorithm, we also check our model for $H_{max}\approx 7$ mm and $H_{max}\approx 13$ mm which are the positions of the highest peaks of the posterior distribution (see Fig. \[fig.AppMH\](b)). For these cases, our results presented in Section \[secResul\] do no qualitatively change. [Fig17.pdf]{} (15,50)[[**[(a)]{}**]{}]{} [Fig18.pdf]{} (15,50)[[**[(b)]{}**]{}]{} [Fig19.pdf]{} (15,50)[[**[(c)]{}**]{}]{} [Fig20.pdf]{} (15,50)[[**[(d)]{}**]{}]{} Fracturing process {#AppFT} ================== Fracturing process (FT) is a stochastic iterative process which generates a finite sequence of numbers with the property that their sum is always a constant ([@borgos2000partitioning; @finley2014exploring]). From a geometrical point of view, this method consists of partitioning an interval of length $\ell$ in a number $D$ of subintervals or segments, with the property that the sum of their lengths is always $\ell$. Following  [@finley2014exploring], the FT process starts with an interval or segment of length $\ell$ which is split into two subintervals of lengths $\widetilde{\ell}$ and $\ell-\widetilde{\ell}$, where $\widetilde{\ell}$ is a stochastic variable generated by the following function: $$\begin{aligned} \label{alphaFTT} \widetilde{\ell}=\ell \times \left\{\begin{array}{ll} \rho \frac{1-\alpha}{\alpha} &\;\;\;\; \text{if}\;\;\;\; 0\leqslant \rho <\frac{\alpha}{2} \\ \frac{1}{2}+\left(\rho-\frac{1}{2}\right)\frac{\alpha}{1-\alpha} &\;\;\;\; \text{if}\;\;\;\; \frac{\alpha}{2}\leqslant \rho\leqslant 1-\frac{\alpha}{2} \\ 1-(1-\rho)\frac{1-\alpha}{\alpha} &\;\;\;\; \text{if}\;\;\;\; 1-\frac{\alpha}{2} < \rho \leqslant 1 \\ \end{array}\right.\end{aligned}$$ Here $\rho$ is a uniform random variable and $\alpha \in (0,1)$ is a parameter that controls the average length $\widetilde{\ell}$. In a second step, the partition function, Eq. (\[alphaFTT\]), is applied again on the intervals resulting from the previous step, generating a total of 4 intervals. This procedure is repeated until the required number of intervals is reached [^1]. In order to model the variability in the daily rainfall $R(t)$, we apply a FT process for each month, in which, - the length of the initial segment $\ell$ is given by Eq. (\[eq.ArtP\]), $i.e.$, the monthly precipitation, - the number $D$ of subintervals is given by Eq. (\[eq.ArtD\]), $i.e.$, the number of rainy days. After we apply the fracturing process, the length of each resulting interval $\widetilde{\ell}_i$ (with $i=1,\cdots,D$) represents the total amount of water that falls in the day $d_i$, which we choose at random as it is shown in the schematic of Fig. \[fig.AppAlpha\], and then we set $R(d_i)=\widetilde{\ell}_i$. [Fig21.pdf]{} (90,56) In Fig. \[fig.AlphaCord\] we plot the distribution of segment lengths obtained from an FT process for $\ell=150$ mm and $D=10$, which are the average rainfall and the number of rainy days in February, respectively. Here we use the value of $\alpha=0.43$ which gives the best fit to the February rainfalls in the city of Córdoba in the period 2001-2015. [Fig22.pdf]{} (90,56) Sensitivity analysis {#Sec.Sensit} ==================== A sensitivity analysis allows us to measure the impact of different parameters on the relevant variables of our model. In this section, we perform a one-way sensitivity analysis on the model of Sec. \[Sec.modd\], by varying a $\pm 25$% of the baseline values of individual parameters one at a time, while keeping the other parameters constant in order to analyze their individual impact on the maximum abundance of mosquitoes $M_{max}$. The parameters examined are: $\beta_L$, $H_{max}$, $H_{min}$, and $K_{L}$. The results of the sensitivity analysis are summarized in Table \[tab.Sens\]. Parameter -25% +25% ----------- ------- ------- $\beta_L$ -2.1% +1% $H_{max}$ -0.4% +0.9% $H_{min}$ -0.7% +0.6% $K_L$ -25% +25% : Variation of the maximum abundance of mosquitoes ($M_{max}=17.3$) when it is applied a one-way sensitivity analysis.[]{data-label="tab.Sens"} It shows, as expected, that for higher values of $\beta_L$, $H_{max}$, $H_{min}$, and $K_L$ the abundance $M_{max}$ increases. Although the influence of the first three is rather weak, $M_{max}$ is heavily influenced by $K_L$, which is therefore a critical parameter for the estimation of mosquito abundance. Acknowledgments {#acknowledgments .unnumbered} =============== This work was supported by SECyT-UNC (Projects 103/15 and 313/16), CONICET (PIP 11220110100794), and PICT Cambio Climático (Ministerio de Ciencia y Técnica de la Provincia de Córdoba), PICT Nro. 2013-1779 (ANPCYT-MYNCYT), Argentina. We also thank Dr. A. M. Visintin and Biól. M. Beranek for useful discussions. Bibliography {#bibliography .unnumbered} ============ [^1]: For example, in order to generate a total of five subintervals, three iterations of the fracturing process must be performed: step 1) the initial interval is divided into two subintervals, step 2) each of the previous subintervals is divided into two parts, and finally step 3) one randomly chosen subinterval of the previous step is split into two subintervals.

ArXiv

--- abstract: 'We establish the existence and regularity properties of a monodromy operator for the linearization of the cubic-quintic complex Ginzburg-Landau equation about a periodically-stationary (breather) solution. We derive a formula for the essential spectrum of the monodromy operator in terms of that of the associated asymptotic linear differential operator. This result is obtained using the theory of analytic semigroups under the assumption that the Ginzburg-Landau equation includes a spectral filtering (diffusion) term. We discuss applications to the stability of periodically-stationary pulses in ultrafast fiber lasers.' author: - John Zweck - Yuri Latushkin - 'Jeremy L. Marzuola' - 'Christopher K.R.T. Jones' bibliography: - '../../BibFiles/PreambleSpaces.bib' - '../../BibFiles/Stability.bib' - '../../BibFiles/Lasers.bib' title: 'The essential spectrum of periodically-stationary solutions of the complex Ginzburg-Landau equation [^1] [^2] [^3] [^4]' --- Introduction ============ The cubic-quintic complex Ginzburg-Landau equation (CQ-CGLE) is a fundamental model for nonlinear waves and coherent structures that arise in fields such as nonlinear optics and condensed matter physics [@akhmediev2008dissipative; @RMP74p99]. The CQ-CGLE supports a wide variety of solutions, including stationary pulses, periodically-stationary pulses, fronts, exploding solitons, and chaotic solutions [@akhmediev2008dissipative]. While stationary pulses (solitons) maintain their shape, periodically-stationary pulses (breathers) change shape as they propagate, returning to the same shape periodically. Periodically-stationary pulses are the solutions of primary interest to engineers designing ultrafast fiber lasers. These lasers generate pulses with widths in the picosecond to femtosecond range and have applications to time and frequency metrology [@chong2015ultrafast; @kartner2004few; @SIREV48p629]. Since the advent of the soliton laser [@Duling:91; @Mollenauer:84], researchers have invented several generations of short-pulse, high-energy fiber lasers for a variety of applications. In the mid 1990’s stretched-pulse (dispersion-managed) lasers were devised to generate pulses with higher energy and shorter duration than can be acheived with soliton lasers [@tamura199377; @tamura1994soliton]. Dissipative soliton lasers, in which effects such as spectral filtering play a significant role, were introduced in about 2005 and are suitable for high energy applications [@chong2006all; @grelu2012dissipative]. Similariton lasers were introduced in 2010 to create femto-second pulses with a high tolerance to noise [@chong2015ultrafast; @fermann2000self; @hartl2007cavity] by exploiting the theoretical discovery of exponentially growing, self-similar pulses in optical fiber amplifiers. The most recent invention is the Mamyshev oscillator which can produce pulses with a peak power in the Megawatt range [@regelskis2015ytterbium; @sidorenko2018self]. The key issue for mathematical modeling of ultrafast lasers is to determine those regions in the design parameter space in which stable pulses exist, and within that space to optimize the pulse parameters. A significant challenge is that from one generation of lasers to the next there has been an increase in the amount by which the pulse changes within each round trip (period). Consequently, soliton perturbation theory [@OptLett11p665; @PLA42p5689], which was developed to analyze the stability of stationary pulses, is no longer applicable. Although low-dimensional reduced ODE models [@tsoy2005bifurcations; @tsoy2006dynamical] and Monte Carlo simulations that use full PDE models [@zweck2018computation] have both been employed to assess pulse stability, to date there is no mathematical theory to determine the stability of periodically-stationary laser pulses. In ultrafast lasers, the different physical effects (dispersion, nonlinearity, spectral filtering, saturable gain and loss) occur in different devices within the laser. For a quantitative model it is therefore necessary to use am equation such as the CQ-CGLE in which the coefficients are piecewise constant in the evolution variable. However, constant coefficient models are also often used to gain qualitative insight into the system behavior [@renninger2008dissipative]. In this paper, we take a first step in the development of a stability theory for ultrafast lasers by calculating the essential spectrum of the monodromy operator of the linearization of the CQ-CGLE about a periodically-stationary pulse solution. Because variable coefficient equations are more challenging to analyze, we restrict attention to the constant coefficient CQ-CGLE, which is a phenomenological, distributed model for short-pulse fiber lasers [@grelu2012dissipative; @SIREV48p629]. This equation has two important classes of solutions that are periodic in the temporal variable. The first class is the family of Kuznetsov-Ma (KM) breathers [@kuznetsov1977solitons; @ma1979perturbed], which are analytical solutions of the focusing nonlinear Schrödinger equation (FNLSE), a special case of the CQ-CGLE. These solutions, which were discovered using integrable systems techniques, have a non-zero background at spatial infinity. A numerical Floquet spectrum computation by Cuevas-Maraver *et al.* [@cuevas2017floquet] suggests that the KM breather is linearly unstable. Muñoz [@munoz2017instability] recently used a Lyapunov functional to prove that because of their non-zero background, these breathers are unstable under small $H^s$ ($s>\frac 12$) perturbations. Integrable systems techniques have also been used to find breather solutions of the modified and higher-order KdV equations and the Gardner hierarchy [@clarke2000chaos; @alejo2019dynamics]. In a recent series of papers [@alejo2018nonlinear; @alejo2013nonlinear; @alejo2017variational], Alejo exploited the integrability structure of these PDE’s and Lyapunov functional techniques to establish the nonlinear stability of several such breathers. The second class consists of the periodically-stationary pulses discovered numerically by Akhmediev and his collaborators [@akhmediev2001pulsating; @tsoy2005bifurcations; @tsoy2006dynamical]. These solutions were found in the case that the CQ-CGLE includes a spectral filtering term. Although Akhmediev *et al.* provided strong numerical evidence for the existence of these solutions, there are no known analytic formulae for these solutions and no mathematical proof of their existence. However, using numerical simulations and reduced ODE models, they provided numerical evidence for the existence of both stable and unstable periodically-stationary pulses. In Floquet theory, the stability of periodic solutions of a system of ODE’s is characterized by the spectrum of the monodromy matrix of the linearization of the system about the solution. Although Floquet methods have been developed for solutions of PDE’s that are periodic in the spatial variables (see for example [@gesztesy1995floquet; @jones2010stability; @kuchment2012floquet; @reed1980methods]), we are only aware of a few results for solutions that are periodic in the temporal (evolution) variable. Wilkening [@wilkeningharmonic] developed a numerical method to study the stability of standing water waves and other time-periodic solutions of the free-surface Euler equations. Motivated by problems from quantum mechanics, Korotyaev [@korotyaev1983spectrum; @korotyaev1985eigenfunctions] studied Schrödinger operators that have a real scalar potential which is periodic in time and rapidly decaying in space. He showed that the monodromy operator has no singular continuous spectrum. However, this result relies heavily on fact that the evolution operator is unitary, which is not the case when the CQ-CGLE includes a spectral filtering term. Similar results are discussed in the book of Kuchment [@kuchment2012floquet]. Finally, Sandstede and Scheel [@doelman2009dynamics; @sandstede2001structure; @sandstede2002stability] have an extensive body of theoretical and numerical results on the stability of time-periodic perturbations of spatially-periodic traveling waves. However, because of the underlying spatial periodicity in their formulation, these results are not applicable to laser systems. The results in this paper can be summarized as follows. In Section \[Sec:Exs\], we review the periodically-stationary solutions of Kuznetsov-Ma and Akhmediev and define the time-periodic operator, $\mathcal L(t)$, which is obtained by linearization of the CQ-CGLE about a periodically-stationary solution. We also discuss the asymptotic operator, $\mathcal L_\infty$, associated with $\mathcal L(t)$. In Section \[Sec:EssSpecL\], we calculate the essential spectrum of $\mathcal L(t)$ with the aid of some results from the text of Kapitula and Promislow [@kapitula2013spectral]. In Section \[Sec:EvolFamily\], we establish the existence of an evolution family for $\mathcal L(t)$ by applying classical results on solutions of initial-value problems for non-autonomous linear differential equations in Banach spaces [@pazy2012semigroups]. In Section \[Sec:EssSpecM\], we use the results obtained in Section \[Sec:EvolFamily\] to define the monodromy operator, $\mathcal M(s)$, and establish the main result of the paper, which is a formula for the essential spectrum of $\mathcal M(s)$ in terms of the essential spectrum of the asymptotic operator, $\mathcal L_\infty$. To obtain this result, we assumed that the CQ-CGLE includes a spectral filtering term, which ensures that the semigroup of the asymptotic operator is analytic. Since all fiber lasers have bandlimited gain, this assumption holds in applications. Based on the numerical results of Cuevas-Maraver *et al.* [@cuevas2017floquet], we conjecture that the formula for the essential spectrum of $\mathcal M(s)$ also holds for the KM breather. However a different approach will be required to prove such a result, since in this case the asymptotic operator is not analytic. Motivating examples {#Sec:Exs} =================== We consider a class of one-dimensional, constant-coefficient nonlinear Schrödinger equations of the form $$i\partial_t \psi \,\,+\,\, \tfrac 12 \partial_x^2 \psi \,\,+\,\, f( | \psi |^2)\psi=0, \label{eq:nls}$$ where $f$ is a polynomial with complex coefficients. We call $t$ the *temporal* or *evolution* variable and $x$ the *spatial* variable. We consider solutions, $\psi$, for which $f( | \psi |^2)\psi \to 0$ at an exponential rate as $x\to \pm\infty$. We say that $\psi$ is a *periodically-stationary solution* of if there is a period, $T$, so that $\psi(t+T,x) = \psi(t,x)$ for all $t$ and $x$. Our analysis is motivated by two important examples. \[ex:KM\] Kuznetsov [@kuznetsov1977solitons] and Ma [@ma1979perturbed] independently discovered a family of periodically-stationary solutions of the FNLSE, $$i\partial_t \psi \,\,+\,\, \tfrac 12 \partial_x^2 \psi \,\,+\,\, ( | \psi |^2-\nu_0^2)\psi=0, \label{eq:KMnls}$$ with the property that $$\lim\limits_{x\to\pm\infty} \psi(t,x) = \nu_0.$$ Here the parameter, $\nu_0>0$, is the background amplitude. The Kuznetsov-Ma (KM) breathers are defined in terms of a second parameter, $\nu > \nu_0$, by [@garnier2012inverse] $$\psi_{\rm{KM}}(t,x)\,\,=\,\, \nu_0 \,\,+\,\, 2\eta\, \frac{\eta\cos(2\nu\eta t) + i\nu \sin(2\nu\eta t) }{\nu_0 \cos(2\nu\eta t) - \nu \cosh(2\eta x)}, \label{eq:KM}$$ where $\eta = \sqrt{\nu^2-\nu_0^2} > 0$. The period of $\psi_{\rm{KM}}$ is $T= {\pi}/{\nu\eta}$. We observe that $f(|\psi_{\rm{KM}}|^2)\psi_{\rm{KM}} = (|\psi_{\rm{KM}}|^2 -\nu_0^2)\psi_{\rm{KM}}\to 0$ at an exponential rate as $x\to\pm\infty$. \[ex:GL\] The CQ-CGLE [@akhmediev2008dissipative; @RMP74p99] is given by $$i\partial_t\psi + (\tfrac D2 - i \beta)\partial_x^2 \psi - i\delta\psi + \left[\gamma-i\epsilon + (\nu-i\mu) |\psi|^2\right]|\psi|^2\psi \,\,=\,\, 0. \label{eq:GL}$$ Among other applications, the CQ-CGLE provides a qualitative model for the generation of short-pulses in mode-locked fiber lasers [@akhmediev2008dissipative; @SIREV48p629]. In this context, the parameters in the equation can be interpreted as follows. The coefficient, $D$, is the fiber dispersion, which is positive in the anomalous or focusing dispersion regime and negative in the normal or defocusing dispersion regime. Spectral filtering is modeling using the term with coefficient $\beta > 0$. The remaining terms model linear gain or loss ($\delta$), saturable nonlinear gain ($\epsilon>0$ and $\mu<0$), and the cubic and quintic nonlinear electric susceptibility of the optical fiber ($\gamma>0$ and $\nu>0$). Using a numerical partial differential equation solver, Akhmediev and his collaborators [@akhmediev2008dissipative; @akhmediev2001pulsating; @tsoy2005bifurcations; @tsoy2006dynamical] provide strong evidence for the existence of periodically-stationary solutions of , which they refer to as pulsating solitons. There are no known analytical formulae for these solutions. However, the numerical results show that these solutions decay at an exponential rate as $x\to\pm\infty$. Equations  and are both special cases of the general nonlinear wave equation $$i\partial_t\psi + (\tfrac D2 - i \beta)\partial_x^2 \psi + (\alpha - i\delta)\psi + \left[\gamma-i\epsilon + (\nu-i\mu) |\psi|^2\right]|\psi|^2\psi \,\,=\,\, 0. \label{eq:NWE}$$ Throughout this paper, we assume that $(D,\beta) \neq (0,0)$. Since the linearization of about a solution involves both the linearized unknown and its complex conjugate, we reformulate as the system of equations for $\boldsymbol\psi = \begin{bmatrix} \Re(\psi) & \Im(\psi)\end{bmatrix}^T$ given by $$\partial_t \boldsymbol\psi \,\,=\,\, \left(\mathbf{B} \, \partial^2_x + \mathbf{N}_0 + \mathbf{N}_1 | \boldsymbol\psi |^2 + \mathbf{N}_2 | \boldsymbol\psi |^4 \right) \boldsymbol\psi, \label{eq:NWEsys}$$ where $$\mathbf{B} = \begin{bmatrix} \beta & -\frac D2 \\ \frac D2 & \beta\end{bmatrix},$$ and $$\mathbf{N}_0 = \begin{bmatrix} \delta & -\alpha \\ \alpha & \delta \end{bmatrix}, \qquad \mathbf{N}_1 = \begin{bmatrix} \epsilon & -\gamma \\ \gamma & \epsilon \end{bmatrix}, \qquad \mathbf{N}_2 = \begin{bmatrix} \mu & -\nu \\ \nu & \mu \end{bmatrix}.$$ The linearization of about a solution, $\boldsymbol\psi$, is of the form $$\partial_t \mathbf p \,\,=\,\,\mathcal L(t) \mathbf p, \label{eq:Linearization}$$ where $\mathcal L=\mathcal L(t)$ is a second-order, linear differential operator in $x$ with real, $t$- and $x$-dependent, matrix-valued coefficients. If the solution $\boldsymbol\psi$ is periodically stationary with period $T$, then $\mathcal L$ is periodic in $t$ with $\mathcal L(t+T) = \mathcal L(t)$. Substituting $\boldsymbol\psi_\varepsilon = \boldsymbol\psi+ \varepsilon \mathbf p$ into are keeping only terms of order $ \varepsilon$ we find that $$\mathcal L(t) \,\,=\,\, \mathbf{B} \, \partial^2_x + \widetilde{ \mathbf{M}}(t), \label{eq:OpL}$$ where $\widetilde{\mathbf{M}}(t)$ is the operator of multiplication by $$\widetilde{ \mathbf{M}}(t,x) = \mathbf{N}_0 + \mathbf{N}_1 | \boldsymbol\psi |^2 + \mathbf{N}_2 | \boldsymbol\psi |^4 + \left(2\mathbf{N}_1 + 4\mathbf{N}_2 |\boldsymbol\psi |^2\right) \boldsymbol\psi \boldsymbol\psi^T. \label{eq:OpM}$$ Essential Spectrum of the Linearized Differential Operator {#Sec:EssSpecL} ========================================================== In this section we introduce the asymptotic operator, $\mathcal L_\infty$, associated with the differential operator, $\mathcal L=\mathcal L(t)$, and determine the essential spectra of $\mathcal L_\infty$ and $\mathcal L$. Our results rely on a general theory summarized by Kapitula and Promislow [@kapitula2013spectral]. Although their results are stated for operators with scalar coefficients, the authors remark that they also hold for operators with matrix-valued coefficients such as the operator, $\mathcal L$ in . We begin with the following assumption. \[hyp:EA\] The solution, $\psi=u+iv$, about which is linearized is assumed to be periodically stationary with period, $T$, and has the following properties: 1. $\psi$ is bounded on $[0,T]\times \mathbb R$, 2. There exist constants $r>0$ and $\psi_\infty \in \mathbb C$ so that $$\lim\limits_{|x|\to\infty} e^{r|x|} |\psi(t,x) - \psi_\infty| \,\,=\,\,0, \qquad \text{for all } t\in[0,T],$$ 3. $\psi$ is differentiable with respect to $x$, and $\psi_x$ is bounded on $[0,T]\times \mathbb R$. The KM breather satisfies Hypothesis \[hyp:EA\]. Although we do not have definitive proof, it is reasonable to assume that the numerically computed periodically-stationary solultions discussed in Example \[ex:GL\] also satisfy Hypothesis \[hyp:EA\]. \[remark:spaces\] Hypothesis \[hyp:EA\] is sufficient to guarantee that for each $t$, the coefficients of the operator, $\mathcal L(t)$, in  lie in $W^{1,\infty}(\mathbb R, \mathbb C^{2\times2}) \cap H^1(\mathbb R, \mathbb C^{2\times2})$. This assumption is used in the proof of Theorem \[thm:EssSpec\] below. In addition, it guarantees that for each $t$, the multiplication operator, $\widetilde{\mathbf M}(t)$, in  is a bounded operator on $L^2(\mathbb R, \mathbb C^2)$. Since $(D,\beta) \neq (0,0)$, it follows from [@kapitula2013spectral Lemma 3.1.2] that the operator $\mathcal L(t) : H^2(\mathbb R, \mathbb C^2) \subset L^2(\mathbb R, \mathbb C^2) \to L^2(\mathbb R, \mathbb C^2)$ is closed. Next, we have the following proposition, whose proof is self-evident. Assume that Hypothesis \[hyp:EA\] is met. Then $$\mathbf{M}_\infty := \lim\limits_{|x|\to\infty} \widetilde{ \mathbf{M}}(t,x)$$ exists and is $t$-independent. Furthermore, the differential operator, $\mathcal L$, is exponentially asymptotic in that the leading coefficient, $ \mathbf{B}$, is constant and there exists a constant $r>0$ so that $$\lim\limits_{|x|\to\infty} e^{r|x|} \,\left|\widetilde{\mathbf M}(t,x) - \mathbf{M}_\infty\right| \,\,=\,\,0 \qquad \text{for all } t \in [0,T].$$ The *asymptotic differential operator*, $\mathcal L_\infty$, associated with the exponentially asymptotic operator, $\mathcal L(t)$, is the $t$-independent operator with constant, matrix-valued coefficients given by $$\mathcal L_\infty \,\,:=\,\, \mathbf{B} \,\partial^2_x + \mathbf{M}_\infty. \label{eq:OpLinf}$$ Furthermore, if we define $ \mathbf{M}(t,x) := \widetilde{ \mathbf{M}}(t,x) - \mathbf{M}_\infty$, then $$\mathcal L(t) = \mathcal L_\infty + \mathbf{M}(t),$$ where $\mathbf{M}(t,x)\to \mathbf 0$ as $|x|\to\infty$. We now review the definition of the essential spectrum we use in this paper. \[SpecDefs\] Let $X$ be a Banach space and let $\mathcal B(X)$ denote the space of bounded linear operators on $X$. Let $\mathcal L: D(\mathcal L)\subset X\to X$ be a closed linear operator with domain $D(\mathcal L)$ that is dense in $X$. The *resolvent set* of $\mathcal L$ is $$\rho(\mathcal L) \,\,:=\,\, \{ \lambda \in \mathbb C \, |\, \mathcal L - \lambda \text{ is invertible and } (\mathcal L - \lambda)^{-1} \in \mathcal B(X) \},$$ and for each $\lambda \in \rho( \mathcal L)$, the *resolvent operator* is $R(\lambda : \mathcal L) := (\mathcal L - \lambda)^{-1}$. The *spectrum* of $\mathcal L$ is $\sigma(\mathcal L) \,\,:=\,\, \mathbb C \setminus \rho(\mathcal L)$. The *point spectrum* of $\mathcal L$ is $$\sigma_{\rm pt}(\mathcal L) \,\,:=\,\, \{ \lambda \in \mathbb C \, |\, \operatorname{Ker}(\mathcal L - \lambda) \neq \{0\} \}.$$ The *Fredholm point spectrum* of $\mathcal L$ is the subset of $\sigma_{\rm pt}(\mathcal L)$ defined by $$\sigma^{\mathcal F}_{\rm pt}(\mathcal L) \,\,:=\,\, \{ \lambda \in \mathbb C \, |\, \mathcal L - \lambda \text{ is Fredholm, } \operatorname{Ind}(\mathcal L - \lambda)=0, \text{ and } \operatorname{Ker}(\mathcal L - \lambda) \neq \{0\} \},$$ and the *essential spectrum* of $\mathcal L$ is $\sigma_{\rm{ess}}(\mathcal L) \,\,:=\,\, \sigma(\mathcal L) \setminus \sigma^{\mathcal F}_{\rm{pt}}(\mathcal L)$. We observe that $\sigma(\mathcal L) = \sigma_{\rm pt}(\mathcal L) \cup \sigma_{\rm{ess}}(\mathcal L)$, but note that this union may not be disjoint. Since the operators we consider are not self-adjoint, there are several non-equivalent definitions of the essential spectrum [@edmunds2018spectral]. An argument that involves the closed graph theorem [@kato2013perturbation] shows that with the definition we use, $\sigma_{\rm{ess}}(\mathcal L)$ consists of those $\lambda\in\mathbb C$ so that either (a) $\mathcal L - \lambda$ is Fredholm but has $\operatorname{Ind}(\mathcal L - \lambda) \neq 0$ or (b) $\mathcal L - \lambda$ is not Fredholm. This definition gives the largest subset of the spectrum that is invariant under compact perturbations [@edmunds2018spectral]. However, the operators, $\mathcal L$, we consider are given as perturbations of constant coefficient differential operators, $\mathcal L_\infty$, by multiplication operators, which are not compact. Nevertheless, it can be shown that $\mathcal L$ is a *relatively compact perturbation* of $\mathcal L_\infty$, by which we mean that $\exists \lambda \in \rho(\mathcal L_\infty)$ so that $(\mathcal L - \mathcal L_\infty) (\mathcal L_\infty-\lambda)^{-1} : X \to X$ is compact. Consequently, by Weyl’s essential spectrum theorem [@kapitula2013spectral], $\sigma_{\rm{ess}}(\mathcal L) = \sigma_{\rm{ess}}(\mathcal L_\infty)$. To calculate the spectrum of the asymptotic operator, $\mathcal L_\infty$, we convert the equation, $(\mathcal L_\infty - \lambda)\mathbf p = \mathbf 0$, to the first-order system, $$\partial_x \mathbf Y \,\,=\,\, \mathbf A_\infty(\lambda) \mathbf Y, \label{eq:Y}$$ where $\mathbf Y = \begin{bmatrix} \mathbf p \\ \mathbf p_x\end{bmatrix} \in \mathbb C^4$ and $\mathbf A_\infty(\lambda)\in \mathbb C^{4\times 4}$ is the constant matrix $$\mathbf A_\infty(\lambda) \,\, := \,\, \begin{bmatrix} \mathbf{0} & \mathbf{I} \\ \mathbf B^{-1}(\lambda - \mathbf M_\infty) & \mathbf{0} \end{bmatrix}.$$ Since does not have solutions that decay as $x\to +\infty$ and as $x\to -\infty$, $\sigma_{\rm{pt}}(\mathcal L_\infty) = \emptyset$. Furthermore, arguing as in the proof of Kapitula and Promislow [@kapitula2013spectral Theorem 3.1.11] and invoking Remark \[remark:spaces\], we have the following theorem which states that $\lambda \in \sigma_{\rm{ess}}(\mathcal L)$ precisely when the matrix $\mathbf A_\infty(\lambda)$ has a pure imaginary eigenvalue. \[thm:EssSpec\] Assume that Hypothesis \[hyp:EA\] is met. Then, $\sigma_{\rm{pt}}(\mathcal L_\infty) = \emptyset$ and $$\sigma_{\rm{ess}}(\mathcal L_\infty) \,\,=\,\, \{ \lambda \in \mathbb C\, |\, \exists \mu\in \mathbb R: \operatorname{det} [\mathbf{A}_\infty(\lambda) - i\mu]=0 \}.$$ Furthermore, the differential operator, $\mathcal L=\mathcal L(t)$, given in is a relatively compact perturbation of the asymptotic operator, $\mathcal L_\infty$, defined in , and $$\sigma_{\rm{ess}}(\mathcal L(t)) \,\,=\,\, \sigma_{\rm{ess}}(\mathcal L_\infty).$$ In particular, the essential spectrum of $\mathcal L(t)$ is $t$-independent. \[cor:EssSpec\] $$\sigma_{\rm{ess}}(\mathcal L) \,\,=\,\, \{ \lambda \in \mathbb C\, |\, \exists \mu\in \mathbb R: \operatorname{det}[\lambda - \mathbf{M}_\infty + \mu^2 \mathbf{B}]=0 \}.$$ Premultiplying $\mathbf{A}_\infty(\lambda) - i\mu$ by the invertible matrix $\begin{bmatrix} \mathbf 0 & \mathbf B \\ \mathbf B & \mathbf 0 \end{bmatrix}$, and applying the Schur determinental formula [@meyer2000matrix], we find that $\lambda \in \sigma_{\rm{ess}}(\mathcal L)$ if and only if $$0 \,\,=\,\, \operatorname{det} \begin{bmatrix} \lambda - \mathbf{M}_\infty & -i\mu \mathbf B \\ -i\mu\mathbf{B} & \mathbf{B} \end{bmatrix} \,\,=\,\,\operatorname{det}\mathbf{B} \operatorname{det}[\lambda - \mathbf{M}_\infty + \mu^2 \mathbf{B}].$$ For the KM breather discussed in Example \[ex:KM\], $$\mathbf{M}_\infty = \begin{bmatrix}0&0\\2\nu_0^2&0\end{bmatrix}.$$ Consequently, $$\sigma_{\rm{ess}}(\mathcal L) \,\,=\,\, \{ \lambda \in \mathbb C \,\,|\,\, \lambda = \pm i \tfrac \mu 2 \sqrt{\mu^2-4\nu_0} \text{ for some } \mu\in\mathbb R \},$$ which is the union of the imaginary axis and the interval $[-\nu_0,\nu_0]$ in the real axis. This result is consistent with a result of Cuevas-Maraver *et al.* [@cuevas2017floquet], who performed a modulation instability analysis to show that the frequency, $\omega$, and wave number, $k$, of a perturbation about a plane wave background, $\psi_\infty(t,x)=\nu_0$, satisfy the dispersion relation $\omega = \pm \frac k 2 \sqrt{k^2-4\nu_0}$. For the CQ-CGL breathers discussed in Example \[ex:GL\], $\mathbf{M}_\infty = \delta \mathbf{I}$. Consequently, $$\sigma_{\rm{ess}}(\mathcal L) \,\,=\,\, \{ \lambda \in \mathbb C \,\,|\,\, \lambda = \delta - \mu^2 (\beta \pm i D/2) \text{ for some } \mu\in\mathbb R \},$$ which is a pair of half-lines in the complex plane that are symmetric about the real axis [@PhysicaD116p95; @shen2016spectra]. If the physical system being modeling includes linear loss ($\delta < 0$) and a spectral filter ($\beta>0$), then the essential spectrum is stable. The Evolution Family {#Sec:EvolFamily} ==================== In this section, we establish the existence of an evolution family for the linearized equation  by applying classical results on the existence, uniqueness, and regularity of solutions of initial-value problems for non-autonomous linear differential equations in Banach spaces. In Section \[Sec:EssSpecM\], the evolution family will be used to define the monodromy operator. We study solutions, $\mathbf u:[s,\infty) \to H^2(\mathbb R,\mathbb C^2)$, of the initial-value problem $$\begin{aligned} \frac{\partial \mathbf u}{\partial t} \,\,&=\,\, \mathcal L(t) \mathbf u, \qquad \text{for } t>s,\\ \mathbf u(s) &= \mathbf v, \end{aligned} \label{eq:IVP}$$ where $\mathcal L(t) = \mathbf B \,\partial_x^2 + \widetilde{\mathbf M}(t)$ is the $t$-dependent family of operators on $L^2(\mathbb R, \mathbb C^2)$ defined in , and $\mathbf v \in H^2(\mathbb R,\mathbb C^2)$. The existence of solutions of linear differential equations in Banach spaces is typically established using semigroup theory. Since the operator, $\mathcal L(t)$, is $t$-dependent, we utilize the theory of evolution families [@pazy2012semigroups], in which the solution of  is represented in the form, $\mathbf u(t) = \mathcal U(t,s)\mathbf v$, where $\mathcal U(t,s)$ is an evolution operator. The following theorem establishes conditions on a solution, $\psi=\psi(t,x)$, of the nonlinear wave equation  that ensure the existence of an evolution operator, $\mathcal U(t,s)$, for the linearized equation, . \[hyp:PsiEvOp\] The $T$-periodic solution, $\psi=u+iv$, about which is linearized satisfies 1. \[hyp:PsiEvOp1\] $\psi$ is bounded and continuous on $[0,T]\times \mathbb R$; 2. \[hyp:PsiEvOp2\] $\psi_{tt}$ exists and is bounded on $[0,T]\times \mathbb R$. In the sequel, for each Banach space, $X$, we let $\| \cdot \|_X$ denote the standard norm on $X$, and we make the following definition. Let $\mathbf A =\mathbf A(t,x): [0,\infty)\times \mathbb R \to \mathbb C^{2\times 2}$ be a bounded matrix-valued function. We define $$\| \mathbf A \|_\infty \,\,:=\,\, \sup\limits_{(t,x)} \| \mathbf A (t,x)\|_{\mathbb C^{2\times 2}},$$ where $\| \cdot \|_{\mathbb C^{2\times 2}}$ denotes the matrix norm induced from the Euclidean norm $\| \cdot \|_{\mathbb C^2}$. \[thm:EvOp\] Assume that Hypothesis \[hyp:PsiEvOp\] is met. Then, there exists a unique evolution operator, $\mathcal U(t,s) \in \mathcal B(L^2(\mathbb R, \mathbb C^2))$, for $0\leq s \leq t < \infty$, such that 1. $\| \mathcal U(t,s) \|_{\mathcal B(L^2(\mathbb R, \mathbb C^2))} \leq \exp [ \, \| \widetilde{\mathbf M} \|_\infty (t-s) \,] $, 2. $\mathcal U(t,s) (H^2(\mathbb R, \mathbb C^2)) \subset H^2(\mathbb R, \mathbb C^2)$, and 3. For each $s$, $\mathcal U(\cdot,s)$ is strongly continuous in that for all $\mathbf v\in L^2(\mathbb R, \mathbb C^2)$, the mapping $t\mapsto \mathcal U(t,s)\mathbf v$ is continuous, and 4. For each $\mathbf v \in H^2(\mathbb R, \mathbb C^2)$, the function $\mathbf u(t) = \mathcal U(t,s)\mathbf v$ is the unique solution of the initial value problem  for which $\mathbf u \in C([s,\infty),H^2(\mathbb R, \mathbb C^2))$ and $\mathbf u \in C^1((s,\infty),L^2(\mathbb R, \mathbb C^2))$. In the case that $\beta > 0$, for any $\mathbf v \in L^2(\mathbb R, \mathbb C^2)$ we have that $\mathbf u(t) = \mathcal U(t,s)\mathbf v \in H^2(\mathbb R, \mathbb C^2)$. However, when $\beta = 0$, we require that $\mathbf v \in H^2(\mathbb R, \mathbb C^2)$ in order that $\mathbf u(t) \in H^2(\mathbb R, \mathbb C^2)$. The theorem follows from a combination of the Hille-Yosida Theorem (see Theorem 3.1 of Pazy [@pazy2012semigroups Ch. 1]) and Theorems 2.3 and 4.8 of [@pazy2012semigroups Ch. 5]. The following three lemmas ensure that the hypotheses of these three theorems hold. \[lemma:HY\] The linear operator, $\mathbf B \partial_{x}^2: H^2(\mathbb R,\mathbb C^2) \subset L^2(\mathbb R,\mathbb C^2) \to L^2(\mathbb R,\mathbb C^2)$ is closed with domain $H^2(\mathbb R,\mathbb C^2)$. Furthermore, $\mathbb R^+ \subset \rho( \mathbf B \partial_{x}^2 )$ and the resolvent operator (see Definition \[SpecDefs\]) satisfies $$\label{eq:HYResBound} \| R(\lambda : \mathbf B \partial_{x}^2)\|_{\mathcal B(L^2(\mathbb R,\mathbb C^2))} \,\,\leq \,\, \frac 1\lambda \qquad\text{for all } \lambda > 0.$$ Consequently, by the Hille-Yosida Theorem, $\mathbf B \partial_{x}^2$ is the infinitesimal generator of a $C_{0}$-semigroup of contractions on $L^2(\mathbb R,\mathbb C^2)$. The closedness of the operator $\mathbf B\partial_x^2$ is discussed in Remark \[remark:spaces\]. By Theorem \[thm:EssSpec\], $\sigma(\mathbf B\partial_x^2) = \{ s(\beta \pm iD/2) \, | \, s\leq 0\}$, and so $\mathbb R^+ \subset \rho(\mathbf B\partial_x^2)$. To establish the bound  on the norm of the resolvent, we use the Fourier transform, $\mathcal F: L^2(\mathbb R,\mathbb C^2) \to L^2(\mathbb R,\mathbb C^2)$, defined by $$\widehat f(\xi)\,\,=\,\,{\mathcal F}[f](\xi)\,\,=\,\, \int_{\mathbb R} f(x) \exp(-2\pi i x\xi)\,dx.$$ Now, for $\mathbf v \in L^2(\mathbb R,\mathbb C^2)$ $$\mathcal F[R(\lambda: \mathbf B \partial_x^2)\mathbf v] (\xi) \,\,=\,\, \mathbf C(\xi) \widehat{\mathbf v}(\xi),$$ where $$\mathbf C(\xi) \,\,=\,\, ( -4\pi\xi^2 \mathbf B - \lambda)^{-1} \,\,=\,\, \frac {1}{d(\xi)} \begin{bmatrix} -4\pi\xi^2\beta - \lambda & -4\pi \xi^2 D/2 \\ 4\pi \xi^2 D/2 & -4\pi\xi^2\beta - \lambda \end{bmatrix},$$ with $d(\xi) = (4\pi \xi^2\beta + \lambda)^2 + (4\pi\xi^2 D/2)^2$. By Parseval’s Theorem, $$\begin{aligned} \| R(\lambda:\mathbf B \partial_x^2)\mathbf v\|_{L^2(\mathbb R,\mathbb C^2)}^2 \,\,&=\,\, \| \mathbf C\widehat{\mathbf v}\|_{L^2(\mathbb R,\mathbb C^2)}^2 \\ \,\,&\leq\,\, \int_{\mathbb R} \| \mathbf C(\xi)\|_{\mathbb C^{2 \times 2}}^2 \| \widehat{\mathbf v}(\xi)\|_{\mathbb C^2}^2\, d\xi \,\,\leq\,\, \| \mathbf C \|_\infty^2 \| {\mathbf v}\|_{L^2(\mathbb R,\mathbb C^2)}^2,\end{aligned}$$ where $\| \mathbf C \|_\infty := \sup\limits_{\xi\in\mathbb R} \| \mathbf C(\xi)\|_{\mathbb C^{2\times 2}}$. Since $\| \mathbf C(\xi)\|_{\mathbb C^{2\times 2}}^2$ is equal to the largest eigenvalue of $\mathbf C(\xi)^T \mathbf C(\xi) = d(\xi)^{-1}\mathbf I_{2\times 2}$, we conclude that $\| \mathbf C(\xi)\|_{\mathbb C^{2\times 2}} = d(\xi)^{-1/2} \leq \lambda^{-1}$, for all $\lambda > 0$. Since $\beta>0$, we conclude that $\| R(\lambda: \mathbf B \partial_{x}^2)\|_{\mathcal B(L^2(\mathbb R,\mathbb C^2))} \,\,\leq \,\,\lambda^{-1}$, as required. \[lemma:Mbound\] Assume that Hypothesis \[hyp:PsiEvOp\] is met. Then, for all $t>0$, $$\| \widetilde{\mathbf M}(t) \|_{\mathcal B(L^2(\mathbb R,\mathbb C^2))} \,\,\leq \,\, \| \widetilde{\mathbf M} \|_\infty \,\, < \,\, \infty.$$ Let $\mathbf u\in L^2(\mathbb R,\mathbb C^2)$. Then, $$\begin{aligned} \| \widetilde{\mathbf M}(t) \mathbf u \|_{L^2(\mathbb R,\mathbb C^2)}^2 %%\,\,&=\,\, \int_{\mathbb R} \| \widetilde{\mathbf M}(t)\mathbf u(x)\|^2_{\mathbb C^2}\, dx \,\,\leq \,\, \int_{\mathbb R} \| \widetilde{\mathbf M}(t,x)\|_{\mathbb C^{2\times 2}}^2 \|\mathbf u(x)\|^2_{\mathbb C^2}\, dx %\\ %& \,\,\leq\,\, \| \widetilde{\mathbf M}\|^2_\infty\, \| \mathbf u\|^2_{L^2(\mathbb R,\mathbb C^2)}.\end{aligned}$$ To show that $\| \widetilde{\mathbf M}\|_\infty < \infty$ we use Hypothesis \[hyp:PsiEvOp\](1) together with the fact that the matrix 2-norm is bounded by the Frobenius norm, $\| \mathbf A \|_{\mathbb C^{2\times 2}} \leq \| \mathbf A \|_F$, where $\| \mathbf A \|_F^2:=\sum\limits_{i,j=1}^2 |A_{ij}|^2$. Taken together, Theorem 2.3 of [@pazy2012semigroups Ch. 5] and Lemmas \[lemma:HY\] and \[lemma:Mbound\] imply that $\{\mathcal L(t)\}_{t\geq 0}$ is a stable family of infinitesimal generators of $C_0$-semigroups on $L^2(\mathbb R,\mathbb C^2)$, which is one of the assumptions required for the application of Theorem 4.8 of Pazy [@pazy2012semigroups Ch. 5]. The remaining assumption in that theorem also holds thanks to the following lemma. \[lemma:LisC1\] Assume that Hypothesis \[hyp:PsiEvOp\] is met. Then, for each $\mathbf v\in H^2(\mathbb R,\mathbb C^2)$, we have that $F(t) := \mathcal L(t)\mathbf v: (0,\infty)\to L^2(\mathbb R,\mathbb C^2)$ is $C^1$. We show that $F$ is differentiable with $F'(t) = \partial_t\widetilde{\mathbf M}(t)\mathbf v$. The proof that $F'$ is continuous is similar. Applying the Mean Value Theorem twice, we find that $$\begin{aligned} &\| F(t+h) - F(t) - h F'(t) \|_{L^2(\mathbb R,\mathbb C^2)}^2 \\ &\leq\,\, \int_{\mathbb R} \| \widetilde{\mathbf M}(t+h,x) - \widetilde{\mathbf M}(t,x) - h \partial_t\widetilde{\mathbf M}(t,x) \|_{\mathbb C^{2\times 2}}^2 \|\mathbf v(x)\|_{\mathbb C^2}^2\, dx \\ &\leq\,\, h^2\, \int_{\mathbb R} \| \partial_t\widetilde{\mathbf M}(t_*(x),x) - \partial_t\widetilde{\mathbf M}(t,x) \|_{\mathbb C^{2\times 2}}^2 \|\mathbf v(x)\|_{\mathbb C^2}^2\, dx \quad\text{with } t_*(x) \in (t,t+h) \\ &\leq\,\, h^4\, \int_{\mathbb R} \| \partial_t^2\widetilde{\mathbf M}(t_{**}(x),x) \|_{\mathbb C^{2\times 2}}^2 \|\mathbf v(x)\|_{\mathbb C^2}^2\, dx \quad\text{with } t_{**}(x) \in (t,t+h).\end{aligned}$$ Therefore, $$\| F(t+h) - F(t) - h F'(t) \|_{L^2(\mathbb R,\mathbb C^2)} \,\,\leq\,\, h^2 \| \partial_t^2\widetilde{\mathbf M}\|_\infty \| \mathbf v\|_{L^2(\mathbb R,\mathbb C^2)},$$ where $ \| \partial_t^2\widetilde{\mathbf M}\|_\infty < \infty$ by Hypothesis \[hyp:PsiEvOp\](2). The essential spectrum of the monodromy operator \[Case: $\beta>0$\] {#Sec:EssSpecM} ==================================================================== In this section, we define the monodromy operator that is the main focus of this paper, and derive a formula for its essential spectrum in terms of that of the asymptotic operator, $\mathcal L_\infty$. Our proof relies on the fact that the semigroup, $e^{t \mathcal L_\infty}$, associated with $\mathcal L_\infty$ is analytic, which only holds when $\beta>0$. Therefore, although the results in this section apply to the CQ-CGL breathers in Example \[ex:KM\], they do not apply to the KM breather in Example \[ex:GL\]. Let $\psi$ be a periodically-stationary solution of the constant coefficient CQ-CGL equation  and let $s\in\mathbb R$. The *monodromy operator*, $\mathcal M(s)$, associated with the linearization of about $\psi$ is the bounded operator on $L^2(\mathbb R,\mathbb C^2)$ defined by $\mathcal M(s)=\mathcal U(s+T,s)$, where $\mathcal U$ is the evolution operator of Theorem \[thm:EvOp\] and $T$ is the period of $\psi$. \[thm:EssSpecU\] Suppose that $\beta > 0$ and Hypothesis \[hyp:PsiEvOp\] is met. Then the $C^0$-semigroup, $e^{t\mathcal L_\infty}$, generated by $\mathcal L_\infty$ is analytic. Furthermore, the essential spectrum of the evolution operator, $\mathcal U(t,s)$, is given by $$\sigma_{\rm{ess}}(\mathcal U(t,s)) \,\,=\,\, \sigma_{\rm{ess}}(e^{(t-s)\mathcal L_\infty}).$$ Therefore, the essential spectrum of the monodromy operator, $\mathcal M(s)$, is given by $$\sigma_{\rm{ess}}(\mathcal M(s)) \,\,=\,\, \sigma_{\rm{ess}}(e^{T\mathcal L_\infty}),$$ which is independent of $s$. Before proving this theorem, we state and prove a Corollary. \[cor:EssSpecM\] Suppose that $\beta > 0$ and Hypothesis \[hyp:PsiEvOp\] is met. Then, the essential spectrum of the monodromy operator, $\mathcal M(s)$, is given by $$\sigma_{\rm{ess}}(\mathcal M(s)) \setminus \{0\} \,\,=\,\, e^{T\sigma_{\rm{ess}}(\mathcal L_\infty)}.$$ In particular, in the case of a CQ-CGL breather, if $\delta < 0$ then the essential spectrum lies inside a circle of radius $e^{\delta T} < 1$. Since the spectral mapping theorem holds for the point spectrum of a $C_0$-semigroup [@engelnagel2000], $\sigma_{\rm{pt}}(e^{T\mathcal L_\infty}) \setminus \{ 0\} = e^{T \sigma_{\rm{pt}} (\mathcal L_\infty)} = \emptyset $, by Theorem \[thm:EssSpec\]. Since $\sigma^{\mathcal F}_{\rm{pt}} \subset \sigma_{\rm{pt}}$, $\sigma^{\mathcal F}_{\rm{pt}}(e^{T\mathcal L_\infty}) \setminus \{0\}= \emptyset$. Consequently, $\sigma_{\rm{ess}}(e^{T\mathcal L_\infty}) \setminus \{ 0\} = \sigma(e^{T\mathcal L_\infty}) \setminus \{ 0\} = e^{T \sigma (\mathcal L_\infty)}$, since the spectral mapping theorem also holds for the spectrum of an analytic semigroup [@engelnagel2000]. The result now follows since $\sigma (\mathcal L_\infty) = \sigma_{\rm{ess}} (\mathcal L_\infty)$. The proof of Theorem \[thm:EssSpecU\] relies on the following two lemmas. Suppose that Hypothesis \[hyp:PsiEvOp\] is met. Then $$\mathcal U(t,s) \,\,=\,\, e^{(t-s)\mathcal L_\infty} \,\,+\,\, \int_s^t e^{(t-\tau)\mathcal L_\infty} \,\mathbf M(\tau) \, \mathcal U(\tau,s)\, d\tau, \qquad\text{in } \mathcal B(L^2(\mathbb R,\mathbb C^2)). \label{eq:VarOfParU}$$ We refer to Engel and Nagel [@engelnagel2000 App. C] for a summary of the theory of Lebesgue integration for functions, $f:J\to X$, from an interval $J\subset \mathbb R$ to a Banach space, $X$. As in the proof of Theorem \[thm:EvOp\], the asymptotic operator, $\mathcal L_\infty$, generates a $C_0$-semigroup on $L^2(\mathbb R, \mathbb C^2)$. For each $\tau\in[s,t]$ and $\mathbf v\in H^2(\mathbb R,\mathbb C^2)$, let $f(\tau) := \mathbf M(\tau) \,\mathcal U(\tau,s)\mathbf v \in L^2(\mathbb R,\mathbb C^2)$. By Hypothesis \[hyp:PsiEvOp\], $f\in L^1([s,t], L^2(\mathbb R,\mathbb C^2))$. Since $\mathcal L(t) = \mathcal L_\infty + \mathbf M(t)$, the result follows from the variation of parameters formula (see Corollary 2.2 of [@pazy2012semigroups Ch. 4], together with Theorem \[thm:EvOp\] above). Corollary 10.6 of [@pazy2012semigroups Ch. 1] implies that $(e^{t\mathcal L_\infty})^* = e^{t\mathcal L_\infty^*}$, where $\mathcal L_\infty^* = \mathbf B^T\,\partial^2_x + \mathbf M_\infty^T$ is the adjoint of $\mathcal L_\infty$. \[lemma:AnalSemiGp\] Suppose that $\beta > 0$ and Hypothesis \[hyp:PsiEvOp\] is met. Then the semigroups $e^{t\mathcal L_\infty}$ and $e^{t\mathcal L_\infty^*}$ are analytic. We will show that for all $\sigma >0$ and $\tau\neq 0$, $$\| R(\sigma + i \tau : \mathbf B \partial_x^2) \|\,\, \leq \,\,\frac{\sqrt{1+(D/2\beta)^2}}{|\tau|}. \label{eq:ASGResolventCond}$$ Therefore, by Theorem 5.2 of [@pazy2012semigroups Ch. 2] (and the discussion preceeding it), $\mathbf B \partial_x^2$ is the infinitesimal generator of an analytical semigroup. Since $\mathcal L_\infty = \mathbf B \partial_x^2 + \mathbf M_\infty$, where $ \mathbf M_\infty$ is a bounded operator, it follows from Corollary 2.2 of [@pazy2012semigroups Ch. 3] that $\mathcal L_\infty$ is the infinitesimal generator of an analytical semigroup. The same argument holds for the adjoint. Note that as $\beta \to 0$, the constant on the right-hand side of blows up. Consequently, this proof cannot be extended to the case $\beta=0$. To establish , we observe that as in the proof of Lemma \[lemma:HY\], $$\begin{aligned} \| R(\sigma + i \tau : \mathbf B \partial_x^2) \|^2 \,\,&\leq\,\, \sup_{\xi\in \mathbb R}\mu_{\text{max}} [\mathbf C(\xi)^* \mathbf C(\xi)] \\ \,\,&=\,\, \left( \inf_{\xi\in \mathbb R} \mu_{\text{min}} [(-4\pi\xi^2\mathbf B - \lambda) ( -4\pi\xi^2\mathbf B^T - \overline{\lambda})]\right)^{-1},\end{aligned}$$ since the largest eigenvalue of a non-negative definite Hermitian matrix is the inverse of the smallest eigenvalue of its inverse. Let $a= 4\pi\xi^2\beta>0$ and $b= 2\pi\xi^2 D$. A calculation shows that $\mu_\text{min} = |a+\lambda|^2 + b^2 - 2|b\tau|$. If we let $z= -a +i\operatorname{sgn}(\tau b) b$, then $\mu_\text{min} = |z-\lambda|^2$. The points, $z$, lie on the half-line in the left-half plane given by $y= - \operatorname{sgn}(\tau) mx$ with $x<0$, where $m= | D/2\beta |$, while $\lambda$ is in the right half-plane. We observe that $z$ and $\lambda$ either both lie above or both lie below the real axis. Consequently, the problem of minimizing $\mu_\text{min}$ as a function of $\xi$ is that of finding the minimum of the square of the distance from $\lambda$ to this half-line, which is greater than the minimum squared distance to the full line. Therefore, $$\| R(\sigma + i \tau : \mathbf B \partial_x^2) \| \,\,\leq \,\,\frac{\sqrt{1+m^2}}{m\sigma + |\tau|} \,\,\leq\,\, \frac{\sqrt{1+m^2}}{|\tau|},$$ as required. *Proof of Theorem \[thm:EssSpecU\]* Since the essential spectrum is invariant under compact perturbations, to prove the theorem, we just need to show that the integral in is a compact operator. By Engel and Nagel [@engelnagel2000 Theorem C.7], it suffices to show that for each $\tau\in[s,t]$ the integrand $\mathcal K(\tau) = e^{(t-\tau)\mathcal L_\infty} \,\mathbf M(\tau) \, \mathcal U(\tau,s)$ is compact, and that the function $\mathcal K:[s,t]\to \mathcal B(L^2(\mathbb R,\mathbb C^2))$ is strongly continuous, in that for all $\mathbf v \in L^2(\mathbb R,\mathbb C^2)$, $\|\mathcal K(\tau)\mathbf v-\mathcal K(\tau_0)\mathbf v\|_{L^2(\mathbb R,\mathbb C^2)} \to 0$ as $\tau\to\tau_0$. To show that $\mathcal K(\tau)$ is compact we show that the adjoint, $\mathcal K^*(\tau)$, is compact. As in Theorem \[thm:EssSpec\], $\mathcal L^*(\tau)$ is a relatively compact perturbation of $\mathcal L_\infty^*$. Therefore there is a $\lambda \in \rho(\mathcal L_\infty^*)$ so that $\mathbf M^*(\tau) (\mathcal L_\infty^* - \lambda)^{-1} \in \mathcal B(L^2(\mathbb R,\mathbb C^2))$ is compact. Since the semigroup $e^{t\mathcal L_\infty^*}$ is analytic, the operator $(\mathcal L_\infty^*-\lambda) e^{(t-\tau)\mathcal L_\infty^*}$ is bounded (see Theorem 5.2 of [@pazy2012semigroups Ch. 2] together with the discussion preceding that result). Therefore the composition, $$\mathcal K^*(\tau) = \mathcal U^*(\tau,s) \,\mathbf M^*(\tau) (\mathcal L_\infty^* - \lambda)^{-1} (\mathcal L_\infty^* - \lambda) e^{(t-\tau)\mathcal L_\infty^*},$$ is also compact. Finally, by Hypothesis \[hyp:PsiEvOp\](1), the function $\mathbf M: [s,t]\to \mathcal B(L^2(\mathbb R,\mathbb C^2))$ is uniformly bounded and strongly continuous. Furthermore, by Theorem \[thm:EvOp\], $\mathcal U(\cdot,s)$ is uniformly bounded and strongly continuous for each $s$. Therefore, $\mathcal K(\tau)$ is strongly continuous, since the composition of uniformly bounded, strongly continuous functions is strongly continuous. [^1]: J.Z. was supported by the NSF under grant DMS 1620293 and thanks the Department of Mathematics at UNC Chapel Hill for hosting his Fall 2018 sabbatical, during which this work began. [^2]: Y.L. was supported by the NSF under grant DMS 1710989 and thanks the Courant Institute of Mathematical Sciences and, especially, Prof. Lai-Sang Young, for the opportunity to visit the Institute where this work was conducted. [^3]: J.L.M. was supported in part by NSF CAREER Grant DMS 1352353 and NSF Grant DMS 1909035. [^4]: C.K.R.T.J. was supported by the US Office of Naval Research under grant N00014-18-1-2204.

ArXiv

--- abstract: 'We determine the structure over ${\mathbb{Z}}$ of the ring of symmetric Hermitian modular forms with respect to $\mathbb{Q}(\sqrt{-1})$ of degree $2$ (with a character), whose Fourier coefficients are integers. Namely, we give a set of generators consisting of $24$ modular forms. As an application of our structure theorem, we give the Sturm bounds of such the modular forms of weight $k$ with $4\mid k$, in the case $p=2$, $3$. We remark that the bounds for $p\ge 5$ are already known.' author: - Toshiyuki Kikuta title: A ring of symmetric Hermitian modular forms of degree $2$ with integral Fourier coefficients --- [**2010 Mathematics subject classification**]{}: Primary 11F30 $\cdot$ Secondary 11F55\ [**Key words**]{}: ring of modular forms, Hermitian modular forms, generators. Introduction ============ Let $e_4$, $e_6$ be the normalized Eisenstein series of respective weight $4$, $6$ for $\Gamma _1:=SL_2(\mathbb{Z})$ and $\delta $ the Ramanujan delta function defined by $\delta =2^{-6}\cdot 3^{-3}(e_4^3-e_6^2)$. For the ${\mathbb{Z}}$-module $M_k(\Gamma _1;{\mathbb{Z}})$ consisting of modular forms of weight $k$ for $\Gamma _1$ whose Fourier coefficients are in ${\mathbb{Z}}$, we define an algebra over ${\mathbb{Z}}$ as $$A(\Gamma _1;{\mathbb{Z}}):=\bigoplus_{k\in {\mathbb{Z}}}M_k(\Gamma _1;{\mathbb{Z}}).$$ Then it is well-known as a classical result that all the Fourier coefficients of the modular forms $e_4$, $e_6$, $\delta $ are integers and they generate $A(\Gamma _1;{\mathbb{Z}})$. Namely we have $$A(\Gamma _1;{\mathbb{Z}})={\mathbb{Z}}[e_4,e_6,\delta ].$$ In the case of Siegel modular forms for the symplectic group $\Gamma _2:=Sp_2(\mathbb{Z})$ of degree $2$, there is a famous result of Igusa. He showed such the ring over ${\mathbb{Z}}$ are generated by $15$ modular forms. He showed also that its set of generators are minimal. In this paper, we consider the ring of symmetric Hermitian modular forms of degree $2$ with respect to $\mathbb{Q}(\sqrt{-1})$ whose Fourier coefficients are in $\mathbb{Z}$. Since it seems to be difficult to give generators of the full space of them, we restrict our selves to the case where the weights are multiples of $4$. We remark that, the ring of Siegel modular forms whose weights are multiples of $4$ are generated over $\mathbb{Z}$ by $23$ modular forms. This is an easy conclusion of Igusa’s result. In our case, there exists a set of generators consisting of $24$ modular forms whose weights are $$\begin{aligned} &4,\ 8,\ 12,\ 12,\ 12,\ 16,\ 16,\ 20,\ 24,\ 24,\ 28,\ 28,\ 32,\\ &36,\ 36,\ 36,\ 40,\ 40,\ 48,\ 48,\ 52,\ 60,\ 60,\ 72,\ 84. \end{aligned}$$ The precise statement can be found in Theorem \[Thm1\]. We construct explicitly these generators in Subsection \[Const\]. As an application of this result, we can obtain the Sturm bounds for $p=2$, $3$, in the Hermitian modular forms whose weights are multiples of $4$ (Theorem \[Thm2\]). We remark that the Sturm bounds for $p\ge 5$ are already known in [@Ki-Na2]. Preliminaries {#sec:4} ============= Hermitian modular forms of degree $2$ {#sec:4.1} ------------------------------------- We deal with Hermitian modular forms of degree $2$ only for ${\boldsymbol K}:=\mathbb{Q}(\sqrt{-1})$. Let ${\mathcal O}$ be the ring of Gauss integers, i.e. ${\mathcal O}=\mathbb{Z}[\sqrt{-1}]$. Let $\mathbb{H}_2$ be the Hermitian upper half-space of degree $2$ defined as $$\mathbb{H}_2:=\{ Z\in M_2(\mathbb{C})\;|\; \tfrac{1}{2i}(Z-{}^t\overline{Z})>0\; \}$$ where ${}^t\overline{Z}$ is the transposed complex conjugate of $Z$. The Hermitian modular group of degree $2$ $$U_2(\mathcal{O}):=\left\{\;M\in M_{4}(\mathcal{O})\;|\; {}^t\overline{M}J_2M=J_2,\; J_2=\binom{\;0_2\;\,-1_2}{1_2\;\;\;0_2}\right\}$$ acts on $\mathbb{H}_2$ by fractional transformation $$\mathbb{H}_2\ni Z\longmapsto M\langle Z\rangle :=(AZ+B)(CZ+D)^{-1}, \;M=\begin{pmatrix} A & B \\ C & D\end{pmatrix}\in U_2(\mathcal{O}).$$ We denote by $M_k(U_2({\mathcal O}))=M_k(U_2({\mathcal O}),{\det}^k)$ the space of symmetric Hermitian modular forms of weight $k$ and character ${\det}^k$ with respect to $U_2({\mathcal O})$. (We deal with modular forms with character ${\det }^{k}$, but we drop this in the notation). Namely, it consists of holomorphic functions $F:\mathbb{H}_2\longrightarrow \mathbb{C}$ satisfying $$F\mid_kM(Z):={\det}(CZ+D)^{-k}F(M\langle Z\rangle )={\det (M)}^k\cdot F(Z),$$ for all $M=\begin{pmatrix}A & B \\ C & D\end{pmatrix} \in U_2({\mathcal O})$ and $F({}^tZ)=F(Z)$. Note that one has $M_k(U_2({\mathcal O}))=\{0\}$ if $k$ is odd. The cusp forms are characterized by the condition $$\Phi \Big(F\mid_k\binom{\!{}^t\overline{U}\;0}{0\;\;U}\Big) \equiv 0\quad \text{for}\; \text{all}\; U\in GL_2(\mathbb{Q}(\sqrt{-1}))$$ where $\Phi$ is the Siegel $\Phi$-operator. We denote by $S_k(U_2({\mathcal O}))$ the subspace consisting of all cusp forms in $M_k(U_2({\mathcal O}))$. Fourier expansion {#sec:2.2} ----------------- Since all of $F$ in $M_k(U_2({\mathcal O}))$ satisfies the condition $$F(Z+B)=F(Z) \quad \text{for}\;\text{all}\; B\in Her_2(\mathcal{O}),$$ it has a Fourier expansion of the form $$F(Z)=\sum_{0\leq H\in\Lambda_2(\boldsymbol{K})}a_F(H)e^{2\pi i\text{tr}(HZ)},$$ where $$\Lambda_2(\boldsymbol{K}):=\{ H=(h_{ij})\in Her_2(\boldsymbol{K})\; |\; h_{ii}\in\mathbb{Z}, 2 h_{ij}\in\mathcal{O} \}.$$ We write $H=(m,r,s,n)$ for $H=\begin{pmatrix} m & \frac{r+si}{2} \\ \frac{r-si}{2} & n \end{pmatrix}\in \Lambda _2({\mathcal O})$ and also $a_F(m,r,s,n)$ for $a_F\begin{pmatrix}m & \frac{r+si}{2} \\ \frac{r-si}{2} & n \end{pmatrix}$ simply. Let $R$ be a subring of ${\mathbb{C}}$, we define $M_k(U_2({\mathcal O});R)$ as an $R$-module of all of $F\in M_k(U_2({\mathcal O}))$ such that $a_F(H)\in R$ for any $H\in \Lambda _2({\mathcal O})$. We put also $S_k(U_2({\mathcal O});R):=M_k(U_2({\mathcal O});R)\cap S_k(U_2({\mathcal O}))$. We put $$\begin{aligned} &\dot{q}_{11}:=\exp(2\pi i z_{11}),\quad \dot{q}_{22}:=\exp(2\pi i z_{22}), \\ &\dot{q}_{12}:=\exp\left( 2\pi i \frac{z_{12}-z_{21}}{-2i} \right),\quad \ddot{q}_{12}:=\exp\left(2\pi i \frac{z_{12}+z_{21}}{2}\right).\end{aligned}$$ Then for $H=(m,r,s,n)$ we have $$e^{2\pi i\text{tr}(HZ)}=\dot{q_{1}}^{m}\dot{q_{12}}^{r}\ddot{q_{12}}^{s}\dot{q_{2}}^{n}.$$ Then any element $F\in M_k(U_2({\mathcal O});R)$ can be regarded as an element of $$R[\![\dot{\boldsymbol{q}}]\!]:= R[\dot{q}_{12}^{\pm 1},\ddot{q}_{12}^{\pm }][\![\dot{q_{1}},\dot{q_{2}}]\!].$$ This notation is useful to calculate the Fourier expansion of Hermitian modular forms. We consider the Hermitian Eisenstein series of degree $2$ $$E_k(Z):=\sum_{M=\left(\begin{smallmatrix} * & * \\ C & D \end{smallmatrix}\right)} ({\det}M)^{\frac{k}{2}}{\det}(CZ+D)^{-k},\quad Z\in\mathbb{H}_2,$$ where $k>4$ is even and $M=\begin{pmatrix} * & * \\ C & D \end{pmatrix}$ runs over a set of representatives of $\left\{\begin{pmatrix} * & * \\ 0_2& * \end{pmatrix} \right\} \backslash U_2(\mathcal{O})$. Then we have $$E_k\in M_k(U_2(\mathcal{O})).$$ Moreover $E_4\in M_4(U_2(\mathcal{O}))$ is constructed by the Maass lift ([@Kri]). The Fourier coefficient of $E_k$ is given by the following formula: \[GHE\] The Fourier coefficient $a_{E_k}(H)$ of $E_k$ is given as follows. $$\begin{aligned} & a_{E_k}(H)\\ &=\begin{cases} 1 & \text{if}\;\; H=0_2,\\ \displaystyle -\frac{2k}{B_k}\,\sigma_{k-1}(\varepsilon (H)) & \text{if}\;\; {\rm rank}(H)=1,\\ \displaystyle \frac{4k(k-1)}{B_k\cdot B_{k-1,\chi_{-4}}}\sum_{0< d|\varepsilon (H)} d^{k-1} G_{\boldsymbol{K}}(k-2,4\,{\det}(H)/d^2) & \text{if}\;\; {\rm rank}(H)=2. \end{cases}\end{aligned}$$ where\ $B_m$ is the $m$-th Bernoulli number,\ $B_{m,\chi_{-4}}$ is the $m$-th generalized Bernoulli number associated with the Kronecker character $\chi_{-4}=\left(\frac{-4}{*}\right)$,\ $\varepsilon (H):={\rm max}\{ l\in\mathbb{N}\,|\, l^{-1}H\in \Lambda_2(\boldsymbol{K})\,\}$,\ and $$\label{GK} \begin{split} &G_{\boldsymbol{K}}(m,N):=\frac{1}{1+|\chi_{-4}(N)|} (\sigma_{m,\chi_{-4}}(N)- \sigma^*_{m,\chi_{-4}}(N))\\ &\sigma_{m,\chi_{-4}}(N):=\sum_{0< d|N}\chi_{-4}(d)d^m,\quad \sigma^*_{m,\chi_{-4}}(N):=\sum_{0< d|N}\chi_{-4}(N/d)d^m. \end{split}$$ We can construct cusp forms by the Hermitian Eisenstein series (cf. [@D-K], Corollary 2); $$\begin{aligned} &E_{10}-E_4E_6\in S_{10}(U_2(\mathcal{O})),\\ &E_{12}-\frac{441}{691}E_4^3-\frac{250}{691}E_6^2\in S_{12}(U_2(\mathcal{O})).\end{aligned}$$ Siegel modular forms of degree $2$ ---------------------------------- Let $M_k(\Gamma_2)$ denote the space of Siegel modular forms of weight $k$ $(\in\mathbb{Z})$ for the Siegel modular group $\Gamma_2:=Sp_2(\mathbb{Z})$ and $S_k(\Gamma_2)$ the subspace of cusp forms. Any Siegel modular form $F$ in $M_k(\Gamma_2)$ has a Fourier expansion of the form $$F(Z)=\sum_{0\leq T\in\Lambda_2}a_F(T)e^{2\pi i\text{tr}(TZ)},$$ where $$\Lambda_2=Sym_2^*(\mathbb{Z}) :=\{ T=(t_{ij})\in Sym_2(\mathbb{Q})\;|\; t_{ii},\;2t_{ij}\in\mathbb{Z}\; \}$$ (the lattice in $Sym_2(\mathbb{R})$ of half-integral, symmetric matrices). We write $T=(m,r,n)$ for $T=\begin{pmatrix}m & \frac{r}{2} \\ \frac{r}{2} & n \end{pmatrix}$ and also $a_F(m,r,n)$ for $a_F\begin{pmatrix}m & \frac{r}{2} \\ \frac{r}{2} & n\end{pmatrix}$. Taking $q_{ij}:=\text{exp}(2\pi iz_{ij})$ with $Z=(z_{ij})\in\mathbb{H}_2$, we have for $T=(m,r,n)$ $$e^{2\pi i\text{tr}(TZ)}=q_{11}^{m}q_{12}^{r}q_{22}^{n}.$$ For any subring $R\subset\mathbb{C}$, we adopt the notation, $$\begin{aligned} & M_k(\Gamma_2;R):=\{ F=\sum_{T\in\Lambda_n}a_F(T)q^T\;|\; a_F(T)\in R\;(\forall T\in\Lambda_2)\;\},\\ & S_k(\Gamma_n;R):=M_k(\Gamma_2)\cap S_k(\Gamma_2).\end{aligned}$$ Any element $F\in M_k(\Gamma_2;R)$ can be regarded as an element of $$R[\![\boldsymbol{q}]\!]:=R[q_{12}^{-1},q_{12}][\![ q_{11},q_{22}]\!].$$ The space $\mathbb{H}_2$ contains the Siegel upper half-space of degree $2$ $$\mathbb{S}_2:=\mathbb{H}_2\cap Sym_2(\mathbb{C}).$$ Hence we can define the restriction map $$\begin{aligned} R[\![\dot{\boldsymbol{q}}]\!]\longrightarrow R[\![\boldsymbol{q}]\!]\end{aligned}$$ via the correspondence $F\mapsto F|_{\mathbb{S}_2}:=F(z_{ij})|_{z_{21}=z_{12}}$ (this means $\dot{q}_{12}\mapsto 1$, $\ddot{q}_{12}\mapsto q_{12}$). In particular, if $F\in M_k(U_2({\mathcal O});R)\subset R[\![\dot{\boldsymbol{q}}]\!]$ then we have $F|_{\mathbb{S}_2}\in M_k(\Gamma _2;R)\subset R[\![\boldsymbol{q}]\!]$. This fact comes from each condition of the modularity. Igusa’s generators over $\mathbb{Z}$ {#sec:3.1} ------------------------------------ Let $k$ be an even integer with $k\ge 4$. The Siegel Eisenstein series $$G_k(Z):=\sum_{M=\left(\begin{smallmatrix}*&* \\ C & D\end{smallmatrix}\right)} {\det}(CZ+D)^{-k},\quad Z\in\mathbb{S}_2$$ defies an element of $M_k(\Gamma _2;\mathbb{Q})$. Here $M=\begin{pmatrix}* & * \\ C & D\end{pmatrix}$ runs over a set of representatives $\left\{\begin{pmatrix}* & * \\ 0_2 & * \end{pmatrix}\right\}\backslash\Gamma _2$. We write $X_4:=G_4$ and $X_6:=G_6$. We set $$\label{Siegel cusp} \begin{split} X_{10}:&=-\frac{43867}{2^{10}\cdot 3^5\cdot 5^2\cdot 7\cdot 53}(G_{10}-G_4G_6), \\ X_{12}:&=-\frac{691\cdot 1847}{2^{13}\cdot 3^6\cdot 5^3\cdot 7^2} (G_{12}-\frac{441}{691}G_4^3-\frac{250}{691}G_6^2). \end{split}$$ Then we have $X_k\in S_k(\Gamma_2;{\mathbb{Z}})$ $(k=10,12)$ and $a_{X_{10}}(1,1,1)=a_{X_{12}}(1,1,1)=1$. Let $k$ be an even integer with $k\ge 4$ and $G_k$ the normalized Siegel Eisenstein series of weight $k$. We set $$\begin{aligned} &Y_{12} := 2^{-6}\cdot 3^{-3}(X_4^3 - X_6^2)+2^4\cdot 3^2X_{12},\\ &X_{16} := 2^{-2}\cdot 3^{-1}(X_4X_{12} - X_6 X_{10}),\\ &X_{18} := 2^{-2}\cdot 3^{-1}(X_6 X_{12}-X_4^2X_{10}),\\ &X_{24} := 2^{-3}\cdot 3^{-1}(X_{12}^2 - X_4 X_{10}^2),\\ &X_{28} := 2^{-1}\cdot 3^{-1}(X_4 X_{24} - X_{10} X_{18}),\\ &X_{30} := 2^{-1}\cdot 3^{-1}(X_6 X_{24} - X_4 X_{10} X_{16}),\\ &X_{36} :=2^{-1}\cdot 3^{-2}(X_{12} X_{24} - X_{10}^2 X_{16}),\\ &X_{40} :=2^{-2}(X_4 X_{36} - X_{10} X_{30}),\\ &X_{42} := 2^{-2}\cdot 3^{-1}(X_{12} X_{30} - X_4 X_{10} X_{28}),\\ &X_{48} := 2^{-2}(X_{12}X_{36} - X_{24}^2).\end{aligned}$$ We write $$\begin{aligned} A^{(m)}(\Gamma _2;\mathbb{Z}):=\bigoplus _{k\in m\mathbb{Z}}M_k(\Gamma _2;{\mathbb{Z}}).\end{aligned}$$ The following structure theorem is due to Igusa. One has $X_k\in M_k(\Gamma _2;{\mathbb{Z}})$ ($k=4$, $6$, $\cdots $, $48$) and $Y_{12}\in M_{12}(\Gamma _2;{\mathbb{Z}})$ and the graded ring $A^{(2)}(\Gamma _2;\mathbb{Z})$ is generated over $\mathbb{Z}$ by them. Moreover, the set of $14$ generators are minimal. Actually, he determined the structure of the full space $A^{(1)}(\Gamma _2;{\mathbb{Z}})$ by using the cusp form of weight $35$. However, since we do not use this result we do not mention its detail. From his result, we have immediately the following property. \[Cor:S\_gen\] The ring $A^{(4)}(\Gamma _2;\mathbb{Z})$ is generated over $\mathbb{Z}$ by the following $23$ generators; $$\begin{aligned} &S_4:=X_4,\quad S_{12}:=X_{12},\quad T_{12}:=Y_{12},\quad U_{12}:=X_6^2,\quad S_{16}:=X_{10}X_6,\\ &T_{16}:=X_{16},\quad S_{20}:=X_{10}^2,\quad S_{24}:=X_{24},\quad T_{24}:=X_6X_{18},\\ &S_{28}:=X_{28},\quad T_{28}:=X_{10}X_{18},\quad S_{36}:=X_{36},\quad T_{36}:=X_{18}^2,\\ &U_{36}:=X_6X_{30},\quad S_{40}:=X_{40},\quad T_{40}:=X_{10}X_{30},\quad S_{48}:=X_{48},\\ &T_{48}:=X_{18}X_{30},\quad S_{52}:=X_{42}X_{10},\quad S_{60}:=X_{30}^2,\quad T_{60}:=X_{18}X_{42},\\ &S_{72}:=X_{30}X_{42},\quad S_{84}:=X_{42}^2.\end{aligned}$$ For later use, we introduce the Sturm bounds for Siegel modular forms of degree $2$. \[Stbd0\] Let $k$ be a positive integer and $p$ an any prime. Let $F\in M_{k}(\Gamma _2;\mathbb{Z}_{(p)})$. Suppose that $a_F(m,r,n)\equiv 0$ mod $p$ for any $m$, $r$, $n$ with $$m,\ n\le [k/10]$$ and $4mn-r^2\ge 0$. Then we have $F\equiv 0$ mod $p$. Structure over ${\mathbb{Z}}[1/2,1/3]$ -------------------------------------- We set $H_{4}:=E_4$ and $$\begin{aligned} &H_8 :=-\frac{61}{2^{10}\cdot 3^2 \cdot 5^2} (E_{8}-H_4^2),\\ &F_{10} := -\frac{277}{2^9\cdot 3^3\cdot 5^2 \cdot 7} (E_{10}-H_4 \cdot E_6),\\ &H_{12} := -\frac{19\cdot 691\cdot 2659}{2^{11}\cdot 3^7\cdot 5^3\cdot 7^2\cdot 73},\\ &~~~~~~~~~\times \left(E_{12}- \frac{3^2\cdot 7^2}{691}H _4^3 -\frac{2\cdot 5^3}{691}H _6^2 + \frac{2^9\cdot 3^4\cdot 5^2\cdot 7^2\cdot 6791}{19\cdot 691\cdot 2659}H _4\cdot H _8\right).\end{aligned}$$ We define the graded ring $A^{(m)}(U_2({\mathcal O});R)$ over $R$ by $$A^{(m)}(U_2({\mathcal O});R)=\bigoplus _{k\in m\mathbb{Z}}M_k(U_2({\mathcal O});R).$$ \[Thm:Ki-Na\] Then all of $H_4$, $E_6$, $H_8$, $F_{10}$, $H_{12}$ have Fourier coefficients in $\mathbb{Z}$ and they generate the graded ring $$A^{(2)}(U_2({\mathcal O});{\mathbb{Z}}[1/2,1/3]).$$ Moreover, these $5$ generators are algebraically independent over $\mathbb{C}$ and we have $$H_4|_{\mathbb{S}_2}=X_4, \quad E_6|_{\mathbb{S}_2}=X_6,\quad H_8|_{\mathbb{S}_2}=0,\quad F_{10}|_{\mathbb{S}_2}=6X_{10},\quad H_{12}|_{\mathbb{S}_2}=X_{12}.$$ The ring $A^{(2)}(U_2({\mathcal O});R)$ coincides with the ring of the full space $A^{(1)}(U_2({\mathcal O});R)$ of symmetric Hermitian modular forms, because of $M_k(U_2({\mathcal O}))=\{0\}$ for odd $k$. Let $p$ be a prime and $\mathbb{Z}_{(p)}$ the localization of ${\mathbb{Z}}$ at the prime ideal $(p)=p{\mathbb{Z}}$, namely, $\mathbb{Z}_{(p)}:=\mathbb{Q}\cap\mathbb{Z}_p$. The following lemma will be needed in later sections. For a formal Fourier series of the form $F=\sum a_F(H)e^{2\pi i {\rm tr}(HZ)}$, we define $v_p(F)\in\mathbb{Z}$ as usual by $$\label{vp} v_p(F):=\underset{{H\in\Lambda_2({\mathcal O})}}{\text{inf}}\text{ord}_p(a_F(H)).$$ \[Lem:ord\] For any $F_i=\sum a_ {F_i}(H)e^{2\pi i {\rm tr}(HZ)}$ ($i=1$, $2$) with $v_p(F_i)<\infty $, we have $$v_p(F_1F_2)=v_p(F_1)+v_p(F_2).$$ We can easily prove this property, if we define an order for two elements of $\Lambda _2({\mathcal O})$ in the same way as in [@Ki-Na2]. We will need the Sturm bounds in the later sections. \[Thm:Na-Ta\] Let $p$ be a prime with $p\ge 5$. Suppose that $F\in M_k(U_2({\mathcal O});\mathbb{Z}_{(p)})$ satisfies that $a_F(m,r,s,n)\equiv 0$ mod $p$ for all $m$, $n\le [k/8]$. Then we have $F\equiv 0$ mod $p$. In [@Ki-Na2] Theorem 2, we obtained the similar type bounds as this statement, but they are not same. We can modify the proof in the similar way as in [@Na-Ta] Proposition 4.5. In general, the Sturm bounds imply the ordinary vanishing conditions. \[Cor:Na-Ta\] Suppose that $F\in M_k(U_2({\mathcal O});\mathbb{Q})$ satisfies that $a_F(m,r,s,n)=0$ for all $m$, $n\le [k/8]$. Then we have $F=0$. We may apply Theorem \[Thm:Na-Ta\] to $F$ for infinitely many primes $p\ge 5$. Structure over $\mathbb{Z}$ =========================== Construction of generators {#Const} -------------------------- We set $$\begin{aligned} &I_{12} := 2^{-6}\cdot 3^{-3}(H _4^3 - E_6^2) + 2^4\cdot 3^2\cdot H _{12},\\ &J_{12}:=E_6^2,\\ &H_{16}:=2^{-1}\cdot 3^{-1}(E_6F_{10}-H _4^2H _8)\\ &I_{16} := 2^{-2}\cdot 3^{-1}(H _4 H _{12} - H_{16}),\\ &H_{20}:=2^{-2} \cdot 3^{-2}(F_{10}^2-H _4 H _8^2 -2^{2}\cdot 3 H _8 H_{12}),\\ &H_{24}:=2^{-3}\cdot 3^{-1} (H_{12}^2- H_4 H_{20}) - 2^{-1}\cdot 3^{-1}H_8\cdot I_{16}.\end{aligned}$$ In order to construct further generators, we use temporarily the alphabets $K$, $L$. $$\begin{aligned} &K_{14}:=2^{-1}\cdot 3^{-1}(H _4 F_{10}-E_6 H_8),\\ &K_{18}:=2^{-2}\cdot 3^{-1}(E_6 H _{12} - H _4 K_{14}),\\ &K_{22}:=2^{-1}\cdot 3^{-1}(F_{10} H_{12}-H_8 K_{14}),\\ &K_{26}:=2^{-1}\cdot 3^{-1}(F_{10} I_{16}-H_{8} K_{18}),\\ &K_{30}:=2^{-1}\cdot 3^{-1}(E_6H_{24}-K_{14}I_{16}) +3^{-1}H_8F_{10}I_{12},\\ &L_{30}:=2^{-1}\cdot 3^{-1}(F_{10} H_{20}-H_8 K_{22}),\\ &K_{34}:=2^{-1}\cdot 3^{-1}(F_{10}H_{24}-H_{8}K_{26}), \\ &K_{42}:=2^{-2}\cdot 3^{-1}(H_{12} K_{30}-K_{14} H_{28})- 2^{-1}H_8I_{12}K_{22}. \end{aligned}$$ From these definition and Theorem \[Thm:Ki-Na\], it is easy to see that $$\begin{aligned} &K_{14}|_{\mathbb{S}_2}=X_4X_{10},\quad K_{18}|_{\mathbb{S}_2}=X_{18},\quad K_{22}|_{\mathbb{S}_2}=X_{10}X_{12},\\ &K_{26}|_{\mathbb{S}_2}=X_6X_{16},\quad K_{30}|_{\mathbb{S}_2}=X_{30},\quad L_{30}|_{\mathbb{S}_2}=X_{10}^3,\\ &K_{34}|_{\mathbb{S}_2}=X_{10}X_{24},\quad K_{42}|_{\mathbb{S}_2}=X_{42}. \end{aligned}$$ Finally we put $$\begin{aligned} &I_{24}:=E_6K_{18},\\ &H_{28}:=2^{-1}\cdot 3^{-1}(H_4H_{24} - I_{28}) - 3^{-1}H_8^2I_{12},\\ &I_{28}:=2^{-1}\cdot 3^{-1}(F_{10}\cdot K_{18}-H_4\cdot H_8\cdot I_{16}),\\ &H_{36}:=2^{-1}\cdot 3^{-2}(H_{12}H_{24} - H_{20}I_{16}) + 7\cdot 3^{-2}H_{8}H_{28}+ 3^{-1}H_8^3H_{12},\\ &I_{36}:=K_{18}^2,\qquad J_{36}:=E_6K_{30}\\ &H_{40}:=2^{-2}(H_4H_{36} - \frac{1}{2\cdot 3}F_{10}\cdot K_{30}) - 5\cdot 2^{-3}\cdot 3^{-1}H_4\cdot H_8\cdot H_{28}, \\ &~~~~~~~~~~~~~~+ 2^{-2}\cdot {H_8}^3\cdot H_{16} + 2^{-1}H_8\cdot I_{12}\cdot H_{20},\\ &I_{40}:=2^{-1}\cdot 3^{-1}(F_{10}K_{30}-H_4\cdot H_8\cdot H_{28}),\\ &H_{48}:=2^{-2}(H_{12}\cdot H_{36}-H_{24}^2)-2^{-3}H_8(H_{12} H_{28}+ 2 H_{40} \\ &~~~~~~~~~~~~~~+4 H_{10}^2 H_{12} H_8- 2 H_{20} H_4 H_8^2 - 2 H_{12} H_4 H_8^3+ 4 H_{20} H_8 I_{12} \\ &~~~~~~~~~~~~~~~~~~~~~~~~~~~~+ 2 H_{12} H_8^2 I_{12} - H_{24} I_{16} - 2 H_8^3 I_{16} + 2 I_{40}),\\ &I_{48}:=K_{18}K_{30},\\ &H_{52}:=2^{-1}\cdot 3^{-1}(F_{10} K_{42} - 2 F_{10}^2 H_{12}^2 H_8 - 2^2 H_{12} H_{20} H_8 I_{12}\\ &~~~~~~~~~~~~~~- 5 H_{10} H_{22} H_8 I_{12} - H_{28} H_8 I_{16} - H_8^3 I_{12} I_{16}),\\ &H_{60}:=K_{30}^2,\qquad I_{60}:=K_{18}K_{42},\qquad H_{72}:=K_{30}K_{42},\qquad H_{84}:=K_{42}^2.\end{aligned}$$ By the definition of them and from Theorem \[Thm:Ki-Na\], we can easily confirm the following property. We have $$\begin{aligned} H_{k_1}|_{\mathbb{S}_2}=S_{k_1},\ I_{k_2}|_{\mathbb{S}_2}=T_{k_2}\quad \text{and}\quad J_{k_3}|_{\mathbb{S}_2}=U_{k_3} \end{aligned}$$ for each $k_1$, $k_2$, $k_3$ with $$\begin{aligned} &k_1\in \{4,12,16,20,24,28,32,36,40,48,52,60,72,84\},\\ &k_2\in \{12,16,24,28,36,40,48,60\},\quad k_3\in \{12,36\}.\end{aligned}$$ Integralities of generators {#Int} --------------------------- First our purpose is to prove that, all Fourier coefficients of the modular forms constructed in the previous subsection are integers. We start with proving several lemmas. We put $H_4=1+2^{4}\cdot 3S$, $E_6=1+2^3\cdot 3^2T$ with $S$, $T\in {\mathbb{Z}}[\![\dot{\boldsymbol{q}}]\!]$. \[Lem0\] We have $S\equiv T$ mod $2^2\cdot 3$. For $H\in \Lambda _2({\mathcal O})$ with ${\rm rank}(H)=1$, we have $$\begin{aligned} &a_{H_4}(H)= 2^4\cdot 3 \cdot 5\sum _{0<d\mid \varepsilon (H)}d^{3}, \\ &a_{E_6}(H)=-2^3\cdot 3^2 \cdot 7 \sum _{0<d\mid \varepsilon (H)}d^{5}. \end{aligned}$$ The assertion (for ${\rm rank}(H)=1$) follows from $5\equiv -7$ mod $2^2 \cdot 3$ and an application of the Euler congruence $$\sum _{0<d\mid \varepsilon (H)}d^{3}\equiv \sum _{0<d\mid \varepsilon (H)}d^{5} \bmod{2^2 \cdot 3}.$$ Let $H\in \Lambda _2({\mathcal O})$ with ${\rm rank}(H)=2$. Then $$\begin{aligned} &a_{H_4}(H)=-2^6\cdot 3\cdot 5\sum _{0<d\mid \varepsilon (H)}d^3G_{{\boldsymbol K}}(3,4\det H/d^2 ),\\ &a_{E_6}(H)=-2^5\cdot 3^2\cdot 5^{-1}\cdot 7 \sum _{0<d\mid \varepsilon (H)}d^5G_{{\boldsymbol K}}(5,4\det H/d^2 ).\end{aligned}$$ The Euler congruence implies that $$\sum _{0<d\mid \varepsilon (H)}d^3G_{{\boldsymbol K}}(3,4\det H/d^2 )\equiv \sum _{0<d\mid \varepsilon (H)}d^5G_{{\boldsymbol K}}(5,4\det H/d^2 ) \bmod 2^2\cdot 3.$$ On the other hand, we have $$\begin{aligned} 2^2\cdot 5\equiv 2^2\cdot 5^{-1}\cdot 7 \bmod{2^2\cdot 3}. \end{aligned}$$ Therefore the assertion holds. By this lemma, we can put $T=S+2^2\cdot 3 U$ with $U\in {\mathbb{Z}}[\![\dot{\boldsymbol{q}}]\!]$. Then we have $$\begin{aligned} &H_4=1+2^4\cdot 3 S, \\ &E_6=1+2^3\cdot 3^2 S+2^5\cdot 3^3 U. \end{aligned}$$ This is the one of important fact for our arguments on integralities of generators. #### Forms of weight $\bold{12}$ We remark that $J_{12}=E_6^2\in M_{12}(U_2({\mathcal O});{\mathbb{Z}})$ follows from $E_6\in M_{6}(U_2({\mathcal O});{\mathbb{Z}})$. \[Lem1\] We have $I_{12}\in M_{12}(U_2({\mathcal O});{\mathbb{Z}})$. We know by Theorem \[Thm:Ki-Na\] that $H_{12}\in M_{12}(U_2({\mathcal O});{\mathbb{Z}})$. Hence, it suffices to prove that $2^{-6}\cdot 3^{-3}(H_4^3 -E_6^2)\in M_{12}(U_2({\mathcal O});{\mathbb{Z}})$. This can be confirmed by the expansion $$2^{-6}\cdot 3^{-3}(H_4^3 -E_6^2)=S^2 + 64 S^3 - U - 72 S U - 432 U^2,$$ where $S$, $U$ is defined as above. #### Forms of weight $\bold{14}$, $\bold{16}$, $\bold{18}$ For the proof of their integralities, we use (as in [@Ki-Na]) the correspondence between the Maass space and the Kohnen plus subspace which given by Krieg [@Kri]. We review it briefly. We define the congruence subgroup of $\Gamma _1=SL_2({\mathbb{Z}})$ with level $N$ ($N\in \mathbb{N}$) as $$\Gamma _0^{(1)}(N):=\left\{\begin{pmatrix} a&b\\ c& d\end{pmatrix}\in \Gamma _1 \left| \right. c\equiv 0 \bmod{N} \right\}.$$ Let $M_k(\Gamma _0^{(1)}(4),\chi _{-4}^k)$ be the space of elliptic modular forms with character $\chi _{-4}^k$ for $\Gamma _0^{(1)}(4)$. Let ${\mathcal M}_k(U_2({\mathcal O}))$ be the Maass space consisting of all of $F \in M_k(U_2({\mathcal O}))$ satisfying the Maass relation. For the precise definition, see [@Kri], p.676. The Hermitian modular forms version of the Kohnen plus subspace is defined as $$\begin{aligned} &M_k^+(\Gamma _0^{(1)}(4),\chi _{-4}^k)\\ &~~~~~~~:=\left\{f=\sum _{n=0}^{\infty} a_f(n)q^n \in M_k(\Gamma _0^{(1)}(4),\chi _{-4}^k) \left| \right. a_f(n)=0\ \forall n\equiv 1 \bmod{4} \right\}\end{aligned}$$ Krieg [@Kri] gave the isomorphism as vector spaces $$M_{k-1}^+(\Gamma _0^{(1)}(4)\chi _{-4}^{k-1})\longrightarrow {\mathcal M}_k(U_2({\mathcal O})).$$ Let $$\theta :=1+2\sum _{n\ge 1}q^{n^2},\quad f_2:=\sum _{n\ge 1}\sigma _1(n)q^n$$ with $\sigma _1(n):=\sum _{0<d\mid n}d$ and $q:=e^{2\pi i \tau }$, $\tau \in \mathbb{H} _1:=\{ \tau =x+iy \; | \; y>0 \}$. Then it is known that $\theta^2\in M_1(\Gamma _0^{(1)}(4),\chi _{-4})$ and $f_2\in M_2(\Gamma _0^{(1)}(4),1)$ and they generate the graded ring $$\begin{aligned} \bigoplus _{k\in \mathbb{Z}}M_k(\Gamma _0^{(1)}(4),\chi _{-4}^k). \end{aligned}$$ Hence we can construct a Hermitian modular form ${\rm Lift}(h) \in M_k(U_2({\mathcal O}))$ from a polynomial $h\in {\mathbb{C}}[\theta ^2, f_2]$ (such that $h\in M_{k-1}^+(\Gamma _0^{(1)}(4),\chi _{-4}^{k-1})$), by the relation between their Fourier coefficients $$a_{{\rm Lift}(h)}(H)=\sum _{0<d \mid \varepsilon (H)}d^{k-1}\frac{1}{1+|\chi _{-4}(4\det H/d^2 )|}a_h(4\det H/d^2).$$ \[Lem2\] We have $I_{16} \in M_{16}(U_2({\mathcal O});\mathbb{Z})$ and $K_{k}\in M_{k}(U_2({\mathcal O});\mathbb{Z})$ for $k=14$, $18$. We set $$\begin{aligned} h_{15}&:=\theta^{14} f_2^4 -28\theta^{10} f_2^5 +192\theta^6f_2^6\\ &=q^4 + 12 q^6 + 64 q^7 + 36 q^8 - 128 q^{10} - 1152 q^{11} - 936 q^{12} - 504 q^{14} \cdots\\ &=\sum _{n\ge 4}a_{h_{15}}(n)q^n. \end{aligned}$$ Then we have $h_{15}\in M_{15}(\Gamma _0^{(1)}(4),\chi _{-4})$. By an easy numerical experiments, we can confirm that $a_{h_{15}}(n)=0$ for all $n$ with $n\le 500$ and $n\equiv 1$ mod $4$. In fact, we can prove $h_{15}\in M^+_{15}(\Gamma _0^{(1)}(4),\chi _{-4})$ as follows. We consider $$\begin{aligned} h_{15}+h_{15}|T_{\chi _{-4}}-h_{15}|U(2)V(2)=\sum _{n\equiv 1 \bmod{4}}a_{h_{15}(n)}q^n \in M_{15}(\Gamma _0^{(1)}(32),\chi _{-4}),\end{aligned}$$ where $U(l)$, $V(l)$ is the usual operators and $T_{\chi }$ is the twisting operator of the Dirichlet character $\chi $ given in Shimura [@Shim]. Namely, their action for $f=\sum _{n=0}^\infty a_f(n) q^n$ is described as $$\begin{aligned} &f|U(l)=\sum _{n=0}^\infty a_{f}(ln)q^n \\ &f|V(l)=\sum _{n=0}^\infty a_{f}(n)q^{ln}, \\ &f|T_{\chi }=\sum _{n=0}^\infty \chi (n)a_{f}(n)q^{n}. \end{aligned}$$ We remark that we have (at least) $f|U(l)\in M_k(\Gamma _0^{(1)}(Nl),\psi )$, $f|V(l)\in M_k(\Gamma _0^{(1)}(Nl^2),\psi )$ and $f|T_{\chi }\in M_k(\Gamma _0^{(1)}(Nl^2),\psi \chi ^2)$ when $f\in M_k(\Gamma ^{(1)}_0(N),\psi )$. Since the Sturm bound for $M_{15}(\Gamma _0^{(1)}(32),\chi _{-4})$ is $$\frac{15}{12}[\Gamma _1:\Gamma _0^{(1)}(32)]=\frac{15}{12}\cdot 32\left(1+\frac{1}{2}\right)=60,$$ our numerical experiment for $n\le 500$ is sufficient. Namely this shows that $\sum _{n\equiv 1 \bmod{4}}a_{h_{15}(n)}q^n=0$ and hence $h_{15}\in M_{15}^+(\Gamma _0^{(4)},\chi _{-4})$. Therefore we can apply the isomorphism constructed by Krieg, there exists ${\rm Lift}(h_{15})\in M_{16}(U_2({\mathcal O}))$ satisfying that $$\begin{aligned} a_{{\rm Lift}(h_{15})}(H) =\sum _{0<d\mid \varepsilon (H)}\frac{d^{15}}{1+|\chi _{-4}(4\det H/d^2)|} a_{h_{15}}(4\det H/d^2).\end{aligned}$$ By the definition of $h_{15}$, we see that $h_{15}\equiv f_2^4$ mod $2$ because of $\theta \equiv 1$ mod $2$. This implies immediately $$\frac{1}{1+|\chi _{-4}(4\det H/d^2)|}a_{h_{15}}(4\det H/d^2)\in \mathbb{Z}$$ for each $d$. Namely ${\rm Lift}(h_{15})\in M_{16}(U_2({\mathcal O});\mathbb{Z})$ follows. By a direct calculation, we see that $$a_{I_{16}}(m,r,s,n)=a_{{\rm Lift}(h_{15})}(m,r,s,n)-56a_{H_8^2}(m,r,s,n)$$ for all $(m,r,s,n)\in \Lambda _{2}({\mathcal O})$ with $m$, $n\le 2=[16/8]$. Applying Corollary \[Cor:Na-Ta\], we obtain $$I_{16}={\rm Lift}(h_{15})-56H_{8}^2.$$ Since ${\rm Lift}(h_{15})-56H_{8}^2\in M_{16}(U_2({\mathcal O});\mathbb{Z})$, we have the assertion $I_{16}\in M_{16}(U_2({\mathcal O});\mathbb{Z})$. Similarly, if we set $$\begin{aligned} &h_{13}:=2\theta^{14}f_2^3 - 60\theta^{10}f_2^4 + 448\theta^6f_2^5\in M^+_{13}(\Gamma _0^{(1)}(4),\chi _{-4})\\ &h_{17}:= \theta^{18} f_2^4 - 36\theta^{14}f_2^5 + 368\theta^{10}f_2^6 - 768\theta^6f_2^7\in M^+_{17}(\Gamma _0^{(1)}(4),\chi _{-4}), \end{aligned}$$ then we can prove the following equalities $$\begin{aligned} &K_{14}={\rm Lift}(h_{13}), \\ &K_{18}={\rm Lift}(h_{17})+256H_8H_{10}. \end{aligned}$$ The assertions for $K_{14}$, $K_{18}$ follow from these fact immediately. We will give numerical data we used in the proofs, in Subsection \[ProofPlus\] \[Lem3\] We have\ (1) $H_{16}\in M_{16}(U_2({\mathcal O});\mathbb{Z})$,\ (2) $6 H_4 H_{12} - E_6 H_{10} + H_4^2 H_8\equiv 0$ mod $2^3\cdot 3^3$. \(1) By the definition of $I_{16}$, we have $$2^2\cdot 3 I_{16}=H_4H_{12}-H_{16}.$$ Since $2^2\cdot 3I_{16}\equiv 0$ mod $2^2\cdot 3$ because of $I_{16}\in M_{16}(U_2({\mathcal O});{\mathbb{Z}})$, we have $H_{16}\in M_{16}(U_2{(\mathcal O});{\mathbb{Z}})$. \(2) By the definition of $H_{16}$, we have $$2\cdot 3 H_{16}=E_6F_{10}-H_4^2H_8.$$ Hence we can write as $$2^3\cdot 3^2 I_{16}=6H_4 H_{12}-E_6F_{10}+H_4^2H_8.$$ Since $I_{16}\in M_{16}(U_2({\mathcal O});{\mathbb{Z}})$, we have $6H_4 H_{12}-E_6F_{10}+H_4^2H_8\equiv 0$ mod $2^3\cdot 3^2$. Using the fact that $H_4\equiv 1$ mod $2^4\cdot 3$, $E_6\equiv 1$ mod $2^3\cdot 3^2$, we get $$6H_{12}-F_{10}+H_4^2H_8 \equiv 0 \bmod{2^3\cdot 3^2}.$$ From (2) in this lemma, we may write as $$6H_{12}-F_{10}+H_4^2H_8=2^3\cdot 3^2 V$$ with $V\in {\mathbb{Z}}[\![\dot{\boldsymbol{q}}]\!]$. This description is another important thing for our arguments. #### Forms of weight $\boldsymbol{k}$ with $\boldsymbol{k \ge 20}$ First we remark that $I_{24}\in M_{24}(U_2({\mathcal O});{\mathbb{Z}})$ is trivial because of $I_{24}=E_6K_{18}$ and $E_6\in M_{6}(U_2({\mathcal O});{\mathbb{Z}})$, $K_{18}\in M_{18}(U_2({\mathcal O});{\mathbb{Z}})$. Similarly, the integralities of $I_{36}=K_{18}^2$, $J_{36}=E_6K_{30}$, $I_{48}=K_{18}K_{30}$, $H_{60}=K_{30}^2$, $I_{60}=K_{18}K_{42}$, $H_{72}=K_{32}K_{42}$, $H_{84}=K_{42}^2$ follow from that of $E_6$, $K_{18}$, $K_{30}$, $K_{32}$, $K_{42}$. We have the integralities of all the generators constructed in Section \[Subsec:gen\]. By the definition of $H_{20}$, we can write as $$H_{20}=2^{-2}\cdot 3^{-2}(F_{10}^2 - 12 H_{12} H_8 - H_4 H_8^2).$$ If we use the descriptions $$\begin{aligned} &F_{10}=6H_{12}+H_{4}^2H_8-2^3\cdot 3^2V,\\ &H_4=1+2^4\cdot 3 S, \\ &E_6=1+2^3\cdot 3^2 S+2^5\cdot 3^3 U, \end{aligned}$$ then we have $$\begin{aligned} H_{20}&= H_{12}^2 + 32 H_{12} H_8 S + 4 H_8^2 S + 768 H_{12} H_8 S^2 + 384 H_8^2 S^2 \\ &+ 12288 H_8^2 S^3 + 147456 H_8^2 S^4 + 24 H_{12} V + 4 H_8 V + 384 H_8 S V \\ &+ 9216 H_8 S^2 V + 144 V^2. \end{aligned}$$ This shows $H_{20}\in M_{20}(U_2({\mathcal O});\mathbb{Z})$. Similarly, we can prove the integralities of all the generators. In fact we can confirm that, all the generators have descriptions as polynomials of $H_{12}$, $H_8$, $S$, $U$, $V\in {\mathbb{Z}}[\![\dot{\boldsymbol{q}}]\!]$ with integral coefficients (see Subsection \[List\]). Now we could prove the integralities of our generators: All of the modular forms $$\begin{aligned} &H_4,\ H_8,\ H_{12},\ I_{12},\ J_{12},\ H_{16},\ I_{16},\ H_{20},\ H_{24},\ I_{24},\ H_{28},\ I_{28},\\ &H_{36},\ I_{36},\ J_{36},\ H_{40},\ I_{40},\ H_{48},\ I_{48},\ H_{52},\ H_{60},\ I_{60},\ H_{72},\ H_{84}\end{aligned}$$ and also $$\begin{aligned} &K_{14},\ K_{18},\ K_{22},\ K_{26},\ K_{30},\ L_{30},\ K_{34},\ K_{42},\ K_{38}\end{aligned}$$ are elements of ${\mathbb{Z}}[\![\dot{\boldsymbol{q}}]\!]$. Proof of the structure theorem ------------------------------ We are now in a position to prove the following main result. \[Thm1\] The graded ring $A^{(4)}(U_2({\mathcal O});\mathbb{Z})$ over $\mathbb{Z}$ is generated by $24$ modular forms $$\begin{aligned} &H_4,\ H_8,\ H_{12},\ I_{12},\ J_{12},\ H_{16},\ I_{16},\ H_{20},\ H_{24},\ I_{24},\ H_{28},\ I_{28},\\ &H_{36},\ I_{36},\ J_{36},\ H_{40},\ I_{40},\ H_{48},\ I_{48},\ H_{52},\ H_{60},\ I_{60},\ H_{72},\ H_{84}. \end{aligned}$$ In other words, for any $F\in M_k(U_2({\mathcal O});\mathbb{Z})$ with $4\mid k$, there exists a polynomial with $24$ variables having coefficients in $\mathbb{Z}$ such that $F=P(H_4,H_8,H_{12},\cdots , H_{84})$. We prove it by an induction on the weight. For $k=4$, the statement is true clearly. Suppose that the statement is true for all $k$ with $k<k_0$. Let $F\in M_{k_0}(U_2({\mathcal O});\mathbb{Z})$. Then there exists a polynomial $P$ with $23$ variables having coefficients in $\mathbb{Z}$ such that $F|_{\mathbb{S}_2}=P(S_4,S_{12},T_{12},\cdots, S_{84})$ because of Corollary \[Cor:S\_gen\]. Then we have $F-P(H_4,H_{12},I_{12},\cdots ,H_{84})\in M_{k_0}(U_2({\mathcal O};\mathbb{Z}))$ and $(F-P(H_4,H_{12},I_{12},\cdots , H_{84}))|_{\mathbb{S}_2}=0$. Therefore there exists $F'\in M_{k_0-8}(U_2({\mathcal O});\mathbb{Q})$ such that $F-P(H_4,H_{12},I_{12},\cdots , H_{84})=H_8F'$. Since all Fourier coefficients of $P(H_4,H_{12},I_{12},\cdots , H_{84})$ are in $\mathbb{Z}$, we have $H_8F'\in M_{k}(U_2({\mathcal O});\mathbb{Z})$. By $v_p(H_8)=0$ for any prime $p$, we have $F'\in M_{k_0-8}(U_2({\mathcal O});\mathbb{Z})$ because of Lemma \[Lem:ord\]. By the induction hypothesis, there exists a polynomial $P'$ such that $F'=P'(H_4,H_8,H_{12},\cdots , H_{84})$. Therefore we have $$F=P(H_4,H_{12},I_{12},\cdots , H_{84})+H_8P'(H_4,H_8,H_{12}\cdots , H_{84}).$$ This completes the proof of Theorem \[Thm1\]. To determine the structure of $A^{(2)}(U_2({\mathcal O});{\mathbb{Z}})$ by our method, we need $K_{46}\in M_{46}(U_2({\mathcal O});{\mathbb{Z}})$ such that $K_{46}|_{\mathbb{S}_2}=X_{10}X_{36}$. However, we predict that there does not exist such $K_{46}$, due to the leading terms of Fourier expansions. This is a main reason why we restricted our selves to the case where the weights are multiples of $4$. We remark also that we can construct $K_{46}'\in M_{46}(U_2({\mathcal O});{\mathbb{Z}})$ such that $K_{46}'|_{\mathbb{S}_2}=3X_{10} X_{36}$. An Application -------------- As an application, we have the following Sturm bounds for any $k$ with $4\mid k$. \[Thm2\] Let $p$ be an any prime and $k$ an integer with $4\mid k$. Suppose that $F\in M_{k}(U_2({\mathcal O});\mathbb{Z})$ satisfies that $a_F(m,r,s,n)\equiv 0$ mod $p$ for all $m$, $n\in \mathbb{Z}$ with $$0\le m,\ n\le \left[\frac{k}{8}\right]$$ Then we have $F\equiv 0$ mod $p$. For the primes $p\ge 5$, we can prove the statement in the similar way. Hence we prove the essential case $p=2$, $3$ only. \[LemA1\] Let $p=2$, $3$ and $k$ be an even integer with $4\mid k$. Suppose that $F\in M_k(U_2({\mathcal O});\mathbb{Z})$ satisfies $F|_{\mathbb{S}_2}\equiv 0$ mod $p$, then there exists $F'\in M_{k-8}(U_2({\mathcal O});\mathbb{Z})$ such that $F\equiv H_8 F'$ mod $p$. For $k=4$, $8$, we have as free $\mathbb{Z}$-modules $$\begin{aligned} &M_{4}(U_2({\mathcal O});\mathbb{Z})=H_4\mathbb{Z},\\ &M_{8}(U_2({\mathcal O});\mathbb{Z})=H_4^2\mathbb{Z}\oplus H_8\mathbb{Z}. \end{aligned}$$ If $k\neq 8$ and $F\not \equiv 0$ mod $p$, then $F|_{\mathbb{S}_2}\equiv 0$ mod $p$ is impossible. If $k=8$, then $F|_{\mathbb{S}_2}\equiv 0$ mod $p$ is possible only if $F\equiv cH_8$ mod $p$ for some $c\in \mathbb{Z}$. Therefore the statements for $k=4$, $8$ are true. We prove the case $k\ge 12$ with $4\mid k$. Since $F|_{\mathbb{S}_2}\equiv 0$ mod $p$, we have $\frac{1}{p} F|_{\mathbb{S}_2}\in M_k(\Gamma_2;\mathbb{Z})$. By Corollary \[Cor:S\_gen\], there exists an isobaric polynomial $P$ with coefficients in $\mathbb{Z}$ such that $\frac{1}{p}F|_{\mathbb{S}_2}=P(S_4,S_{12},\cdots ,S_{84})$. If we put $$G:=P(H_4,H_{12},\cdots ,H_{84}),$$ then we have $G\in M_{k}(U_2({\mathcal O});\mathbb{Z})$ and $(F-pG)|_{\mathbb{S}_2}=0$. By the result of Dern-Krieg [@D-K], there exists $F'\in M_{k-8}(U_2({\mathcal O});\mathbb{Q})$ such that $F-pG=H_8F'$. Since $v_p(F-pG)\ge 0$ and $v_p(H_8)=0$ for any $p\ge 2$, it should be that $F'\in M_{k-8}(U_2({\mathcal O});\mathbb{Z})$. Then we have $F\equiv H_8F'$ mod $p$. This competes the proof of Lemma \[LemA1\]. We prove Theorem \[Thm2\]. For $k=4$, $8$, we have as free $\mathbb{Z}$-modules $$\begin{aligned} &M_{4}(U_2({\mathcal O});\mathbb{Z})=H_4\mathbb{Z},\\ &M_{8}(U_2({\mathcal O});\mathbb{Z})=H_4^2\mathbb{Z}\oplus H_8\mathbb{Z}.\end{aligned}$$ Since $H_4\equiv 1$ mod $p$ and $H_8\not \equiv c$ mod $p$ for any $c\in {\mathbb{Z}}$, the statements for $k=4$, $8$ are trivial. Let $k\ge 12$. From $[k/8]\ge [k/10]$, we can apply the Sturm bound in Theorem \[Stbd0\] to $F|_{\mathbb{S}_2}$ and then we have $F|_{\mathbb{S}_2}\equiv 0$ mod $p$. By Lemma \[LemA1\], there exists $F'\in M_{k-8}(U_2({\mathcal O});\mathbb{Z})$ such that $F\equiv H_8F'$ mod $p$. Then $F'$ has the property that $a_{F'}(m,r,s,n)\equiv 0$ mod $p$ for any $m$, $n\in \mathbb{Z}$ with $$0\le m,\ n\le \left[\frac{k}{8}\right]-1=\left[\frac{k-8}{8}\right].$$ This is due to the explicit form of the Fourier expansion of $H_8$ (the same reason as in [@Ki-Ta] Lemma 5.1); $$\begin{aligned} H_8&=\dot{q}_1 \dot{q}_2 (4 -2\dot{q}_{12}^{-1}- 2\dot{q}_{12} - 2\ddot{q}_{12}^{-1} \\ &+ \dot{q}_{12}^{-1}\ddot{q}_{12}^{-1} + \dot{q}_{12}\ddot{q}_{12}^{-1} - 2 \ddot{q}_{12} + \dot{q}_{12}^{-1}\ddot{q}_{12} + \dot{q}_{12}\ddot{q}_{12})+\cdots. \end{aligned}$$ Note here that $4\mid k-8$ and we can apply the above argument to $F'$. If we apply this argument repeatedly, we have $F\equiv 0$ mod $p$. This completes the proof of Theorem \[Thm2\]. Completion of the proofs by numerical data ========================================== Fourier expansions of $\boldsymbol{h_{13}}$, $\boldsymbol{h_{15}}$, $\boldsymbol{h_{17}}$ {#ProofPlus} ----------------------------------------------------------------------------------------- In the proof of Lemma \[Lem2\], we relied on the numerical data. Hence we give its data here. Let $$b_k:=\frac{k}{12}[\Gamma _1:\Gamma _0^{(1)}(32)]$$ be the Sturm bounds we mentioned in the proof of Lemma \[Lem2\]. Then we have $b_{13}=52$, $b_{15}=60$, $b_{17}=68$. Therefore the following numerical data are sufficient for our purpose. [ $$\begin{aligned} h_{13}&:=2\theta^{14}f_2^3 - 60\theta^{10}f_2^4 + 448\theta^6f_2^5\\& =2q^3 - 4 q^4 + 112 q^6 - 4 q^7 - 432 q^8 - 640 q^{10} - 594 q^{11} + 5504 q^{12} - 4320 q^{14} + 9380 q^{15} - 20288 q^{16} \\& + 46848 q^{18} - 71622 q^{19} - 16200 q^{20} - 123376 q^{22} + 331668 q^{23} + 282112 q^{24} - 65664 q^{26} - 978492 q^{27} \\& - 453376 q^{28} + 709600 q^{30} + 1749808 q^{31} - 1112832 q^{32} - 120064 q^{34} - 1329480 q^{35} + 3895356 q^{36} \\& - 2315088 q^{38} - 1756316 q^{39} - 152160 q^{40} - 2846208 q^{42} + 7579934 q^{43} - 11366784 q^{44} + 16414816 q^{46} \\& - 17552376 q^{47} + 10176512 q^{48} + 5875200 q^{50} + 33105284 q^{51} + 3775288 q^{52}+\cdots ,\end{aligned}$$ $$\begin{aligned} h_{15}&:=\theta^{14} f_2^4 -28\theta^{10} f_2^5 +192\theta^6f_2^6\\& =q^4 + 12 q^6 + 64 q^7 + 36 q^8 - 128 q^{10} - 1152 q^{11} - 936 q^{12} - 504 q^{14} + 7872 q^{15} + 8144 q^{16} + 16128 q^{18}\\& - 18816 q^{19} - 32022 q^{20} - 121100 q^{22} - 51264 q^{23} + 26976 q^{24} + 464256 q^{26} + 408960 q^{27} + 258448 q^{28} \\&- 909576 q^{30} - 577024 q^{31} - 971712 q^{32} + 355072 q^{34} - 2085120 q^{35} + 525753 q^{36} + 2238876 q^{38}\\& + 7869888 q^{39} + 4278504 q^{40} - 5027328 q^{42} - 853760 q^{43} - 9440856 q^{44} + 8767832 q^{46} - 36277632 q^{47}\\& - 1162368 q^{48} - 26012160 q^{50} + 46803840 q^{51} + 24912602 q^{52} + 40240728 q^{54} + 71676992 q^{55}\\& - 22735296 q^{56} + 47704960 q^{58} - 187329024 q^{59} + 8247408 q^{60}+\cdots , \end{aligned}$$ $$\begin{aligned} h_{17}&:= \theta^{18} f_2^4 - 36\theta^{14}f_2^5 + 368\theta^{10}f_2^6 - 768\theta^6f_2^7\\ &=q^4 - 12 q^6 - 128 q^7 - 228 q^8 - 800 q^{10} - 768 q^{11} + 1872 q^{12} + 15576 q^{14} + 36480 q^{15} + 9296 q^{16} \\& - 108864 q^{18} - 297216 q^{19} - 178110 q^{20} + 356140 q^{22} + 845952 q^{23} + 816576 q^{24} - 682656 q^{26}\\& + 1071360 q^{27} - 803744 q^{28} + 3381480 q^{30} - 12461056 q^{31} - 5338176 q^{32} - 23163968 q^{34} + 20912640 q^{35} \\& + 16663617 q^{36} + 79051812 q^{38} + 40330368 q^{39} + 2424120 q^{40} - 99195264 q^{42}\\& - 169433600 q^{43} - 64675536 q^{44} - 142870072 q^{46} + 63431424 q^{47} - 965376 q^{48} + 629961600 q^{50} \\& + 381400320 q^{51} + 220457666 q^{52} - 671789592 q^{54} - 295596160 q^{55} + 283752576 q^{56} + 90976480 q^{58}\\& + 62678016 q^{59} - 1557183840 q^{60} - 135149088 q^{62} - 2319442560 q^{63} - 394334976 q^{64} - 99539136 q^{66} \\& + 1338126080 q^{67} + 6624813570 q^{68}+\cdots. \end{aligned}$$ ]{} Proof of integralities of the generators {#List} ---------------------------------------- In this subsection, we list the descriptions of our generators as polynomials with variables $H_{12}$, $H_8$, $S$, $U$, $V\in {\mathbb{Z}}[\![\dot{\boldsymbol{q}}]\!]$, where $S$, $U$, $V$ are defined by $$\begin{aligned} &F_{10}=6H_{12}+H_{4}^2H_8-2^3\cdot 3^2V,\\ &H_4=1+2^4\cdot 3 S, \\ &E_6=1+2^3\cdot 3^2 S+2^5\cdot 3^3 U. \end{aligned}$$ The list below shows that the integralities of corresponding generators as in Subsection \[Int\]. Namely we prove that our generators are elements of the ring ${\mathbb{Z}}[H_{12},H_8,S,U,V]$ in the following. $$\begin{aligned} K_{22}&=H_{12}^2 + 8 H_{12} H_8 S - 2 H_8^2 S + 384 H_{12} H_8 S^2 - 192 H_8^2 S^2 - 3072 H_8^2 S^3 + 24 H_8^2 U + 12 H_{12} V - 2 H_8 V - 96 H_8 S V, \\ H_{24}&= -2 H_{12}^2 S + H_{12} H_8 S + 96 H_{12} H_8 S^2 + 8 H_8^2 S^2 + 1536 H_{12} H_8 S^3 + 896 H_8^2 S^3 + 30720 H_8^2 S^4 + 294912 H_8^2 S^5 - 12 H_{12} H_8 U\\ & - 2 H_8^2 U - 192 H_8^2 S U - 4608 H_8^2 S^2 U + H_{12} V + 48 H_{12} S V+ 12 H_8 S V + 1152 H_8 S^2 V + 18432 H_8 S^3 V - 144 H_8 U V + 6 V^2\\& + 288 S V^2, \\ K_{26}&=2 H_{12}^2 S + H_{12} H_8 S + 96 H_{12} H_8 S^2 + 8 H_8^2 S^2 + 3072 H_{12} H_8 S^3+ 1280 H_8^2 S^3 + 61440 H_8^2 S^4 + 884736 H_8^2 S^5 + 72 H_{12}^2 U\\& + 36 H_{12} H_8 U + 4 H_8^2 U + 2304 H_{12} H_8 S U + 480 H_8^2 S U + 55296 H_{12} H_8 S^2 U + 27648 H_8^2 S^2 U + 884736 H_8^2 S^3 U + 10616832 H_8^2 S^4 U \\ &+ H_{12} V + 96 H_{12} S V + 24 H_8 S V + 2304 H_8 S^2 V+ 55296 H_8 S^3 V + 1728 H_{12} U V + 288 H_8 U V + 27648 H_8 S U V\\& + 663552 H_8 S^2 U V + 12 V^2 + 864 S V^2 + 10368 U V^2,\\ H_{28}&=-48 H_{12} H_8^2 + 16 H_{12}^2 S^2 + 8 H_{12} H_8 S^2 + H_8^2 S^2 + 640 H_{12} H_8 S^3+ 192 H_8^2 S^3 + 12288 H_{12} H_8 S^4 + 12288 H_8^2 S^4\\ &+ 294912 H_8^2 S^5 + 2359296 H_8^2 S^6- 12 H_{12}^2 U - 4 H_{12} H_8 U - 288 H_{12} H_8 S U - 24 H_8^2 S U - 4608 H_{12} H_8 S^2 U\\ &- 2304 H_8^2 S^2 U - 36864 H_8^2 S^3 U + 144 H_8^2 U^2 + 4 H_{12} S V + 2 H_8 S V +384 H_{12} S^2 V + 288 H_8 S^2 V + 12288 H_8 S^3 V + 147456 H_8 S^4 V \\ &- 144 H_{12} U V- 24 H_8 U V - 1152 H_8 S U V + V^2 + 96 S V^2 + 2304 S^2 V^2,\\ I_{28}&=-2 H_{12}^2 S - H_{12} H_8 S - 192 H_{12}^2 S^2 - 192 H_{12} H_8 S^2 - 16 H_8^2 S^2- 9984 H_{12} H_8 S^3 - 2560 H_8^2 S^3 - 147456 H_{12} H_8 S^4 - 147456 H_8^2 S^4 \\ &- 3538944 H_8^2 S^5 - 28311552 H_8^2 S^6 + 72 H_{12}^2 U + 36 H_{12} H_8 U + 4 H_8^2 U + 2304 H_{12} H_8 S U + 576 H_8^2 S U + 27648 H_{12} H_8 S^2 U \\& + 27648 H_8^2 S^2 U + 442368 H_8^2 S^3 U - H_{12} V - 120 H_{12} S V - 24 H_8 S V - 4608 H_{12} S^2 V - 3456 H_8 S^2 V - 147456 H_8 S^3 V\\ &- 1769472 H_8 S^4 V + 864 H_{12} U V+ 288 H_8 U V + 13824 H_8 S U V - 12 V^2 - 1152 S V^2 - 27648 S^2 V^2,\end{aligned}$$ $$\begin{aligned} K_{30}&=288 H_{12}^2 H_8 + 48 H_{12} H_8^2 + 4608 H_{12} H_8^2 S - 8 H_{12}^2 S^2 + 2 H_{12} H_8 S^2+ H_8^2 S^2 + 110592 H_{12} H_8^2 S^2 + 256 H_{12} H_8 S^3 + 192 H_8^2 S^3 \\& + 6144 H_{12} H_8 S^4+ 13056 H_8^2 S^4 + 368640 H_8^2 S^5 + 3538944 H_8^2 S^6 + 12 H_{12}^2 U + 2 H_{12} H_8 U + 288 H_{12}^2 S U + 240 H_{12} H_8 S U\\& + 13824 H_{12} H_8 S^2 U + 1152 H_8^2 S^2 U + 221184 H_{12} H_8 S^3 U+ 129024 H_8^2 S^3 U + 4423680 H_8^2 S^4 U + 42467328 H_8^2 S^5 U- 864 H_{12} H_8 U^2\\ & - 144 H_8^2 U^2- 13824 H_8^2 S U^2- 331776 H_8^2 S^2 U^2 + 3456 H_{12} H_8 V + 4 H_{12} S V + 2 H_8 S V + 192 H_{12} S^2 V+ 312 H_8 S^2 V\\& + 15360 H_8 S^3 V + 221184 H_8 S^4 V + 144 H_{12} U V+ 6912 H_{12} S U V+ 1728 H_8 S U V + 165888 H_8 S^2 U V + 2654208 H_8 S^3 U V\\& - 10368 H_8 U^2 V + V^2 + 120 S V^2+ 3456 S^2 V^2 + 864 U V^2 + 41472 S U V^2,\\ L_{30}&=H_{12}^3 + H_{12}^2 H_8 + 48 H_{12}^2 H_8 S + 16 H_{12} H_8^2 S - H_8^3 S + 1152 H_{12}^2 H_8 S^2 + 1344 H_{12} H_8^2 S^2 - 32 H_8^3 S^2 + 36864 H_{12} H_8^2 S^3+ 7168 H_8^3 S^3\\& + 442368 H_{12} H_8^2 S^4 + 368640 H_8^3 S^4 + 7077888 H_8^3 S^5 + 56623104 H_8^3 S^6 + 20 H_8^3 U + 36 H_{12}^2 V + 18 H_{12} H_8 V - H_8^2 V + 1152 H_{12} H_8 S V \\ &+ 96 H_8^2 S V + 27648 H_{12} H_8 S^2 V + 13824 H_8^2 S^2 V + 442368 H_8^2 S^3 V + 5308416 H_8^2 S^4 V + 432 H_{12} V^2 + 72 H_8 V^2 + 6912 H_8 S V^2 \\& + 165888 H_8 S^2 V^2 + 1728 V^3,\\ K_{34}&=-2 H_{12}^3 S - H_{12}^2 H_8 S - 128 H_{12}^2 H_8 S^2 - 24 H_{12} H_8^2 S^2 - 2304 H_{12}^2 H_8 S^3- 2560 H_{12} H_8^2 S^3 - 64 H_8^3 S^3 - 92160 H_{12} H_8^2 S^4\\& - 12288 H_8^3 S^4 - 884736 H_{12} H_8^2 S^5 - 737280 H_8^3 S^5 - 16515072 H_8^3 S^6 - 113246208 H_8^3 S^7 + 24 H_{12}^2 H_8 U + 10 H_{12} H_8^2 U+ H_8^3 U\\& + 768 H_{12} H_8^2 S U + 144 H_8^3 S U + 18432 H_{12} H_8^2 S^2 U + 9216 H_8^3 S^2 U + 294912 H_8^3 S^3 U + 3538944 H_8^3 S^4 U - H_{12}^2 V - 72 H_{12}^2 S V\\& - 32 H_{12} H_8 S V + 2 H_8^2 S V - 3456 H_{12} H_8 S^2 V - 96 H_8^2 S^2 V - 55296 H_{12} H_8 S^3 V - 27648 H_8^2 S^3 V - 1105920 H_8^2 S^4 V - 10616832 H_8^2 S^5 V\\& + 576 H_{12} H_8 U V + 96 H_8^2 U V + 9216 H_8^2 S U V + 221184 H_8^2 S^2 U V - 18 H_{12} V^2 + H_8 V^2 - 864 H_{12} S V^2 - 144 H_8 S V^2 - 20736 H_8 S^2 V^2\\& - 331776 H_8 S^3 V^2 + 3456 H_8 U V^2\\& - 72 V^3 - 3456 S V^3,\\ H_{36}&=-37 H_{12} H_8^3 + 16 H_{12}^2 H_8 S^2 + 8 H_{12} H_8^2 S^2 + H_8^3 S^2 + 128 H_{12}^2 H_8 S^3+ 704 H_{12} H_8^2 S^3 + 192 H_8^3 S^3 + 17408 H_{12} H_8^2 S^4 + 12800 H_8^3 S^4\\& + 98304 H_{12} H_8^2 S^5 + 352256 H_8^3 S^5 + 4194304 H_8^3 S^6 + 18874368 H_8^3 S^7 + 4 H_{12}^3 U - 8 H_{12}^2 H_8 U - 3 H_{12} H_8^2 U + 192 H_{12}^2 H_8 S U \\& - 176 H_{12} H_8^2 S U - 16 H_8^3 S U+ 4608 H_{12}^2 H_8 S^2 U + 768 H_{12} H_8^2 S^2 U - 1280 H_8^3 S^2 U + 147456 H_{12} H_8^2 S^3 U + 10240 H_8^3 S^3 U \\& + 1769472 H_{12} H_8^2 S^4 U + 1474560 H_8^3 S^4 U + 28311552 H_8^3 S^5 U + 226492416 H_8^3 S^6 U + 112 H_8^3 U^2 + 4 H_{12}^2 S V + 6 H_{12} H_8 S V+ 2 H_8^2 S V\\& + 576 H_{12} H_8 S^2 V + 304 H_8^2 S^2 V + 6144 H_{12} H_8 S^3 V + 14848 H_8^2 S^3 V + 270336 H_8^2 S^4 V + 1769472 H_8^2 S^5 V + 144 H_{12}^2 U V - 72 H_{12} H_8 U V \\& - 16 H_8^2 U V + 4608 H_{12} H_8 S U V- 192 H_8^2 S U V + 110592 H_{12} H_8 S^2 U V + 55296 H_8^2 S^2 U V + 1769472 H_8^2 S^3 U V + 21233664 H_8^2 S^4 U V \\& + H_{12} V^2 + H_8 V^2 + 96 H_{12} S V^2 + 120 H_8 S V^2 + 4608 H_8 S^2 V^2 + 55296 H_8 S^3 V^2 + 1728 H_{12} U V^2 + 288 H_8 U V^2 + 27648 H_8 S U V^2\\& + 663552 H_8 S^2 U V^2 + 8 V^3 + 576 S V^3 + 6912 U V^3,\\ K_{38}&=-96 H_{12}^2 H_8^2 - 16 H_{12} H_8^3 - 1536 H_{12} H_8^3 S + 16 H_{12}^3 S^2 + 12 H_{12}^2 H_8 S^2 + 2 H_{12} H_8^2 S^2 - 36864 H_{12} H_8^3 S^2 + 896 H_{12}^2 H_8 S^3\\& + 384 H_{12} H_8^2 S^3+ 16 H_8^3 S^3 + 18432 H_{12}^2 H_8 S^4 + 26624 H_{12} H_8^2 S^4 + 3328 H_8^3 S^4 + 737280 H_{12} H_8^2 S^5 + 258048 H_8^3 S^5 \\& + 7077888 H_{12} H_8^2 S^6 + 9240576 H_8^3 S^6 + 150994944 H_8^3 S^7 + 905969664 H_8^3 S^8 - 12 H_{12}^3 U - 8 H_{12}^2 H_8 U - H_{12} H_8^2 U - 528 H_{12}^2 H_8 S U\\& - 176 H_{12} H_8^2 S U - 4 H_8^3 S U - 9216 H_{12}^2 H_8 S^2 U - 11520 H_{12} H_8^2 S^2 U - 960 H_8^3 S^2 U - 258048 H_{12} H_8^2 S^3 U - 73728 H_8^3 S^3 U \\& - 1769472 H_{12} H_8^2 S^4 U - 2211840 H_8^3 S^4 U - 21233664 H_8^3 S^5 U + 288 H_{12} H_8^2 U^2 + 48 H_8^3 U^2 + 4608 H_8^3 S U^2 + 110592 H_8^3 S^2 U^2\\& - 1152 H_{12} H_8^2 V + 4 H_{12}^2 S V + 2 H_{12} H_8 S V + 576 H_{12}^2 S^2 V + 480 H_{12} H_8 S^2 V + 40 H_8^2 S^2 V + 27648 H_{12} H_8 S^3 V + 7168 H_8^2 S^3 V \\& + 442368 H_{12} H_8 S^4 V + 442368 H_8^2 S^4 V + 10616832 H_8^2 S^5 V + 84934656 H_8^2 S^6 V - 288 H_{12}^2 U V \ - 120 H_{12} H_8 U V - 4 H_8^2 U V\\& - 8064 H_{12} H_8 S U V - 1152 H_8^2 S U V - 110592 H_{12} H_8 S^2 U V - 82944 H_8^2 S^2 U V - 1327104 H_8^2 S^3 U V + 3456 H_8^2 U^2 V \\& + H_{12} V^2 + 144 H_{12} S V^2 + 36 H_8 S V^2 + 6912 H_{12} S^2 V^2 + 5184 H_8 S^2 V^2 + 221184 H_8 S^3 V^2 + 2654208 H_8 S^4 V^2 - 1728 H_{12} U V^2 \\& - 432 H_8 U V^2 - 20736 H_8 S U V^2 + 12 V^3 + 1152 S V^3 + 27648 S^2 V^3,\\ H_{40}&=-24 H_{12}^2 H_8^2 - H_{12} H_8^3 - 42 H_{12} H_8^3 S + 3 H_8^4 S + 2 H_{12}^3 S^2 + H_{12}^2 H_8 S^2 + 288 H_8^4 S^2 + 64 H_{12}^2 H_8 S^3 + 8 H_{12} H_8^2 S^3 + 6912 H_8^4 S^3 \\ &+ 768 H_{12}^2 H_8 S^4 + 512 H_{12} H_8^2 S^4 - 64 H_8^3 S^4 - 6144 H_{12} H_8^2 S^5 - 10240 H_8^3 S^5- 294912 H_{12} H_8^2 S^6 - 573440 H_8^3 S^6 - 13369344 H_8^3 S^7\\& - 113246208 H_8^3 S^8 - 2 H_{12}^3 U - H_{12}^2 H_8 U + 216 H_{12} H_8^3 U + 36 H_8^4 U - 24 H_{12}^3 S U - 84 H_{12}^2 H_8 S U - 14 H_{12} H_8^2 S U - H_8^3 S U + 3456 H_8^4 S U \\& - 2304 H_{12}^2 H_8 S^2 U - 1632 H_{12} H_8^2 S^2 U - 176 H_8^3 S^2 U + 82944 H_8^4 S^2 U - 27648 H_{12}^2 H_8 S^3 U - 55296 H_{12} H_8^2 S^3 U - 12032 H_8^3 S^3 U\\& - 1105920 H_{12} H_8^2 S^4 U - 466944 H_8^3 S^4 U - 10616832 H_{12} H_8^2 S^5 U - 12386304 H_8^3 S^5 U - 198180864 H_8^3 S^6 U - 1358954496 H_8^3 S^7 U\\ & + 72 H_{12} H_8^2 U^2 + 4 H_8^3 U^2 + 192 H_8^3 S U^2 + 3 H_8^3 V + 216 H_8^3 S V + 24 H_{12}^2 S^2 V - 2 H_8^2 S^2 V - 384 H_{12} H_8 S^3 V - 416 H_8^2 S^3 V \\ &- 18432 H_{12} H_8 S^4 V - 30720 H_8^2 S^4 V - 958464 H_8^2 S^5 V - 10616832 H_8^2 S^6 V - 36 H_{12}^2 U V - 12 H_{12} H_8 U V - H_8^2 U V + 2592 H_8^3 U V\\& - 864 H_{12}^2 S U V - 1152 H_{12} H_8 S U V - 168 H_8^2 S U V - 41472 H_{12} H_8 S^2 U V - 12672 H_8^2 S^2 U V - 663552 H_{12} H_8 S^3 U V - 552960 H_8^2 S^3 U V\\& - 13271040 H_8^2 S^4 U V - 127401984 H_8^2 S^5 U V - 6 H_{12} S V^2 - 3 H_8 S V^2 - 288 H_{12} S^2 V^2 - 432 H_8 S^2 V^2 - 20736 H_8 S^3 V^2 \\& - 331776 H_8 S^4 V^2 - 216 H_{12} U V^2 - 36 H_8 U V^2 - 10368 H_{12} S U V^2 - 5184 H_8 S U V^2 - 248832 H_8 S^2 U V^2 - 3981312 H_8 S^3 U V^2 - V^3 \\ &- 120 S V^3 - 3456 S^2 V^3 - 864 U V^3 - 41472 S U V^3,\end{aligned}$$ $$\begin{aligned} I_{40}&=288 H_{12}^3 H_8 + 96 H_{12}^2 H_8^2 + 16 H_{12} H_8^3 + 9216 H_{12}^2 H_8^2 S + 1920 H_{12} H_8^3 S - 8 H_{12}^3 S^2 - 2 H_{12}^2 H_8 S^2+ 221184 H_{12}^2 H_8^2 S^2\\& + 110592 H_{12} H_8^3 S^2 + 96 H_{12} H_8^2 S^3 + 8 H_8^3 S^3 + 3538944 H_{12} H_8^3 S^3 + 3072 H_{12}^2 H_8 S^4 + 11776 H_{12} H_8^2 S^4 + 2048 H_8^3 S^4 \\& + 42467328 H_{12} H_8^3 S^4 + 466944 H_{12} H_8^2 S^5 + 196608 H_8^3 S^5 + 5898240 H_{12} H_8^2 S^6 + 8749056 H_8^3 S^6 + 179306496 H_8^3 S^7\\& + 1358954496 H_8^3 S^8 + 12 H_{12}^3 U + 6 H_{12}^2 H_8 U + H_{12} H_8^2 U + 288 H_{12}^3 S U + 576 H_{12}^2 H_8 S U + 152 H_{12} H_8^2 S U + 4 H_8^3 S U\\& + 23040 H_{12}^2 H_8 S^2 U + 11136 H_{12} H_8^2 S^2 U + 768 H_8^3 S^2 U + 331776 H_{12}^2 H_8 S^3 U + 516096 H_{12} H_8^2 S^3 U + 64512 H_8^3 S^3 U \\ &+ 13271040 H_{12} H_8^2 S^4 U + 3538944 H_8^3 S^4 U + 127401984 H_{12} H_8^2 S^5 U + 127401984 H_8^3 S^5 U + 2378170368 H_8^3 S^6 U\\& + 16307453952 H_8^3 S^7 U - 864 H_{12}^2 H_8 U^2 - 288 H_{12} H_8^2 U^2 - 48 H_8^3 U^2 - 27648 H_{12} H_8^2 S U^2 - 5760 H_8^3 S U^2 - 663552 H_{12} H_8^2 S^2 U^2 \\& - 331776 H_8^3 S^2 U^2 - 10616832 H_8^3 S^3 U^2 - 127401984 H_8^3 S^4 U^2 + 6912 H_{12}^2 H_8 V + 1152 H_{12} H_8^2 V + 4 H_{12}^2 S V + 2 H_{12} H_8 S V \\:& + 110592 H_{12} H_8^2 S V + 96 H_{12}^2 S^2 V + 336 H_{12} H_8 S^2 V + 32 H_8^2 S^2 V + 2654208 H_{12} H_8^2 S^2 V + 19968 H_{12} H_8 S^3 V + 6272 H_8^2 S^3 V \\& + 368640 H_{12} H_8 S^4 V + 436224 H_8^2 S^4 V + 12681216 H_8^2 S^5 V + 127401984 H_8^2 S^6 V + 288 H_{12}^2 U V + 72 H_{12} H_8 U V + 4 H_8^2 U V\\& + 10368 H_{12}^2 S U V + 9216 H_{12} H_8 S U V + 672 H_8^2 S U V + 497664 H_{12} H_8 S^2 U V + 78336 H_8^2 S^2 U V + 7962624 H_{12} H_8 S^3 U V \\& + 5308416 H_8^2 S^3 U V + 159252480 H_8^2 S^4 U V + 1528823808 H_8^2 S^5 U V - 20736 H_{12} H_8 U^2 V - 3456 H_8^2 U^2 V \\& - 331776 H_8^2 S U^2 V - 7962624 H_8^2 S^2 U^2 V + H_{12} V^2 + 41472 H_{12} H_8 V^2 + 168 H_{12} S V^2 + 36 H_8 S V^2 + 5760 H_{12} S^2 V^2\\& + 5472 H_8 S^2 V^2 + 267264 H_8 S^3 V^2 + 3981312 H_8 S^4 V^2 + 2592 H_{12} U V^2 + 144 H_8 U V^2 + 124416 H_{12} S U V^2 + 41472 H_8 S U V^2\\& + 2985984 H_8 S^2 U V^2 + 47775744 H_8 S^3 U V^2 - 124416 H_8 U^2 V^2 + 12 V^3 + 1440 S V^3 + 41472 S^2 V^3 + 10368 U V^3 + 497664 S U V^3,\\ K_{42}&=-48 H_{12}^3 H_8 + 8 H_{12}^2 H_8^2 + 192 H_{12} H_8^3 S - 2 H_{12}^3 S^2 - H_{12}^2 H_8 S^2 - 18432 H_{12}^2 H_8^2 S^2 + 18432 H_{12} H_8^3 S^2 - 64 H_{12}^3 S^3 - 112 H_{12}^2 H_8 S^3 \\& - 16 H_{12} H_8^2 S^3 + 294912 H_{12} H_8^3 S^3 - 4608 H_{12}^2 H_8 S^4 - 2560 H_{12} H_8^2 S^4 - 128 H_8^3 S^4 - 73728 H_{12}^2 H_8 S^5 - 141312 H_{12} H_8^2 S^5 \\& - 24576 H_8^3 S^5 - 3244032 H_{12} H_8^2 S^6 - 1671168 H_8^3 S^6 - 28311552 H_{12} H_8^2 S^7 - 49545216 H_8^3 S^7 - 679477248 H_8^3 S^8 - 3623878656 H_8^3 S^9 \\& + 2 H_{12}^3 U + H_{12}^2 H_8 U - 2304 H_{12} H_8^3 U + 72 H_{12}^3 S U + 108 H_{12}^2 H_8 S U + 10 H_{12} H_8^2 S U - H_8^3 S U + 4032 H_{12}^2 H_8 S^2 U + 1632 H_{12} H_8^2 S^2 U\\& - 144 H_8^3 S^2 U + 55296 H_{12}^2 H_8 S^3 U + 82944 H_{12} H_8^2 S^3 U - 2304 H_8^3 S^3 U 1548288 H_{12} H_8^2 S^4 U + 331776 H_8^3 S^4 U+ 10616832 H_{12} H_8^2 S^5 U\\& + 10616832 H_8^3 S^5 U + 84934656 H_8^3 S^6 U - 72 H_{12} H_8^2 U^2 + 12 H_8^3 U^2 - 3456 H_{12} H_8^2 S U^2 - 82944 H_8^3 S^2 U^2 - 1327104 H_8^3 S^3 U^2\\& + 6912 H_8^3 U^3 - 576 H_{12}^2 H_8 V + 192 H_{12} H_8^2 V + 9216 H_{12} H_8^2 S V - 48 H_{12}^2 S^2 V - 24 H_{12} H_8 S^2 V - 2 H_8^2 S^2 V - 2304 H_{12}^2 S^3 V\\& - 3072 H_{12} H_8 S^3 V - 608 H_8^2 S^3 V - 129024 H_{12} H_8 S^4 V - 61440 H_8^2 S^4 V - 1769472 H_{12} H_8 S^5 V - 2654208 H_8^2 S^5 V - 49545216 H_8^2 S^6 V\\& - 339738624 H_8^2 S^7 V + 36 H_{12}^2 U V + 12 H_{12} H_8 U V - H_8^2 U V + 1728 H_{12}^2 S U V + 1440 H_{12} H_8 S U V - 48 H_8^2 S U V+ 55296 H_{12} H_8 S^2 U V\\& + 6912 H_8^2 S^2 U V + 663552 H_{12} H_8 S^3 U V + 442368 H_8^2 S^3 U V + 5308416 H_8^2 S^4 U V - 864 H_8^2 U^2 V - 41472 H_8^2 S U^2 V - 6 H_{12} S V^2 \\& - 3 H_8 S V^2 - 864 H_{12} S^2 V^2 - 576 H_8 S^2 V^2 - 27648 H_{12} S^3 V^2 - 39168 H_8 S^3 V^2 - 1105920 H_8 S^4 V^2 - 10616832 H_8 S^5 V^2\\& + 216 H_{12} U V^2 + 36 H_8 U V^2 + 10368 H_{12} S U V^2 + 3456 H_8 S U V^2 + 82944 H_8 S^2 U V^2 - V^3 - 144 S V^3 - 6912 S^2 V^3 - 110592 S^3 V^3,\\ H_{48}&=-162 H_{12}^3 H_8^2 - 63 H_{12}^2 H_8^3 - 4 H_{12} H_8^4 - 5172 H_{12}^2 H_8^3 S - 834 H_{12} H_8^4 S - H_{12}^4 S^2 + 3 H_{12}^3 H_8 S^2 + H_{12}^2 H_8^2 S^2\\& - 124416 H_{12}^2 H_8^3 S^2 - 61632 H_{12} H_8^4 S^2 + 48 H_8^5 S^2 - 64 H_{12}^3 H_8 S^3 + 40 H_{12}^2 H_8^2 S^3 - 12 H_{12} H_8^3 S^3 - 3 H_8^4 S^3 - 1981440 H_{12} H_8^4 S^3\\& + 5376 H_8^5 S^3 - 1536 H_{12}^3 H_8 S^4 - 1152 H_{12}^2 H_8^2 S^4 - 1792 H_{12} H_8^3 S^4 - 624 H_8^4 S^4 - 23887872 H_{12} H_8^4 S^4 + 184320 H_8^5 S^4 - 79872 H_{12}^2 H_8^2 S^5 \\& - 99328 H_{12} H_8^3 S^5 - 51712 H_8^4 S^5 + 1769472 H_8^5 S^5 - 884736 H_{12}^2 H_8^2 S^6 - 2441216 H_{12} H_8^3 S^6 - 2170880 H_8^4 S^6 - 33030144 H_{12} H_8^3 S^7\\& - 48758784 H_8^4 S^7 - 226492416 H_{12} H_8^3 S^8 - 594542592 H_8^4 S^8 - 4529848320 H_8^4 S^9 - 21743271936 H_8^4 S^{10} + H_{12}^4 U - 3 H_{12}^3 H_8 U\\& - H_{12}^2 H_8^2 U - 72 H_{12} H_8^4 U - 12 H_8^5 U + 12 H_{12}^3 H_8 S U - 78 H_{12}^2 H_8^2 S U - H_{12} H_8^3 S U + H_8^4 S U- 1152 H_8^5 S U + 1152 H_{12}^3 H_8 S^2 U \\& - 2496 H_{12}^2 H_8^2 S^2 U- 464 H_{12} H_8^3 S^2 U + 160 H_8^4 S^2 U - 27648 H_8^5 S^2 U - 4608 H_{12}^2 H_8^2 S^3 U - 38912 H_{12} H_8^3 S^3 U + 7552 H_8^4 S^3 U \\& + 442368 H_{12}^2 H_8^2 S^4 U - 1290240 H_{12} H_8^3 S^4 U - 24576 H_8^4 S^4 U - 8847360 H_{12} H_8^3 S^5 U - 11501568 H_8^4 S^5 U + 56623104 H_{12} H_8^3 S^6 U\\& - 297271296 H_8^4 S^6 U - 2038431744 H_8^4 S^7 U + 288 H_{12}^2 H_8^2 U^2 + 124 H_{12} H_8^3 U^2+ 7 H_8^4 U^2 + 9216 H_{12} H_8^3 S U^2 + 1488 H_8^4 S U^2\\& + 221184 H_{12} H_8^3 S^2 U^2 + 110592 H_8^4 S^2 U^2+ 3538944 H_8^4 S^3 U^2 + 42467328 H_8^4 S^4 U^2 - 3888 H_{12}^2 H_8^2 V - 642 H_{12} H_8^3 V\\& - 61920 H_{12} H_8^3 S V+ 72 H_8^4 S V - 48 H_{12}^3 S^2 V + 12 H_{12}^2 H_8 S^2 V - 24 H_{12} H_8^2 S^2 V - 7 H_8^3 S^2 V - 1492992 H_{12} H_8^3 S^2 V \\& + 6912 H_8^4 S^2 V - 2688 H_{12}^2 H_8 S^3 V- 3072 H_{12} H_8^2 S^3 V - 1360 H_8^3 S^3 V + 110592 H_8^4 S^3 V - 55296 H_{12}^2 H_8 S^4 V - 116736 H_{12} H_8^2 S^4 V \\& - 96768 H_8^3 S^4 V - 2211840 H_{12} H_8^2 S^5 V - 3133440 H_8^3 S^5 V - 21233664 H_{12} H_8^2 S^6 V - 48955392 H_8^3 S^6 V\\& - 452984832 H_8^3 S^7 V - 2717908992 H_8^3 S^8 V + 36 H_{12}^3 U V - 48 H_{12}^2 H_8 U V - 3 H_{12} H_8^2 U V + H_8^3 U V - 864 H_8^4 U V - 144 H_{12}^2 H_8 S U V \\& - 480 H_{12} H_8^2 S U V + 180 H_8^3 S U V + 27648 H_{12}^2 H_8 S^2 U V - 48384 H_{12} H_8^2 S^2 U V + 5184 H_8^3 S^2 U V - 552960 H_{12} H_8^2 S^3 U V \\& - 387072 H_8^3 S^3 U V + 5308416 H_{12} H_8^2 S^4 U V - 19906560 H_8^3 S^4 U V - 191102976 H_8^3 S^5 U V + 6912 H_{12} H_8^2 U^2 V + 1152 H_8^3 U^2 V\\& + 110592 H_8^3 S U^2 V + 2654208 H_8^3 S^2 U^2 V - 23328 H_{12} H_8^2 V^2 + 36 H_8^3 V^2 - 6 H_{12}^2 S V^2 - 15 H_{12} H_8 S V^2 - 6 H_8^2 S V^2 + 1728 H_8^3 S V^2\\& - 864 H_{12}^2 S^2 V^2 - 1296 H_{12} H_8 S^2 V^2 - 1032 H_8^2 S^2 V^2 - 41472 H_{12} H_8 S^3 V^2 - 59136 H_8^2 S^3 V^2 - 663552 H_{12} H_8 S^4 V^2\\& - 1327104 H_8^2 S^4 V^2 - 15925248 H_8^2 S^5 V^2 - 127401984 H_8^2 S^6 V^2 + 432 H_{12}^2 U V^2 - 252 H_{12} H_8 U V^2 + 42 H_8^2 U V^2 \\& - 8640 H_{12} H_8 S U V^2 - 864 H_8^2 S U V^2 + 165888 H_{12} H_8 S^2 U V^2 - 373248 H_8^2 S^2 U V^2 - 5971968 H_8^2 S^3 U V^2 + 41472 H_8^2 U^2 V^2 \\& - H_{12} V^3 - 2 H_8 V^3 - 144 H_{12} S V^3 - 276 H_8 S V^3 - 6912 H_{12} S^2 V^3 - 12096 H_8 S^2 V^3 - 221184 H_8 S^3 V^3 - 2654208 H_8 S^4 V^3 \\& + 1728 H_{12} U V^3 - 1296 H_8 U V^3 - 62208 H_8 S U V^3 - 9 V^4 - 864 S V^4 - 20736 S^2 V^4\end{aligned}$$ $$\begin{aligned} K_{52}&=-876 H_{12}^4 H_8 - 124 H_{12}^3 H_8^2 + H_{12}^2 H_8^3 - 21504 H_{12}^3 H_8^2 S + 384 H_{12}^2 H_8^3 S + 288 H_{12} H_8^4 S \\& - 2 H_{12}^4 S^2 - 7 H_{12}^3 H_8 S^2 - H_{12}^2 H_8^2 S^2 - 672768 H_{12}^3 H_8^2 S^2- 23040 H_{12}^2 H_8^3 S^2 \\& + 53760 H_{12} H_8^4 S^2 - 64 H_{12}^4 S^3 - 512 H_{12}^3 H_8 S^3- 240 H_{12}^2 H_8^2 S^3 - 5750784 H_{12}^2 H_8^3 S^3\\& + 2 H_8^4 S^3 + 3588096 H_{12} H_8^4 S^3 - 6400 H_{12}^3 H_8 S^4 - 18432 H_{12}^2 H_8^2 S^4 - 704 H_{12} H_8^3 S^4 \\& - 129171456 H_{12}^2 H_8^3 S^4 + 480 H_8^4 S^4 + 100270080 H_{12} H_8^4 S^4 - 98304 H_{12}^3 H_8 S^5 - 538624 H_{12}^2 H_8^2 S^5 \\& - 131072 H_{12} H_8^3 S^5 + 43008 H_8^4 S^5+ 962592768 H_{12} H_8^4 S^5 - 6193152 H_{12}^2 H_8^2 S^6 - 8241152 H_{12} H_8^3 S^6\\& + 1630208 H_8^4 S^6 - 56623104 H_{12}^2 H_8^2 S^7 - 208404480 H_{12} H_8^3 S^7 + 10616832 H_8^4 S^7 \\& - 2378170368 H_{12} H_8^3 S^8 - 1019215872 H_8^4 S^8- 14495514624 H_{12} H_8^3 S^9 - 29595009024 H_8^4 S^9 \\& - 318901321728 H_8^4 S^{10} - 1391569403904 H_8^4 S^{11} + 2 H_{12}^4 U + 7 H_{12}^3 H_8 U + H_{12}^2 H_8^2 U\\& - 18432 H_{12}^2 H_8^3 U - 3072 H_{12} H_8^4 U + 72 H_{12}^4 S U + 556 H_{12}^3 H_8 S U + 242 H_{12}^2 H_8^2 S U - H_{12} H_8^3 S U \\& - 2 H_8^4 S U - 294912 H_{12} H_8^4 S U + 6144 H_{12}^3 H_8 S^2 U + 18784 H_{12}^2 H_8^2 S^2 U + 272 H_{12} H_8^3 S^2 U \\& - 528 H_8^4 S^2 U - 7077888 H_{12} H_8^4 S^2 U + 82944 H_{12}^3 H_8 S^3 U + 516096 H_{12}^2 H_8^2 S^3 U + 89344 H_{12} H_8^3 S^3 U\\& - 52480 H_8^4 S^3 U + 4202496 H_{12}^2 H_8^2 S^4 U + 6316032 H_{12} H_8^3 S^4 U - 2383872 H_8^4 S^4 U \\& + 31850496 H_{12}^2 H_8^2 S^5 U + 136249344 H_{12} H_8^3 S^5 U - 44236800 H_8^4 S^5 U + 934281216 H_{12} H_8^3 S^6 U\\& - 9437184 H_8^4 S^6 U + 4076863488 H_{12} H_8^3 S^7 U + 7247757312 H_8^4 S^7 U + 43486543872 H_8^4 S^8 U\\& + 2304 H_{12}^3 H_8 U^2 + 216 H_{12}^2 H_8^2 U^2 +100 H_{12} H_8^3 U^2 + 20 H_8^4 U^2 + 51840 H_{12}^2 H_8^2 S U^2\\& + 4032 H_{12} H_8^3 S U^2 + 2400 H_8^4 S U^2 + 1769472 H_{12}^2 H_8^2 S^2 U^2 - 165888 H_{12} H_8^3 S^2 U^2 + 4608 H_8^4 S^2 U^2 \\& + 11501568 H_{12} H_8^3 S^3 U^2 - 8699904 H_8^4 S^3 U^2 + 339738624 H_{12} H_8^3 S^4 U^2 - 336199680 H_8^4 S^4 U^2\\& - 3227516928 H_8^4 S^5 U^2 + 55296 H_{12} H_8^3 U^3 + 9216 H_8^4 U^3 + 884736 H_8^4 S U^3 \\& + 21233664 H_8^4 S^2 U^3 - 21024 H_{12}^3 H_8 V - 240 H_{12}^2 H_8^2 V + 288 H_{12} H_8^3 V - 179712 H_{12}^2 H_8^2 S V \\& + 59904 H_{12} H_8^3 S V - 72 H_{12}^3 S^2 V - 176 H_{12}^2 H_8 S^2 V - 6 H_{12} H_8^2 S^2 V- 8073216 H_{12}^2 H_8^2 S^2 V \\& + 2 H_8^3 S^2 V + 3760128 H_{12} H_8^3 S^2 V - 3072 H_{12}^3 S^3 V - 13952 H_{12}^2 H_8 S^3 V - 2656 H_{12} H_8^2 S^3 V \\& + 480 H_8^3 S^3 V + 60162048 H_{12} H_8^3 S^3 V - 239616 H_{12}^2 H_8 S^4 V \\& - 280576 H_{12} H_8^2 S^4 V + 36352 H_8^3 S^4 V - 3538944 H_{12}^2 H_8 S^5 V - 10887168 H_{12} H_8^2 S^5 V + 417792 H_8^3 S^5 V\\& - 166330368 H_{12} H_8^2 S^6 V - 59768832 H_8^3 S^6 V - 1358954496 H_{12} H_8^2 S^7 V - 2378170368 H_8^3 S^7 V \\& - 32614907904 H_8^3 S^8 V - 173946175488 H_8^3 S^9 V + 60 H_{12}^3 U V + 164 H_{12}^2 H_8 U V + 3 H_{12} H_8^2 U V \\& - 2 H_8^3 U V - 221184 H_{12} H_8^3 U V+ 2592 H_{12}^3 S U V + 13248 H_{12}^2 H_8 S U V + 1816 H_{12} H_8^2 S U V - 544 H_8^3 S U V \\ &+ 152064 H_{12}^2 H_8 S^2 U V + 202368 H_{12} H_8^2 S^2 U V - 52992 H_8^3 S^2 U V + 1990656 H_{12}^2 H_8 S^3 U V\\& + 6967296 H_{12} H_8^2 S^3 U V - 1916928 H_8^3 S^3 U V + 61046784 H_{12} H_8^2 S^4 U V \\& - 7077888 H_8^3 S^4 U V + 382205952 H_{12} H_8^2 S^5 U V + 509607936 H_8^3 S^5 U V \\& + 4076863488 H_8^3 S^6 U V + 55296 H_{12}^2 H_8 U^2 V - 1728 H_{12} H_8^2 U^2 V \\& + 480 H_8^3 U^2 V + 359424 H_{12} H_8^2 S U^2 V - 96768 H_8^3 S U^2 V + 21233664 H_{12} H_8^2 S^2 U^2 V \\& - 12607488 H_8^3 S^2 U^2 V - 201719808 H_8^3 S^3 U^2 V + 663552 H_8^3 U^3 V \\& - 126144 H_{12}^2 H_8 V^2 + 19584 H_{12} H_8^2 V^2 - 6 H_{12}^2 S V^2 - 3 H_{12} H_8 S V^2 + 940032 H_{12} H_8^2 S V^2 - 1440 H_{12}^2 S^2 V^2 \\& - 1776 H_{12} H_8 S^2 V^2 + 56 H_8^2 S^2 V^2 - 55296 H_{12}^2 S^3 V^2 - 143616 H_{12} H_8 S^3 V^2 - 2944 H_8^2 S^3 V^2\\& - 3428352 H_{12} H_8 S^4 V^2 - 1155072 H_8^2 S^4 V^2- 42467328 H_{12} H_8 S^5 V^2 - 63700992 H_8^2 S^5 V^2\\& - 1189085184 H_8^2 S^6 V^2 - 8153726976 H_8^2 S^7 V^2 + 648 H_{12}^2 U V^2 + 1020 H_{12} H_8 U V^2 - 128 H_8^2 U V^2 \\& + 31104 H_{12}^2 S U V^2 + 84672 H_{12} H_8 S U V^2 - 13632 H_8^2 S U V^2 + 1078272 H_{12} H_8 S^2 U V^2 \\& - 193536 H_8^2 S^2 U V^2 + 11943936 H_{12} H_8 S^3 U V^2 + 10616832 H_8^2 S^3 U V^2 + 127401984 H_8^2 S^4 U V^2 \\ &+ 331776 H_{12} H_8 U^2 V^2 - 65664 H_8^2 U^2 V^2 - 3151872 H_8^2 S U^2 V^2\\& - H_{12} V^3 - 216 H_{12} S V^3- 48 H_8 S V^3 - 17280 H_{12} S^2 V^3 - 9216 H_8 S^2 V^3 - 442368 H_{12} S^3 V^3\\& - 626688 H_8 S^3 V^3 - 17694720 H_8 S^4 V^3 - 169869312 H_8 S^5 V^3 + 2592 H_{12} U V^3 + 576 H_8 U V^3\\& + 124416 H_{12} S U V^3 + 55296 H_8 S U V^3 + 1327104 H_8 S^2 U V^3 - 12 V^4 - 1728 S V^4 \\& - 82944 S^2 V^4 - 1327104 S^3 V^4.\end{aligned}$$ Acknowledgement {#acknowledgement .unnumbered} =============== The idea of proof of $h_{15}\in M_{15}^+(\Gamma _0^{(1)}(4),\chi _{-4})$ using the twisting operator is due to Professor S. Böcherer. This makes it possible to prove Lemma \[Lem2\]. The author is supported by JSPS KAKENHI Grant Number JP18K03229. D. Choi, Dohoon, Y. Choie, T. Kikuta, Sturm type theorem for Siegel modular forms of genus $2$ modulo $p$, Acta Arith. 158 (2013), no. 2, 129-139. T. Dern, Hermitesche Modulformen zweiten Grades, Verlag Mainz, Wissenschaftsverlag, Aachen, 2001. T. Dern, A. Krieg, Graded rings of Hermitian modular forms of degree $2$, Manuscripta Math. 110 (2003), no. 2, 251-272. J.-I. Igusa, On the ring of modular forms of degree two over ${\mathbb{Z}}$, Amer. J. Math. 101 (1979), no. 1, 149-183. T. Kikuta, S. Nagaoka, On Hermitian modular forms mod $p$. J. Math. Soc. Japan 63 (2011), no. 1, 211-238. T. Kikuta, S.  Nagaoka, On the theta operator for Hermitian modular forms of degree 2, Abh. Math. Semin. Univ. Hambg. 87 (2017), no. 1, 145-163. T. Kikuta, S. Takemori, Sturm bounds for Siegel modular forms of degree 2 and odd weights, to apper in Math. Z. S. Nagaoka, S. Takemori, Theta operator on Hermitian modular forms over the Eisenstein field, to appear in Ramanujan J. A. Krieg, The Maass spaces on the Hermitian half-space of degree $2$. Math. Ann. 289 (1991), no. 4, 663-681. G. Shimura, Introduction to the arithmetic theory of automorphic functions. Reprint of the 1971 original. Publications of the Mathematical Society of Japan, 11. Kano Memorial Lectures, 1. Princeton University Press, Princeton, NJ, 1994. xiv+271

ArXiv

--- author: - Núria Sidro for the MAGIC Collaboration bibliography: - 'biblio.bib' title: First results of galactic observations with MAGIC --- Introduction: The MAGIC Telescope {#sec:intro} ================================= The Major Atmospheric Gamma Imaging Cherenkov (MAGIC) telescope [@performance] is a very high energy (VHE) $\gamma$-ray telescope, operating in an energy band from 100 GeV to 10 TeV, exploiting the Imaging Air Cherenkov technique. Located on the Canary Island of La Palma, at $28^\circ 45^\prime 30^{\prime\prime}$N, $17^\circ$ $52^\prime$ $48^{\prime\prime}$W and 2250 m above sea level. The telescope has a 17-m diameter tessellated parabolic mirror, and is equipped with a 3.5$^\circ$-3.8$^\circ$ field of view camera. See [@martinez] for a complete description of the instrument. In this work we show that MAGIC has the capability to contribute to the growing VHE -ray source catalogue by exploring the part of the Galactic sources observable from the Northern Hemisphere. The physic program developed with the MAGIC telescope includes both, topics of fundamental physics and astrophysics. In this paper we present the results regarding the observations of galactic targets. The results from extragalactic observations are presented elsewhere in these proceedings [@firpo]. The Crab nebula and pulsars {#sec:crab} =========================== The Crab nebula is a steady emitter at GeV and TeV energies, what makes it into an excellent calibration candle. This object has been observed extensively in the past over a wide range of wavelengths, up to nearly 100 TeV. Nevertheless, some of the relevant physics phenomena are expected to happen in the VHE domain, namely the spectrum showing an Inverse Compton (IC) peak close to 100 GeV, a cut-off of the pulsed emission somewhere between 10-100 GeV, and the verification of the flux stability down to the percent level. The existing VHE $\gamma$-ray experimental data is well described by electron acceleration followed by the IC scattering of photons generated by synchrotron radiation (synchrotron self Compton process). Along the first cycle of MAGIC’s regular observations, a significant amount of time has been devoted to observe the Crab nebula, both for technical and astrophysical studies. A sample of 12 hours of selected data has been used to measure with high precision the spectrum down to $\sim$100 GeV, as shown in Figure \[fig:crab\] [@wagner]. We have also carried out a search for pulsed $\gamma$-ray emission from Crab pulsar, albeit with negative results. The derived upper limits (95% C.L.) are 2.0$\times 10^{-10}$ ph s$^{-1}$cm$^{-2}$ at 90 GeV and 1.1$\times 10^{-10}$ ph s$^{-1}$cm$^{-2}$ at 150 GeV. We also carried out a search for pulsed $\gamma$-ray emission from two milisecond pulsars [@ona] PSR B1957+20 and PSR J0218+4232, albeit without positive result. The corresponding upper limits are $F_{\mathrm{PSR~B1957+20}} \sim 2.3 \times 10^{-11}$ and $F_{\mathrm{PSR~J0218+4232}} \sim 2.9 \times 10^{-11}$ ph s$^{-1}$cm$^{-2}$ for the steady emission and $F_{\mathrm{PSR~B1957+20}} \sim 5.1 \times 10^{-12}$ and $F_{\mathrm{PSR~J0218+4232}} \sim 6.5 \times 10^{-12}$ ph s$^{-1}$cm$^{-2}$ for the pulsed one. ![ Energy spectrum above 100 GeV from the Crab nebula measured by MAGIC in two different observation seasons.[]{data-label="fig:crab"}](gal_post_sidro_1_fig2.eps){width="7.5cm"} Supernova remnants {#sec:sn} ================== Shocks produced at supernova explosions are assumed to be the source of the galactic component of the cosmic ray flux [@zwicky]. The proof that this is the case could be provided by observations in the VHE domain. The rationale is that the hadronic component of the cosmic rays –enhanced close to their source, i.e. the SNR– should produce VHE $\gamma$-rays by the interaction with nearby dense molecular clouds. Although recent data seem to indicate that this is the case, it is difficult to disentangle the VHE component initiated by hadrons from that produced by Bremsstrahlung and IC processes by accelerated electrons. Therefore more data in the TeV regime together with multi-wavelength studies are needed to finally solve the long-standing puzzle of the origin of galactic cosmic rays. Within its program of observation of galactic sources, MAGIC has observed a number of supernova remnants. In particular, we are observing several of the brightest EGRET sources associated to SNRs, and the analysis of the acquired data is in progress. On the other hand, we have confirmed the VHE $\gamma$-ray emission from the SNRs HESS J1813-178 [@magic_hess1813] and HESS J1834-087 (W41) [@magic_hess1834]. Our results have confirmed SNRs as a well established population of VHE $\gamma$-ray emitters. The energy spectra measured by MAGIC are both well described by an unbroken power law and an intensity of about 10$\%$ of the Crab nebula flux. Furthermore, MAGIC has proven its capability to study moderately extended sources by observing HESS J1834-087. Interestingly, the maximum of the VHE emission for this object has been correlated with a maximum in the density of a nearby molecular cloud. Although the mechanism responsible for the VHE radiation remains yet to be clarified, this is a hint that it could be produced by high energy hadrons interacting with the molecular cloud. Galactic Center {#sec:gc} --------------- We have also measured the VHE $\gamma$-ray flux from the galactic center [@magic_gc]. The possibility to indirectly detect dark matter through its annihilation into VHE $\gamma$-rays has risen the interest to observe this region during the last years. Our observations have confirmed a point-like $\gamma$-ray excess whose location is spatially consistent with Sgr A\* as well as Sgr A East. The energy spectrum of the detected emission is well described by an unbroken power law of photon index $\alpha=-2.2$ and intensity about $10\%$ of that of the Crab nebula flux at 1 TeV. The power law spectrum disfavours dark matter annihilation as the main origin of the detected flux. There is no evidence for variability of the flux on hour/day time scales nor on a year scale. This suggests that the acceleration takes place in a steady object such as a SNR or a PWN, and not in the central black hole. The $\gamma$-ray binary LS I +61 303 {#sec:lsi} ==================================== ![ Smoothed maps of $\gamma$-ray excess events above 400 GeV around LS I +61 303 (from [@lsi]), for observations around periastron (A) and latter orbital phases (B). []{data-label="fig:lsi-skymap"}](gal_post_sidro_1_fig1.eps){width="10.5cm"} This $\gamma$-ray binary system is composed of a B0 main sequence star with a circumstellar disc, i.e. a Be star, located at a distance of $\sim$2 kpc. A compact object of unknown nature (neutron star or black hole) is orbiting around it, in a highly eccentric ($e=0.72\pm0.15$) orbit. LS I +61 303 was observed with MAGIC for 54 hours between October 2005 and March 2006 [@lsi]. The reconstructed $\gamma$-ray map is shown in Figure \[fig:lsi-skymap\]. The data were first divided into two different samples, around periastron passage (0.2-0.3) and at higher (0.4-0.7) orbital phases. No significant excess in the number of $\gamma$-ray events is detected around periastron passage, whereas there is a clear detection (9.4$\sigma$ statistical significance) at later orbital phases. Two different scenarios have been involved to explain this high energy emissions: the microquasar scenario where the $\gamma$-rays are produced in a radio-emitting jet; or the pulsar binary scenario, where they are produced in the shock which is generated by the interaction of a pulsar wind and the wind of the massive companion. See [@sidro] for more details. #### Acknowledgements. We thank the IAC for the excellent working conditions at the ORM in La Palma. The support of the German BMBF and MPG, the Italian INFN, the Spanish CICYT is gratefully acknowledged. This work was also supported by ETH research grant TH-34/04-3, and the Polish MNiI grant 1P03D01028.

ArXiv

--- abstract: | We discuss the fundamental constants in the Standard Model of particle physics, in particular possible changes of these constants on the cosmological time scale. The Grand Unification of the observed strong, electromagnetic and weak interactions implies relations between time variations of the finestructure constant $\alpha$ and the QCD scale $\Lambda_c$. The astrophysical observation of a variation of $\alpha $ implies a time variation of $\Lambda_c$ of the order of at least $10^{-15} / year$. Several experiments in Quantum Optics, which were designed to look for a time variation of $\Lambda_c$, are discussed.\ \ 1. Introduction 2. Fundamental Constants in the Standard Model 3. Does the Finestructure Constant depend on Time? 4. Grand Unification 5. Results from Quantum Optics 6. Conclusions and Outlook author: - | H. Fritzsch\ Universität München, Department für Physik, Munich, Germany title: The Fundamental Constants in Physics and their Time Dependence --- The Standard Model ================== The Standard Model of particle physics consists of 1. the theory of the strong interactions (QCD)[@Fritzsch], 2. the theory of the electroweak interactions, based on the gauge group $SU(2) \times U(1)$[@Glash]. The theory of QCD is an unbroken gauge theory, based on the gauge group $SU(3)$, acting in the internal space of ”color“. The basic fermions of the theory are the various quarks, which form color triplets. The 8 massless gauge bosons are $SU(3)$– octets, the gluons. The interactions of the quarks and gluons are dictated by the gauge properties of the theory. The quarks and gluons interact through the vertex $g_s \cdot \bar q \gamma_{\mu} \frac{\lambda_i}{2} q \cdot A^{\mu}_i$, where $q$ are the quark fields and $A^{\mu}_i$ the eight gluon fields. The eight $SU(3)$–matrices are denoted by $\lambda_i$. The strength of the coupling is given by the coupling constant $g_s$. QCD is a non–Abelian gauge theory. This implies that there is a direct coupling of the gluons among each other. There is a trilinear coupling, proportional to $g_s$, and a quadrilinear coupling, proportional to $g^2_s$. It is assumed, but thus far not proven, that the QCD gauge interaction leads to a confinement of all colored quanta, in particular of the quarks and the gluons. Replacing the continuous space–time continuum by a lattice, one can solve the QCD field equations with the computer. The results confirm the confinement hypothesis, but it remains unclear whether one can perform the limit, in which the lattice spacing goes to zero, without changing the result. The experiments strongly support the QCD theory. It has the property of asymptotic freedom: the strength of the interactions goes to zero on a logarithmic scale at high energies. At low energies the interaction gets large, and the confinement property might indeed be true. The equations, describing the renormalization of the coupling constant, give for $\alpha_s = \frac{g_s \, ^2}{4 \pi}$: $$\begin{aligned} \mu \cdot \frac{\partial \alpha_s}{\partial_{\mu}} & = & - \frac{\beta_0}{2 \pi} \alpha_s^2 - \frac{\beta_1}{4 \pi^2} \alpha_s^3 - \ldots \nonumber \\ \beta_0 & = & 11 - \frac{2}{3} n_f \nonumber \\ \beta_1 & = & 51 - \frac{19}{3} n_f\end{aligned}$$ ($n_f$: number of quark flavors with mass less than the energy scale $\mu$ ). Since the interactions gets weak at high energies, the quarks and also the gluons appear nearly as pointlike objects at small distances. This has been observed in the experiments of deep inelastic scattering of electrons, myons and neutrinos off nuclear targets. Since the strong coupling constant at high energies is small, but not zero, one expects small departures from the pointlike structure, i. e. violations of the scaling behaviour of the cross-sections. This has been observed in many experiments, supporting the QCD theory. The value of the QCD coupling constant $\alpha_s = \frac{g_s \, ^2}{4 \pi}$ depends on the energy. One has found in the analysis of the scaling violations: $$\alpha_s \left( M_z^2 \right) \approx 0.118 \pm 0.011$$ ($M_z$: mass of the $Z$–boson, $M_z \cong 91.2 \, \,\, {\rm GeV}$). One can express $\alpha_s (\mu)$ as a function of the QCD scale parameter $\Lambda_c$: $$\begin{aligned} \alpha_s (\mu )^{-1} & \approx & \left( \frac{\beta_0}{4 \pi} \right) ln \left( \frac{\mu^2}{\Lambda_c^2} \right) \nonumber \\ \beta_0 & = & \left( 11 - \frac{2}{3} n_f \right)\end{aligned}$$ The experiment give: $$\Lambda_c \approx 217^{+25}_{-23} \, \, \, {\rm MeV} \, .$$ The standard electroweak model is based on the gauge group $SU(2) \times U(1)$. There are three $W$–bosons, related to the $SU(2)$ group, and one $B$–boson, related to the $U(1)$–group. The lefthanded quarks and leptons are $SU(2)$–doublets, the righthanded ones are singlets. Thus parity is violated maximally. The gauge invariance of the $SU(2) \times U(2)$–model is broken spontaneously by the ”Higgs“–mechanism[@Higgs]. The masses for the gauge bosons are generated by the spontaneous symmetry breaking. Goldstone bosons end up as longitudinal components of the gauge bosons. In the ”Higgs“ mechanism there exists a self–interacting complex doublet of scalar fields. In the process of symmetry breaking the neutral component of the scalar doublet acquires a vacuum expectation value $v $, which is directly related to the Fermi constant of the weak interactions. Thus the vacuum expectation value is known from the experiments, if the theory is correct: $$v \cong 246 \, \, \, {\rm GeV}$$ This energy sets the scale for the symmetry breaking. Three massless Goldstone bosons are generated, which are absorbed to give masses to the $W^+, W^-$ and $Z$–bosons. The remaining component of the complex doublet is the ”Higgs“–boson, thus far a hypothetical particle. It would be the elementary scalar boson in the Standard Model. One hopes to find this particle with the new accelerator LHC at CERN, which starts to operate in 2008. In the $SU(2) \times U(1)$ model there are two neutral gauge bosons, which are mixtures of $W_3$ and $B$, the $Z$–boson and the photon. The associated mixing angle $\Theta_w$ is a fundamental parameter which has to be fixed by experiment. It is given by the $Z$–mass, the Fermi constant and $\alpha$: $$sin^2 \Theta_w \cdot cos^2 \Theta_w = \frac{\pi \alpha \left( M_z \right)}{\sqrt{2} \cdot G_F \cdot M_Z^2} \, .$$ Using $\alpha \left( M_Z \right)^{-1} = 128,91 \pm 0.02$, one finds: $$sin^2 \Theta_w = 0.23108 \pm 0.00005 \, .$$ Note that $sin^2 \Theta_w $ is also related to the mass ratio $M_W / M_Z$: $$\begin{aligned} sin^2 \Theta_w & = & 1 - M_W^2 / M_Z^2 \nonumber \\ M_Z & = & M_W / cos \Theta_w \, . \end{aligned}$$ In the Standard Model the interactions depend on 28 fundamental constants, e. g. on the finestructure constant $\alpha $. There were many attempts to get further insights, how these constants arise and whether one might calculate them, but thus far all without success. At the moment we can only determine these constants by the experiments. In physics we are dealing with the laws of nature, but little thought is given to the boundary condition of the universe, related directly to the Big Bang. We do not know at the moment, what role is played by the fundamental costants, but these constants could form a bridge between these boundary conditions and the local laws of nature. Thus they would be relicts of the Big Bang. What are the fundamental constants? Some physicists believe that at least some of them are just cosmic accidents, fixed by the dynamics of the Big Bang. If the Big Bang would be repeated, these constants would take different values. Thus the constants are arbitrary, depending on details of the Big Bang. Obviously in this case there is no way to calculate the fundamental constants. Indeed, some fundamental constants might be cosmic accidents, but it is unlikely, that this is the case for all the 28 fundamental constants. New interactions, discovered e. g. with the new LHC–accelerator at CERN, might offer a way to calculate at least some of the fundamental constants. We also do not understand, why the fundamental constants are constant in time. Small time variations are indeed possible and even suggested by astrophysical experiments. In the theory of superstrings one expects time variations of the fundamental constants, in particular of the finestructure constant, of the QCD scale parameter, and of the weak interaction coupling constants. If one finds that the fundamental constants are changing in time, then they are not just numbers, set into the fabric of basic physics – rather they would be dynamical quantities which change according to some deeper laws that we have to understand. These laws would be truly fundamental and may even point the way to a unified theory including gravity. Fundamental Constants in the Standard Model =========================================== The Standard Model is the successful theory of the observed particle physics phenomena. However, the electroweak interactions and QCD depend on 28 fundamental constants, and within the Standard Model there is no way to calculate these constants or to derive relations between them. The most famous fundamental constant is the finestructure constant $\alpha$, introduced in 1916 by Arnold Sommerfeld: $$\alpha = \frac{e^2}{\hbar c} \, .$$ In this constant the electromagnetic coupling $e$ enters, as well as the Planck constant $h$, the constant of the quantum physics, and the speed of light $c$, the basic constant of the theory of relativity. Sommerfeld realized that $\alpha$ is a dimensionless number, close to the inverse of the prime number 137. The experiments give the following value for $\alpha^{-1}$: 137,03599907(9). As early as 1905 Max Planck pointed out in a letter to Paul Ehrenfest that $h$ and $\frac{e^2}{c}$, both having the physical dimension of an action, were roughly of the same magnitude. Adams and Lewis[@Adams] proposed in 1914 a relation between $h, e^2$ and $c$: $$h = \frac{e^2}{c} \left( \frac{ 8 \pi^5}{15} \right)^{1/3} \, .$$ This can be rewritten as follows: $$\alpha^{-1} = \left( \frac{8 \pi^5}{15} \right)^{1/3} \, .$$ The numerical value for $\alpha ^{-1}$ is 137,348, which is rather close to the modern value 137,036. Heisenberg proposed in 1936: $$\alpha^{-1} = 2^{-4} \, 3^{-3} \pi \, ,$$ which gives $\alpha^{-1} = 137,51$. In 1971 Wyler[@Wyler] published the following expression for $\alpha$, $$\alpha = \frac{9}{8 \pi ^4} \left( \frac{\pi ^ 5}{2^4 \cdot 5!} \right)^{1/4} \, ,$$ which gives $\alpha^{-1} = 137,03608$ and agrees well with the experimental value. Richard P. Feynman wrote about the finestructure constant[@Feyn]: ”It has been a mystery ever since it was discussed more than fifty years ago, and all good theoretical physicists put this number up on their wall and worry about it. Immediately you would like to know where this number for a coupling comes from: is it related to $\pi $ or perhaps to the base of the natural logarithms? Nobody knows. It’s one of the greatest mysteries of physics: a magic number that comes to us with no understanding by man …”. Based on $\alpha$ the theory of Quantum Electrodynamics (QED) was developed, merging electrodynamics, quantum theory and relativity theory. It is the most successful theory in science, tested with a precision of $10^{-7}$. In quantum field theory the number, describing the strength of the interactions, is not a fixed constant, but a function of the energy involved. This can be easily understood. The groundstate of a system, described by a quantum field theory, is filled with virtual pairs of quanta, e. g. with $e^+e^-$–pairs in QED. Thus an electron is surrountded by $e^+e^-$–pairs. The virtual electrons are repelled by the electrons, the virtual positrons are attracted, and the electron charge is partially shielded by the virtual positrons. Thus at relatively large distances (larger than the Compton wavelength of an electron $\lambda_c = \frac{h}{m_ec} \approx 2.43 \cdot 10^{-12} \, m$) the electron charge is smaller than at distances less than $\lambda_c$. The dependence on the energy is described by the renormalization group equations, first considered by Murray Gell–Mann and Francis Low[@Gell]: $$\frac{d}{d \, ln \left(q/M \right)} e(q) = \beta(e) \, ,$$ where $$\beta(e) = \frac{e^3}{12 \pi ^ 2} + \ldots \, .$$ In QED at high energies one has to include not only virtual $e^+e^-$–pairs, but also the $\mu^+ \mu^-$– and $\tau^+ \tau^-$–pairs, as well as the quark–antiquark–pairs. Doing so, one finds that the finestructure constant $\alpha$ at the mass of the $Z$–boson should be the inverse of 127.8, in very good agreement with the experimental data taken with the LEP–accelerator. Another fundamental parameter of the Standard Model is the proton mass. In the theory of the strong interaction QCD the proton mass is a parameter, which can be calculated as a function of the QCD scale parameter $\Lambda_c$ and of the light quark masses. The QCD scale parameter has been determined by the experiments: $$\Lambda_c = 217 \pm 25 \, \,\, {\rm MeV} \, .$$ The QCD–Theory gives a very clear picture of mass generation. In the limit, where the quark masses are set to zero, the nucleon mass is just the confined field energy of the gluons and quarks and can be written as: $$M(Nucleon) = const. \cdot \Lambda_c \, .$$ The [*const.*]{} has been calculated using the lattice approach to QCD. It is about 3,9, predicting a nucleon mass in the limit $m_q = 0$ of about 860 MeV. The observed nucleon mass (about 940 MeV) is higher, due to the contributions of the mass terms of the light quarks $u, d, s$, which in reality are not massless. The mass of the proton can be decomposed as follows: $$\begin{aligned} M_p & = & \, \, \, {\rm const.} \, \, \, \Lambda_c + \\ & & < p \mid m_u \bar u u \mid p > + < p \mid m_d \bar dd \mid p > + < p \mid m_s \bar s s \mid p > + c_{\rm elm} \, \, \, \cdot \Lambda_c \, . \nonumber\end{aligned}$$ The last term describes the electromagnetic self–energy of the proton. This self–energy is also proportional to the QCD–scale $\Lambda $. Calculations give: $$c_{\rm elm} \cdot \Lambda_c \approx 2.0 \, \, \, {\rm MeV} \, .$$ The up–quark mass term contributes about 20 MeV to the proton mass, according to the chiral perturbation theory, the d-.quark mass term about 19 MeV. Thus the $d$–contribution to the proton mass is about as large as the $u$–contribution, although there are two $u$–quarks in the proton, and only one $d$–quark. This originates from the fact that the $d$–mass is larger than the $u$–mass. In chiral perturbation theory the $u$– and $d$–masses can be estimated: $$\begin{aligned} m_u & \approx & 3 \pm 1 \, \, \, {\rm MeV} \nonumber \\ m_d & \approx & 6 \pm 1.5 \, \, \, {\rm MeV} \, .\end{aligned}$$ These masses are normalized at the scale $\mu = 2 \, \, \, {\rm GeV}$. Note that quark masses are not the masses of free particles, but of the confined quarks. The quark masses depend on the energy scale $\mu $, relevant for the discussion. The mass of the strange quark can also be estimated in the chiral perturbation theory. On finds for $\mu = 2 \, \, \, {\rm GeV}$: $$m_s \approx 103 \pm 20 \, \, \, {\rm MeV} \, .$$ Thus the mass of the strange quark is about 20 times larger than the $d$–mass. Although there are no valence $s$–quarks in the proton, the numerous $\bar ss$–pairs contribute about 35 MeV to the proton mass, i. e. more than the $\bar uu$– or $\bar dd$–pairs, due to the large ratio $m_s / m_d$. The heavy quarks $\bar cc$ and $\bar bb$ do not contribute much to the proton mass. We can decompose the proton mass as follows: $$\begin{aligned} M_p & = & 938 \, \, \, {\rm MeV} \nonumber \\ & = & (862 \qquad + \qquad 20 \qquad + \qquad 19 \qquad + \qquad 35 \qquad + \qquad 2) \, \, \, {\rm MeV} \nonumber \\ & & \, \, \, \, \uparrow \, \, \, \, \, \, \, \, \hspace{2cm} \uparrow \, \, \, \, \, \hspace{2cm} \uparrow \, \, \, \, \, \hspace{2cm} \uparrow \, \, \, \hspace{2cm} \uparrow \nonumber \\ & & QCD \, \, \, \hspace{1cm} u-quarks \, \, \, \hspace{0.5cm} d-quarks \, \, \, \hspace{0.4cm} s-quarks \, \, \, \, \, \hspace{0.2cm} QED\end{aligned}$$ The masses of the heavy quarks $c$ and $b$ can be estimated by considering the spectra of the particles, containing $c$– or $b$–quarks, e. g. the charm–mesons or the $B$–mesons. One finds: $$\begin{aligned} m_c (m_c) & \approx & 1,24 \pm 0.09 \, \, \, {\rm GeV} \nonumber \\ m_b (m_b) & \approx & 4,2 \pm 0.07 \, \, \, {\rm GeV}\end{aligned}$$ The dark corner of the Standard Model is the sector of the fermion masses. There are the six quark masses, the three charged fermion masses, the three neutrino masses, the four flavor mixing parameters of the quarks and the six flavor mixing parameters of the leptons (if neutrinos are Majorana particles). Altogether these parameters make up 22 of the 28 fundamental constants. The remaining six constants are $\Lambda_c, \alpha_2, \alpha_1$, the mass of the $W$–boson, the mass of the hypothetical ”Higgs“–boson and the constant $G$ for gravity. What are the fermion masses? We do not know. They might also be due to a confined field energy, but in this case the quarks and leptons would have to have a finite radius, as e. g. in composite models. The masses would be generated by a new interaction. The experiments give a limit on the internal radius of the leptons and quarks, which is of the order of $10^{-17}$ cm. In the Standard Model the masses of the leptons and quarks are generated spontaneously, like the $W$ and $Z$–masses. Each fermion couples with a certain strength to the scalar ”Higgs“–boson via a Yukawa coupling. The fermion mass is then given by: $$m(fermion) = g \cdot V \, ,$$ where $V$ is the vacuum expectation value of the ”Higgs“–field. For the electron this Yukawa coupling constant must be very small, since $V$ is about 246 GeV: $$g(electron) = 0,00000208 \, .$$ Nobody understands, why this constant is so tiny. The problem of fermion masses remains to be solved. It seems to be the most fundamental problem we are facing at the present time. If one is interested only in stable matter, as e. g. in solid state physics, only seven fundamental constants enter: $$G, \Lambda, \alpha, m_e, m_u, m_d, m_s \, .$$ The mass of the $s$–quark has been included, since the $\left( \bar ss \right)$–pairs contribute to the nucleon mass about 40 MeV. These seven constants describe all atoms and stable nuclei. It is possible, that there exist relations between the fundamental constants. Relations, which seem to work very well, are the relations between the flavor mixing angles and the quark masses, which where predicted some time ago[@Frit]: $$\begin{aligned} \Theta_u & = & \sqrt{m_u / m_c} \nonumber \\ \Theta_d & = & \sqrt{m_d / m_s} \, .\end{aligned}$$ Similar relations also exist for the neutrino masses and the associated mixing angles[@Frixi]. These relations are obtained if both for the $u$–type and for the $d$–type quarks the following mass matrices are relevant (texture 0 matrices): $$M = \left( \begin{array}{lll} 0 & A & 0 \\ A^* & C & B \\ 0 & B^* & D \end{array} \right) \, .$$ It would be interesting to know whether these mass matrices are indeed realized in nature. Relations like the ones discussed above would reduce the number of fundamental constant from 28 to about 20. Does the Finestructure Constant depend on Time? =============================================== Recent observations in astrophysics[@Webb] indicate that the finestructure constant $\alpha$ is a slow function of time. Billions of years ago it was smaller than today. A group of researchers from Australia, the UK and the USA analysed the spectra of distant quasars, using the Keck telescope in Hawaii. They studied about 150 quasars, some of them about 11 billion lightyears away. The redshifts of these objects varied between 0.5 and 3.5. This corresponds to ages varying between 23% and 87% of the age of our universe. They used the ”many multiplet method“ and in particular studied the spectral lines of iron, nickel, magnesium, zinc and aluminium. It was found that $\alpha$ changes in time: $$\frac{\Delta \alpha}{\alpha} = (- 0.72 \pm 0.18 ) \cdot 10^{-5} \, .$$ Taking into account the ages of the observed quasars, one concludes that in a linear approximation the absolute magnitude of the relative change of $\alpha $ must be: $$\left| \frac{d \alpha / dt }{\alpha} \right| \approx 1.2 \cdot 10^{-15} / year \, .$$ We like to mention that recent observations of quasar spectra, performed by different groups, seem to rule out a time variation of $\alpha $ at the level given above[@Chasid]. The idea that fundamental constants are not constant in time, but have a cosmological time dependence, is not new. In the 1930s P. Dirac[@Dirac] discussed a time variation of Newtons constant $G$. Dirac argued that $G$ should vary by about a factor 2 during the lifetime of the universe. The latter is now fixed to about $1.4 \cdot 10^{10}$. Only a few years ago it was found that a possible time variation of $G$ must be less than $10^{-11}$ per year, and Dirac’s hypothesis is now excluded. In the 1950s L. Landau discussed a possible time variation of the finestructure constant $\alpha $ in connection with the renormalization of the electric charge[@Land]. In the 1970s French nuclear physicists discovered that about 1.8 billion years ago a natural reactor existed in Gabon, West–Africa, close to the river Oklo. About 2 billion years ago uranium -235 was more abundant than today (about 3,7%). Today it is only 0,72%. The water of the river Oklo served as a moderator for the reactor. The natural reactor operated for about 100 million years. The isotopes of the rare earths, for example the element Samarium, were produced by the fission of uranium. The observed distribution of the isotopes today is consistent with the calculation, assuming that the isotopes were exposed to a strong neutron flux. Especially the reaction of Samarium with a neutron is interesting: $$Sm(149) + n \rightarrow Sm (150) + \gamma \, .$$ It was known that the very large cross–section for this reaction (about $60 \ldots 90$ kb) is due to a nuclear resonance just above threshold. The energy of this resonance is very small: $E = 0.0973 \, \, \, {\rm eV}$. The position of this resonance cannot have changed in the past 2 billion years by more than 0.1 eV. Suppose $\alpha $ has changed during this time. The energy of the resonance depends also on the strength of the electromagnetic interaction. One concludes: $$\frac{ \alpha \left( Oklo) \right) - \alpha (now)}{\alpha (now)} < 10^{-7} \, .$$ Taking into account the two billion years, the relative change of $\alpha $ per year must be less than $10^{-16}$ per year, as estimated by T. Damour and F. Dyson[@Dam]. This conclusion is correct only if no other fundamental parameters changed in the past two billion years. If other parameters, like the strong interaction coupling constant, changed also, the constraint mentioned above does not apply. The Oklo constraint for $\alpha $ is not consistent with the astrophysical observation for the relative changes of $\alpha $ of order $10^{-15}$ per year. No problem exists, however, if other fundamental ”constants“ also changed. Below we shall see that this was indeed the case. Recently one has also found a time change of the mass ratio $$\mu = \frac{M(proton)}{m(electron)} \, .$$ One observed the light from a pair of quasars, which are 12 billion light years away from the earth[@Rein]. The light was emitted, when the universe was only 1.7 billion years old. The study of the spectra revealed, that the ratio $\mu $ has changed in time: $$\frac{\Delta \mu}{\mu} \approx \left( 2 \pm 0.6 \right) \cdot 10^{-5} \, .$$ Taking into account the lifetime of 12 billion years, the change of $\mu $ per year in a linear approximation would be $10^{-15}$ / year. We shall return to this time change of $\mu $ later. Grand Unification ================= In the Standard Model we have three basic coupling constants, one for QCD $\left( \alpha_3 = \alpha_s \right) $ and two for the electroweak interactions, based on the gauge group $SU(2) \times U(1)$: $\alpha_2$ and $\alpha_1$. The gauge group of the Standard Model is $SU(3)_c \times SU(2) \times U(1)$. The three gauge interactions are independent of each other. Since 1974 the idea is discussed that the gauge group of the Standard Model is a subgroup of a larger simple group, and the three gauge interactions are embedded in a Grand Unified Theory (GUT). Such a Grand Unification implies that $\alpha_3, \alpha_2$ and $\alpha_1$ are related and can be expressed in terms of the unified coupling constant $\alpha_{un}$ and the energy scale of the unification $\Lambda_u$. The simplest theory of Grand Unification is a theory based on the gauge group $SU(5)$[@Geor]. The quarks and leptons of one generation can be described by two $SU(5)$–representations. Let us consider the 5–representation of $SU(5)$. After the breakdown of $SU(5)$ to $SU(3) \times SU(2) \times U(1)$ we obtain: $$\begin{aligned} 5 & \rightarrow & (3,1) + (1,2) \nonumber \\ \bar 5 & \rightarrow & (\bar 3,1) + (1,2) \, .\end{aligned}$$ Thus the 5–representation contains a color triplet, which is a singlet under $SU(2)$, and a color singlet, which is a $SU(2)$–doublet. We can identify the associated particles, neglecting the flavor mixing, and obtain for the first lepton–quark family: $$(5) = \left( \begin{array}{l} \bar d_r\\ \bar d_g\\ \bar d_b\\ \nu_e\\ e^- \end{array} \right) \, .$$ The representation with the next higher dimension is the 10–representation, which is an antisymmetric second–rank tensor. The 10–representation decomposes after the symmetry breaking as follows: $$(10) \rightarrow (5,1) = (3,2) + (1,1)$$ In terms of the lepton and quark fields of the first generation we can write the 10–representation (a $5 \times 5$–matrix) as follows: $$(10) = \frac{1}{\sqrt{2}} \left( \begin{array}{ccccc} 0 & \bar u_b & - \bar u_g & - \bar u_r & - \bar d_r \\ - \bar u_b & 0 & \bar u_r & \- \bar u_g & - \bar d_g \\ \bar u_g & - \bar u_r & 0 & - \bar u_b & - \bar d_b \\ u_r & u_g & u_b & 0 & e^+\\ d_r & d_g & d_b & -e^+ & 0 \, . \end{array} \right) \, .$$ Combining these two representations, one finds precisely the lepton and quarks, forming one generation: $$\bar 5 + 10 \rightarrow (3,2) + 2 \left( \bar 3,1 \right) + \left(1,2 \right) + (1,1) \, .$$ For the first generation one has: $$\bar 5 + 10 \rightarrow \left( {u \atop d} \right)_L + \bar u_L + \bar d_L + \left( {\nu_e \atop e^-} \right)_L + e^+_L \, .$$ The second and third generation are analogous. The unification based on the gauge gruop $SU(5)$ has a number of interesting features: 1. The electric charge is quantized. $$t r Q = O \rightarrow \, Q(d) = \frac{1}{3} \, Q \left( e^- \right)$$ 2. At some high mass scale $\Lambda_{un}$ the gauge group of the Standard Model turns into the group $SU(5)$, and there is only one single gauge coupling. The three coupling constants $g_3, g_2, g_1$ for $SU(3), SU(2)$ and $U(1)$ must then be of the same order of magnitude, related to each other by constants. The rather different values of the coupling constants $g_3, g_2, g_1$ at low energies must be due to renormalization effects. This would also give a natural explanation of why the strong interactions are strong and the weak interactions are weak. It has to do with the size of the corresponding group. Apart from normalization constants the three coupling constants $g_3, g_2$ and $g_1$, are equal at the unification mass $\Lambda_{un}$. Thus the $SU(2) \times U(1)$ mixing angle, given by $tan \Theta_w = \frac{g_1}{g_2}$, is fixed at or above $\Lambda_{un}$: $$sin^2 \Theta_w = tr T_3^2 / tr Q^2 = \frac{3}{8} \, .$$ At an energy scale $\mu << \Lambda_{un}$ the parameter $sin^2 \Theta$ changes along with the three coupling constants: $$\begin{aligned} \frac{sin^2 \Theta_w}{\alpha} - \frac{1}{\alpha_s} & = & \frac{11}{6 \pi} \, ln \, \left( \frac{M}{\mu} \right) \nonumber \\ \alpha / \alpha_s & = & \frac{3}{10} \left( 6 sin^2 \Theta_w - 1 \right) \, . \end{aligned}$$ At $\mu = M_z$ the electroweak mixing angle has been measured: $sin^2\Theta_w \left( M_u \right) \cong 0.2312$. Note that above the unification energy $\alpha$ and $\alpha_s$ are related: $$\alpha / \alpha_s = 3/8 \, .$$ This relation can be checked by experiment. In order to get a rough agreement between the observed values for $g_3, g_2$ and $g_1$ and the values predicted by the $SU(5)$ theory, one can easily see that the unification scale must be very high. Note that $$\begin{aligned} ln \left( \frac{M}{\mu}\right) & = & \frac{6 \pi}{11} \left( \frac{sin^2 \Theta_w}{\alpha} - \frac{1}{\alpha_s} \right) \nonumber \\ \mu & = & M_Z \nonumber \\ ln \left( M / M_Z \right) & \cong & 39,9 \nonumber \\ M & \approx & 2 \cdot 10^{15} \, \, \, {\rm GeV} \, .\end{aligned}$$ However, the precise values of the three coupling constants, determined by the LEP–experiments, disagree with the $SU(5)$ prediction. The three coupling constants do not converge at high energies to a single coupling constant $\alpha_{un}$. However, a convergence takes place, if supersymmetric particles are added. Supersymmetry implies that for each fermion a boson is added (s–leptons, s-quarks), and for each boson a new fermion is introduced (photino, etc.). These new particles are not observed in the experiments. It is assumed that they have a mass of about 1 TeV or more. The new particles contribute to the renormalization of the gauge coupling constants at high energies (about 1 TeV), and a convergence of the three coupling constants taken place. Thus a supersymmetric version of the $SU(5)$–theory is consistent with the experiments. In theories of Grand Unification like the $SU(5)$–theory one has typically quarks, antiquarks and leptons in one fermion representation. This implies that the proton can decay, e. g. $p \rightarrow e^+ \pi^0$. The lifetime depends on the mass scale for the unification. For $\Lambda_{un} = 5 \cdot 10^{14} \, \, \, {\rm GeV}$ in the $SU(5)$–theory without supersymmetry one finds $10^{30}$ years for the proton lifetime. The experimental lower limit is about $10^{31}$ years. There is a natural embedding of a group $SU(n)$ into $SO(2n)$, corresponding to the fact that $n$ complex numbers can be represented by $2n$ real numbers. Thus one may consider to use the gauge group $SO(10)$ instead of $SU(5)$. This was discussed first in 1975 by P. Minkowski and the author[@Frimin]. The fermions of one generation are described by a 16–dimensional spinor representation of $SO(10)$. Since $SU(5)$ is a subgroup of $SO(10)$, one has the following decomposition: $$16 \rightarrow \bar 5 + 10 + 1 \, .$$ Thus the fermions of the $SU(5)$–theory are obtained, plus one additional fermion (per family). This state is an $SU(5)$–singlet and describes a lefthanded antineutrino field. Using the leptons and quarks of the first generation we can write the 16–representation as follows in terms of lefthanded fields: $$(16) = \left( \begin{array}{lllllllll} \bar \nu_e & \bar u_r & \bar u_g & \bar u_b & \vdots & u_r & u_g & u_b &\nu_e \\ e^+ & \bar d_r & d_y & \bar d_b & \vdots & d_r & d_g & d_b &e^- \end{array} \right)$$ An interesting feature of the $SO(10)$–theory is that the gauge group for the electroweak interactions is larger than in the $SU(5)$–theory. $SO(10)$ has the subgroup $SO(6) \times SO(4)$. Since $SO(4)$ is isomorphic into $SU(2) \times SU(2)$, one finds: $$SO(10) \rightarrow SU(4) \times SU(2)_L \times SU(2)_R \, .$$ The group $SU(4)$ must contain the color group $SU(3)^c$. The 16–representation of the fermions decomposes under $SU(4)$ into two 4–representations. These contain three quarks and one lepton, e. g. $\left( d_r, d_g, d_b \right)$ and $e^-$. One can interpret the leptons as the fourth color. However, the gauge group $SU(4)$ must be broken at high energies: $$SU(4) \rightarrow SU(3) \times U(1) \, .$$ We obtain at low energies the gauge group $$SU(3)^c \times SU(2)_L \times SU(2)_R \times U(1) \, .$$ However, the masses of the gauge bosons for the group $SU(2)_R$ must be much larger than the observed $W$–bosons, related to the group $SU(2)_L$. In the $SU(5)$–theory the minimal number of fermions of the Standard Model is included. But in the $SO(10)$–theory a new righthanded neutrino is added (the $\bar \nu_{eL}$–particle). This righthanded fermion is interpreted as a heavy Majorana particle. In this case a mass for the lefthanded neutrino is generated by the ”see–saw“–mechanism[@Yana]. Thus in the $SO(10)$–theory the neutrinos are massive, while in the $SU(5)$–theory they must be massless. The $SO(10)$–theory is much more symmetrical than the $SU(5)$–theory, and it is hard to believe that Nature would stop at $SU(5)$, if Nature has chosen to unify the basic interactions. In the $SO(10)$–theory there is one additional free parameter, related to the masses of the righthanded $W$–bosons. Since righthanded charged currents are not observed, the masses of the associated $W$–bosons must be rather high, at least 300 GeV. Since there is this new parameter $M_R$ in the $SO(10)$–theory, it can be chosen such that the coupling constant converges at very high energies, without using supersymmetric particles. If one chooses $M_R \sim 10^9 \ldots 10^{11}$ GeV, the convergence occurs. The idea of Grand Unification leads to the reduction of the fundamental constants by one. The three gauge coupling constants of the Standard Model can be expressed in terms of a unified coupling constant $\alpha_u$ at the energy $\Lambda_u$, where the unification takes place. The three coupling constants are replaced by $\alpha_u$ and $\Lambda_u$. In a Grand Unified Theory the three coupling constants of the Standard Model are related to each other. If e. g. the finestructure constant shows a time variation, the other two coupling constants should also vary in time. Otherwise the unification would not be universal in time. Knowing the time variation of $\alpha $, one should be able to calculate the time variation of the other coupling constants. We shall investigate only the time change of the QCD coupling constant $\alpha_s $. We can also calculate the time change of the weak coupling constant, but this would be useless, since we have no information about the weak coupling constant billions of years ago, at least not about small changes of the order of $10^{-4}$ or less. We use the supersymmetric $SU(5)$–theory to study the time change of the coupling constants[@Cal; @Lang]. The change of $\alpha $ is traced back to a change of the unified coupling constant at the energy of unification and to a change of the unification energy. These changes are related to each other: $$\frac{1}{\alpha} \frac{\dot{\alpha}}{\alpha} = \frac{8}{3} \cdot \frac{1}{\alpha_s} - \frac{10}{\pi } \frac{\dot \Lambda_{un}}{\Lambda_{un}} \, .$$ We consider the following three scenarios: 1. $\Lambda_{un}$ is kept constant, $\alpha_u = \alpha_u(t)$. Then we obtain: $$\frac{1}{\alpha } \frac{\dot \alpha}{\alpha} = \frac{8}{3} \frac{1}{\alpha_s} \frac{ \dot{\alpha}_s}{\alpha_s} \, .$$ Using the experimental value $\alpha_s \left( M_Z \right) \approx 0.121$, we find for the time variation of the QCD scale[@Cal]:\ $$\begin{aligned} \frac{\dot \Lambda}{\Lambda} & \approx & R \cdot \frac{\dot{\alpha}}{\alpha} \nonumber \\ \nonumber\\ R & \approx & 38 \pm 6 \, .\end{aligned}$$ The uncertainty in $R$ comes from the uncertainty in the determination of the strong interaction coupling constant $\alpha_s$. A time variation of the QCD scale $\Lambda$ implies a time change of the proton mass and of the masses of all atomic nuclei. The change of the nucleon mass during the last 10 billion years amounts to about 0.3 MeV. In QCD the magnetic moments of the nucleon and of the atomic nuclei are inversely proportional to the QCD scale parameters $\Lambda $. Thus we find for the nuclear magnetic moments: $$\frac{\dot{\mu}}{\mu} = \frac{\frac{d}{dt} \left(\frac{1}{\Lambda} \right)}{\Lambda} = - \frac{\dot{\Lambda}}{\Lambda} = - R \cdot \frac{\dot{\alpha}}{\alpha} \, .$$ Taking the astrophysics result for $\left( \dot{\alpha} / \alpha \right)$, we would obtain: $$\frac{\dot{\Lambda}}{\Lambda} \approx 4 \cdot 10^{-14} / yr \, .$$ 2. The unified coupling constant is kept invariant, but $\Lambda_{un}$ changes in time. In that case we find[@Osk]: $$\frac{\dot{\alpha}}{\alpha} \cong - \alpha \cdot \frac{10}{\pi} \frac{\dot{\Lambda}_{un}}{\Lambda_{un}}$$ and $$\frac{\dot{\Lambda}}{\Lambda} \approx - 31 \cdot \frac{\dot{\alpha}}{\alpha} \, .$$ The change of the unification mass scale $\Lambda_{un}$ can be estimated, taking as input the time variation of the finestructure constant $\alpha $. One finds that this mass was 8 billion years ago about $8 \cdot 10^{12}$ GeV higher than today. In a linear approximation $\Lambda_{un}$ is decreasing at the rate $$\dot{\Lambda}_{un} / \Lambda_{un} \approx - 7 \cdot 10^{-14} / yr \, .$$ The relative changes of $\Lambda $ and $\alpha $ are opposite in sign. While $\alpha $, according to ref. [@Webb], is increasing with a rate of $10^{-15} / yr$, the QCD scale $\Lambda $ and the nucleon mass are decreasing with a rate of about $3 \cdot 10^{-14} / yr$. The magnetic moments of the nucleons and of nuclei would increase: $$\frac{\dot{\mu}}{\mu} \approx 3 \cdot 10^{-14} / yr \, .$$ 3. The third possibility is that both $\alpha_u$ and $\Lambda_{un}$ are time–dependent. In this case we find: $$\frac{\dot{\Lambda}}{\Lambda} \cong 46 \cdot \frac{\dot{\alpha}}{\alpha} + 1,07 \cdot \frac{\dot{\Lambda}_{un}}{\Lambda_{un}} \, .$$ On the right two terms appear: $\left( \dot{\alpha} / \alpha \right)$ and $\left( \dot{\Lambda}_{un} / \Lambda_{un} \right)$. These two terms might conspire in such a way that $\left( \dot{\Lambda} / \Lambda \right)$ is smaller than about $\left( 40 \cdot \dot{\alpha} / \alpha \right)$. In the first two scenarios the QCD–scale $\Lambda $ changes faster than the finestructure constant. In the third scenario the change of the QCD–scale depends both on the change of $\alpha $ and on the change of the unification scale. The signs of both changes can be different, and one could have a partial cancellation. In this case the change of $\Lambda $ could be less than in the first two scenarios. This possibility will be discussed later. The question arises, whether a time change of the QCD scale parameter could be observed in the experiments. The mass of the proton and the masses of the atomic nuclei as well as their magnetic moments depend linearly on the QCD scale. If this scale changes, the mass ratio $M_{p} / m_e = \mu$ would change as well, if the electron mass remains constant. We mentioned before that the mass ratio $\mu $ seems to show a time variation – in a linear approximation one has about $$\frac{\Delta \mu}{\mu} \approx 10^{-15} / year \, .$$ If we take the electron mass to be constant in time, this would imply that the QCD–scale $\Lambda $ changes with the rate $$\frac{\Delta \Lambda}{\Lambda} \approx 10^{-15} / year \, .$$ The connection between a time variation of the finestructure constant and of the QCD scale, discussed above, is only valid, if either the unified coupling constant or the unification scale depends on time, not both. If both the unification scale and the unified coupling constant are time dependent, we should use instead eqs. (51) and (60). As seen in particular in eq. (60), there might be a cancellation between the two terms, and in this case the time variation of the QCD–scale would be smaller than $10^{-14} / year$. If the two terms cancel exactly, the QCD–scale would be constant, but this seems unlikely. Thus a time variation of the QCD–scale of the order of $10^{-15} / year$ (see eq. (62)) is quite possible. Can such a small time variation be observed in experiments here on the earth? In Quantum Optics one can carry out very precise experiments with lasers. In the next chapter we shall describe such an experiment at the Max–Planck–Institute of Quantum Optics in Munich, which was designed especially to find a time variation of the QCD scale $\Lambda $ as expected by the Grand Unified Theory. Results from QuantumOptics ========================== Being the simplest of all stable atoms, the hydrogen atom is a very good test object for checking fundamental theories. Its atomic properties can be calculated with unprecedented accuracy. But at the same time the level structure of the hydrogen atom can be very accurately probed, using spectroscopy methods in the visible, infrared and ultraviolett regions. Thus the hydrogen atom plays an important rôle in determining the fundamental constants like the finestructure constant. Measurements of the Lamb shift and the 2S hyperfine structure permit very sensitive tests of quantum electrodynamics. Combining optical frequency measurements in hydrogen with results from other atoms, stringent upper limits for a time variation of the finestructure constant[@Fisch] and of the QCD scale parameter can be derived. The employment of frequency combs turned high–precision frequency measurements into a routine procedure. The unprecedented accuracy of the frequency comb have opened up wide perspectives for optical atomic clock applications in fundamental physics. Frequency measurements in the laboratory have become competitive recently in terms of sensitivity to a possible time variation of the fine–structure constant in the present epoch. Though the time interval covered by these measurements is restricted to a few years, very high accuracy compensates for this disadvantage. Their sensitivity becomes comparable with astrophysical and geological methods operating on a billion–year time scale. Important advantages of the laboratory experiments are: The variety of different systems that may be tested, the possibility to change parameters of the experiments in order to control systematic effects, and the determination of the drift rates from the measured data. Modern precision frequency measurements deliver information about the stability of the present values of the fundamental constants, which can only be tested with laboratory measurements. At the same time only non-laboratory methods are sensitive to processes that happened in the early universe, which can be much more severe as compared to the present time. As both classes of experiments probe the constants at different epochs, they supplement each other to get a more detailed view of the possible time variation of the fundamental constants. In the experiment of the MPQ–group in Munich[@Osk] one was able to determine the frequency of the hydrogen 1S–2S–transition to 2466061102474851(34) Hz. A comparison with the experiment performed in 1999 gives an upper limit on a time variation of the transition frequency in the time between the two measurements, 44 months apart. One finds for the difference $(-29 \pm 57) Hz$, i. e. it is consistent with zero. One can compare this frequency with the frequency of the hyperfine transition in $^{133}$ Cs, and one obtains for the fractional time variation: $\left( - 3.2 \pm 6.3 \right) \cdot 10^{-15} yr^{-1}$. Comparing the $1S - 2S$ hydrogen transition with the hyperfine transition of Cesium $^{133} Cs$, one can obtain information about the time variation of the ratio $\alpha / \alpha_s$. This is the case, since the Cesium hyperfine transition depends on the magnetic moment of the Cesium nucleus, and the magnetic moment is proportional to ($ 1 / \Lambda_c$, ($\Lambda_c$: QCD scale parameter). If $\Lambda_c$ varies in time, the magnetic moment will also vary. One has also found interesting limits[@Osk] on the time variation of $\alpha $: $$- 1.5 \cdot 10^{-15} / yr < \frac{\delta \alpha}{\alpha} < 0.4 \cdot 10^{-15} / yr \, .$$ Recently one has obtained a limit for the time variation of the magnetic moment of the Cesium nucleus[@Kola]: $$\frac{\delta \mu}{\mu} = \left( 1.5 \pm 2.0 \right) \cdot 10^{-15} / yr \, .$$ These results are consistent with zero. However, the limit on the time variation of $\alpha $ is of the same order as the astrophysics result. The result concerning the magnetic moment implies a limit on the time variation of $\Lambda_c$: $$\frac{\Delta \Lambda_c}{\Lambda_c} = \left( - 1.5 \pm 2.0 \right) \cdot 10^{-15} / yr \, .$$ This result is in disagreement with our results, based on the assumption, that either $\alpha_u$ or $\Lambda_{un}$ change in time. We obtained about $10^{-14} / yr$, which is excluded by the experiments. The result given above is consistent with no time change for $\Lambda_c$, but it also agrees with a small time change of the order of $10^{-15}$ per year. If we assume that the electron mass does not change in time, such a change of $\Lambda_c$ would agree with the astrophysics result on the time variation of the ratio M(proton) / m(electron). Theoretically we would expect such a time variation if both $\Lambda_{un}$ and $\alpha_u$ change in time, as discussed at the end of chapter 4. Conclusions and Outlook ======================= In this review we have summarized our present knowledge about the fundamental constants and their possible time variation. Today we do not know how these constants are generated or whether they might depend on time. The phenomenon of the fundamental constants in physics remains a mystery. There might be relations between these constants, e. g. between the flavor mixing angles and the fermion masses, or relations between the three coupling constants, implied by the idea of Grand Unification. This would reduce the number of basic constants from 28 down to a smaller number, but at least 18 fundamental constants would still exist. A possible time variation of the fundamental constants must be rather slow, at least for those fundamental constants, which are measured very precisely, i. e. the finestructure constant, the QCD–scale $\Lambda $, and the electron mass. The constant of gravity $G$ is known with a precision of $10^{-11}$. All other fundamental constants, e. g. the masses of the other leptons or the masses of the quarks, are not known with a high precision. We would not notice, if for example the mass of the $\tau$–lepton would change with a rate of $10^{-5}$ per year, or if the mass of the $b$–quark would change with a rate of $10^{-4}$ per years. The present limits on the time variation of the finestructure constant, the QCD scale or the electron mass are of the order of $10^{-15}$ / year. These limits should be improved by at least two orders of magnitude in the near future. The astrophysics experiment [@Rein] indicates a time variation of $\Lambda_c$ of the order of $10^{-15} / year$. Is this experiment correct? We do not know. Other astrophysics experiments should be carried out. These experiments have a time scale of about $10^{10}$ years. They cannot determine a time variation at the present time, 14 billion years after the Big Bang. The experiment in quantum optics, however, have a time scale of only a few years, but a very high precision. If the astrophysics experiments indicate a time variation of the order of $10^{-15}/ year$, it does not mean that experiments in quantum optics should also give such a time variation. It might be that until about 10 billion years after the Big Bang the constants did vary slowly, but after that they remain constant. No theory exists thus far for a time variation, and there is no reason to believe that a time variation should be linear, i. e. $10^{-15} / year$ throughout the history of our universe. If the fundamental constants do vary, one would expect that the variation is rather large very close to the Big Bang. In the first microseconds after the Big Bang constants like $\alpha $ or $\Lambda_c$ might have changed by a factor 2, and we would not know. There are limits from nucleosynthesis on possible variations of $\alpha $ or $\Lambda_c$, but they are rather weak and test these constants about one minute after the Big Bang, not earlier. In cosmology one should consider time variations of fundamental parameters in more detail. They arise naturally in the superstring theories. Perhaps allowing a suitable time variation of the constants leads to a better understanding of the cosmic evolution immediately after the Big Bang. Some, but not all fundamental constants might simply be cosmic accidents. But in this case the constants had a rapid time variation right after the Big Bang. If this is true, we shall never be able to determine theoretically the values of those constants. But allowing time variations might lead to better cosmological theories and to a better understanding of particle physics. Particle physics and cosmology together would give a unified view on our universe. [99]{} -2pt A. Fritzsch and M. Gell–Mann, .\ H. Fritzsch, M. Gell–Mann and H. Leutwyler, . S. L. Glashow, . S. L. Glashow and J. C. Ward, .\ S. Weinberg, .\ A. Salam, , ed. N. Svartholm, Stockholm. P. Higgs, . R. Brout and F. Englert, . G.S. Guralnik, C. R. Hagen, and T. W. B. Kibble, . E. Adams and G. Lewis, . A. Wyler, . R. P. Feynman, . M. Gell–Mann and F. Low, . H. Fritzsch, . H. Fritzsch and Z. Xing, .\ H. Fritzsch, Proc. of JCFP Peking, (2007), to appear. J. K. Webb et al., . H. Chasid, R. Srianand, P. Petitjeau and B. Aracil, .\ R. Quast, D. Reimers and S.A. Levshakow, . P. Dirac, . L. D. Landau, . T. Damour and F. Dyson, . E. Reinhold, R. Buning, U. Hollenstein, A. Ivanchik, P. Petitjean and W. Ubachs, . H. Georgi and S. Glashow, . H. Fritzsch and P. Minkowski, . T. Yanagida, .\ M. Gell–Mann, P. Ramond and R. Slansky, . X. Calmet and H. Fritzsch, . P. Langacker, G. Segre and M. J. Strassler, . M. Fischer et al., . W. H. Oskay et al., . N. Kolachevsky, J. Alnis, A. Matveev, Th. Udem, R. Holzwarth and T. W. Hänsch: Precision Measurements in Atomic Hydrogen, in: Optical Atomic Clock, Y. Ovchinnikov. ed., to appear.

ArXiv

--- abstract: 'The problem of coordination without *a priori* information about the environment is important in robotics. Applications vary from formation control to search and rescue. This paper considers the problem of search by a group of solitary robots: self-interested robots without *a priori* knowledge about each other, and with restricted communication capacity. When the capacity of robots to communicate is limited, they may obliviously search in overlapping regions (i.e. be subject to interference). Interference hinders robot progress, and strategies have been proposed in the literature to mitigate interference [@IEEEexample:wellman2011using; @IEEEexample:hourani2013serendipity]. Interaction of solitary robots has attracted much interest in robotics, but the problem of mitigating interference when time for search is limited remains an important area of research. We propose a coordination strategy based on the method of cellular decomposition [@IEEEexample:choset2001coverage] where we employ the concept of soft obstacles: a robot considers cells assigned to other robots as obstacles. The performance of the proposed strategy is demonstrated by means of simulation experiments. Simulations indicate the utility of the strategy in situations where a known upper bound on the search time precludes search of the entire environment.' author: - - - bibliography: - 'IEEEabrv.bib' - 'jordan\_soft\_obstacles.bib' title: 'A Coordinated Search Strategy for Multiple Solitary Robots: An Extension' --- Introduction {#sec:introduction} ============ In recent years, researchers and engineers have been increasingly interested in coordinating systems of intelligent robots, with applications ranging from formation control to search and rescue. The problem of coordination is important in robotics, for example, where a group of robots need to achieve some task [@IEEEexample:arkin2002line; @IEEEexample:choset2001coverage; @IEEEexample:feinerman2012collaborative; @IEEEexample:hao2008cooperative; @IEEEexample:kong2006distributed; @IEEEexample:rekleitis2004limited; @IEEEexample:roy2001collaborative; @IEEEexample:solanas2004coordinated; @IEEEexample:wurm2008coordinated]. This work considers the case of search by multiple robots. Search of an environment consists of finding targets by exploring the environment. In this work, coverage is a proxy for finding targets: the exploration performance of a robot is taken to be proportional to its area covered. This paper considers situations where solitary robots: self-interested robots without *a priori* knowledge about each other and limited communication, search an unknown environment. Real mobile robots have limited communication capacity [@IEEEexample:mosteo2008multi]. A recent survey of solutions to problems of limited communication can be found in [@IEEEexample:amigoni2017multirobot]. We assume that when a robot starts searching, the only information it has about the environment is the region that it can perceive from its initial location. The scale and boundary locations of the environment are also unknown, although a hard boundary can be sensed when within perception range. It is also assumed that robots have a limited time for search: the environment is wide and robots cannot explore it entirely, even were the environment known [@IEEEexample:wellman2011using]. The constraint on time could be, for example, due to limited battery power of robots. One of the main concerns induced when robots with limited communication explore an unknown environment is interference: robots may obliviously search in overlapping regions which hinders robot progress. Interference can also be a concern for robots which some *priori* knowledge [@IEEEexample:wellman2011using; @IEEEexample:hourani2013serendipity]. In the case of solitary robots, the only moment that robots detect interference is when they meet. Under limited communication, the meeting of solitary robots is just a fortunate coincidence (as in accidental or symmetric rendezvous [@IEEEexample:alpern1995rendezvous]). In this work, the accidental rendezvous strategy will refer to a strategy where robots meet only accidentally. When search involves multiple robots, one would like a group of robots that know about each other to explore non-overlapping regions – this is termed sustainability in [@IEEEexample:hourani2013serendipity]. It is assumed that robots are effective in making use of information they possess. Consequently, each group member typically should cover almost the same area in size. (We ignore gradient in the exploration of regions.) This paper proposes a coordination strategy based on the method of cellular decomposition [@IEEEexample:choset2001coverage] and the use of soft obstacles. The proposed strategy provides sustainable performance when the time for search is limited (insufficient to cover the entire search area) and known. A shorter version of this work can be found in [@IEEEexample:masakuna2019coordinated]. Here we extend the theoretical and practical investigation of the proposed coordination strategy. The rest of the paper is organised as follows: Section \[sec:relatedwork\] discusses related work, focusing particularly on the accidental rendezvous and periodic rendezvous strategies. Section \[sec:softobstacle\] motivates and presents the proposed strategy. Section \[sec:experiments\] compares the approaches and discusses the results. In Section \[sec:conclusion\], we conclude and propose further work. Related Work {#sec:relatedwork} ============ The accidental rendezvous strategy applies a frontier-based approach for exploration of unknown environments [@IEEEexample:yamauchi1997frontier; @IEEEexample:yamauchi1998frontier]. Frontiers are regions at the boundary between explored and unexplored regions. Frontier-based strategies are greedy algorithm and differ primarily in how frontiers are selected. Greedy algorithms are optimal for coverage problems, as they are a special form of submodular optimization [@IEEEexample:nemhauser1978analysis]. The main work using rendezvous in exploring unknown environments can be found in [@IEEEexample:wellman2011using; @IEEEexample:hourani2013serendipity; @IEEEexample:de2010selection]. An interesting approach which provides a solution to mitigate interference with sustainable exploration performance was proposed in [@IEEEexample:wellman2011using]: at each rendezvous, the search space is divided into sectors (i.e. unbounded regions) to which robots are assigned. For example, the most westerly part of the environment might be assigned to a robot. The soft obstacle strategy proposed here also applies this idea of assigning robots to sectors. However, dividing areas into sectors can result in uneven assignment, which promotes interference. When uneven assignment, robots have different-sized areas to explore. For example, an uneven assignment arises when robots interact close to a boundary of the environment which they ignore while coordinating. In Figure \[fig:exampleARS2\] for instance, the areas assigned to two of the robots (the green and red trajectories indicate these robots) are partly outside the environment, something the robots are unaware of when coordinating, as boundaries of the environment are unknown until encountered. Thus, assigning sectors under the accidental rendezvous strategy may lead to interference, and hence to non-sustainable performance. [0.24]{} ![*Illustration of assignment of sectors with four robots (coloured circles are robots, the lines show their respective trajectories). **(\[fig:exampleARS1\])**: even assignment, robots start at the centre of the environment. The chance of overlap is small. **(\[fig:exampleARS2\])**: uneven assignment, robots ignore that they start close to a boundary, so the chance of overlap is high (see crossed lines).*[]{data-label="fig:exampleARS"}](figs/ARS_even.png "fig:"){width="\textwidth"} [0.24]{} ![*Illustration of assignment of sectors with four robots (coloured circles are robots, the lines show their respective trajectories). **(\[fig:exampleARS1\])**: even assignment, robots start at the centre of the environment. The chance of overlap is small. **(\[fig:exampleARS2\])**: uneven assignment, robots ignore that they start close to a boundary, so the chance of overlap is high (see crossed lines).*[]{data-label="fig:exampleARS"}](figs/ARS_uneven.png "fig:"){width="\textwidth"} To address uneven assignment, robots could schedule further meetings to share additional information. However, scheduled meetings induce interruptibility: robots cease new knowledge acquisition when the rendezvous time arrives. They stop knowledge acquisition to travel to the rendezvous point. It has been shown that interruptibility can consume up to half of the search time [@IEEEexample:hourani2013serendipity]. Methods exist which mitigate interruptibility. For example, the exploration method proposed in [@IEEEexample:de2010selection] indirectly mitigated interruptibility. The method consists of two types of robots (explorer and relay) and a base station. The explorers intentionally meet and share knowledge with the relays. The latter transfer the shared information to the base station. Interruptibility is mitigated by the fact that the explorer and relay meet closer to the frontiers of the explorer. In the approach taken in this work, all robots are explorers and no base station is present. Another solution in the literature which mitigates interruptibility has robots plan to meet other robots before the rendezvous time arrives [@IEEEexample:hourani2013serendipity]. Two challenging aspects need to be handled in this case: forecasting of the current positions of the robots involved; and forecasting of the paths taken by the robots from their current positions. To address these challenges, the method proposed by Hourani *et al.* is based on serendipity [@IEEEexample:hourani2013serendipity]. Usually, when robots interrupt exploration to attend a rendezvous, new information might not be gained. To address this negative impact, some robots, which are referred to as serendip robots, plan to interact with other members before the rendezvous time arrives by forecasting their paths. Thus interruptibility is mitigated. For the contribution here, this work applies the cellular decomposition strategy to search of an unknown environment by solitary robots. The approach aims to produce an interruptibility-free strategy with sustainable exploration performance. Cellular decomposition allows robots to explore non-overlapping regions, thus interference is mitigated. Cellular decomposition also allows robots to avoid the need to schedule further rendezvous, thus reducing interruptibility. The region assigned to a robot is termed an exploration region. Here, a robot considers exploration regions assigned to others as soft obstacles. Thus, it eschews those regions, although a robot can move through soft obstacles if it would otherwise be trapped because surrounded by explored regions. Soft-obstacle Strategy {#sec:softobstacle} ====================== The proposed coordination strategy is an heuristic technique, which means it seeks for a satisfactory solution. Solitary robots, being self-interested, do not have a common goal. Solitary robots work under bounded rationality. Due to limited knowledge of a solitary robot in terms of the tractability of the problem and information about others, a satisfactory solution may be needed [@IEEEexample:simon1956rational]. This work assumes the following: - the environment is a static, wide and initially unknown $2$D gridworld, although boundaries can be detected when encountered; - targets are uniformly and independently distributed. - robots are solitary, autonomous, homogeneous and uniquely identifiable. A noise-free system is considered, but uncertainty can be incorporated; - there could be many robots searching, but coordination involves only a small number of robots at a time (coordination of up to ten robots will be considered here). The key concern with sectors (i.e. unbounded regions) which this work aims to handle is that of the behaviour of robots in cases of uneven or unbalanced assignment. We propose a way to allow a robot to explore outside of its sector, taking into account other robot sectors, in order to reduce interference. It is to be noted that the main difference between the soft obstacle strategy and the strategy of sectors applied in [@IEEEexample:wellman2011using] is that the soft obstacle strategy assigns bounded regions to robots. Sectors are not bounded. In summary, the problem considered in this paper is the provision of a coordination strategy for multiple solitary robots. In the case studied here, robots can coordinate only when they are in the same vicinity. The strategy targets the situation where search time is known and bounded, for example, by an attribute of the robot such as available battery time. Intuitive considerations in the soft obstacle strategy ------------------------------------------------------ To contribute effectively to a group search, each robot should avoid searching areas others have already covered. To achieve this, we make use of the following two insights: 1. since the upper bound on search time is known in advance, a robot can use this information to evaluate the maximum area it can cover; 2. if a robot knows the regions assigned to other robots, it can plan to avoid exploring those regions. A robot is aware of other robots’ regions only through rendezvous. These insights are used in our approach when constructing the exploration region points for the cellular decomposition strategy [@IEEEexample:choset2001coverage]. However, just as with the approach of sectors applied in [@IEEEexample:wellman2011using], there remain other major concerns when applying cellular decomposition in an unknown environment. - First, the region assigned to a robot could be partly outside the environment, which the robot is not aware of at the start ($X_3$ in Figure \[fig:illustration\]). This is because robots might not have information about the boundaries of the environment while coordinating. - Second, the exploration region of a robot could be occupied by obstacles (black boxes and circle in Figure \[fig:illustration\]). ![*Illustration of a coordination based on the method of cellular decomposition with four robots. Black boxes and circle denote obstacles. The most thickened lines denote boundaries of the search space and the squares with solid lines are exploration regions.*[]{data-label="fig:illustration"}](figs/illustration1.png) Areas assigned to a robot might include parts which are inaccessible because they are occupied by obstacles or because they lie outside the search region. This can lead to robots having uneven assignments. When an uneven assignment is not addressed, robots may explore overlapping regions. Interference is more likely to occur in cases of uneven assignment. Let $\mathcal{W}$ and $\mathcal{O}$ denote the search environment and the set of obstacles in $\mathcal{W}$ respectively. With $\mathcal{W}^c$ the complement of $\mathcal{W}$, the inaccessible region is given by $$\mathcal{O}\cup \mathcal{W}^c\,.$$ A robot first creates a virtual world, meaning that some parts of its region might not be in the real environment. When its exploration region is partly accessible, the robot explores the accessible part first (details on exploring accessible regions will be provided later). Subsequently, the robot looks for another exploration region, taking into account the exploration regions assigned to others (i.e. soft obstacles) that it is aware of. For a robot, the soft obstacles are prohibited regions. An important consequence of the above approach is that no matter what might prevent a robot from exploring its assigned exploration region (such as uneven assignment), other robots are prohibited from searching in that exploration region. This is an important feature of the strategy. In Figure \[fig:illustration\], for example, most of the exploration region of robot $R_3$ is *outside* the environment (i.e. it is inaccessible). The robot $R_3$, after exploring the accessible area of its exploration region, needs to get another exploration region. Given that $R_3$ knows the exploration regions of the other three robots, to get a fresh exploration region $R_3$ should follow a direction leading to the complement of the soft obstacles. In Figure \[fig:illustration\] for example, $R_3$ can follow the direction indicated by the arrow on top of its exploration region. In coordination, the robots concerned choose a leader. To choose a leader, we consider the approach used in [@IEEEexample:kim2013leader]. In [@IEEEexample:kim2013leader], a leader is elected based on closeness centralitythe closeness centrality of a robot is the reciprocal average of the depths between the robot and all other robots. Robots involved in interaction build a spanning tree first by message passing, then they choose a leader which has the highest closeness centrality. We consider situations where choosing a leader at the central position is important, for efficient transmission. The elected leader assigns exploration regions to the participating robots using the Hungarian algorithm [@IEEEexample:kuhn1955hungarian] to minimize robot-to-region travelling time. Robots use straight lines to reach their respective exploration regions while avoiding obstacles or soft obstacles. In addition to the problem of assignment of regions to robots, an obstacle avoidance system must be incorporated in order to enable robots to avoid obstacles encountered. Techniques used to avoid obstacles are introduced next. Obstacle avoidance techniques {#sec:avoidance} ----------------------------- We consider two situations for obstacle avoidance: when a robot is moving to an exploration region and when a robot is exploring its region. For the former case, various methods can be applied which include the potential field, the vector field histogram methods and the local navigation method [@IEEEexample:khatib1986real; @IEEEexample:ulrich1998vfh; @IEEEexample:fujimori1997adaptive]. For simplicity, we use a Bug algorithm [@IEEEexample:oroko2014obstacle]. Bug algorithms are simple methods used for path planning to avoid an unexpected obstacle in the robot motion by updating the directional angle of a robot when an obstacle is detected. Bug algorithms assume only local knowledge of the search environment (i.e. obstacles are unknown) and a goal, and the robot knows directions (and can evaluate the Euclidean distance) towards the goal. The robot has a limited perception range. When a robot detects an obstacle, it follows the trajectory of the obstacle boundary. Bug algorithms are based on the two following principles. To avoid a detected obstacle the robot, - follow the closest edge of the obstacle, towards right, left, up or down. - move in a straight line toward a goal. Here, goal can be either the exploration region of the robot (when the robot is travelling to its exploration region) or a boundary of its exploration region (when the robot is exploring its region using zigzag). There exist various variants of Bug algorithms, including Bug 1, Bug 2, and Distance Bug. Some Bug algorithms, as listed above, can be seen as improvement to others to some extent. For instance Distance Bug is an improvement of Bug 2 and the latter is an improvement to Bug 1. The three variants are described as follows. In Bug 1, when a robot encounters an obstacle while travelling towards the goal, it follows a canonical direction until the location of the initial encounter is reached. After reaching the initial encounter, the robot follows the boundary of that obstacle to reach the point along the boundary that is closest to the goal. Once reached that closest point, the robot moves directly toward the goal. It repeats the same behaviour when new obstacles are encountered (Figure \[fig:bug1\]). The revolving behaviour of the robot around each and every obstacle is computationally costly. Let $x_i$ and $a_i$ be the initial point of the robot and the goal respectively. Let $C(x_i, a_i)$ denote the performance of a robot to get to $a_i$ from $x_i$. The worst case performance is upper bounded by $$C(x_i, a_i)\leq \delta(x_i, a_i) + \frac{3}{2} \sum_{j=1}^{k} p_j\,,$$ where $\delta(x_i, a_i)$ is the Euclidean distance $x_i$ to $a_i$, $p_j$ is the perimeter of the $ij$th obstacle, and $k$ is the number of obstacles encountered by the robot. Bug 2 addresses this issue in Bug 1. In Bug 2, when a robot encounters an obstacle, it starts moving along the edge of the obstacle until it finds a point along the boundary of the obstacle with the same slope (Figure \[fig:bug2\]). Then the robot moves to the goal. The worst case performance is upper bounded by $$C(x_i, a_i)\leq\delta(x_i, a_i) + \frac{1}{2} \sum_{j=1}^{k} n_jp_j\,,$$ where $n_j$ is the number of times the $j$th obstacle crosses the line segment between $x_i$ and $a_i$. In Bug 2, looking for a point along the boundary generating a line to the goal with the same slope as that of the line from $x_i$ and $a_i$ can slow the procedure of obstacle avoidance. Distance Bug, which we use in this work, improves Bug 2 in that regards. In Distance Bug, When a robot encounters an obstacle , it chooses points which minimise the distance to the goal while avoiding obstacle (Figure \[fig:bugdistance\]). The worst case performance is upper bounded by $$C(x_i, a_i)\leq\delta(x_i, a_i) + \frac{1}{3} \sum_{j=1}^{k} n_jp_j\,,$$ where $n_j$ is the number of times the $j$th obstacle crosses the line segment between $x_i$ and $a_i$. [0.24]{} ![*Illustration of obstacle avoidance using Bug algorithms where red filled circle denotes a robot $R_i$, black rectangle an obstacle and white square with black boundary the exploration region of $R_i$, $X_i$. **(\[fig:bug1\])**: Bug 1. **(\[fig:bug2\])**: Bug 2. **(\[fig:bugdistance\])**: Distance Bug.*[]{data-label="fig:bugs"}](figs/bug1_1.pdf "fig:"){width="\textwidth"} [0.23]{} ![*Illustration of obstacle avoidance using Bug algorithms where red filled circle denotes a robot $R_i$, black rectangle an obstacle and white square with black boundary the exploration region of $R_i$, $X_i$. **(\[fig:bug1\])**: Bug 1. **(\[fig:bug2\])**: Bug 2. **(\[fig:bugdistance\])**: Distance Bug.*[]{data-label="fig:bugs"}](figs/bug2_.pdf "fig:"){width="\textwidth"} [0.2]{} ![*Illustration of obstacle avoidance using Bug algorithms where red filled circle denotes a robot $R_i$, black rectangle an obstacle and white square with black boundary the exploration region of $R_i$, $X_i$. **(\[fig:bug1\])**: Bug 1. **(\[fig:bug2\])**: Bug 2. **(\[fig:bugdistance\])**: Distance Bug.*[]{data-label="fig:bugs"}](figs/bug2_2.pdf "fig:"){width="\textwidth"} Regarding obstacle avoidance during a robot’s exploration of its assigned region, we propose a simple strategy to avoid obstacles. A robot applies a motion pattern (the motion pattern will be described in Subsection \[sec:zigzag\]) to search its assigned region. If it encounters an obstacle, it attempts to maintain the pattern to the extent possible while avoiding the obstacle. This means that when it completes the current motion pattern some sub-regions might still be relatively unexplored. When it finishes applying its current motion pattern, it therefore must determine whether its current exploration region still contains significant unexplored areas, and to focus on these, or to seek a fresh search region based on the remaining search time. To determine the unexplored area, a robot generates seeds over its exploration region, sampled from a uniform distribution. The seeds indicate the size and location of the unexplored area. Figure \[fig:avoidance\] and Figure \[fig:avoidance2\] illustrate obstacle avoidance. Soft-obstacle strategy: cellular decomposition ---------------------------------------------- Let $X_i$, $\mathcal{C}^{(t)}_i$ and $\mathcal{Z}^{(t)}_i$ denote the exploration region, the explored region and the interference region of robot $R_i$ at time $t$ respectively. Two components make up soft obstacles: the region $\mathcal{C}^{(t)}_i$ already explored and known to the robot $R_i$ and the interference regions $\mathcal{Z}^{(t)}_i$, i.e., exploration regions of other robots. The soft obstacles $\mathcal{B}^{(t)}_i$ of robot $R_i$ are $$\label{eq:constructsoftobstacle} \mathcal{B}^{(t)}_i = \mathcal{C}^{(t)}_i \cup \mathcal{Z}^{(t)}_i\,.$$ $R_i$’s exploration region $X_i$ is a grid and is composed of at least one cell. Initially, $X_i$ has a single cell. It is decomposed into other cells when $R_i$ avoids encountered obstacles in $X_i$ as explained above in Subsection \[sec:avoidance\]. Assume that the coordination procedure lasts for $\Delta T$. At the end of coordination, the interference region of a robot $R_i$ is updated to include the other robots’ exploration regions. $$\label{eq:interferenceregion} \mathcal{Z}^{(t+\Delta T)}_i=\mathcal{Z}^{(t)}_i\cup \bigg(\bigcup_{j\neq i}X_{j}\bigg)\,,$$ An exploration region of a robot is characterised by a width, a height, and its left lower corner. The sensing area by a robot is a circle. [^1]The maximum area that a robot can scan in $\tau$ time units is $$\label{eq:area} A(\tau)= \pi \, r^2+2\gamma r\tau\,,$$ where $\gamma$ ($\gamma>0$) denotes the motion scale and $r$ denotes the perception range of robots. A motion scale is a positive real number that denotes the number of points scanned on a line by a robot per unit of time and it is fixed. Robots can have different motion scales. In this work homogeneous solitary robots have the same motion scale, $\gamma$. In interaction, robots fuse their individual information and assign to each other fresh exploration regions. Algorithms used to fuse data and to mission robots are described next. Data fusion and mission plans ----------------------------- A local leader has three roles: it receives individual data from others, it coordinates the plans for their next actions and replicates decisions (missions) to others. Since search is the use case of this paper, robots’ data also contains maps. Robots need to combine their individual maps. The following step is to send individual data to the leader. Each robot has its own internal coordinate system. When two robots intend to merge their maps, unification of internal coordinate systems must be done. To fuse data, we consider a general framework proposed in [@IEEEexample:konolige2003map]. The framework uses robot-to-robot measurements to achieve map fusion. The process of data fusion starts from robots with lower degrees. The leader must be the last to receive. A robot $R_i$ sends data to robot $R_j$ following format $(\mathcal{H}^{(t)}_i, \mathcal{C}^{(t)}_i)$ where $\mathcal{C}^{(t)}_i$ denotes the map from $R_i$ and $\mathcal{H}^{(t)}_i$ the interaction history of $R_i$ at time $t$ from their respective coordinate systems. The algorithm of data fusion applied by a single robot $R_i$ is illustrated in Algorithm \[algo:sending\]. In Algorithm \[algo:sending\], $\mathcal{N}_i$ denote the immediate neighbours of the robot $R_i$. $\mathcal{C}^{(t)}_i\gets \mathcal{C}^{(t)}_i \cup \mathcal{C}^{(t)}_j $ $\mathcal{H}^{(t)}_i \gets \mathcal{H}^{(t)}_i \cup \mathcal{H}^{(t)}_j $ When the leader has received all the information, it applies the coordination strategy. After combining individual data and making plans, the leader replicates the results to others. The algorithm that a robot $R_i$ applies for mission plans is given in Algorithm \[algo:reply\]. The involved robots adopt the coordinate system of the leader. In Algorithm \[algo:reply\], $(XY)_i$ denotes the internal coordinate system of the robot $R_i$. $\mathcal{H}^{(t)}_j \gets \mathcal{H}^{(t)}_i $ $(XY)_j \gets (XY)_i $ gets the exploration region $X_j$ of $R_j$ The interaction history of $R_i$ is given as follows, $$\label{eq:history} \mathcal{H}^{(t)}_i = \{(R_j, l_j, h_{j}, a_j):\, \forall R_j\in \mathcal{N}_i\}\,,$$ where $R_j$ denotes robots already interacted with $R_i$ (or robots which $R_i$ has histories of), $l_{j}$, $h_{j}$ and $a_j$ denote length, height and centroid of the exploration region of the robot $R_j$. In applying cellular decomposition in the proposed coordination strategy, the elected leader intentionally leaves unexplored spaces (or margins) between robots’ exploration regions (the dotted blue lines around an exploration region in Figure \[fig:illustration\]). Margins are to be used by robots whose exploration regions are partly inaccessible as each robot plans independently. Use of margins is part of the novelty of the proposed approach. Measure of margins {#subsec:decomposition} ------------------ When the region $X_i$ of $R_i$ is partly accessible, the margin $M_i$ will allow the robot $R_i$ to avoid interference when it is travelling to a new exploration region. It should be noted that robots also explore margins. Once in a margin, the robot considers the margin as an exploration region and explores it entirely before moving to its fresh exploration region. In other words, a margin is a path from the current location of the robot to its new exploration region. The area of the margin $M_i$ of a robot $R_i$ is proportional to a time $\tau_0$. The value of $\tau_0$, given in Equation \[eq:tau0\], is the amount of time a robot will take to explore its margin. The width of a margin is $m$ ($m\geq 2r$). $$\label{eq:tau0} \tau_0=\frac{2}{r\gamma} \bigg(m^2 + m\sqrt{\pi r^2+2\gamma r (\tau-t-\Delta T)}\bigg)\,.$$ The augmented exploration region of $R_i$, which is $M_i\cup X_i$, has the following area $$\label{eq:newarea} \pi \, r^2+2\gamma r(\tau+\tau_0-t-\Delta T)\,,$$ where $t$ is the amount of time already spent. The approach used by a robot to explore its region is discussed next. Zigzag search {#sec:zigzag} ------------- To cover its exploration region, a robot applies a zigzag search[^2] [@IEEEexample:choset2001coverage]. For the benchmark coordination algorithms, a robot will use a frontier-based search algorithm to cover its exploration region. In zigzag search, the robot moves back and forth from one side of its region to the other, along parallel paths. For instance, a robot may move from west to east back and forth towards the north (Figure \[fig:descrzigzag1\]). [0.12]{} ![*Application of zigzag search. **(\[fig:descrzigzag1\])**: an obstacle-free exploration region. **(\[fig:avoidance2\])**: a region occupied by a regular polygonal obstacle. **(\[fig:avoidance\])**: a region occupied by a non-regular polygonal obstacle.*[]{data-label="fig:zigzag"}](figs/zigzag_case1.png "fig:"){width="\textwidth"} [0.12]{} ![*Application of zigzag search. **(\[fig:descrzigzag1\])**: an obstacle-free exploration region. **(\[fig:avoidance2\])**: a region occupied by a regular polygonal obstacle. **(\[fig:avoidance\])**: a region occupied by a non-regular polygonal obstacle.*[]{data-label="fig:zigzag"}](figs/avoidance2.png "fig:"){width="\textwidth"} [0.12]{} ![*Application of zigzag search. **(\[fig:descrzigzag1\])**: an obstacle-free exploration region. **(\[fig:avoidance2\])**: a region occupied by a regular polygonal obstacle. **(\[fig:avoidance\])**: a region occupied by a non-regular polygonal obstacle.*[]{data-label="fig:zigzag"}](figs/avoidance12.png "fig:"){width="\textwidth"} The soft obstacle strategy used by a robot $R_i$ is given in Algorithm \[algo:softobstacle\]. $t\gets 0$ the location $x_i$ of $R_i$ is set to its initial position $\mathcal{C}^{(t)}_i \gets B_r(x_i), \mathcal{Z}^{(t)}_i \gets \emptyset$ initialise the exploration region $X_i$ from Equation \[eq:area\] $R_i$ applies the zigzag search choose a leader $R_k$ among the interacting robots $R_k$ combines the explored region $\mathcal{C}^{(t)}_j$ and interaction histories $\mathcal{H}^{(t)}_j$ of other robots $R_j$ using Algorithm \[algo:sending\] $R_i$ assigns mission plans to other robots using Algorithm \[algo:reply\] $R_i$ waits for its exploration region from the leader $R_i$ updates its soft obstacle $\mathcal{B}^{(t+\Delta T)}_i$ $t\gets t+\Delta T$ $R_i$ applies the zigzag search $t\gets t+\Delta t$ Regarding research issues, the soft obstacle strategy inherits completeness from the zigzag search and cellular decomposition [@IEEEexample:choset2001coverage]. However, the soft obstacle strategy can suffer from high computational complexity. The soft obstacle strategy is based on cellular decomposition. Decomposition of exploration regions in the soft obstacle strategy application is computationally costly. In interaction, robots determine exploration regions to travel to. Also, the coordination strategy is not recommended for situations where coordination involves a large number of robots, because the unknown environment might not be decomposed efficiently. If many robots have inaccessible regions, it might not be possible for each robot to acquire a non-overlapping region independently before the end of the search. Experiments on this aspect will be investigated in future. Results and Discussion {#sec:experiments} ====================== For experimental investigation, the accidental rendezvous strategy (ARS), the periodic rendezvous strategy (PRS) and our proposed soft obstacle strategy (SOS) are considered. Two aspects will then be investigated: (1) Experimental investigation of the sustainability of the exploration performance of robots which start searching within the same vicinity. (2) Theoretical analysis of the most appropriate strategy for coordination of solitary robots. For the experimental investigation, we consider four ways to compare the three strategies (following the representation of [@IEEEexample:wellman2011using; @IEEEexample:hourani2013serendipity]): (i) We run a one-shot experiment and we plot robot individual coverage at the end of the search. (ii) We show the robot coverage over time during the search. (iii) We run a one-shot experiment with each strategy and we show the robots’ individual trajectories. (iv) We run $150$ experiments, $50$ each on three different environments, and plot means and standard deviations of coverage per strategy (Figure \[fig:error\]). In terms of results, we expect ARS and SOS to outperform PRS in terms of coverage, because PRS suffers from interruptibility. On the other hand, PRS should outperform ARS and SOS in sustainability. We expect SOS to outperform ARS in both coverage and sustainability. We use the standard deviations of robot coverages to measure sustainability. This should be small in the case of sustainable performance. Performance will be evaluated based on coverage area. The maximum area that a robot can cover, $A(\tau)$, can be computed as in Equation \[eq:area\]. Coverage will be reported as a percentage of $A(\tau)$. Experimental setup ------------------ The perception range was set to $r=20$ and the motion scale to $\gamma=1$. Three unstructured worlds were considered (Figure \[fig:environment\]) and location-based sensing was simulated. The environment used to run simulations was $480\times 600$. The simulator was built in Python and Qt, a C++ cross-platform application framework for graphical user interfaces. We included our own implementation of ARS and PRS into the simulator. In PRS, the rendezvous time $t_r$ was set as done in [@IEEEexample:hourani2013serendipity] as, $$t_r = 2\times a + b\,,$$ where $a$ is the time required to travel between the rendezvous point and the current locations of the involved robots, and $b$ is a threshold which provides a time that robots can use to explore as they move to the rendezvous point. For the first rendezvous, $a$ is set to $50$ steps. For further rendezvous, it depends on the area of the shared explored regions between the robots involved. The rendezvous point is chosen by the leader in the shared explored region. The known upper bound amount of time for search is determined as follows, $$\tau= k\times \bigg(\frac{ w h}{2\gamma rN}-\frac{\pi r}{2\gamma}\bigg)\,,$$ where $w$ and $h$ denote the width and the height of the search environment, and $k$ is defined in $ [0.5, 0.8]$. The factor $k$ is used to vary the width of the area to explore. Experiments were conducted (Table \[tab:experiments\]) with $k=0.6$. $N$ $2$ $3$ $4$ $7$ $8$ $10$ -------- --------- --------- --------- --------- --------- --------- -- -- $\tau$ $2141$ $1421$ $1061$ $598$ $521$ $413$ $A$ $86896$ $58096$ $43696$ $25176$ $22096$ $17776$ : *Experimental setup: for each number of robots ($N$) and the time for search ($\tau$), the maximum possible coverage per robot $A(\tau)$ for this time will be used to evaluate robots’ performance.*[]{data-label="tab:experiments"} [0.1]{} ![*The simulation environments used.*[]{data-label="fig:environment"}](figs/world.png "fig:"){width="\textwidth"} [0.1]{} ![*The simulation environments used.*[]{data-label="fig:environment"}](figs/world2.png "fig:"){width="\textwidth"} [0.1]{} ![*The simulation environments used.*[]{data-label="fig:environment"}](figs/world3.png "fig:"){width="\textwidth"} It should be noted that, this work does not define a process for determining environments to run experiments. Environments are randomly generated. Results ------- Figure \[fig:experiment2\] shows the results for two robots in the unknown environment illustrated in Figure \[fig:environment\]. The three strategies perform well in terms of sustainability. The performance (in intervals in $\%$) of $A(\tau)$ of SOS, ARS and PRS are $87.857\pm 0.437$, $78.224\pm 3.715$ and $50.755\pm 0.326$ respectively. Robots had an even assignment. ARS and SOS have good performance in terms of coverage. But PRS has less coverage due to interruptibility which took roughly $49\%$ of the search time. In Figure \[fig:experiment31\], robots also had an even assignment (the horizontal straight lines on PRS correspond to interruptibility), while Figure \[fig:experiment32\] shows the results for an uneven assignment. With ARS for the latter, two robots ($R_1$ and $R_3$) suffered from high interference. [0.24]{} ![*Individual coverage performance for a team of two robots. **(\[fig:indicov2\])**: final coverage achieved by individual robots. **(\[fig:progresscov2\])**: coverage trends of coverage of individual robots over time. **(\[fig:sostraj2\])**: trajectories of robots using SOS. Coloured dots are initial locations and coloured arrows are ending points of robot trajectories. **(\[fig:arstraj2\])**: trajectories of robots using ARS. **(\[fig:prstraj2\])**: trajectories of robots using PRS.*[]{data-label="fig:experiment2"}](figs/experiment2_coverage.pdf "fig:"){width="\textwidth"} [0.24]{} ![*Individual coverage performance for a team of two robots. **(\[fig:indicov2\])**: final coverage achieved by individual robots. **(\[fig:progresscov2\])**: coverage trends of coverage of individual robots over time. **(\[fig:sostraj2\])**: trajectories of robots using SOS. Coloured dots are initial locations and coloured arrows are ending points of robot trajectories. **(\[fig:arstraj2\])**: trajectories of robots using ARS. **(\[fig:prstraj2\])**: trajectories of robots using PRS.*[]{data-label="fig:experiment2"}](figs/experiment2_progress.pdf "fig:"){width="\textwidth"} [0.24]{} ![*Individual coverage performance for a team of two robots. **(\[fig:indicov2\])**: final coverage achieved by individual robots. **(\[fig:progresscov2\])**: coverage trends of coverage of individual robots over time. **(\[fig:sostraj2\])**: trajectories of robots using SOS. Coloured dots are initial locations and coloured arrows are ending points of robot trajectories. **(\[fig:arstraj2\])**: trajectories of robots using ARS. **(\[fig:prstraj2\])**: trajectories of robots using PRS.*[]{data-label="fig:experiment2"}](figs/experiment2_traj_SOS.pdf "fig:"){width="\textwidth"} [0.24]{} ![*Individual coverage performance for a team of two robots. **(\[fig:indicov2\])**: final coverage achieved by individual robots. **(\[fig:progresscov2\])**: coverage trends of coverage of individual robots over time. **(\[fig:sostraj2\])**: trajectories of robots using SOS. Coloured dots are initial locations and coloured arrows are ending points of robot trajectories. **(\[fig:arstraj2\])**: trajectories of robots using ARS. **(\[fig:prstraj2\])**: trajectories of robots using PRS.*[]{data-label="fig:experiment2"}](figs/experiment2_traj_ARS.pdf "fig:"){width="\textwidth"} [0.24]{} ![*Individual coverage performance for a team of two robots. **(\[fig:indicov2\])**: final coverage achieved by individual robots. **(\[fig:progresscov2\])**: coverage trends of coverage of individual robots over time. **(\[fig:sostraj2\])**: trajectories of robots using SOS. Coloured dots are initial locations and coloured arrows are ending points of robot trajectories. **(\[fig:arstraj2\])**: trajectories of robots using ARS. **(\[fig:prstraj2\])**: trajectories of robots using PRS.*[]{data-label="fig:experiment2"}](figs/experiment2_traj_PRS.pdf "fig:"){width="\textwidth"} [0.24]{} ![*Individual coverage performance for a team of three robots.*[]{data-label="fig:experiment31"}](figs/experiment32_coverage.pdf "fig:"){width="\textwidth"} [0.24]{} ![*Individual coverage performance for a team of three robots.*[]{data-label="fig:experiment31"}](figs/experiment32_progress.pdf "fig:"){width="\textwidth"} [0.24]{} ![*Individual coverage performance for a team of three robots.*[]{data-label="fig:experiment31"}](figs/experiment31_traj_SOS.pdf "fig:"){width="\textwidth"} [0.24]{} ![*Individual coverage performance for a team of three robots.*[]{data-label="fig:experiment31"}](figs/experiment31_traj_ARS.pdf "fig:"){width="\textwidth"} [0.24]{} ![*Individual coverage performance for a team of three robots.*[]{data-label="fig:experiment31"}](figs/experiment31_traj_PRS.pdf "fig:"){width="\textwidth"} [0.24]{} ![*Individual coverage performance for a team of three robots*.[]{data-label="fig:experiment32"}](figs/experiment32_coverage.pdf "fig:"){width="\textwidth"} [0.24]{} ![*Individual coverage performance for a team of three robots*.[]{data-label="fig:experiment32"}](figs/experiment32_progress.pdf "fig:"){width="\textwidth"} [0.24]{} ![*Individual coverage performance for a team of three robots*.[]{data-label="fig:experiment32"}](figs/experiment32_traj_SOS.pdf "fig:"){width="\textwidth"} [0.24]{} ![*Individual coverage performance for a team of three robots*.[]{data-label="fig:experiment32"}](figs/experiment32_traj_ARS.pdf "fig:"){width="\textwidth"} [0.24]{} ![*Individual coverage performance for a team of three robots*.[]{data-label="fig:experiment32"}](figs/experiment32_traj_PRS.pdf "fig:"){width="\textwidth"} [0.24]{} ![*Individual coverage performance for a team of four robots.*[]{data-label="fig:experiment4"}](figs/experiment4_coverage.pdf "fig:"){width="\textwidth"} [0.24]{} ![*Individual coverage performance for a team of four robots.*[]{data-label="fig:experiment4"}](figs/experiment4_progress.pdf "fig:"){width="\textwidth"} [0.24]{} ![*Individual coverage performance for a team of seven robots*.[]{data-label="fig:experiment7"}](figs/experiment7_coverage.pdf "fig:"){width="\textwidth"} [0.24]{} ![*Individual coverage performance for a team of seven robots*.[]{data-label="fig:experiment7"}](figs/experiment7_progress.pdf "fig:"){width="\textwidth"} [0.24]{} ![*Error bars for the coordination strategies’ performance. Robots start from random locations in close proximity to each other. **(\[fig:mc8\])**: error bars with a team of eight robots. **(\[fig:mc10\])**: error bars with a team of ten robots. The normalised coverage (in $\%$) intervals obtained with SOS, ARS and PRS are $83.140\pm 6.913$, $68.108\pm 13.189$ and $37.016\pm 4.269$ for the eight robots respectively; and $80.239\pm 5.909$, $63.053\pm 13.046$ and $31.090\pm 2.574$ for the ten robots respectively.*[]{data-label="fig:error"}](figs/monte1.pdf "fig:"){width="\textwidth"} [0.24]{} ![*Error bars for the coordination strategies’ performance. Robots start from random locations in close proximity to each other. **(\[fig:mc8\])**: error bars with a team of eight robots. **(\[fig:mc10\])**: error bars with a team of ten robots. The normalised coverage (in $\%$) intervals obtained with SOS, ARS and PRS are $83.140\pm 6.913$, $68.108\pm 13.189$ and $37.016\pm 4.269$ for the eight robots respectively; and $80.239\pm 5.909$, $63.053\pm 13.046$ and $31.090\pm 2.574$ for the ten robots respectively.*[]{data-label="fig:error"}](figs/monte2.pdf "fig:"){width="\textwidth"} Figure \[fig:experiment4\] and Figure \[fig:experiment7\] show results for teams of four and seven robots respectively. In Figure \[fig:experiment4\] robots had an uneven assignment. An uneven assignment does not affect the sustainability of PRS significantly because robots meet on a regular basis. Also ARS provides a good coverage performance ( $\approx 25\%$ of interference which includes overlap and backtracking). In Figure \[fig:experiment7\], $R_1$ has a low coverage as it was exploring an already explored region. As expected, PRS and SOS provide more sustainable performance than ARS. Trajectories of robots (Figures \[fig:experiment2\] – \[fig:experiment7\]) show that when SOS robots may explore non-overlapping regions. In the ARS case there is overlapping of regions. Figure \[fig:error\] shows summary coverage results from $150$ experiments. The results confirm that PRS provides the most sustainable performance (its standard deviation is very small) but with less coverage due to interruptibility. The robot performance obtained with SOS is more sustainable than that with ARS. SOS outperforms other strategies in coverage. Discussion ---------- ### Sustainability of exploration performance the soft obstacle strategy works well when the search time is known in advance. Robots evaluate their exploration regions based on the amount of remaining search time. The case of unlimited time for search suits PRS the most. ARS provides sustainable performance provided that robots have an even assignment in interaction [@IEEEexample:wellman2011using]. Otherwise, robots’ explorations will overlap. But sustainable performance is guaranteed in PRS [@IEEEexample:hourani2013serendipity], since robots maintain collaboration through scheduled meetings. However, the approach is affected by interruptibility, which can hinder fast coverage. SOS can provide sustainable performance where the amount of search time is limited and known. While coordinating, the size of exploration regions assigned to robots by cellular decomposition is a function of the amount of time for search and the number of robots in interaction. In the case of an uneven assignment with SOS, each of the robots concerned considers the regions of others as soft obstacles and avoids exploring these regions. However, in the case of uneven assignment with a large number of interacting robots, SOS might not outperform the other two strategies. In addition, when considering a large number of robots, some robots may take long paths to reach their assigned regions. Consequently, these robots may spend more time traversing than searching. ### Applicability of the strategies for solitary robots PRS is used to manage a team of robots which can schedule meetings among themselves. In the case of solitary robots, a robot can encounter different robots at different times. For such robots to be incorporated into the team, a means would be required to schedule them into future rendezvous. Moreover, a robot encountering more than one team could have the problem of being a candidate for many other rendezvous. For instance, what happens if a robot which is going to rendezvous encounters another robot from another local group also going to rendezvous with its mates? If both decide to interact and share information, they might miss their respective meetings. If they do not interact, each misses the additional information from the other. Based on our knowledge, this issue is not yet fully addressed in the literature. However, for situations such as map building where robots are bound by a team goal, ARS and SOS might not be recommended. Bound by a team goal, robots need to maintain collaboration. In ARS and SOS, collaboration of robots is always accidental. Suppose a robot $R_j$ (that $R_i$ has previously met) meets another robot $R_k$ which $R_i$ is not aware of. How can $R_i$ learn from $R_k$? This situation can be nicely handled with PRS. Another potential difficulty of strategies which utilise accidental rendezvous, such as ARS and SOS, is that in some cases a large amount of time may be spent in these accidental interactions. This might particularly be so if there is a large number of robots relative to the coverage area. Conclusion {#sec:conclusion} ========== A novel interruptibility-free strategy was designed to mitigate interference with sustainable performance in situations where solitary robots search an unknown environment when a known upper bound on the search time is insufficient to allow search of the entire search space. An improvement is observed when robots have an uneven assignment. An uneven assignment is when search areas assigned to robots are unbalanced. This can arise, for example, when robots ignore the fact that they meet close to a boundary of the search environment. However, the soft obstacle strategy is not recommended when robots are bound by a team goal (e.g. map construction) as robots need to maintain collaboration. In the periodic rendezvous strategy, exchange of individual information between robots is maintained. The soft obstacle strategy is recommended for self-interested robots. The soft obstacle strategy proposed focuses on a limited number of robots for coordination. A further avenue of research could be to extend the work by considering how a large number of robots could apply the strategy for coordination. 0.1in Acknowledgements {#acknowledgements .unnumbered} ================ The authors would like to thank Prof. Jeff Sanders for fruitful exchanges and his sharp scientific opinions. S.W. Utete and J.F. Masakuna are members of the AIMS[^3] Research Centre which receives support from the NRF[^4]. This support is acknowledged with thanks. [^1]: But the sensing area can take other shapes. This should only have a slight impact on the initial and final conditions of the solution. [^2]: But other algorithms could be applied such as the spiral search and frontier-based search. [^3]: African Institute for Mathematical Sciences (South Africa). [^4]: National Research Foundation.

ArXiv

--- abstract: 'Identifying a patient’s key problems over time is a common task for providers at the point care, yet a complex and time-consuming activity given current electric health records. To enable a problem-oriented summarizer to identify a patient’s comprehensive list of problems and their salience, we propose an unsupervised phenotyping approach that jointly learns a large number of phenotypes/problems across structured and unstructured data. To identify the appropriate granularity of the learned phenotypes, the model is trained on a target patient population of the same clinic. To enable the content organization of a problem-oriented summarizer, the model identifies phenotype relatedness as well. The model leverages a correlated-mixed membership approach with variational inference applied to heterogenous clinical data. In this paper, we focus our experiments on assessing the learned phenotypes and their relatedness as learned from a specific patient population. We ground our experiments in phenotyping patients from an HIV clinic in a large urban care institution (n=7,523), where patients have voluminous, longitudinal documentation, and where providers would benefit from summaries of these patient’s medical histories, whether about their HIV or any comorbidities. We find that the learned phenotypes and their relatedness are clinically valid when assessed qualitatively by clinical experts, and that the model surpasses baseline in inferring phenotype-relatedness when comparing to existing expert-curated condition groupings.' author: - 'Gal Levy-Fix' - Jason Zucker - Konstantin Stojanovic - Noémie Elhadad bibliography: - 'ctm\_bib2.bib' title: Towards Patient Record Summarization Through Joint Phenotype Learning in HIV Patients --- Introduction ============ Electronic health records (EHR) have improved the availability of patient records, but this has not always translated to increased availability of relevant information to clinicians [@christensen_Instant_2008]. This is partly because increased amounts of data in EHRs has made it more difficult for clinicians to review patients’ previous medical histories and obtain an overview of the patient record [@JENSEN201644]. Increased amounts patient data have also raised concerns regarding clinician information overload [@farri_Qualitative_2012], having effects on care quality[@cj_Use_2014], and patient safety[@rj_Cognitive_2011]. Patient record summarization has been suggested as a valuable tool to support clinicians in making sense of increasingly large patient records[@feblowitz_Summarization_2011]. There are a number of open challenges associated with robust summarization of clinical documentation [@pivovarov_Automated_2015], including content selection—identifying the right summary elements at the right granularity in the input patient record— and content organization—organizing summary output in a coherent and actionable fashion for the clinicians, all the while preserving data provenance. Previous work has shown that problem-oriented summary supports the needs of clinicians [@weed_Medical_1968; @hirsch_HARVEST_2014; @lee_Public_2015; @li_Impact_2018]. High-throughput computational phenotyping methods are attractive for identifying a patient’s problems in a robust and scalable fashion [@ho_Limestone_2014; @pivovarov_Learning_2015; @joshi_Identifiable_2016]. Considering the characteristics of electronic health record (EHR) data (missingness, heterogeneity, uncertainty), Bayesian generative approaches are attractive to handle them and provide easily interpretable outputs that quantify their uncertainty. To enable a problem-oriented summarizer to identify a target patient’s comprehensive list of problems and their salience, we propose a probabilistic machine learning approach that can identify a large number of phenotypes/problems using patients’ structured and unstructured data in an unsupervised fashion. The machine learning model is trained on the EHR data of many patients to simultaneously learn probabilistic definitions of many phenotypes at the same time. Figure \[fig:approach\] shows a graphical schema of the proposed approach . To identify the appropriate granularity of the learned phenotypes, the model is trained on a target patient population of the same clinic. Each phenotype definition is composed of diagnoses, medications, laboratory tests, and clinical notes that have been observed to commonly co-occur in the training patient population. Figure \[fig:ModelOutput1\] shows an example phenotype learned by the model. Figure \[fig:ModelOutput2\] shows phenotype-phenotype correlations learned by the model. Phenotypes are labeled with their most probable diagnosis code. The learned phenotypes from the model are then used to summarize a single patient EHR data over time. Figure \[fig:summary\] shows an automatically generated example summary of a single patient record over a five year period that leverages the proposed approach.\ ![Example of five learned phenotypes and their learned correlations[]{data-label="fig:approach"}](pics/approach_vis_extended.pdf){width="\linewidth"} ![Example of learned phenotype and its probabilistic definition across the four data types (yellow for diagnosis codes, green for notes, purple for medications, and blue for laboratory tests). The mostly likely diagnosis code is assigned as label for the phenotype.)[]{data-label="fig:ModelOutput1"}](pics/phenotypeexample.pdf){width="\linewidth"} ![Example of five learned phenotypes and their learned correlations[]{data-label="fig:ModelOutput2"}](pics/corr_example.pdf){width="\linewidth"} ![Example of patient-specific summary over five years. The top five most salient problems in 2019 are visualized and how their documentation has evolved through time. In this setup, the summary was produced at the year level by binning the patient’s documentation for that time resolution. The patient has HIV-specific problems, although their HIV is becoming asymptomatic, as well as comorbidities, all cardiac in nature. (Relations among the inferred phenotypes are not shown. Dates are changed to maintain patient privacy.)[]{data-label="fig:summary"}](pics/ptLabeledSummary.pdf){width="\linewidth"} In this work we focus on phenotyping a population of patients from an HIV clinic in a large urban care institution. HIV-positive patient experience a high burden of disease, with many comorbidities due to the inflammatory nature of the virus and the toxicity of their medications [@gallant_Comorbidities_2017]. Moreover, due to high healthcare utilization, patients have complex and long medical histories that are difficult to sift through, exacerbating the need for summarization. We hypothesize that **1)** the model will learn many clinically valid phenotypes and phenotype relationships; **2)** training the model on an HIV-positive population will result in the identification of several HIV phenotypes, representing the different presentations and progression stages of HIV—a granularity that would likely be missed if trained on a more general and heterogeneous patient population; **3)** the model will also learn non-HIV phenotypes, representative of the many comorbidities of HIV; **4)** the model will identify correlations among phenotypes that indicate clinically valid relations of different types beyond simple is-a relationships. Methods ======= The model --------- The model we proposed is based on the correlated topic model (CTM) [@blei_Correlated_2007]. In our context, topics are equivalent to phenotypes and documents are the patient records. We make a methodological contribution by expanding the CTM to support multiple input sources, beyond the single input source usually assumed in topic modeling. We make this important expansion to the model since unlike topic identification in general text, clinical documentation is more than just clinical notes. Instead, our model is able to learn phenotype definitions through identifying co-occuring patterns in clinical notes, laboratory tests, ordered medications, and diagnosis codes. Incorporating multiple sources of data into the phenotypes definitions allows for more robust phenotype definitions that can help overcome the inaccuracies present in just relaying on an single source of patient data. This is supported by previous work that has shown that incorporating heterogenous data yields superior phenotypes [@pivovarov_Learning_2015]. It has been previously proposed to leverage topic-model like models to learn clinical phenotypes. Our model differs in that we do not assume that phenotypes identified in each patient record are independent from one another. We remove the assumption of independence by allowing for phenotypes to be correlated. To do this our model, like the original CTM, replaces the traditional Dirichlet distribution used in Latent Dirchelet Allocation (LDA) [@blei_Latent_2003] to govern topic proportions with a logit-normal distribution[@blei_Correlated_2007]. The logit-normal distribution allows for phenotype proportions in each patient record to be correlated with one another (through the normal covariance matrix) but also sum up to 1 or 100% of the patient record, as desired when modeling proportions. Changing the previously assumed Dirchelet distribution with a logit-normal distribution removes the conditional conjugacy between the posterior distribution and prior distribution of the phenotype proportions. To perform posterior inference Wang and Blei [@wang_Variational_2013] propose Laplace variational inference, a generalized form of variational inference that can handle non-conjugate models. In this paper we generalize the proposed Laplace Variational Inference even further to allow for multiple input types. This makes the model inference especially relevant to clinical data which contains many different data types. The model training is time-agnostic and treats each patient record as bag of observations, one for each data type. While motivation behind the model is to assign phenotypes on a single patient level for patient-level summarization, in this paper we focus on the learned phenotypes on the population level. Each phenotype is labeled using the most probably diagnosis code. The generative process of each patient record (D) with $N_m$ number of tokens for $M$ data types is provided below. The graphical representation of the model is presented in Figure \[fig:model\]. 1. Draw log phenotype proportion $\nu_d \sim N(\mu_0, \Sigma_0)$ 2. For each $n_m$ token ($x_{d,n_m}$) in data type ($m=1,...,M$): 1. Draw phenotype assignment $z_{d,n_m}|\nu_d \sim Mult(\pi(\nu_d))$ 2. Draw token $x_{n_m}|z_{n_m}, \beta_{k,m} \sim Mult(\beta_{z_{n_{m}}})$ ![The graphical representation of the multi-input correlated topic model. Multiple inputs are represented by the additional plate notation M that is not present in the single-input CTM model. []{data-label="fig:model"}](pics/model.pdf){width="\linewidth"} Probabilistic inference ----------------------- The phenotype definitions and their correlations with one another are obtained through performing Bayesian posterior inference which estimates the conditional probability of the unobserved or latent model variables given the observed model variables. In the case of the proposed model this means calculating the probability of the phenotype proportions of each patient record ($\nu$) and phenotype assignment of each input ($Z_{nm}$) given the observed patient data ($X_{nmd}$) and phenotype distributions ($\beta$), or $p(\nu,z_{}| x_{d,n_m}, \beta)$. When the posterior distribution has a conjugate prior this greatly simplifies the Bayesian analysis and allows for the use of popular sampling methods for approximate inference such as Markov chain Monte Carlo sampling such as Gibbs sampling as employed in [@pivovarov_Learning_2015]. However, conjugacy limits the types of distributions used in the model, and thus restricts the flexibility of data modeling. In order to allow for phenotypes to be correlated with one another the prior distribution used to model the phenotype proportions in the patient record needs to allow for phenotype correlations. Since that is not possible with the Dirchelet distribution, it needs to be replaced with a different distribution that meets this criteria. However since the Dirichlet distribution is the conjugate prior to the multivariate distribution used to model the phenotype data assignments, this modeling change means that the model losses its conditional conjugacy. Hence, deterministic approximate inference methods such as variational inference is more feasible than other sampling methods. By contrast to sampling approximation methods for inference, the theoretical guaranties of convergence of variational inference methods to the true posterior have been less studied. However, variational inference has become a popular inference method in Bayesian statistics as it tends to be faster and scale better with large and complex data [@blei_Variational_2017]. Even in variational inference, some popular implementations such as mean-field variational depend of conjugate models. Wang and Blei [@wang_Variational_2013] propose Laplace variational inference, a generalized form of variational inference that can handle non-conjugate models. The method uses Laplace approximations in the coordinate ascent updates within the variational optimization problem. This methods was shown to generalize to different types of non-conjugate model and have superior performance compared to the original ad-hoc inference method previously proposed here [@blei_Correlated_2007]. In this paper we generalized the Laplace Variational Inference for multiple input types. The mathematical derivation of the Laplace variational inference with multiple input types is shown below. Inference mathematical derivation --------------------------------- As presented in the graphical model (see Figure \[fig:model\]), the under-script $m$ represents the $m$-th input type, where $m=1,...,M$. The derivation below contributed to the previously proposed inference by [@wang_Variational_2013] by allowing for M input types instead of a single input type. The model is represented by the joint probability distribution in equation (1). The inference problem is to solve for the posterior distribution which is the conditional distribution of the latent variables $\nu$ and $z$ given $x$ in equation (2). $$p(\nu, z, x)= \prod_{m=1}^{M}p(x_m|z_m)p(z_m|\nu)p(\nu)$$ $$p(\nu, z|x)=\frac{p(\nu, z, x)}{\int p(\nu,z,x)dz d\nu}$$ The integral in the denominator of equation (2) is intractable to compute exactly [@blei_Correlated_2007]. As proposed by [@wang_Variational_2013] the posterior is approximated using Laplace Variational Inference through optimization. A family of densities are posited over the latent variables. The model assumptions include: 1. The variational distribution is fully factorized: $$q(\nu, z)=q(\nu)\prod_{m=1}^{M}q(z_m)$$ 2. $\nu$ is real valued and $p(\nu)$ is twice differentiable with respect to $\nu$ 3. The distribution $p(z_m|\nu)$ is in the exponential family: $$p(z_m|\nu)=h(z_m)exp\{\eta(\nu)^Tt(z_m)-a(\eta(\nu))\}$$ 4. The distribution $p(x_m|z_m)$ is in the exponential exponential family such that: $$p(x_m|z_m)=h(x_m)exp\{t(z_m)^T<t(x_m),1>\}$$ In variational inference the approximation for the posterior distribution is obtained through minimizing the Kullback-Leiber (KL) divergence to the exact posterior. $$q^*(\nu,z)=argmin KL(Q(\nu, z) || P(\nu, z|x)$$ Under standard variational inference theory minimizing the KL divergence between $q(\nu,z)$ and the true posterior $p(\nu,z|x)$ is the same as maximizing the lower bound of the log marginal likelihood of observed data $x$. Using Jensens’s inequality the variational object $L(q)$ is defined by equation (5). $$\begin{aligned} log p(x) & = log \int p(\nu,z,x)dz d\nu \\ & \ge E_q[log (p(\nu, z, x))] - E_q[log(q(\nu,z))] \\ & = E_q[log (p(\nu, z, x))] - E_q[log(q(\nu)\prod_{m=1}^{M}q(z_m)] \\ & \equiv L(q) \end{aligned}$$ Setting the partial derivative of $L(q)$ with respect to $q$ to zero provides the optimal variational updates to $q(\nu)$ and $q(z_m)$ seen in Equations (9) and (10). When $p(\nu)$ is conjugate to $p(z_m|\nu)$ then equations (5) and (6) have closed form solutions. In the case of this non-conjugate model [@wang_Variational_2013] put forward approximates to the updates using Laplace approximation. $$q^*(\nu) \propto exp \{ E_{q(z)}[log \prod_{m=1}^{M} p(z_m|\nu)p(\nu)]\}$$ $$\begin{aligned} q^*(z_1) &\propto exp \{ E_{q(\nu)}[log p(x_1|z_1)p(z_1|\nu)]\} \\ &\vdots \\ q^*(z_m) &\propto exp \{ E_{q(\nu)}[log p(x_m|z_1)p(z_m|\nu)]\} \end{aligned}$$ The following is the derivation of the variational update to $q^*(\nu)$ using the previously stated assumption that $p(z_m|\nu)$ is assumed to belong to the exponential family. $$\begin{aligned} q^*(\nu) &\propto exp \{E_{q(z)}[log\prod_{m=1}^Mp(z_m|\nu)p(\nu)]\} \\ &=exp\{E_{q(z)}[logp(\nu)+\sum_{m=1}^Mlog(z_m|\nu)]\} \\ &=exp\{E_{q(z)}[\sum_{m=1}^M log (h(z_m)exp\{\eta(\nu)^Tt(z_m)-a(\eta(\nu))\})) +log p(\nu)]\} \\ &= exp\{E_{q(z)}[\sum_{m=1}^M (\eta(\nu)^Tt(z_m)-a(\eta(\nu)) ) +logp(\nu) ] \} \\ &=exp\{E_{q(z)}f(\nu)\} \end{aligned}$$ The function $f(\nu)$ in Equation (10) has no closed form and this is approximated with the following 2nd order Taylor approximation around $\hat{\nu}$ which is the $\nu$ that maximizes $\nabla f(\nu)$. $$f(\nu)\approx f(\hat{\nu})+ \nabla f(\hat{\nu})(\nu - \hat{\nu}) + \frac{1}{2}(\nu - \hat{\nu})^T\nabla^2 f(\hat{\nu})(\nu - \hat{\nu})$$ Thus the update for $q(\nu)$ is approximate with $\mathcal{N}(\hat{\nu},-\nabla^2f(\hat{\nu})^{-1}) $ The sufficient statistics of the exponential family are: $$\begin{aligned} h(z_m)&=1 \\ t(z_m)&=\sum_n z_{mn} \\ \eta(\nu)&=\nu - log \{\sum_k exp \{ \nu \} \\ a(\eta(\nu))&=0 \end{aligned}$$ Using the sufficient statistics above $f(\nu)$ is the following: $$\begin{aligned} f(\nu) &= \sum_{m=1}^M (\eta(\nu)^T - a(\eta(\nu))-\frac{1}{2} (\nu - \mu_0)^T\Sigma_0^{-1}(\nu-\mu_0) \\ &= \eta(\nu)^T\sum_{m=1}^M \{E_{q(z)}[t(z_m)]\} - \frac{1}{2}(\nu-\mu_0)^T\Sigma_0^{-1}(\nu-\mu_0) \end{aligned}$$ The first derivative and second derivative of $f(\nu)$ are the following: $$\begin{aligned} \nabla f(\nu) &= \pi_i (1_{[i=j]}- \pi_j)\sum_{m=1}^M E_{q(z)}[t(z_m)]-\Sigma_0^{-1}(\nu - \mu_0) \\ &=\sum_{m=1}^M E_{q(z)}[t(z_m]- \pi\sum_{k=1}^K[\sum_{m=1}^ME_{q(z)}[t(z_m)]]_k - \Sigma_0^{-1}(\nu -\mu_0) \end{aligned}$$ where $\pi \propto exp\{\eta(\nu)\}$ $$\nabla^2 f(\nu)_{ij} = (-\pi_i 1_{i=j}+ \pi_i \pi_j)\sum_{k=1}^K[\sum_{m=1}^M E_{q(z)}[t(z)]]_k - (\Sigma_0^{-1})_{ij}$$ The update to $q(z_m)$ where $m= 1,...,m$ is the following: $$\begin{aligned} q^*(z_m) &\propto exp \{ E_{q(\nu)} [log p(x_m |z_m)p(z_m|\nu)] \} \\ &= exp \{ log p(x_m |z_m)+ E_{q(\nu)}[log p(z_m|\nu)]\} \end{aligned}$$ Using the exponential form of $p(z_m|\nu)$ and $p(x_m|z_m)$: $$\begin{aligned} log q(z_m)&= log p(x_m| z_m)+ E_{q(\nu)}[log p(z_m|\nu)] \\ &=log p(x_m|z_m)+log h(z_m)+ E_{q(\nu)}[\eta(\nu)^T]t(z_m)- C \\ &=log h(z_m) + t(z_m)^T<t(x_m), 1> + log h(z_m) + \\ & E_{q(\nu)}[\eta(\nu)^T]t(z_m)-C \end{aligned}$$ $$q(z_m)\propto h(z_m)exp\{E_{q(\nu)}[\eta(\nu)^T]+ t(x_m)^Tt(z_m)\}$$ Dataset ------- The model was trained on the EHR data of 7,523 patients from an HIV clinical from a large urban care institute. Patient data used was fully identified for which the use was approved by the Institutional Review Board of our institution. The data spanned 8 years and included the data types: words from clinical notes, laboratory tests ordered, medication orders, and assigned diagnoses codes from across all clinical settings (inpatient, outpatient, emergency). For the purpose of the model training each patient record was restricted to the most recent 2.5 years data. The final training dataset included the following total data counts and unique vocabulary size in brackets: total words from clinical notes: 128,034,516 (unique: 25,894); total laboratory tests: 463,524 (unique: 129); total medications: 510,820 (unique: 6,714); and total diagnosis codes: 246,623 (unique: 2,956). Model training and parameter selection -------------------------------------- The parameters of the normal distribution governing phenotype proportions $\nu_d$ were initialized with $\mu_0$ equal to a zero vector and $\Sigma_0$ set to the identity matrix. The phenotype distribution $\beta_{k,m}$ for each input type was initialized with a Uniform distribution over the (K-1) simplex. This equivalent to initializing topics with a Dirichlet distribution with parametrization of 1. A small amount of random positive noise was added to each uniform distribution so there was a small variation in the initial phenotypes. Three alternatives of the model were estimated (K=50, 100, 250). To identify the best performing model of the three alternative number of phenotype (K=50, 100, 250), a clinical expert reviewed 20 randomly selected phenotypes from each model. The best performing model is further evaluated for the clinical correctness of the phenotypes and phenotype-relatedness learned by the model. Evaluation Setup ================ We evaluate our hypotheses 1 through 4 using a mixture of qualitative and quantitative evaluations. Qualitative evaluation of the phenotypes and phenotype relatedness was performed by two clinical experts. The quantitative evaluation was performed through a comparison to the Clinical Classification Software (CCS), which provides expert-curated manual classification of diagnosis codes into largely clinically homogeneous groups [@agencyforhealthcareresearchandquality_HCUP_2017] Hypothesis 1: clinical validity ------------------------------- To evaluate the clinical validity of the learned phenotypes, 50 randomly selected phenotypes were evaluated independently by two clinicians. The phenotypes were evaluated according to their coherence, granularity, and label quality [@pivovarov_Learning_2015]. Previous works citing clinical evaluation of phenotypes by experts have reported the scoring of a single clinician [@pivovarov_Learning_2015; @ho_Limestone_2014]. Since this scoring can very subjective, we opted for two clinicians to score the phenotypes and the final score assigned is the average of the two clinicians. Since we did not want the opinion of one clinicians to be influenced by the other, there was not adjudication stage in the scoring (common on qualitative rating tasks made by more than one reviewer). This made the qualitative evaluation a very stringent task. We provide an analysis of the agreement between the clinicians scoring which can illuminate the level of subjectivity of this type of evaluation. #### Phenotype coherence. Phenotype coherence is meant to capture the quality of each learned phenotype according to its most probably observations. A coherent phenotype is defined to describe a single condition with few or no unrelated observations (clinical words, labratory tests, medications, and diagnosis codes). The expert was asked to rate each phenotype as having: ‘bad coherence’ (score=1) , ‘some coherence’ (score=2), ‘good coherence’ (score=3), or ‘excellent coherence’ (score=4). Phenotypes with ‘bad coherence’ should look like a random combination of observations, ‘some coherence’ indicates the observations assigned to the phenotype are somewhat related to one another, ‘good coherence’ indicates the phenotype is a very good representation of a disease, and ‘excellent coherence’ indicates the phenotype definition has almost no unrelated observations assigned to it. #### Phenotype granularity. The clinical experts were asked to characterize the granularity of each randomly selected phenotype by assessing whether the model learned a ‘single disease’ (score=3), a ‘group of diseases’ (score=2), or a ’non-disease’ phenotype (score=1). #### Label quality. The representativeness of the automatically assigned phenotype label of the phenotype as a whole was evaluated. Each label was categorized by the clinical experts as ‘unrelated’ to the rest of the phenotype (score=1), ‘related’ to the rest of the phenotype (score=2), or ‘actionable’ (score=3). Labels that were deemed as actionable are those representative of a single phenotype and have the appropriate granularity to provide a clinician information that could be used without additional information to guide further testing, diagnosis, or counseling.\ #### Phenotype relatedness. Next, the clinical validity of the phenotypes-relatedness were evaluated by a single clinical expert. The expert reviewed all phenotypes relationships that were indicated to have a correlation greater than 0.5 correlation coefficient. Two sets of phenotype-relationships evaluated: 1) positive phenotype relationships learned between “more common” non-hiv phenotypes, defined as phenotypes that were represented in more than 5% of the patient population in our dataset; and 2) positive relationships learned between “rarer” non-hiv phenotypes, represented in 5% of sample population or less. The justification for evaluating relationships between “more common” phenotypes is that the model findings are grounded in more patient record, which could result in more robust findings. However, evaluating phenotype relationships identified between “rarer” phenotypes could still be interesting to assess in case the model is able to identify less known clinical relationships. Hypothesis 2: focus on HIV phenotypes ------------------------------------- Our second hypothesis was that since HIV is a complex disorder with diverse presentations and severity among patients, the model would identify several distinct HIV phenotypes. In evaluating this hypothesis, we wished to understand to what extent the model is able to learn multiple clinically valid HIV phenotypes and also characterize what those phenotypes were. To do so we had an HIV clinical expert review all the phenotypes automatically labeled as ’HIV’. The clinical expert was asked to i) indicate if the phenotype was clinically valid, ii) indicate if the phenotype was indeed an ’HIV’ phenotypes; and iii) give a more granular description of the phenotype if it was indeed an ’HIV’ phenotype in order to assess if the model identified disease progression, presentation, or acuity. Hypothesis 3: focus on non-HIV phenotypes ----------------------------------------- To assess if the model was able to learn diverse phenotypes, representative of the many comorbidities of HIV we quantitatively compared the phenotypes learned to the disease groups identified in the CCS. We did this by categorizing all 250 learned phenotypes according to their labels’ corresponding CCS level-1 category. If the model was able to learn phenotypes that fit into many CCS categories, we would could conclude that the model was able to learn diverse types of phenotypes, beyond HIV. Hypothesis 4: types of phenotype- relatedness ---------------------------------------------- We performed two evaluations to assess whether the model was able to identify correlations among phenotypes that indicate clinically valid relations of different types beyond simple is-a relationships. The first evaluation included a clinical expert review phenotypes identified by the model as highly related and determine what kind of relationship type the model learned. Example relationship types include comorbidities, same phenotype, phenotype subtype, and others. The second evaluation we counted how many significant relations learned by the model indicated an is-a relationship, as evidenced by same level-1 CCS categories, versus a more diverse relation type such as comorbidity when spanning different CCS categories. Results ======= The qualitative evaluation by the clinical expert indicated that the 250-phenotype model yielded the most coherent and granular phenotypes of the three models (K=50, 100, 250). All results below are described for the evaluation performed for the K=250 phenotype model. Hypothesis 1: clinical validity ------------------------------- #### Phenotype quality Of the 50 evaluated phenotypes from the 250-phenotype model, 10% of the phenotypes (n=5) were deemed to have no coherence (average coherence score of 1 or 1.5) while the large majority of evaluated phenotypes (n=45) were deemed to be coherent (with average coherence score of 2 or above) (see Figure \[fig:coherence\]). The most number of phenotypes were scored as having ’good coherence’ (n=13), followed by 12 phenotypes with an average of 3.5 (between ‘good coherence’ and ‘excellent coherence’). The ‘bad coherence’ phenotypes were found to be non-disease specific, but instead captured documentation related to general primary care visits. Figure \[fig:coherence\_examples\] shows the diagnosis codes of example phenotypes with coherence scores 1 (‘bad coherence’) through 4 (‘excellent coherence’) by both the clinical experts. The phenotypes in the example identified a clinic visit phenotype (scored 1), grouping of cancers phentoype (scored 2), grouping of heart diseases phenotype (scored 3), and an Aterial fibrillation phenotype (scored 4). Comparing the coherence phenotype scoring assigned by the two clinicians we found that the two clinicians had a low agreement on the exact coherence score assigned to the phenotypes (scores 1 through 4) but that the average difference between the scores was less than 1 point (0.9). This indicates that the clinicians evaluation of the phenotypes was not far apart. When comparing the clinician agreement on whether a phenotype was identified as not coherent (score of 1) versus coherent (score of 2 and above) the agreement was high, at 90% of the evaluated phenotypes (see Table  \[tab:agr\]). Of the 5 phenotypes that the reviewers did not agree on, 4 looked like HIV clinic well visits. The disagreement seemed to stem from whether the model identified a disease phenotype or a clinical-settings phenotype. An example such phenotype had the following top 5 diagnosis codes: ‘Human immune virus disease’, ‘Obesity NOS’, ‘Elevated blood pressure w/o hypertension’, ‘Hypertension NOS’, and ‘Laboratory exam NOS’. ![Phenotype coherence scores. Average score across the two clinical expert scores. Score 1=‘bad coherence’, 2=‘good coherence’, 3= ‘very good coherence’, 4=‘excellent coherence’.[]{data-label="fig:coherence"}](pics/coherence_scores.pdf){width="\linewidth"} ![Example phenotypes by coherence score assigned by the clinical experts. Each phenotype is represented here by its top diagnosis codes rather than all 4 data types for the sake of space. Score 1=‘bad coherence’, 2=‘good coherence’, 3= ‘very good coherence’, 4=‘excellent coherence’.[]{data-label="fig:coherence_examples"}](pics/phenotypeCoherenceexample.pdf){width="\linewidth"} ------------- -------------- -------------- ---------- Not Coherent Coherent Clinician 2 Not Coherent 1 1 Coherent 4 45 ------------- -------------- -------------- ---------- : Comparison of 2 clinician scoring for phenotype coherence[]{data-label="tab:agr"} The phenotype granularity scores indicated that 90% of the evaluated phenotypes (n=47) had a granularity score 2 or greater (see Figure \[fig:gran\]). This means that almost all of the evaluated phenotypes were deemed by both reviewers to identify a single or a group of diseases. The most number of phenotypes were assigned an average score of 2.5 (n=31), the next most prevalent score was 2 (n=13). This indicates that the model mostly identified phenotypes that were a group of diseases rather than a single disease. An example of a phenotype that had an average granularity score of 2.5 had different diagnoses codes identifying a fall or accident and different body parts such as shoulder, forearm, limb and hand. One reviewer scored the phenotype as identifying a single disease being ‘limb injury due to accident’, while the other reviewer believed that since multiple body parts were identified the phenotype represented a group of diseases. ![Phenotype granularity scores. Average score across the two clinical expert scores. Score 1=’non disease’, 2=’group of diseases’, 3= ’single disease’.[]{data-label="fig:gran"}](pics/gran_scores.pdf){width="\linewidth"} The phenotype labels were mostly found to be ‘related’ with a score of 2.5 (n=21) or 2 (n=19) (see Figure \[fig:label\]). Only 3 phenotype labels were identified as ‘actionable’ with a score of 3 by both reviewers. The feedback from the reviewers was that the diagnosis code used for the phenotypes was too granular to adequately represent the entire phenotype. ![Phenotype coherence scores. Histogram of average phenotype coherence scores assigned by the two clinical expert. Score 1=‘not related’, 2=‘related’, 3= ‘actionable’.[]{data-label="fig:label"}](pics/label_scores.pdf){width="\linewidth"} #### Phenotype-relatedness quality Of the learned phenotype-pair correlations, 471 (1.5% of all possible phenotype-phenotype pairs) were significant (correlation coefficient above 0.5 in absolute value). Of the 471 significantly correlated phenotype-pairs, 395 where positive correlated (Figure \[fig:all\_positive\_corr\]) and 76 were negatively correlated. We had a clinical expert perform clinical validity of the learned phenotype relationships. ![All significant pairwise-positive correlations visualized[]{data-label="fig:all_positive_corr"}](pics/all_corr.pdf){width="\linewidth"} In the “more common” phenotype set, 82 phenotype pairs were found to have a correlation greater than 0.5 (Figure \[fig:more\_than5\]). These 82 correlations resulted from 61 unique phenotypes, hence on average each phenotype had more than significant correlation with more than one phenotype. Of the 82 reviewed relations 80 (98%) were found clinically valid. One relation rated non clinically valid was the high correlation between a non-disease phenotype for outpatient visits and a non-disease phenotype for inpatient visits. The other non clinically valid relation was between a joint disease phenotype and a phenotype that seemed to be a mix of hepatitis C, liver disease, and obesity. ![Significant pairwise-positive correlations evaluated by clinician for clinical correctness.[]{data-label="fig:more_than5"}](pics/corr_more5_big.pdf){width="\linewidth"} In the “more rare” phenotype set, 21 phenotype pairs were found be have a correlation greater than 0.5 (Figure \[fig:less\_than5\]). These 21 correlations results from 23 unique phenotypes. Of the 21 reviewed relations 12 (57%) were found to clinically valid. Most of phenotype pairs that the clinician deemed as unrelated were not very coherent phenotypes which could be expected from phenotypes that were assigned to less of the 5% of the training set. ![All significant pairwise-positive correlations for ‘rare’ phenotypes (defined as present in less than 5% of the training set).[]{data-label="fig:less_than5"}](pics/corr_less5_big.pdf){width="\linewidth"} Hypothesis 2: focus on HIV phenotypes ------------------------------------- Of the 250 phenotypes, 73 where identified as ’HIV’ according to their automatically generated label. The clinical expert evaluation of the these phenotypes showed that most of the identified phenotypes represented a routine primary care visit of an HIV patient. Three phenotypes were clear representations of HIV phenotype and two other phenotypes representing AIDS, the development of HIV into a disease. The rest of the phenotypes were of HIV comorbidities (psychiatric, cancer, renal, neurological, etc) mixed with HIV related observations. A few phenotypes captured behavioral phenotypes (substance abuse) and 11 phenotypes were deemed as non-coherent. Hypothesis 3: focus on non-HIV phenotypes ----------------------------------------- When categorized into CCS categories according to their ICD label, the learned phenotypes were were found to cover 16 out of the 18 CCS level 1 classifications (Table \[tab:freq\]). The two CCS level 1 categories not captured in the phenotype labels pertained to pediatric conditions. Beyond the most prevalent CCS category related to HIV, ‘Mental Illness’ (which include substance use) and ‘Disease of the circulatory system’ were the most frequent disease groups identified by the model (Figure \[fig:ccs1\]). This finding reflects the high coverage of the learned phenotypes related to the types of conditions characteristics of the input population. ------------------------------------------------ ------------ CCS level 1 Category Number of Category phenotypes Infectious and parasitic diseases 83 Mental illness 32 Circulatory system 26 Neoplasms 23 Respiratory sys. 13 Endocrine; nutritional; and metabolic diseases 12 Digestive sys. 10 Musculoskeletal system and connective tissue 10 Genitourinary system 10 Nervous system and sense organs 8 Symp; signs; and ill-defined conditions 8 Blood and blood-forming organs 6 Injury and poisoning 4 Skin and subcutaneous tissue 3 Complications of pregnancy 1 Resid. codes; unclassified; all E codes 1 Certain cond. originating in perinatal period 0 Congenital anomalies 0 ------------------------------------------------ ------------ : 250 phenotypes by their CCS category[]{data-label="tab:freq"} ![250 learned phenotypes colored by their labels’ corresponding CCS category. Size of the circle indicates proportion of phenotype represented in the training set.[]{data-label="fig:ccs1"}](pics/ccs_bubbles_label.pdf){width="\linewidth"} Hypothesis 4: types of phenotype- relatedness --------------------------------------------- Of the 82 relations evaluated, 63 fit into the same CCS multi-level classification, level 1 category and thus could be inferred using the CCS. However 19 relations were not of the same level 1 category. Out of those 19, 2 were deemed to be unrelated by the clinical expert, 17 relations (21%) were clinically correct and could not be inferred from the CCS and showed more diversity in the relation type learned: the phenotype for severe HIV and one representing the non-disease ICU visits, as well as comorbidity relations like in the pair for ‘end-stage renal disease’ and ‘acute respiratory failure.’ Patient-level summarization --------------------------- After the model learns 250 phenotypes from the patient population in the HIV clinic, the model can be applied to the data found in a single patient record. Running the model inference on patient level data (without re-learning the model parameters) provides a 250 dimensional summarization of the patient record. To summarize the patient record over time, we can run the model inference on the patient data after segmenting the patient data at the desired time granularity. The identified phenotype proportions over time is inputed to a sankey visualization presented in Figure \[fig:summary\]. Each sankey line represents a phenotype identified to be relevant in the patient record. The hight of each sankey link indicates the proportion of the phenotype in that period. In order to be actionable and avoid information overload the summary showcases the patient’s top 5 problems. Top problems are defined at the phenotypes that are found by the model to have the highest probability among the 250 phenotypes learned by the model. The visualization then illustrates how the proportion of the phenotypes increased, decreased, or stayed the same from one period to the next. As a clinical decision support tool, this visualization of the change in phenotypes identified in the patient record could signal to users what health problems the patient possess and how they have changed in salience over time. The described approach for patient summarizing using the proposed phenotyping model benefits from several of the key characteristics of the model. Since the phenotyping model is fully unsupervised the model can easily be utilized for other patient populations by re-training the model on relevant patient data. For instance if patient record summarization was desired for oncology patients, the model can be retrained on oncology patients to learn cancer-specific phenotypes as well relevant co-morbidity phenotypes. The patient summary benefits from the high-throughput nature of the model in that the model learns many phenotypes at the same time and is able to summarize the patient record according to all the phenotypes found to be prevalent in the patient record. Finally the model provides a probabilistic summary of the patient record. The generated patient summary is probabilistic in two senses; 1) each data point has a probability of being associated with the phenotypes; and 2) the phenotype assignments to the patient is also probabilistic which can be interpreted as the salience of the associated phenotype in that time period. To ensure that the model provides a digestible summary of the patient record we analyzed how many phenotypes are required to capture a large majority of the patient record in our training set. Since if we find that the model assigns a large number of phenotypes to each patient record then the proposed summary may still provide too much information be useful at the bedside. In our analysis we found that more than half of patients in our dataset were almost completely described by 1-5 phenotypes (Figure \[fig:ninty\]). The large majority of the remaining patients were described by 6-20 phenotypes. Hence, even though is trained to learn a large number of phenotypes (K=250), each patient record is summarized by only a few phenotypes. This indicates that the model has the potential to reduce many thousands of data points in the record of each patient to a list of a handful of problems and how they have changed over time. ![Number of phenotypes needed to explain 90% of a given patient record. For example, 65% of the patient records in the training set are almost fully explained (90% of data) by 1-5 phenotypes. Where each patient record may be explained by a different 1-5 phenotypes from the 250 phenotypes the model learned from the entire patient cohort. []{data-label="fig:ninty"}](pics/ninty.pdf){width="\linewidth"} Discussion ========== Evaluation results show the model simultaneously identifies 250 phenotypes with good coherence and coverage. Learned phenotype relatedness were found clinically meaningful and diverse, identifying some relations out of scope of the baseline resource. Our experimentation shows that when training the model on a cohort of HIV patients, the model learns multiple HIV phenotypes that can provide good granularity when used for single-patient problem-oriented summarization. The model was also found to identify a wide range of non-HIV phenotypes, yet commonly encountered in HIV patients. The learned phenotype-phenotype correlations learned from the patient cohort could be used to group and organize highly-related phenotypes in the patient-level summary, to provide a clearer overview of the patient’s problems. In many settings but notably urgent care and emergency settings in particular, patient summarization enabled by this model, could provide clinicians a tool for more rapid understanding of the patient comorbidities, leading to better diagnosis, expedited referrals, and potentially a reduction in over-testing. Our future work includes performing an evaluation study of the patient record summarization in assisting clinicians to review patient records more effectively and accurately. This work was supported by NLM T15 LM007079 Data Science Supplement (GL) and NSF award \#1344668 (NE). Online Resources ================ Code and phenotype correlations will be made available at publication time.

ArXiv

--- address: | Valentino Magnani: Dipartimento di Matematica\ Largo Bruno Pontecorvo 5\ $\;\;$ I-56127 Pisa author: - Valentino Magnani title: '[**Towards Differential Calculus in stratified groups**]{}' --- -8truecm Introduction ============ The relationship between the geometry of stratified groups and several branches of Mathematics, as PDEs, Differential Geometry, Complex Analysis, Control Theory and Geometric Measure Theory has known an increasing interest in the last decade. The initial motivation of the present work was the study of an implicit function theorem between stratified groups with respect to the corresponding notion differentiability. Then we realized that to formulate and prove this theorem a series of basic results were needed. These ones naturally lead us to the development of an Intrinsic Differential Calculus for group-valued mappings defined on stratified groups. This project belongs to the general program of developing Analysis in metric spaces, as explained in several books [@AmbTil], [@BBI], [@DavSem], [@HajKos], [@Hein1], [@Semmes]. Other monographs are focused on Lie groups and sub-Riemannian manifolds, [@BelRis], [@CDPT], [@FS], [@Gr2], [@Montgom], [@VSC]. Calculus and Analysis in stratified groups have been developed by several authors with a vast literature, here we mention a list of references for a rough overview that is certainly far from being complete, [@ASCV], [@ArcMor], [@Bal], [@BHIT], [@BSCV], [@CapCow], [@CG], [@CHMY], [@CitMan], [@DGN1], [@DGN3], [@FSSC4], [@FSSC5], [@FSSC6], [@GN], [@Hein], [@KirSer], [@LeoMas], [@LeoRig], [@Mag5], [@Mag8A], [@MagVit], [@Mat05], [@MonRick], [@MonSer], [@Pau2], [@RitRos]. In the present work the notion of differentiability plays a central role and essentially enters every result. This notion in the setting of stratified groups has been introduced by P. Pansu to obtain a rigidity theorem for quasi-isometries of the quaternionic hyperbolic space, where he proved a Rademacher type theorem for Lipschitz mappings with respect to his notion of differentiability, [@Pan2]. Our [*intrinsically $C^1$-smooth Calculus*]{} starts from the following \[PdifContact\] Let $\Omega\subset{\mathbb{G}}$ be an open set and let $f:\Omega{\longrightarrow}{\mathbb{M}}$. Then the following statements are equivalent - $f$ is continuously P-differentiable, - letting $f(x)=\exp\circ F(x)=\exp\sum_{j=1}^\upsilon F_j(x)$, the mappings $F_j:\Omega{\longrightarrow}W_j$, equal to $\pi_j\circ F$, are continuously P-differentiable and the formulae $$\begin{aligned} &&\pi_1\circ df(x)=dF_1(x) \label{eqF_11} \\ &&dF_j(x)(h)=\sum_{n=2}^\upsilon\frac{(-1)^n}{n!}\;\pi_j \left([F(x),dF(x)(h)]_{n-1}\right)\,. \label{eqDF_j1} \label{eqdF}\end{aligned}$$ hold for every $j=2,\ldots,\upsilon$ and every $h\in{\mathbb{G}}$. - $f$ is continuously h-differentiable contact. The integer $\upsilon$ denotes the step of ${\mathbb{M}}$ and the symbol $d$ denotes the P-differential read between Lie algebras. We have denoted by ${\mathbb{G}}$ and ${\mathbb{M}}$ a stratified group and a graded group, respectively. The corresponding Lie algebras ${\mathcal{G}}$ of ${\mathbb{G}}$ and ${\mathcal{M}}$ of ${\mathbb{M}}$ are decomposed into the direct sums of layers $V_i$ and $W_i$, respectively. The mappings $\pi_j$ indicate the canonical projection onto the $j$-th layer of a graded algebra. Notice that h-differentiability, introduced in Definition \[hdiff\], is a much weaker notion than P-differentiability. According to Theorem \[PdifContact\], we have that continuous h-differentiability and contact property given by yield continuous P-differentiability. The main point of Theorem \[PdifContact\] is that it allows for the study of P-differentiable mappings regardless of the notion of P-differentiability and using instead the following system of first order nonlinear PDEs $$\label{hcondF} \left\{\begin{array}{c} X_1F_j=\sum_{n=2}^\upsilon\frac{(-1)^n}{n!}\;\pi_j\left([F,X_1F]_{n-1}\right)\\ X_2F_j=\sum_{n=2}^\upsilon\frac{(-1)^n}{n!}\;\pi_j\left([F,X_2F]_{n-1}\right)\\ \!\!\!\!\!\vdots\qquad\vdots\qquad\vdots\qquad\vdots\qquad\vdots\qquad\vdots\qquad\vdots \\ X_mF_j=\sum_{n=2}^\upsilon\frac{(-1)^n}{n!}\;\pi_j\left([F,X_mF]_{n-1}\right) \end{array}\right.$$ for every $j=2,\ldots,\upsilon$, where $(X_1,\ldots,X_m)$ is a basis of the first layer of ${\mathcal{G}}$. In fact, formulae recursively define the P-differential $dF_j$ for every $j=2,\ldots,\upsilon$, starting from $dF_1$ and in particular yield the system . These $m\times(\upsilon-1)$ equations exactly characterize the contact property of $f$ and can be used to study existence of Lipschitz extensions for mappings between stratified groups, [@Mag10]. Another feature of these equations is that they allow to transmit the regularity of $F_1$ to the remaining components $F_j$ of higher order. The same phenomenon occurs in the “algebraic regularity” proved in Theorem 6.1 of [@CapCow]. The crucial point in the proof of Theorem \[PdifContact\] is the quantitative estimate for the P-difference quotient of a horizontal curve, that is obtained in Theorem \[pansu\]. This estimate plays a key role also in the proof of the mean value inequality and it allows for the extension of Pansu result, [@Pan2], on a.e. P-differentiability of Lipschitz mappings to the case when the target is a graded group, see Corollary \[Pansugrad\]. On the other hand, this key estimate is obtained in a rather “simple” way, once the technical Lemma \[keylemma\] is adopted, and it boils down to integrate recursively the differential equations $$\begin{aligned} \label{cntctcurve} \dot\gamma_i(t)=\sum_{n=2}^\upsilon\frac{(-1)^n}{n!}\; \pi_i\left([\gamma(t),\dot\gamma(t)]_{n-1}\right), \qquad i=2,\ldots,\upsilon,\end{aligned}$$ that must be satisfied by every horizontal curve $\Gamma=\exp\gamma\in{\mathbb{M}}$, where $\gamma_i=\pi_i\circ \gamma$ and $i=2,\ldots,\upsilon$, according to Proposition \[horprop\]. Lemma \[keylemma\] is one of the major technical results of this paper and will appear several times throughout the present work. It corresponds to the linearization of the homogeneous addends $c_n$ appearing in the Baker-Campbell-Hausdorff formula . If one assumes apriori $C^1$ smoothness of $f$, then the characterization of contact mappings through P-differentiabiltiy is already shown in [@War]. The intriguing feature of Theorem \[PdifContact\], along with its technical difficulty, stems from the fact that we do not assume any “extrinsic” regularity of $f$, making this characterization sharp. Continuously h-differentiable mappings are continuous, since their components are locally Lipschitz with respect to the homogeneous distance of the domain. Furthermore, adding the contact property to the continuous h-differentiability, then Theorem \[PdifContact\] and Corollary \[corlip\] yield the local Lipschitz property of $f$ with respect to homogeneous distances. However, let us point out that one can find continuously h-differentiable contact mappings that are not differentiable on sets of positive measure with respect to the Euclidean notion of differentiability, [@Mag3]. Our next result is the mean value inequality for P-differentiable mappings, that represents a crucial tool in the proof of our implicit function theorem. Even in this case, the proof relies on a preliminary study of horizontal curves. In Corollary \[hdifflip\] we characterize Lipschitz curves in graded groups with respect to a homogeneous distance as Lipschitz curves in the Euclidean sense that also satisfy the differential equations . In the case of stratified groups, where the Lie bracket condition is satisfied, this is a well known fact that can be found in Section 11.1 of [@HajKos] for the more general Carnot-Carathéodory spaces. \[unifestim\] Let $\Omega\subset{\mathbb{G}}$ be an open subset and consider a continuously P-differentiable mapping $f:\Omega{\longrightarrow}{\mathbb{M}}$. Let $\Omega_1,\Omega_2\subset{\mathbb{G}}$ be open subsets such that $$\label{inclomega1} \Big\{x\in{\mathbb{G}}\,\Big|\,d(x,\Omega_1)\leq c({\mathbb{G}},d)\, N\,\mbox{\rm diam}(\Omega_1)\Big\}\subset\Omega_2\,,$$ where $\Omega_2$ is compactly contained in $\Omega$. Here $c({\mathbb{G}},d)$ and $N$ are geometric constants only depending on the metric space $({\mathbb{G}},d)$, see both Lemma \[140FS\] and Definition \[defcaDef\]. Then there exist a constant $C$, only depending on ${\mathbb{G}}$, $\max_{x\in\overline{\Omega}_2}\|dF_1(x)\|$ and on the modulus of continuity $\omega_{\overline{\Omega}_2,dF_1}$ of $x{\longrightarrow}df(x)$, defined in , such that $$\label{keyunifest} \frac{\rho\Big(f(x)^{-1}f(y),Df(x)(x^{-1}y)\Big)}{d(x,y)}\leq C\;\;\big[\omega_{\overline{\Omega}_2,df}\big(N\,c({\mathbb{G}},d)\,d(x,y)\big) \big]^{1/\iota^2}\,.$$ for every $x,y\in\overline{\Omega}_1$, with $x\neq y$. The integer $\iota$ denotes the step of ${\mathbb{G}}$. Several difficulties are hidden in this estimate. First, P-differentiable mappings in general are not $C^1$ smooth in the classical sense. Second, working by single components does not suffice, since this would lead to estimates on the Euclidean norm of the difference quotient, that does not fit into the notion of P-differentiability. Third, the family of horizontal curves along which one integrates the P-differential is not manageable. In fact, condition has only the technical motivation to make sure that the special family of piecewise horizontal lines that connect points of $\Omega_1$ and along which we apply are all contained in $\Omega_2$. An immediate corollary of the mean value inequality is a short proof of the inverse mapping theorem in stratified groups. On the other hand, the difficulty in applying Theorem \[unifestim\] to prove an intrinsic implicit function theorem arises from the algebraic problems related to a proper factorization of the group ${\mathbb{G}}$. In the commutative case, that is the classical viewpoint, the implicit mapping is defined by decomposing the space into a product of two linear subspaces, naturally given by the tangent space to the level set and one of its complementary subspaces, as for instance the orthogonal subspace, if a scalar product is fixed. Extending this argument to stratified groups does not work when we seek the second subspace. In fact, the first one is automatically defined as the kernel of the P-differential of the defining mapping, that is also a normal subgroup, but in general we cannot claim a group structure in the “complementary subspace”. For this reason, according to our terminology, we consider the special class of [*h-epimorphisms*]{}, as the “regular surjective h-homomorphisms", that yield the natural splitting of ${\mathbb{G}}$, when they are the P-differential of the defining mapping. Notice that this splitting of the group is a necessary condition in order to state an implicit function theorem. In fact, the existence of the complementary subgroup $H$ in Theorem \[implth\] is a consequence of Proposition \[GlinH\], since $Df(\overline x)$ is an h-epimorphism. To emphasize this special class of surjective h-homomorphisms, in Example \[exahr2\], we provide a surjective h-homomorphism that is not an h-epimorphism. Notice that not all Lie subgroups are considered, but only the special class of homogeneous subgroups, that are closed under dilations. This turns out to be rather natural, in the perspective to interpret a subgroup as blow-up of a suitable intrinsically regular set, according to Theorem \[blwlevelimage\]. Using the terminology of Section \[homsubgr\], two homogeneous subgroups $P$ and $H$ are [*complementary*]{} if they satisfy the conditions ${\mathbb{G}}=PH$ and $P\cap H=\{e\}$. Now we are in the position to establish one of the central results of this paper. \[implth\] Let $\Omega\subset{\mathbb{G}}$ be an open set and let $f:\Omega{\longrightarrow}{\mathbb{M}}$ be continuously P-differentiable, where $\overline{x}\in\Omega$ and the P-differential $Df(\overline{x}):{\mathbb{G}}{\longrightarrow}{\mathbb{M}}$ is an h-epimorphism. Let $N$ be the kernel of $Df(\overline{x})$ and let $H$ be a complementary subgroup. Then there exist $r,s>0$, with $D^N_{\overline{n},r}D^H_{\overline{h},s}\subset\Omega$, along with a unique mapping ${\varphi}:D^N_{\overline{n},r}{\longrightarrow}D^H_{\overline{h},s}$, such that $$\begin{aligned} \label{impleq} f^{-1}{\mbox{\large $($}}f(\overline{x}){\mbox{\large $)$}}\cap D^N_{\overline{n},r}D^H_{\overline{h},s} =\{n {\varphi}(n)\mid n\in D^N_{\overline{n},r}\}.\end{aligned}$$ Furthermore, there exists a constant $\kappa>0$ such that the Lipschitz-type estimate $$\begin{aligned} \label{intrlip} d\big({\varphi}(n),{\varphi}(n')\big)\leq \kappa \;d\big({\varphi}(n')^{-1}n^{-1}n'{\varphi}(n')\big)\end{aligned}$$ holds. In particular, the mapping ${\varphi}$ is $1/\iota$-Hölder continuous with respect to the metrics $d$ in $D^H_{\overline{h},r}$ and $\|\cdot\|$ in $D^N_{\overline{n},s}$. The induced splitting of ${\mathbb{G}}$ is precisely an inner semidirect product, that is in general not direct. This should explain why the classical contraction-mapping principle here seems to be not applicable. More precisely, to use this argument, in the case $\overline{x}$ equals the unit element $e\in{\mathbb{G}}$, one should consider the mapping $F(h)=L(h)^{-1}f(nh)$, where $L$ is the restriction of the P-differential $Df(e)$ to the complementary subgroup $H$. In the classical case, one shows that the mapping $F$ is a contraction for every $n$ sufficiently close to $e$. In stratified groups, using the notion of P-differentiability, the analogous argument does not work, due to noncommutativity. To overcome this point, we show that the mapping $F_n(h)=f(nh)$ is uniformly biLipschitz with respect to $n$ and has constant nonvanishing topological degree as $n$ varies in a compact neighbourhood of the unit element of $N$. This gives existence and uniqueness of the implicit mapping. We also wish to point out how inequality surprisingly fits into the intrinsic notion of Lipschitz mapping given in Definition 3.1 of [@FSSC7], when the ambient space is an Heisenberg group. It is now natural to investigate the natural counterpart of the implicit function theorem, namely, the rank theorem. To translate this theorem in terms of continuously P-differentiable mappings, we first single out the class injective h-homomorphisms that provide the splitting of the target group, according to Proposition \[GlinHmono\]. \[Pembed\] Let $f:\Omega{\longrightarrow}{\mathbb{M}}$ be a continuously P-differentiable mapping, where $\Omega$ is an open subset of ${\mathbb{G}}$. Let $\overline x\in\Omega$ and let $Df(\overline{x}):{\mathbb{G}}{\longrightarrow}{\mathbb{M}}$ be the P-differential. Let us assume that $Df(\overline x)$ is an h-monomorphism of image $H$ and let $N$ be a normal complementary subgroup. Let $p:{\mathbb{M}}{\longrightarrow}\ H$ be the associated canonical projection. Then there exist neighbourhoods $V\subset\Omega$ of $\overline{x}$ and $W\subset H$ of $p(f(\overline{x}))$ along with mappings ${\varphi}:W{\longrightarrow}N$, $J:H{\longrightarrow}{\mathbb{G}}$ and $\Psi:f(V){\longrightarrow}{\mathbb{M}}$ such that we have $$\begin{aligned} \label{imagerep} f(V)=\{h{\varphi}(h)\mid h\in W\}\quad\mbox{and}\quad \Psi\circ f\circ J_{|V'}=I_{|V'},\end{aligned}$$ where $I:H\hookrightarrow{\mathbb{M}}$ is the restriction of the identity mapping $\mbox{Id}_{\mathbb{M}}$ to $H\subset{\mathbb{M}}$, the open subset $V'\subset H$ is equal to $J^{-1}(V)$ and $J=\big(p\circ Df(\overline{x})\big)^{-1}$. Furthermore, setting $F=\exp^{-1}\circ{\varphi}$, then there exists $C>0$ such that for every $h,h'\in W$ we have $$\begin{aligned} \label{lipparam} \|F(h)-F(h')\|\leq C\, d(h,h')\,.\end{aligned}$$ The proof of this result uses completely different tools with respect to the implicit function theorem. Here the key observation is that the projection $p:{\mathbb{M}}{\longrightarrow}H$ is an h-epimorphism, hence it is Lipschitz. Then we apply our inverse mapping theorem. Implicit function theorem and rank theorem are the standard tools to define differentiable manifolds. Analogously, in an obvious way Theorem \[implth\] and Theorem \[Pembed\] define subsets that are expected to possess some intrinsic regularity. These are the $({\mathbb{G}},{\mathbb{M}})$-regular sets, that we distinguish into those contained in ${\mathbb{G}}$, that are suitable level sets and those of ${\mathbb{M}}$, that are suitable images, see Section \[GMregsets\] for more details. By estimate , that immediately follows from , $({\mathbb{G}},{\mathbb{M}})$-regular sets of ${\mathbb{G}}$ can be locally parametrized by $1/\iota$-Hölder mappings. In the case of $({\mathbb{H}}^1,{\mathbb{R}})$-regular sets of ${\mathbb{H}}^1$, hence $\iota=2$, B. Kirchheim and F. Serra Cassano have proved that the embedding Hölder exponent $1/2$ cannot be improved in general. In the same work [@KirSer], the authors provide through a nontrivial construction an example of $({\mathbb{H}}^1,{\mathbb{R}})$-regular set of ${\mathbb{H}}^1$ of Euclidean Hausdorff dimension $5/2$, that is clearly not rectifiable in the Federer sense, [@Fed]. Understanding the intrinsic regularity of these sets is certainly hard and very far from being completely understood. For instance, a fine characterization of $({\mathbb{H}}^n,{\mathbb{R}})$-regular sets as suitable 1-codimensional intrinsic graphs has been recently established in [@ASCV]. As an immediate consequence of both rank theorem and implicit function theorem, we have the following \[GMreggraph\] Every $({\mathbb{G}},{\mathbb{M}})$-regular set is locally an intrinsic graph. Intrinsic graphs and $({\mathbb{G}},{\mathbb{M}})$-regular sets of ${\mathbb{G}}$ first appeared in the works of Franchi, Serapioni and Serra Cassano, [@FSSC3], [@FSSC4] in the case ${\mathbb{M}}={\mathbb{R}}$ and more recently in Heisenberg groups with ${\mathbb{M}}={\mathbb{R}}^k$, [@FSSC6]. In general stratified groups, these sets have been considered in [@Mag5], where the problem of studying their metric and topological properties has been raised. For instance, Corollary \[GMreggraph\] shows that the topological codimension of $({\mathbb{G}},{\mathbb{M}})$-regular sets corresponds to the topological dimension of ${\mathbb{M}}$. A detailed study of $({\mathbb{H}}^n,{\mathbb{R}}^k)$-regular sets, $1\leq k\leq n$, has been accomplished in [@FSSC6], where a corresponding implicit function theorem and an area-type formula have been established. In the same work, following the Rumin complex [@Rum90], the authors lay the foundations for an intrinsic theory of currents, where “rectifiable” sets correspond to both $({\mathbb{H}}^n,{\mathbb{R}}^k)$-regular sets and $({\mathbb{R}}^k,{\mathbb{H}}^n)$-regular sets. In the terminology of [@FSSC6], these two class of sets correspond to [*low codimensional ${\mathbb{H}}$-regular surfaces*]{} and [*low dimensional ${\mathbb{H}}$-regular surfaces*]{}, respectively. As an example of another geometry to be considered, Theorem \[intregh21\] provides the complete list of all possible intrinsically regular sets of the 6-dimensional complexified Heisenberg group ${\mathbb{H}}_2^1$. To obtain this classification we need both algebraic and analytical tools. In fact, we first find all possible factorizations of ${\mathbb{H}}_2^1$ into an inner semidirect product of two complementary subgroups, then we apply our implicit function theorem joined with Theorem \[legalg\]. Low dimensional ${\mathbb{H}}$-regular surfaces of ${\mathbb{H}}^n$ are nontrivial examples of $({\mathbb{G}},{\mathbb{M}})$-regular sets of ${\mathbb{M}}$, where ${\mathbb{G}}={\mathbb{R}}^k$ and ${\mathbb{M}}={\mathbb{H}}^n$. More generally, Theorem \[legalg\] characterizes $({\mathbb{R}}^k,{\mathbb{M}})$-submanifolds as $k$-dimensional $C^1$-Legendrian submanifolds of ${\mathbb{M}}$. This theorem is a straightforward consequence of Theorem \[PdifContact\]. As another consequence of Theorem \[legalg\], Legendrian submanifolds are intrinsic graphs. If we consider general couples of stratified groups, then $({\mathbb{G}},{\mathbb{M}})$-regular sets of ${\mathbb{M}}$ can be thought of as a smooth version of “$N$-rectifiable sets” studied in [@Pau2], where ${\mathbb{G}}=N$. From the metric viewpoint, estimate shows a stronger regularity of image sets with respect to level sets, where the “nonlinear estimate” holds. However, in the direction of intrinsic regularity we have the following \[blwlevelimage\] Under hypotheses of Theorem \[implth\], we consider the set $S=f^{-1}\big(f(\overline{x})\big)$. Then for every $R>0$ we have $$D_R\cap \delta_{1/\lambda}\big((\overline{x})^{-1}S\big) {\longrightarrow}D_R\cap N\quad\mbox{as}\quad\lambda{\rightarrow}0^+$$ with respect to the Hausdorff convergence of sets. In particular, $\mbox{\rm Tan}(S,\overline{x})=N$. Under hypotheses of Theorem \[Pembed\], setting $S=f(V)$, for every $R>0$, we have $$D_R\cap \delta_{1/\lambda}\big((\overline{x})^{-1}S\big){\longrightarrow}D_R\cap H \quad\mbox{as}\quad\lambda{\rightarrow}0^+$$ with respect to the Hausdorff convergence of sets. In particular, $\mbox{\rm Tan}(S,\overline{x})=H$. As a consequence of this theorem, we notice another strong difference between level sets and image sets in general. In fact, all homogeneous tangent cones to a $({\mathbb{G}},{\mathbb{M}})$-regular set of ${\mathbb{M}}$ are clearly h-isomophic to ${\mathbb{M}}$ and in particular have all the same Hausdorff dimension. Furthermore, by the area formula of [@Mag], their Hausdorff dimension coincides with that of ${\mathbb{G}}$ and we have an integral formula for their Hausdorff measure. On the other hand, homogeneous tangent cones to a $({\mathbb{G}},{\mathbb{M}})$-regular set of ${\mathbb{G}}$ are not necessarily h-isomorphic to each other, as Example \[nnhisotan\] shows. By the way, Corollary \[GMGTan\] shows that all of the homogeneous tangent cones to a $({\mathbb{G}},{\mathbb{M}})$-regular set of ${\mathbb{G}}$ have the same Hausdorff dimension and this one equals to $$\mbox{${\mathcal{H}}$-dim$({\mathbb{G}})-{\mathcal{H}}$-dim(${\mathbb{M}}$)}.$$ This suggests that the Hausdorff dimension of a $({\mathbb{G}},{\mathbb{M}})$-regular set of ${\mathbb{G}}$ should coincide with that of its homogeneous tangent cones, but this problem still claims to be investigated. From the previous results, it is clear that the richness of $({\mathbb{G}},{\mathbb{M}})$-regular sets is connected to the richness of h-homomorphisms between ${\mathbb{G}}$ and ${\mathbb{M}}$ along with their factorizing properties. To advertize novel applications of both the rank theorem and the implicit function theorem, in Section \[sectfactor\] we consider some special couples of groups where all injective h-homomorphisms are h-monomorphisms and all surjective h-homomorphisms are h-epimorphisms. We believe that possibly many more “geometries” can be discovered through a deeper algebraic investigation. Finally, we give a concise overview of the paper. In Section \[preldef\], we recall the main definitions that will be used throughout. Section \[techlem\] develops the technical machinery of the paper, that plays a key role in the proofs of our main results. In Section \[estimhcurve\], we establish quantitative estimates on the P-difference quotient of horizontal curves and we apply them to characterize continuously P-differentiable mappings. In Section \[ABSC\] we characterize absolutely continuous curves with respect to a homogeneous distance and through this characterization we prove the Lipschitz property of continuously P-differentiable mappings with values in a graded group. Section \[MVI\] contains the proof of the mean value inequality for continuously P-differentiable mappings. As an immediate application, in Subsection \[invmapthe\] we give a proof of the inverse mapping theorem in stratified groups. Section \[homsubgr\] recalls the notion of homogeneous subgroup, introduces complementary subgroups and gives simple characterizations of both h-epimorphisms and h-monomorphisms. In Section \[hquot\] we show that the quotient of a graded group by a normal homogeneous subgroup is still a graded group and the analogous statement holds for stratified groups. Section \[MProofs\] is devoted to the proof of both the implicit function theorem and the rank theorem. In Section \[GMregsets\] we prove the everywhere existence of the homogeneous tangent cone to $({\mathbb{G}},{\mathbb{M}})$-regular sets and we show by an example that they might not be h-isomorphic to each other in the case of $({\mathbb{G}},{\mathbb{M}})$-regular sets of ${\mathbb{G}}$. In Section \[sectfactor\] we give the notions of h-quotients and of h-embeddings. We introduce the notion of factorizing group as a quotient and factorizing group as a subgroup, providing corresponding examples. In Section \[SecEx\], we characterize all $C^1$ smooth Legendrian submanifolds in graded groups as $({\mathbb{R}}^k,{\mathbb{M}})$-regular sets and we find all intrinsically regular sets of Heisenberg groups and of the complexified Heisenberg group. .2truecm [**Acknowledgements.**]{} I am grateful to Fulvio Ricci for his fruitful comments and pleasant discussions. I thank Yu. L. Sachkov for the interesting discussions we had on Lie groups theory when I was at ISAS of Trieste. I thank Raul Serapioni for his subtle observations on the issue of higher codimensional intrinsically regular level sets. I wish to thank Alessandro Ottazzi for some inspiring discussions on factorizations of stratified groups. Preliminaries and definitions {#preldef} ============================= All Lie groups we consider in this paper are real, connected, simply connected and finite dimensional. A [*graded group*]{} is a Lie group ${\mathbb{G}}$, whose Lie algebra ${\mathcal{G}}$ can be written as the direct sum of subspaces $V_i$, called layers, such that $$\begin{aligned} \label{grading} [V_i,V_j]\subset V_{i+j}\end{aligned}$$ and ${\mathcal{G}}=V_1\oplus\cdots\oplus V_\iota$. The integer $\iota$ is the step of nilpotence of ${\mathbb{G}}$, [@FS]. A graded group ${\mathbb{G}}$ is [*stratified*]{} if its layers satisfy the stronger condition $[V_i,V_j]=V_{i+j}$. The grading of ${\mathcal{G}}$ allows us to introduce a one-parameter group of Lie algebra automorphisms $\delta_r:{\mathcal{G}}{\longrightarrow}{\mathcal{G}}$, defined as $\delta_r(X)=r^i$ if $X\in V_i$, where $r>0$. These mappings are called [*dilations*]{}. Taking into account that the exponential mapping $\exp:{\mathcal{G}}{\longrightarrow}{\mathbb{G}}$ is a diffeomorphism for simply connected nilpotent Lie groups, we can read dilations in the group ${\mathbb{G}}$ through the mapping $\exp$ and mantain the same notation. Recall from Theorem 2.14.3 of [@Vara] that the differential of the exponential mapping is given by the followig formula $$\label{dexp} d\exp\,(X)=\mbox{Id}-\sum_{n=2}^\infty\frac{(-1)^n}{n!}\,\mbox{ad}(X)^{n-1}\,,$$ \[versub\][Let ${\mathbb{H}}^1$ denote the Heisenberg group and let ${\mathfrak{h}}^1={\mbox{span}}\{X,Y,Z\}$ be its Lie algebra with $[X,Y]=Z$. Then the “vertical” subgroup $\exp\big({\mbox{span}}\{X,Z\}\big)$ is a graded group, but it is not stratified.]{} The group operation can be read in the algebra as follows $$\begin{aligned} \label{absBCH} X{\circledcirc}Y=\sum_{j=1}^\iota c_n(X,Y)\end{aligned}$$ where $c_1(X,Y)=X+Y$ and the addends $c_n$ are given by induction through the Baker-Campbell-Hausdorff formula $$\begin{aligned} \label{kBCH} &&(n+1)\,c_{n+1}(X,Y)=\frac{1}{2}\;[X-Y,c_n(X,Y)] \\ &&+\sum_{\substack{p\geq 1\\ 2p\leq n}}K_{2p} \sum_{\substack{k_1,\ldots,k_{2p}>0 \\ k_1+\cdots k_{2p}=n}}[c_{k_1}(X,Y),[\cdots,[c_{k_{2p}}(X,Y),X+Y],],\ldots,], \nonumber\end{aligned}$$ see Lemma 2.15.3 of [@Vara]. Analyzing , one easily notices that $$\begin{aligned} c_n(\lambda X,\lambda Y)=\lambda^n\,c_n(X,Y) \end{aligned}$$ for every $X,Y\in{\mathcal{G}}$ and $\lambda\in{\mathbb{R}}$. These formulae will be important in the next section. The metric structure of a graded group is given by a continuous, left invariant distance $d:{\mathbb{G}}\times{\mathbb{G}}{\longrightarrow}{\mathbb{R}}$ such that $d(\delta_rx,\delta_ry)=r\,d(x,y)$ for every $x,y\in{\mathbb{G}}$ and $r>0$. Every distance satisfying these properties is a [*homogeneous distance*]{}. The Carnot-Carathéodory distance is an important example of homogeneous distance that can be defined in stratified groups, since they satisfy the Lie bracket generating condition, see for instance [@Gr1]. Notice that graded groups may not satisfy this condition. On the other hand, according to Example \[versub\], when a graded group is contained in a stratified group, the restriction of the Carnot-Caratheodory distance to the graded subgorup provides an example of homogeneous distance. In general, it is possible to construct homogeneous distances in every graded group, [@FSSC5], [@RicPC]. We denote by $e$ the unit element and to simplify notation we set $d(x)=d(x,e)$. Notice that left invariance and symmetry of $d$ imply the equality $d(x^{-1})=d(x)$. An open ball of center $x$ and radius $r$ with respect to a homogeneous distance will be denoted by $B_{x,r}$. The corresponding closed ball will be denoted by $D_{x,r}$ h-homomorphisms and notions of differentiability {#homodiff} ------------------------------------------------ [Let ${\mathbb{G}}$ and ${\mathbb{M}}$ be graded groups with dilations $\delta_r^{\mathbb{G}}$ and $\delta_r^{\mathbb{M}}$, respectively. We say that a group homomorphism $L:{\mathbb{G}}{\longrightarrow}{\mathbb{M}}$ such that $L(\delta_r^{\mathbb{G}}x)=\delta_r^{\mathbb{M}}L(x)$ for every $x\in{\mathbb{G}}$ and $r>0$ is a [*homogeneous homomorphism*]{}, in short [*h-homomorphism*]{}. Invertible h-homomorphisms will be called [*h-isomorphisms*]{}. ]{} [Analogous terminology will be used for the corresponding Lie algebra homomorphisms of graded algebras that commute with dilations.]{} \[disthomo\] [Concerning injective and surjective h-homomorphisms, we will use the classical terminology of h-epimorphism and h-monomorphisms to indicate special classes of surjective and injective h-homomorphisms. In fact, a surjective linear mapping of vector spaces is characterized by the existence of a right inverse that is also linear. Analogously, injective linear mappings are characterized by the existence of a linear left inverse mapping. The analogous characterization for either surjective or injective algebra homomorphisms does not work, as we will see in the next example.]{} \[exahr2\][Let $l:{\mathfrak{h}}^1{\longrightarrow}{\mathbb{R}}^2$ be a surjective h-homomorphism. Clearly $l({\mathfrak{v}})={\mathbb{R}}^2$ and $l({\mathfrak{z}})=\{0\}$, where ${\mathfrak{h}}^1={\mathfrak{v}}\oplus{\mathfrak{z}}$. We show that there is no right inverse that is also an h-homomorphism. Assume by contradiction that there exists an h-homomorphism $\tau:{\mathbb{R}}^2{\longrightarrow}{\mathfrak{h}}^1$ that is a right inverse. Then the property $l\circ\tau={\mbox{Id}}_{{\mathbb{R}}^2}$ and the fact that $\tau$ is an h-homomorphism imply that $\tau({\mathbb{R}}^2)$ is a 2-dimensional homogeneous subalgebra of ${\mathfrak{h}}^1$. Clearly $\tau({\mathbb{R}}^2)$ cannot intersect $\ker l={\mathfrak{z}}$, but this conflicts with Example \[nnexcmpl\], where we show that any 2-dimensional homogeneous subalgebra of ${\mathfrak{h}}^1$ contains ${\mathfrak{z}}$. ]{} As a consequence of the previous example, in the category of graded algebras and h-homomorphisms, requiring the existence of a right inverse homomorphism is a stronger condition than surjectivity. This motivates the following \[hepihmono\][We say that an h-homomorphism is an [*h-epimorphism*]{} if it has a right inverse that is also an h-homomorphism. We say that an h-homomorphism is an [*h-monomorphism*]{} if it has a left inverse that is also an h-homomorphism.]{} In Subsection \[hepimono\], we will see how either h-epimorphisms or h-monomorphisms can be characterized by their property of factorizing either the domain or the codomain. [Let ${\mathbb{G}}$ and ${\mathbb{M}}$ be graded groups with homogeneous distances $d$ and $\rho$, respectively. Let $\Omega$ be an open subset of ${\mathbb{G}}$ and consider $f:\Omega{\longrightarrow}{\mathbb{M}}$. We say that $f$ is [*P-differentiable*]{} at $x\in\Omega$ if there exists an h-homomorphism $L:{\mathbb{G}}{\longrightarrow}{\mathbb{M}}$ such that $$\frac{\rho\left(f(x)^{-1}f(xh),L(h)\right)}{d(h)}{\longrightarrow}0 \quad\mbox{as}\quad h{\rightarrow}e\,.$$ The h-homomorphism $L$ satisfying this limit is unique and it is called [*P-differential*]{} of $f$ at $x$. We denote $L$ by $Df(x)$. When we read the P-differential between Lie algebras, we will denote it by $df(x)$. ]{} \[hdiff\][Let $\Omega\subset{\mathbb{G}}$ be an open set and let $M$ be a smooth manifold. We consider a mapping $f:\Omega{\longrightarrow}M$ and $x\in\Omega$. We say that $f$ is [*horizontally differentiable*]{} at $x$, in short [*h-differentiable*]{} at $x$, if there exists a neighbourhood $U$ of the origin in the first layer $V_1$ such that the restriction $$U\ni X{\longrightarrow}f(x\exp X)\in M$$ is differentiable at the origin. The differential $L:H_x{\mathbb{G}}{\longrightarrow}T_{f(x)}M$ will be denoted by $d_Hf(x)$. We say that $f$ is [*continuously*]{} h-differentiable in the case $x{\longrightarrow}d_Hf(x)$ is a continuous mapping. If $M$ is a graded group ${\mathbb{M}}$, then we also use the notation $D_Hf(x):\exp V_1{\longrightarrow}{\mathbb{M}}$ to denote the h-differential between the corresponding groups. ]{} \[Pdifhdif\][We notice that P-differentiability of ${\mathbb{R}}^k$-valued mappings on an open subset $\Omega\subset{\mathbb{G}}$ implies h-differentiability. It suffices to restrict P-differentiability to horizontal directions and to observe that $d(\exp X)/\|X\|$ is bounded away from zero and from above with constants independent from $X$ as it varies in $V_1$. This yields $$\begin{aligned} \label{dhfP} D_Hf(x)(\exp X)=Df(x)(\exp X)\qquad \mbox{for every}\qquad X\in V_1.\end{aligned}$$ Then h-differentiability in general is a weaker notion. On the other hand, the little regularity of h-differentiable mappings suffices to introduce contact mappings.]{} [Let $f:\Omega{\longrightarrow}{\mathbb{M}}$ be an h-differentiable mapping. We say that $f$ is a [*contact mapping*]{} if the inclusion $d_Hf(x)(H_x{\mathbb{G}})\subset H_{f(x)}{\mathbb{M}}$ holds for every $x\in\Omega$. ]{} [Notice that ${\mathbb{R}}^k$-valued continuously h-differentiable mappings are automatically contact. This family of mappings have been already considered in [@FSSC4] as mappings of ${\bf C}^1_{\mathbb{G}}$.]{} \[chain\] Let $f:\Omega{\longrightarrow}{\mathbb{U}}$ be P-differentiable at $x\in\Omega$ and let $g:\Upsilon{\longrightarrow}{\mathbb{M}}$ be P-differentiable at $f(x)\in\Upsilon$, where $\Omega\subset{\mathbb{G}}$ and $\Upsilon\subset{\mathbb{U}}$ are open subsets and $f(\Omega)\subset\Upsilon$. Then $g\circ f:\Omega{\longrightarrow}{\mathbb{M}}$ is P-differentiable at $x$ and $D(g\circ f)(x)=Dg\big(f(x)\big)\circ Df(x)$. The proof of this result is straightforward, see for instance Proposition 3.2.5 of [@MagPhD]. Technical lemmata {#techlem} ================= Throughout this section, we denote by ${\mathbb{G}}$ a graded group, equipped with a homogeneous distance $d$. Its Lie algebra ${\mathcal{G}}$ of layers $V_i$ has step $\iota.$ On ${\mathcal{G}}$, seen as a finite dimensional real vector space we fix a norm $\|\cdot\|$. Bilinearity of brackets yields a constant $\beta>0$, such that for every $X,Y\in{\mathcal{G}}$ we have $$\label{lcst} \|[X,Y]\|\leq\beta\, \|X\|\;\|Y\|.$$ Let $\nu>0$ and let $n=2,\ldots,\iota$. Then there exists a constant $\alpha_n(\nu)$ only depending on $n$ and $\nu$ such that $$\begin{aligned} \label{bilestim} \|c_n(X,Y)\|\leq \alpha_n(\nu)\;\|[X,Y]\|\end{aligned}$$ whenever $\|X\|,\|Y\|\leq\nu$. [Proof.]{} Our statement is trivial for $n=2$, being $c_2(X,Y)=[X,Y]/2$. Assume that it is true for every $j=2,\ldots,n$, with $n\geq2$. It suffices to observe that $[c_{k_{2p}}(X,Y),X+Y]\neq0$ in implies $k_{2p}>1$, then inductive hypothesis yields $$\|c_{k_{2p}}(X,Y)\|\leq \alpha_{k_{2p}}(\nu)\; \|[X,Y]\|\,.$$ Using this estimate in and observing that $\|c_{k_i}(X,Y)\|\leq2\nu$, whenever $k_i=1$, the thesis follows. $\Box$ Let $c_n(X,Y)$ be as in . Then for each $n=2,\ldots,\iota$ there exists a set of real numbers $\big\{e_{n,\alpha}\mid \alpha\in \{1,2\}^{n-1}\big\}$ only depending on ${\mathcal{G}}$ such that for every $A_1,A_2\in{\mathcal{G}}$ we have $$\begin{aligned} \label{mlinc} c_n(A_1,A_2)=\sum_{\alpha\in\{1,2\}^{n-1}}e_{n,\alpha}\; L_n(A_\alpha,A_1+A_2)\,,\end{aligned}$$ where $A_\alpha=(A_{\alpha_1},\ldots,A_{\alpha_{n-1}})$, $L_1=\mbox{Id}_{\mathcal{G}}$ and for $n\geq2$ the $n$-linear mapping $L_n:{\mathcal{G}}^n{\longrightarrow}{\mathcal{G}}$ is defined by $$\label{Ln} L_n(X_1,X_2,\ldots,X_n)=[X_1,[X_2,[\ldots,[X_{n-1},X_n]]\ldots]\,.$$ [Proof.]{} Let $L_1:{\mathcal{G}}{\longrightarrow}{\mathcal{G}}$ be the identity mapping and let $k_1,\ldots,k_p$ be positive integers with $n=k_1+k_2\cdots+k_p$ where $p\in{\mathbb{N}}$. Iterating Jacoby identity of the Lie product, it is not difficult to check that $$\begin{aligned} \label{nnassp} &&[L_{k_1}(X^1_1,\ldots,X^1_{k_1}),[L_{k_2},(X^2_1,\ldots,X^2_{k_2}),[\cdots,\\ &&[L_{k_{p-1}}(X^{p-1}_1,\ldots,X^{p-1}_{k_{p-1}}), L_{k_p},(X^p_1,\ldots,X^p_{k_p})]]\ldots] \nonumber \\ &&=\sum_{\sigma_1\in S_{k_1},\ldots,\sigma_{p-1}\in S_{k_{p-1}}} s_{\sigma_1,\ldots,\sigma_{p-1}} \;\;L_n\big(X_{\sigma_1}^1,X^2_{\sigma_2},\ldots,X^{p-1}_{\sigma_1}, X^p\big)\,,\nonumber\end{aligned}$$ where $S_j$ is the set of all permutations on $j$ elements, $s_{\sigma_1,\ldots,\sigma_{p-1}}\in\{-1,0,1\}$, $$X^p=(X^p_1,\ldots,X^p_{k_p})\quad\mbox{ and}\quad X^j_{\sigma_j}=(X^j_{\sigma_j(1)},X^j_{\sigma_j(2)},\ldots, X^j_{\sigma_j(k_j)})\,.$$ Our statement can be proved by induction. It is clearly true for $n=2$ taking $e_1=1/2$ and $e_2=0$, due to the formula $c_2(A_1,A_2)=[A_1,A_2]/2$. Let us assume that holds for all $c_j(A_1,A_2)$ with $j\leq n$. By recurrence equation , we have $$\begin{aligned} &&(n+1)\,c_{n+1}(A_1,A_2)=\frac{1}{2}\sum_{\alpha\in\{1,2\}^{n-1}} e_{n,\alpha}\;\left[A_1-A_2, L_n(A_\alpha,A_1+A_2)\right] \\ &&+\sum_{\substack{p\geq 1\\ 2p\leq n}}K_{2p} \sum_{\substack{k_1,\ldots,k_{2p}>0 \\ k_1+\cdots k_{2p}=n}} \sum_{\substack{\;\alpha_i\in\{1,2\}^{k_i-1}\\i=1,\ldots,2p}} e_{k_1,\alpha_1}\;e_{k_2,\alpha_2}\;\cdots\; e_{k_{2p},\alpha_{2p}}\nonumber \\ &&[L_{k_1}(A_{\alpha_1},A_1+A_2),[L_{k_2}(A_{\alpha_2},A_1+A_2,), [\cdots,[L_{k_{2p}}(A_{\alpha_{2p}},A_1+A_2),A_1+A_2]]\ldots] \nonumber\end{aligned}$$ Then applying we have proved that $c_{n+1}(A_1,A_2)$ can be represented as a linear combination of terms $L_{n+1}(A_\alpha,A_1+A_2)$ where $\alpha\in\{1,2\}^n$. This concludes our proof. $\Box$ .2truecm Following the notation of [@Pan2], Section 4.5, in the next definition we introduce the iterated Lie bracket. [Let $X,Y\in{\mathcal{G}}$. The $k$-th bracket is defined by $$\begin{aligned} [X,Y]_k=\underbrace{[X,[X,[\cdots,[X}_{\mbox{\tiny $k$ times}},Y],],\ldots,] \quad\mbox{and}\quad[X,Y]_0=Y\,.\end{aligned}$$ ]{} \[keylemma\] Let $c_n(X,Y)$ be as in , where $n=2,\ldots,\iota$. Then we have $$\begin{aligned} \label{cnrest} c_n(X,Y)=\frac{(-1)^{n-1}}{n!}\; \left[\frac{Y-X}{2},X+Y\right]_{n-1}+R_n(X,Y)\end{aligned}$$ and for every $\nu>0$ there exists a positive nondecreasing function $C(n,\cdot)$ such that $$\begin{aligned} \label{estimsmall} \|R_n(X,Y)\|\leq C(n,\nu)\, \|X+Y\|^3\end{aligned}$$ whenever $\|X\|,\|Y\|\leq\nu$. [Proof.]{} Our statement is trivial for $n=2$, being $R_2(X,Y)$ vanishing for any $X,Y\in{\mathcal{G}}$. Let us consider $n\geq3$. We apply formula to , getting $$\begin{aligned} \label{cnrestpar} c_n(X,Y)=-\frac{1}{n}\;\left[\frac{Y-X}{2},c_{n-1}(X,Y)\right]+E_n(X,Y)\end{aligned}$$ where for every $A_1,A_2\in{\mathcal{G}}$ we have $$\begin{aligned} &&E_n(A_1,A_2)=\sum_{\substack{p\geq 1\\ 2p\leq n-1}}K_{2p} \sum_{\substack{k_1,\ldots,k_{2p}>0 \\k_1+\cdots k_{2p}=n-1}} \sum_{\substack{\;\alpha_i\in\{1,2\}^{k_i-1}\\i=1,\ldots,2p}} e_{k_1,\alpha_1}\;e_{k_2,\alpha_2}\;\cdots\; e_{k_{2p},\alpha_{2p}}\\ &&[L_{k_1}(A_{\alpha_1},A_1+A_2),[L_{k_2}(A_{\alpha_2},A_1+A_2,), [\cdots,[L_{k_{2p}}(A_{\alpha_{2p}},A_1+A_2),A_1+A_2]]\ldots]\nonumber\,.\end{aligned}$$ As a consequence, there exist constants $\widetilde C_p>0$ such that $$\begin{aligned} \label{enest} &&\|E_n(X,Y)\|\leq\sum_{1\leq p\leq(n-1)/2} \widetilde C_p \;\; \nu^{n-2p-1}\;\|X+Y\|^{2p+1}\\ &&\leq\sum_{1\leq p\leq(n-1)/2} 4^p\;\widetilde C_p \;\; \nu^{n-3}\;\|X+Y\|^3\,.\nonumber\end{aligned}$$ Iterating , we get $$\begin{aligned} && c_n(X,Y)=\frac{(-1)^2}{n(n-1)}\left[\frac{Y-X}{2},c_{n-2}(X,Y)\right]_2 +E_n(X,Y) \\ && -\frac{1}{n}\left[\frac{Y-X}{2},E_{n-1}(X,Y)\right] =\frac{(-1)^{n-1}}{n!}\left[\frac{Y-X}{2},c_1(X,Y)\right]_{n-1} +E_n(X,Y)\nonumber \\ &&+\sum_{k=1}^{n-2}\frac{(-1)^k}{n(n-1)\cdots(n-k+1)} \left[\frac{Y-X}{2},E_{n-k}(X,Y)\right]_k\nonumber\end{aligned}$$ therefore concludes our proof. $\Box$ [As an immediate consequence of Lemma \[keylemma\], we achieve the estimate $$\begin{aligned} \label{leftinveucl} \|(-\xi){\circledcirc}\eta\|\leq C(\nu) \|\xi-\eta\|\end{aligned}$$ for every $\xi,\eta\in{\mathcal{G}}$ satisfying $\|\xi\|,\|\eta\|\leq\nu$. To see this, it suffices to apply and to the Baker-Campbell-Hausdorff formula $(-\xi){\circledcirc}\eta=-\xi+\eta+\sum_{n=2}^\iota c_n(-\xi,\eta)$. ]{} As an immediate extension of inequality (+) at p.13 of [@Pan2], see also [@Pan1], we have the following lemma. \[estimi\] Let $\pi^i:{\mathcal{G}}{\longrightarrow}V_i\oplus\cdots\oplus V_\iota$ be the natural projection and let $U$ be a bounded open neighbourhood of the unit element $e\in{\mathbb{G}}$. Then there exists a constant $K_U>0$, depending on $U$, such that $$\begin{aligned} \label{estimil} \|\pi^i\left(\exp^{-1}(x)\right)\|\leq K_U\;d(x)^i \end{aligned}$$ holds for every $x\in U$. [Proof.]{} We first define ${\mathcal{S}}=\{v\in{\mathcal{G}}\mid d(\exp v)=1\}$. Let us fix $v\in{\mathcal{S}}$ and $s>0$ satisfying the condition $s\leq M$, with $M=\max_{x\in U} d(x)$. We have $$\|\pi^i(\delta_sv)\|=s^i\;\Big\|\sum_{j=i}^\iota s^{j-i}\,\pi_j(v)\Big\| \leq C s^i\,,$$ where $C=\sum_{j=i}^\iota M^{j-i}\,\max_{u\in{\mathcal{S}}}\|\pi_j(u)\|$. This concludes the proof. $\Box$ [From Proposition 1.5 of [@FS], it is easy to check that there exists a constant $\kappa(\nu)$ such that whenever $\|\xi\|\leq\nu$ there holds $$\begin{aligned} \label{rhonormiota} d(\exp\xi)\leq\kappa(\nu)\,\|\xi\|^{1/\iota}\,.\end{aligned}$$ ]{} [ As a byproduct of and we obtain the well known estimate $$\begin{aligned} \label{rhoestnorm} d\big(\exp\xi,\exp\eta\big)\leq C(\nu)\; \|\xi-\eta\|^{1/\iota}\end{aligned}$$ for every $\xi,\eta\in{\mathcal{G}}$ satisfying $\|\xi\|,\|\eta\|\leq\nu$, see [@RotStein] for its proof in the more general setting of vector fields satisfying the Hörmander condition. ]{} [Notice that for every homogeneous distance $d$, there exists a constant $C_{d,\|\cdot\|}$ such that $$C_{d,\|\cdot\|}^{-1}\;\sum_{j=1}^\iota\|\pi_j(\xi)\|^{1/j}\leq d(\exp\xi)\leq C_{d,\|\cdot\|}\;\sum_{j=1}^\iota\|\pi_j(\xi)\|^{1/j}\,.$$ In particular, we have $$\label{lip1} d(\exp\xi,\exp\eta)\geq C_{d,\|\cdot\|}^{-1}\;\|\pi_1(\xi-\eta)\|\,.$$ for every $\xi,\eta\in{\mathcal{G}}$. ]{} \[lemconjest\] Let $x,y\in{\mathbb{G}}$ and let $\nu>0$ be such that $d(x),d(y)\leq\nu$. Then there exists a constant $C(\nu)$ only depending on ${\mathbb{G}}$ and on $\nu$ such that $$\begin{aligned} \label{conjest1} d(y^{-1}xy)\leq C(\nu)\;\|\exp^{-1}(x)\|^{1/\iota}\,.\end{aligned}$$ [Proof.]{} Let us fix $x=\exp\xi$ and $y=\exp h$. By Lemma \[estimi\] we find a constant $k_\nu>0$ such that $\|\xi\|,\|h\|\leq k_\nu$. The Baker-Campbell-Hausdorff fomuls yields $$\begin{aligned} \label{bchassoc} (-h){\circledcirc}\big(\xi{\circledcirc}h\big) =\xi+\sum_{n=2}^\iota c_n(\xi,h)+\sum_{n=2}^\iota c_n(-h,\xi{\circledcirc}h).\end{aligned}$$ The same formula along with , also gives the estimate $$\|\xi{\circledcirc}h\|\leq k_\nu\,\Big(2+\sum_{n=2}^\iota \alpha_n(k_\nu)\,\beta\,k_\nu\Big) =A(\nu).$$ The bilinear estimate also yields $$\begin{aligned} \label{iterest} \|c_n(-h,\xi{\circledcirc}h)\|\leq\alpha_n\big(A(\nu)\big)\, \|[h,\xi{\circledcirc}h]\|.\end{aligned}$$ Observing that $$[h,\xi{\circledcirc}h]=\big[h,\xi+\sum_{n=2}^\iota c_n(\xi,h)\big]$$ we achieve the estimate $$\|[h,\xi{\circledcirc}h]\|\leq\beta\,\|h\|\,\|\xi\|\, \big(1+\sum_{n=2}^\iota\alpha_n(k_\nu)\,\|h\|\big).$$ Joining this estimate with we achieve $$\begin{aligned} \label{iterest1} \|c_n(-h,\xi{\circledcirc}h)\|\leq \beta\,\alpha_n\big(A(\nu)\big)\,\Big(1+\sum_{n=2}^\iota\nu\,\alpha_n(\nu)\Big)\,\|h\|\,\|\xi\|\end{aligned}$$ Thus, formula , estimates and give a constant $B(\nu)>0$ such that $$\begin{aligned} \label{fingope} \|(-h){\circledcirc}\xi{\circledcirc}h\|\leq \|\xi\|\;\big(1+B(\nu)\|h\|\big).\end{aligned}$$ The previous inequality yields $$\|(-h){\circledcirc}\xi{\circledcirc}h\|\leq k_\nu\,\big(1+k_\nu\;B(\nu)\big),$$ then gives a constant $B_1(\nu)>0$ such that $$d(y^{-1}xy)\leq B_1(\nu)\;\|(-h){\circledcirc}\xi{\circledcirc}h\|^{1/\iota}$$ Thus, applying we achieve a constant $B_2(\nu)>0$ such that $$d(y^{-1}xy)\leq B_2(\nu)\;\|\xi\|^{1/\iota}\,.$$ This finishes the proof. $\Box$ [The previous lemma also provides another variant of its estimate. In fact, the condition $d(x)\leq\nu$ implies that $\|\exp^{-1}(x)\|$ is bounded by $d(x)$ up to a factor only depending on $\nu$, due to Lemma \[estimi\]. As a result, the assumption $d(x),d(y)\leq\nu$ gives a constant $C(\nu)$ such that the following estimate holds $$\begin{aligned} \label{conjest} d(y^{-1}xy)\leq C(\nu)\;d(x)^{1/\iota}.\end{aligned}$$ ]{} \[lestimprod\] Let $N$ be a positive integer and let $A_j,B_j\in{\mathbb{G}}$ with $j=1,\ldots,N$. Let $\nu>0$ be such that $d(B_jB_{j+1}\cdots B_N)\leq\nu$ and $d(A_j,B_j)\leq\nu$ for every $j=1,\ldots,N$. Then there exists $K_\nu>0$ such that $$\begin{aligned} \label{estimprod} d\big(A_1A_2\cdots A_N,B_1B_2\cdots B_N\big)\leq K_\nu\; \sum_{j=1}^Nd(A_j,B_j)^{1/\iota}.\end{aligned}$$ [Proof.]{} We define $$\hat{B}_j=B_jB_{j+1}\cdots B_N,\qquad \hat{A}_j=A_jA_{j+1}\cdots A_N$$ and use left invariance of $d$ to obtain $$d\big(\hat{A}_1,\hat{B}_1\big)\leq d(A_N,B_N) +\sum_{j=1}^{N-1}d\big(\hat{B}_{j+1}A_j^{-1}B_j\hat{B}_{j+1}\big)\,.$$ By our hypothesis and we get $$\begin{aligned} &&d\big(\hat{A}_1,\hat{B}_1\big)\leq d(A_N,B_N) +C(\nu)\;\sum_{j=1}^{N-1}d\big(A_j,B_j\big)^{1/\iota} \leq d(A_N,B_N)^{1/\iota}\nu^{1-1/\iota}\\ &&+C(\nu)\;\sum_{j=1}^{N-1}d\big(A_j,B_j\big)^{1/\iota} \leq\max\{C(\nu),\nu^{1-1/\iota}\}\sum_{j=1}^Nd(A_j,B_j)^{1/\iota}.\end{aligned}$$ This shows the validity of . $\Box$ \[bchdifflem\] Let $X,Y,D_1,D_2\in{\mathcal{G}}$ and let $\nu\geq\max\{\|X\|,\|Y\|,\|D_1\|,\|D_2\|\}$. Then there exists a constant $\kappa_n>0$ such that $$\begin{aligned} \label{bchdiffprod} \|c_n(X+D_1,Y+D_2)-c_n(X,Y)\|\leq \gamma_n\; \nu^{n-1}\; \max\{\|D_1\|,\|D_2\|\}\end{aligned}$$ for every $n=2,\ldots,\iota$. [Proof.]{} We set $X=A_1^1$, $D_1=A_1^2$, $Y=A_2^1$ and $D_2=A_2^2$ and apply . Taking into account that $L_n$ is a multilinear mapping defined in , we obtain $$\begin{aligned} &&c_n(A_1^1+A_1^2,A_2^1+A_2^2)\\ &&=\sum_{\alpha\in\{1,2\}^{n-1}}e_{n,\alpha}\; L_n(A^1_{\alpha_1}+A^2_{\alpha_1},\ldots,A^1_{\alpha_{n-1}}+A^2_{\alpha_{n-1}},A_1^1+A_1^2+A_2^1+A_2^2) \\ &&=\sum_{\alpha\in\{1,2\}^{n-1}}e_{n,\alpha} \sum_{\gamma\in\{1,2\}^{n-1}} L_n(A^{\gamma_1}_{\alpha_1},\ldots,A^{\gamma_{n-1}}_{\alpha_{n-1}},A_1^1+A_1^2+A_2^1+A_2^2)\\ &&=\sum_{\alpha\in\{1,2\}^{n-1}}e_{n,\alpha} \sum_{\gamma\in\{1,2\}^{n-1}} L_n(A^{\gamma_1}_{\alpha_1},\ldots,A^{\gamma_{n-1}} _{\alpha_{n-1}},A_1^1+A_2^1)\\ &&+\sum_{i=1,2}\sum_{\alpha\in\{1,2\}^{n-1}}e_{n,\alpha} \sum_{\gamma\in\{1,2\}^{n-1}} L_n(A^{\gamma_1}_{\alpha_1},\ldots,A^{\gamma_{n-1}}_{\alpha_{n-1}},A_i^2)\\ &&=c_n(A_1^1,A_2^1)+\sum_{\alpha\in\{1,2\}^{n-1}}e_{n,\alpha} \sum_{\substack{\gamma\in\{1,2\}^{n-1}\\\gamma\neq(1,1,\ldots,1)}} L_n(A^{\gamma_1}_{\alpha_1},\ldots,A^{\gamma_{n-1}} _{\alpha_{n-1}},A_1^1+A_2^1)\\ &&+\sum_{i=1,2}\sum_{\alpha\in\{1,2\}^{n-1}}e_{n,\alpha} \sum_{\gamma\in\{1,2\}^{n-1}} L_n(A^{\gamma_1}_{\alpha_1},\ldots,A^{\gamma_{n-1}}_{\alpha_{n-1}},A_i^2)\end{aligned}$$ where the equality in the line before last we have applied again formula . Taking into account the multilinear estimate $$\|L_n(X_1,X_2,\ldots,X_n)\|\leq\beta^n\;\prod_{j=1}^n\|X_j\|,$$ where $\beta$ is given in . As a consequence, a short calculation yields $$\|c_n(A_1^1+A_1^2,A_2^1+A_2^2)-c_n(A_1^1,A_2^1)\|\leq 2^{n+1}\,\beta^n\,\nu^{n-1}\,\max\{\|A_1^2\|,\|A_2^2\|\}\, \sum_{\alpha\in\{1,2\}^{n-1}}|e_{n,\alpha}|\,,$$ leading us to the conclusion. $\Box$ P-differentiability of curves in graded groups {#estimhcurve} ============================================== In this section we study differentiability properties of curves $\Gamma:[a,b]{\longrightarrow}{\mathbb{M}}$, where ${\mathbb{M}}$ is a graded group. We denote by $\gamma$ the corresponding curve $\exp^{-1}\circ\Gamma$ with values in the Lie algebra ${\mathcal{M}}$ of ${\mathbb{M}}$. The components of $\gamma$ taking values in the layers $W_i$ are denoted by $\gamma_i$, hence $\gamma=\sum_{i=1}^\upsilon\gamma_i$. We have $\gamma_i=\pi_i\circ \gamma$, where $\pi_i:{\mathcal{M}}{\longrightarrow}W_i$ is the canonical projection on layers of degree $i$ and ${\mathcal{M}}=W_1\oplus\cdots\oplus W_\upsilon$. We denote by $\|\cdot\|$ a fixed norm on ${\mathcal{M}}$. The [*horizontal subspace*]{} of ${\mathbb{M}}$ at $x\in{\mathbb{M}}$ is defined as follows $$H_x{\mathbb{M}}=\{U(x)\mid U\in W_1\}.$$ [We say that a curve $\Gamma:[a,b]{\longrightarrow}{\mathbb{M}}$ is [*horizontal*]{} if $\gamma:[a,b]{\longrightarrow}{\mathcal{M}}$ is absolutely continuous and for a.e. differentiability point $t\in[a,b]$ the inclusion $\dot\gamma(t)\in H_{\gamma(t)}{\mathbb{M}}$ holds. ]{} Identifying any $T_x{\mathbb{M}}$ with ${\mathcal{M}}$ and applying formula for the differential of the exponential mapping, we have $$\label{dexpgamma} \dot\Gamma(t)= \dot\gamma(t)-\sum_{n=2}^\upsilon\frac{(-1)^n}{n!}\,\mbox{ad}(\gamma(t))^{n-1}(\dot\gamma(t))= \dot\gamma(t)-\sum_{n=2}^\upsilon\frac{(-1)^n}{n!}\, [\gamma(t),\dot\gamma(t)]_{n-1}\,.$$ Then $\dot\Gamma(t)\in W_1$ if and only if $$\pi_i\Big(\dot\gamma(t)-\sum_{n=2}^\upsilon\frac{(-1)^n}{n!}\, [\gamma(t),\dot\gamma(t)]_{n-1}\Big)=0$$ for all $i\geq2$. This immediately proves the following \[horprop\] Let $\gamma:[a,b]{\longrightarrow}{\mathcal{M}}$ be an absolutely continuous curve. Then $\Gamma$ is a horizontal curve if and only if the differential equation $$\begin{aligned} \label{eqhdiffcurve} \dot\gamma_i(t)=\sum_{n=2}^\upsilon\frac{(-1)^n}{n!}\; \pi_i\left([\gamma(t),\dot\gamma(t)]_{n-1}\right) \end{aligned}$$ a.e. holds for each $i=2,\ldots,\upsilon$. \[mcw\][Let $\gamma:[a,b]{\longrightarrow}{\mathcal{M}}$ be a locally summable curve and let $\lambda\neq0$. The [*sup-average*]{} of $\gamma$ at $t$ is defined as follows $$\begin{aligned} \label{cMt} {\mathcal{A}}_t^\lambda(\gamma)=\left\{\begin{array}{ll} \displaystyle\sup_{0\leq\tau\leq \lambda} \;\;{\hbox{\vrule height3.5pt depth-2.8pt width4pt}\mkern-14mu\int\nolimits}_t^{t+\tau}\|\gamma(l)\|\,dl & \mbox{if $\;\lambda>0$} \\ \displaystyle\sup_{\lambda\leq\tau\leq 0} \;\;{\hbox{\vrule height3.5pt depth-2.8pt width4pt}\mkern-14mu\int\nolimits}_{t+\tau}^t\|\gamma(l)\|\,dl & \mbox{if $\;\lambda<0$} \end{array} \right.\,.\end{aligned}$$ ]{} Notice that for some $t$ the sup-average ${\mathcal{A}}_t^\lambda(\gamma)$ takes values in $[0,+\infty]$, as $\gamma$ is not necessarily bounded in a neighbourhood of $t$. \[theoestim\] There exist strictly increasing functions $\Upsilon_i:[0,+\infty[{\longrightarrow}[0,+\infty[$, with $i=2,\ldots,\upsilon$, which only depend on ${\mathcal{M}}$, are infinitesimal at zero and satisfy the following properties. For any horizontal curve $\Gamma:[-\alpha,\alpha]{\longrightarrow}{\mathbb{M}}$ such that $\Gamma(0)=e$, the estimates $$\begin{aligned} \label{estimhor} \left|\int_0^\lambda\|\dot\gamma_i(t)\|\,dt\right|\leq \Upsilon_i(L)\, {\mathcal{A}}_0^\lambda(\dot\gamma_1-X)\,|\lambda|^i\end{aligned}$$ hold, where $L=\max \left\{{\mathcal{A}}_0^{-\alpha}(\dot\gamma_1),{\mathcal{A}}_0^\alpha(\dot\gamma_1)\right\}$, $X\in{\mathcal{M}}$, $\|X\|\leq L$ and the function ${\mathcal{A}}_0^\lambda(\cdot)$ has been introduced in Definition \[mcw\]. [Proof.]{} First of all, we assume that ${\mathcal{A}}_0^\lambda(\dot\gamma_1)<+\infty$, otherwise the proof becomes trivial. Due to Proposition \[horprop\], the differential equations hold. Applying for $i=2$, we achieve $$\label{gmdot2} \dot\gamma_2(t)=\frac{t}{2}\;\Big[\frac{\gamma_1(t)}{t},\dot\gamma_1(t)\Big]= \frac{t}{2} \left(\left[{\hbox{\vrule height3.5pt depth-2.8pt width4pt}\mkern-14mu\int\nolimits}_0^t\dot\gamma_1(\tau)\,d\tau,X\right] +\left[{\hbox{\vrule height3.5pt depth-2.8pt width4pt}\mkern-14mu\int\nolimits}_0^t\dot\gamma_1(\tau)\,d\tau, \dot\gamma_1(t)-X\right]\right)\,.$$ For every $0<s<\alpha$, it follows that $$\begin{aligned} &&\!\!\!\int_0^s\|\dot\gamma_2(t)\|\,dt \leq\frac{s}{2}\int_0^s\left\|\left[{\hbox{\vrule height3.5pt depth-2.8pt width4pt}\mkern-14mu\int\nolimits}_0^t\dot\gamma_1(\tau)\,d\tau,X\right]\right\|\,dt +\frac{s}{2}\int_0^s\left\|\left[{\hbox{\vrule height3.5pt depth-2.8pt width4pt}\mkern-14mu\int\nolimits}_0^t\dot\gamma_1(\tau)\, d\tau,\dot\gamma_1(t)-X\right]\right\|\,dt \\ &&\leq\frac{\beta Ls}{2}\left(\int_0^s{\mathcal{A}}^t_0(\dot\gamma_1-X)\,dt +s\,{\mathcal{A}}_0^s(\dot\gamma_1-X)\right)\,.\end{aligned}$$ This shows that $$\begin{aligned} \label{os2} \int_0^s\|\dot\gamma_2(t)\|\,dt\leq \beta\,L\,s^2\,{\mathcal{A}}_0^s(\dot\gamma_1-X).\end{aligned}$$ From the grading property , equations can be stated as follows $$\begin{aligned} \label{confshape} \dot\gamma_j(t)=\sum_{n=2}^\upsilon\frac{(-1)^n}{n!} \sum_{\substack{1\leq l_1,\ldots,l_n\leq\upsilon\\ l_1+\cdots +l_n=j}} [\gamma_{l_1}(t),[\gamma_{l_2}(t),[\cdots,[\gamma_{l_{n-1}}(t), \dot\gamma_{l_n}(t)],],\ldots,]\end{aligned}$$ for every $j=2,\ldots,\upsilon$. We will proceed by induction, assuming that for every $j=2,\ldots,i$ and every $\upsilon\geq i$, there exists a positive constant $\kappa_j(\beta)$ such that $$\begin{aligned} \label{indsj} \int_0^s\|\dot\gamma_j(t)\|\,dt\leq\kappa_j(\beta)\,F_j(L) \;{\mathcal{A}}^s_0(\dot\gamma_1-X)\,s^j,\end{aligned}$$ where $F_j:[0,+\infty[{\longrightarrow}[0,+\infty[$ is a strictly increasing function, that is infinitesimal at zero. We have already proved this statement for $i=2$. Using for $j=i+1$ and applying the inductive hypothesis , we get $$\begin{aligned} &&\int_0^s\|\dot\gamma_{i+1}(t)\|\,dt\leq\sum_{n=2}^\upsilon\frac{\beta^n}{n!} \sum_{\substack{1\leq l_1,\ldots,l_n\leq\upsilon\\ l_1+\cdots +l_n=i+1}} \int_0^s\|\gamma_{l_1}(t)\|\|\gamma_{l_2}(t)\|\cdots\|\gamma_{l_{n-1}}(t)\| \|\dot\gamma_{l_n}(t)\|\,dt\\ &&\!\!\!\leq\sum_{n=2}^\upsilon\frac{\beta^n}{n!} \sum_{\substack{1\leq l_1,\ldots,l_n\leq\upsilon\\ l_1+\cdots +l_n=i+1}} \int_0^s\|\dot\gamma_{l_1}(t)\|\,dt\int_0^s\|\dot\gamma_{l_2}(t)\|\,dt\cdots \int_0^s\|\dot\gamma_{l_{n-1}}(t)\|\,dt\int_0^s\|\dot\gamma_{l_n}(t)\|\,dt\\ &&\leq\sum_{n=2}^\upsilon\frac{\beta^n}{n!} \sum_{\substack{1\leq l_1,\ldots,l_n\leq\upsilon\\ l_1+\cdots +l_n=i+1}} \kappa_{l_1}\cdots\kappa_{l_n}\,F_{l_1}(L)\cdots F_{l_n}(L)\ {\mathcal{A}}_0^s(\dot\gamma_1-X)^n\,s^{i+1}\\ &&\leq\sum_{n=2}^\upsilon\frac{\beta^n}{n!} \sum_{\substack{1\leq l_1,\ldots,l_n\leq\upsilon\\ l_1+\cdots +l_n=i+1}} \kappa_{l_1}\cdots\kappa_{l_n}\,F_{l_1}(L)\cdots F_{l_n}(L)\ (2L)^{n-1}\,{\mathcal{A}}_0^s(\dot\gamma_1-X)\,s^{i+1}\,.\end{aligned}$$ This proves estimates for every $j=2,\ldots,\upsilon$. To complete the proof, it suffices to apply these estimates to the Lipschitz curve $\tilde\Gamma(t)=\Gamma(-t)$ and replace $X$ by $-X$. $\Box$ \[corlt\] There exists a strictly increasing function $\Upsilon:[0,+\infty[{\longrightarrow}[0,+\infty[$, infinitesimal at zero and only depending on ${\mathcal{M}}$ that has the following property. Let $\Gamma:[a,b]{\longrightarrow}{\mathbb{M}}$ be a horizontal curve and assume that $t\in[a,b]$ is an approximate continuity points $t$ of $\dot\gamma_1$. Then $$\begin{aligned} \label{estimdif} \;\;\left\|\pi_i\left(\exp^{-1}\Big( \delta_{1/h}\big(\Gamma(t)^{-1}\Gamma(t+h)\big)\Big)\right)\right\| \leq\Upsilon(L_t)\;{\mathcal{A}}^h_t\big(\dot\gamma_1-\dot\gamma_1(t)\big)\end{aligned}$$ for every $i=2,\ldots,\upsilon$, where $L_t=\max\left\{{\mathcal{A}}^{b-t}_t(\dot\gamma_1),{\mathcal{A}}^{a-t}_t(\dot\gamma_1)\right\}$. [Proof.]{} Clearly, $\gamma_1$ is absolutely continuous. Let $t$ be an approximate continuity point of $\dot\gamma_1$ and then also a differentiability point. It suffices to apply Lemma \[theoestim\] to the curve $h{\longrightarrow}\Gamma(t)^{-1}\Gamma(t+h)$ with $X=\dot\gamma_1(t)$. In fact, we observe that $$\pi_1\left(\exp^{-1}\big(\Gamma(t)^{-1}\Gamma(t+h)\big)\right) =\gamma_1(t+h)-\gamma_1(t)$$ and that $${\mathcal{A}}^\lambda_0\big(\dot\gamma_1(t+\cdot)-\dot\gamma_1(t)\big) ={\mathcal{A}}^\lambda_t\big(\dot\gamma_1(\cdot)-\dot\gamma_1(t)\big)\,.$$ This implies that $$\begin{aligned} \;\;\left\|\pi_i\Big(\exp^{-1}\big(\Gamma(t)^{-1}\Gamma(t+h)\big)\Big)\right\|\leq\Upsilon_i(L_t)\,|h|^i\, {\mathcal{A}}^h_t\big(\dot\gamma_1-\dot\gamma_1(t)\big)\end{aligned}$$ where $\Upsilon_i$ are given in Lemma \[theoestim\] and $$L_t=\max\left\{{\mathcal{A}}^{b-t}_t(\dot\gamma_1),{\mathcal{A}}^{a-t}_t(\dot\gamma_1)\right\}\,.$$ Thus, setting $\Upsilon=\max_{i=2,\ldots,\upsilon}\Upsilon_i$, and using the definition of dilations, our claim follows. $\Box$ \[pansu\] There exists a nondecreasing function $\Upsilon:[0,+\infty[{\longrightarrow}[0,+\infty[$ only depending on ${\mathcal{M}}$ with the following property. If $\Gamma:[a,b]{\longrightarrow}{\mathbb{M}}$ is a horizontal curve, then for all approximate continuity points $t$ of $\dot\gamma_1$ the following estimate $$\begin{aligned} \label{pansuestim} \;\qquad\left\|\delta_{1/h} \Big(-h\,\dot\gamma_1(t){\circledcirc}\big(-\gamma(t)\big){\circledcirc}\gamma(t+h)\Big) \right\|\leq\Upsilon(L_t)\;{\mathcal{A}}^h_t\big(\dot\gamma_1-\dot\gamma_1(t)\big)\end{aligned}$$ holds, where $L_t=\max\left\{{\mathcal{A}}^{b-t}_t(\dot\gamma_1),{\mathcal{A}}^{a-t}_t(\dot\gamma_1)\right\}$. In particular, $\Gamma$ is a.e. P-differentiable. [Proof.]{} Let $t$ be an approximate continuity point of $\dot\gamma_1$ and define $$h{\longrightarrow}\Gamma(t)^{-1}\Gamma(t+h) =\exp\theta(h) =\exp\big(\theta_1(h)+\cdots+\theta_\upsilon(h)\big)\,,$$ with $\theta_i(h)\in V_i$. Notice that in particular $\theta_1(h)=\gamma_1(t+h)-\gamma_1(t)$. In view of , there exists a constant $K>0$ depending on ${\mathcal{M}}$ and a strictly increasing function $\Upsilon:[0,+\infty[{\longrightarrow}[0,+\infty[$ infinitesimal at zero such that $$\begin{aligned} \label{thetahi} \left\|\frac{\theta_i(h)}{h^i}\right\|\leq \Upsilon(L_t)\, {\mathcal{A}}^h_t\big(\dot\gamma_1-\dot\gamma_1(t)\big).\end{aligned}$$ for every $i=2,\ldots,\upsilon$, where we have set $$L_t=\max\left\{{\mathcal{A}}^{b-t}_t(\dot\gamma_1), {\mathcal{A}}^{a-t}_t(\dot\gamma_1)\right\}.$$ Thus, taking into account that $t$ is a differentiability point of $\gamma_1$ and that $0$ is an approximate continuity point of $\dot\theta_1$, we obtain that $\Gamma$ is P-differentiable at $t$ and that $$D\Gamma(t)(\lambda)=\exp\big(\lambda\,\dot\gamma_1(t)\big)\,.$$ Notice also that $$\left\|\frac{\theta_1(h)}{h}-\dot\gamma_1(t)\right\| \leq {\mathcal{A}}^h_t\big(\dot\gamma_1-\dot\gamma_1(t)\big),$$ then yields $$\begin{aligned} \label{thetamin} \left\|\delta_{1/h}\theta(h)-\dot\gamma_1(t)\right\|\leq \Big[1+(\upsilon-1)\Upsilon(L_t)\Big] \;{\mathcal{A}}^h_t\big(\dot\gamma_1-\dot\gamma_1(t)\big)\,.\end{aligned}$$ Applying and to $$-\dot\gamma_1(t){\circledcirc}\delta_{1/h}\theta(h) =\delta_{1/h}\theta(h)-\dot\gamma_1(t)+ \sum_{n=2}^\upsilon c_n\big(-\dot\gamma_1(t),\delta_{1/h}\theta(h)\big)\,,$$ we get the estimate $$\begin{aligned} \label{longestdot} &&\left\|-\dot\gamma_1(t){\circledcirc}\delta_{1/h}\theta(h)\right\|\leq \Big[1+(\upsilon-1)\Upsilon(L_t)\Big] \;{\mathcal{A}}^h_t\big(\dot\gamma_1-\dot\gamma_1(t)\big)\\ &&+\sum_{n=2}^\upsilon\frac{\beta^{n-1}}{n!} \left\|\frac{\delta_{1/h}\theta(h)+\dot\gamma_1(t)}{2}\right\|^{n-1} \Big[1+(\upsilon-1)\Upsilon(L_t)\Big] {\mathcal{A}}^h_t\big(\dot\gamma_1-\dot\gamma_1(t)\big)\nonumber\\ &&+\sum_{n=2}^\upsilon \left\|R_n\left(-\dot\gamma_1(t),\delta_{1/h}\theta(h)\right)\right\|\,.\nonumber\end{aligned}$$ Observing that $\|\dot\gamma_1(t)\|\leq L_t$ and ${\mathcal{A}}^h_t\big(\dot\gamma_1-\dot\gamma_1(t)\big)\leq2L_t$, inequality implies that $$\begin{aligned} &&\left\|\delta_{1/h}\theta(h)\right\|\leq 3L_t\Big[1+(\upsilon-1)\Upsilon(L_t)\Big]=\Upsilon_1(L_t),\end{aligned}$$ then we apply to the third line of , obtaining $$\begin{aligned} &&\left\|-\dot\gamma_1(t){\circledcirc}\delta_{1/h}\theta(h)\right\|\leq \bigg(1+\sum_{n=2}^\upsilon\frac{\beta^{n-1}}{n!} \left(\frac{\Upsilon_1(L_t)+L_t}{2}\right)^{n-1}\bigg)\\ &&\,\Big[1+(\upsilon-1)\Upsilon(L_t)\Big] {\mathcal{A}}^h_t\big(\dot\gamma_1-\dot\gamma_1(t)\big)\nonumber \\ && +\Big(\sum_{n=2}^\upsilon C{\mbox{\Large $($}}n,\Upsilon_1(L_t){\mbox{\Large $)$}}\Big)\, \Big[1+(\upsilon-1)\Upsilon(L_t)\Big]^3\,4\,L_t^2\; {\mathcal{A}}^h_t\big(\dot\gamma_1-\dot\gamma_1(t)\big)\nonumber\,.\end{aligned}$$ This concludes the proof. $\Box$ \[Pansugrad\] Let $A$ be a measurable set of ${\mathbb{G}}$ and let $f:A{\longrightarrow}{\mathbb{M}}$ be a Lipschitz mapping. Then $f$ is a.e. P-differentiable. [Proof.]{} By Theorem \[pansu\], rectifiable curves in graded groups are in particular a.e. P-differentiable. This allows us to extend Proposition 3.6 of [@Mag] to the case of mappings with values in a graded group ${\mathbb{M}}$. Thus, repeating the same arguments of the proof of Theorem 3.9 of [@Mag] the claim is achieved. $\Box$ Characterization of P-differentiable mappings {#contactmap} --------------------------------------------- Here we denote by ${\mathbb{G}}$ and ${\mathbb{M}}$ two arbitrary graded groups, where ${\mathbb{G}}$ is stratified and $\Omega\subset{\mathbb{G}}$ is an open set. To obtain estimates on the P-difference quotient of P-differentiable mappings, we will use the following family of piecewise horizontal lines. \[PN\][Let $N$ be a positive integer, $X_1,\ldots,X_m$ be a basis of the first layer $V_1$, $i_1,\ldots,i_N\in\{1,\ldots,m\}$ be fixed integers and for every $a=(a_1,\ldots,a_N)\in{\mathbb{R}}^N$ define $$\begin{aligned} P^s(a)=\left\{\begin{array}{ll} e & \mbox{if $s=0$} \\ \delta_{a_1}h_{i_1}\delta_{a_2}h_{i_2}\cdots\delta_{a_s}h_{i_s} & \mbox{if $s=1,\ldots,N$} \end{array}\right.\,,\end{aligned}$$ where $h_1=\exp X_1,\ldots,h_m=\exp X_m$ and we have assumed that $d(h_i)=1$. ]{} Lemma 1.40 of [@FS] yields the following \[140FS\] For every stratified group ${\mathbb{G}}$ there exists an integer $N$ and a family of integers $\{i_1,\ldots,i_N\}\subset\{1,\ldots,m\}$ depending on ${\mathbb{G}}$, such that the mapping $P^N$ of Definition \[PN\] sends a neighbourhood of the origin in ${\mathbb{R}}^N$ onto a neighbourhood of the identity $e\in{\mathbb{G}}$. \[defcaDef\][Under conditions of Lemma \[140FS\], we define the number $$\begin{aligned} \label{defca} c({\mathbb{G}},d)=\max_{\substack{s=1,\ldots,N\\ a=(a_s)\in (P^N)^{-1}(D_1)}}|a_s|,\end{aligned}$$ that only depends on the algebraic structure of ${\mathbb{G}}$ and on the homogeneous distance $d$ used to define both $P^N$ and the closed unit ball $D_1$. ]{} \[rmkPN\][Being the mapping $P^N$ homogeneous with respect to dilations, namely, $P^N(ra)=\delta_r\big(P^N(a)\big)$, it follows that $P^N$ is surjective onto ${\mathbb{G}}$. Homogeneity yields $$\begin{aligned} \label{defcar} \max_{\substack{s=1,\ldots,N\\ a=(a_s)\in (P^N)^{-1}(D_r)}}|a_s| =r\;c({\mathbb{G}},d)\end{aligned}$$ for every $r>0$ and implies that the preimage of compact sets of ${\mathbb{G}}$ is compact in ${\mathbb{R}}^N$. This shows that definition is well posed. ]{} [Let $f:K{\longrightarrow}Y$ be a vector valued continuous function on a compact metric space $(K,\rho)$. Then we define the [*modulus of continuity*]{} of $f$ on $K$ as $$\begin{aligned} \label{modcont} \omega_{K,f}(t)=\max_{\substack{x,y\in K\\ \rho(x,y)\leq t}} \|f(x)-f(y)\|\,.\end{aligned}$$ ]{} \[PdifPdifj\] Let $f:\Omega{\longrightarrow}{\mathbb{M}}$ be P-differentiable at $x\in\Omega$, where $f=\exp\circ F$ and $F=F_1+\cdots+F_\upsilon$ with $F_j:\Omega{\longrightarrow}W_j$. Then every $F_j$ is P-differentiable at $x$ and we have the formulae $$\begin{aligned} &&\pi_1\circ df(x)=dF_1(x) \label{eqF_1}\\ &&dF_i(x)(h)=\sum_{n=2}^\upsilon\frac{(-1)^n}{n!}\;\pi_i \left([F(x),dF(x)(h)]_{n-1}\right)\,. \label{eqDF_j}\end{aligned}$$ for every $i=2,\ldots,\upsilon$ and every $h\in{\mathbb{G}}$ [Proof.]{} We will use the notation $|h|=d(h)$. By definition of P-differentiability we have that $$\delta_{1/|h|}\left(f(x)^{-1}f(xh)\right)$$ converges to $Df(x)(h)$ as $h{\rightarrow}e$, uniformly with respect to the parameter $\delta_{1/|h|}h$ varying in a compact set. Then the difference quotients $$\pi_i\circ\exp^{-1}\Big[\delta_{1/|h|}\left(f(x)^{-1}f(xh)\right)\Big]\,,$$ uniformly converge for every $i=1,\ldots,\upsilon$. In other words, the difference quotient $$\frac{F_1(xh)-F_1(x)}{|h|}$$ along with $$\label{convi} \frac{F_i(xh)-F_i(x)+\sum_{n=2}^\upsilon \pi_i\left(c_n{\mbox{\large $($}}-F(x),F(xh){\mbox{\large $)$}}\right)}{|h|^i}$$ converge as $h{\rightarrow}0$ whenever $i=2,\ldots,\upsilon$. In particular, $F_1$ is $P$-differentiable at $x$ and $$\pi_1\circ\exp^{-1}\big(Df(x)(h)\big)=DF_1(x)(h)$$ that implies . Writing for $i=2$, we notice that this expression can be written as $$\begin{aligned} \label{conv2} \frac{F_2(xh)-F_2(x)- \frac{1}{2}[F_1(x),F_1(xh)]}{|h|^2}\,.\end{aligned}$$ The convergence of the previous quotient and P-differentiability of $F_1$ at $x$ imply that $F_2$ is also differentiable at $x$ and there holds $$DF_2(x)(h)=\frac{1}{2}\,[F_1(x),DF_1(x)(h)].$$ By induction, we assume that for $j\geq2$, the vector-valued mapping $F_i$ is differentiable at $x$ and that $$\begin{aligned} DF_i(x)(h)=\sum_{n=2}^\upsilon\frac{(-1)^n}{n!}\; \pi_i\left([\overline F_{i-1}(x), D\overline F_{i-1}(x)(h)]_{n-1}\right)\end{aligned}$$ holds for every $i=1,\ldots,j$, where we have set $$\overline F_i=F_1+F_2+\cdots+F_i.$$ By P-differentiability , we have $$\label{convip1} \frac{F_{i+1}(xh)-F_{i+1}(x)+\sum_{n=2}^\upsilon \pi_{i+1}\left(c_n{\mbox{\large $($}}-F(x),F(xh){\mbox{\large $)$}}\right)}{|h|}{\longrightarrow}0\quad \mbox{as}\quad h{\rightarrow}0\,.$$ Now we notice that $$\label{eqci} \pi_{i+1}\left(c_n{\mbox{\large $($}}-F(t),F(t+h){\mbox{\large $)$}}\right) =\pi_{i+1}\left(c_n{\mbox{\large $($}}-\overline{F}_i(t), \overline{F}_i(t+h){\mbox{\large $)$}}\right)\,.$$ Furthermore, inductive hypothesis and Lemma \[keylemma\] ensure the existence of $$\label{indhypl} \lim_{|h|{\rightarrow}0} \frac{c_n(-\overline F_i(x),\overline F_i(x+h))}{|h|}= \frac{(-1)^{n-1}}{n!} \left[\overline{F}_i(x),D\overline{F}_i(x)(h)\right]_{n-1}.$$ Thus, joining , and , it follows that there exists $DF_{i+1}(x)$ and $$DF_{i+1}(x)(h)=\sum_{n=2}^{i+1}\frac{(-1)^n}{n!}\;\pi_{i+1} \left([\overline{F}_i(x),D\overline F_i(x)]_{n-1}\right)\,.$$ This shows that $F:\Omega{\longrightarrow}{\mathcal{M}}$ is $P$-differentiable at $t$, hence in the previous formula we can replace $\overline F_i$ by $F$, achieving . This ends the proof. $\Box$ \[Pansuimpl\] Let $\Gamma:[a,b]{\longrightarrow}{\mathbb{M}}$ be P-differentiable at $t\in[a,b]$. Then $\gamma$ is differentiable at t and satisfies equations . This corollary is an immediate consequence of Theorem \[PdifPdifj\]. Now we can prove the first result stated in the introduction. .2truecm [Proof of Theorem \[PdifContact\].]{} If $f:\Omega{\longrightarrow}{\mathbb{M}}$ is continuously P-differentiable, then Theorem \[PdifPdifj\] implies the P-differentiability of all $F_j$’s and the validity of both formulae and . These equations in turn give the continuoity of the P-differential of every $F_j$. This shows that the first condition implies the second one. Now, we assume that the second condition holds. As observed in Remark \[Pdifhdif\], P-differentiability of vector-valued mappings implies h-differentiability, hence all $F_j$’s are in particular h-differentiable. In addition, formula yields the continuity of the horizontal differential. By our assumption and hold, therefore combining these equations with Proposition \[horprop\], it follows that $f$ also has the contact property. We have shown that $f$ is continuously h-differentiable contact, that is the third condition. Finally, we assume that $f$ is a continuously h-differentiable contact mapping and consider $$\Gamma_X(t)=\exp\big(\sum_{j=1}^\iota\gamma_{X,j}(t)\Big) =f\big(c_X(t)\big) =\exp\Big(\sum_{j=1}^\iota F_j\big(c_X(t)\big)\Big)$$ where $c_X(t)=x\exp(tX)$ and $\exp^{-1}\circ\Gamma_X=\gamma_X=\sum_{j=1}^\iota\gamma_{X,j}$ with $t{\longrightarrow}\gamma_{X,j}(t)\in V_j$. By definition of h-differentiability we immediately observe that also $F_j$ are continuously h-differentiable and we have $$\dot\gamma_{X,j}(t)=XF_j(c_X(t))$$ By our assumptions, we know that $\Gamma_X$ is a horizontal curve, then we apply the key estimate to $\Gamma_X$ at $t=0$, achieving $$\left\|\delta_{1/h} \Big(-h\,XF_1(x){\circledcirc}\big(-\gamma_X(0)\big){\circledcirc}\gamma_X(h)\Big) \right\|\leq\Upsilon(L)\;{\mathcal{A}}^h_0\big((XF_1)\circ c_X-XF_1(x)\big)$$ where $\Upsilon:[0,+\infty[{\longrightarrow}[0,+\infty[$ is a nondecreasing function depending on ${\mathbb{M}}$, ${\mathcal{A}}_0^h$ is defined in and $$\begin{aligned} L=\max\left\{{\mathcal{A}}^{\varepsilon}_0(\dot\gamma_{X,1}),{\mathcal{A}}^{-{\varepsilon}}_0(\dot\gamma_{X,1})\right\}.\end{aligned}$$ Here ${\varepsilon}>0$ is fixed such that $c_X([-{\varepsilon},{\varepsilon}])\subset U$ for every $X\in V_1$ with $\|X\|=1$, where $U\subset\Omega$ is a fixed compact neighbourhood of $x$. For every $|h|\leq{\varepsilon}$, we have $$\begin{aligned} {\mathcal{A}}^h_0\big((XF_1)\circ c_X-XF_1(x)\big) \leq \max_{\substack{y,z\in U\\ \|\exp^{-1}(y)-\exp^{-1}(z)\|\leq|h|}} \|XF_1(z)-XF_1(y)\|\,.\end{aligned}$$ We notice that $\dot\Gamma_X(0)=XF_1(x)=d_Hf(x)(X)$, then setting $$\begin{aligned} \omega_{d_Hf,U}(|h|) =\max_{\substack{y,z\in U\\ \|\exp^{-1}(y)-\exp^{-1}(z)\|\leq{\varepsilon}}} \|d_Hf(y)-d_Hf(z)\|\end{aligned}$$ and observing that $$L\leq\max_{y\in U}\|d_Hf(y)\|$$ $$\label{uniflipest} \left\|\delta_{1/h} \Big(\!-h\,XF_1(x){\circledcirc}\big(\!-\gamma_X(0)\big){\circledcirc}\gamma_X(h)\Big) \right\|\leq\Upsilon\big(\max_{y\in U}\|d_Hf(y)\|\big)\; \omega_{d_Hf,U}(|h|)\,.$$ The right hand side is independent of $x$ and tends to zero as $h{\rightarrow}0$, then taking the exponential of elements in the left hand side, we get the uniform convergence of $$\delta_{1/h}\bigg(\exp\big(-h XF_1(y)\big) \Big(f(y)^{-1}f(y\exp(hX)\Big)\bigg)$$ to the unit element $e$ as $h{\rightarrow}0$, where $y$ varies in the smaller neighbourhood $$U_{-{\varepsilon}}=\{y\in U\mid \mbox{dist}_{\|\cdot\|}(y,U^c)\geq{\varepsilon}\}\,.$$ of $x$. Thus, arguing as in Corollaire 3.3 of [@Pan2], we achieve the P-differentiability of $f$ in a possibly smaller neighbourhood of $x$. This shows the everywhere P-differentiability of $f$. Observing that $$\begin{aligned} \label{hpdif} d_Hf(x)(X)=df(x)(X)\end{aligned}$$ for every $x\in\Omega$ and $X\in V_1$, we get that $x{\longrightarrow}Df(x)_{|\exp V_1}$ is continuous. The fact that $df(x)_{|V_j}$ polynomially depends on $df(x)_{|V_1}$ concludes the proof. $\Box$ Absolutely continuous curves in graded groups {#ABSC} ============================================= This section is devoted to the characterization of absolutely continuous curves in graded groups. ${\mathbb{M}}$ and ${\mathcal{M}}$ denote a graded group along with its Lie algebra, equipped with a homogeneous distance $\rho$ and a norm $\|\cdot\|$, respectively. [Let $\gamma:[0,s]{\longrightarrow}{\mathcal{M}}$ be an absolutely continuonus curve in ${\mathcal{M}}$ and let $\Sigma=(t_0,t_1,\ldots,t_N)$ be a partition of $[0,s]$, where $t_0=0$, $t_i<t_{i+1}$ and $t_N=s$ for some $N\in\mathbb N$. The associated “sum” with respect to the group operation in ${\mathcal{M}}$ is defined by $$\begin{aligned} \label{ARiemsum} \sigma_\Sigma(\gamma)(s)=\sum_{k=0}^{N-1}\; \gamma(t_k)^{-1}{\circledcirc}\gamma(t_{k+1})\,.\end{aligned}$$ We also set $\|\Sigma\|=\max_{1\leq i\leq N}(t_i-t_{i-1})$. ]{} \[laireb\] Let $\gamma:[0,s]{\longrightarrow}{\mathcal{M}}$ be an absolutely continuous curve. Then there exists $$\begin{aligned} \label{aireb} \lim_{\|\Sigma\|{\rightarrow}0^+} \sigma_\Sigma(s) =\gamma(s)-\gamma(0)+\sum_{n=2}^\iota\frac{(-1)^{n-1}}{n!} \int_0^s[\gamma(l),\dot\gamma(l)]_{n-1}\,dl\,.\end{aligned}$$ [Proof.]{} Applying and , we get $$\sigma_\Sigma(s)=\sum_{k=0}^{N-1} \sum_{n=1}^\iota c_n\big(-\gamma(t_k),\gamma(t_{k+1})\big)\,,$$ then formula yields $$\begin{aligned} \label{developc} &&\sigma_\Sigma(s)=\gamma(s)-\gamma(0) +\sum_{k=0}^{N-1}\sum_{n=2}^\iota\frac{(-1)^{n-1}}{n!} \left[\frac{\gamma(t_k)+\gamma(t_{k+1})}{2}, \gamma(t_{k+1})-\gamma(t_k)\right]_{n-1}\\ &&+\sum_{k=0}^{N-1}\sum_{n=2}^\iota R_n\big(-\gamma(t_k),\gamma(t_{k+1})\big) \,.\nonumber\end{aligned}$$ We define $\mu_k=\big(\gamma(t_k)+\gamma(t_{k+1})\big)/2$, then the absolute continuity of $\gamma$ yields $$\sum_{k=0}^{N-1}\left[\mu_k,\gamma(t_{k+1})-\gamma(t_k)\right]_{n-1}= \sum_{k=0}^{N-1}\left[\mu_k,\int_{t_k}^{t_{k+1}}\dot\gamma(l)\,dl\right]_{n-1}=\sum_{k=0}^{N-1} \int_{t_k}^{t_{k+1}}\left[\mu_k,\dot\gamma(l)\right]_{n-1}\,dl\,.$$ Now we notice that $$\begin{aligned} &&\sum_{k=0}^{N-1} \int_{t_k}^{t_{k+1}}\left[\mu_k,\dot\gamma(l)\right]_{n-1}\,dl= \int_0^s[\gamma(l),\dot\gamma(l)]_{n-1}\,dl\\ &&+\sum_{p=0}^{n-2}\sum_{k=0}^{N-1}\int_{t_k}^{t_{k+1}} [\mu_k,[\mu_k-\gamma(l),[\gamma(l),\dot\gamma(l)]_p]]_{n-p-2}\,dl,\end{aligned}$$ hence the estimate $$\begin{aligned} &&\left\|\;\sum_{p=0}^{n-2}\sum_{k=0}^{N-1}\int_{t_k}^{t_{k+1}} [\mu_k,[\mu_k-\gamma(l),[\gamma(l),\dot\gamma(l)]_p]]_{n-p-2}\,dl\;\right\|\\ &&\leq\,(p-1)\,\beta^{n-2}\,\eta(\|\Sigma\|)\, \max_{t\in[0,s]}\|\gamma(t)\|^{n-2}\, \int_{0}^{s}\|\dot\gamma(l)\|\,dl\end{aligned}$$ immediately proves the validity of the following $$\begin{aligned} \label{airebalaye} \lim_{\|\Sigma\|{\rightarrow}0^+} \sum_{k=0}^{N-1}\left[\frac{\gamma(t_k)+\gamma(t_{k+1})}{2}, \gamma(t_{k+1})-\gamma(t_k)\right]_{n-1}= \int_0^s[\gamma(l),\dot\gamma(l)]_{n-1}\,dl\,,\end{aligned}$$ where we have set $$\eta(\tau)=\sup_{|\beta-\alpha|\leq\tau} \left|\,\int_\alpha^\beta\|\dot\gamma(l)\|\,dl\right|$$ which goes to zero as $\tau{\rightarrow}0^+$, due to the absolute continuity of $\gamma$. Taking into account , and , then limit follows. $\Box$ [Incidentally, the limit exactly coincides with the “$(n$$-$$1)$-ième aire balayée” of Definition 4.6 in [@Pan2].]{} [Let $\Gamma:[a,b]{\longrightarrow}{\mathbb{M}}$ be a continuous curve. Then we define $$\mbox{\rm Var}_a^t\Gamma=\sup_{\substack{t_0=a<t_1<\cdots<t_N=t\\ N\in{\mathbb{N}}}}\sum_{k=1}^N \rho\left(\Gamma(t_k),\Gamma(t_{k-1})\right)\,.$$ ]{} \[hdiffcurve\] Let $\Gamma:[a,b]{\longrightarrow}{\mathbb{M}}$ be a curve and define $\gamma=\exp^{-1}\circ\Gamma=\sum_{i=1}^\upsilon\gamma_i$, where $\gamma_i$ takes values in $V_i$. Then the following statements are equivalent: 1. $\Gamma$ is absolutely continuous, 2. $\gamma$ is absolutely continuous and the differential equation $$\begin{aligned} \label{eqhdiffcurve1} \dot\gamma_i(t)=\sum_{n=2}^\upsilon\frac{(-1)^n}{n!}\; \pi_i\left([\gamma(t),\dot\gamma(t)]_{n-1}\right) \end{aligned}$$ is satisfied a.e. for every $i\geq2$. If one of the previous conditions holds, then there exists a constant $C>0$ only depending on $\rho$ and $\|\cdot\|$, such that for any $\tau_1<\tau_2$, we have $$\begin{aligned} \label{absestim} \rho\big(\Gamma(\tau_1),\Gamma(\tau_2)\big)\leq C\,\int_{\tau_1}^{\tau_2}\|\dot\gamma_1(t)\|\,dt\,.\end{aligned}$$ [Proof.]{} We first assume that $\Gamma$ is absolutely continuous with respect to a homogeneous distance $\rho$ fixed in ${\mathbb{M}}$. Observing that the image of $\Gamma$ is bounded and applying Lemma \[estimi\] with $i=1$, we immediately conclude that $\gamma$ is also absolutely continuous with respect to the norm $\|\cdot\|$ fixed in ${\mathcal{M}}$; then $\gamma$ is a.e. differentiable. Let $a<\tau<b$ be both an approximate continuity point of $\dot\gamma$ and a differentiability point of the total variation $$\begin{aligned} \label{veegamma} [a,b]\ni l{\longrightarrow}\mbox{\rm Var}_a^l\Gamma\end{aligned}$$ We fix ${\varepsilon}>0$ suitable small. We fix a bounded open neighbourhood $U$ of $e$ and choose $\delta>0$ such that $\exp\big(\gamma(\tau+t_{i-1})^{-1}{\circledcirc}\gamma(\tau+t_i)\big)\in U$ whenever $\Sigma=(t_0,t_1,\ldots,t_N)$ is an arbitrary partition of $[0,s]$ with $\|\Sigma\|\leq\delta$ and $0<s\leq{\varepsilon}$. For every $i=1,\ldots,\upsilon$, there exists a constant $k_i>0$ such that $$\begin{aligned} &&\|\pi_i\big(\sigma_\Sigma(\gamma(\tau+\cdot))(s)\big)\|\leq k_i\; \|\pi^i\big(\sigma_\Sigma(\gamma(\tau+\cdot))(s)\big)\|\\ &&\leq k_i\, K_U\,\sum_{j=1}^N \rho\big(\Gamma(\tau+t_j),\Gamma(\tau+t_{j-1})\big)^i\leq k_i\, K_U\, \bigg(\mbox{\rm Var}_{\tau}^{\tau+s}\Gamma\bigg)^i\,.\end{aligned}$$ Taking the limit in the previous inequality as $\|\Sigma\|{\longrightarrow}0^+$ and applying Lemma \[laireb\], we obtain $$\begin{aligned} &&\Big\|\gamma_i(\tau+s)-\gamma_i(\tau) +\sum_{n=2}^\upsilon\frac{(-1)^{n-1}}{n!} \int_0^s\,\pi_i\left([\gamma(\tau+l),\dot \gamma(\tau+l)]_{n-1}\right)\,dl\Big\| \\ &&\leq\,k_i\, K_U\,\bigg(\mbox{\rm Var}_{\tau}^{\tau+s}\Gamma\bigg)^{i-1} \bigg(\mbox{\rm Var}_a^{\tau+s}\Gamma-\mbox{\rm Var}_a^\tau\Gamma\bigg)\end{aligned}$$ By our assumptions, $\tau$ is a differentiability point of both $\gamma$ and of the total variation and it is an approximate continuity point of $\dot\gamma$. Thus, dividing by $s$ the last inequality and taking the limit as $s{\rightarrow}0^+$ for every $i\geq2$, equations follow. Now we assume the $\gamma$ is absolutely continuous and that a.e. holds for every $i\geq2$. Let us fix $a<\tau_1<\tau_2<b$ and consider $$h{\longrightarrow}\Theta(h)=\Gamma(\tau_1)^{-1}\Gamma(\tau_1+h)=\exp\theta(h)= \exp\big(\theta_1(h)+\theta_2(h)+\cdots+\theta_\upsilon(h)\big)\,,$$ where $\theta_i(h)\in V_i$. In view of Proposition \[horprop\] and observing that $\Theta$, as left translated of $\Gamma$, is a.e. horizontal, we have that $$\begin{aligned} \label{eqhdiffcurvetheta} \dot\theta_i(h)=\sum_{n=2}^\upsilon\frac{(-1)^n}{n!}\; \pi_i\left([\theta(h),\dot\theta(h)]_{n-1}\right)\end{aligned}$$ a.e. holds and $\theta$ is absolutely continuous. Now, we wish to prove by induction that there exists a constant $C_i>0$, only depending on the norm constant $\beta$, such that $$\begin{aligned} \label{piabs} \int_0^{h_0}\|\dot\theta_i(t)\|\,dt\leq C_i\left(\int_0^{h_0}\|\dot\theta_1\|\right)^i\end{aligned}$$ for every $i\geq1$, with $h_0=\tau_2-\tau_1$. To prove , we first write as follows $$\label{confshapetheta} \dot\theta_i(h)=\sum_{n=2}^\upsilon\frac{(-1)^n}{n!} \sum_{\substack{1\leq l_1,\ldots,l_n\leq\upsilon\\l_1+\cdots+l_n=i}} [\theta_{l_1}(h),[\theta_{l_2}(h),[\cdots,[\theta_{l_{n-1}}(h),\dot\theta_{l_n}(h)],],\ldots,]\,,$$ since ${\mathbb{M}}$ is a graded group. For $i=1$, inequality is trivial, choosing $C_1\geq1$. Let us assume that holds for every $i\leq j$. Due to for $i=j+1$ and our induction hypothesis, we have $$\begin{aligned} &&\int_0^{h_0}\|\dot\theta_{j+1}(t)\|\,dt\leq \sum_{n=2}^\upsilon\frac{\beta^{n-1}}{n!}\!\!\!\!\! \sum_{\substack{1\leq l_1,\ldots,l_n\leq\upsilon\\l_1+\cdots+l_n=j+1}} \!\!\!\int_0^{h_0}\|\theta_{l_1}(t)\|\,\|\theta_{l_2}(t)\|\cdots \|\theta_{l_{n-1}}(t)\|\,\|\dot\theta_{l_n}(t)\|\,dt\\ &&\leq\sum_{n=2}^\upsilon\frac{\beta^{n-1}}{n!} \sum_{\substack{1\leq l_1,\ldots,l_n\leq\upsilon\\l_1+\cdots+l_n=j+1}} C_{l_1}\cdots C_{l_n} \Big(\int_0^{h_0}\|\dot\theta_1(t)\|\,dt\Big)^{j+1}\\ &&=\left(\int_0^{h_0}\|\dot\theta_1(t)\|\,dt\right)^{j+1}\, \Big(\sum_{n=2}^\upsilon\frac{\beta^{n-1}}{n!} \sum_{\substack{1\leq l_1,\ldots,l_n\leq\upsilon\\l_1+\cdots+l_n=j+1}} C_{l_1}\cdots C_{l_n}\Big)\,.\end{aligned}$$ Thus, we have proved that $$\begin{aligned} \label{piabsfinal} \|\theta_i(h_0)\| =\left\|\pi_i\big(-\gamma(\tau_1){\circledcirc}\gamma(\tau_2)\big)\right\|\leq C_i\left(\int_{\tau_1}^{\tau_2}\|\dot\gamma_1\|\right)^i\,,\end{aligned}$$ for every $i=1,\ldots,\upsilon$. These estimates show the validity of , hence $\Gamma$ is absolutely continuous. $\Box$ \[hdifflip\] Let $\Gamma:[a,b]{\longrightarrow}{\mathbb{M}}$ be a curve and define $\gamma=\exp^{-1}\circ\Gamma=\sum_{i=1}^\upsilon\gamma_i$, where $\gamma_i$ takes values in $V_i$. Then the following statements are equivalent: 1. $\Gamma$ is Lipschitz, 2. $\gamma$ is Lipschitz and the differential equation $$\begin{aligned} \label{eqhdiffcurve2} \dot\gamma_i(t)=\sum_{n=2}^\upsilon\frac{(-1)^n}{n!}\; \pi_i\left([\gamma(t),\dot\gamma(t)]_{n-1}\right) \end{aligned}$$ is satisfied a.e. for every $i\geq2$. If one of the previous conditions holds, then there exists a constant $C>0$ only depending on $\rho$ and $\|\cdot\|$, such that $$\begin{aligned} \label{lipestim} C^{-1}\;\mbox{\rm Lip}(\gamma_1) \leq \mbox{\rm Lip}(\Gamma)\leq C\;\mbox{\rm Lip}(\gamma_1)\end{aligned}$$ [Proof.]{} In view of Theorem \[hdiffcurve\], if $\Gamma$ is Lipschitz, then $\gamma$ is absolutely continuous and a.e. hold. By the general estimate , $\gamma_1$ is Lipschitz and we have a constant $C_1>0$, only depending on $\rho$ and $\|\cdot\|$, such that $$C_1^{-1}\;{\mbox{Lip}}(\gamma_1)\leq {\mbox{Lip}}(\Gamma)\,.$$ Thus, from the estimate follows. Applying recursively the equations , the Lipschitz property of $\gamma_1$ implies the Lipschitz property of all $\gamma_j$’s. Conversely, if $\gamma$ is Lipschitz and a.e. hold, then Theorem \[hdiffcurve\] implies that $\Gamma$ is absolutely continuous and the estimate yields the Lipschitz property of $\Gamma$. $\Box$ Let $\Gamma:[a,b]{\longrightarrow}{\mathbb{M}}$ be an absolutely continuous curve. Then the following formula holds $$\mbox{\rm Var}_a^b\Gamma=\int_a^b\rho\Big(\exp\big(\dot\gamma_1(t)\big)\Big)\,dt\,.$$ where $\mbox{\rm Var}_a^b\Gamma$ is computed with respect to the homogeneous distance $\rho$. [Proof.]{} Our claim is a consequence of a.e. P-differentiability of $\Gamma$ proved in Theorem \[pansu\] and the general formula $$\begin{aligned} \label{VarGamma} \mbox{\rm Var}_a^b\Gamma=\int_a^b |\dot\Gamma(t)|\end{aligned}$$ that holds in metric spaces, where $|\dot\Gamma(t)|$ is the metric derivative, that in this case coincides with $\rho\big(\exp\big(\dot\gamma_1(t)\big)\big)$, by definition of P-differentiability. Formula can be found in [@AmbTil] for Lipschitz curves in metric spaces. Its extension to absolutely continuous curves is straightforard and can be proved in a similar manner, taking into account formula . $\Box$ Lipschitz property of P-differentiable mappings {#LpPD} ----------------------------------------------- In the sequel, ${\mathbb{G}}$ will denote a stratified group with Lie algebra ${\mathcal{G}}$ and $\Omega\subset{\mathbb{G}}$ will be an open set. The mapping $f:\Omega{\longrightarrow}{\mathbb{M}}$ will be represented in the algebra by $F=\exp^{-1}\circ f:\Omega{\longrightarrow}{\mathcal{M}}$ and $F_j=\pi_j\circ F$, where $\pi_j:{\mathcal{M}}{\longrightarrow}W_j$ are the canonical projections onto the layers. \[PlipV\] Let $V$ be a finite dimensional normed vector space and let $F:\Omega{\longrightarrow}V$ be continuously P-differentiable. There exist constants $\mu_0>1$ and $C_0>0$, only depending on $({\mathbb{G}},d)$, such that $$\|F(x)-F(y)\|\leq C_0\max_{z\in D_{\xi,\mu_0r}}\|dF(z)\|\;\,d(x,y)$$ for every $x,y\in D_{\xi,r}$, where $D_{\xi,\mu_0r}\subset\Omega$. [Proof.]{} We consider $c_0=c({\mathbb{G}},d)$ as in and fix $\mu_0=2(1+c_0N)$. If $D_{\xi,\mu_0r}\subset\Omega$ for a suitable small $r>0$, then one can check that $$A_{\xi,r}=\{y\in{\mathbb{G}}\mid {\mbox{\rm dist}}(y,B_{\xi,r})\leq 2\,c_0\,N\,r\}\subset B_{\xi,\mu_0r}\,.$$ Let $x,y\in B_{\xi,r}$ and observe that Remark \[rmkPN\] implies the existence of $a\in (P^N)^{-1}(D_1)$ such that $y=x\delta_{d(x,y)}P^N(a)$. We fix $\lambda=d(x,y)$, then taking into account that $x\in B_{\xi,r}$ one easily realizes that $$[0,\lambda\, a_s]\ni l{\longrightarrow}x\delta_\lambda P^{s-1}(a)\delta_lh_{i_s}\in A_{\xi,r}\subset B_{\xi,\mu_0r}\subset\Omega$$ for every $s=1,\ldots,N$. Now, for every horizontal direction $h=\exp X\in\exp (V_1)$ and every $z\in\Omega$ such that $$[0,1]\ni t{\longrightarrow}\Gamma(t)=z\exp tX=z\delta_th\in\Omega$$ we have the integral formula $$F(xh)-F(x)=\int_0^1 DF(x\exp t X)(h)\,dt\,,$$ since $t{\longrightarrow}F\circ \Gamma(t)$ is continuously differentiable with $\big(F\circ\Gamma\big)'(t)=DF(x\exp tX)(h)$. It follows that $$\begin{aligned} \label{estFh} \|F(xh)-F(x)\| \leq\|X\|\,\int_0^1\|dF(x\delta_th)\|\,dt\,.\end{aligned}$$ For every $s=1,\ldots,N$, we apply to the curve $$\Gamma(t)=x\delta_\lambda P^{s-1}(a)\delta_{t\lambda a_s}h_{i_s}\in D_{\xi,\mu_0r}\quad\mbox{for every}\quad t\in[0,1]$$ obtaining the final estimates $$\begin{aligned} &&\|F(y)-F(x)\|\leq\sum_{s=1}^N \|\exp^{-1}\big(\delta_{\lambda a_s}h_{i_s}\big)\|\; \int_0^1\|dF(x\delta_\lambda P^{s-1}(a)\delta_{t\lambda a_s}h_{i_s})\|\,dt \\ &&\leq c_0\,d(x,y)\,\max_{1\leq s\leq N}\|\exp^{-1}\big(h_{i_s}\big)\|\, \max_{y\in D_{\xi,\mu_0r}}\|dF(y)\|=C_0\,d(x,y)\, \max_{y\in D_{\xi,\mu_0r}}\|dF(y)\|\,,\nonumber\end{aligned}$$ that conclude the proof. $\Box$ \[lipFcurve\] Let $c:[a,b]{\longrightarrow}\Omega$ be a Lipschitz curve and let $f:\Omega{\longrightarrow}{\mathbb{M}}$ be a continuously P-differentiable mapping. Then there exists a constant $C>0$, only depending on $d$, $\rho$ and the norms on ${\mathcal{G}}$ and ${\mathcal{M}}$, such that $\Gamma=f\circ c$ is Lipschitz and the following formulae hold $$\begin{aligned} &&\rho\big(\Gamma(t),\Gamma(\tau)\big)\leq C\;\mbox{\rm Lip}(c) \max_{x\in c([a,b])} \|dF_1(x)\|\;|t-\tau|\,, \label{Liplip} \\ && \dot\gamma_1(t)=dF_1\big(c(t)\big)\circ dc(t)=dF_1\big(c(t)\big) \circ\dot\alpha_1(t) \label{dF_1}\,,\end{aligned}$$ where we have set $\gamma_j=\pi_j\circ\exp^{-1}\circ\Gamma$, $\alpha_j=\pi_j\circ\exp^{-1}\circ c$, $$\Gamma=\exp\big(\gamma_1+\cdots+\gamma_\upsilon\big)\qquad \mbox{and}\qquad c=\exp(\alpha_1+\cdots+\alpha_\iota)\,.$$ [Proof.]{} Let $c:[a,b]{\longrightarrow}\Omega$ be a horizontal curve, where $f:\Omega{\longrightarrow}{\mathbb{M}}$ is continuously P-differentiable. Due to Theorem \[PdifPdifj\], $F_j$ are continuously P-differentiable, hence Lemma \[PlipV\] yields their local Lipschitz property. Thus, the composition $$F\circ c:[a,b]{\longrightarrow}{\mathcal{M}}$$ is locally Lipschitz, hence it is absolutely continuous. By Corollary \[hdifflip\], $c$ is horizontal, then Theorem \[pansu\] yields the a.e. P-differentiability of $c$. In view of the chain rule stated in Proposition \[chain\], it follows that $\Gamma$, equal to $f\circ c$, is a.e. P-differentiable and $$d\Gamma(t)=df\big(c(t)\big)\big(dc(t)\big)\,.$$ for a.e. $t\in[a,b]$. Applying Corollary \[Pansuimpl\] and the characterization of Proposition \[horprop\], it follows that $\Gamma$ is horizontal. In view of , the formula above yields the a.e. validity of , since $c$ is horizontal. Due to and , it follows that $\Gamma$ is Lipschitz and formula follows. $\Box$ \[corlip\] Every continuously P-differentiable mapping $f:\Omega{\longrightarrow}{\mathbb{M}}$ is locally Lipschitz. [Proof.]{} It suffices to observe that $\Omega$ is locally connected by geodesics with respect to a Carnot-Carathéodory distance fixed on ${\mathbb{G}}$ and then apply Proposition \[lipFcurve\]. $\Box$ Mean value inequality {#MVI} ===================== Throughout this section, ${\mathbb{G}}$ and ${\mathcal{G}}$ denote a stratified group along with its Lie algebra, $\Omega\subset{\mathbb{G}}$ is a fixed open set, $d$ is a homogeneous distance on ${\mathbb{G}}$. ${\mathbb{M}}$ is a graded group equipped with a homogeneous distance $\rho$. When the norm $\|\cdot\|$ is applied to a vector of either $T{\mathbb{G}}$ or $T{\mathbb{M}}$, it is understood that it represents the left invariant Finsler norm generated by the fixed norm in the Lie algebra of either ${\mathbb{G}}$ or ${\mathbb{M}}$. The mapping $f:\Omega{\longrightarrow}{\mathbb{M}}$ will be represented in the algebra by $F=\exp^{-1}\circ f:\Omega{\longrightarrow}{\mathcal{M}}$ and $F_j=\pi_j\circ F$, where $\pi_j:{\mathcal{M}}{\longrightarrow}W_j$ are the canonical projections onto the layers. \[unifestimth\] Let $f:\Omega{\longrightarrow}{\mathbb{M}}$ be a continuously P-differentiable mapping and let $\Omega'$ be an open subset compactly contained in $\Omega$. Let $c:[a,b]{\longrightarrow}\Omega'$ be a Lipschitz curve with $\|\dot c(t)\|\leq1$ a.e. Then there exists a constant $C$, depending on the modulus of continuity of $dF_1$ on $\overline{\Omega'}$ and on $L=\max_{x\in\overline{\Omega'}}\|dF_1(x)\|$, such that at approximate continuity points $t$ of $\dot c$ the following estimate holds $$\begin{aligned} \label{unifestimch} &&\quad\rho\left(Df{\mbox{\large $($}}c(t){\mbox{\large $)$}}\big(\exp \dot c(t)\,h\big),f{\mbox{\large $($}}c(t){\mbox{\large $)$}}^{-1}f{\mbox{\large $($}}c(t+h){\mbox{\large $)$}}\right)\\ &&\leq C\Bigg[L^{1/\iota} \sup_{s\in I_{t,t+h}}\left|\,{\hbox{\vrule height3.5pt depth-2.8pt width4pt}\mkern-14mu\int\nolimits}_t^{t+h}\|\dot c(l)-\dot c(t)\|\,dl\,\right|^{1/\iota} \!\!\!+\!\max_{s\in I_{t,t+h}}\|dF_1{\mbox{\large $($}}c(s){\mbox{\large $)$}}-dF_1{\mbox{\large $($}}c(t){\mbox{\large $)$}}\|^{1/\iota} \Bigg]\!|h|, \nonumber\end{aligned}$$ where we have set $I_{t,t+h}=[\min\{t,t+h\},\max\{t,t+h\}]$. [Proof.]{} By Proposition \[lipFcurve\], it follows that $\Gamma=f\circ c=\exp\circ \gamma$ is Lipschitz and $$\begin{aligned} \label{gammalpha} \dot\gamma_1(t)=dF_1(c(t))\big(\dot\alpha_1(t)\big)\,,\end{aligned}$$ where $df=\exp^{-1}\circ Df\circ \exp$ and $c=\exp\big(\alpha_1+\cdots+\alpha_\iota)$. Observing that $\|\dot\alpha_1\|\leq 1$ a.e., we then apply Theorem \[pansu\] to $\Gamma$. As a consequence, taking into account , the estimate applied to $\Gamma$ yields $$\begin{aligned} \label{pansuestim1} &&\left\|\delta_{1/h} \Big[-h\,dF_1{\mbox{\large $($}}c(t){\mbox{\large $)$}}\,\big(\dot\alpha_1(t)\big){\circledcirc}\big(-F{\mbox{\large $($}}c(t){\mbox{\large $)$}}{\circledcirc}F{\mbox{\large $($}}c(t+h){\mbox{\large $)$}}\Big]\right\|\\ &&\leq\Upsilon(L)\;\sup_{s\in I_{t,t+h}}\left|\,{\hbox{\vrule height3.5pt depth-2.8pt width4pt}\mkern-14mu\int\nolimits}_t^s \|dF_1\big(c(l)\big)\big(\dot\alpha_1(l)\big)-dF_1\big(c(t)\big) \big(\dot\alpha_1(t)\big)\|dl\right| \nonumber \\ &&\leq\Upsilon(L)\; \Bigg[L\sup_{s\in I_{t,t+h}}\left|\,{\hbox{\vrule height3.5pt depth-2.8pt width4pt}\mkern-14mu\int\nolimits}_t^s \|\dot\alpha_1(l)-\dot\alpha_1(t)\|\,dl\,\right| +\max_{s\in I_{t,t+h}}\|dF_1{\mbox{\large $($}}c(s){\mbox{\large $)$}}-dF_1{\mbox{\large $($}}c(t){\mbox{\large $)$}}\|\Bigg]\nonumber\end{aligned}$$ where $\Upsilon$ is the a nondecreasing function defined in Theorem \[pansu\]. Now, applying to , we obtain a constant $C>0$ such that holds. It is not difficult to notice that $C$ only depends on the quantities stated in our claim. $\Box$ Let $f:\Omega{\longrightarrow}{\mathbb{M}}$ be a continuously P-differentiable mapping and let $\Omega_1$, $\Omega_2$ be open subsets such that $\Omega_2$ is compactly contained in $\Omega$ and for every $x,y\in\Omega_1$ such that $x^{-1}y\in\exp(V_1)$ we have $x\delta_l(x^{-1}y)\in\overline{\Omega}_2$ whenever $0\leq l\leq1$. Then there exists a constant $C$, depending on the modulus of continuity of $\omega_{\overline{\Omega}_2,dF_1}$ and on $\max_{x\in\overline{\Omega_2}}\|dF_1(x)\|$ such that $$\label{unifestimh} \rho\left(Df{\mbox{\large $($}}x{\mbox{\large $)$}}\big(x^{-1}y\big), f{\mbox{\large $($}}x{\mbox{\large $)$}}^{-1}f{\mbox{\large $($}}y{\mbox{\large $)$}}\right)\leq C\; \omega_{\overline{\Omega}_2,dF_1}\big(d(x,y)\big)^{1/\iota}\,d(x,y)\,.$$ whenever $x,y\in\Omega_1$ and $x^{-1}y\in\exp V_1$. [Proof.]{} It suffices to apply Theorem \[unifestimth\] to the curve $c(l)=x\delta_lu$, with $[a,b]=[0,\rho(x,y)]$, $t=0$, $h=d(x,y)$ and $u=\delta_{1/\rho(x,y)}(x^{-1}y)$. Then $$\rho\left(Df{\mbox{\large $($}}x{\mbox{\large $)$}}\big(x^{-1}y\big), f{\mbox{\large $($}}x{\mbox{\large $)$}}^{-1}f{\mbox{\large $($}}y{\mbox{\large $)$}}\right)\leq C\;\Bigg[ \max_{l\in I_{t,t+h}}\|dF_1{\mbox{\large $($}}c(l){\mbox{\large $)$}}-dF_1{\mbox{\large $($}}x{\mbox{\large $)$}}\|^{1/\iota} \Bigg]\,d(x,y)\,.$$ Observing that $d(c(l),x)\leq d(x,y)$, estimate follows. $\Box$ .2truecm [Proof of Theorem \[unifestim\].]{} Let $x,y\in\Omega_1$ and let $a\in(P^N)^{-1}(D_1)$ be such that $P^N(a)=\delta_{1/d(x,y)}\big(x^{-1}y\big)$. Thus, due to Definition \[PN\], we have $$d(P^s(a))\leq N\,c({\mathbb{G}},d)$$ where $c({\mathbb{G}},d)$ and $N$ are defined in and in Lemma \[140FS\], respectively. For the sequel, we set $c_0=c({\mathbb{G}},d)$. Analogously, we have $$d\big(\delta_tP^{s-1}(a)\delta_lh_{i_s}\big)\leq c_0\,N\,{\mbox{diam}}(\Omega_1)$$ if $l\in [0,ta_s]$ and $0\leq t\leq{\mbox{diam}}(\Omega_1)$. We fix $t_0=d(x,y)\leq{\mbox{diam}}(\Omega_1)$, hence from our assumptions it follows that each curve $$[0,t_0a_s]\ni l{\longrightarrow}x\delta_{t_0}P^{s-1}(a)\delta_lh_{i_s}$$ is contained in $\Omega_2$. Applying to this curve, there exists a constant $C_1$, only depending on both $\omega_{\overline\Omega_2,dF_1}$ and $\max_{x\in\overline{\Omega_2}}\|dF_1(x)\|$, such that $$\begin{aligned} && \rho\Big(f(x\delta_{t_0}P^{s-1}(a))^{-1}f\big(x\delta_{t_0}P^s(a)\big), Df(x\delta_{t_0}P^{s-1}(a))\big(\delta_{a_st_0}h_{i_s}\big)\Big) \\ &&\leq C_1\;|a_s\,t_0|\; \omega_{\overline{\Omega}_2,dF_1}\big(c_0\,t_0\big)^{1/\iota}.\nonumber\end{aligned}$$ Then we obtain the key uniform estimate $$\begin{aligned} \label{keyunif} && \rho\Big( \delta_{1/t_0}{\mbox{\Large $($}}f(x\delta_{t_0}P^{s-1}(a))^{-1}f\big(x\delta_{t_0}P^s(a)\big) {\mbox{\Large $)$}},Df(x\delta_{t_0}P^{s-1}(a))\big(\delta_{a_s}h_{i_s}\big)\Big) \\ && \leq C_1\,c_0\; \omega_{\overline{\Omega}_2,dF_1}\big(c_0\,t_0\big)^{1/\iota}.\nonumber \end{aligned}$$ We fix $\nu=\max_{x\in\overline{\Omega_2}}\|dF_1(x)\|$, so that yields a constant $K_\nu>0$ such that $$\rho\big(Df(x)\big)\leq k_\nu\|dF_1(x)\|^{1/\iota}$$ where we have set $\rho(Df(x))=\max_{d(h)=1}\rho\big(Df(x)(h)\big)$. We have the estimate $$\begin{aligned} \label{diffest} \rho\big(Df(x)(\delta_{a_s}h_{i_s}\cdots\delta_{a_N}h_{i_N})\big)\leq c_0\,k_\nu\,N\,\nu^{1/\iota} \end{aligned}$$ In view of and , we use , getting a constant $C_2$, only depending on ${\mathbb{G}}$, $\nu$ and $\omega_{\overline{\Omega},dF_1}$ such that $$\begin{aligned} \label{firstmult} &&\rho\Big(\delta_{1/t_0}{\mbox{\Large $($}}f(x)^{-1}f\big(x\delta_{t_0}P^N(a)\big){\mbox{\Large $)$}}, Df(x)\big(P^N(a)\big)\Big)\\ &&\leq C_2 \sum_{s=1}^N \rho\Big(\delta_{1/t_0} {\mbox{\Large $($}}f(x\delta_{t_0}P^{s-1}(a))^{-1}f\big(x\delta_{t_0}P^s(a)\big){\mbox{\Large $)$}}, Df(x)\big(\delta_{a_s}h_{i_s}\big)\Big)^{1/\iota}\,,\nonumber\end{aligned}$$ where we have used the decomposition $$\begin{aligned} \label{hordec} \delta_{1/t_0}{\mbox{\Large $($}}f(x)^{-1}f\big(x\delta_{t_0}P^N(a)\big){\mbox{\Large $)$}}= \prod_{s=1}^N\delta_{1/t_0}{\mbox{\Large $($}}f\big(x\delta_{t_0}P^{s-1}(a)\big)^{-1}f\big(x\delta_{t_0}P^s(a)\big){\mbox{\Large $)$}}.\end{aligned}$$ By we have $$\begin{aligned} \label{previntest} &&\rho\Big(\delta_{1/t_0} {\mbox{\Large $($}}f\big(x\delta_{t_0}P^{s-1}(a)\big)^{-1}f\big(x\delta_{t_0}P^s(a)\big){\mbox{\Large $)$}}, Df(x)\big(\delta_{a_s}h_{i_s}\big)\Big)^{1/\iota}\\ &&\leq \big(C_1\,c_0\big)^{1/\iota}\; \omega_{\overline{\Omega}_2,dF_1}\big(c_0\,t_0\big)^{1/\iota^2} +\rho\Big(Df(x\delta_{t_0}P^{s-1}(a)(\delta_{a_s}h_{i_s}), Df(x)\big(\delta_{a_s}h_{i_s}\big)\Big)^{1/\iota}\nonumber\end{aligned}$$ By we have a constant $C_3>0$ depending on $c$ and $\nu$, such that $$\begin{aligned} \label{intermest} \rho\Big(Df(x\delta_{t_0}P^{s-1}(a))(\delta_{a_s}h_{i_s}), Df(x)\big(\delta_{a_s}h_{i_s}\big)\Big)\leq c_0\;C_3\; \omega_{\overline{\Omega_2},dF_1}(N\,c_0\,t_0)^{1/\iota}\end{aligned}$$ Joining , and , we achieve $$\begin{aligned} &&\rho\Big(\delta_{1/t_0}{\mbox{\Large $($}}f(x)^{-1}f\big(x\delta_{t_0}P^N(a)\big){\mbox{\Large $)$}}, Df(x)\big(P^N(a)\big)\Big)\\ &&\leq N\,C_2\Big(c_0^{1/\iota}\,C_1^{1/\iota} \omega_{\overline{\Omega}_2,dF_1}(c_0t_0)^{1/\iota^2} +c_0^{1/\iota}\,C_3^{1/\iota}\omega_{\overline\Omega_2dF_1} (N\,c_0\,t_0)^{1/\iota^2}\Big)\,.\nonumber\end{aligned}$$ Taking into account that $y=x\delta_{t_0}P^N(a)$, then the previous estimate can be precisely written as follows $$\begin{aligned} &&\frac{\rho\Big(f(x)^{-1}f(y),Df(x)(x^{-1}y)\Big)}{d(x,y)}\\ &&\leq N\,C_2\,c_0^{1/\iota} \Big(C_1^{1/\iota}\omega_{\overline\Omega_2,dF_1}(c_0\,d(x,y))^{1/\iota^2} +C_3^{1/\iota}\omega_{\overline\Omega_2dF_1} (N\,c_0\,d(x,y))^{1/\iota^2}\Big)\,.\nonumber\end{aligned}$$ The estimate obtained in $\Omega_1$ easily extends to its closure. This concludes the proof. $\Box$ Inverse mapping theorem {#invmapthe} ----------------------- As an immediate consequence of the mean value inequality we achieve the inverse mapping theorem for P-differentiable mappings. \[inverse\] Let $\Omega$ be an open subset of ${\mathbb{G}}$ and let $f:\Omega{\longrightarrow}{\mathbb{G}}$ be a continuously P-differentiable mapping and assume that $Df(x):{\mathbb{G}}{\longrightarrow}{\mathbb{G}}$ is invertible at some $\overline{x}\in\Omega$. Then there exists a neighbourhood $U$ of $\overline{x}$ such that the restriction $f_{|U}$ has an inverse mapping $g$ that is also P-differentiable and for every $y\in f(U)$ we have $$Dg(y)=Df(g(y))^{-1}.$$ [Proof.]{} By continuity of $x{\longrightarrow}Df(x)$, there exists a compact neighbourhood $U'$ of $\overline{x}$ and a number $\mu>0$ such that $$\min_{x\in U'}\min_{d(u)=1}\rho\big(Df(x)(u)\big)=\mu.$$ Then triangle inequality and estimates give us a constant $\beta>0$ and an open neighbourhood $U$ contained in $U'$ and depending on the modulus of continuity $\omega_{df,U'}$, such that $$d\big(f(x),f(y)\big)\geq \beta\;d(x,y)$$ whenever $x,y\in U$. Then $f_{|U}$ is invertible onto its image $f(U)$ and by the domain invariance theorem, see for instance Theorem 3.3.2 of [@LLoyd], $f_{|U}$ is an open mapping. Then the inverse mapping $g:f(U){\longrightarrow}U$ is continuous and the chain rule of Proposition \[chain\] concludes the proof. $\Box$ Homogeneous subgroups {#homsubgr} ===================== Throughout this section, we assume that all subgroups are closed, connected and simply connected Lie subgroups, if not otherwise stated. Here ${\mathbb{G}}$ denotes an arbitrary graded group, that is not necessarily stratified. Its Lie algebra is denoted by ${\mathcal{G}}$. From 5.2.4 of [@Stein], we recall the notion of homogeneous subalgebra and the corresponding notion of homogeneous subgroup. [Let ${\mathfrak{p}}\subset{\mathcal{G}}$ be a Lie subalgebra. We say that ${\mathfrak{p}}$ is a [*homogeneous subalgebra*]{} if $\delta_r{\mathfrak{p}}\subset{\mathfrak{p}}$ for every $r>0$. ]{} [It is not difficult to find examples of subalgebras which are not homogeneous. It suffices to consider ${\mathcal{L}}={\mbox{span}}\{X+Z\},$ which is a subalgebra of the Heisenberg algebra ${\mathfrak{h}}^1$ of brackets $[X,Y]=Z$. However, $\delta_2(X+Z)=2X+4Z\notin{\mathcal{L}}$. ]{} \[pfactrs\] Let ${\mathfrak{a}}$ be a homogeneous subalgebra of ${\mathcal{G}}$, where ${\mathcal{G}}$ is decomposed into the direct sum $V_1\oplus\cdots\oplus V_\iota$. Then we have $${\mathfrak{a}}=(V_1\cap{\mathfrak{a}})\oplus (V_2\cap{\mathfrak{a}})\oplus\cdots\oplus(V_\iota\cap{\mathfrak{a}})\,.$$ [Proof.]{} We have only to prove that each vector $\xi\in{\mathfrak{a}}$ with the unique decomposition $\xi=\sum_{j=1}^\iota\xi_j$, with $\xi_j\in V_j$, satisfies the property $\xi_j\in{\mathfrak{a}}$. By hypothesis, $\delta_r\xi\in{\mathfrak{a}}$ whenever $r>0$, then the closedness of ${\mathfrak{a}}$ implies that $$\lim_{r{\rightarrow}0^+}\frac{1}{r}\;\delta_r\xi=\xi_1\in{\mathfrak{a}}\,.$$ This implies that $\xi-\xi_1\in{\mathfrak{a}}$, then $$\lim_{r{\rightarrow}0^+}\frac{1}{r^2}\;\delta_r(\xi-\xi_1)=\xi_2\in{\mathfrak{a}}\,.$$ Iterating this process by induction, our claim follows. $\Box$ \[homsub\] Every homogeneous subalgebra ${\mathfrak{a}}\subset{\mathcal{G}}$ is a graded algebra. [Let ${\mathfrak{a}}={\mathfrak{a}}_1\oplus\cdots\oplus{\mathfrak{a}}_\iota$ be a homogeneous subalgebra. In general, some factor ${\mathfrak{a}}_j$ might be the null space. Consider for instance the homogeneous subalgebra ${\mathfrak{a}}=V_2\oplus V_4\oplus\cdots\oplus V_{2[\iota/2]}$, where ${\mathcal{G}}=V_1\oplus V_2\oplus\cdots\oplus V_\iota$. ]{} [Let $P\subset{\mathbb{G}}$ be a Lie subgroup. We say that $P$ is a [*homogeneous subgroup*]{} if $\delta_rP\subset P$ for every $r>0$. ]{} It is clear that all properties of homogeneous subalgebras are exactly translated to homogeneous subgroups. Thus, in the sequel we will equivalently work with either homogeneous subalgebras or homogeneous subgroups. Complementary subgroups {#complemsbsect} ----------------------- This subsection is devoted to the notion of complementary subgroup, that plays an important role in the algebraic side of this paper. When $A$ and $B$ are subsets of an abstract group $G$ we will use the notation $$AB=\{ab\,|\, a\in A, b\in B\}.$$ \[complem\][Let $P$ be a homogeneous subgroup of ${\mathbb{G}}$. If there exists a homogeneous subgroup $H$ of ${\mathbb{G}}$ satisfying the properties $PH={\mathbb{G}}$ and $P\cap H=\{e\}$, then we say that $H$ is a [*complementary subgroup*]{} to $P$. ]{} [Recall that for abstract subgroups $A,B$ of an abstract group $G$, the subset $AB$ is an abstract subgroup if and only if $AB=BA$, see for instance [@Her]. As a consequence, $H$ is complementary to $P$ if and only if $P$ is complementary to $H$. ]{} Due to the previous remark, being complementary is a symmetric relation and we can say that two subgroups are [*complementary*]{}. \[uniqued\] [Let $H$ and $P$ be complementary subgroups of ${\mathbb{G}}$ and let $g\in{\mathbb{G}}$. Then it is immediate to check that there exist unique elements $h,h'\in H$ and $p,p'\in P$ such that $g=hp=p'h'$. ]{} The following proposition characterizes complementary subgroups through their Lie subalgebras. \[homdec\] Let ${\mathfrak{p}}$ and ${\mathfrak{h}}$ be homogeneous subalgebras of ${\mathcal{G}}$ and let $P$ and $H$ denote their corresponding homogeneous subgroups, respectively. Then the condition ${\mathfrak{p}}\oplus{\mathfrak{h}}={\mathcal{G}}$ is equivalent to require that $P$ and $H$ are complementary subgroups. Furthermore, if one of these conditions hold, then the mapping $$\begin{aligned} \label{phidecomp} \phi:{\mathfrak{p}}\times{\mathfrak{h}}{\longrightarrow}{\mathbb{G}}, \qquad \phi(W,Y)=\exp W\;\exp Y\end{aligned}$$ is a diffeomophism. [Proof.]{} Assume that ${\mathfrak{p}}\oplus{\mathfrak{h}}={\mathcal{G}}$. By Theorem 2.10.1 of [@Vara] the exist bounded neighbourhoods $\Omega_1$ and $\Omega_2$ of the origin in ${\mathfrak{p}}$ ad in ${\mathfrak{h}}$, respectively, such that the mapping restricted to $\Omega_1\times\Omega_2$ is a diffeomorphism onto some neighbourhood $U$ of $e$. Observing that $$\delta_r\big(\phi(W,Y)\big)=\phi(\delta_rW,\delta_rY)$$ for every $r>0$, it follows that $\phi$ is invertible. This shows that $P$ and $H$ are complementary subgroups. Viceversa, suppose that $P$ and $H$ are complementary subgroups and assume by contradiction that there exists $X\in{\mathfrak{p}}\cap{\mathfrak{h}}{\setminus}\{0\}$. This implies that $\exp X\in H\cap P=\{e\}$, that is absurd. Now, we wish to show that ${\mathfrak{p}}\oplus{\mathfrak{h}}={\mathcal{G}}$. By contradiction, we assume that $\dim\big({\mathfrak{p}}\oplus{\mathfrak{h}}\big)<\dim{\mathcal{G}}$. We notice that the mapping $\phi$ defined in has everywhere injective differential, since ${\mathfrak{h}}\cap{\mathfrak{p}}=\{0\}$ and both $\exp_{|{\mathfrak{p}}}$ and $\exp_{|{\mathfrak{h}}}$ have everywhere injective differential. Thus, $\phi({\mathfrak{p}}\times{\mathfrak{h}})\subset{\mathbb{G}}$ is an open subset, whose dimension is less than $\dim{\mathcal{G}}$. As a consequence, $\phi$ cannot be surjective and this conflicts with hypothesis $PH={\mathbb{G}}$. $\Box$ [Joining Proposition \[pfactrs\] and Proposition \[homdec\], we have the following property. If $K$ and $P$ are complementary subgroups, then $$\begin{aligned} \label{QKP} Q=\mbox{${\mathcal{H}}$-$\dim(K)$}+\mbox{${\mathcal{H}}$-$\dim(P)$},\end{aligned}$$ where $\mbox{${\mathcal{H}}$-$\dim$}$ denotes the Hausdorff dimension with respect to a fixed homogeneous distance. One can interpret this fact as a “proper splitting" of the ambient group even with respect to the metric point of view. ]{} Through Proposition \[homdec\], it is easy to observe that not any subgroup admits a complementary subgroup. \[nnexcmpl\][Let us consider the second layer ${\mathfrak{n}}={\mbox{span}}\{Z\}$ of the Heisenberg algebra ${\mathfrak{h}}^1$, with bracket relations $[X,Y]=Z$. Then the normal subgroup $N=\exp({\mathfrak{n}})$ does not possess any complementary subgroup, see also [@FSSC6]. In fact, let ${\mathfrak{a}}$ be any 2-dimensional homogeneous subalgebra of ${\mathfrak{h}}^1$. Then Proposition \[pfactrs\] gives the decomposition ${\mathfrak{a}}={\mathfrak{a}}_1\oplus{\mathfrak{a}}_2$, where ${\mathfrak{a}}_j$ are contained in the $j$-th layer, $j=1,2$. If ${\mathfrak{a}}_2=\{0\}$, then ${\mathfrak{a}}={\mathfrak{a}}_1={\mathfrak{v}}$ is not a subalgebra of ${\mathfrak{h}}^1$. Thus, ${\mathfrak{a}}$ must contain ${\mathfrak{n}}$ and this conflicts with existence of a 2-dimensional subalgebra complementary to ${\mathfrak{n}}$. ]{} [More generally there do not exist 2-dimensional subalgebras ${\mathfrak{a}}$ of ${\mathfrak{h}}^1$ such that ${\mathfrak{a}}\oplus{\mathfrak{n}}={\mathfrak{h}}^1$, even if we do not require the homogeneity of ${\mathfrak{a}}$. In fact, let ${\mathfrak{a}}={\mbox{span}}\{U+\alpha Z,U'+\alpha'Z\}$ be a 2-dimensional subalgebra of ${\mathfrak{h}}^1$, where $U,U'$ belong to the first layer ${\mathfrak{v}}$. If $[U,U']=0$ then they are proportional and ${\mathfrak{a}}\oplus{\mathfrak{n}}$ is 2-dimensional. It follows that $[U,U']=\gamma Z$, with $\gamma\neq0$. On the other hand, ${\mathfrak{a}}$ is a subalgebra, hence $Z\in{\mathfrak{a}}$ and this would imply ${\mathfrak{a}}={\mathfrak{h}}^1$. This is also a contradiction, then such a complementary subalgebra cannot exist. ]{} We also point out that a complementary subgroup, when it exists need not be unique, as we show in the next [All the homogeneous subalgebras $${\mathfrak{a}}_\lambda={\mbox{span}}\{X+\lambda Y\}\qquad \lambda\in{\mathbb{R}},$$ of ${\mathbb{H}}^1$ yield complementary subgroups $A_\lambda=\exp{\mathfrak{a}}_\lambda$ of $S=\exp\big({\mbox{span}}\{Y,Z\}\big).$ ]{} Next, we wish to see how the notion of complementary subgroup is translated for the corresponding subalgebras, when ${\mathbb{G}}$ is a graded group. First of all, we notice that a decomposition of ${\mathcal{G}}$ as a direct sum of two subspaces does not ensure a corresponding decomposition of ${\mathbb{G}}$ as the product of their images through the exponential mapping. This simple fact is shown in the next example. [We consider the Heisenberg algebra ${\mathfrak{h}}^1$ with basis $(X,Y,Z)$ and bracket relation $[X,Y]=Z$. We consider the following subspaces of ${\mathfrak{h}}^1$ $$\left\{\begin{array}{l} {\mathfrak{u}}={\mbox{span}}\{X,Y\} \\ {\mathfrak{w}}={\mbox{span}}\{Y+\frac{1}{2}Z\} \end{array}\right..$$ It is immediate to check that ${\mathfrak{u}}\oplus{\mathfrak{w}}={\mathfrak{h}}^1$, but $\exp{\mathfrak{u}}\;\exp{\mathfrak{w}}\neq{\mathbb{H}}^1$, since $$\exp(-X+\lambda Z)\notin \exp{\mathfrak{u}}\;\exp{\mathfrak{w}}$$ for every $\lambda\neq0$. ]{} Even the decomposition of ${\mathcal{G}}$ as the direct sum of two subalgebras does not imply that their corresponding subgroups factorize ${\mathbb{G}}$, as we show in the next [We consider the Heisenberg algebra ${\mathfrak{h}}^2$ with basis $(X_1,Y_1,X_2,Y_2,Z)$ and bracket relations $[X_1,Y_1]=[X_2,Y_2]=Z$. We consider the following subalgebras of ${\mathfrak{h}}^2$ $$\left\{\begin{array}{l} {\mathfrak{a}}={\mbox{span}}\{X_1,X_2,Z+Y_1\} \\ {\mathfrak{b}}={\mbox{span}}\{Y_1,Y_2\} \end{array}\right..$$ It is immediate to check that ${\mathfrak{a}}\oplus{\mathfrak{b}}={\mathfrak{h}}^2$, but $$\exp(2X_1+Z)\notin AB,$$ where $A=\exp{\mathfrak{a}}$ and $B=\exp{\mathfrak{b}}$ are the corresponding Lie subgroups. ]{} h-epimorphisms and h-monomorphisms {#hepimono} ---------------------------------- In this subsection we use complementary subgroups to characterize both h-epimorphisms and h-monomorphisms. These characterizations will be used in the proof of both the implicit function theorem and the rank theorem. \[GlinH\] Let $L:{\mathbb{G}}{\longrightarrow}{\mathbb{M}}$ be a surjective h-homomorphism and let $N$ be its kernel. The conditions 1. there exists a subgroup $H$ complementary to $N$ 2. $L$ is an h-epimorphism are equivalent and if one of them holds, then the restriction $L_{|H}$ is an h-isomorphism. [Proof.]{} We first assume the validity of the first condition. Then we consider the restriction $T=L_{|H}:H{\longrightarrow}{\mathbb{M}}$. Let $m\in{\mathbb{M}}$ and let $g\in{\mathbb{G}}$ such that $L(g)=m$. Then the property $NH={\mathbb{G}}$ implies that $g=nh$, where $(n,h)\in N\times H$. As a consequence, we have $$L(nh)=L(h)=T(h)=m,$$ hence $T$ is surjective. If we have $T(h)=e$, then $$h\in H\cap N=\{e\}.$$ We have shown that $L_{|H}$ is an h-homomorphism. Clearly, $T^{-1}:{\mathbb{M}}{\longrightarrow}H$ is an h-homomorphism and satisfies $L\circ T^{-1}={\mbox{Id}}_{\mathbb{M}}$, hence $L$ is an h-epimorphism. Conversely, if $L$ is an h-epimorphism, then there exists a right inverse $R:{\mathbb{M}}{\longrightarrow}{\mathbb{G}}$ that is also an h-homomorphism. We set $H=R({\mathbb{M}})$ and easily observe that $H\cap N=\{e\}$. Let $g\in{\mathbb{G}}$ and consider $$m=L(g)=L\big(R(m)\big),$$ therefore $g^{-1}R(m)\in N$ and this implies that $g\in HN$. It follows that ${\mathbb{G}}=HN=NH$ and this concludes the proof. $\Box$ \[GlinHmono\] Let $T:{\mathbb{G}}{\longrightarrow}{\mathbb{M}}$ be an injective h-homomorphism and let $H$ be its image. The conditions 1. there exists a normal subgroup $N$ complementary to $H$, 2. there exists an h-epimorphism $p:{\mathbb{M}}{\longrightarrow}H$ such that $p_{|H}={\mbox{Id}}_H$, 3. $T$ is an h-monomorphism are equivalent. [Proof.]{} We show that (1) implies (2). We define the projection $p:{\mathbb{M}}{\longrightarrow}H$ that associates to any element $m=nh\in{\mathbb{M}}$, with $(n,h)\in N\times H$ the element $h\in H$. This definition is well posed, since $N$ and $H$ are complementary. The fact that $N$ is normal and the uniqueness of representation of the product $nh$ give $$p\big((n_1h_1)(n_2h_2)\big)=p\big((n_1h_1n_2h_1^{-1})(h_1h_2)\big)=h_1h_2 =p(n_1h_1)\,p(n_2h_2)\,.$$ It is trivial to observe that $p$ is homogeneous and that its restriction to $H$ is exactly ${\mbox{Id}}_H$. Then $p$ is a surjective h-homomorphism. Furthermore, $H$ is a complementary subgroup of the kernel $N$, then Proposition \[GlinH\] implies that $p$ is an h-epimorphism. To prove that (2) implies (3), we define the mapping $L=J\circ p$, where the injectivity of $T$ allows us to define the h-isomorphism $J:H{\longrightarrow}{\mathbb{G}}$ such that $J\big(T(g)\big)=g$ for every $g\in{\mathbb{G}}$. Then $L$ is an h-homomorphism as composition of h-homomorphisms. In addition, one can easily verify that $L\circ T={\mbox{Id}}_{\mathbb{G}}$. We are left to show that (3) implies (1). Bt definition of h-monomorphism we have an h-homomorphism $L:{\mathbb{M}}{\longrightarrow}{\mathbb{G}}$ that is a left inverse of $T$. Let $N$ be its kernel, that is a normal homogeneous subgroup of ${\mathbb{M}}$. We have to show that $N$ and $H$ are complementary subgroups. Let $m\in{\mathbb{M}}$ and consider $L(m)=g\in{\mathbb{G}}$. We have $$L\big(m\,T(g^{-1})\big)=L(m)g^{-1}=e\,,$$ then $m\,T(g^{-1})=n\in N$, namely, $m=n\,T(g)\in NH$. Let $x\in N\cap H$ and let $g\in{\mathbb{G}}$ such that $x=T(g)$. Thus, we get $$e_{\mathbb{G}}=L(x)=L\circ T(g)=g\,,$$ that implies $g=e_{\mathbb{G}}$ and $x=T(e_{\mathbb{G}})=e_{\mathbb{M}}$, namely, $N\cap H=\{e_{\mathbb{M}}\}$. $\Box$ Quotients of graded groups {#hquot} ========================== In this section we show that the group quotient of graded groups is still graded with a natural left invariant and homogeneous distance. We first recall some elementary facts of Lie groups Theory in order to study the relationship between quotients of Lie algebras and quotients of Lie groups. Let $G$ be a real Lie group and let $H$ be a Lie subgroup of $G$. Recall that the quotient $G/H$ has a unique manifold structure that makes the projection $\pi:G{\longrightarrow}G/H$ a smooth mapping. $G/H$ is called homogeneous manifold, see Theorem 3.58 of [@Warn]. If we consider a normal Lie subgroup $N$, then $G/N$ is in addition a Lie group, according to Theorem 3.64 of [@Warn] and in this case $$\pi:G{\longrightarrow}G/N$$ is clearly is a Lie group homomorphism. As a result, by Theorem 3.14 of [@Warn] $$d\pi:{\mathcal{G}}{\longrightarrow}\overline{\mathcal{G}}$$ is a Lie algebra homomorphism, where ${\mathcal{G}}$ and $\overline{\mathcal{G}}$ are the Lie algebras of $G$ and $G/N$, respectively. [The mapping $d\pi$ is also surjective. In fact, Theorem 3.58 ensures the existence of a neighbourhood $W\subset G/N$ of the unit element $\overline{e}$ of $G/N$ and a smooth mapping $\tau:W{\longrightarrow}G$ such that $\pi\circ\tau={\mbox{Id}}_W$. Let $\Gamma:]-{\varepsilon},{\varepsilon}[{\longrightarrow}W$ be a smooth curve with $\dot\Gamma(0)=\overline{X}_{\overline e}\in T_{\overline e}\big(G/N\big)$. Then the curve $\gamma=\tau\circ\Gamma:]-{\varepsilon},{\varepsilon}[{\longrightarrow}G$ satisfies $\pi\circ\gamma=\Gamma$ and $d\pi(X_e)=\overline{X}_{\overline e}$, where $X_e=\dot\gamma(0)$. This shows surjectivity of $d\pi$. ]{} \[algquot\] The Lie algebra $\overline{{\mathcal{G}}}$ of $G/N$ is isomophic to the quotient algebra ${\mathcal{G}}/{\mathcal{N}}$, where ${\mathcal{G}}$ is the Lie algebra of $G$ and ${\mathcal{N}}$ is the ideal corresponding to the normal subgroup $N$. Furthermore, we can represent the exponential mapping of $G/N$ on ${\mathcal{G}}/{\mathcal{N}}$ by setting ${\mbox{\rm Exp}}:{\mathcal{G}}/{\mathcal{N}}{\longrightarrow}G/N$ and $${\mbox{\rm Exp}}(X+{\mathcal{N}})=\pi\big(\exp(X)\big)$$ where $\exp:{\mathcal{G}}{\longrightarrow}G$ is the exponential mapping of $G$, $\pi:G{\longrightarrow}G/N$ is the canonical projection and the diagram $$\xymatrix@C=30pt@R=30pt{{\mathcal{G}}/{\mathcal{N}}\ar_{{\rm Exp}}[rd] \ar^{\overline{d\pi}}[r] & \overline{\mathcal{G}}\ar^{\overline{\exp}}[d] \\ & G/N }$$ commutes, where $\overline{\exp}:\overline{{\mathcal{G}}}{\longrightarrow}G/N$ is the canonical exponential mapping of $G/N$. [Proof.]{} We have seen above that $d\pi:{\mathcal{G}}{\longrightarrow}\overline{{\mathcal{G}}}$ is surjective, hence $$\dim\big(\ker(d\pi)\big)=\dim({\mathcal{N}}).$$ For every $U\in{\mathcal{N}}$ the curve $\pi\big(\exp(tX)\big)$ is constantly equal to the unit element $\overline e$ of $G/N$, then $d\pi(X)=0$ and clearly ${\mathcal{N}}\subset\ker d\pi$. It follows that ${\mathcal{N}}=\ker d\pi$ and it is well defined the algebra isomorphism $\overline{d\pi}:{\mathcal{G}}/{\mathcal{N}}{\longrightarrow}\overline{{\mathcal{G}}}$, where $p:{\mathcal{G}}{\longrightarrow}{\mathcal{G}}/{\mathcal{N}}$ is the canonical projection and $$[p(X),p(Y)]=p\big([X,Y]\big)$$ for every $X,Y\in{\mathcal{G}}$. If we consider the diagram $$\begin{aligned} \label{comdiag} \xymatrix@C=30pt@R=30pt{{\mathcal{G}}\ar_{p}[d] \ar^{d\pi}[rd] & \\ {\mathcal{G}}/{\mathcal{N}}\ar_{{\rm Exp}}[rd] \ar^{\overline{d\pi}}[r] & \overline{\mathcal{G}}\ar^{\overline{\exp}}[d] \\ & G/N }\end{aligned}$$ then we are left to show that its lower part commutes. In fact, by definition of $\overline{d\pi}$, for every $X\in{\mathcal{G}}$ we have $$\begin{aligned} \overline{\exp}\circ\overline{d\pi}\big(p(X)\big)= \overline{\exp}\circ d\pi(X).\end{aligned}$$ On the other hand, due to Theorem 3.32 of [@Warn], the diagram $$\xymatrix@C=30pt@R=30pt{G \ar^{\pi}[r] & G/N \\ {\mathcal{G}}\ar^{\exp}[u] \ar^{d\pi}[r] & \overline{{\mathcal{G}}} \ar_{\overline{\exp}}[u] }$$ commutes, namely $\overline{\exp}\circ d\pi(X)=\pi\circ\exp(X)$ for every $X\in{\mathcal{G}}$. It follows that $$\overline{\exp}\circ\overline{d\pi}=\pi\circ\exp={\mbox{\rm Exp}}.$$ This concludes the proof. $\Box$ \[quotgrad\] Let ${\mathbb{M}}$ be a graded group of algebra ${\mathcal{M}}=W_1\oplus\cdots\oplus W_\upsilon$ and let $N$ be a normal homogeneous subgroup of corresponding ideal ${\mathcal{N}}={\mathcal{N}}_1\oplus\cdots\oplus{\mathcal{N}}_\upsilon$. Then the surjective Lie algebra homomorphism $d\pi:{\mathcal{M}}{\longrightarrow}\overline{\mathcal{M}}$ induces a grading $$\overline W_1\oplus\cdots\oplus \overline W_\upsilon$$ on the Lie algebra $\overline{\mathcal{M}}$ of ${\mathbb{M}}/N$, that is h-isomorphic to $$W_1/{\mathcal{N}}_1\oplus\cdots\oplus W_\upsilon/{\mathcal{N}}_\upsilon\,,$$ where $d\pi(W_i)=\overline W_i$ is linearly isomorphic to $W_i/{\mathcal{N}}_i$. Furthermore, a one parameter group of dilations $\overline\delta_r$ can be defined on ${\mathbb{M}}/N$ such that $\pi:{\mathbb{M}}{\longrightarrow}{\mathbb{M}}/N$ is an h-homomorphism. Finally, if ${\mathbb{M}}$ is stratified then so is ${\mathbb{M}}/N$. [Proof.]{} The decomposition of ${\mathcal{N}}$ into factors ${\mathcal{N}}_i$ follows from Corollary \[homsub\], where some of ${\mathcal{N}}_i$’s are possibly vanishing. By Proposition \[algquot\] the Lie algebra $\overline{\mathcal{M}}$ of the quotient ${\mathbb{M}}/N$ is isomorphic to ${\mathcal{M}}/{\mathcal{N}}$ and $$p:{\mathcal{M}}{\longrightarrow}{\mathcal{M}}/{\mathcal{N}}\qquad\mbox{is equivalent to}\qquad d\pi:{\mathcal{M}}{\longrightarrow}\overline{\mathcal{M}}\,,$$ by commutativity of diagram , hence we consider the grading induced by $p$. Let us check that $p(W_i)\cap p(W_j)=\{0\}$ whenever $i\neq j$. By contradiction, let $\overline X\in p(W_i)\cap p(W_j)$ be nonvanishing. Then there exist $X_i\in W_i{\setminus}{\mathcal{N}}$ and $X_j\in W_j{\setminus}{\mathcal{N}}$ such that $X_i=X_j+Z$, where $Z\in{\mathcal{N}}$. Since ${\mathcal{N}}$ is homogeneous, we must have $Z=Z_i+Z_j$, where $Z_i\in W_i\cap{\mathcal{N}}$ and $Z_j\in W_j\cap{\mathcal{N}}$, hence $Z_i=X_i$ and $Z_j=-X_j$ this conflicts with our initial assumption on $X_i$ and $X_j$. This shows that $\overline{\mathcal{M}}=\overline W_1\oplus\cdots\oplus\overline W_\upsilon$ is a grading, where $\overline W_i=d\pi(W_i)$. This immediately implies that $\overline {\mathcal{M}}$ is stratified when so is ${\mathcal{M}}$. If we set $\overline\delta_r(\overline X_i)=r^i\overline X_i$ for every $\overline X_i\in\overline W_i$, then one can easily check that $d\pi\circ\delta_r=\overline\delta_r\circ d\pi$ and $\pi$ is an h-homomorphism. Finally, we consider the following commutative diagram $$\xymatrix@C=30pt@R=30pt{{\mathcal{M}}\ar_{p_0}[d] \ar^{p}[r] & {\mathcal{M}}/{\mathcal{N}}\\ W_1/{\mathcal{N}}_1\oplus\cdots\oplus W_\upsilon/{\mathcal{N}}_\upsilon \ar^{J}[ru] & }$$ where we have set $$p_0(w_1+\cdots+w_\upsilon)=\big(w_1+{\mathcal{N}}_1,\ldots,w_\upsilon+{\mathcal{N}}_\upsilon\big)$$ and the following definition $$J\big(w_1+{\mathcal{N}}_1,\ldots,w_\upsilon+{\mathcal{N}}_\upsilon\big)= w_1+\cdots+ w_\upsilon+{\mathcal{N}}$$ is well posed. It is immediate to observe that $\ker J=\{0\}$, then $J$ is a linear isomorphism. The commutativity of the diagram above implies that $$p(W_i)=J\big(W_i/{\mathcal{N}}_i\big)\,.$$ This concludes the proof. $\Box$ [ The induced grading on ${\mathcal{M}}/{\mathcal{N}}$ considered on Proposition \[quotgrad\] is given by the subspaces $\overline W_j=d\pi(W_j)$, that satisfy ${\mathcal{M}}/{\mathcal{N}}=\overline W_1\oplus\cdots\oplus\overline W_\upsilon$, where some of these factors are possibly vanishing. On the other hand, when ${\mathcal{N}}$ is not homogeneous ${\mathcal{M}}/{\mathcal{N}}$ still can have a grading with dilations, but these ones do not commute with $\pi$, namely, $\pi$ is no longer an h-homomorphism. ]{} [Let $W_1={\mbox{span}}\{X_1,X_2,X_3\}$ and $W_2={\mbox{span}}\{Z\}$ where the only nontrivial bracket relation is $[X_1,X_2]=Z$. ${\mathcal{M}}=W_1\oplus W_2$ is a 2-step stratified algebra and ${\mathcal{N}}={\mbox{span}}\{X_3-Z\}$ is an ideal of ${\mathcal{M}}$ that is clearly not homogeneous. Elements of ${\mathcal{M}}/{\mathcal{N}}$ are $\overline X_i=X_i+{\mathcal{N}}$, where $i=1,2,3$ and $\overline X_3=Z+{\mathcal{N}}$. Clearly, the only nontrivial bracket relation is $[\overline X_1,\overline X_2]=\overline X_3$ and ${\mathcal{M}}/{\mathcal{N}}$ is isomorphic to the 3-dimensional Heisenberg algebra. On the other hand, there are no dilations $\overline\delta_r$ on ${\mathcal{M}}/{\mathcal{N}}$ such that $d\pi\circ\delta_r=\overline\delta_r\circ d\pi$, since $$d\pi\big(\delta_rX_3\big)=r\,d\pi(X_3)=r\,\overline X_3\qquad \mbox{and}\qquad d\pi\big(\delta_rZ\big)=r^2\,d\pi(Z)=r^2\,\overline X_3.$$ We also notice that in this case $d\pi$ cannot induce any grading on $\overline{\mathcal{M}}$, since $d\pi(W_1)\cap d\pi(W_2)=\{\overline X_3\}$. ]{} Implicit Function Theorem and Rank Theorem {#MProofs} ========================================== In this section we prove both the implicit function theorem and the rank theorem stated in the introduction. We will use the following notation $D_{\overline{n},r}^N=D_{\overline{n},r}\cap N$ and $D_{\overline{h},s}^H=D_{\overline{h},s}\cap H$. .2truecm [Proof of Theorem \[implth\].]{} Our arguments are divided into two main steps. [**Step 1**]{}: [*Existence.*]{} Due to Proposition \[GlinH\], the restriction $Df(\overline{x})(h):H{\longrightarrow}{\mathbb{G}}$ is invertible, then $$\min_{\substack{d(h)=1\\ h\in H}} \rho{\mbox{\large $($}}Df(\overline{x})(h){\mbox{\large $)$}}>0.$$ By continuity of $ \Omega\times H\ni (x,h){\longrightarrow}\rho{\mbox{\large $($}}Df(x)(h){\mbox{\large $)$}}, $ there exists $R>0$ such that $$\label{mumu} \mu=\min_{x\in D_{\overline{x},R}}\min_{\substack{d(h)=1\\ h\in H}} \rho{\mbox{\large $($}}Df(x)(h){\mbox{\large $)$}}>0\quad\mbox{and}\quad \overline{\mu}=\max_{x\in D_{\overline{x},R}}\max_{\substack{d(h)=1\\ h\in H}} \rho{\mbox{\large $($}}Df(x)(h){\mbox{\large $)$}}>0$$ with $D_{\overline{x},R}\subset\Omega$. Now we set $R_0=R/(2+2\,c({\mathbb{G}},d)\,N)$, where $c({\mathbb{G}},d)$ and $N$ are as and Lemma \[140FS\], respectively. Then condition is satisfied and Theorem \[unifestim\] yields the estimate $$\label{applunifestim} \max_{x,y\in D_{\overline{x},R_0}} \rho{\mbox{\large $($}}f(x)^{-1}f(y),Df(x)(x^{-1}y){\mbox{\large $)$}}\leq C\;\;\big[\omega_{D_{\overline{x},R},df}\big(N\,c\,d(x,y)\big) \big]^{1/\iota^2}\;d(x,y)\,.$$ Thus, by definition of $\overline{\mu}$ in we have $$\begin{aligned} \label{frabove} \rho\big(f(x),f(y)\big)\leq \left(C\big[\omega_{D_{\overline{x},R},df}\big(N\,c\,d(x,y)\big) \big]^{1/\iota^2}+\overline{\mu}\right)\;d(x,y)\end{aligned}$$ for every $x,y\in D_{\overline{x},R_0}$. Let $(\overline{n},\overline{h})$ be the unique couple of $N\times H$ such that $\overline{n}\overline{h}=\overline{x}$ and let $r,s>0$ be such that $D^N_{\overline{n},r}D^H_{\overline{h},s}\subset D_{\overline{x},R_0}$. Then, using definition of $\mu$ in , for every $n\in D^N_{\overline{n},r}$ and every $h,h'\in D^H_{\overline{h},s}$ there holds $$\begin{aligned} \label{frbelow} \rho\big(f(nh),f(nh')\big))\geq \left(\mu-C\big[\omega_{D_{\overline{x},R},df}\big(N\,c\,d(h,h')\big) \big]^{1/\iota^2}\right)\;d(h,h')\end{aligned}$$ observing that $d(nh,nh')=d(h,h')$. Using and and possibly taking a smaller $s>0$ depending on $C, N, c>0$ we get a constant $\beta>0$ such that $$\begin{aligned} \label{biLip} \beta^{-1}\;d(h,h')\leq\rho\big(F_n(h),F_n(h')\big)\leq\beta\;d(h,h')\end{aligned}$$ where for each $n\in D^N_{\overline{n},r}$ we have defined $$F_n:D^H_{\overline{h},s}{\longrightarrow}{\mathbb{M}},\qquad F_n(h)=f(nh).$$ Observing that $F_{\overline{n}}(\overline{h})=f(\overline{x})$ and taking into account the biLipschitz estimate we have $$F_{\overline{n}}(h)\neq f(\overline{x})\quad\mbox{for every}\quad h\in\partial D^H_{\overline{h},s}\,.$$ Injectivity of $F_{\overline{n}}$ implies that $$\deg\big(F_{\overline{n}},B^H_{\overline{h},s},f(\overline{x})\big) \in\{-1,1\},$$ see for instance Theorem 3.3.3 of [@LLoyd]. By continuity of $$n{\longrightarrow}\max_{h\in\partial D^H_{\overline{h},s}} \rho\big(f(nh),f(\overline{n}h)\big)\,,$$ up to choosing a smaller $r>0$, we can assume that $$\begin{aligned} \max_{h\in\partial D^H_{\overline{h},s}} \rho\big(f(nh),f(\overline{n}h)\big)<\frac{s}{2\beta}\end{aligned}$$ for every $n\in D^N_{\overline{n},r}$. As a consequence, applying , it follows that $$\rho\big(F_n(h),f(\overline{x})\big)\geq \rho\big(F_{\overline{n}}(h),f(\overline{x})\big) -\rho\big(F_{\overline{n}}(h),F_n(h)\big)>\frac{s}{2\beta}$$ for every $h\in\partial D^H_{\overline{h},s}$ and every $n\in D^N_{\overline{n},r}$. For an arbitrary $n\in D^N_{\overline{n},r}{\setminus}\{\overline{n}\}$ one can consider the continuous curve $\gamma:[0,1]{\longrightarrow}D^N_{\overline{n},r}$, defined by $\gamma(t)=\overline{n}\delta_t\big((\overline{n})^{-1}n\big)$. Notice that $\gamma$ has image in $D^N_{\overline{n},r}$, since $N$ is a homogeneous subgroup of ${\mathbb{G}}$. By previous estimates, the mapping $\Phi:[0,1]\times D_{\overline{h},s}{\longrightarrow}{\mathbb{M}}$ defined by $$\Phi(t,h)=f\big(\gamma(t)h\big)$$ is a homotopy between $F_{\overline{n}}$ and $F_n$ such that $\Phi(t,h)\neq f(\overline{x})$ for every $t\in[0,1]$ and every $h\in\partial D^H_{\overline{h},s}$. Thus, homotopy invariance of topological degree, [@JTSch], implies $$\deg\big(F_n,B^H_{\overline{h},s},f(\overline{x})\big)= \deg\big(F_{\overline{n}},B^H_{\overline{h},s},f(\overline{x})\big)\neq0,$$ hence there exists at least one element $h'\in D^H_{\overline{h},s}$, depending on $n$, such that $F_n(h')=f(\overline{x})$. Injectivity of $F_n$ gives uniqueness of $h'$, hence there exists ${\varphi}: D^N_{\overline{n},r}{\longrightarrow}D^H_{\overline{h},s}$, uniquely defined, such that $$F_n\big({\varphi}(n)\big)=f\big(n\,{\varphi}(n)\big)=f(\overline{x}).$$ This concludes the proof of the first step. [**Step 2**]{}: [*Regularity.*]{} We keep the same notation of the previous step. By definition of $\mu>0$ in the restriction $L(x)=\big(Df\big(x)\big)_{|H}:H{\longrightarrow}{\mathbb{M}}$ is invertible for every $x\in D_{\overline{x},R_0}$ and setting $T(x)=L(x)^{-1}$, we have $$\begin{aligned} \label{unifinvest} d\big(T(x)(m)\big)\leq \mu^{-1}\;\rho(m)\end{aligned}$$ for every $m\in{\mathbb{M}}$. Due to , we can choose a possibly smaller $R>0$, hence a smaller $R_0=R/(2+2cN)$ such that $$\begin{aligned} \max_{x,y\in D_{\overline{x},R_0}} \rho{\mbox{\large $($}}f(x)^{-1}f(y),Df(x)(x^{-1}y){\mbox{\large $)$}}\leq\frac{\mu\;d(x,y)}{2}\,.\end{aligned}$$ It follows that the remainder $E(n,n')$ in the expression $$\begin{aligned} f\big(n{\varphi}(n)\big)^{-1}f(n{\varphi}(n')) =L\big(n{\varphi}(n)\big)\big({\varphi}(n)^{-1}{\varphi}(n')\big)E(n,n')\end{aligned}$$ satisfies the uniform estimate $$\begin{aligned} \label{appl1unifestim} \rho\big(E(n,n')\big)\leq\frac{\mu}{2}\;d\big({\varphi}(n),{\varphi}(n')\big)\end{aligned}$$ for every $n,n'\in D^N_{\overline{n},r}$. The implicit mapping ${\varphi}$ satisfies $$f\big(n{\varphi}(n)\big)^{-1}f(n{\varphi}(n')\big) =\left(f\big(n{\varphi}(n')\big)^{-1}f\big(n'{\varphi}(n')\big)\right)^{-1},$$ therefore one easily gets $${\varphi}(n)^{-1}{\varphi}(n')=T\big(n{\varphi}(n)\big)\left(f\big(n{\varphi}(n')\big)^{-1}f\big(n'{\varphi}(n')\big)\right)^{-1}\,T\big(n{\varphi}(n)\big)\big(E(n,n')\big)^{-1}.$$ As a result, in view of and , we obtain $$\begin{aligned} d\big({\varphi}(n),{\varphi}(n')\big)\leq \frac{2}{\mu}\; \rho\left(f\big(n{\varphi}(n')\big),f\big(n'{\varphi}(n')\big)\right).\end{aligned}$$ By , up to choosing a possibly smaller $R>0$, we can suppose that $$\rho\big(f(x),f(y)\big)\leq 2\,\overline{\mu}\,\,d(x,y)$$ for every $x,y\in D_{\overline{x},R_0}$, hence $$\begin{aligned} d\big({\varphi}(n),{\varphi}(n')\big)\leq \frac{4\,\overline{\mu}}{\mu}\; d\big({\varphi}(n')^{-1}n^{-1}n'{\varphi}(n')\big).\end{aligned}$$ Finally, formulae and lead us to the conclusion. $\Box$ [Under hypotheses of Theorem \[implth\], one immediately gets $$\begin{aligned} \label{holdergraph} d\big(n{\varphi}(n),n'{\varphi}(n')\big)\leq(1+\kappa)\; d\big({\varphi}(n')^{-1}n^{-1}n'{\varphi}(n')\big)\,,\end{aligned}$$ where $\kappa$ is as in the statement of the above mentioned theorem. ]{} .2truecm [Proof of Theorem \[Pembed\].]{} By Proposition \[GlinHmono\] we have a projection $p:{\mathbb{M}}{\longrightarrow}H$ that is an h-epimorphism and $p_{|H}={\mbox{Id}}_H$. We also know that the kernel $N$ of $p$ is complementary to $H$. As a first consequence, the mapping $p\,\circ f:\Omega{\longrightarrow}H$ is continuously P-differentiable and $D(p\,\circ f)(\overline{x}):{\mathbb{G}}{\longrightarrow}H$ is invertible. Due to Theorem \[inverse\], there exist open neighbourhoods $V\Subset\Omega$ and $W\Subset H$ of $\overline{x}$ and of $p\big(f(\overline{x})\big)$, respectively, such that the restriction $$(p\circ f)_{|\overline{V}}:\overline{V}{\longrightarrow}\overline{W}$$ is invertible. We denote by $\psi:\overline{W}{\longrightarrow}\overline{V}$ its inverse function. By Remark \[uniqued\], there exists a unique function $f_N:\Omega{\longrightarrow}N$ such that $f(x)=p\big(f(x)\big)f_N(x)$ for every $x\in\Omega$, hence we consider $g:\overline{W}{\longrightarrow}{\mathbb{M}}$ defined by $$g(h)=f\big(\psi(h)\big)=h\,f_N\big(\psi(h)\big)$$ for every $h\in\overline{W}$. Uniqueness of factorization implies that $$g(W')=g(W')\cap W'N,$$ whenever $W'\subset\overline{W}$. If $W'$ is open, then so is $W'N$, since the mapping is open. As a consequence, $g$ is an open mapping on its image, hence so is the restriction $f_{|V}$. This implies that $f_{|V}$ is a topological embedding. If we set $${\varphi}=f_N\circ\psi\quad\mbox{and}\quad\Psi=I\circ J^{-1}\circ\psi\circ p_{|f(V)}$$ then ${\varphi}(h)=h^{-1}g(h)$ and holds, where $I$ and $J$ are defined in the statement of Theorem \[Pembed\]. We notice that $g$ is continuously P-differentiable on $\overline{W}$, then in particular it is Lipschitz on $\overline{W}$ up to a suitable choice of $W$, for instance a closed ball. Our next computations are performed in the Lie algebra ${\mathcal{G}}$ of ${\mathbb{G}}$. We set ${\varphi}=\exp\circ F$ and $g=\exp\circ G$, therefore Lemma \[estimi\] gives a constant $C>0$, depending on both $\overline{W}$ and the Lipschitz constant of $g$ such that $$\begin{aligned} \label{lipesth} \|G(h)-G(h')\|\leq C\; d(h,h')\quad\mbox{and}\quad \|\xi-\xi'\|\leq C\;d(h,h'),\end{aligned}$$ where $h=\exp\xi$ and $h'=\exp\xi'$, with $h,h'\in\overline{W}.$ We have $$\begin{aligned} &&F(h)-F(h')=(-\xi){\circledcirc}G(h)-\big((-\xi'){\circledcirc}G(h')\big)\\ &&=\xi'-\xi+G(h)-G(h') +\sum_{n=2}^\iota \left[c_n\big(-\xi,G(h)\big)-c_n\big(-\xi',G(h')\big)\right],\end{aligned}$$ due to the formula . From , it follows that $$\|F(h)-F(h')\|\leq\|\xi'-\xi\|+\|G(h)-G(h')\| +\sum_{n=2}^\iota \gamma_n\,\nu^{n-1} \big(\|\xi-\xi'\|+\|G(h)-G(h')\|\big),\nonumber$$ where we have set $$\nu=2\,\Big(\max_{\eta\in\overline{W}}\|G(\eta)\|+\sup_{\eta\in\overline{W}}\|\eta\|\Big).$$ By virtue of , we get a constant $C_1>0$, depending on $\overline{W}$, $\nu$ and the Lipschitz constant of $g$ such that $$\|F(h)-F(h')\|\leq C_1\,d(h,h').$$ This concludes the proof. $\Box$ Intrinsic graphs and $({\mathbb{G}},{\mathbb{M}})$-regular sets {#GMregsets} =============================================================== The notion of complementary subgroup allows us to set properly the well established notion of intrinsic graph in those graded groups that admit a factorization by two complementary subgroups. \[intrgraph\][Let $P$ and $H$ be complementary subgroups of ${\mathbb{G}}$ and let $S\subset{\mathbb{G}}$. We say that $S$ is an [*intrinsic graph*]{} with respect to $(P,H)$ if there exists a subset $A\subset P$ and a mapping ${\varphi}:A{\longrightarrow}H$ such that $$S=\{p\,{\varphi}(p)\mid p\in A\}.$$ ]{} \[trintrgr\][As already observed in [@FSSC6], intrinsic graphs are preserved under left translations. If $P$ and $H$ are complementary subgroups of ${\mathbb{G}}$ and $S=\{p\,{\varphi}(p)\mid p\in A\}$ is an intrinsic graph with respect to $(P,H)$, then the left translated $xp\,{\varphi}(p)$, for some $x\in{\mathbb{G}}$, is of the form $\tilde p\,\psi(\tilde p)$ where $\tilde p\in\tilde A\subset P$ and $\psi:\tilde A{\longrightarrow}H$. In fact, by unique decomposition of elements in ${\mathbb{G}}$ with respect to $P$ and $H$, see Remark \[uniqued\], there exist functions $\psi_1:A{\longrightarrow}P$ and $\psi_2:A{\longrightarrow}H$ such that $$xp\,{\varphi}(p)=\psi_1(p)\psi_2(p)\,.$$ To conclude, it suffices to show that $\psi_1$ is injective. If $\psi_1(p)=\psi_1(p')$, then $$xp\,{\varphi}(p)\psi_2(p)^{-1}=xp'{\varphi}(p')\psi_2(p')^{-1},$$ where ${\varphi}(p)\psi_2(p)^{-1},{\varphi}(p')\psi_2(p')^{-1}\in H$. Again, uniqueness yields $p=p'$. ]{} \[gmsing\][Let ${\mathbb{G}}$ be a stratified group and let ${\mathbb{M}}$ be a graded group such that ${\mathbb{M}}$ is an h-quotient of ${\mathbb{G}}$. We say that a subset $S\subset{\mathbb{G}}$ is [*$({\mathbb{G}},{\mathbb{M}})$-regular*]{} if for every point $\overline{x}\in S$, there exists an open neighbourhood $U$ of $\overline{x}$ and a continuously P-differentiable mapping $f:U{\longrightarrow}{\mathbb{M}}$ such that $S\cap U=f^{-1}(e)$ and $Df(x):{\mathbb{G}}{\longrightarrow}{\mathbb{M}}$ is an h-epimorphism for every $x\in U$. ]{} \[gmsinm\][Let ${\mathbb{G}}$ be a stratified group and let ${\mathbb{M}}$ be a graded group such that ${\mathbb{G}}$ h-embeds into ${\mathbb{M}}$. A subset $S\subset{\mathbb{M}}$ is [*$({\mathbb{G}},{\mathbb{M}})$-regular*]{} if for every point $\overline{x}\in S$, there exist open neighbourhoods $U\subset{\mathbb{M}}$ of $\overline{x}$ and $V\subset{\mathbb{G}}$ of $e\in{\mathbb{G}}$, along with a continuously P-differentiable topological embedding $f:V{\longrightarrow}{\mathbb{M}}$, such that $S\cap U=f(V)$ and $Df(y):{\mathbb{G}}{\longrightarrow}{\mathbb{M}}$ is an h-monomorphism for every $y\in V$. ]{} [When a subset $S$ is either $({\mathbb{G}},{\mathbb{M}})$-regular in ${\mathbb{G}}$ or $({\mathbb{G}},{\mathbb{M}})$-regular in ${\mathbb{M}}$, we simply say that it is $({\mathbb{G}},{\mathbb{M}})$-regular without further specification, or we can say that it is an [*intrinsically regular set*]{}. ]{} Tangent cone to $({\mathbb{G}},{\mathbb{M}})$-regular sets {#intrblw} ---------------------------------------------------------- The notion of “tangent cone" given in 3.1.21 of [@Fed] can be easily extended to graded groups. \[tancone\][Let ${\mathbb{G}}$ be a graded group and let $S\subset{\mathbb{G}}$ with $a\in{\mathbb{G}}$. The [*homogeneous tangent cone*]{} of $S$ at $a$ is the homogeneous subset $$\begin{aligned} &&\mbox{Tan}(S,a)=\Big\{v\in{\mathbb{G}}\mid v=\lim_{k{\rightarrow}\infty}\delta_{r_k}(a^{-1}s_k),\, \mbox{for some sequences}\; (r_k)\subset{\mathbb{R}}^+,\\ &&(s_k)\subset S,\;\mbox{where}\;s_k{\rightarrow}a\Big\}. \nonumber\end{aligned}$$ ]{} [If $a\in\overline{S}$, then $\mbox{Tan}(S,a)\neq\emptyset$.]{} [Proof of Theorem \[blwlevelimage\].]{} Replacing $f$ with $f\circ l_x$, where $l_x(y)=xy$, it is not restrictive to assume that $\overline{x}=e$, since $$(\overline{x})^{-1}S=(f\circ l_{\overline{x}})^{-1}(f\circ l_{\overline{x}})(e)$$ and $Df(\overline{x})=D(f\circ l_{\overline{x}})(e)$. Now we use the notation of Theorem \[implth\] applying it in the case $\overline{x}=e$. From P-differentiability of $f$ at $e$ and representation , we have $$f(e)=f(n{\varphi}(n))=f(e) Df(e)\big({\varphi}(n)\big)o\big(n{\varphi}(n)\big).$$ Setting $L=Df(e)$, we obtain $$\begin{aligned} d\big({\varphi}(n)\big)= d\left(L_{|H}^{-1}\big(o(n{\varphi}(n))\big)\right)\leq {\varepsilon}\big(d(n)+d({\varphi}(n))\big)\end{aligned}$$ for $n$ suitable close to $e$ and ${\varepsilon}>0$ small, arbitrarily fixed. From this, we easily conclude that ${\varphi}$ is P-differentiable at $e_N$ and $D{\varphi}(e_N)$ is the null h-homomorphism. Now, we arbitrarily fix $R>0$ and consider $\lambda>0$ sufficiently small, such that $$D_R\cap\delta_{1/\lambda}S=D_R\cap \{\delta_{1/\lambda}n\delta_{1/\lambda}{\varphi}(n)\mid n\in D_r^N\}.$$ Let $(\lambda_k)$ be an infinitesimal sequence of positive numbers. Using Proposition 4.5.5 of [@AmbTil], we have to prove the following conditions: - if $x=\lim_{k\to\infty} x_k$ for some sequence $(x_k)$ such that $x_k\in D_R\cap\delta_{1/\lambda_k}S$, then we have $x\in D_R\cap N$; - if $x\in D_R\cap N$, then there exists a sequence $(x_k)$ such that $x_k\in D_R\cap\delta_{1/\lambda_k}S$ and $x_k\to x$. To prove (i), we write $x_k=\big(\delta_{1/\lambda_k}n_k\big)\big(\delta_{1/\lambda_k}{\varphi}(n_k)\big)$ and use Proposition \[homdec\] with respect to our complementary subgroups $N$ and $H$. Through the global diffeomorphism $\phi$ in , convergence of the sequence $(x_k)$ implies convergence of both $(\delta_{1/\lambda_k}n_k)$ and $\big(\delta_{1/\lambda_k}{\varphi}(n_k)\big)$. We notice that the limit $n_0$ of $(\delta_{1/\lambda_k}n_k)$ belongs to $N$. P-differentibility of ${\varphi}$ at $e_N$ gives $$\lim_{k{\rightarrow}\infty}\delta_{1/\lambda_k}{\varphi}(\delta_{\lambda_k}n'_k\big) =D{\varphi}(e_N)(n_0)=e_H$$ where we have set $\delta_{1/\lambda_k}n_k=n'_k$. Then $x_k{\rightarrow}x=n_0\in D_R\cap N$ and our claim is achieved. Now we choose $x\in N\cap D_R$ and consider a sequence $(x_l)$ contained in $N\cap B_R$ converging to $x$. We observe that for every fixed $l$, there exists a sufficiently large $k_l$ such that $\delta_{\lambda_{k_l}}x_l\in D^N_r$ and $$\delta_{1/\lambda_{k_l}} \big(\delta_{\lambda_{k_l}} x_l{\varphi}(\delta_{\lambda_{k_l}}x_l)\big) \in B_R\cap\delta_{1/\lambda_k}S,$$ since P-differentiability of ${\varphi}$ at $e_N$ implies that the sequence $\delta_{1/\lambda_{k}}\big(\delta_{\lambda_k}x_l {\varphi}(\delta_{\lambda_k}x_l)\big)$ converges to $x_l$ as $k{\rightarrow}\infty$. The fact that $D{\varphi}(e_N)$ is the null h-homomorphism also implies that $$\delta_{1/\lambda_{k_l}} \big(\delta_{\lambda_{k_l}} x_l{\varphi}(\delta_{\lambda_{k_l}}x_l)\big){\longrightarrow}xD{\varphi}(e_N)(x)=x\quad\mbox{as}\quad l{\rightarrow}\infty.$$ Finally, we notice that conditions (i) and (ii) exactly prove that the two inclusions between $N$ and $\mbox{Tan}(S,e)$. This concludes the proof of the first part of Theorem \[blwlevelimage\]. Now, we keep the same notation used in the proof of Theorem \[Pembed\] and observe that it is not restrictive assuming that $f(\overline{x})=e_{\mathbb{M}}$. In fact, in the general case it suffices to replace $f$ with the left translated $x{\rightarrow}f(\overline{x})^{-1}f(x)$, that holds the same assumptions. We know that the mapping $p\circ f_{|V}$ is invertible with P-differentiable inverse $\psi:W{\longrightarrow}V$ and $g(h)=f\circ \psi(h)=h{\varphi}(h)$ for every $h\in W$. By Proposition \[chain\], we have that $Dg(e_H):H\hookrightarrow{\mathbb{M}}$ identically embeds $H$ into ${\mathbb{M}}$, then $$g(h)=h\,o(h)=h\,{\varphi}(h)$$ so that ${\varphi}(h)=o(h)$. This implies that there exists $D{\varphi}(e_H)$ and it is constantly equal to the unit element $e_N\in N$. To show the remaining claims, one can argue exactly as in the first part of this proof, using the fact that ${\varphi}$ is P-differentiable at $e_H$ with null P-differential at this point and replacing $N$ with $H$. $\Box$ \[GMGTan\] Let $S$ be a $({\mathbb{G}},{\mathbb{M}})$-regular set of ${\mathbb{G}}$. Then for every $x\in S$, we have $${\mathcal{H}}\mbox{\rm -dim}\big(\mbox{\rm Tan}(S,x)\big) ={\mathcal{H}}\mbox{\rm -dim}{\mathbb{G}}-{\mathcal{H}}\mbox{\rm -dim}{\mathbb{M}}\,.$$ [Proof.]{} By Theorem \[blwlevelimage\], $\mbox{\rm Tan}(S,x)$ is h-isomorphic to $N=\ker Df(x)$, where $f$ is the continuously P-differentiable mapping defining $S$ around $x$ as a level set. Due to Proposition \[GlinH\], the factorizing property of h-epimorphisms yields a complementary subgroup $H$ that is h-isomorphic to ${\mathbb{M}}$. Due to formula , our claim follows. $\Box$ \[nnhisotan\][The fact that homogeneous tangent cones to a $({\mathbb{G}},{\mathbb{M}})$-regular set of ${\mathbb{G}}$ have a fixed Hausdorff dimension does not mean that they are all algebraically h-isomorphic. Let us consider the mapping $$f:{\mathbb{H}}^2{\longrightarrow}{\mathbb{R}}^2,\qquad (x_1,x_2,x_3,x_4,x_5){\longrightarrow}\Big(\sqrt{x_2^2+x_3^2},x_4\Big)$$ where $p=\exp\big(\sum_{j=1}^5x_jX_j)\in{\mathbb{H}}^2$ and $[X_1,X_2]=[X_3,X_4]=X_5$ are the nontrivial bracket relations of ${\mathbb{H}}^2$. The Pansu differential $Df(x)$ is represented by the matrix $$\left(\begin{array}{ccccc} 0 & \frac{x_2}{\sqrt{x_2^2+x_3^2}} & \frac{x_3}{\sqrt{x_2^2+x_3^2}} & 0 & 0 \\ 0 & 0 & 0 & x_4 & 0 \end{array}\right)\,.$$ Then, defining the connected open set $$\Omega=\left\{(x_1,x_2,x_3,x_4,x_5)\in{\mathbb{H}}^2\mid x_4>0,\;x_2^2+x_3^2>0 \right\}$$ we notice that $S=\Omega\cap f^{-1}\big((1,1)\big)$ is $({\mathbb{H}}^2,{\mathbb{R}}^2)$-regular in ${\mathbb{H}}^2$. Now we fix $\xi=(0,1,0,1,0)$ and $\eta=(0,0,1,1,0)$, observing that $\xi,\eta\in S$. Thus, it is easy to check that $\mbox{Tan}(S,\xi)$ is commutative and $\mbox{Tan}(S,\eta)$ is h-isomorphic to ${\mathbb{H}}^1$. ]{} Factorizing groups {#sectfactor} ================== In this section we investigate the algebraic conditions under which either surjective or injective h-homomorphisms are h-monomorphisms or h-epimorphisms, respectively. Let ${\mathbb{G}}$ and ${\mathbb{M}}$ be arbitrary graded groups with algebras ${\mathcal{G}}$ and ${\mathcal{M}}$. \[hquotembed\][We say that ${\mathbb{M}}$ is an [*h-quotient*]{} of ${\mathbb{G}}$ if there exists a normal homogeneous subgroup $N\subset{\mathbb{G}}$ such that ${\mathbb{G}}/N$ is h-isomorphic to ${\mathbb{M}}$. Analogously, ${\mathbb{G}}$ [*h-embeds*]{} into ${\mathbb{M}}$ if and only if there exists a homogeneous subgroup $H$ of ${\mathbb{M}}$ which is h-isomorphic to ${\mathbb{G}}$. ]{} \[normhsu\] There exists a normal subgroup $N$ of ${\mathbb{G}}$ such that ${\mathbb{G}}/N$ is h-isomorphic to ${\mathbb{M}}$ if and only if there exists a surjective h-homomorphism $L:{\mathbb{G}}{\longrightarrow}{\mathbb{M}}$. [Proof.]{} If the first condition holds, then we denote by $T:{\mathbb{G}}/N{\longrightarrow}{\mathbb{M}}$ the canonical h-isomorphism and notice that $L=T\circ \pi:{\mathbb{G}}{\longrightarrow}{\mathbb{M}}$ is a surjective h-homomrphism, where $\pi:{\mathbb{G}}{\longrightarrow}{\mathbb{G}}/N$ is the canonical projection, that is also an h-homomorphism, according to Proposition \[quotgrad\]. The converse is trivial. $\Box$ \[hhomosub\][The previous proposition shows that ${\mathbb{M}}$ is an [*h-quotient*]{} of ${\mathbb{G}}$ if and only if there exists a surjective h-homomorphism $L:{\mathbb{G}}{\longrightarrow}{\mathbb{M}}$. In view of Proposition \[quotgrad\], we also notice that ${\mathbb{M}}$ is stratified when so is ${\mathbb{G}}$. On the other side, it is easy to check that ${\mathbb{G}}$ h-embeds into ${\mathbb{M}}$ if and only if there exists an injective h-homomorphism $T:{\mathbb{G}}{\longrightarrow}{\mathbb{M}}$. ]{} \[hepim\] If $m$ is the dimension of the first layer of ${\mathcal{G}}$, then ${\mathbb{R}}^k$ is an h-quotient of ${\mathbb{G}}$ if and only if $k\leq m$. [Proof.]{} Let $V_1\oplus V_2\oplus\cdots\oplus V_\iota$ be the direct decomposition of ${\mathcal{G}}$ into its layers. Let ${\mathfrak{u}}$ be an $(m-k)$-dimensional subspace of $V_1$. Then ${\mathcal{N}}={\mathfrak{u}}\oplus V_2\oplus\cdots\oplus V_\iota$ is an ideal and clearly ${\mathcal{G}}/{\mathcal{N}}$ is h-isomorphic to ${\mathbb{R}}^k$. By Proposition \[algquot\], ${\mathbb{G}}/N$ is also h-isomorphic to ${\mathbb{R}}^k$, where $N=\exp{\mathcal{N}}$. Conversely, assume that ${\mathbb{R}}^k$ is an h-quotient of some ${\mathbb{G}}$. From Proposition \[quotgrad\], it follows that there exists $N$ of ideal ${\mathcal{N}}={\mathcal{N}}_1\oplus\cdots\oplus{\mathcal{N}}_\iota$ such that ${\mathbb{R}}^k$ is linearly isomorphic to $V_1/{\mathcal{N}}_1$, then $k\leq\dim V_1=m$. $\Box$ [The Heisenberg group ${\mathbb{H}}^k$ is not an h-quotient of ${\mathbb{H}}^n$, whenever $n>k$. By contradiction, assume that there exists a normal subgroup $N=\exp{\mathfrak{n}}$ such that ${\mathbb{H}}^n/N$ is h-isomorphic to ${\mathbb{H}}^k$. In terms of algebras, we have a $2(n-k)$-dimensional ideal ${\mathfrak{n}}$ of ${\mathfrak{h}}^n$ such that ${\mathfrak{h}}^n/{\mathfrak{n}}$ is h-isomorphic to ${\mathfrak{h}}^k$. This implies that there exists ${\mathfrak{n}}+Z'\in{\mathfrak{h}}^n/{\mathfrak{n}}$ that corresponds to a nonvanishing element of the second layer of ${\mathfrak{h}}^k$. It follows that ${\mathfrak{n}}\subset{\mathfrak{v}}$ and $[{\mathfrak{n}},{\mathfrak{n}}]=\{0\}$, since ${\mathfrak{v}}$ is the first layer of ${\mathfrak{h}}^n$. If we pick an element $X\in{\mathfrak{v}}$ such that $[X,{\mathfrak{n}}]\neq\{0\}$, then we meet a contradiction. In fact, $X\notin{\mathfrak{n}}$, since ${\mathfrak{n}}$ is commutative and this conflicts with the fact that ${\mathfrak{n}}$ is an ideal. ]{} [In view of Remark \[hhomosub\], the previous example shows the nonexistence of surjective h-homomorphisms from ${\mathbb{H}}^n$ to ${\mathbb{H}}^k$ whenever $n>k$, as it was proved in Theorem 2.8 of [@Mag1], by explicit representation of h-homomorphisms from ${\mathbb{H}}^n$ to ${\mathbb{H}}^k$. ]{} \[factorquot\][We say that ${\mathbb{M}}$ [*factorizes ${\mathbb{G}}$ as a quotient*]{} if it is an h-quotient of ${\mathbb{G}}$ and every normal subgroup $N$ of ${\mathbb{G}}$ such that ${\mathbb{G}}/N$ is h-isomorphic to ${\mathbb{M}}$ has a complementary subgroup $H$. ]{} \[factorsubg\][We say that ${\mathbb{G}}$ [*factorizes ${\mathbb{M}}$ as a subgroup*]{} if it h-embeds into ${\mathbb{M}}$ and every subgroup $H$ of ${\mathbb{M}}$ which is h-isomorphic to ${\mathbb{G}}$ has a complementary normal subgroup $N$. ]{} As a corollary of both Proposition \[GlinH\] and Proposition \[GlinHmono\], we have the following \[factorpro\] ${\mathbb{M}}$ factorizes ${\mathbb{G}}$ as a quotient if and only if every surjective h-homomorphism is an h-epimorphism and ${\mathbb{G}}$ factorizes ${\mathbb{M}}$ as a subgroup if and only if every injective h-homomorphism is an h-monomorphism. \[rksub\] If ${\mathbb{R}}^k$ h-embeds into ${\mathbb{G}}$, then it factorizes ${\mathbb{G}}$ as a subgroup. [Proof.]{} By hypothesis, the class of $k$-dimensional subgroups $H=\exp {\mathfrak{h}}$ of ${\mathbb{G}}$ which are h-isomorphic to ${\mathbb{R}}^k$ is nonempty and it correspond to $k$-dimensional commutative subgroups contained in $\exp V_1$. Let ${\mathcal{N}}_1$ be a subspace of $V_1$ such that ${\mathfrak{h}}\oplus{\mathcal{N}}_1=V_1$. Then ${\mathcal{N}}={\mathcal{N}}_1\oplus V_2\oplus\cdots\oplus V_\iota$ is an ideal and $N=\exp {\mathcal{N}}$ is complementary to $H$. $\Box$ \[rksubrm\][If ${\mathbb{R}}^k$ h-embeds into ${\mathbb{M}}$, then every injective h-homomorphism from ${\mathbb{R}}^k$ to ${\mathbb{M}}$ is an h-monomorphism. This immdiately follows joining Proposition \[factorpro\] and Proposition \[rksub\]. ]{} \[factor\][We say that ${\mathbb{G}}$ [*factorizes ${\mathbb{M}}$*]{} if it factorizes ${\mathbb{M}}$ both as a subgroup and as a quotient. ]{} [It might be interesting from an algebraic viewpoint to investigate whether factorizing groups as quotient are also factorizing as a subgroup and viceversa. ]{} \[rfactg\] ${\mathbb{R}}$ factorizes every stratified group. [Proof.]{} By Lemma \[hepim\], ${\mathbb{R}}$ is an h-quotient of an arbitrary stratified group ${\mathbb{G}}$. Let $N=\exp\big({\mathcal{N}}_1\oplus\cdots\oplus{\mathcal{N}}_\iota\big)$ be a homogeneous normal subgroup such that ${\mathbb{G}}/N$ is h-isomorphic to ${\mathbb{R}}$. By Proposition \[quotgrad\], we must have that $V_1/{\mathcal{N}}_1$ is one-dimensional. The Lie algebra ${\mathcal{G}}$ of ${\mathbb{G}}$ is the direct sum $V_1\oplus\cdots\oplus V_\iota$. We choose $X\in V_1{\setminus}{\mathcal{N}}_1$ and define $${\mathfrak{h}}=\{tX\mid t\in{\mathbb{R}}\}.$$ It follows that ${\mathfrak{n}}$ and ${\mathfrak{h}}$ are homogeneous subalgebras such that ${\mathfrak{n}}\oplus{\mathfrak{h}}={\mathcal{G}}$. By virtue of Proposition \[homdec\], ${\mathbb{R}}$ factorizes ${\mathbb{G}}$ as a quotient. Finally, by Proposition \[rksub\] ${\mathbb{R}}$ factorizes ${\mathbb{G}}$ as a subgroup, hence our proof is complete. $\Box$ \[cntrex\][One can easily find ${\mathbb{G}}$ and ${\mathbb{M}}$ such that there exist h-epimorphisms $L:{\mathbb{G}}{\longrightarrow}{\mathbb{M}}$, although ${\mathbb{M}}$ does not factorize ${\mathbb{G}}$ as a quotient. In fact, let us consider the 2-step stratified group ${\mathbb{G}}$ with Lie algebra ${\mathcal{G}}=V_1\oplus V_2$, where $$V_1={\mbox{span}}\{X_1,X_2,X_3,X_4\},\qquad V_2={\mbox{span}}\{Z_{23},Z_{24},Z_{34}\}$$ and the only nontrivial bracket relations are the following $$[X_2,X_3]=Z_{23},\quad[X_2,X_4]=Z_{24}, \quad\mbox{and}\quad [X_3,X_4]=Z_{34}\,.$$ Let $L_1:{\mathbb{G}}{\longrightarrow}{\mathbb{R}}^2$ be the h-epimorphism defined by $$L_1\Big(\exp \big(\sum_{i=1}^4x_iX_i+z_{23}Z_{23}+z_{24}Z_{24}+z_{34}Z_{34}\big)\Big) =(x_1,x_2)$$ and let $\ker L_1=\exp{\mathfrak{n}}=\exp({\mathfrak{n}}_1\oplus{\mathfrak{n}}_2)=N$. We have ${\mathfrak{n}}_2=V_2$ and ${\mathfrak{n}}_1={\mbox{span}}\{X_3,X_4\}$, then $H=\exp\{X_1,X_2\}$ is complementary to $N$, as it easily follows from Proposition \[homdec\]. However, if we consider the h-epimorphism $L_2:{\mathbb{G}}{\longrightarrow}{\mathbb{R}}^2$ defined by $$L_2\Big(\exp \big(\sum_{i=1}^4x_iX_i+z_{23}Z_{23}+z_{24}Z_{24}+z_{34}Z_{34}\big)\Big) =(x_3,x_4),$$ then $\ker L_2=\exp{\mathfrak{n}}$, with ${\mathfrak{n}}={\mbox{span}}\{X_1,X_2\}\oplus V_2$ and we will check that $N=\exp{\mathfrak{n}}$ does not admit any complementary subgroup. By contradiction, if $H$ is complementary to $N$, then Proposition \[GlinH\] shows that the restriction $T:H{\longrightarrow}{\mathbb{R}}^2$ of $L_2$ is a group isomorphism. In particular, $H$ is commutative. In addition, Proposition \[homdec\] shows that ${\mathfrak{h}}\oplus{\mathfrak{n}}={\mathcal{G}}$, where ${\mathfrak{h}}$ denotes the Lie algebra of $H$. Then ${\mathfrak{h}}$ is a 2-dimensional commutative subalgebra of ${\mathcal{G}}$. We consider a basis $(v,w)$ of ${\mathfrak{h}}$, given by $$v=\sum_{j=1}^4\alpha_{j}X_j+Z,\quad\mbox{and}\quad w=\sum_{j=1}^4\beta_jX_j+T,$$ where $T,Z\in V_2$. The decomposition ${\mathfrak{h}}\oplus{\mathfrak{n}}={\mathcal{G}}$ implies that $(v,w,X_1,X_2,Z_{23},Z_{24},Z_{34})$ is a basis of ${\mathcal{G}}$, hence we must have $\alpha_3\beta_4-\beta_3\alpha_4\neq0$. As a consequence, $$[v,w]=(\alpha_2\beta_3-\alpha_3\beta_2)Z_{23}+ (\alpha_2\beta_4-\alpha_4\beta_2)Z_{24}+ (\alpha_3\beta_4-\alpha_4\beta_3)Z_{34}\neq0\,.$$ This conflicts with the fact that ${\mathfrak{h}}$ is commutative. ]{} Factorizations in some H-type groups {#htypeg} ------------------------------------ Our aim here is to present some factorizing properties of ${\mathbb{R}}^k$ in some important H-type groups. These groups were introduced in [@Kap]. \[htype\][Let ${\mathfrak{g}}$ be a Lie algebra equipped with an inner product ${\langle}\cdot,\cdot{\rangle}$ and let ${\mathfrak{z}}$ be a nontrivial subspace of ${\mathfrak{g}}$ such that $[{\mathfrak{g}},{\mathfrak{z}}]=0$ and $[{\mathfrak{g}},{\mathfrak{g}}]\subset{\mathfrak{z}}$. Let ${\mathfrak{v}}$ be the orthogonal space of ${\mathfrak{z}}$ and define $J:{\mathfrak{z}}{\longrightarrow}\mbox{End}({\mathfrak{v}})$ by the formula $$\begin{aligned} {\langle}J_ZX,Y{\rangle}={\langle}Z,[X,Y]{\rangle}\end{aligned}$$ for every $Z\in{\mathfrak{z}}$ and $X,Y\in{\mathfrak{v}}$. If $J$ satisfies the condition $$|J_ZX|=|Z|\,|X|\,,$$ then we say that ${\mathfrak{g}}$ is an [*H-type algebra*]{}. ]{} From the previous definition, it follows that $$\begin{aligned} \label{keyhtype} [X,J_ZX]=|X|^2\;Z\qquad\mbox{for every}\qquad (X,Z)\in {\mathfrak{v}}\times{\mathfrak{z}}\,.\end{aligned}$$ Then ${\mathfrak{g}}={\mathfrak{v}}\oplus{\mathfrak{z}}$ is a 2-step stratified algebra of first layer ${\mathfrak{v}}$ and second layer ${\mathfrak{z}}$. In the sequel, we will utilize another well known formula $$\begin{aligned} \label{polH} J_ZJ_W+J_WJ_Z=-2\,{\langle}Z,W{\rangle}\,\mbox{Id}_{\mathfrak{v}}.\end{aligned}$$ \[heishtype\] [Let $(X,Y,Z)$ be an orthonormal basis of the Heisenberg algebra ${\mathfrak{h}}^1$, where $[X,Y]=Z$. Then setting ${\mathfrak{z}}={\mbox{span}}\{Z\}$, ${\mathfrak{v}}={\mbox{span}}\{X,Y\}$ and $$J_ZX=Y,\qquad J_ZY=-X$$ extended by linearity, it follows that $J$ makes ${\mathfrak{h}}^1$ an H-type group. The higher dimensional Heisenberg algebras ${\mathfrak{h}}^n$ can be seen as direct product of the irreducible Heisenberg algebras isomorphic to ${\mathfrak{h}}^1$ as follows $${\mathfrak{h}}^n={\mathfrak{v}}_1\oplus\cdots\oplus{\mathfrak{v}}_n\oplus{\mathfrak{z}},$$ where $(X_i,Y_i)$ is an orthonormal basis of ${\mathfrak{v}}_i$ and $J_ZX_i=Y_i$, $J_ZY_i=-X_i$. Here we notice that setting $\omega(U_1,U_2)={\langle}J_Z U_1,U_2{\rangle}$, we define a symplectic form on the $2n$-dimensional first layer ${\mathfrak{v}}={\mathfrak{v}}_1\oplus\cdots\oplus{\mathfrak{v}}_n$. ]{} \[h12k\][It is easy to check that ${\mathbb{R}}^2$ does not factorize ${\mathfrak{h}}^1$ as a quotient, although it is an h-quotient of ${\mathfrak{h}}^1$. In fact, the only 1-dimensional ideal ${\mathfrak{n}}$ of ${\mathfrak{h}}^1$ is the second layer ${\mathfrak{z}}$ and we have already shown in Example \[nnexcmpl\] that there do not exist 2-dimensional subalgebras complementary to ${\mathfrak{z}}$. ]{} However, as first observed in [@FSSC6], ${\mathbb{R}}^k$ factorizes the Heisenberg group ${\mathbb{H}}^n$ as a quotient, whenever $k\leq n$. In view of Proposition \[GlinH\], the next statement translates Proposition 3.24 of [@FSSC6] into our setting, with a different proof. \[kfacth\] If $1\leq k\leq n$, then ${\mathbb{R}}^k$ factorizes ${\mathbb{H}}^n$. [Proof.]{} By Lemma \[hepim\], ${\mathbb{R}}^k$ is an h-quotient ${\mathbb{H}}^n$. Let $N$ be a normal homogeneous subgroup of ${\mathbb{H}}^n$ such that ${\mathbb{H}}^n/N$ is h-isomorphic to ${\mathbb{R}}^k$. Let ${\mathfrak{n}}={\mathfrak{n}}_1\oplus{\mathfrak{n}}_2$ be the Lie algebra of $N$, which is also an ideal of ${\mathfrak{h}}^n$. It is clear that ${\mathfrak{n}}_2={\mathfrak{z}}$, where ${\mathfrak{v}}$ and ${\mathfrak{z}}$ denote the first and the second layers of ${\mathfrak{h}}^n$, respectively. If we prove that there exists a $k$-dimensional commutative subalgebra ${\mathfrak{s}}$ contained in ${\mathfrak{v}}$ such that ${\mathfrak{s}}\oplus{\mathfrak{n}}_1={\mathfrak{v}}$, then Proposition \[homdec\] concludes the proof. To prove this, we fix a unit vector $Z\in{\mathfrak{z}}$ and consider the mapping $J_Z:{\mathfrak{v}}{\longrightarrow}{\mathfrak{v}}$ that makes ${\mathfrak{h}}^n$ an H-type algebra, see Example \[heishtype\]. We set ${\mathfrak{n}}_1\cap J_Z({\mathfrak{n}}_1)={\mathfrak{w}}$ and notice that $J_Z({\mathfrak{w}})={\mathfrak{w}}$. If ${\mathfrak{w}}\neq\{0\}$, then ${\mathfrak{w}}$ is a symplectic subspace with respect to the symplectic form $\omega(X,Y)={\langle}J_ZX,Y{\rangle}$. In fact, one can easily construct the following orthogonal basis $(e_1,\ldots,e_l,J_Z(e_1),\ldots,J_Z(e_l))$ of ${\mathfrak{w}}$. We proceed as follows, choosing a unit vector $e_1\in{\mathfrak{w}}$ and then considering the orthogonal $J_Z(e_1)\in{\mathfrak{w}}$. The procedure continues selecting a unit vector $e_2\in{\mathfrak{w}}\cap{\mbox{span}}\{e_1,J_Z(e_1)\}^\bot$ and checking that $J_Z(e_2)\in{\mathfrak{w}}\cap{\mbox{span}}\{e_1,J_Z(e_1)\}^\bot$ and it is repeated up to reaching the desired basis. Notice that this basis is orthonormal by construction. If $p=\dim({\mathfrak{n}}_1)>2l$, then we choose $u_1\in {\mathfrak{n}}_1\cap{\mathfrak{w}}^\bot$ and observe that $J_Z(u_1)\notin{\mathfrak{n}}_1$, otherwise $u_1,J_Z(u_1)$ would belong to ${\mathfrak{w}}$, that is a contradiction. One can iterate this argument up to reaching $u_1,\ldots,u_{p-2l}$, where for every $j=1,\ldots,p-2l$ we have $$J_Z(u_j)\notin{\mathfrak{n}}_1.$$ By construction $u_1,\ldots,u_{p-2l},J_Z(u_1),\ldots,J_Z(u_{p-2l})$ are orthogonal vectors and $$(e_1,\ldots,e_l,J_Z(e_1),\ldots,J_Z(e_l),u_1,\ldots,u_{p-2l})$$ is an orthonormal basis of ${\mathfrak{n}}_1$. If $(p-l)<n$, then repeating the previous argument we can complete $$\begin{aligned} \label{symplbasis} \left(e_1,\ldots,e_l,J_Z(e_1),\ldots,J_Z(e_l),u_1,\ldots,u_{p-2l}, J_Z(u_1),\ldots,J_Z(u_{p-2l})\right)\end{aligned}$$ to the following orthonormal basis of ${\mathfrak{v}}$ $$\begin{aligned} &&\big(e_1,\ldots,e_l,J_Z(e_1),\ldots,J_Z(e_l),u_1,\ldots,u_{p-2l}, J_Z(u_1),\ldots,J_Z(u_{p-2l}),\\ &&w_1,\ldots,w_r,J_Z(w_1),\ldots,J_Z(w_r)\big)\,.\end{aligned}$$ This is precisely a symplectic basis with respect to $\omega(\cdot,\cdot)$, since $${\langle}Z,[w_s,J_Zw_s]{\rangle}={\langle}J_Zw_s,J_Zw_s{\rangle}=1 ={\langle}Z,[e_i,J_Ze_i]{\rangle}={\langle}Z,[u_j,J_Zu_j]{\rangle}$$ implies that $[w_s,J_Zw_s]=[e_i,J_Ze_i]=[u_j,J_Zu_j]=Z$ and in the same way one can check that these are the only nontrivial brackets. For instance, we have $${\langle}Z,[J_Z(u_i),w_j]{\rangle}={\langle}J_Z^2(u_i),w_j{\rangle}=-{\langle}u_i,w_j{\rangle}=0$$ that gives $[J_Z(u_i),w_j]=0$. Finally, the dimensional condition $2l+2(p-2l)+2r=2n$ gives $l-r=p-n\geq0$, by our hypothesis on $p\geq n$. Thus, the basis $$\Big(J_Z(u_1),\ldots,J_Z(u_{p-2l}), w_1+e_1,\ldots,w_r+e_r,J_Z(w_1)-J_Z(e_1),\ldots,J_Z(w_r)-J_Z(e_r)\Big)$$ defines a $k$-dimensional subspace ${\mathfrak{s}}$ that is commutative and by construction satisfies ${\mathfrak{s}}\cap{\mathfrak{n}}_1=\{0\}$. This shows that ${\mathbb{R}}^k$ factorizes ${\mathbb{H}}^n$ as a quotient. Due to Proposition \[rksub\], ${\mathbb{R}}^k$ also factorizes ${\mathbb{H}}^n$ as a subgroup. This concludes the proof. $\Box$ [Joining Proposition \[factorpro\] and Proposition \[rfactg\], it follows that every surjective h-homomorphism $L:{\mathbb{G}}{\longrightarrow}{\mathbb{R}}$ and every injective h-homomorphism $L:{\mathbb{R}}{\longrightarrow}{\mathbb{G}}$ are an h-epimorphism and an h-monomoprhism, respectively. Joining Proposition \[factorpro\] and Proposition \[kfacth\], for every $1\leq k\leq n$, it follows that every surjective h-homomorphism $L:{\mathbb{H}}^n{\longrightarrow}{\mathbb{R}}^k$ and every injective h-homomorphism $L:{\mathbb{R}}^k{\longrightarrow}{\mathbb{H}}^n$ are an h-epimorphism and an h-monomoprhism, respectively. On the other hand, Example \[h12k\] shows that ${\mathbb{R}}^k$ is not an h-quotient of ${\mathbb{H}}^n$ whenever $k>n$. ]{} Another example of H-type group is the complexified Heisenberg group ${\mathbb{H}}_2^1$, where the center ${\mathfrak{z}}$ of its Lie algebra ${\mathfrak{h}}_2^1$ is 2-dimensional and the first layer ${\mathfrak{v}}$ has dimension four. More information on this group can be found in [@ReiRic]. \[baslemcompheis\] Let ${\mathfrak{h}}_2^1={\mathfrak{v}}\oplus{\mathfrak{z}}$ be the 6-dimensional real Lie algebra of the complexified Heisenberg group ${\mathbb{H}}_2^1$ and let $X\in{\mathfrak{v}}$ with $|X|=1$. Let $(Z_1,Z_2)$ be an orthonormal basis of ${\mathfrak{z}}$. Then the following vectors $$\begin{aligned} \label{R_i} R_0=X,\quad R_1=J_{Z_1}X,\quad R_2=J_{Z_2}X,\quad R_3=J_{Z_1}J_{Z_2}X,\end{aligned}$$ form an orthonormal basis of ${\mathfrak{v}}$ and the only nontrivial bracket relations are given by $$\begin{aligned} [R_0,R_1]=[R_2,R_3]=Z_1, \quad [R_0,R_2]=-[R_1,R_3]=Z_2.\end{aligned}$$ [Proof.]{} Using just the properties of $J$ and formula $[Y,J_ZY]=|Y|^2Z$ for every $Y\in{\mathfrak{v}}$ and $Z\in{\mathfrak{z}}$ it is easy to check that $R_i$ form an orthonormal basis of ${\mathfrak{v}}$. The previous formula also yields $[R_0,R_1]=Z_1$ and $[R_0,R_2]=Z_2$. In addition, we have $$\begin{aligned} &&{\langle}Z_1,[R_1,R_2]{\rangle}={\langle}J_{Z_1}^2X,J_{Z_2}X{\rangle}=-{\langle}X,J_{Z_2}X{\rangle}=0, \\ &&{\langle}Z_2,[R_1,R_2]{\rangle}={\langle}J_{Z_2}J_{Z_1}X,J_{Z_2}X{\rangle}={\langle}J_{Z_1}X,X{\rangle}=0, \end{aligned}$$ then $[R_1,R_2]=0$. To prove that $[R_0,R_3]=0$, $[R_1,R_3]=-Z_2$ and $[R_2,R_3]=Z_1$, one argues in the same way, using also formula . $\Box$ \[h21r2\] ${\mathbb{R}}^2$ factorizes the complexified Heisenberg group ${\mathbb{H}}_2^1$ as a quotient. [Proof.]{} Lemma \[hepim\] ensures that ${\mathbb{R}}^2$ is an h-quotient of ${\mathbb{H}}_2^1$. Let $N$ be the kernel of an h-epimorphism $L:{\mathbb{H}}_2^1{\longrightarrow}{\mathbb{R}}^2$ and set ${\mathfrak{n}}=\exp^{-1}(N)={\mathfrak{n}}_1\oplus{\mathfrak{n}}_2$, with ${\mathfrak{n}}_1={\mathfrak{n}}\cap V_1$ and ${\mathfrak{n}}_2={\mathfrak{n}}\cap V_2$, due to Proposition \[pfactrs\]. Clearly ${\mathfrak{n}}_2={\mathfrak{z}}$ and $\dim({\mathfrak{n}}_1)=2$. We have either $\dim([{\mathfrak{n}}_1,{\mathfrak{n}}_1])=0$ or $\dim([{\mathfrak{n}}_1,{\mathfrak{n}}_1])=1$. In the first case we choose an orthonormal basis $(X,Y)$ of ${\mathfrak{n}}_1$ and represent $Y$ as a linear combination of $R_i$, according to . Then the fact that $(X,Y)$ is an orthonormal basis and $[X,Y]=0$ imply that $Y=J_{Z_1}J_{Z_2}X$ for a fixed orthonormal basis $(Z_1,Z_2)$ of the center ${\mathfrak{z}}$. Now we simply notice that the commutative subalgebra ${\mathfrak{h}}={\mbox{span}}\{J_{Z_1}X,J_{Z_2}X\}$ satisfies ${\mathfrak{h}}\oplus{\mathfrak{n}}={\mathfrak{h}}_2^1$. Let us consider the remaining case and take an orthonormal basis $(X,Y)$ of ${\mathfrak{n}}_1$. We have $[X,Y]=Z\neq0$. Let $(T_1,T_2)$ be an orthonormal basis of ${\mathfrak{z}}$ such that $T_1=Z/|Z|$. Replacing $(Z_1,Z_2)$ in Lemma \[baslemcompheis\] with the orthonormal basis $(T_1,T_2)$, we get $Y=\alpha_1 J_{T_1}X+\alpha_3J_{T_1}J_{T_2}X$, with $|\alpha_1|=|Z|>0$. By direct computation, one can check that the commutative subalgebra ${\mathfrak{h}}={\mbox{span}}\{X-\lambda J_{T_2}X,\lambda J_{T_1}X+J_{T_1}J_{T_2}X\}$ satisfies the condition ${\mathfrak{h}}\oplus{\mathfrak{n}}={\mathfrak{h}}_2^1$ if we fix $\lambda\neq0$ and $\lambda^{-1}\neq\alpha_3/\alpha_1$. This concludes the proof. $\Box$ \[nnexcomm\] Let ${\mathfrak{h}}_2^1={\mathfrak{v}}\oplus{\mathfrak{z}}$ be the complexified Heisenberg algebra. Then there do not exist commutative subalgebras of ${\mathfrak{v}}$ with dimension greater than two. [Proof.]{} By the general formula , for every $X\in{\mathfrak{v}}{\setminus}\{0\}$ the mapping $$\mbox{ad X}:{\mathfrak{v}}{\longrightarrow}{\mathfrak{z}}\,,\qquad Y{\longrightarrow}[X,Y]$$ is surjective, then its kernel is 2-dimensional. By contradiction, the existence of a commutative subalgebra of ${\mathfrak{v}}$ with dimension greater than two would conflict with the dimension of the kernel. $\Box$ For $k=3,4$ we have that ${\mathbb{R}}^k$ does not h-embeds into ${\mathbb{H}}_2^1$ and does not factorize ${\mathbb{H}}^1_2$ as a quotient. [Proof.]{} The first assertion immediately follows from both Definition \[hquotembed\] and Proposition \[nnexcomm\]. Concerning the proof of the second assertion, in view of Proposition \[factorpro\] we consider a surjective h-homomorphism $L:{\mathfrak{h}}_2^1{\longrightarrow}{\mathbb{R}}^3$. By contradiction, if $L$ is an h-epimorphism, then we get a 3-dimensional subalgebra of ${\mathfrak{v}}$ that is h-isomorphic to ${\mathbb{R}}^3$. This conflicts with Proposition \[nnexcomm\] and concludes the proof. $\Box$ Factorizations in some free stratified groups {#factfree} --------------------------------------------- We denote by ${\mathfrak{g}}_{p,\upsilon}$ the free $\upsilon$-step stratified algebra on $p$ generators. The corresponding simply connected Lie group will be denoted by ${\mathbb{G}}_{p,\upsilon}$. \[rpupsilon\][We notice that ${\mathfrak{g}}_{p,\upsilon}$ is an h-quotient of ${\mathfrak{g}}_{r,\upsilon}\oplus{\mathfrak{a}}$ for every $p\leq r$, where ${\mathfrak{a}}$ is a stratified algebra and $\oplus$ denotes the direct product of Lie algebras. It suffices to consider the basis $$\begin{aligned} &&{\mathcal{B}}_{r,\upsilon}=\big\{[X_{j_1},[X_{j_2},[\cdots,[X_{j_{s-1}},X_{j_s}],],\ldots,] \mid 1\leq s\leq \upsilon,\\ && (j_1,\ldots,j_s)\in {\mathcal{A}}^s_{r,\upsilon}\subset\{1,\ldots,r\}^s\big\}\end{aligned}$$ of ${\mathfrak{g}}_{r,\upsilon}$ and the basis $$\begin{aligned} &&{\mathcal{B}}_{p,\upsilon}'=\big\{[X_{j_1}',[X_{j_2}',[\cdots,[X_{j_{s-1}}', X_{j_s}'],],\ldots,] \mid 1\leq s\leq \upsilon,\\ && (j_1,\ldots,j_s)\in {\mathcal{A}}'^s_{p,\upsilon}\subset\{1,\ldots,p\}^s\big\}\end{aligned}$$ of ${\mathfrak{g}}_{p,\upsilon}$, setting $$\begin{aligned} L\big([X_{j_1},[X_{j_2},[\cdots,[X_{j_{s-1}},X_{j_s}],],\ldots,]\big):= [X_{j_1}',[X_{j_2}',[\cdots,[X_{j_{s-1}}',X_{j_s}'],],\ldots,]\,,\end{aligned}$$ for every $1\leq s\leq \upsilon$ and every $(j_1,\ldots,j_s)\in {\mathcal{A}}'^s_{p,\upsilon}\subset{\mathcal{A}}^s_{r,\upsilon}$, $$\begin{aligned} L\big([X_{j_1},[X_{j_2},[\cdots,[X_{j_{s-1}},X_{j_s}],],\ldots,]\big):=0\end{aligned}$$ for every $1\leq s\leq \upsilon$ and every $(j_1,\ldots,j_s)\in{\mathcal{A}}^s_{r,\upsilon}{\setminus}{\mathcal{A}}'^s_{p,\upsilon}$ and $L({\mathfrak{a}})=\{0\}$. It is easy to check that $L:{\mathfrak{g}}_{r,\upsilon}\oplus{\mathfrak{a}}{\longrightarrow}{\mathfrak{g}}_{p,\upsilon}$ is a surjective h-homomorphism. ]{} \[Gpupsilon\] Let ${\mathbb{P}}$ be a stratified group such that ${\mathbb{G}}_{p,\upsilon}$ is an h-quotient of ${\mathbb{P}}$. Then ${\mathbb{G}}_{p,\upsilon}$ factorizes ${\mathbb{P}}$ as a quotient. [Proof.]{} We apply Proposition \[factorpro\]. By hypothesis, we have $L:{\mathcal{P}}{\longrightarrow}{\mathfrak{g}}_{p,\upsilon}$ that is a surjective h-homomorphism, where ${\mathcal{P}}=W_1\oplus\cdots\oplus W_\iota$ is the stratified algebra of ${\mathbb{P}}$. Let $X_1,\ldots,X_p$ be generators of ${\mathfrak{g}}_{p,\upsilon}$. Then there exist $U_1,\ldots, U_p\in W_1$ such that $L(U_i)=X_i$. By hypothesis, we consider the basis $${\mathcal{B}}=\{[X_{j_1},[X_{j_2},[\cdots,[X_{j_{s-1}},X_{j_s}],],\ldots,]\mid 1\leq s\leq \upsilon,\;(j_1,\ldots,j_s)\in {\mathcal{A}}^s\subset\{1,\ldots,p\}^s\}$$ of ${\mathfrak{g}}_{p,\upsilon}$. The homomorphism property of $L$ implies that $$L\big([U_{j_1},[U_{j_2},[\cdots,[U_{j_{s-1}},U_{j_s}],],\ldots,]\big)= [X_{j_1},[X_{j_2},[\cdots,[X_{j_{s-1}},X_{j_s}],],\ldots,]\,,$$ then the family $${\mathcal{B}}'=\{[U_{j_1},[U_{j_2},[\cdots,[U_{j_{s-1}},U_{j_s}],],\ldots,]\mid 1\leq s\leq \upsilon,\;(j_1,\ldots,j_s)\in {\mathcal{A}}'^s\subset\{1,\ldots,p\}^s\}$$ is a basis of ${\mathfrak{h}}=\mbox{Lie-}{\mbox{span}}\{U_1,\ldots,U_p\}$, then $L$ maps ${\mathfrak{h}}$ h-isomorphically onto ${\mathfrak{g}}_{p,\upsilon}$. $\Box$ ${\mathbb{H}}^1$ factorizes ${\mathbb{G}}_{r,2}\times {\mathbb{G}}'$ as a quotient for every $r\geq2$. [Proof.]{} Observe that ${\mathfrak{h}}^1={\mathfrak{g}}_{2,2}$, take into account Remark \[rpupsilon\] and apply Proposition \[Gpupsilon\]. $\Box$ .2truecm The following corollaries are straightforward. ${\mathbb{H}}^1$ factorizes ${\mathbb{G}}_{r,2}\times {\mathbb{G}}_{r,2}\times\cdots\times{\mathbb{G}}_{r,2}$ as a quotient for every $r\geq2$. ${\mathbb{H}}^1$ factorizes ${\mathbb{H}}^1\times {\mathbb{H}}^1\times\cdots\times{\mathbb{H}}^1$ as a quotient. [Although ${\mathbb{H}}^1$ factorizes ${\mathbb{H}}^1\times {\mathbb{H}}^1\times\cdots\times{\mathbb{H}}^1$ as a quotient, one notice that there are a few nontrivial h-epimorphisms between these groups. In fact, they are all of the form $$L:\overbrace{{\mathbb{H}}^1\times {\mathbb{H}}^1\times\cdots\times{\mathbb{H}}^1}^{n-times}{\longrightarrow}{\mathbb{H}}^1\,\quad L(a_1,\ldots,a_n)=J(a_k)$$ for some fixed $k\in\{1,\ldots,n\}$ and some h-isomorphism $J:{\mathbb{H}}^1{\longrightarrow}{\mathbb{H}}^1$. ]{} Examples of $({\mathbb{G}},{\mathbb{M}})$-regular sets {#SecEx} ====================================================== Existence of different types of intrinsically regular sets in a given graded group ${\mathbb{P}}$ depends on the corresponding algebraic factorizations. In correspondence to the fact that ${\mathbb{R}}^k$ factorizes a graded group ${\mathbb{M}}$ as a subgroup, we have the following \[legalg\] Let ${\mathbb{M}}$ be a graded group and let $n$ be the maximum over all dimensions of commutative subalgebras contained in the first layer. Then the family of $k$-dimensional Legendrian $C^1$ smooth submanifolds is nonempty if and only if $1\leq k\leq n$ and it coincides with that of $({\mathbb{R}}^k,{\mathbb{M}})$-regular sets of ${\mathbb{M}}$. [Proof.]{} Every $k$-dimensional commutative subalgebra of the first layer also represents a trivial example of $k$-dimensional Legendrian submanifold. This shows, by definition of $n$, that this family is nonempty for every $1\leq k\leq n$. Recall that a $k$-dimensional Legendrian submanifold is locally parametrized by a $C^1$ contact mapping defined on an open subset of ${\mathbb{R}}^k$. In other words, it is locally parametrized by a continuously h-differentiable contact mapping with injective differential. Then Theorem \[PdifContact\] shows that this mapping is continuously P-differentiable. Furthermore, since in this case classical differentiability coincides with h-differentiability, then formula shows that the classical differential coincides with the P-differential. Then the P-differential of the local parametrization is an injective h-homomorphism. As a result, by Remark \[rksubrm\] the P-differential is an h-monomorphism. Taking into account Definition \[gmsinm\], we have shown that every $C^1$ smooth Legendrian submanifold contained in ${\mathbb{M}}$ is an $({\mathbb{R}}^k,{\mathbb{M}})$-regular set of ${\mathbb{M}}$. Clearly, we have also proved that there do not exist $k$-dimensional Legendrian submanifold when $k>n$, since the injective P-differential of the local parametrization would imply the existence of a commutative subalgebra in the first layer of ${\mathcal{M}}$ with dimension greater than $k$, that is a contradiction. Now, let $S$ be an $({\mathbb{R}}^k,{\mathbb{M}})$-regular set, where clearly $1\leq k\leq n$. Again Theorem \[PdifContact\] shows that continuously P-differentiable mappings on ${\mathbb{R}}^k$ exactly correspond to contact mappings of class $C^1$, then $S$ is a $C^1$ smooth Legendrian submanifold. $\Box$ .2truecm It is well known that in the Heisenberg group ${\mathbb{H}}^n$ the only intrinsically regular sets are $({\mathbb{H}}^n,{\mathbb{R}}^k)$-regular sets and $({\mathbb{R}}^k,{\mathbb{H}}^n$-regular sets with $1\leq k\leq n$, [@FSSC6]. [We say that a homogeneous subgroup of ${\mathbb{H}}^n$ is [*horizontal*]{} if its Lie algebra is contained in the first layer ${\mathfrak{v}}$ and [*vertical*]{} if it contains the second layer ${\mathfrak{z}}$.]{} [Notice that horizontal subgroups are always commutative, whereas homogeneous normal subgroups exactly characterize vertical subgroups. ]{} \[anyfact\] Every couple of complementary subgroups of ${\mathbb{H}}^n$ is formed by a horizontal and a vertical subgroup. [Proof.]{} Let ${\mathfrak{a}}$ and ${\mathfrak{b}}$ be the corresponding homogeneous subalgebras of $A$ and $B$, respectively. From our hypothesis and Proposition \[homdec\], it follows that ${\mathfrak{a}}\oplus{\mathfrak{b}}={\mathfrak{h}}^n$. In particular, $Z\in{\mathfrak{a}}\oplus{\mathfrak{b}}$ hence we can suppose that ${\mathfrak{a}}$ is the subalgebra containing a vector $W=W_0+\lambda Z$ with $\lambda\neq0$. Homogeneity of ${\mathfrak{a}}$ gives $W_0+r\lambda Z\in{\mathfrak{a}}$ for every $r>0$, therefore $W_0\in{\mathfrak{a}}$ and $Z\in{\mathfrak{a}}$. This shows that ${\mathfrak{a}}$ is an ideal of ${\mathfrak{h}}^n$. Clearly, the same argument shows that ${\mathfrak{b}}$ cannot have vectors of the form $U_0+\mu Z$ with $\mu\neq0$, otherwise $Z$ would belong to ${\mathfrak{b}}$, that conflicts with the decomposition ${\mathfrak{a}}\oplus{\mathfrak{b}}={\mathfrak{h}}^n$. This shows that ${\mathfrak{b}}$ is contained in the first layer of ${\mathfrak{h}}^n$ and concludes the proof. $\Box$ .2truecm The special factorization of ${\mathbb{H}}^n$ cannot be extended to all H-type groups. \[nnorhtype\][Let us consider the complexified Heisenberg group ${\mathbb{H}}^1_2$ along with the bases $(R_0,R_1,R_2,R_3)$ and $(Z_1,Z_2)$, defined in Lemma \[baslemcompheis\]. We define the homogeneous commutative subalgebras ${\mathfrak{a}}={\mbox{span}}\{R_0,R_3,Z_1\}$ and ${\mathfrak{b}}={\mbox{span}}\{R_1,R_2,Z_2\}$. Clearly, ${\mathfrak{a}}\oplus{\mathfrak{b}}={\mathfrak{h}}_2^1$, hence from Proposition \[homdec\] it follows that $A=\exp{\mathfrak{a}}$ and $B=\exp{\mathfrak{b}}$ are complementary subgroups of ${\mathbb{H}}_2^1$. On the other hand, $[R_0,R_2]=Z_2\notin{\mathfrak{a}}$ and $[R_0,R_1]=Z_1\notin{\mathfrak{b}}$ impliy that both ${\mathfrak{a}}$ and ${\mathfrak{b}}$ are not ideals of ${\mathfrak{h}}_2^1$.]{} On the other hand, we have the following \[clsh21\] Let $N$ and $H$ be complementary subgroups of ${\mathbb{H}}_2^1$, where $N$ is a normal. Then $N$ contains the center of the group, $\mbox{\rm top-}\dim(N)\in\{4,5\}$ and $H$ is commutative and horizontal, namely, $H\subset\exp{\mathfrak{v}}$. [Proof]{} Let $N=\exp {\mathfrak{n}}$ and $H=\exp {\mathfrak{h}}$. If $\dim({\mathfrak{n}})\geq3$, then ${\mathfrak{n}}\supset{\mathfrak{z}}$ by formula . Then in this case $\dim({\mathfrak{h}})\leq 3$ and ${\mathfrak{h}}\subset{\mathfrak{v}}$. By Proposition \[nnexcomm\], it follows that $\dim({\mathfrak{h}})\leq 2$. Then the case $\dim({\mathfrak{n}})=3$ cannot occur. If $\dim({\mathfrak{n}})\leq2$, then ${\mathfrak{n}}\subset{\mathfrak{z}}$, then Proposition \[nnexcomm\] prevents the existence of ${\mathfrak{h}}\subset{\mathfrak{v}}$, that should be commutative. As a result, the only allowed possibilities are $\dim({\mathfrak{n}})=4$ and $\dim({\mathfrak{n}})=5$. $\Box$ \[intregh21\] The only intrinsically regular sets of ${\mathbb{H}}_2^1$ are contained in the following list 1. $({\mathbb{H}}_2^1,{\mathbb{R}})$-regular sets, 2. $({\mathbb{H}}_2^1,{\mathbb{R}}^2)$-regular sets, 3. $({\mathbb{R}}^2,{\mathbb{H}}_2^1)$-regular sets, 4. $({\mathbb{R}},{\mathbb{H}}_2^1)$-regular sets. [Proof.]{} By Proposition \[GlinH\] and Proposition \[clsh21\], we are allowed to consider only level sets defined through continuously P-differentiable mappings with values in ${\mathbb{R}}^k$, $k=1,2$ and defined on an open set of ${\mathbb{H}}_2^1$. Every $C^1$ mapping $f:\Omega\subset{\mathbb{H}}_2^1{\longrightarrow}{\mathbb{R}}^2$ with surjective differential defines a $({\mathbb{H}}_2^1,{\mathbb{R}}^2)$-regular set, since it is continuously P-differential and its P-differential is an h-epimorphism due to Proposition \[h21r2\]. This allows us to apply Theorem \[implth\]. We argue in the same way for $C^1$ mappings $f:\Omega\subset{\mathbb{H}}_2^1{\longrightarrow}{\mathbb{R}}$, due to Proposition \[rfactg\]. Analogously, Proposition \[GlinHmono\] and Proposition \[clsh21\] allow us to consider only image sets from open subsets of ${\mathbb{R}}^k$, $k=1,2$ with values in ${\mathbb{H}}_2^1$. By Theorem \[legalg\], $({\mathbb{R}},{\mathbb{H}}_2^1)$-regular sets and $({\mathbb{R}}^2,{\mathbb{H}}_2^1)$-regular sets of ${\mathbb{H}}_2^1$ exist and correspond to one dimensional and two dimensional $C^1$ smooth Legendrian submanifolds of ${\mathbb{H}}_2^1$. $\Box$ [99]{} , [*Intrinsic regular hypersurfaces in Heisenberg groups*]{}, J. Geom. Anal. 16 (2006), no. 2, 187–232. , [*Selected Topics on Analysis in Metric Spaces*]{}, Oxford University Press, (2003). , [*Stability of isometric maps in the Heisenberg group*]{}, preprint (2005) , [*Size of characteristic sets and functions with prescribed gradients*]{}, J. Reine Angew. Math., [**564**]{}, 63-83, (2003) eds, [*Sub-Riemannian geometry*]{}, Progress in Mathematics, [**144**]{}, Birkhäuser Verlag, Basel, 1996. , [*A course in metric geometry*]{}, Graduate Studies in Mathematics, [**33**]{}, American Mathematical Society, Providence, RI, (2001) , [*Lifts of Lipschitz maps and horizontal fractals in the Heisenberg group*]{}, Ergodic Theory Dynam. Systems [**26**]{}, n.3, 621-651, (2006) , [*The Bernstein problem for intrinsic graphs in Heisenberg groups and calibrations*]{}, preprint (2006) , [*Conformality and Q-Harmonicity in Carnot groups*]{}, Duke Math. J. [**135**]{}, n.3, 455-479, (2006) , [*An Introduction to the Heisenberg Group and the Sub-Riemannian Isoperimetric Problem*]{}, Progress in Mathematics, [**259**]{}, (2007) , [*Ahlfors type estimates for perimeter measures in Carnot-Carathéodory spaces*]{}, J. Geom. Anal. 16 (2006), no. 3, 455–497. , [*Minimal surfaces in pseudohermitian geometry*]{}, Ann. Sc. Norm. Super. Pisa Cl. Sci. (5) [**4**]{}, n.1, 129-177 (2005) , [*Blow-Up in Non-Homogeneous Lie Groups and Rectifiability*]{}, Houston J. Math. [**31**]{}, n.2, 333-353, (2005) , [*Non-doubling Ahlfors measures, Perimeter measures, and the characterization of the trace spaces of Sobolev functions in Carnot-Carathéodory spaces*]{}, Mem. Amer. Math. Soc. [**182**]{}, n.857, (2006) , [*A notable family of entire intrinsic minimal graphs in the Heisenberg group which are not perimeter minimizing*]{}, to appear on Amer. Journal Math. , [*Fractured Fractals and Broken Dreams. Self-Similar Geometry through Metric and Measure*]{}, Oxford University Press, 1997. , [*Geometric Measure Theory*]{}, Springer, (1969). , [*Hardy Spaces on Homogeneous groups*]{}, Princeton University Press, 1982 , [*Rectifiability and Perimeter in the Heisenberg group*]{}, Math. Ann. [**321**]{}, n.3, 479-531, (2001) , [*Regular hypersurfaces, intrinsic perimeter and implicit function theorem in Carnot groups*]{}, Comm. Anal. Geom. [**11**]{}, n.5, 909-944, (2003) , [*On the structure of finite perimeter sets in step 2 Carnot groups*]{}, J. Geom. Anal. [**13**]{}, n.3, 421-466, (2003) , [*Regular submanifolds, graphs and area formula in Heisenberg groups*]{}, Adv. Math. [**211**]{}, n.1, 152-203, (2007) , [*Intrinsic Lipschitz graphs in Heisenberg groups*]{}, preprint (2006) , [*Isoperimetric and Sobolev Inequalities for Carnot-Carathéodory Spaces and the Existence of Minimal Surfaces*]{}, Comm. Pure Appl. Math. [**49**]{}, 1081-1144 (1996) , [*Carnot-Carathéodory spaces seen from within*]{}, in [*Subriemannian Geometry*]{}, Progress in Mathematics, [**144**]{}. ed. by A.Bellaiche and J.Risler, Birkhauser Verlag, Basel, 1996. , [*Metric structures for Riemannian and non-Riemannian spaces*]{}, with appendices by [M.Katz, P.Pansu, S.Semmes]{}, Progress in Mathematics, ed. by Lafontaine and Pansu, Birkäuser, Boston, (1999) , [*Sobolev met Poincaré*]{}, Mem. Amer. Math. Soc. [**145**]{}, (2000) , [*Calculus on Carnot groups*]{}, Ber. Univ. Jyväskylä Math. Inst., [**68**]{}, 1-31, (1995). , [*Lectures on analysis on metric spaces*]{}, Springer-Verlag, New York, (2001) , [*Algebra*]{}, Editori riuniti (1994), translation from [*Topics in algebra*]{} , [*Fundamental solutions for a class of hypoelliptic PDE generated by composition of quadratic forms*]{}, Trans. Amer. Math. Soc. [**258**]{}, n.1, 147-153, (1980) , [*Foundation for the Theory of Quasiconformal Mappings on the Heisenberg Group*]{}, Adv. Math., [**111**]{}, 1-87, (1995). , [*Rectifiability and parametrization of intrinsic regular surfaces in the Heisenberg group*]{}, Ann. Sc. Norm. Super. Pisa Cl. Sci. (5), [**3**]{}, n.4, 871-896, (2004) , [*Algebra*]{}, Springer-Verlag, New York, (2002) , [*On the isoperimetric problem in the Heisenberg group ${\mathbb H}^n$*]{}. Ann. Mat. Pura Appl. (4) [**184**]{}, n.4, 533-553 (2005) , [*Isoperimetric sets on Carnot groups*]{}, Houston J. Math. [**29**]{}, n.3, 609-637 (2003) , [*Degree theory*]{}, Cambridge University Press, (1978) , [*Elements of Geometric Measure Theory on Sub-Riemannian groups*]{}, PhD theses series of Scuola Normale Superiore, (2002) , [*Differentiability and Area formula on stratified Lie groups*]{}, Houston J. Math., [**27**]{}, n.2, 297-323, (2001) , [*On a general coarea inequality and applications*]{}, Ann. Acad. Sci. Fenn. Math., [**27**]{}, 121-140, (2002) , [*The coarea formula for real-valued Lipschitz maps on stratified groups*]{}, Math. Nachr., [**278**]{}, n.14, 1-17, (2005) . [*Characteristic points, rectifiability and perimeter measure on stratified groups*]{}, J. Eur. Math. Soc. [**8**]{}, n.4, 585-609, (2006) , [*An intrinsic measure for submanifolds in stratified groups*]{}, to appear on J. Reine Angew. Math. , [*Non-horizontal submanifolds and coarea formula*]{}, preprint (2007) , [*Contact equations and Lipschitz extensions*]{}, in preparation , [*Measures with unique tangent measures in metric groups*]{}, Math. Scand. [**97**]{}, n.2, 298-308 (2005) , [*Convex isoperimetric sets in the Heisenberg group*]{}, preprint (2006) , [*Surface measures in some CC spaces*]{}, Calc. Var. Partial Differential Equations [**13**]{}, n.3, 339-376, (2001). , [*A Tour of Subriemannian Geometries, Their Geodesics and Applications*]{}, American Mathematical Society, (2002) , [*Croissance des boules et des géodésiques fermeés dans les nilvariété*]{}, Ergod. Dinam. Syst., [**3**]{}, 415-445 (1983) , [*Métriques de Carnot-Carathéodory quasiisométries des espaces symétriques de rang un*]{}, Ann. Math., [**129**]{}, 1-60, (1989) , [*A notion of rectifiability modelled on Carnot groups*]{}, Indiana Univ. Math. J., [**53**]{}, n.1, 49-81, (2004). , [*Nonlinear functional analysis*]{}, Courant Institute of Mathematical Sciences, New York University, (1965) , [*The complexified Heisenberg group*]{}, Proceedings on Analysis and Geometry (Russian) (Novosibirsk Akademgorodok, 1999), 465–480, Izdat. Ross. Akad. Nauk Sib. Otd. Inst. Mat., Novosibirsk, (2000). , [*Personal communication*]{} , [*Area-stationary surfaces in the Heisenberg group $\mathbb H^1$*]{}, preprint (2005) , [*Hypoelliptic differential operators and nilpotent groups*]{}, Acta Math. [**137**]{}, 247-320, (1976). , [*Un complexe de formes différentielles sur les variétés de contact,*]{} C. R. Acad. Sci. Paris Sér. I Math. [**310**]{}, n.6, 401-404, (1990) , [*Formes différentielles sur les variétés de contact*]{}, J. Differential Geom. [**39**]{}, n.2, 281-330, (1994) , [*Differential geometry on C-C spaces and application to the Novikov-Shubin numbers of nilpotent Lie groups*]{}, C. R. Acad. Sci. Paris Sér. I Math. [**329**]{}, n.11, 985-990, (1999) , [*Some Novel Types of Fractal Geometry*]{}, Oxford University Press, (2001). , [*Harmonic Analysis*]{}, Princeton Univ. Press, (1993). , [*Lie groups, Lie algebras and their representation*]{}, Springer-Verlag, New York, (1984). , [*Analysis and Geometry on Groups*]{}, Cambridge University Press, Cambridge, 1992. , [*Contact and Pansu differentiable maps on Carnot groups*]{}, to appear on Bul. Austr. Math. Soc. , [*Foundations of differentiable manifolds and Lie groups*]{}, Foresman and Company, London, (1971).

ArXiv

--- author: - 'Alice C. Schwarze' - 'Philip S. Chodrow' - 'Mason A. Porter' title: 'Log-minor distributions and an application to estimating mean subsystem entropy' --- AMS 2010 Subject classification: 15B99, 15A15, 60E15, 93A10\ Keywords: empirical distributions, determinants, sampling error, positive-definite matrices, random matrices Abstract {#abstract .unnumbered} ======== A common task in physics, information theory, and other fields is the analysis of properties of subsystems of a given system. Given the covariance matrix $M$ of a system of $n$ coupled variables, the covariance matrices of the subsystems are principal submatrices of $M$. The rapid growth with $n$ of the set of principal submatrices makes it impractical to exhaustively study each submatrix for even modestly-sized systems. It is therefore of great interest to derive methods for approximating the distributions of important submatrix properties for a given matrix. Motivated by the importance of differential entropy as a systemic measure of disorder, we study the distribution of log-determinants of principal $k\times k$ submatrices when the covariance matrix has bounded condition number. We derive upper bounds for the right tail and the variance of the distribution of minors, and we use these in turn to derive upper bounds on the standard error of the sample mean of subsystem entropy. Our results demonstrate that, despite the rapid growth of the set of subsystems with $n$, the number of samples that are needed to bound the sampling error is asymptotically independent of $n$. Instead, it is sufficient to increase the number of samples in linear proportion to $k$ to achieve a desired sampling accuracy. Introduction {#sec:introduction} ============ In many fields of study, researchers use matrices to represent systems of interest. Statisticians and data scientists represent large tabular data sets as matrices [@Skiena2017]. In network science, it is common to use adjacency matrices to represent the structure of a network [@Newman2018]. In dynamical systems, researchers use Jacobian matrices in the study of the linearized dynamics of a system of coupled variables [@Strogatz2015]. For networks, dynamical systems, statistical analysis of large data sets, and other applications, it can be insightful (and even necessary) to examine their components (as subnetworks, subsystems, reduced data sets, and so on). Several researchers have used subsystem properties to characterize robustness and other salient properties of dynamical systems [@Tononi1994; @Tononi1999; @Lucia2005; @Randles2011; @Li2012; @Li2016]. Network scientists count and analyze motifs and other subgraphs in networks to characterize a network’s structure [@Shen-Orr2002; @Milo2002]. Several prominent tools in data science are based on linear sketching, an approach to data dimensionality reduction whereby one obtains a reduced data set via matrix multiplication [@Woodruff2014; @Francis2018] or as a linear combination of submatrices [@Sarlos2006; @Clarkson2009; @Kyrillidis2014]. An example of such a tool for dimensionality reduction is principal component analysis [@Liberty2013]. The various applications of submatrices motivate the mathematical study of their properties. In this paper, we study the distribution of log-determinants of principal submatrices of a positive definite matrix and show that our results lead to controllable sampling guarantees for computing the mean differential entropy of subsystems for a dynamical system. Researchers have studied the differential entropy of subsystems in areas such as physics [@Page1993; @Sen1996], biology [@Li2012; @Li2016], neuroscience [@Tononi1994; @Tononi1999; @Lucia2005], computer science [@Randles2011], and coding theory [@Koetter2003]. For example, Tononi et al.(1999) computed a measure of network redundancy from the mean differential entropy of its subsystems of fixed size [@Tononi1999]. Teschendorff et al.(2014) [@Teschendorff2014] used differential entropy to define a measure of network robustness for protein-interaction networks . For several symmetric multivariate distributions, estimates of differential entropy are affine functions of the log-determinant of a system’s covariance matrix. Examples include the multivariate normal distribution [@Ahmed1989], the multivariate $t$ distribution [@guerrero1996measure; @Nadarajah2005], and the multivariate Cauchy distribution [@Nadarajah2005]. For the $n$-variate normal distribution with covariance matrix $M$, for example, the differential entropy is [@Ahmed1989] $$\begin{aligned} h(M) = \frac{1}{2} \log(\det M) + \frac{n}{2}\left(1 + \log (2\pi)\right)\,, \label{eq:diff_ent}\end{aligned}$$ where the base of the logarithm can be any finite positive number[^1]. The logarithm of the covariance matrix is thus sufficient to approximate the differential entropy of several multivariate distributions. The principal submatrices of $M$ are covariance matrices of subsystems that correspond to subsets of coupled variables. One can compute the differential entropy of a subsystem by computing $h$ in for a principal submatrix of $M$. A system of $n$ coupled variables possesses $\binom{n}{k}\approx \frac{n^k}{k!}$ subsystems of $k$ variables; each of these subsystems corresponds to one of the $\binom{n}{k}$ principal $k\times k$ submatrices of $M$. The exact computation of the distribution of differential subsystem entropy or its moments thus requires one to compute $O(n^k)$ distinct determinants, an infeasible task for large $n$ and $k$. This task can be computationally prohibitive even for modestly-sized systems. To our knowledge, the largest system for which researchers have exactly computed the differential entropy of subsystems is a synthetic network with $n=12$ variables and subsystems with $k\leq 12$ variables [@Tononi1999]. To address this problem, we study the distribution of log-determinants of principal $k\times k$ submatrices. We refer to these log-determinants as *log-minors* of *size* $k$. As we noted above, these log-determinants are sufficient to determine the subsystem entropy for many important multivariate distributions. Knowledge of the properties of this distribution thus enables the derivation of bounds on the sampling error when estimating subsystem entropy in many applications. We show that, given a bound on the condition number of $M$, the standard error of a sample mean of differential entropy is independent of $n$ and sublinear in $k$, implying that one needs a sublinear number of samples in $k$ to ensure a desired accuracy. Our paper proceeds as follows. In , we introduce some notation that we use throughout this paper. In , we give several upper bounds on the tail and variance of the distribution of log-minors of a positive-definite matrix with bounded condition number. We present proofs for these bounds in and show numerical examples in . In , we apply our theorems to provide probabilistic guarantees on the sample mean and relative error, and we discuss implications for the design of practical schemes for estimating mean subsystem entropy. We conclude and discuss possible extensions in . Notation {#sec:notation} ======== Let $M \in {\mathbbm{R}}^{n\times n}$ be a positive-definite matrix. Let $\lambda_1(M)\geq\lambda_2(M)\cdots\geq\lambda_n(M)\geq 0$ be the eigenvalues of $M$. Because $M$ is positive definite, it is also nonsingular; its condition number is $\kappa(M) = \lambda_1(M)/\lambda_n(M)$. For a given index set $I\in[n]^k := \{1, \ldots, n\}^k$, the matrix $M_{I} := [M_{i,j}]_{i,j\in I}$ is the corresponding principal submatrix of $M$. For any fixed $k\leq n$, let $\mathcal{A}_k(M)$ denote the set of all such $k\times k$ submatrices of $M$, and let $A_k(M)$ denote a uniformly-random element of this set. We define a random variable ${Y_{k}(M)} := \log(\det A_k(M))$ and denote its empirical distribution by ${\mu}_{M,k}$. For convenience, we define $\wedge_{n,k}:=\min\{k,n-k\}$. Bounds on the distribution of log-minors {#sec:results} ======================================== In this section, we state bounds on the distribution ${\mu}_{M,k}$ for a positive-definite matrix $M$ with bounded condition number. We give upper bounds for the distribution’s support, variance, and right tail. We also show that we can improve these bounds if $M$ is diagonal. Let $M$ be a positive-definite $n\times n$ matrix with condition number $\kappa(M)\leq{\tilde\kappa}$. For every $r\geq 0$, we have $${\operatorname{Pr}}\left(|{Y_{k}(M)} -\mathbbm E [{Y_{k}(M)}]|\geq r\right)\leq 3\exp \left(-\frac{r}{\log{\tilde\kappa}}\sqrt{\frac{n}{k(n-k)}}\right)\,.\label{eq:ldi}$$ Furthermore, the variance of ${Y_{k}(M)}$ satisfies $${\operatorname{var}}({Y_{k}(M)})\leq 6\left(\frac{k(n-k)}{n}\right)(\log{\tilde\kappa})^2\,. \label{eq:ldi_varbound}$$ \[th:ldi\] The tail bound in does not guarantee that ${Y_{k}(M)}$ concentrates[^2] on $\mathcal A_k(M)$. This is because the bound in is increasing with respect to $k$ and asymptotically constant with respect to $n$. Indeed, for the bound to approach $0$ for a sequence $\{M_i\}$ of matrices, it is both necessary and sufficient that $\lim_{i\rightarrow \infty} \sqrt{k_i}\log \kappa_i(M) \rightarrow 0$. Because $k$ cannot be smaller than $1$, this condition requires the condition number $\kappa_i(M)$ to approach $1$. The condition $\lim_{i\rightarrow\infty}\kappa_i(M)\rightarrow 1$ severely constrains the sequence $\{M_i\}$. In that limit, all eigenvalues of $M$ are equal to each other and all log-minors are equal to $k\log\lambda_1$. Let $M$ be a positive-definite $n\times n$ matrix with condition number $\kappa(M)\leq{\tilde\kappa}$. For any $k$, the random variable ${Y_{k}(M)}$ and its distribution ${\mu}_{M,k}$ satisfy the following properties: 1. The distribution ${\mu}_{M,k}$ has bounded support that is contained in an interval whose length is no greater than $(\wedge_{n,k}\times\log{\tilde\kappa})$; and 2. the variance of ${Y_{k}(M)}$ satisfies $${\operatorname{var}}({Y_{k}(M)})\leq \frac{1}{4}\left(\wedge_{n,k}\times \log {\tilde\kappa}\right)^2\,. \label{eq:support_varbound}$$ \[th:support\] The variance bound in is much sharper than the one in . Both variance bounds are asymptotically constant with respect to $n$. For fixed $k$, the two variance bounds differ by a factor of $24$ in the large-$n$ limit. For even $n$ and $k\in\{1,n-1\}$, the bound on the variance in is sharp when $M$ is a $2\ell\times 2\ell$ (where $\ell=n/2$) diagonal matrix with entries $\lambda_1 = \cdots = \lambda_\ell = {\tilde\kappa}$ and $\lambda_{\ell+1} = \cdots = \lambda_{2\ell}=1$. When $M$ is diagonal, we can derive a variance bound that is sharper than the bounds in . Let $D$ be a positive-definite $n\times n$ diagonal matrix with condition number $\kappa(D)\leq{\tilde\kappa}$. The variance of ${Y_{k}(D)}$ satisfies $${\operatorname{var}}({Y_{k}(D)})\leq\frac{k}{4}\left(\frac{n-k}{n-1}\right)(\log{\tilde\kappa})^2\,. \label{eq:diag_varbound}$$ \[th:diag\] The two variance bounds in are asymptotically constant with respect to $n$ and converge to the same limiting value of $k(\log{\tilde\kappa})^2/4$. The variance bound for diagonal positive-definite matrices in is sharper than the variance bound for general positive-definite matrices in . The former differs from the latter by a factor of $\max\{k,n-k\}/(n-1)\leq 1$. For even $n$ and any $k\leq n$, the bound on the variance in is sharp when $M$ is a $2\ell\times 2\ell$ (where $\ell=n/2$) diagonal matrix with entries $\lambda_1 = \cdots = \lambda_\ell = {\tilde\kappa}$ and $\lambda_{\ell+1} = \cdots = \lambda_{2\ell}=1$. The sharpness of the bound for diagonal matrices indicates a limit to possible improvements for the variance bound for general positive-definite matrices. Specifically, one cannot hope to improve the variance bound in by more than a factor of $\max\{k,n-k\}/(n-1)$. The variance bound in is sharp for a diagonal matrix. This observation and several examples in motivate the following conjecture. Let $\mathcal M_{\kappa}$ be the set of positive-definite $n\times n$ matrices with condition number $\kappa$. For all $k<n$, $\kappa$, and $M\in\mathcal M_{\kappa}$, there exists a diagonal matrix $D\in\mathcal M_{\kappa}$ such that $${\operatorname{var}}({Y_{k}(M)}) \leq {\operatorname{var}}({Y_{k}(D)})\,. \label{eq:ineq}$$ \[conj:diag\] The variance bounds (see ) have important implications for the accuracy of sample means of log-minors. We discuss these implications in . Proofs of bounds on the distribution of log-minors {#sec:proofs} ================================================== Proof of {#sec:proof2} --------- To prove , we use Cauchy’s interlacing theorem and results on Markov chains on countable sets. Chatterjee and Ledoux (2009) previously used this approach to prove a concentration result for empirical cumulative eigenvalue spectra of Hermitian matrices [@Chatterjee2009]. \[prop:cauchy\] Let $M$ be a Hermitian $n\times n$ matrix, and let $A$ be a principal $(n-1)\times (n-1)$ submatrix of $M$. If $M$ has eigenvalues $\lambda_1(M)\geq\lambda_2(M)\geq\dots\geq\lambda_n(M)$ and $A$ has eigenvalues $\lambda_1(A)\geq\lambda_2(A)\geq\dots\geq\lambda_{n-1}(A)$, then $$\lambda_1(M)\geq\lambda_1(A)\geq\lambda_2(M)\geq\lambda_2(A)\geq\lambda_3(M)\geq\dots\geq\lambda_{n-1}(M)\geq\lambda_{n-1}(A)\geq\lambda_n(M)\,.\nonumber$$ \[prop:ledoux\] Let $(\Pi,\mu_{\mathcal A})$ be a reversible Markov chain on a finite or countable set $\mathcal A$. Let $(\Pi,\mu_{\mathcal A})$ have a spectral gap[^3] of $\overline{\lambda}>0$. It follows that , whenever $f: \mathcal A\rightarrow\mathbbm R$ is a function such that $$|||f|||^2_\infty:=\frac{1}{2}\sup_{A\in \mathcal A}\sum_{B\in \mathcal A} \left(f(A)-f(B)\right)^2\Pi(A,B)\leq 1\,,\nonumber$$ it is also true that $f$ is integrable with respect to $\mu_{\mathcal A}$ and that, for every $r\geq 0$, the probability measure $$\mu\left(\{f\geq \int_{\mathcal A} f\,d\mu_{\mathcal A} + r\}\right)\leq 3e^{-\frac{r}{2}\sqrt{\overline{\lambda}}}\,. \nonumber$$ The expected squared distance in $f$ between $A\in\mathcal A$ and its adjacent states in the Markov chain is $\sum_{B\in\mathcal A}\left(f(A)-f(B)\right)^2\Pi(A,B)$. One can thus think of $|||f|||_{\infty}^2$ as a measure of the expected squared distance between the greatest “outlier” $A$ and adjacent states in $\mathcal A$. We thus refer to $|||f|||_{\infty}^2$ as the *squared outlier deviation* of $f$ on $(\Pi,\mu_{\mathcal A})$. \[prop:gap\] Let $\mathcal S_n$ be the set of permutations of $n$ elements, and let $s\in\mathcal S_n$. Let the “random-transposition walk” be a reversible Markov chain $(\Pi, \mu_{\mathcal S_n})$ with kernel $$\begin{aligned} \Pi=\left\{\begin{matrix} 1/n\,,&\textrm{if } s'=s\,,\\ 2/n^2\,,&\textrm{if } s'=\tau s \textrm{ for some transposition }\tau\,,\\ 0\,,&\textrm{otherwise}\,. \end{matrix}\right. \nonumber\end{aligned}$$ The random-transposition walk has a spectral gap of $\overline \lambda=2/n$. Every principal $k\times k$ submatrix of $M$ is the top-left principal $k\times k$ submatrix of $M$ after a permutation of its rows and columns. We denote the permutated matrix by $s M$ and its top-left principal $k\times k$ submatrix by $\hat A_k(s M)$, where $s\in \mathcal S_n$ is a permutation of $n$ elements. For the top-left principal $k\times k$ submatrix, only the first $k$ elements of $s$ are relevant. There are $(n-k)!$ permutations $s\in \mathcal S_n$ that are identical in their first $k$ elements, so there is a $1$-to-$(n-k)!$ correspondence between $\mathcal A_k(M)$ and $\mathcal S_n$. Because of the correspondence between $\mathcal A_k(M)$ and $\mathcal S_n$, we obtain the same distribution for a function $f(A_k(M))$, where we choose $A_k(M)$ uniformly at random from $\mathcal A_k(M)$, and for $f(\hat A_k(sM))$, where we choose $s$ uniformly at random from $\mathcal S_n$. Let $f_{\alpha}: \mathcal S_n \rightarrow \mathbbm R$ be such that $$\label{eq:fdef} f_{\alpha}(s) := \alpha{\log(\det {\hat A_k(sM)})}$$ for some $\alpha\in\mathbbm R$. To find an upper bound on the squared outlier deviation for $f_{\alpha}$ on the random-transposition walk, we make two observations: 1. Consider two permutations, $s$ and $s'$, that are adjacent in the random-transposition walk; that is, $s'=\tau s$ for some transposition $\tau$. The determinant is invariant under basis transformation, so the value of $f_{\alpha}(s')$ can differ from $f_{\alpha}(s)$ only if $\tau$ is a transposition that swaps one of the first $k$ elements in $s$ with one of the last $n-k$ elements in $s$. There are $n^2$ possible transpositions for a sequence of $n$ elements; and $2k(n-k)$ of these transpositions swap one of the first $k$ elements of the sequence with one of the last $n-k$ elements of the sequence. Consequently, the fraction of transpositions that change the value of $f_{\alpha}$ has an upper bound of $$b_1:= \frac{2k}{n^2}(n-k)\,.\nonumber$$ 2. Using Cauchy’s interlacing theorem (see ), one can find an upper bound $b_2$ for $|f_{\alpha}(A)-f_{\alpha}(B)|$. For any $k<n$ and any pair $A, B\in\mathcal A_k(M)$, there exists a matrix $C\in A_{k+1}(M)$ such that $A$ and $B$ are principal submatrices of $C$. Cauchy’s interlacing theorem implies that 1. $\lambda_1(M)$ is an upper bound on the largest eigenvalue of $C$; 2. $\lambda_n(M)$ is a lower bound on the smallest eigenvalue of $C$; 3. $\sum_{i=1}^{k}\log\lambda_i(C)$ is an upper bound on ${\log(\det {A})}$ and ${\log(\det {B})}$; and 4. $\sum_{i=2}^{k+1}\log\lambda_i(C)$ is a lower bound on ${\log(\det {A})}$ and ${\log(\det {B})}$. Therefore, $$\begin{aligned} |{\log(\det {A})}-{\log(\det {B})}|\leq & \sum_{i=1}^{k}\log\lambda_i(C) - \sum_{i=2}^{k+1}\log\lambda_i(C)\\ \leq & \log\lambda_1(M) - \log\lambda_n(M) \leq \log {\tilde\kappa}\,.\end{aligned}$$ This upper bound for $|{\log(\det {A})}-{\log(\det {B})}|$ holds for arbitrary $A,B\in\mathcal A_k(M)$. We can thus set the upper bound to be $b_2=\alpha\log{\tilde\kappa}$. We obtain an upper bound for the squared outlier deviation of $f_{\alpha}$ of $$\begin{aligned} |||f_{\alpha}|||_\infty^2 &\leq \frac{1}{2}b_1b_2^2\\ &=k(n-k)\left(\frac{\alpha\log{\tilde\kappa}}{n}\right)^2\,.\end{aligned}$$ Let $\alpha'=n\log{\tilde\kappa}/\sqrt{k(n-k)}$. The function $f_{\alpha'}$ on $(\Pi,\mu_{\mathcal S_n})$ has a squared outlier deviation of $|||f_{\alpha'}|||^2_\infty\leq 1$. We can thus use the tail bound for functions on countable sets (see ) for $f_{\alpha'}$. Therefore, $${\operatorname{Pr}}(|f_{\alpha'}(\hat A_k(\sigma M)) -\mathbbm E[f_{\alpha'}(A_k(\sigma M)]|\geq \alpha'r)\leq 3e^{-\frac{\alpha'r}{2}\sqrt{\overline{\lambda}}}\,.$$ We can substitute $f_{\alpha'}(\hat A_k(\sigma M))$ in by $\alpha'{Y_{k}(M)}$, because of the correspondence between $\mathcal A_k(M)$ and $\mathcal S_n$. Applying , we obtain $$\begin{aligned} {\operatorname{Pr}}(|\alpha'{Y_{k}(M)} -\alpha'\mathbbm E [{Y_{k}(M)}]\geq \alpha'r)&\leq 3e^{-\frac{\alpha'r}{\sqrt{2n}}}\\ \implies{\operatorname{Pr}}(|{Y_{k}(M)} -\mathbbm E[{Y_{k}(M)}]|\geq r)&\leq 3\exp \left(-\frac{r}{\log{\tilde\kappa}}\sqrt{\frac{n}{k(n-k)}}\right)\,.\end{aligned}$$ This proves the first statement of (see ). We derive a bound on the variance of ${\log(\det {A})}$ from from a direct calculation. First, we write $$\begin{aligned} {\operatorname{var}}({Y_{k}(M)}) &= \mathbbm{E}[({Y_{k}(M)}-\mathbbm{E}[{Y_{k}(M)}])^2]\\ &= \int_{0}^\infty {\operatorname{Pr}}({Y_{k}(M)}-\mathbbm{E}[{Y_{k}(M)}])^2 \geq u)du\\ &= \int_{0}^\infty {\operatorname{Pr}}({Y_{k}(M)}-\mathbbm{E}[{Y_{k}(M)}])\geq \sqrt u\} du\,.\end{aligned}$$ Using the tail bound in , it follows that $$\begin{aligned} {\operatorname{var}}({Y_{k}(M)}) &\leq \int_{0}^\infty3\exp \left(-\frac{\sqrt{u}}{\log{\tilde\kappa}}\sqrt{\frac{n}{k(n-k)}}\right) du\\ & = 6\left(\frac{k(n-k)}{n}\right)(\log{\tilde\kappa})^2\end{aligned}$$ Proof of {#sec:proof1} --------- We prove using Cauchy’s interlacing theorem and Popoviciu’s inequality. \[prop:popoviciu\] Let $X$ be a real-valued random variable supported on the interval $[x_\mathrm{min}, x_{\mathrm{max}}]$. It then follows that $X$ has variance $${\operatorname{var}}(X)\leq (x_{\operatorname{max}}-x_{\operatorname{min}})^2/4\,.\nonumber$$ For a proof of this version of Popoviciu’s inequality, see Ref.[@stackPopoviciu]. For any finite $n$ and $k$, the set $\mathcal A_k(M)$ of principal $k\times k$ submatrices of an $n\times n$ matrix $M$ has finite cardinality $\binom{n}{k}$. It follows that the distribution of any function of $A_k(M)$ has finite support. We define an interval $[r_1,r_2]$ with $r_1:=\min{Y_{k}(M)}$ and $r_2:=\max{Y_{k}(M)}$, such that the support of ${\mu}_{M,k}$ is a finite subset of $[r_1,r_2]$. We can obtain any principal $k\times k$ submatrix $A$ of $M$ by removing $n-k$ row–column pairs from $M$. Successive applications of Cauchy’s interlacing theorem show that $\lambda_1(A)\leq\lambda_1(M)$ and $\lambda_k(A)\geq\lambda_n(M)$. It follows that $$\begin{aligned} r_1=k\log\lambda_n(M)\,, \qquad r_2=k\log\lambda_1(M)\,.\nonumber\end{aligned}$$ Therefore, $$\begin{aligned} r_2-r_1\leq k\left(\log{\tilde\kappa}+\log\lambda_n(M)\right)-k\log\lambda_n(M)=k\log {\tilde\kappa}\,.\nonumber\end{aligned}$$ If $k>n/2$, any two principal $k\times k$ submatrices share $2k-n$ rows and columns. They can thus differ in at most $n-k$ rows and columns. It follows that one can refine the lower and upper bounds on the support of ${\mu}_{M,k}$ so that $[r_1,r_2]\leq\wedge_{n,k}\times\log{\tilde\kappa}$. We have thus proven the first part of Theorem 1. Applying Popiviciu’s inequality to $X={Y_{k}(M)}$ with $x_{\operatorname{max}}-x_{\operatorname{min}}\leq r_2-r_1$ yields the variance bound in . Proof of {#sec:proof3} --------- For our proof of , we maximize ${\operatorname{var}}({Y_{k}(D)})$ with respect to the eigenvalues of $D$. Let $D$ be a positive-definite diagonal matrix with entries $\lambda_1(D)\geq\lambda_2(D)\geq\dots\geq\lambda_n(D)>0$. Define $x_i = \log \lambda_i(D)$ for each $i \in [n]$; and let $I \in [n]^k$. It then follows that $$\begin{aligned} {\log(\det {D_{I}})} &= \sum_{i \in I}x_{i}\,.\label{eq:xi}\end{aligned}$$ We now consider the function $v(x_1,x_2,\ldots,x_n) := {\operatorname{var}}({Y_{k}(D)})$. From , we see that every value of ${Y_{k}(D)}$ is a sum of a subset of the variables $x_i$. Therefore, the function $v(x_1,x_2,\ldots,x_n)$ is convex (i.e., concave up) in the variables $x_1,x_2,\ldots,x_n$. Furthermore, the variance is translation-invariant. We may therefore, without loss of generality, suppose that $x_n = 0$ (corresponding to $\lambda_n(D) = 1$) and $x_1 = \log\kappa(D)$ (corresponding to $\lambda_n(D) =\kappa(D)$). Consequently, the maximization of the variance ${\operatorname{var}}({Y_{k}(D)})$ amounts to the maximization of $v$ over the volume $[0,\log \kappa(D)]^n$ associated with an $n$-dimensional hypercube with edge length $\log \kappa(D)$. The solutions lie at the vertices of this hypercube. Therefore, $$\begin{aligned} x_i = \begin{cases} \log \kappa(D)\,, &\quad i \leq \ell \,, \\ 0 \,, &\quad \text{otherwise}\,, \end{cases}\end{aligned}$$ for some $\ell \in [n]$. We may now view ${Y_{k}(D)}/(\log \kappa(D))$ as a hypergeometric random variable on a population of size $n$ for which $\ell$ elements have the value $1$ and $n-l$ elements have the value $0$. The variance of this hypergeometric random variable is $$\begin{aligned} {\operatorname{var}}({Y_{k}(D)}) = \frac{k(n-k)}{n^2(n-1)}\left(n-\ell\right)\ell (\log \kappa(D))^2\,,\end{aligned}$$ which is maximal at $$\begin{aligned} \ell^* = \begin{cases} \frac{n}{2} \,, &\quad n \text{ even}\,, \\ \frac{n\pm1}{2}\,, &\quad n\text{ odd}\,. \end{cases}\end{aligned}$$ The maximal value of ${\operatorname{var}}({Y_{k}(D)})$ for even $n$ leads to the variance bound $$\begin{aligned} \label{eq:vbd} {\operatorname{var}}({Y_{k}(D)}) &\leq \frac{k(n-k)}{4(n-1)}(\log \kappa(D))^2 \\ &\leq\frac{k(n-k)}{4(n-1)}(\log {\tilde\kappa})^2\,. \end{aligned}$$ Comparing the maximal values of ${\operatorname{var}}({Y_{k}(D)})$ for even $n$ and for odd $n$ shows that is a variance bound for all $n$. Examples {#sec:examples} ======== In this section, we compare the tail of the distribution ${\mu}_{M,k}$ for several example matrices $M$ to the bounds in Theorems \[th:ldi\] and \[th:diag\]. We consider four examples of positive-definite $n\times n$ matrices with $n=20$ and fixed condition number $\kappa=3$. #### Example E1. Consider the diagonal matrix $M_{\textrm{E1}}$ that maximizes the variance of ${Y_{k}(M_{\textrm{E1}})}$. (See the proof of .) For even $n$, this matrix has eigenvalues $\lambda_1, \lambda_2, \ldots, \lambda_n$, where $\lambda_1,\ldots,\lambda_{n/2}={\tilde\kappa}$ and $\lambda_{n/2+1},\ldots,\lambda_n=1$. #### Example E2. Consider a diagonal matrix $M_{\textrm{E2}}$ with eigenvalues $\lambda_1, \lambda_2, \ldots, \lambda_n$. We set $\lambda_1=\kappa$ and $\lambda_n=1$. We draw $\lambda_2,\ldots,\lambda_{n-1}$ from a uniform distribution on $[1,\kappa]$. #### Example E3. We obtain a non-diagonal positive-definite matrix $M_{\textrm{E3}}$ with condition number $\kappa$ via an orthogonal transformation of $M_{\textrm{E1}}$. That is, $$M_{\textrm{E3}} := Q^{-1}M_{\textrm{E1}}Q\,,\nonumber$$ where $Q$ is an orthogonal matrix that we choose from the Haar measure over the group of orthogonal matrices. We use Stewart’s algorithm [@Stewart1980] to generate $Q$. #### Example E4. We again generate a random orthogonal matrix $Q$ using Stewart’s algorithm. We obtain another non-diagonal positive-definite matrix $M_{\textrm{E4}}:=Q^{-1}M_{\textrm{E2}}Q$ via an orthogonal transformation of $M_{\textrm{E2}}$. In , we show the empirical probability densities of ${Y_{k}(M)}$ for Examples E1, E2, E3, and E4 using four different values of $k$. For all four examples, we observe that the interval on which ${\mu}_{M,k}$ is supported shifts to the right for progressively larger $k$. The length of the supported interval increases with $\wedge_{n,k}$. For $k=5$ and $k=10$ — the cases in which $\wedge_{n,k}$ is larger than $1$ — the distribution ${\mu}_{M,k}$ are almost symmetric about ${\mathbbm{E}}[{Y_{k}(M)}]$ for all four examples. For Example E1, the distribution ${\mu}_{M,k}$ is symmetric about its mean for all examined values of $k$. Its density is nonzero at $\wedge_{n,k}+1$ equidistant points. [ccccccc]{} ------------------------------------------------------------------------ ------------------------------------------------------------------------ & & & &\ & & & & Theorem 1 & Theorem 2 & Theorem 3\ ------------------------------------------------------------------------ ------------------------------------------------------------------------ 1 & E1 & $0.549$ & $0.302$ & $6.880$ & $0.302$ & $0.302$\ ------------------------------------------------------------------------ ------------------------------------------------------------------------ & E2 & $0.689$ & $0.115$ & $6.880$ & $0.302$ & $0.302$\ ------------------------------------------------------------------------ ------------------------------------------------------------------------ & E3 & $0.683$ & $0.021$ & $6.880$ & $0.302$ & N/A\ ------------------------------------------------------------------------ ------------------------------------------------------------------------ & E4 & $0.739$ & $0.005$ & $6.880$ & $0.302$ & N/A\ ------------------------------------------------------------------------ ------------------------------------------------------------------------ 5 & E1 & $2.747$ & $1.191$ & $27.156$ & $1.509$ & $1.191$\ ------------------------------------------------------------------------ ------------------------------------------------------------------------ & E2 & $3.446$ & $0.454$ & $27.156$ & $1.509$ & $1.191$\ ------------------------------------------------------------------------ ------------------------------------------------------------------------ & E3 & $3.283$ & $0.091$ & $27.156$ & $1.509$ & N/A\ ------------------------------------------------------------------------ ------------------------------------------------------------------------ & E4 & $3.649$ & $0.020$ & $27.156$ & $1.509$ & N/A\ ------------------------------------------------------------------------ ------------------------------------------------------------------------ 10 & E1 & $5.493$ & $1.588$ & $36.208$ & $3.017$ & $1.588$\ ------------------------------------------------------------------------ ------------------------------------------------------------------------ & E2 & $6.893$ & $0.605$ & $36.208$ & $3.017$ & $1.588$\ ------------------------------------------------------------------------ ------------------------------------------------------------------------ & E3 & $6.213$ & $0.128$ & $36.208$ & $3.017$ & N/A\ ------------------------------------------------------------------------ ------------------------------------------------------------------------ & E4 & $7.176$ & $0.031$ & $36.208$ & $3.017$ & N/A\ ------------------------------------------------------------------------ ------------------------------------------------------------------------ 19 & E1 & $10.437$ & $0.302$ & $6.880$ & $0.302$ & $0.302$\ ------------------------------------------------------------------------ ------------------------------------------------------------------------ & E2 & $13.096$ & $0.115$ & $6.880$ & $0.302$ & $0.302$\ ------------------------------------------------------------------------ ------------------------------------------------------------------------ & E3 & $10.570$ & $0.022$ & $6.880$ & $0.302$ & N/A\ ------------------------------------------------------------------------ ------------------------------------------------------------------------ & E4 & $13.155$ & $0.008$ & $6.880$ & $0.302$ & N/A\ In , we show ${\mathbbm{E}}[{Y_{k}(M)}]$ and ${\operatorname{var}}({Y_{k}(M)})$ for the distributions in . We first consider the expectation of ${Y_{k}(M)}$. For all four examples, we observe that ${\mathbbm{E}}[{Y_{k}(M)}]$ increases with $k$. For all examined values of $k$, we see that ${\mathbbm{E}}[{Y_{k}(M_{\textrm{E4}})}]>{\mathbbm{E}}[{Y_{k}(M_{\textrm{E2}})}]>{\mathbbm{E}}[{Y_{k}(M_{\textrm{E3}})}]>{\mathbbm{E}}[{Y_{k}(M_{\textrm{E1}})}]$. Our observations thus suggest that the expectation of ${Y_{k}(M)}$ is large when we choose eigenvalues of $M$ uniformly at random from the interval $[1,\kappa]$ and small when we set half of the eigenvalues of $M$ to $1$ and the other half to $\kappa$. We now give several observations about the variance of ${Y_{k}(M)}$. For all examined values of $k$, we see that ${\operatorname{var}}({Y_{k}(M_{\textrm{E1}})})>{\operatorname{var}}({Y_{k}(M_{\textrm{E2}})})>{\operatorname{var}}({Y_{k}(M_{\textrm{E3}})})>{\operatorname{var}}({Y_{k}(M_{\textrm{E4}})})$. Our observation of larger ${\operatorname{var}}({Y_{k}(M)})$ for the examples with diagonal matrices (Examples E1 and E2) than for the examples with non-diagonal matrices (Examples E3 and E4) gives intuitive support for . Our observation that ${\operatorname{var}}({Y_{k}(M_{\textrm{E1}})})>{\operatorname{var}}({Y_{k}(M_{\textrm{E2}})})$ reflects the fact that Example E1 maximizes the variance in this case (see ). For all examined $k$, the value of the variance bound in (see ) is at least 12 times larger than the value of the variance bound in (see ). For $k=1$ and $k=19$, the cases in which $\wedge_{n,k}=1$, the value of the variance bound in is equal to the value of the variance bound for diagonal positive-definite matrices (). Additionally, it is sharp in Example E1. In Fig.\[fig:large\_deviations\], we show the empirical tails ${\operatorname{Pr}}(|{Y_{k}(M)}-\mathbbm E[{Y_{k}(M)}]|\geq r)$ for our four examples. We also show the tail bound B1 from and two Chebyshev bounds[^4], B2 and B3, which we obtain from the variance bounds in , respectively. Consistent with our observations in on ${\operatorname{var}}({Y_{k}(M)})$, we observe that the tail probability tends to be larger for the examples with diagonal matrices (Examples E1 and E2) than for the examples with non-diagonal matrices (Examples E3 and E4). The difference in functional form guarantees that the bound B1 intersects with the Chebyshev bound B2 at two values of $r$. If we denote these values by $r'$ and $r''>r'$, the bound B1 is sharper than B2 on $[0,r']$ and $[r'',\infty)$. In our observations, both bounds exceed the trivial bound ${\operatorname{Pr}}(|{Y_{k}(M)}-\mathbbm E[{Y_{k}(M)}]|\geq r)\leq 1$ on $[0,r']$. The value $r''$ lies outside the support of ${\operatorname{Pr}}(|{Y_{k}(M)}-\mathbbm E[{Y_{k}(M)}]|\geq r)$. We thus see that B1 is sharper than B2 only for values of $r$ for which neither bound is informative. For $k=1$ and $k=19$, the bounds B2 and B3 coincide and are sharp at $r=(\log\kappa)/2$ when $M=M_{\textrm{E1}}$. For $k=5$ and $k=10$, the bound (B3) for diagonal positive-definite matrices is sharper than the bound (B2) for general positive-definite matrices. The difference between the two bounds is most visible for $k=10$, which is the case that maximizes $\wedge_{n,k}$. Estimating mean subsystem entropy {#sec:sampling} ================================= We now consider the implications of our results in Section \[sec:results\] for the problem of estimating the mean subsystem entropy of a given system of coupled variables. When the joint distribution of variables is a multivariate normal distribution, one can compute the differential entropy of a subsystem by applying to the corresponding sub-covariance matrix. We are interested in the *mean subsystem entropy* ${\mathbbm{E}}[h(A_k(M))]$ for subsystems of $k$ variables. As we noted previously, the large number of subsystems for even modest values of $n$ and $k$ render it prohibitive to exactly compute ${\mathbbm{E}}[h(A_k(M))]$. Fortunately, the tail and variance bounds in Section \[sec:results\] allow us to instead provide sampling guarantees, through which one can achieve a prescribed sampling accuracy. We give upper bounds on the standard error and on the coefficient of variation for both a sample mean of ${Y_{k}(M)}$ and a sample mean of subsystem entropy. Fix a subsystem size $k$ and sample size $q\geq 1$. The $q$-sample mean of ${Y_{k}(M)}$ is $$\begin{aligned} S_Y := \frac{1}{q}\sum_{i = 1}^q {\log(\det {A_i})}\,,\end{aligned}$$ where we choose each $A_i$ uniformly at random from $\mathcal A_k(M)$. The $q$-sample mean of subsystem entropy is $$\begin{aligned} S_h := \frac{1}{q}\sum_{i = 1}^q h(A_i)\,.\end{aligned}$$ We use $S_Y$ and $S_h$ as estimators of the population means ${\mathbbm{E}}[{Y_{k}(M)}]$ and ${\mathbbm{E}}[h(A_k(M))]$, respectively. These estimators are unbiased, as ${\mathbbm{E}}[S_Y]={\mathbbm{E}}[{Y_{k}(M)}]$ and ${\mathbbm{E}}[S_h]={\mathbbm{E}}[h(A_k(M))]$. A measure of reliability of an estimator is the *standard error*, which one computes as the estimator’s standard deviation. Because $h(A_k(M))$ differs from ${Y_{k}(M)}/2$ by a constant, the sample mean $S_h$ has the standard error $$\begin{aligned} \hat{\sigma}(S_h) = \frac{1}{2}\hat{\sigma}(S_Y) = \frac{1}{2}\sqrt{\frac{{\operatorname{var}}\left({Y_{k}(M)}\right)}{q}}\,.\end{aligned}$$ We may therefore use the bounds of to derive bounds on the standard error for $S_Y$ and $S_h$. Let $M$ be a covariance matrix of an $n$-variate normal distribution, and suppose that the condition number of $M$ satisfies $\kappa(M)\leq{\tilde\kappa}$. Let $S_h$ be the $q$-sample mean of the entropy of subsets of $k$ variables; and let $S_Y$ be the $q$-sample mean of log-determinants of $k\times k$ principal submatrices of $M$. It then follows, for any subsystem size $k$, that the standard error of the mean subsystem entropy is $\hat\sigma(S_h)=\hat\sigma(S_Y)/2$ and that $\hat\sigma(S_Y)$ satisfies $$\hat{\sigma}(S_Y)\leq\sqrt{\frac{6k(n-k)}{qn}}\log {\tilde\kappa}\,,\label{eq:se1}$$ and $$\hat{\sigma}(S_Y)\leq\frac{1}{2}\sqrt{\frac{\wedge_{n,k}}{q}}\log {\tilde\kappa}\,.\label{eq:se2}$$ Furthermore, if $M$ is diagonal, $$\hat{\sigma}(S_Y)\leq\frac{1}{2}\sqrt{\frac{k(n-k)}{q(n-1)}}\log{\tilde\kappa}\,.\label{eq:se3}$$ The *coefficient of variation* ${c_v(S)}$ is another measure of reliability for estimators. It measures the size of the typical error of an estimator $S$ as a fraction of the magnitude of ${\mathbbm{E}}[S]$. As a formula, it is given by $$\begin{aligned} {c_v(S)} := \frac{\hat{\sigma}(S)}{{\mathbbm{E}}[S]}\,. \label{eq:cv_def}\end{aligned}$$ The coefficient of variation for $S_Y$ arises from the standard deviation of the relative error $$\begin{aligned} \mathcal{E} := {\left|\frac{{Y_{k}(M)} - {\mathbbm{E}}[{Y_{k}(M)}]}{{\mathbbm{E}}[{Y_{k}(M)}]}\right|}\,\end{aligned}$$ of ${Y_{k}(M)}$ because $$\begin{aligned} \frac{\hat{\sigma}(S_Y)}{{\mathbbm{E}}[S_Y]} =\frac{\hat{\sigma}({Y_{k}(M)})- {\mathbbm{E}}[{Y_{k}(M)}]}{{\mathbbm{E}}[{Y_{k}(M)}]\sqrt{q}} =\hat{\sigma}\left(\left|\frac{{Y_{k}(M)}-{\mathbbm{E}}[{Y_{k}(M)}]}{{\mathbbm{E}}[{Y_{k}(M)}]}\right|\right)/\sqrt{q} =\hat{\sigma}(\mathcal E)/\sqrt{q}\,. \end{aligned}$$ For a multivariate Gaussian distribution, the following corollaries give bounds on the coefficient of variation for the sample mean of log-minors and for the sample mean of subsystem entropy. \[cor:rel\_err\] Let $\lambda_1(M),\ldots,\lambda_n(M)$ be the eigenvalues of $M$; we order them from largest to smallest. Let $\ell(M) = \min\{{\left|\log \lambda_1(M)\right|}, {\left|\log \lambda_n(M)\right|}\}$. If $\ell(M)\neq0$, the coefficient of variation for a $q$-sample mean $S_Y$ of ${Y_{k}(M)}$ satisfies $$\begin{aligned} {c_v(S_Y)} \leq \frac{\log \hat{\kappa}}{\ell(M)}\sqrt{\frac{6(n-k)}{qkn}} \label{eq:cvy1} \end{aligned}$$ and $$\begin{aligned} {c_v(S_Y)} \leq \frac{\log \hat{\kappa}}{2\ell(M)}\sqrt{\frac{\wedge_{n,k}}{qk^2}}\,.\label{eq:cvy2} \end{aligned}$$ This corollary follows from . We use as upper bounds on the numerator. For all $k$, a lower bound on the denominator is ${\left|{\mathbbm{E}}[Y_k(M)]\right|} \geq k\ell(M)$. For an $n$-variate Gaussian distribution with covariance matrix $M$, the coefficient of variation $c_v(S_h)$ for a $q$-sample mean $S_h$ of subsystem entropy satisfies $$\begin{aligned} c_v(S_h) \leq \frac{2\log \hat{\kappa}}{\ell(M) + \log 2e\pi}\sqrt{\frac{6(n-k)}{qkn}} \label{eq:cvh1} \end{aligned}$$ and $$\begin{aligned} c_v(S_h) \leq \frac{\log \hat{\kappa}}{\ell(M) + \log 2e\pi}\sqrt{\frac{\wedge_{n,k}}{qk^2}}\,.\label{eq:cvh2} \end{aligned}$$ We derive this result from ; we use to bound the numerator, and we use to bound the expectation in the denominator. The bounds in are sharper bounds than . From , we see that both ${c_v(S_Y)}$ and ${c_v(S_h)}$ decay in proportion to $\sqrt{k}$. Indeed, under a certain regularity condition (which we specify in ), the coefficient of variation decays to $0$ in the limit of large $n$ and large $k$. \[cor:rel\_err\_concentration\] Let $\{M_i\}$ be a sequence of positive-definite matrices of dimension $n(i)$. Let $k = k(i) \leq n(i)$ be a function of $i$. Suppose that the sequence $$\begin{aligned} a_i := \sqrt{k(i)} \ell(M_i) \label{eq:criterion} \end{aligned}$$ is nondecreasing and unbounded. It then follows that $c_v(S_Y) \rightarrow 0$ and $\mathcal{E}$ converges in probability to $0$ as $i$ becomes large. A sufficient condition for the concentration of $\mathcal E$ is that the sequence $\{M_i\}$ has fixed condition number and the smallest eigenvalue $\lambda_n$ is bounded away from both $0$ and $1$. Formally, the latter condition is $$\begin{aligned} \label{eq:ccon} \text{there exists} \,\, \delta > 0 \,\, \text{such that, for all} \,\, i\,, \text{ we have}\,\, \lambda_{n(i)}(M_i) \in [\delta, 1 - \delta]\cap[1+\delta, \infty)\ \,. \end{aligned}$$ A popular model for sample covariance matrices is the Wishart ensemble[^5]. A sequence $\{M_i\}$ of Wishart matrices can satisfy the condition in if the ratio $c:=n/n_f$ of the number $n$ of variables and the number $n_f$ of degrees of freedom is $c\notin\{1/4,1\}$ [@Katzav2010; @Majumdar2010]. One can use these bounds on the standard error to choose a sample size $q$ that guarantees a desired accuracy of a sample mean. In Fig.\[fig:sampling\_error\], we show our bounds on the standard error and the coefficient of variation of $S_Y$ and $S_h$ with $q=2000k$ and $\ell(M)=1$. In the left panels, we show the bounds B1 from and B2 from on the standard error of $S_Y$ and $S_h$. In the right panels, we show the bounds B1’ (see ) and B2’ (see ) on the coefficient of variation of $S_Y$ and $S_h$. In panels [([**A**]{})]{} and [([**B**]{})]{}, we vary the system size $n$ for fixed subsystem size $k$. We observe for $n\leq 2k$ that the values of the bounds B2 and B2’ increase with $n$. For $n>2k$, the bounds B2 and B2’ are independent of $n$. The bounds B1 and B1’ are less sharp than the bounds B2 and B2’. The values of B1 and B1’ increase with $n$ and approach their asymptotic values from below. For example, the bound B1 for $\hat\sigma(S_Y)$ has a limiting value of $\sqrt{6k/q}\times\log{\tilde\kappa}\approx0.06$. In panels [([**C**]{})]{} and [([**D**]{})]{}, we vary $k$ for fixed $n$. We observe for $k\leq n/2$ that the value of the bound B2 on the standard error is independent of $k$. For $k>n/2$, the value of B2 decreases with increasing $k$ and is $0$ for $k=n$. The bound B1 is less sharp than B2. Its value decreases with increasing $k$ for any $k\leq n$. The values of the bounds B1’ and B2’ on the coefficient of variation decrease with increasing subsystem size and are $0$ for $k=n$. In panels [([**E**]{})]{} and [([**F**]{})]{}, we vary $k$ for fixed ratio $k/n$, and we observe that the bounds on the standard error are independent of $k$ if the ratio $k/n$ is constant. The values of the bounds B1’ and B2’ decrease with increasing $k$. This is consistent with our previous observation that ${c_v(S_Y)}$ vanishes if the sequence $a_i$ (see ) becomes unbounded. It is important to note that all of our bounds on the standard error and the coefficient of variation of $S_Y$ and $S_h$ are asymptotically constant in $n$. It is thus not necessary to sample proportionally more minors from a larger matrix. Instead, to guarantee a desired accuracy of a sample mean of log-minors or subsystem entropy, one can choose $q$ to be a function of $k$. To ensure that the standard error is constant or decreases with growing $n$ and $k$, it is sufficient to choose $q$ in linear proportion to $k$. When the smallest and largest eigenvalues of a system’s correlation matrix are fixed, one can ensure that the coefficient of variation is constant or decreasing with growing $n$ and $k$ by choosing $q$ in linear proportion to $k^{-1}$. Conclusions {#sec:conclusions} =========== We examined the problem of estimating the mean subsystem entropy of a system of $n$ coupled variables with covariance matrix $M$. When the joint distribution of a system’s variables is an $n$-variate Gaussian, $t$, or Cauchy distribution, the mean differential entropy of subsystems is an affine function of the log-minors of the covariance matrix [@Ahmed1989; @Nadarajah2005]. We derived tail and variance bounds on the distribution of log-minors of fixed size of a positive-definite matrix with bounded condition number. Using our variance bounds, we provided upper bounds on the standard error and on the coefficient of variation of both the sample mean of log-minors and the sample mean of subsystem entropy. Our results indicate that, despite the rapid growth of the number of subsystems with $n$, the accuracy of these sample means is asymptotically independent of a system’s size. Instead, it is sufficient to increase the number of samples in linear proportion to the size of subsystems to achieve a desired sampling accuracy. Our results are salient to studies that use mean subsystem entropy to examine systems of coupled variables [@Koetter2003; @Tononi1994; @Tononi1999]. Even for a system with as few as $50$ variables, sampling just 0.001% of its subsystem entropies can require the computation of over a billion log-determinants. Using the largest and smallest eigenvalues of a system’s covariance matrix to determine the number of samples that are needed to achieve a prescribed accuracy for a sample mean can thus facilitate a quantitative study of mean subsystem entropy when it would otherwise be impossible. Throughout our paper, we relied only on knowledge of the largest and smallest eigenvalues of a system’s covariance matrix. We expect that it is possible to derive sharper bounds than our current results when one knows the complete spectrum of a system’s covariance matrix, likely by relying on Cauchy’s interlacing theorem () to control the log-minors. We presented two bounds on the variance of a log-minor that we choose uniformly at random from the set of log-minors of size $k$ of an $n\times n$ positive-definite matrix. The variance bound in is sharper than the one in , but either bound is sufficient to deduce that the accuracy of a sample mean of subsystem entropy is asymptotically independent of a system’s size and that one can achieve a prescribed accuracy by choosing the number of samples in linear proportion to the size of subsystems. The proof of our first bound (see ) relies on the existence of an upper bound for the difference between ${\log(\det {A})}$ for two different principal submatrices $A\in\mathcal A_k(M)$ and the invariance of ${\log(\det {A})}$ under a basis transformation of $A$. The proof of our second bound (see ) relies on the existence of an upper bound and a lower bound for the support of the distribution ${\mu}_{M,k}$. Similar bounds and the invariance under basis transformation hold for several other matrix properties, including the largest and smallest eigenvalues. It is thus plausible that one can derive similar results for the standard error and coefficient of variation for many spectral properties of principal submatrices. For example, Chatterjee and Ledoux (2009) proved a large-deviation inequality for the empirical cumulative eigenvalue distribution of principal submatrices of Hermitian matrices [@Chatterjee2009]. These and other variance and tail bounds on submatrix properties offer welcoming possibilities to enhance computational studies that characterize complex systems based on the mean properties of their subsystems. For example, they can provide guarantees for linear sketching techniques, which are relevant for data dimensionality reduction. They can also facilitate the use of methods of spectral graph analysis in the study of subgraphs, graphlets, and motifs in networks. Acknowledgements {#acknowledgements .unnumbered} ================ We thank Clément Canonne, Kameron Decker Harris, Michael Neely, and participants of the *IPAM Quantitative Linear Algebra Tutorials* for helpful discussions. A.C.S. was supported by the Clarendon Fund, e-Therapeutics plc, and funding from the Engineering and Physical Sciences Research Council under grant number EP/L016044/1. P.S.C. was supported by the National Science Foundation under Graduate Research Fellowship Grant 1122374. [36]{} \[1\][\#1]{} \[1\][`#1`]{} urlstyle \[1\][doi: \#1]{} S. S. Skiena. *[The Data Science Design Manual]{}*. Springer, Cham, Switzerland, 2017. M. E. J. Newman. *Networks*. Oxford University Press, Oxford, United Kingdom, 2018. S. H. Strogatz. *[Nonlinear Dynamics and Chaos: With Applications to Physics, Biology, Chemistry, and Engineering]{}*. Westview Press, Boulder, CO, USA, 2018. G. Tononi, O. Sporns, and G. M. Edelman. A measure for brain complexity: [R]{}elating functional segregation and integration in the nervous system. *Proceedings of the [N]{}ational [A]{}cademy of [S]{}ciences of the [U]{}nited [S]{}tates of [A]{}merica*, 910 (11):0 5033–5037, 1994. G. Tononi, O. Sporns, and G. M. Edelman. Measures of degeneracy and redundancy in biological networks. *[Proceedings of the National Academy of Sciences of the United States of America]{}*, 960 (6):0 3257–3262, 1999. M. De Lucia, M. Bottaccio, M. Montuori, and L. Pietronero. Topological approach to neural complexity. *[Physical Review E]{}*, 710 (1):0 016114, 2005. M. Randles, D. Lamb, E. Odat, and A. Taleb-Bendiab. Distributed redundancy and robustness in complex systems. *[Journal of Computer and System Sciences]{}*, 770 (2):0 293–304, 2011. Y. Li, G. Dwivedi, W. Huang, M. L. Kemp, and Y. Yi. Quantification of degeneracy in biological systems for characterization of functional interactions between modules. *[Journal of Theoretical biology]{}*, 302:0 29–38, 2012. Y. Li and Y. Yi. Systematic measures of biological networks [II]{}: [D]{}egeneracy, complexity, and robustness. *[Communications on Pure and Applied Mathematics]{}*, 690 (10):0 1952–1983, 2016. S. S. Shen-Orr, R. Milo, S. Mangan, and U. Alon. Network motifs in the transcriptional regulation network of [Escherichia coli]{}. *Nature genetics*, 310 (1):0 64, 2002. R. Milo, S. Shen-Orr, S. Itzkovitz, N. Kashtan, D. Chklovskii, and U. Alon. Network motifs: [S]{}imple building blocks of complex networks. *Science*, 2980 (5594):0 824–827, 2002. D. P. Woodruff. Sketching as a tool for numerical linear algebra. *[Foundations and Trends in Theoretical Computer Science]{}*, 100 (1–2):0 1–157, 2014. D. P. Francis and K. Raimond. A practical streaming approximate matrix multiplication algorithm. *[Journal of King Saud University-Computer and Information Sciences]{}*, 2018. T. Sarlos. Improved approximation algorithms for large matrices via random projections. In *[47th Annual IEEE Symposium on Foundations of Computer Science (FOCS)]{}*, pages 143–152. IEEE, 2006. K. L. Clarkson and D. P. Woodruff. Numerical linear algebra in the streaming model. In *[Proceedings of the 41st Annual ACM Symposium on Theory of Computing]{}*, pages 205–214. ACM, 2009. A. Kyrillidis, M. Vlachos, and A. Zouzias. Approximate matrix multiplication with application to linear embeddings. In *[IEEE International Symposium on Information Theory (ISIT)]{}*, pages 2182–2186. [IEEE]{}, 2014. E. Liberty. Simple and deterministic matrix sketching. In *[Proceedings of the 19th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining]{}*, pages 581–588. ACM, 2013. D. N. Page. Average entropy of a subsystem. *[Physical Review Letters]{}*, 710 (9):0 1291–1294, 1993. S. Sen. Average entropy of a quantum subsystem. *[Physical Review Letters]{}*, 770 (1):0 1–3, 1996. R. Koetter and M. M[é]{}dard. An algebraic approach to network coding. *[IEEE/ACM Transactions on Networking (TON)]{}*, 110 (5):0 782–795, 2003. A. E. Teschendorff, P. Sollich, and R. Kuehn. Signalling entropy: [A]{} novel network-theoretical framework for systems analysis and interpretation of functional omic data. *Methods*, 670 (3):0 282–293, 2014. N. A. Ahmed and D. V. Gokhale. Entropy expressions and their estimators for multivariate distributions. *[IEEE Transactions on Information Theory]{}*, 350 (3):0 688–692, 1989. J.-L. Guerrero-Cusumano. A measure of total variability for the multivariate $t$ distribution with applications to finance. *[Information Sciences]{}*, 920 (1–4):0 47–63, 1996. S. Nadarajah and S. Kotz. Mathematical properties of the multivariate $t$ distribution. *Acta Applicandae Mathematica*, 890 (1–3):0 53–84, 2005. M. Ledoux. *[The Concentration of Measure Phenomenon]{}*. American Mathematical Society, Providence, RI, USA, 2001. S. Chatterjee and M. Ledoux. An observation about submatrices. *[Electronic Communications in Probability]{}*, 140 (48):0 495–500, 2009. S.-G. Hwang. Cauchy’s interlace theorem for eigenvalues of [H]{}ermitian matrices. *[The American Mathematical Monthly]{}*, 1110 (2):0 157–159, 2004. P. Diaconis and M. Shahshahani. Generating a random permutation with random transpositions. *Zeitschrift f[ü]{}r [W]{}ahrscheinlichkeitstheorie und verwandte [G]{}ebiete*, 570 (2):0 159–179, 1981. T. Popoviciu. Sur les équations algébriques ayant toutes leurs racines réelles. *[Mathematica]{}*, 9:0 129–145, 1935. R. Sharma, M. Gupta, and G. Kapoor. Some better bounds on the variance with applications. *[Journal of Mathematical Inequalities]{}*, 40 (3):0 355–363, 2010. P. C. Marques. Variance of a bounded random variable. Cross Validated. URL <https://stats.stackexchange.com/q/50552>. (version: 2016-04-23). G. W. Stewart. The efficient generation of random orthogonal matrices with an application to condition estimators. *[SIAM Journal on Numerical Analysis]{}*, 170 (3):0 403–409, 1980. J. K. Blitzstein and J. Hwang. *[Introduction to Probability]{}*. CRC Press, Boca Raton, FL, USA, 2014. N. Goodman. The distribution of the determinant of a complex wishart distributed matrix. *[The Annals of Mathematical Statistics]{}*, 340 (1):0 178–180, 1963. E. Katzav and I. P. Castillo. Large deviations of the smallest eigenvalue of the [W]{}ishart-[L]{}aguerre ensemble. *[Physical Review E]{}*, 820 (4):0 040104, 2010. S. N. Majumdar. Extreme eigenvalues of [W]{}ishart matrices: [A]{}pplication to entangled bipartite system. In G. Akemann, J. Baik, and P. Di Francesco, editors, *[The Oxford Handbook of Random Matrix Theory]{}*. Oxford University Press, Oxford, United Kingdom, 2011. [^1]: The base $b$ of the logarithm determines the units of entropy. If one chooses $b=2$, one measures entropy values in bits. If one chooses $b=e$, one measures entropy values in nats. [^2]: Ledoux defined concentration of measure in Ref.[@Ledoux2001] (on page 3) as follows. Let $(X,d)$ be a metric space with probability measure $\mu$ on Borel sets of $(X,d)$. The *concentration function* is defined as $\alpha_{(X,d,\mu)}:=\sup\{1-\mu(A_r);A\subset X, \mu(A)\geq\frac{1}{2}\}$, where $r>0$ and $A_r:=\{x\in X;d(x,A)<r\}$ is the open $r$-neighborhood of $A$. The measure $\mu$ has *normal concentration* on $(X,d)$ if there are constants $c$ and $C$ such that $\alpha_{(X,d,\mu)}\leq C\exp(-cr^2)$ for every $r$. [^3]: The spectral gap (also called the “Poincaré constant”) of a Markov chain $(\Pi,\mu_{\mathcal A})$ on a space $\mathcal A$ is the constant $\overline{\lambda}$ such that, for all functions $f$, we have $\overline{\lambda}\times{\operatorname{var}}_\mu(f)\leq\frac{1}{2}\sum_{A,B\in\mathcal A}[f(A)-f(B)]^2\Pi(A,B)\mu(\{A\})$. See, for example, Ref.[@Ledoux2001] (page 50). [^4]: We can obtain a tail bound from a variance bound by using Chebyshev’s inequality [@Blitzstein2014] (page 429), $\operatorname{Pr}(|X-\mathbbm E[X]|\geq r)\leq \frac{{\operatorname{var}}(X)}{r^2}$, for an integrable random variable $X$ and $r\in\mathbbm R_+$. [^5]: The Wishart ensemble $W_n(V,n_f)$ with scale matrix $V$ and $n_f$ degrees of freedom is the ensemble of random matrices $M:=n_f^{-1}\sum_{i=1}^{n_f}X_{i}^TX_{i}$, where the $X_1,X_2,\dots,X_{n_f}$ are $n_f$ realizations of an $n$-variate random variable with 0-mean Gaussian distribution $N_n(0,V)$ [@Goodman1963; @Katzav2010].

ArXiv

--- abstract: 'The ultra-strong light-matter coupling regime has been demonstrated in a novel three-dimensional inductor-capacitor (LC) circuit resonator, embedding a semiconductor two-dimensional electron gas in the capacitive part. The fundamental resonance of the LC circuit interacts with the intersubband plasmon excitation of the electron gas at $\omega_c = 3.3$ THz with a normalized coupling strength $2\Omega_R/\omega_c = 0.27$. Light matter interaction is driven by the quasi-static electric field in the capacitors, and takes place in a highly subwavelength effective volume $V_{\mathrm{eff}} = 10^{-6}\lambda_0^3$ . This enables the observation of the ultra-strong light-matter coupling with $2.4\times10^3$ electrons only. Notably, our fabrication protocol can be applied to the integration of a semiconductor region into arbitrary nano-engineered three dimensional meta-atoms. This circuit architecture can be considered the building block of metamaterials for ultra-low dark current detectors.' author: - Mathieu Jeannin - Giacomo Mariotti Nesurini - Stéphan Suffit - Djamal Gacemi - Angela Vasanelli - Lianhe Li - Alexander Giles Davies - Edmund Linfield - Carlo Sirtori - Yanko Todorov bibliography: - 'Bib\_USC-THz-LC.bib' title: 'Ultra-Strong Light-Matter Coupling in Deeply Subwavelength THz LC resonators' --- Metamaterials were introduced to enable new electromagnetic properties of matter which are not naturally found in nature. Celebrated examples of such achievements are, for instance, negative refraction [@shelby_experimental_2001] and artificial magnetism.[@pendry_magnetism_1999] The unit cells of metamaterials are artificially designed meta-atoms that have dimensions ideally much smaller than the wavelength of interest $\lambda_0$.[@cai_optical_2010] Such meta-atoms act as high frequency inductor-capacitor (LC) resonators which sustain a resonance close to $\lambda_0\propto\sqrt{LC}$.[@cai_optical_2010] The resonant behaviour, occurring into highly subwavelength volumes, generates high electromagnetic field intensities which, as pointed out by the seminal paper of @pendry_magnetism_1999, [@pendry_magnetism_1999] are crucial to implement artificial electromagnetic properties of a macroscopic ensemble of meta-atoms. Moreover, the ability to control and enhance the electromagnetic field at the nanoscale is beneficial for optoelectronic devices, such as nano-lasers [@wang_unusual_2017] electromagnetic sensors [@chen_metamaterials_2012; @belacel_optomechanical_2017; @alves_mems_2018] and detectors. [@shrekenhamer_experimental_2012; @viti_nanowire_2014; @luxmoore_graphenemetamaterial_2016; @palaferri_ultra-subwavelength_2016; @palaferri_room-temperature_2018] For instance, metamaterial architectures have lead to a substantial decrease of the thermally excited dark current in quantum infrared detectors, resulting in higher temperature operation. [@palaferri_ultra-subwavelength_2016; @palaferri_room-temperature_2018] The LC circuit can be seen as a quantum harmonic oscillator sustaining vacuum electric field fluctuations that scale as $1/V_{\mathrm{eff}}^{1/2}$, where $V_{\mathrm{eff}}$ is the effective volume of the capacitive parts. [@louisell_quantum_1990] For an emitter/absorber inserted between the capacitor plates, the light-matter interaction is proportional to $1/V_{\mathrm{eff}}^{1/2}$, and thus strongly enhanced. Fundamental electro-dynamical phenomena, such as the Purcell effect[@purcell_resonance_1946] and strong light-matter coupling regime[@yamamoto_semiconductor_2000] can therefore be observed. In the strong coupling regime, energy is reversibly exchanged between the matter excitation and the electromagnetic resonator at the Rabi frequency $\Omega_R$. This results in an energy splitting of the circuit resonance into two polaritons states separated by $2\hbar\Omega_R$. The regime of strong coupling has been observed in many physical systems which have been reviewed e.g. in Refs. , and some specific realizations with metamaterial resonators were achieved in the sub-THz, [@scalari_ultrastrong_2012] THz[@todorov_ultrastrong_2010; @strupiechonski_sub-diffraction-limit_2012; @dietze_ultrastrong_2013] and the Mid-IR[@benz_control_2015; @askenazi_ultra-strong_2014; @askenazi_midinfrared_2017] part of the spectrum. In these systems, the highly subwavelength interaction volumes combined with the collective effect of $N_e$ identical electronic transitions result into high coupling constants $\Omega_R\propto(N_e/V_{\mathrm{eff}})^{1/2}$, and allow reaching the ultra-strong coupling regime where the Rabi splitting becomes of the same order of magnitude as the frequency of the material excitation $\tilde{\omega}$, $2\Omega_R/\tilde{\omega}\approx1$. [@ciuti_quantum_2005] Since there is virtually no lower limit for the interaction volume $V_{\mathrm{eff}}$ in LC resonators, the fascinating regime of ultra-strong coupling can be realized in structures having few electrons only. [@todorov_few-electron_2014; @keller_few-electron_2017] In such limit, the effective bosonization procedure employed to describe the properties of the two-dimensional electron gas breaks down, and one can investigate the unique regime where the few electrons in the system have to be exactly treated as fermions.[@todorov_few-electron_2014] Recently, the ultra-strong coupling regime with a small number of electrons has been experimentally observed by coupling transitions between Landau levels in a two dimensional electron gas under a high magnetic field and nanogap complementary bow-tie antennas, with a record low number of 80 electrons.[@keller_few-electron_2017] Those studies were performed in the sub-THz part of the spectrum (300GHz) using resonators based on a planar geometry. Here, we demonstrate a three-dimensional metamaterial architecture that has the potential to go beyond this limit in the THz range (3THz), without the need for a magnetic field. Our metamaterial allows confining the electric field in all directions of space into nanoscale volumes, on the order of $V_{\mathrm{eff}}=10^{-6}\lambda_0^3$. The resonance of the structure is coupled to an intersubband (ISB) transition of high density electron gas in the ground state of semiconductor quantum wells (QWs). A relative Rabi frequency of $2\Omega_R/\tilde{\omega}=0.27$ is attained with a record low overall number of electrons $N_e\approx10^3$ for intersubband systems. Other reports on coupled LC resonators-ISB transitions systems in the THz spectral range reached similar coupling constants of $2\Omega_R/\tilde{\omega}=0.2$ with a much greater mode volume and electron number ($V_{\mathrm{eff}}\approx10^{-5}\lambda_0^3$ and $N_e=4.6\times 10^4$).[@paulillo_room_2016] Comparable number of coupled electrons have been reported in the MIR spectral range[@benz_strong_2013; @malerba_towards_2016] using very small mode volume cavities with $V_{\mathrm{eff}}\approx 6-9 \times 10^{-6}\lambda_0^3$, but at the expense of reducing the coupling constant ($2\Omega_R/\tilde{\omega}=0.1$ in Ref.  and $2\Omega_R/\tilde{\omega}=0.05$ in Ref. ). We use the dependence of the polariton splitting on the effective mode volume as a near field probe to estimate the highly subwavelength volume of our resonators in comparison with reference microcavity systems. [@todorov_ultrastrong_2010; @feuillet-palma_extremely_2012] These results are obtained thanks to a novel fabrication process that allows structuring metal patterns on both sides of a very thin semiconductor layer, which opens many degrees of freedom in the design and functionalization of 3D metamaterial architectures into optoelectronic devices. {width="\linewidth"} Our THz LC resonator is introduced in Fig. \[fig:LCPresentation\] (a), along with a sketch of the equivalent circuit. The bottom metallic ground plate is formed by two square capacitor plates of width $W$, connected by a thin stripe. The top metallic part is composed of two rectangular capacitor plates ($W \times W +0.5\mu$m) connected by a bent wire. Two parallel plate capacitors are thus formed at the overlap between the metallic pads, while the bent wire acts as an inductive loop. Note that the top metal plates are 500 nm wider than the bottom one, resulting in a 500 nm wide extension over the outer parts of the capacitor, as shown in Fig. \[fig:LCPresentation\] (c). The capacitors area is shown by the blue squares on the left image of this panel. The small extensions allow engineering the fringing fields between the two capacitor plates. By breaking the symmetry of the in-plane component of the electric field, they allow efficient coupling between the resonator mode and far-field radiation, as determined by previous work on SiN-based resonators. [@mottaghizadeh_nanoscale_2017] The circuit-like mode of the structure which oscillates at the lowest frequency confines the electric field in the capacitive areas while the magnetic field loops around the inductive wire, [@todorov_three-dimensional_2015; @mottaghizadeh_nanoscale_2017] as shown in Fig. \[fig:LCPresentation\] (b), calculated using finite-elements method simulations. They represent the electric (left) and the magnetic (right) energy density in the center $xy$ plane of the resonator. The matter resonance is provided by GaAs/AlGaAs QWs inserted inside the capacitive elements. Note that the $z$-component of the electric field is the only active component for ISB absorption. The resonator mode described above thus naturally fulfills the ISB polarization selection rule, which requires the electric field to be oriented along the growth axis of the QWs. To fabricate the LC resonator, the ground plate is first patterned using e-beam lithography and used as a mask to etch the absorbing region with an inductively-coupled plasma. The structure is then encapsulated in a 3 $\mu$m thick SiN layer, and the surface of the sample is metalized with an optically thick Au layer. The latter serves as a mirror which blocks the transmission into the substrate, such as the reflected beam contains only the spectral features of the metamaterial array. The sample (grown on a GaAs substrate) is then flipped and transferred to a host substrate using an epoxy, and the growth substrate is selectively etched away, revealing the bottom of the patterned absorbing region. The top inductive loop and rectangular capacitor plates are then defined using e-beam lithography. Figure \[fig:LCPresentation\] (c) shows scanning electron microscope images of typical LC resonators. In the following, we keep the lateral size of the capacitor $W=1~\mu$m. The internal perimeter P$_{int}=2L_y+L_x+4\mu$m of the inductive loop is varied from 9 to 14$\mu$m to tune the resonant frequency across the 2-6 THz spectral region, as explained further. A single resonator fits in a square with a diagonal of 4.2 $\mu$m (P$_{int}$=9 $\mu$m) to 7 $\mu$m (P$_{int}$=12 $\mu$m), much smaller than the vacuum wavelength $\lambda_0$=100 $\mu$m. The footprint of a single resonator ranges from $3\times3$ to 5$\times$5 $\mu$m$^2$, e.g. $10^{-3}\lambda_0^2$ to $2.5\times 10^{-3}\lambda_0^2$. {width="\linewidth"} We first probe the optical properties of our system at room temperature, where the ISB absorption can be neglected as the thermal energy is sufficiently high to equally populate the first few energy levels of the QWs, and we can study the electromagnetic modes of the resonator alone. [@todorov_polaritonic_2012] We perform reflectivity experiments using a dry-air purged Fourier Transform Spectrometer (FTIR) (Brucker Vertex 70v) and a globar source. The FTIR is equipped with a custom made reflectivity setup, which allows focusing the globar beam on the sample with the help of a pair of F1 parabolic mirrors, and the reflected beam is measured with a He cooled Ge bolometer (QMC Instruments). Light is linearly polarized and impinging at 45° on the sample, as sketched in Fig. \[fig:Reflectivity\] (f). To ensure good spatial overlap with the globar beam, we fabricate dense arrays composed of $\approx$50 000 resonators separated by 2 $\mu$m from each other, with a total area of 2$\times$2 mm$^2$. All spectra are normalized by the reference from a Au mirror. In Fig. \[fig:Reflectivity\] (a)-(b) we show typical spectra obtained for resonators with P$_{int}=10$ $\mu$m, with light polarized respectively along the line formed by the capacitors ($E_x$), and orthogonally to that line ($E_y$). By comparing the two spectra, we first note the presence of a dip in the reflectivity spectrum for $E_x$-polarized light at 3.5 THz, which is absent in the $E_y$-polarized spectrum. This corresponds to the LC mode of the resonator represented in Fig.\[fig:LCPresentation\]. A broad resonance is observed in both polarizations at 6.5 THz, whose origin will be discussed later. A strong dip is observed just above 8 THz in both polarizations. This corresponds to the lower frequency edge of the GaAs reststrahlen band arising from the optical phonons,[@kittel_introduction_1996] confirming the very strong localization of the electric field in the semi-conductor layer enabled by our device, as the GaAs only constitutes 3.5% of the surface probed by the THz beam. The baseline in all spectra arises from the 3 $\mu$m thick SiN layer, as confirmed by the reflectivity spectra from an area without any patterns (dashed curves in Fig. \[fig:Reflectivity\] (a)-(b)). Indeed the SiN layer has a residual absorption in this spectral region.[@cataldo_infrared_2012] To clarify the origin of the various resonances observed in the experiments we simulated the reflectivity of the structure using a commercial finite element method software (Comsol v.5.3a). Maps of the $E_z$ component of the electric field for the three modes are shown in Figs. \[fig:Reflectivity\] (c)-(d) in two cut planes located at the center of the capacitors. Note that in the right panel of Fig. \[fig:Reflectivity\] (c) and in Fig. \[fig:Reflectivity\] (d) the magnitude of the electric field $E_z$ has been multiplied by two for clarity. The $xy$-plane color maps show the expected electric field distribution for the LC mode, and reveal that the resonance at 6.5 THz actually corresponds to two different modes (P1 and P2) excited with different polarizations ($E_x$ and $E_y$ respectively). Comparing the $xz$-plane maps, we see that only the LC mode efficiently confines the electric field inside the capacitors. P1 and P2 are modes where the electric field lies mainly in the plane of the inductive loop, reminiscent of the modes observed in planar metamaterial resonators.[@zhou_magnetic_2007] Notably, simulations including the top loop alone (and not the ground plate) show the same P1 and P2 modes but not the three-dimensional LC resonance. These two modes couple to the isotropic phonon absorption in the GaAs regions below the Au pads, as they still have a spectral overlap with the GaAs reststrahlen band. A detailed survey of the different modes sustained by the structure is beyond the scope of this paper, and from now on we will restrict the analysis to the LC mode, which is the only one providing electric field almost exclusively localized in the capacitive parts and satisfying the selection rule for the QW ISB absorption. The LC resonance is tuned by changing the internal perimeter P$_{int}$ of the inductive loop, while keeping the capacitance parts fixed. The resonant frequencies as a function of P$_{int}$ are reported in Fig. \[fig:Reflectivity\] (e) (symbols). We compare the experimental results with those of a lumped element model which corresponds to the equivalent circuit sketched in Fig. \[fig:LCPresentation\](a) (see Supplementary Materials for more information). The resonant frequencies of the model, provided by $\omega_{LC} = \sqrt{2/LC}$ are plotted in dashed lines in Fig. \[fig:Reflectivity\] (c), and are in good agreement with those extracted from the measurements. The calculated inductance varies from $L=11$ pH to $L=6$ pH. The calculated capacitance is $C=480$ aF, which compares well with the value for an ideal parallel plate capacitor $C=\varepsilon \varepsilon_0 W^2/d=374$ aF, with $\varepsilon$ and $\varepsilon_0$ the material and vacuum permittivities. The difference between these two values can be explained by the contribution of the fringing fields and of the in-plane parasitic gap capacitance between the two metal pads. This evidences that the fundamental mode of our resonators operates in the near quasi-static limit. To explore the ultra-strong light-matter coupling regime, we have inserted in our resonators an absorbing region consisting of 5 repetitions of 32nm GaAs quantum wells separated by 20nm Al$_{0.15}$Ga$_{0.85}$As barriers, similar to the design described in Ref. . The QWs are modulation-doped by Si $\delta$-doped regions placed 5 nm away from the QW, with a nominal sheet carrier density of $4\times10^{11}$ cm$^{-2}$. In such thin double-metal structures the presence of the two metal-semiconductor interfaces creates a depletion layer on each side of the semiconductor region, bending the conduction band profile and usually depleting 2-3 QWs.[@sze_physics_2006; @todorov_polaritonic_2012] To compensate for this effect, we introduced a doped AlGaAs spacing layer and a doped GaAs external layer (both doped at $2\times10^{18}$ cm$^{-3}$) on each side of the absorbing region, which is of total thickness $d=290$ nm. Additional details on the band structure and sample design are provided in the Supplementary materials. The matter excitation coupled to the LC resonator is an intersubband plasmon of frequency $\tilde{\omega}=\sqrt{\omega_{12}^2+\omega_{P}^2}$, where $\omega_{12}$ is the bare ISB transition frequency and $\omega_P$ is the plasma frequency of the two-dimensional electron gas in the QWs.[@Ando_electronic_1982] The latter is provided by the expression $\omega_P = \sqrt{(n_1-n_2)e^2/(\varepsilon \varepsilon_0m^{\ast}L_\mathrm{QW,eff})}$, where $e$ is the electron charge, $m^{\ast}$ is the electron effective mass in the QW, $n_1$ (resp. $n_2$) is the surface electron density in the first (second) subband, and $L_\mathrm{QW,eff}$ is an effective length of the quantum well as defined in Ref. . The quantity $L_\mathrm{QW,eff}$ can be seen as an effective thickness of the quantum confined electron plasma and depends on the wave functions of the first and second subbands (see Ref. 41). We find $L_\mathrm{QW,eff}\approx25$ nm, smaller than the physical thickness of the QW (32 nm). For the following discussion, it is important to note that $\omega_P$ and hence $\tilde{\omega}$ depend on the charge density in a single quantum well only. The characteristic equation of the coupled intersubband plasmon-resonator system is written in the general case: $$\left( \omega^2-\tilde{\omega}^{2} \right) \left( \omega^2-\omega_c^2 \right) = \Psi^2 f_{w} f_{12} \omega_P^2 \omega_c^2 \label{eq:dispersion}$$ where $\omega_c$ is the resonator frequency, $f_{12}$ is the $1\rightarrow2$ transition oscillator strength ($f_{12}=0.96$ for an infinite QW), $f_{w}=N_{QW}L_\mathrm{QW,eff}/d$ describes the filling factor of the QWs inside the absorbing region with $N_{QW}$ charged quantum wells, and $\Psi^2$ describes the optical overlap of the resonator mode with the absorbing region, which is related to the effective mode volume $V_{\mathrm{eff}}$ as explained further. Note that the coefficient $f_w$ quantifies the *filling factor* of the QWs inside the absorbing region that takes into account the fact that the active dipoles are not homogeneously distributed in the semiconductor, but are only localized inside the QWs. Instead, the dimensionless overlap factor $\Psi^2$ represents the fraction of electromagnetic energy coupled into the $z$-component of the electric field and spatially overlapping with the semiconductor layers inside the capacitor. It is defined by:[@zanotto_analysis_2012] $$\Psi^2 = \frac{\int_{AR}\frac{\varepsilon\varepsilon_0}{2}|E_z|^2}{U_e} \label{eq:Psi}$$ where the energy stored in the vertical component of the electric field $E_z$ (the sole component of the field to couple to ISB transitions) is integrated over the absorbing region (AR) volume $V_{AR}=2W^2d$ and normalized by the total electric energy $U_e$ of the mode. The effective mode volume is then determined by the relation $V_{\mathrm{eff}}=V_{AR}/\Psi^2$. Since $\Psi^2 \leq 1$ this definition accounts for field leakage outside the capacitive parts of the resonator. Eq. \[eq:dispersion\] provides the upper (UP) and lower (LP) polariton frequencies as a function of $\omega_c$. The minimum splitting between UP and LP is exactly the vacuum Rabi splitting $2\Omega_R$: $$2\Omega_R = \sqrt{\Psi^2 f_w f_{12}} \omega_P = \sqrt{\frac{\Psi^2 f_{12} e^2 N_{QW}(n_1-n_2)}{\varepsilon \varepsilon_0m^{\ast}d}} \label{eq:Rabi}$$ The Rabi splitting can also be expressed as $2\Omega_R = \sqrt{f_{12} e^2 /\varepsilon \varepsilon_0m^{\ast}} \times \sqrt{N_{QW}(N_1-N_2)/V_{\mathrm{eff}}}$, where $N_{1,2}$ is the total number of electrons populating the subband $1,2$. The Rabi splitting that is experimentally determined from spectroscopic studies, as described further, is thus directly linked to the effective mode volume $V_{\mathrm{eff}}$ which is an important quantity in nano-photonic systems.[@liu_fundamental_2016] However, the analysis of our experimental data is easier to perform with the help of Eq. . Ultimately, two parameters govern the maximum coupling strength: the overlap factor $\Psi^2$ and the number of available dipoles in the microcavity volume $N_{QW}(n_1-n_2)$. The strength of the coupling can thus be controlled by tuning the population difference through the change of the temperature of the sample. In order to asses the parameters $N_{QW}$ and $\Psi^2$ in our LC resonators, we compare the spectroscopic features of the ultra-strong coupling regime in the LC resonators with square patch double-metal microcavities. Such double-metal cavities sustain a resonance at $\lambda=2n_{eff}s$, where $s$ is the size of the patch, and $n_{eff}=3.9$ the effective index of the confined mode.[@todorov_optical_2010] They have been shown to reach the ultra-strong coupling regime with similar absorbing regions with $\Psi^2 \approx 1$, however with a much larger effective mode volume. [@todorov_ultrastrong_2010; @todorov_polaritonic_2012] They will serve as a reference to our current LC samples.[@feuillet-palma_extremely_2012] {width="\linewidth"} A comparison between a LC resonator and a patch cavity with the same absorbing region is shown in Fig. \[fig:TSweep\]. Optical microscope images of the two types of resonators, which have identical resonant frequencies (3.5 THz) are shown in Fig. \[fig:TSweep\] (a) (to scale). While the size of the patch cavity is set by the diffraction limit, clearly the lateral confinement of the electric field in the capacitors of the LC resonator is well below that limit. For spectroscopic studies at cryogenic temperatures, the samples are mounted on the cold finger of a liquid helium continuous flow cryostat, and probed in a reflectivity experiment. We report in Fig. \[fig:TSweep\] (b) (resp. (c)) the reflectivity of the patch cavities array (resp. LC resonator) tuned near the intersubband plasmon resonance as a function of temperature, ranging from 300 K to 7 K. In the case of the LC resonator, the baseline induced by the SiN has been removed for clarity. In both cases the room temperature spectra show a single resonance around 3.5 THz as the population difference between the first two subbands is zero ($n_1 \approx n_2$), owing to the thermal electron distribution. The matter excitation thus vanishes, and one only sees the LC and the patch cavity resonances. This single resonance splits when decreasing the temperature as the population difference $n_1-n_2$ increases, and the maximum separation is obtained at low temperature (7K), reaching 0.92 THz for the patch cavities and 0.75 THz for the LC resonators. {width="\linewidth"} Tuning the LC and patch resonant frequencies allows us to more precisely map the dispersion relation given in equation . In Fig. \[fig:USCLCs\] the position of the UP and LP modes at T=7 K are plotted as a function of the cavity frequency, obtained by modifying either the size $s$ of the patch cavities, or the inductance of the LC resonators through the internal perimeter P$_{int}$. We observe a clear anticrossing and the opening of a polaritonic gap between the two polariton branches. By fitting the data with equation , we can extract the Rabi splittings $2\Omega_{R-LC}=0.72$ THz and $2\Omega_{R-patchs}=0.85$ THz, representing respectively 21% and 24% of the intersubband plasmon frequency $\tilde{\omega}$ determined from the fit of the dispersion relation. The slight reduction of the Rabi splitting for LC resonators can be due to both lateral depletion of the QW, as well as to the overlap factor $\Psi^2$ that is less than unity. We now want to determine separately the total charge and the overlap factor in our systems. We start by deducing the total charge left in the 5 QW patch cavity sample, by comparing it to a reference sample having exactly the same QWs, but repeated 25 times. The 25 QWs sample is also processed into patch cavities (Refs. ), and we can safely assume that $\Psi_{ref}^2=\Psi^2=1$. The data for that sample are provided in the Supplementary Materials, along with a detailed derivation of the method to extract the electronic population. We deduce that 21 QWs are populated, with an equivalent doping of $1.37\times10^{11}$cm$^{-2}$. Using equation , (see also Ref. ) a proportionality rule yields the total surface charge density at low temperature: $$\left(\frac{\Omega_R}{\Omega_{R-ref}}\right)^2 = \frac{\Psi^2 f_{12}N_{QW} n_1}{\Psi_{ref}^2 f_{12-ref}N_{QW-ref} n_{1-ref}} \frac{d_{ref}}{d} \label{eq:OmegaR_ratio}$$ In that case $\Psi^2=\Psi_{ref}^2=1$, and we find an equivalent total charge density of $N_{QW} n_1=2 n_1$, meaning that only $2/5$ of the total charge is left is the QWs, confirming the importance of the depletion effects at the metal-semiconductor interfaces. Furthermore, for both samples we observe the same matter excitation frequency $\tilde{\omega}$, which means the surface charge density per quantum well is the same. We therefore conclude that we have $N_{QW}=$2 charged quantum wells in the 5QW absorbing region. A more detailed analysis is given in the Supplementary Information. Having determined the total charge in the case of the patch cavities, we use the same proportionality rule to compare the 5QW patch and LC resonators. We first assume that we have the same total charge in both samples. Then, according to equation we can derive the geometric overlap factor $\Psi^2$. Comparing the data from Fig. \[fig:USCLCs\] we derive a confinement factor $\Psi^2=0.7$. A more careful comparison of Fig. \[fig:USCLCs\] (a) and (b) shows that the fits of the dispersion relations yields a plasmon frequency $\tilde{\omega}$ at a slightly lower frequency for the LC resonators ($\tilde{\omega}=$3.55 THz) than the one of the patch cavities ($\tilde{\omega}=$3.65 THz). Since the plasmon frequency is related to the plasma frequency through the formula $\tilde{\omega}=\sqrt{\omega_{12}^2+\omega_{P}^2}$, we conclude that the red shift is due to a lower plasma frequency of the LC as compared to the patch cavity. This is due to a further lateral depletion of the QWs originating from surface traps at the etched surface of the absorbing region, as already reported in the case of etched pillar or nanowire structures. [@dietze_ultrastrong_2013; @amanti_electrically_2013; @lahnemann_near-infrared_2017] From our data, we infer a plasma frequency for the LC resonators that is 7.6% lower than the one for the patch cavities. Correcting for this effect yields an overlap factor $\Psi^2=0.79$. This value is slightly larger than the one predicted in numerical simulations, $\Psi_{LC}^2=0.64$. The effective mode volume of the LC resonators is thus determined to be $V_{\mathrm{eff}}=1.2\times10^{-6}\lambda_0^3$, almost two orders of magnitude smaller than the patch cavities. The low value of the effective mode volume is a striking feature of our resonator, since achieving an effective volume very close to the physical volume of the semiconductor absorbing region ($9.2\times10^{-7}\lambda_0^3$) represents a critical trade-off in double-metal geometries, owing to the leakage of the electric energy in fringing fields. Square or wire patch cavities indeed lead to $\Psi^2$ factors close to unity at the expense of a large mode volume, while other systems report very small mode volume, sacrificing the overlap factor down to a few percents. [@malerba_towards_2016; @paulillo_room_2016] We can also estimate the intersubband plasmon frequency $\tilde{\omega}$ in the case where one or three QWs would be populated. The detail of the calculation is given in Supplementary Information, and the results are shown in stars in Fig. \[fig:USCLCs\] (b). We can see that the intersubband plasmon frequencies obtained in the two cases strongly differ from our measurement, confirming the analysis presented above. From the knowledge of the equivalent surface charge density we can deduce the total number of electrons participating in the coupling with the cavity and resonator modes. The calculation yields $N_{e-patchs} = 3.3 \times 10^5 $ e$^{-}$/patch and $N_{e-LC} = 2.4 \times 10^3 $ e$^{-}$/capacitor. Our newly developed LC resonators thus allow us to greatly decrease the number of electrons involved in the coupling while maintaining a large vacuum Rabi splitting, making a step towards the few electrons regime beyond previous results on double-metal cavities. [@feuillet-palma_extremely_2012; @paulillo_room_2016; @malerba_towards_2016] {width="\linewidth"} In order to improve further the strength of the light-matter coupling, we designed a second sample with an identical absorbing region, but with an increased thickness of the cavity (320 nm) and and increased doping density in the outermost GaAs layers ($6\times10^{18}$ cm$^{-3}$). After processing into patch cavities and LC metamaterials, we perform the same experiments as described above (see Supplementary materials for details). Our results show that in the present sample, all 5 QWs are populated. This means that thanks to the increased charge density in the interface layers, the populated QWs participating in the optical absorption occupy a larger fraction of the total cavity volume, increasing the filling factor $f_w$. However, we deduce a lower value of the optical overlap factor, $\Psi^2$=0.56, meaning a slightly higher effective volume than that of the previous LC sample. This could be explained by a partial screening of the penetration of the electric field inside the capacitors, owing to the higher number of free carriers in the contact layers. Nevertheless, as shown in Fig. \[fig:USCDop\], the Rabi splitting $2\Omega_R$ is increased to 0.89 THz in the new LC sample, reaching values of 0.27 of the ISB plasmon frequency. The total electron number is $N_e=6\times10^3$e$^-$/capacitor. Remarkably, this value is one of the lowest achieved so far using ISB transitions coupled to metamaterials, while retaining a large $\Omega_R/\omega_c$ ratio. In conclusion, we demonstrate a deeply subwavelength confinement of electromagnetic energy with an effective mode volume $V_{\mathrm{eff}}=10^{-6}\lambda_0^3$ and a reduction of the number of interacting electrons, down to a few thousands while keeping $2\Omega_R/\tilde{\omega}=0.27$. The electron number is an order of magnitude larger than that recently reported in systems exploring the coupling between Landau level transitions and planar metamaterials.[@keller_few-electron_2017] However, our structures will allow further reduction of the number of electrons while maintaining a large ratio $\Omega_R/\tilde{\omega}$. This can be achieved by reducing the size of the capacitive elements to a few hundreds of nanometers in our LC resonators. For instance, by maintaining a surface equivalent doping of $\approx2\times10^{11}$cm$^{-2}$ and working with a single quantum well, we could achieve ultra-strong light-matter coupling with only 10 electrons in 100 nm wide capacitors. Devices with such small capacitive elements have already been fabricated in SiN-based LC resonators.[@mottaghizadeh_nanoscale_2017] An important asset of our device architectures is that they naturally provide the possibility to implement DC current input/output from the semiconductor region. Indeed, the top and bottom metallic parts are direct current uncoupled, and contacts can be taken directly on the inductive parts without hindering the sub-wavelength confinement in the capacitors. Such device architectures will allow us to study new effects related to the ultra-strong coupling, beyond the spectroscopic studies performed so far,[@keller_few-electron_2017] such as polariton-assisted fermionic transport.[@feist_extraordinary_2015; @orgiu_conductivity_2015; @bruhat_cavity_2016; @stockklauser_strong_2017; @paravicini-bagliani_magneto-transport_2018; @hagenmuller_intrinsic_2018] The reduced number of electrons leads to a reduced number of dark states, optimizing the coupling between the collective electronic excitation and the electronic current.[@de_liberato_quantum_2009] Furthermore, in the weak coupling regime, these device architectures can be very beneficial for ultra-low dark current ISB detectors.[@palaferri_ultra-subwavelength_2016] The authors acknowledge the help of the technical staff from the cleanroom facility of Université Paris Diderot. We thank Li Chen for the wafer bonding of the samples used to fabricate patch antennas. This work was supported by the French National Research Agency under the contract ANR-16-CE24-0020. We acknowledge support from the EPSRC (UK) programme grant HyperTerahertz EP/P021859/1. EHL acknowledges support from the Royal Society and the Wolfson Foundation.

ArXiv

--- abstract: | The paper deals with an integrodifferential operator which models numerous phenomena in superconductivity, in biology and in viscoelasticity. Initial-boundary value problems with Neumann, Dirichlet and mixed boundary conditions are analyzed. An asymptotic analysis is achieved proving that for large t, the influences of the initial data vanish, while the effects of boundary disturbances are everywhere bounded. :[Initial- boundary problems for higher order parabolic equations; Laplace transform; Superconductivity;FitzHugh Nagumo model.]{} **Mathematics Subject Classification (2000)**44A10,35K57,35A08,35K35 author: - 'Monica De Angelis[^1]' title: Asymptotic estimates related to an integro differential equation --- Introduction ============ If $ u=u(x,t), $ let us consider the following integrodifferential equation $$\label{11} \mathcal L u \equiv \,\, u_t - \varepsilon u_{xx} + au +b \int^t_0 e^{- \beta (t-\tau)}\, u(x,\tau) \, d\tau \,=\, F(x,t,u) \,$$ where $ \varepsilon, a, b, \beta $ are positive constants, $ x $ denotes the direction of propagation and $ t $ is the time. According to the meaning of $ F(x,t,u)$, equation (\[11\]) describes the evolution of several linear or non linear physical models. For instance, when $ F=f(x,t), $ (\[11\]) is related to the following linear phenomena: - motions of viscoelastic fluids or solids [@bcf; @dr1; @dmr; @mor]; - heat conduction at low temperature [@mps; @fr; @bs], - sound propagation in viscous gases [@l]. When $ F=F(x,t,u), $ some non linear phenomena involve equation (\[11\]) both in superconductivity and biology. $ \bullet $ [*Superconductivity*]{} – Let $\, u\, $ be the difference between the wave functions phases of two superconductors in a Josephson junction. The equation describing tunnel effects is the following one: $$\label{122} \varepsilon u_{xxt}\, - \, u_{tt} \, +\, u_{xx}-\, \alpha u_t = \, \, \sin u \ - \gamma$$ where constant $\, \gamma \, $ is a forcing term proportional to a bias current, while the $ \varepsilon -term$ and the $\, \alpha -term $ account for the dissipative normal electron current flow, respectively along and across the junction [@bp; @df13]. Equation (\[122\]) can be obtained by (\[11\]) as soon as one assumes $$\label{133} a \,=\, \alpha \, - \dfrac{1}{\varepsilon} \, \,\quad\,\, b = \, - \, \dfrac{a}{\varepsilon} \,\,\quad \displaystyle \, \beta \,= \dfrac{1}{\varepsilon}\,\,$$ and $ F $ is such that $$\label{144} F(x,t,u)\,=\, -\, \int _0^t \, e^{\,-\,\frac{1}{\varepsilon}\,(t-\tau\,)}\,\,[\, \, sen \, u (x, \tau)\,- \gamma\, \,]\, \, d\tau.$$ Besides, when the case of an exponentially shaped Josephson junction (ESJJ) is considered, the evolution of the phase inside this junction is described by the third order equation: $$\label{12} (\partial_{xx} \, - \,\lambda\, \partial _x\,)\,\,(\varepsilon u _{t}+ u) - \partial_t(u_{t}+\alpha\,u)\, = \, \, \sin u \ - \gamma$$ where $ \lambda $ is a positive constant generally less than one and the terms $ \lambda u_{xt} $ and $ \lambda u_{x} $ represent the current due to the tapering junction. In particular $ \lambda u_{x} $ corresponds to a geometrical force driving the fluxons from the wide edge to the narrow edge. [@df13; @df213; @bcs] An (ESJJ) provides several advantages with respect to a rectangular junction ([@mda13] and reference therein). For instance, in [@bcs] it has been proved that it is possible to obtain a voltage which is not chaotic anymore, but rather periodic excluding, in this way, some among the possible causes of large spectral width. It is also proved that the problem of trapped flux can be avoided. Numerous applications and devices involve Josephson junctions, for example SQUIDs which are very versatile and can be used in a lot of fields. (see f.i.[@cd] and references therein). Moreover, if $\,\,u= e^{\lambda\,x/2 \,}\,\,\overline{u},\,\, $ (\[12\]) turns into an equation like (\[122\]) and hence into (\[11\]). $ \bullet $ [*Biology*]{} – Let us consider the FitzHugh-Nagumo system (FHN) which models the propagation of nerve impulses. [@m1]: $$\label{15} \left \{ \begin{array}{lll} \displaystyle{\frac{\partial \,u }{\partial \,t }} =\, \varepsilon \,\frac{\partial^2 \,u }{\partial \,x^2 } \,-\, v\,\, + f(u ) \, \\ \\ \displaystyle{\frac{\partial \,v }{\partial \,t } }\, = \, b\, u\, - \beta\, v\,. \\ \end{array} \right.$$ Here, $\, u\,( \,x,t\,)$ models the transmembrane voltage of a nerve axon at a distance x and time t, while $\, v\,(\,x,t\,)$ is an auxiliary variable acting as a recovery variable. Besides, the function $\,f (u)\,$ has the qualitative form of a cubic polynomial $$\label{16} f(u)\, =\,-\,a\,u\, +\,\varphi(u) \quad with \quad \varphi \,=\, u^2\, (\,a+1\,-u\,),$$ while $ \varepsilon,\, b,\, \beta\, $ are non negative and the parameter $ a, $ representing the threshold constant, is generally $ \,0<a<1.$ (see f.i. [@rda08] and references therein) Denoting by $\,v_0 \,$ the initial value of v, system (\[15\]) (\[16\]) can be given the form of the integrodifferential equation (\[11\]) as soon as one puts: $$\label{51} F(x,t,u)\, =\,\varphi (u) \, -\, v_0(x) \, e^{\,-\,\beta\,t\,}.$$ In this paper, initial value problems with Neumann, Dirichlet and mixed boundary conditions for (\[11\]) are considered. By means of properties of the fundamental solution $ K_0(x,t,)$ of the operator $ {\cal L} ,$ appropriate estimates are obtained. The function $ K_0(x,t)$ has already been determined and analyzed in [@dr8] and an analysis related to a Neumann boundary problem has been conducted in [@dr13]. Aim of this paper is an asymptotic analysis for the initial boundary value problem both with Dirichlet conditions and with mixed conditions. These cases involve x-derivative of theta functions $ \theta(x,t) $ and $ \theta^*(x,t) $ which are determined in sec (\[sec3\]). So, effects of boundary perturbations can be evaluated by means of a well known theorem on asymptotic behavior of convolutions. As an example, according to the equivalence between operator $ \mathcal{L} $ and the FHN system, an estimate of the solution related to the reaction-diffusion system (\[15\]) is obtained proving that, for large $ t, $ effects determined by boundary disturbance are bounded. Some models of superconductivity and biology ============================================ Let $\, T\, $ be an arbitrary positive constant and $$\, \Omega_T \, \equiv \{\,(x,t) : \, 0\,\leq \,x \,\leq L \,\,; \ 0 < t \leq T. \,$$ \(I) A first example is related to [*Neumann*]{} boundary conditions (NBC) $$\label{21} \left \{ \begin{array}{lll} { \cal L}\,u\, \,=\, F(x,t,u) \, & (x,t) \in \Omega_T \, \\ \\ \,u (x,0)\, = u_0(x)\, \,\,\, & x\, \in [0,L], \\ \\ \, u_x(0,t)\,=\,\psi_1(t) \qquad u_x(L,t)\,=\,\psi_2(t) & 0<t\leq T. \end{array} \right.$$ In superconductivity, this problem occurs when the magnetic field, proportional to the phase gradient, is assigned. [@j; @ddf]. In mathematical biology, it can refer to a two-species reaction diffusion system subjected to flux boundary conditions [@m1]. The same conditions are present in case of pacemakers [@ks] and are applied also to study distributed (FHN) systems [@ns] or to solve FHN systems by means of numerical calculations [@d]. \(II) Another example concerns [*Dirichlet*]{} boundary conditions (DBC) $$\label{22} \left \{ \begin{array}{lll} { \cal L}\,u\, \,=\, F(x,t,u) \, & (x,t) \in \Omega_T \, \\ \\ \,u (x,0)\, = u_0(x)\, \,\,\, & x\, \in [0,L], \\\\ \,u(0,t)\,=\,g_1(t) \qquad u(L,t)\,=\,g_2(t) & 0<t\leq T. \end{array} \right.$$ In superconductivity, $(\ref{22})_3$ refer to the phase boundary specifications[@df213; @df13; @mda13]. In excitable systems these conditions occur when the behavior of a single dendrite has to be determined and the voltage level is fixed[@ks] or when the pulse propagation in a continuum of heart cells is studied [@a; @ks]. Besides, the Dirichlet problem is also considered to determine universal attractors both for Hodgkin-Huxley equations and for FHN systems,[@m] and for stability analysis and asymptotic behavior of reaction-diffusion systems solutions, [@dm13; @t; @ccd; @ra; @mda14], or in hyperbolic diffusion [@gs]. \(III) At last, [*mixed*]{} boundary conditions (MBC) as $$\label{23} \left \{ \begin{array}{lll} { \cal L}\,u\, \,=\, F(x,t,u) \, & (x,t) \in \Omega_T \, \\ \\ \,u (x,0)\, = u_0(x)\, \,\,\, & x\, \in [0,L], \\ \\ \,u(0,t)\,=\,h_1(t) \qquad u_x(L,t)\,=\,h_2(t) & 0<t\leq T, \end{array} \right.$$ occur in many physical examples both in superconductivity (see,f.i.[@bv] and references therein) and in biology, as shown in [@m1; @ks]. In particular, in [@rs], mixed boundary conditions are considered in order to give qualitative information concerning both the threshold problem and the asymptotic behavior of large solutions for the FHN system. When $ F\,= f(x,t) $ is a linear function, problems (\[21\])-(\[23\]) can be solved by Laplace transformation with respect to $ t. $ Let $ z(x,t) $ be an arbitrary function admitting Laplace transform $ \hat z(x,s) $ $$\label{24} \hat z (x,s) \, = \int_ 0^\infty \, e^{-st} \, z(x,t) \,dt \,\,= \mathcal{L}_t\,z$$ Referring to the parameters $ a, \,\beta,\, b, \,\,\varepsilon\,\, $ of the operator $ \mathcal{L}, $ if $$\label{25} \sigma^2 \ \,=\, s\, +\, a \, + \, \frac{b}{s+\beta},\,\quad \,\,\displaystyle\,\tilde{\sigma}^2\,=\, \sigma^2/{\varepsilon,}\,\,$$ we denote by $ \theta(x,s) $ and $ \theta ^* (x,s) $ the following Laplace transforms: $$\, \label{26} \displaystyle \hat \theta \,(\,y,\tilde\sigma)\,= \,\dfrac{\cosh\,[\, \tilde\sigma \,\,(L-y)\,]}{\,2\, \, {\varepsilon} \,\,\tilde \sigma\,\,\, \sinh\, (\,\tilde \sigma \,L\,)}\,\,=$$ $$=\, \frac{1}{2 \,\, \sqrt\varepsilon \,\,\,\sigma } \, \biggl\{\, e^{- \frac{y}{\sqrt \varepsilon} \,\,\sigma}+\, \sum_{n=1}^\infty \,\, \biggl[ \,e^{- \frac{2nL+y}{\sqrt \varepsilon} \,\,\sigma} \, +\, e^{- \frac{2nL-y}{\sqrt \varepsilon} \,\,\sigma}\, \biggr] \, \biggr\},$$ $$\, \label{27} \displaystyle \hat \theta^* \,(\,y,\tilde\sigma)\,= \,\dfrac{\sinh\,[\, \tilde\sigma \,\,(L-y)\,]}{\,2\, \, {\varepsilon} \,\,\tilde \sigma\,\,\, \cosh\, (\,\tilde \sigma \,L\,)}\,\,=$$ $$=\, \frac{1}{2 \sqrt\varepsilon \sigma } \, \biggl\{ e^{- \frac{y}{\sqrt \varepsilon}\sigma}+\,\, 2\,\,\sum_{n=1}^\infty \,\, \biggl( \,e^{- \frac{4nL+y}{\sqrt \varepsilon} \,\,\sigma} \, +\, e^{- \frac{4nL-y}{\sqrt \varepsilon} \,\,\sigma}\, \biggr) \, - \sum_{n=1}^\infty \,\, \biggl( \,e^{- \frac{2nL+y}{\sqrt \varepsilon} \,\,\sigma} \, +\, e^{- \frac{2nL-y}{\sqrt \varepsilon} \,\,\sigma}\, \biggr) \biggr\}.$$ Then, the Laplace transform solutions of the linear problems (\[21\])-(\[23\]) can be obtained by means of standard techniques and it results: $ \bullet $ Formal solution for initial boundary problem with (NBC) $$\label{28} \begin{split} \hat u (x,s) = &\,\int _0^L \, [\,\hat \theta\,(\,|x-\xi|, \,s\,)\,+\,\,\,\hat \theta\,(\,|x+\xi|,\, s\,)\,] \, \,[\,u_0(\,\xi\,) \,+\,\hat f(\,\xi,s)\,]\,d\xi\, \\ \,& -\,\,\ 2 \,\,\varepsilon \, \,\hat \psi_1 \,(s) \,\, \hat \theta (x,s)\,+ \, 2 \,\, \varepsilon \,\, \hat \psi_2 \, (s)\,\,\hat \theta \,(x-L,s\,).\, \,\, \end{split}$$ $ \bullet $ Formal solution for (DBC) $$\label{29} \begin{split} \hat u (x,s) = &\,\int _0^L \, [\,\hat \theta\,(\,|x-\xi|, \,s\,)\,-\,\,\,\hat \theta\,(\,x+\xi,\, s\,)\,] \, \,[\,u_0(\,\xi\,) \,+\,\hat f(\,\xi,s)\,]\,d\xi\, - \\ \, & -\,\,\ 2 \,\,\varepsilon \, \,\hat g_1 \,(s) \,\, \hat\theta_x (x,s)\,+ \, 2 \,\, \varepsilon \,\, \hat g_2 \, (s)\,\,\,\hat \theta_x \,(x-L,s\,).\, \,\, \end{split}$$ $ \bullet $ Formal solution for (MBC) $$\label{210} \begin{split} \hat u (x,s) = &\,\int _0^L \, [\,\hat \theta^*\,(\,x+\xi, \,s\,)\,-\,\,\,\hat \theta^*\,(\,|x-\xi|,\, s\,)\,] \, \,[\,u_0(\,\xi\,) \,+\,\hat f(\,\xi,s)\,]\,d\xi\, + \\ \,& \,-\,\ 2 \,\,\varepsilon \, \,\hat h_1 \,(s) \,\, \,\,\hat \theta^*_x (x,s)\,+ \, 2 \,\,\varepsilon \hat h _2 \, (s)\,\,\hat \theta^* \,(L-x,s\,) . \,\, \end{split}$$ $K_0 (x,t)$ and $ \theta(x,t) $ properties {#sec3} =========================================== The Neumann boundary value problem has already been solved in [@dr13]. Let us consider now cases (II) and (III). Let $ K_0(x,t) $ be the fundamental solution of the linear operator $\, \cal L \,$ defined in (\[11\]). It has already been determined in [@dr8] and one has: $$\label{31} K_0(r,t)= \frac{1}{2 \sqrt{\pi \varepsilon } }\biggl[ \frac{ e^{- \frac{r^2 }{4 t}-a\,t}}{\sqrt t}-\,\sqrt b \int^t_0 \frac{e^{- \frac{r^2}{4 y}\,- ay}}{\sqrt{t-y}} \, e^{-\beta (\, t \,-\,y\,)} J_1 (2 \sqrt{\,by\,(t-y)\,})\,\,dy \biggr]$$ where $\, r \, = |x| \, / \sqrt \varepsilon \, \, $ and $ J_n (z) \,$ is the Bessel function of first kind. Function $ K_0 \, $ has the same basic properties of the fundamental solution of the heat equation, and in the half-plane $ \Re e \,s > \,max(\,-\,a ,\,-\beta\,)\,$ it results: $$\label{32} \,{\cal L }_t\,\,K_0\,\equiv \,\,\int_ 0^\infty e^{-st} \,\, K_0\,(r,t) \,\,dt \,\,= \, \frac{e^{- \,r\,\sigma}}{2 \, \sqrt\varepsilon \,\sigma \, }$$ where $ \sigma $ is defined in $(\ref{25})_1$. Among other properties, in [@dr8] the following estimates have been proved: $$\label{33} \int_\Re|\,K_0(x-\xi,t)|d\xi\leq e^{\,-\,at} +\, \sqrt b\, \pi t e^{\,-\,\omega \, t } \quad \int_0^t\,d \tau\, \int_\Re |K_0(x-\xi,t)| \, \,d\xi \leq \, \beta_0$$ $$\label{34} |K_0| \, \leq \, \frac{e^{- \frac{r^2}{4 t}\,}}{2\,\sqrt{\pi \varepsilon t}} \,\, [ \, e^{\,-\,at}\, +\, b t \,E(t)\, ]$$ where constants $ \omega, \,\,\beta_0\,\, \mbox {and}\,\, E(t) $ are given by: $$\label{35} \omega = min(a,\beta),\,\,\qquad \beta _0 =\,\, \frac{1}{a}\, +\, \pi \sqrt b \, \, \displaystyle {\frac{a+\beta}{2(a\beta)^{3/2}}},$$ $$E(t) \,=\, \frac{e^{\,-\,\beta t}\,-\,e^{\,-\,at}}{a\,-\,\beta}\,\,>0.\,$$ Moreover, denoting by $$\label {36} K_i( r,t) \, = \,\,\int^t_0 \,\,e^{-\,\beta \,(\,t-\tau)\,}\,K_{i-1}\,(x,\tau\,) \, d \tau\,\, \qquad ( i=1,2)$$ kernels $ K_1(x,t) $ and $ K_2(x,t) $ have the same properties of $ K_0(x,t). $ Hence, the following theorem holds [@dr8]: For all the positive constants $ a, \,b,\,\, \varepsilon,\, \beta $ it results: $$\label{37} \int_\Re |K_1| \, \ d\xi \leq \, \,E(t);\, \qquad \int _0^t\\d\tau \,\int_\Re |K_1| \, \ d\xi \leq \, \beta_1\,$$ $$\label{38} \int_\Re \left|K_2 (x-\xi,t)\right| \, d\xi \, \leq \, t\, E(t)$$ where $ \beta_1\,=\, ({a\,\beta})^{\,-1}.\,$ So that, in order to obtain inverse formulae of (\[29\]) and (\[210\]), let us apply (\[32\]) to (\[26\])(\[27\]). Then, one deduces the following functions which are similar to [*theta functions*]{}: $$\label{310} \begin{split} \theta (x,t) \,=\,& K_0(x,t) \ +\, \sum_{n=1}^\infty \,\, \ [\, K_0(x \,+2nL,\,t) \, + \, K_0 ( x-2nL, \,t)\,] \, \\&= \sum_{n=-\infty }^\infty \,\, \ K_0(x \,+2nL,\,t). \end{split}$$ $$\label{3100} \begin{split} \theta^* (x,t) \,= \,2\,\sum_{n=-\infty }^\infty \, K_0(x \,+4nL,\,t)\, - \,\sum_{n=-\infty }^\infty \,\, K_0(x \,+2nL,\,t). \end{split}$$ Some of the properties of function $ \theta(x,t) $ have already been evaluated in [@dr13]. Precisely, denoting by $C= 2 \varepsilon \,\,\pi^2 / (\, 6\,e L ^2\,) $ and letting $$\label{311} C_0\, = \, \frac{1}{ 2 \sqrt{\varepsilon \, \omega }}\,+\,\frac{ b\,\,\omega^{-\, 3/2}}{4 \sqrt{ \varepsilon } \,\,|a-\beta|} \,\biggl[\,1\,\, +\, \dfrac{C}{b}\,|a-\beta|\,+ \frac{3 \,C}{2\, \omega }\,\,\biggr],$$ the $\theta (x,t)$ function, defined in $( \ref{310}), $ satisfies the following inequalities: $$\label{312} \int_0^L |\theta (|x-\xi|,\,t)|\ \, d\xi \leq \, ( 1\, +\, \sqrt b \,\pi \,t \, ) \,\,e^{- \omega \, t\,}$$ $$\label{313} \int_0^t\,d \tau\, \int_0^L |\theta (|x-\xi|,\,t)|\ \, d\xi \leq \, \beta_0;\,\quad \quad \int_0^ \infty \, | \theta ( x,\tau )| \,\, d \tau \,\, \leq \,\,C_0,$$ and, it results: $$\label{314} \lim _{t \to \infty} \theta ( x, t ) \,\, = \,\,0;\qquad\lim _{t \to \infty} \int _0^t \theta ( x, \tau ) \,\, d\tau \,\,= \frac{1}{2 \, \varepsilon \,\,\sigma _0\, }\,\,\ \frac{\cosh \sigma_0 \,\,(L-x)}{\sinh\,( \sigma_0 \, L). }$$ where $ \sigma_0 = \sqrt{\biggl(\,a\,\,+ \dfrac{b}{\beta}\biggr)\dfrac{1}{\varepsilon}}.$ Furthermore, as for $ \frac{\partial \theta}{\partial x } $, from (\[31\]), it is well-rendered that the x derivative of the integral term vanishes for $ x\,\rightarrow 0 \,$, while the first term represents the derivative with respect to $ x $ of the fundamental solution related to the heat equation. So, by means of classic theorems (see,f.i. [@c] p. 60), conditions $(\ref{22})_3$ are surely satisfied. Moreover one has: $$\label{317} \begin{split} \lim _{t \to \infty} \int _0^t \theta_x ( x, \tau ) \,\, d\tau \,\,= \frac{1}{2 \, \varepsilon \, }\,\,\ \frac{\sinh \sigma_0 \,\,(x-L)}{\sinh\,( \sigma_0 \, L) }\\ \\ \lim _{t \to \infty} \int _0^t \theta^*_x ( x, \tau ) \,\, d\tau \,\,= \,-\,\frac{1}{2 \, \varepsilon \, }\,\,\ \frac{\cosh \sigma_0 \,\,(L-x)}{\cosh\,( \sigma_0 \, L) }\end{split}$$ Asymptotic behaviour {#sec4} ==================== When the source term $ F =f(x,t) $ is a prefixed function depending only on $ x $ and $ t $, then, initial boundary value problems (\[22\]) (\[23\]) are linear and can be solved explicitly. Moreover, when $\, F\, =\, F(x,t,u) \,$ depends also on the unknown function $ u(x,t), $ then these problems admit integral differential formulations and one has: $ \bullet $ Integro differential equation for problem (\[22\]) (DBC): $$\label{41} \begin{split} u(\, x,\,t\,)\, = \,\,\int^L_0 \, [\theta \,(|x-\xi|,\, t)\,- \theta (x+\xi,\,t)\,]\, \,u_0(\xi)\,\, d\xi \,\,- \,\\ \\ \,2 \, \varepsilon \,\int^t_0 \theta_x\, (x,\, t-\tau) \,\,\, g_1 (\tau )\,\,d\tau\,+\, 2\,\, \varepsilon \int^t_0 \theta_x\, (x-L,\, t-\tau) \,\,\, g_2 (\tau )\,\,d\tau\, \\\\ +\,\int^t_0 d\tau\int^L_0 \, [\,\theta\, (|x-\xi|,\, t-\tau)- \theta (x+\xi,\,t-\tau )] \,\,\, F\,(\,\xi,\tau,\,u(x,\tau))\, \,\,d\xi. \end{split}$$ $ \bullet $ Integro differential equation for (\[23\]) (MBC): $$\label{42} \begin{split} u(\, x,\,t\,)\, = \,\,\int^L_0 \, [\theta^* \,(|x-\xi|,\, t)\,- \theta^* (x+\xi,\,t)\,]\, \,u_0(\xi)\,\, d\xi \,\,- \,\\ \\ \,2 \, \varepsilon \,\int^t_0 \theta^*_x\, (x,\, t-\tau) \,\,\, h_1 (\tau )\,\,d\tau\,+\, 2\,\, \varepsilon \int^t_0 \theta^*\, (L-x,\, t-\tau) \,\,\, h_2 (\tau )\,\,d\tau\, \\\\ +\,\int^t_0 d\tau\int^L_0 \, [\,\theta^*\, (|x-\xi|,\, t-\tau)- \theta^* (x+\xi,\,t-\tau )] \,\,\, F\,(\,\xi,\tau,\,u(x,\tau))\, \,\,d\xi. \end{split}$$ Now, if $ \,{\cal B}_ T \, $ denotes the Banach space $$\label{43} \,{\cal B}_ T \, \equiv \, \,\bigg\{\, z\,(\,x,t\,) : \, z\, \in C \,(\Omega_T), \, \,\, ||\,z\,|| \,= \displaystyle \sup _{ \Omega_T\,}\, | \, z \,(\,x,\,t) \,|, \,\, < \infty \bigg \}$$ and $ D $ is the following set: $$D\equiv \{(x,t,u) : (x,t) \in \Omega_T , -\infty < u <\infty\,$$ then, let assume the source term $ F(x,t,u) \, $ be defined and continuous on $ D $ and uniformly Lipschitz continuous in $(x,t,u)$ for each compact subset of $ \Omega_T. $ Besides, let $ F $ be a bounded function for bounded $ u $ and there exists a constant $ C $ such that: $$|F(x,t,u_1)-F(x,t,u_2)| \leq\,\, C\,\, |u_1-u_2|.$$ So, by means of standard methods related to integral equations and owing to basic properties of $ K_0,$ it is possible to prove that the mappings defined by (\[41\]) (\[42\]) are a contraction of $ {\cal B}_ T $ in $ {\cal B}_ T $ and so they admits a unique fixed point $ u(x,t) \, \in {\cal B}_ T $. [@c; @dmm] In order to enable a quicker reading, attention will be paid only to the initial boundary value problem with Dirichlet conditions. However, all the following analysis can be applied to the mixed problem,too. At first, let us consider $ g_i =0\,\,(i=1,2) $ and let $$\,\,\,||\,u_0\,|| \,= \displaystyle \sup _{ 0\leq\,x\,\leq \,L\,}\, | \,u_0 \,(\,x\,) \,|, \,\,\qquad||\,F\,|| \,= \displaystyle \sup _{ \Omega_T\,}\, | \,F \,(\,x,\,t,\, u) \,|, \,$$. In [@dr8] the following theorem has been proved: When $ g_i\,=\,0 \, \,\,\,(i=1,2),\, $ solution (\[41\]), for large $ t , $ verifies the following estimate: $$\label{44} |u(x,t)| \, \leq \,\, 2 \,\,\bigl[\,\,||\,F\,|| \,\, \beta_0\,+\, \, \,\,||\,u_0\,|| \,\, ( 1\, +\, \sqrt b \,\pi \,t \, ) \,\,e^{- \omega \, t\,}\, \bigr]$$ where $ \, \omega = \min \,(a, \beta )$ and $ \beta_0\,$ is defined by $(\ref{35})_{2}$. As for contributes of boundary data, the well known theorem will be considered: [@b] Let $ h(t) $ and $ \chi(t) $ be two continuous functions on $ [0,\infty [.$ If they satisfy the following hypotheses $$\label{hp} \exists \,\, \displaystyle{\lim_{t \to \infty}}\chi(t) \, = \, \chi(\infty)\qquad\exists \,\, \displaystyle{\lim_{t \to \infty}}h(t) \, = \, h(\infty),$$ $$\label{hp2} \dot h(t)\, \in \, L_1 [ \,0, \infty),$$ then, it results: $$\label{47} \lim_{t \to \infty} \,\, \int_o^t \,\chi(t-\tau ) \, \dot h ( \tau ) \, d \tau \,\, = \, \,\chi(\infty) \,\, [\,\,h(\infty) - h(0)\,\,].$$ According to this, it is possible to state: \[theorem asintotico\] Let $ g_ i \,\,\ (i=1,2) \,\,$ be two continuous functions converging for $ t \rightarrow \, \infty . $ In this case one has: $$\label{48} \lim_{t \to \infty } \,\int_0^t \,\theta_x \,(x,\tau)\,\,\, g_i \,(t-\tau)\, \,d\,\tau \, = \, g_{i, \infty} \,\,\,\,\, \frac{1}{2 \, \varepsilon }\,\,\, \frac{\sinh \sigma_0 \,\,(x-L)}{\sinh\ \sigma_0 \, L }$$ where $ \sigma_0 = \sqrt{\biggl(\,a\,\,+ \dfrac{b}{\beta}\biggr) \dfrac{1}{\varepsilon}}.$ Let us apply (\[47\]) with $ h= \int _0^t \theta_x(x,\tau) d\tau \, \,\,\mbox{and}\,\, \chi = g_i \,\,(i=1,2) $. Then, (\[48\]) follows by $(\ref{317})_1.$ An example: estimate for the FitzHugh Nagumo system ==================================================== When $u(x,t) $ is determined, by means (\[15\]), the $v(x,t)\,$ component is given by $$\label{52} v\,(x,t) \, =\,v_0 \, e^{\,-\,\beta\,t\,} \,+\, b\, \int_0^t\, e^{\,-\,\beta\,(\,t-\tau\,)}\,u(x,\tau) \, d\tau.$$ To achieve the expression of the solution $ (u,v), $ let us denote with $\, f_1 \, \ast f_2 \, $ the convolution $$f_1 ( \cdot, t) \ast \, f_2 ( \cdot ,t) \, = \int_0^t \, f_1 ( \cdot, t) \, f_2 \,\,( \cdot , t -\tau) \,d\,\tau.$$ So that, referring to Dirichlet conditions, if $$G(x,\xi, t) \, = \, \theta \,(\,|x-\xi|,\, t\,)\,- \, \theta \,(\,x+\xi,\,t\,),$$ and denoting by $ N(x,t) $ the following known function depending on the data $( u_0, v_0, g_1, g_2)$: $$\label{53} N(x,t)\, =\, -2 \,\varepsilon \, g_1 (t) \, \ast \, \theta_x (x,t) \,+$$ $$+ \, 2\, \varepsilon \, \,g_2 (t) \, \ast \, \theta_x ( x-L , t) \, +\,\int^L_0 \, \,u_0\,(\xi)\,\, G( x, \xi,t) \, d\xi \, - \, e^{\,-\, \beta\, t\, } \, \ast \, \int^L_0 \, v_0( \xi)\,\, G( x, \xi,t) \, d\xi \,,$$ it results: $$\label{54} \begin{split} v\,(x,t) \, = &\,\, v_0 \, e^{\,-\,\beta\,t\,} \,+\, b\, e^{\,-\, \beta\, t\, } \, \ast \, \,N(x,t) \\ \\& +\,b\,\,\, e^{\,-\, \beta\, t\, } \, \ast \,\int _0^L \, G \, (\, x, \xi , t-\tau) \,\, \ast \,\varphi\,[\,\xi,\,\tau,\,u(\xi, \tau)]\, \,]\}\,\,d\xi\,. \end{split}$$ So, the asymptotic effects due to initial disturbances are vanishing, while the effects of the source terms are bounded. Indeed, letting $$||\,u_0\,|| \,= \displaystyle \sup _{ 0\leq\,x\,\leq \,L\,}\, | \,u_0 \,(\,x\,) \,|, \,\,\qquad ||\,v_0\,|| \,= \displaystyle \sup _{ 0\leq\,x\,\leq \,L\,}\, | \,v_0 \,(\,x\,) \,|, \,$$ and $$||\,\varphi\,|| \,= \displaystyle \sup _{ \Omega_T\,}\, | \,\varphi \,(\,x,\,t,\,u) \,|,$$ by means of (\[51\]) (\[41\]) and (\[54\]) and owing to the estimates $(\ref{33})_1$, (\[37\]), (\[38\]), the following theorem holds: For regular solution $ (u,v) $ of the (FHN) model, when $g_1\,\,= g_2 \,= \,0,\, \, $ the following estimates hold: $$\label{55} \left\{ \begin{array}{lll} \left| u \, \right| \, \leq 2\,[\,\left\| u_0 \right\| \, (1+\pi \sqrt b \, t ) \, e^ {\,-\omega\,t\,}\,+\,\left\| v_0 \right\|\,E(t) \, +\, \beta_0 \,\left\| \varphi \right\|\,] \\ \\ \left| v \, \right| \, \leq \left\| v_0 \right\|\, e^ {\,-\,\beta\,t\,}\,+\,2\,[\,b\,(\,\left\| u_0 \right\|\,+\, t\, \left\| v_0 \right\|\,) \, E(t) \, + \, b\, \beta_1\, \left\| \varphi \right\| \,] \\ \end{array} \right.$$ As for the asymptotic effects of boundary perturbations $ g_1, \, g_2 \, $ by means of (\[48\]), when $ u_0 =0 $ and $ F=0, $ one has $$\label{56} \left\{ \begin{array}{lll} u \, = g_{1,\infty}\,\,\frac{\sinh \sigma_0 \,\,(L-x)}{\sinh\ \sigma_0 \, L } + g_{2,\infty}\,\,\frac{\sinh \sigma_0 \, x}{\sinh\ \sigma_0 \, L } \big| \\ \\ v \, = \, \,\dfrac{b}{\beta} \,\,\bigg[\,\,g_{1, \infty} \,\,\, \,\, \frac{\sinh \sigma_0 \,\,(L-x)}{\sinh\ \sigma_0 \, L } \,\,+ \,\,g_{2, \infty} \,\,\, \,\, \frac{\sinh \sigma_0 \,\,(x)}{\sinh\ \sigma_0 \, L } \,\bigg]. \\ \end{array} \right.$$ Remarks ======= $ \bullet $ The paper is concerned with the nonlinear integral equation (\[11\]) whose kernel is a Green function with numerous basic properties typical of the diffusion equation. $ \bullet $ Neumann, Dirichlet and mixed boundary conditions are considered, and integro differential formulations of [*non linear*]{} problems are obtained. $ \bullet $ The asymptotic behavior for initial boundary value problem with Dirichlet conditions is evaluated, showing that effects due to initial disturbances vanish, while the influences of the source term and boundary perturbations are everywhere bounded. $ \bullet $The analysis related to Dirichlet conditions can be applied to mixed problem, too. Indeed, like $ \theta(x,t), $ also the Green function $ \theta^*(x,t) $ defined in (\[27\]) depends on the fundamental solution $ K_0. $ $ \bullet $ The equivalence among equation (\[11\]) and numerous models allow us to apply asymptotic theorems to many other problems related to various physical fields. Acknowledgment {#acknowledgment .unnumbered} ============== This work has been performed under the auspices of G.N.F.M. of I.N.d.A.M. and of Programma F.A.R.O. (Finanziamenti per l’ Avvio di Ricerche Originali, III tornata) “Controllo e stabilita’ di processi diffusivi nell’ambiente”, Polo delle Scienze e Tecnologie, Universita’ degli Studi di Napoli Federico II (2012). [99]{} Bini D., Cherubini C., Filippi S.Viscoelastic Fizhugh-Nagumo models. Physical Review E 041929 (2005) P.Renno, M. De Angelis, [*Diffusion and wave behavior in linear Voigt model.*]{} C. R. Mecanique **330** (2002)21-26 De Angelis, M. , Monte, A.M. , Renno, P. [*On fast and slow times in models with diffusion* ]{}Mathematical Models and Methods in Applied Sciences 12,12 (2002), 1741-1749 J.A. Morrison, Wave propagations in rods of Voigt material and visco-elastic materials with three-parameters models, Quart. Appl. Math. 14 (1956) 153-169. Flavin J.N., S.Rionero [*Qualitative estimates for Partial Differential Equations: an introduction*]{}. Boca Raton, Florida: CRC Press(1996) Morro, A., Payne.L. E., Straughan,B.: [ *Decay, growth,continuous dependence and uniqueness results of generalized heat theories*]{}. Appl. Anal.,[ 38]{} (1990)231-243 B. Straughan, Heat Waves, in: Springer Series in Applied Mathematical Sciences, vol. 177, 2011. Lamb,H.: [ *Hydrodynamics*]{}. Cambridge University Press (1971) A. Barone, G. Paternó, Physics and Applications of the Josephson Effect, Wiley-Interscience, New-York, 1982, and references therein. M.D. Angelis, G. Fiore, [*Existence and uniqueness of solutions of a class of third order dissipative problems with various boundary conditions describing the Josephson effect,*]{} J. Math. Anal. Appl. , **404, Issue 2,** (2013) 477-490 A. Benabdallah; J.G.Caputo; A.C. Scott [*Laminar phase flow for an exponentially tapered Josephson oscillator*]{} Appl. Phys. **88**,6 (2000) 3527-3540 M.De Angelis, G. Fiore, [*Diffusion effects in a superconductive model*]{} Communications on Pure and Applied Analysis Volume 13, Number 1, January 2014 DOI[10.3934/cpaa.2014.13.]{} M. De Angelis, [*On exponentially shaped Josephson junctions*]{} Acta Appl. Math **122 iussue I** (2012) 179-189 S.A. Cybart et al., Dynes Series array of incommensurate superconducting quantum interference devices Appl. Phys Lett 93 (2008) 182502 1-3 Murray, J.D. : [ *Mathematical Biology. I. and II* ]{}. Springer-Verlag, N.Y (2002) M. De Angelis, P. Renno,[*O*n the Fitzhugh Nagumo model]{} in: 14 Conferences on Waves and stability in Continuous Media Word Sci. Publ. Hackensack,N.Y. 2008 193-198 P. Renno, M. De Angelis [*Existence, uniqueness and a priori estimates for a non linear integro - differential equation* ]{}Ricerche di Mat. **5**7 (2008) 95-109 P. Renno, M.De Angelis, [*Asymptotic effects of boundary perturbations in excitable systems*]{} submitted to Discrete and Continuous Dynamical Systems - Series B http://arxiv.org/pdf/1304.3891v1.pdf A.D’Anna,M. De Angelis,G. Fiore, [*Existence and Uniqueness for Some 3rd Order Dissipative Problems with Various Boundary Conditions*]{} Acta Appl. Math. **122** (2012), 255-267. Jaworski M. Fluxon dynamics in an exponentially shaped Josephson junction. Physical Review B 71(2005)214515 6 pages Keener, J. P. Sneyd,J. [ *Mathematical Physiology* ]{}. Springer-Verlag, N.Y (1998) O. Nekhamkina and M. Sheintuch [*Boundary-induced spatiotemporal complex patterns in excitable systems*]{} Phys. Rev. E **73**, (2006).066224 1-4 A. Dikansky [*Fitzhugh-Nagumo equations in a nonhomogeneous medium*]{} Discrete and continuous dynamical systems Supplement Volume 2005 pag 216- 224 R. Artebrant [*Bifurcating Solutions to the Monodomain Model Equipped with FitzHugh-Nagumo Kinetics* ]{} J. Appl.Math. doi:10.1155/2009/292183 M Marion [*F*inite dimensional attractors associated to partly dissipative reaction diffusion equations]{}, SIAM J Math Anal **20,no 4** (1989), 816-844. M.De Angelis [On a parabolic operator of dissipative system]{} http://arxiv.org/abs/1307.1887 Torcicollo I., On the Dynamics of the nonlinear duopoly game. International Journal of Non-Linear Mechanics., 2013, doi: 10.1016/j.ijnonlinmec.2013.06.011 F. Capone · V. De Cataldis · R. De Luca [*On the nonlinear stability of an epidemic SEIR reaction-diffusion model*]{} Ricerche mat. (2013) 62: 161-181 M.De Angelis [*A priori estimates for excitable models*]{} Meccanica (2013) DOI: 10.1007/s11012-013-9763-2 A. Raheem [*Existence and Uniqueness of a Solution of Fisher-KKP Type Reaction Diffusion Equation*]{} Nonlinear Dynamics and Systems Theory, [**13**]{}(2) (2013) 193–202 M. Gentile. B. Straughan, [*Hyperbolic diffusion with Christov - Morro theory*]{} Mathematics and Computers in Simulation (2012) doi.org/10.1016/j.matcom.2012.07.010 Bergeret, Volkov, Efetov [*Exotic Proximity Effects in Superconductor*]{}Science Series II: Mathematics, Physics and Chemistry Volume 154 Springer 2004 J Rauch J. Smoller [*Q*ualitative Theory of the FitzHugh-Nagumo Equations]{} Advances in mathematics 27, 12-44 (1978) J. R. Cannon, [*The one-dimensional heat equation* ]{}, Addison-Wesley Publishing Company (1984) M.De Angelis, A.Maio and E.Mazziotti *Existence and uniqueness results for a class of non linear models* in “ Mathematical Physics models and engineering sciences" (eds. Liguori, Italy),(2008), 191–202. L.Berg [*Introduction to the operational calculus,*]{} North Holland Publ. Comp 1967 \[Sample\_NDST: LastPage\] [^1]: University of Naples “Federico II”, Dep. Mat. Appl. “R.Caccioppoli”, Via Claudio n.21, 80125, Naples, Italy. [[`modeange@unina.it`](unina:modeange@unina.it)]{}

ArXiv

--- abstract: 'We introduce a new fingerprint that allows distinguishing between liquid-like and solid-like atomic environments. This fingerprint is based on an approximate expression for the entropy projected on individual atoms. When combined with a local enthalpy, this fingerprint acquires an even finer resolution and it is capable of discriminating between different crystal structures.' author: - 'Pablo M. Piaggi' - Michele Parrinello title: Entropy based fingerprint for local crystalline order --- Introduction ============ Atomistic computer simulation is an important technique used in the study of a broad range of phenomena in materials science, chemistry, and condensed matter physics. In these fields, very often one is faced with the problem of identifying different local arrangements. A paradigmatic case is that of the nucleation of a crystal from the liquid where one is required to distinguish between solid-like and liquid-like atomic environments. The situation is even more complicated in systems exhibiting polymorphism since in these cases it is desirable to classify the atoms as belonging to one of the different polymorphic structures. This is a common occurrence in nucleation studies where Ostwald’s step rule is observed [@tenWolde99; @Giberti15] or where clusters exhibit a core-shell structure [@tenWolde95; @Lechner11]. Another area where the ability to distinguish between different local arrangements plays a role is in the identification of crystallites in nanocrystalline materials [@Meyers06]. Several methods have been proposed to distinguish between liquid-like and solid-like atoms and to identify local crystalline structures. One such method is the common neighbor analysis (CNA) [@Honeycutt87; @Stukowski12] which is an efficient algorithm able to distinguish between liquid, bcc, fcc, and hcp phases. However, it lacks robustness with respect to particle displacements such as those arrising from thermal motion or stresses. Another popular method is based on the local Steinhardt parameters [@Lechner08] which are local, averaged versions of the original Steinhardt parameters [@Steinhardt83]. However, they also come at a high computational cost and presume that the nature of the crystal structure is known beforehand. This work is inspired by a recent progress in the study of nucleation using metadynamics[@Laio02; @Barducci08] to enhance the probability of inducing the crystal formation in an accessible computer time. Metadynamics relies on the identification of appropriate collective variables (CVs). In Ref.  we found that enthalpy and an approximate expression for entropy based on the two body correlation function, were useful CVs in this context. One of the features of this work was that the CVs did not contain any information on the geometry of the crystal structure. This suggested that perhaps from these two quantities one could extract fingerprints able to distinguish between different local atomic arrangements. Enthalpy and entropy are global properties and in order to be able to use them as local parameters we have to project them onto each atom. We propose a method that is able to do so. We find that the local entropy thus defined is able to distinguish extremely well between solid-like and liquid-like atoms. Furthermore, in conjuction with local enthalpy it can distinguish well between different polymorphs, even in the subtle case of the difference between fcc-like and hcp-like arrangements. Entropy approximation based on the two body correlation function ================================================================ {width="95.00000%"} \[sec:PairEntropy\] Ref.  was based on the consideration that in the liquid to solid transition there is a trade-off between entropy and enthalpy. The role of metadynamics was there to enhance the fluctuations of these two quantities so as to accelerate crystallization. This required designing CVs able to describe these two quantities. Enthalpy is easy to compute but entropy is extremely costly to evaluate. However, an expression that gives an approximate evaluation of the entropy is sufficient for the purpose of driving crystallization. Such an expression was derived from an expansion of the configurational entropy in terms of multibody correlation functions[@Green52; @Nettleton58; @Baranyai89]. In simple liquids the second term of the expansion, often called two-body excess entropy, involves only the pair correlation function and accounts for about 90% of the configurational entropy [@Wallace87; @Wallace94; @Laird92; @Baranyai89]. This term is given by, $$S_2 = -2\pi\rho k_B \int\limits_0^{\infty} \left [ g(r) \ln g(r) - g(r) + 1 \right ] r^2 dr, \label{eq:pair_entropy}$$ where $\rho$ is the system’s density, and $g(r)$ is the radial distribution function. Extensions of the expansion to multicomponent[@Hernando90; @Prestipino04] and inhomogeneus[@Morita61] systems are also available. We also recall that entropy series expansions have been used to study order-disorder phenomena starting with the landmark work of Kikuchi[@Kikuchi51]. In order to come to grasp with $S_2$ and understand better why it works, we first contrast in Fig. \[fig:Figure1\] the different behaviors of $g(r)$ and the integral in Eq. \[eq:pair\_entropy\] $I(r)=[ g(r) \ln g(r) - g(r) + 1 ] r^2$. The data were taken from a system with Lennard-Jones interactions at temperature $T=1.15$ and pressure $P=5.68$, that corresponds to the solid-liquid coexistence point[@Hansen69]. The Lennard-Jones potential was truncated at 2.5 and tail corrections were included. We refer the reader to Appendix \[sec:appendixA\] for further computational details. As usual we use Lennard-Jones units [@FrenkelBook], i.e. $\sigma=1$ and $\epsilon=1$. We have chosen these thermodynamic conditions because at this temperature and pressure the fcc, hcp, bcc, and liquid phases are all metastable allowing a fair comparison. The first observation is that while $g(r)$ has some difficulty at distinguishing betweem solid and liquid, it strikes the eye that $I(r)$ in the liquid phase is much more short ranged than in the solid phases. Furthermore, the $g(r)$ for the solid phases can hardly distinguish between the different polymorphs. In contrast, the bcc $I(r)$ appears clearly different from that of the closed packed structures. More subtle is the difference between fcc and hcp, that is revealed only if one goes as far out as the third neighbor shell. Entropy fingerprint for solid-like and liquid-like environments {#sec:entropy_fingerprint} =============================================================== The analysis of $I(r)$ suggests that, if properly projected onto the different atoms, $S_2$ could be used as a fingerprint to identify local structures. The projection on atom $i$ can be achieved using the expression: $$s_S^i = -2\pi\rho k_B \int\limits_0^{r_{m}} \left [ g_m^i(r) \ln g_m^i(r) - g_m^i(r) + 1 \right ] r^2 dr, \label{eq:entropy_parameter}$$ where $r_{m}$ is an upper integration limit that in principle should be taken to infinity, and $g_m^i$ is the radial distribution function centered at the $i$-th atom. To obtain a continuous and differentiable order parameter, we define a mollified version of the radial distribution function[@Piaggi17], $$g_m^i(r) = \frac{1}{4 \pi \rho r^2} \sum\limits_{j} \frac{1}{\sqrt{2 \pi \sigma^2}} e^{-(r-r_{ij})^2/(2\sigma^2)} , \label{eq:mollified_rdf}$$ where $j$ are the neighbors of atom $i$, $r_{ij}$ is the distance between atoms $i$ and $j$, and $\sigma$ is a broadening parameter. We shall choose $\sigma$ so small that $g_m(r) \sim g(r)$ yet large enough for the derivatives relative to the atomic positions to be manageable[@Piaggi17]. A similar projection of $S_2$ has been used in Ref. . If we use $s_S^i$ as defined in Eq. (\[eq:entropy\_parameter\]) it can be seen in Fig. \[fig:Figure2\] that, in the cases of Na[@Wilson15] at 350 K and Al[@Sturgeon00] at 900 K (see Appendix \[sec:appendixA\] for technical details), the distribution of $s_S^i$ in the liquid and solid phases are peaked at two different positions but exhibit a large overlap. {width="95.00000%"} In order to calculate local order parameters whose distributions are more clearly distinct, we take cue from Lechner and Dellago [@Lechner08] and define an average local entropy: $$\bar{s}_S^i = \frac{\sum_j s_S^j f(r_{ij}) + s_S^i}{\sum_j f(r_{ij})+1} \label{eq:entropy_parameter_avg}$$ where $j$ runs over the neighbors of atom $i$ and $f(r_{ij})$ is a switching function with cutoff $r_{a}$. Switching functions have a value of 1 for $r_{ij} \ll r_{a}$, 0 for $r_{ij} \gg r_{a}$, and decay smoothly from 1 to 0 for $r_{ij} \approx r_{a}$. We have used a switching function with the functional form: $$f(r_{ij}) = \frac{1-(r_{ij}/r_{a})^N}{1-(r_{ij}/r_{a})^M}$$ with $N=6$ and $M=12$. Such a form has proven useful in many other contexts[@Tribello14]. At variance with $s_S^i$, the distributions of $\bar{s}_S^i$ of the liquid and solid phases now have a negligible overlap (see Fig. \[fig:Figure2\]). Henceforth, we shall drop the index $i$ when referring to distributions and we shall refer to $\bar{s}_S$ as entropy fingerprint. The ability to distinguish sharply between solid-like and liquid-like molecules depend on a wise choice of the parameters $r_m$ and $r_a$. As $r_m$ is increased, more of the long range part of the integrand is included making the difference between liquid and solid more and more evident. On the other hand by increasing $r_a$, more neighbors are included in the summation in Eq. (\[eq:entropy\_parameter\_avg\]) and eventually the locality of $\bar{s}_S$ is lost. In the practice we have chosen for $r_m$ and $r_a$ the smallest values that still ensure sharp distinction between solid-like and liquid-like atoms. The parameters $r_{m}$, $r_{a}$, and $\sigma$ that were used are summarized in Table \[tab:table1\]. Structure Model T (K) $r_{m}$ ($a$) $r_{a}$ ($a$) ----------- ------- ------- --------------- --------------- bcc Na 350 1.8 (5NS) 1.2 (2NS) fcc Al 900 1.4 (3NS) 0.9 (1NS) : \[tab:table1\] Parameters in the definition of $s_S$ and $\bar{s}_S$ for different structures. The columns represent the crystal structure, the model system, the temperature (T) at which the distributions of solid and liquid phases are compared, and the parameters $r_{m}$ and $r_{a}$ defined in Eq. (\[eq:entropy\_parameter\]) and (\[eq:mollified\_rdf\]). $r_{m}$ and $r_{a}$ are in units of the lattice constant, $a=4.23$ Å for Na and $a=4.05$ Å for Al. We report the number of neighbor shells (NS) corresponding to $r_m$ and $r_a$. For both cases $\sigma=0.02$ nm. It is interesting to investigate whether the entropy fingerprint can identify ordered structures in a complex situation, in a context different from nucleation. To this effect we generated a nanocrystalline structure (see Fig. \[fig:Figure3\]) using a procedure described in Appendix \[sec:appendixA\]. ![\[fig:Figure3\] Nanocrystalline Al with mean grain size 5 nm at 300 K. Atoms are colored according to $\bar{s}_S$ (see text for details). The colorscale is such that green and blue atoms have ordered and disordered environments, respectively. Image obtained with <span style="font-variant:small-caps;">OVITO</span> [@Stukowski09]. ](Figure3.png){width="48.00000%"} The system is Al, as described by the potential in Ref. . It can be seen that the entropy fingerprint clearly brings out the nanostructure of the system and the network of grain boundaries. This indicates that the entropy fingerprint can also work in inhomogeneous situations where different atomic environments coexist. Identification of crystal structures ==================================== In the previous section we have shown that $\bar{s}_S$ is able to distinguish liquid-like from solid-like atomic environments. We will now explore the possibility of distinguishing between fcc, hcp, bcc and liquid-like atomic environments. As we shall see, this is best achieved if we accompany our definition of local entropy with a measure of local enthalpy. The local enthalpy is easily defined if we consider an interatomic potential $U(\mathbf{R})$ that can be decomposed into energies $U_i(\mathbf{R})$ associated to individual atoms. Here $\mathbf{R}$ denotes the atomic coordinates of an $N$ atom system. The expression that we shall use is then, $$s_H^i = U_i(\mathbf{R})+ PV/N \label{eq:enthalpy_parameter}$$ where $P$ and $V$ are the system’s pressure and volume, respectively and, for simplicity, we have partitioned the volume of the system into $N$ equal parts. A more complex partition criterion is also possible. As done for the local entropy, we define an average local enthalpy, $$\bar{s}_H^i = \frac{\sum_j s_H^j f(r_{ij}) + s_H^i}{\sum_j f(r_{ij})+1} \label{eq:enthalpy_parameter_avg}$$ where the symbols have the same meaning as in Eq. (\[eq:entropy\_parameter\_avg\]). We calculated the joint probability distributions of $\bar{s}_H$ and $\bar{s}_S$ ($P(\bar{s}_H,\bar{s}_S)$) of the fcc, hcp, bcc, and liquid phases of the Lennard-Jones system described in Section \[sec:PairEntropy\]. For this purpose we simulated systems in each of those phases for 200 ps. The thermodynamic conditions were the same as described in Section \[sec:PairEntropy\]. We used the following parameters to define $\bar{s}_H$ and $\bar{s}_S$: $r_{m}=r_{a}=2.5$, and $\sigma=0.1$. The $P(\bar{s}_H,\bar{s}_S)$ of each phase are shown in Fig. \[fig:Figure4\]. {width="95.00000%"} Each $P(\bar{s}_H,\bar{s}_S)$ was normalized to one. We now discuss the results in Fig. \[fig:Figure4\]. We first notice that the distributions of the different phases in Fig. \[fig:Figure4\] have minimal overlap and therefore $\bar{s}_H$ and $\bar{s}_S$ are useful fingerprints. As in the case of Na and Al, the distributions of liquid and solid phases are very far apart and therefore the fingerprints distinguish very well between liquid-like and solid-like environments. The distributions in the solid phases are clustered together in the region of low enthalpy and entropy, and it is easy to distinguish between the structures using $\bar{s}_H$ and $\bar{s}_S$. We analyze in detail the challenging case of fcc and hcp. Both fcc and hcp structures are formed by stacking of close-packed planes. However, they differ in the way the close-packed planes are stacked. For this reason, these structures are usually not easy to discriminate. As seen in Fig. \[fig:Figure4\], the fingerprints introduced in this work discriminate well between fcc and hcp configurations. However, a large value of $r_a$ was necessary. Conclusions =========== To conclude, the degree of success of the entropy based fingerprint is at first sight surprising. However, the root of this success must lie on the point of view taken here that does not directly focus on the local geometry but on properties of deeper thermodynamic significance, like local entropy and enthalpy. It also points to the usefulness of looking at old problems from a different standpoint. Computational details {#sec:appendixA} ===================== We performed molecular dynamics (MD) simulations using <span style="font-variant:small-caps;">LAMMPS</span> [@Plimpton95]. We employed an anisotropic Parrinello-Rahman barostat [@Parrinello81] and the stochastic velocity rescaling thermostat [@Bussi07]. The fingerprints were programmed in a development version of <span style="font-variant:small-caps;">PLUMED 2</span> [@Tribello14]. The Lennard-Jones simulations were performed at temperature $T=1.15$ and pressure $P=5.68$ (solid-liquid coexistence[@Hansen69]). As usual, we use Lennard-Jones units [@FrenkelBook], i.e. $\sigma=1$ and $\epsilon=1$. The Lennard-Jones potential was truncated at 2.5 and tail corrections were included. The time step for the integration of the equations of motion was 0.002. The relaxation times of the barostat and thermostat were 5 and 0.05, respectively. Na and Al were simulated using embedded atom models (EAM)[@Wilson15; @Sturgeon00]. The time step for the integration of the equations of motion was 2 fs. For Na we set the temperature at 350 K, close to the melting temperature (366 K) of the model. For Al the temperature was set to 900 K, near the melting temperature 931 K. In both cases the pressure was set to its standard atmospheric value. The relaxation times of the barostat and thermostat were 10 ps and 0.1 ps, respectively. The results presented in Fig. \[fig:Figure2\] were obtained by performing independent simulations in the liquid and solid phases of Na and Al at the above cited temperatures. Each simulation had a length of 200 ps and the distributions of $s_S$ and $\bar{s}_S$ were calculated taking samples every 1 ps. The configuration of the nanocrystalline Al was constructed using Voronoi tesselation[@Piaggi15; @Meyers06]. The mean grain size was 5 nm and the system contained 255064 atoms. We performed an annealing at 600 K for 0.2 ns, then the temperature was ramped to 300 K in 0.2 ns, and finally the temperature was kept constant at 300 K for 0.2 ns. For these simulations we employed a different EAM potential[@Mendelev08]. The configuration in Fig. \[fig:Figure3\] corresponds to the last in this trajectory. The simulation details were the same as those used for Al above. EAM potentials [@Daw84; @Finnis84] have a natural way to partition the energy between the atoms as needed in Eq. (\[eq:enthalpy\_parameter\]), i.e. $$U_i(\mathbf{R}) = \sum\limits_{j\neq i} \phi(r_{ij}) + F \left (\sum\limits_{j\neq i}\rho_{\mathrm{atom}}(r_{ij}) \right) \label{eq:energy_partition}$$ where $\phi$ is a pairwise potential, $F$ is the embedding energy function, and $\rho_{\mathrm{atom}}$ is the electron charge density function. We have used this partition criterion. This research was supported by the NCCR MARVEL funded by the Swiss National Science Foundation. The authors also acknowledge funding from the European Union Grant No. ERC-2014-AdG-670227 / VARMET. The computational time for this work was provided by the Swiss National Supercomputing Center (CSCS) under project ID mr3. Calculations were performed in CSCS cluster Piz Daint. [36]{}ifxundefined \[1\][ ifx[\#1]{} ]{}ifnum \[1\][ \#1firstoftwo secondoftwo ]{}ifx \[1\][ \#1firstoftwo secondoftwo ]{}““\#1””@noop \[0\][secondoftwo]{}sanitize@url \[0\][‘\ 12‘\$12 ‘&12‘\#12‘12‘\_12‘%12]{}@startlink\[1\]@endlink\[0\]@bib@innerbibempty [****,  ()](\doibase 10.1039/a809346f) [****,  ()](\doibase 10.1016/j.ces.2014.08.032) [****,  ()](\doibase 10.1103/PhysRevLett.75.2714) [****,  ()](\doibase 10.1103/PhysRevLett.106.085701) [****,  ()](\doibase 10.1016/j.pmatsci.2005.08.003) [****,  ()](\doibase 10.1021/j100303a014) [****,  ()](\doibase 10.1088/0965-0393/20/4/045021) [****,  ()](\doibase 10.1063/1.2977970) [****,  ()](\doibase 10.1103/PhysRevB.28.784) [****,  ()](\doibase 10.1073/pnas.202427399) [****,  ()](\doibase 10.1103/PhysRevLett.100.020603) [****,  ()](\doibase 10.1103/PhysRevLett.119.015701) @noop [**]{} (, ) [****,  ()](\doibase 10.1063/1.1744724) [****, ()](\doibase 10.1103/PhysRevA.40.3817) [****,  ()](\doibase 10.1063/1.453158) [****,  ()](\doibase 10.1002/qua.560520215) [****,  ()](\doibase 10.1103/PhysRevA.45.5680) [****,  ()](\doibase 10.1080/00268979000100211) [****,  ()](\doibase 10.1088/1742-5468/2004/09/P09008) [****,  ()](\doibase 10.1143/PTP.25.537) [****, ()](\doibase 10.1103/PhysRev.81.988) [****, ()](\doibase 10.1103/PhysRev.184.151) @noop [**]{}, Vol.  (, ) [****,  ()](\doibase 10.1063/1.4769981) [****,  ()](\doibase 10.1063/1.4916741) [****, ()](\doibase 10.1103/PhysRevB.62.14720) [****, ()](\doibase 10.1016/j.cpc.2013.09.018) [****,  ()](\doibase 10.1088/0965-0393/18/1/015012) [****,  ()](\doibase 10.1080/14786430802206482) [****,  ()](\doibase 10.1006/jcph.1995.1039) [****,  ()](\doibase 10.1063/1.328693) [****,  ()](\doibase 10.1063/1.2408420) [****,  ()](\doibase 10.1016/j.jnucmat.2014.12.069) [****, ()](\doibase 10.1103/PhysRevB.29.6443) [****,  ()](\doibase 10.1080/01418618408244210)

ArXiv

--- abstract: 'We propose a multielectron approach to calculate superexchange interaction in magnetic Mott-Hubbard insulator La$_2$CuO$_4$(further La214) that allows to obtain the effect of optical pumping on the superexchange interaction. We use the cell perturbation theory with exact diagonalization of the multiband $pd$ Hamiltonian inside each CuO$_6$ unit cell and treating the intercell hopping as perturbation. To incorporate effect of optical pumping we include in this work the excited single-hole local states as well as all two-hole singlets and triplets. By projecting out the interband intercell electron hopping we have obtained the effective Heisenberg-like Hamiltonian with the local spin at site $R_i$ being a superposition of the ground and excited single-hole states. We found that antiferromagnetic contribution to the exchange energy in La214 will increase in accordance to [$\sim 4\cdot10^{-3} eV(\%)^{-1}$]{} at the resonance light occupation of the excited single hole in-gap state.' author: - 'Vladimir A. Gavrichkov' - 'Semen I. Polukeev' - 'Sergey G. Ovchinnikov' bibliography: - 'my.bib' title: 'In-gap excitation effect on a superexchange in La$_2$CuO$_4$ by creating nonequilibrium photoexcited centers' --- \[sec:intr\]Introduction\ ========================= Understanding the energy transfer between charge, orbital, and spin degrees of freedom is the important problem for many fields of solid state physics. Since the first experiments  [@Beaurepaire_etal1996; @Hohlfeld_etal1997] optical excitation of electronic spins and ultrafast magnetization dynamics have obtained much attention.  [@Stanciu_etal2007; @Ostler_etal2012] A possibility to control the exchange interaction by light is important in many physics areas, from quantum computing [@Duan_etal2003; @Trotzky_etal2008; @Chen_etal2011] to strongly correlated materials. [@Wall_etal2009; @Forst_etal2013; @Li_etal2013] In many experiments the effect of optical pumping on the exchange interaction in the Mott-Hubbard insulators like manganites, [@Wall_etal2009] ferroborates, [@Kalashnikova_etal2007; @Kalashnikova_etal2008] TmFeO3, ErFeO3 [@Mikhaylovskiy_etal2014] etc. has been found. The origin of interatomic exchange interaction in all these oxides is related to the superecxhange mechanism via oxygen. [@Anderson_1950] There are some simplified model calculations of the super exchange interaction under light irradiation in the three atomic model cation1-oxygen-cation2, [@Moskvin_etal1979] that in complete theory should be extended to the crystal lattice. The calculation of the superexchange interaction for the crystal lattice can be easily done for some simplified model like the Hubbard model. [@Bulaevskii_etal1968; @Chao_etal1977; @Hirsch_1987] Within the LDA+DMFT approach the first-priciple calculations of the exchange interaction in correlated materials has been carried out in the work. [@Katsnelson_etal2000] An idea of generalization of this approach to nonequilibrium optically excited magnetics has been also proposed in work. [@Secchi_etal2013] without any practical conclusions. Nevertheless up to now the microscopic calculation of the superexchange interaction in La214 under light irradiation is absent. It is known that in the Hubbard model the superexchange $J$ results from the projecting out the interatomic hopping $t^{ab}$ accompanying the interband excitation from the low Hubbard band (LHB=a) to the upper Hubbard band (UHB=b). Due to the large insulator gap $U>>t^{ab}$ the interband excitation requires too much energy, and only virtual interband excitations from LHB to UHB and back are possible providing the exchange coupling $J \sim (t_{ab})^2/U$. [@Bulaevskii_etal1968; @Chao_etal1977; @Hirsch_1987] The convenient mathematical tool for projecting out the irrelevant at large $U$ UHB is given by the projection operators. [@Chao_etal1977] In this paper we calculate the exchange interaction in La214 under optical pumping within the hybrid LDA+GTB (generalized tight binding) approach. Previously we have carried out similar calculation for La214 in the ground state. [@Gavrichkov_etal2008] The LDA+GTB method allows to calculate the electronic structure of strongly correlated oxides like cuprates, [@Korshunov_etal2005] manganites, [@Gavrichkov_etal2010] boroxide [@Ovchinnikov_2003; @Ovchinnikov_Zabluda2004] and cobaltates. [@Orlov_etal2013] We use the cell perturbation theory with exact diagonalization of the multiband $pd-$ Hamiltonian inside each CuO$_6$ unit cell with $ab~initio$ calculated parameters and treating the intercell hopping as perturbation. We restrict ourselves here by the antiferromagnetic undoped cuprate La214, nevertheless all ideas and methods used may be applied to any Mott-Hubbard insulator. To incorporate effect of optical pumping we include in this work the excited single-hole local states as well as all excited two-hole singlets and triplets. It requires a generalization of the projection operators used here in comparison to the papers. [@Chao_etal1977; @Gavrichkov_etal2008] Finally we have obtained the modification of exchange interaction induced by the light irradiation. \[sec:II\]Effective superexchange hamiltonian\ ============================================== In the GTB approach one can assume that the quasiparticles are unit cell excitations which can be represented graphically as single-particle excitations (transitions) between different sectors $N_h=...(N_{0}-1), N_{0}, (N_{0}+1),...$ of the configuration space of the unit cell ($N_{0}$ is hole number per cell in the undoped material, see Fig.\[fig:1\]). [@Ovchinnikov_etal2012] Each of these transitions forms a $r$-th quasiparticle band, where the vector band index $r=\{i,i'\}$ in configurational space  [@Zaitsev_1975] numerates the initial $i$ and final $i'$ many-electron states in the transition. The transitions, with the number of electrons increasing or decreasing, form the conduction or valence bands, respectively. For undoped La214 due to electroneutrality the proper subspace is $d9p6+d10p5$ with one hole per CuO$_6$ cluster, it has one hole, $N_0=1$. The hole addition requires $N_{+} = 2$ states $d9p5+d10p4+d8p6$. The hole removal results in $N_- =0$ states that for cuprates is given by a hole vacuum $d10p6$. In the LDA+GTB method the Hamiltonian parameters are calculated ab initio[@Korshunov_etal2005] and the GTB cell approach  [@Gavrichkov_etal1998; @Ovchinnikov_etal2012] is used to take into account strong electron correlations explicity. A crystal lattice is divided into unit cells, so that the Hamiltonian is represented by the sum $H_0+H_1$, where the component $H_0$ is the sum of intracell terms and component $H_1$ takes into account the intercell hoppings and interactions. The component $H_0$ is exactly diagonalized. The exact multielectron cell states $|i\rangle$ ($|i'\rangle$) and energies $\xi_{i}$ are determined. Then these states are used to construct the Hubbard operators of the unit cell $\vec{R}_{f}: X^{i,i'}_f = |i\rangle\langle i'|$, where the meaning of the indexes $i$ and $i'$ is clear from Fig.\[fig:1\]. ![\[fig:1\]](Fig_1_exchange_under_light) $$H_0=\sum_{f} \left\{% \varepsilon_0X^{00}_f+\sum_{l\sigma}\left(\epsilon_l-\mu\right)X_f^{l\sigma,l\sigma} +\sum_{\nu}^{N_\nu}(E_{\nu}-2\mu)X_f^{\nu,\nu} \right\} \label{eq:1}$$ is the sum of intracell terms and component $H_1$ takes into account the intercell hoppings and interactions. Here $$\begin{aligned} {H_1} &&= \sum\limits_{fg} {} \sum\limits_{\lambda \lambda '\sigma } {} t_{fg}^{\lambda \lambda '}c_{f\lambda \sigma }^ + {c_{g\lambda '\sigma }} + h.c. \nonumber \\ &&= \sum\limits_{fg} {} \sum\limits_{rr'} {} t_{fg}^{rr'}\mathop {X_f^r}\limits^ + X_g^{r'} \label{eq:2},\end{aligned}$$ where $t_{fg}^{\lambda \lambda'}$ is the matrix of hopping integrals, and $$\begin{aligned} t_{fg}^{rr'}& = & \sum\limits_{\lambda \lambda '} \sum\limits_\sigma t_{fg}^{\lambda \lambda '}\nonumber\\ &\times &\left[ \gamma _{\lambda \sigma }^*\left( r \right)\gamma _{\lambda '\sigma }\left( r' \right) + \gamma _{\lambda '\sigma }^*\left( r \right)\gamma _{\lambda \sigma }\left( r' \right) \right], \label{eq:3}\end{aligned}$$ where matrix elements: $$\begin{aligned} {\gamma _{\lambda \sigma }}\left( {{r}} \right)& = &\left\langle ({{N_{+},{{M'}_{S'}}})_\mu} \right|c_{f\lambda \sigma } \left|({{N_{0},{M_S}})_l}\right\rangle\times \nonumber \\ & \times &\delta \left( {S',S \pm |\sigma| } \right)\delta \left( {M', M + \sigma } \right), \label{eq:4}\end{aligned}$$ Consideration is restricted by the case with one hole per cell $N_{0}=1$ in the undoped materials and an arbitrary number $N_\lambda$ of the occupied $\lambda$ orbitals, i.e. number of electrons $N_e=2N_\lambda-1$. This is relevant for the high-$T_c$ cuprates. In this case of one hole per cell, the ${\left| {({N_{0}},{M_S})_i} \right\rangle} $ cell states are a superposition of different hole configurations of the same orbital (l) symmetry: $$\left| ({N_{0},{M_{S}}})_{l} \right\rangle = \sum\limits_\lambda ^{} {{\beta _l }\left( {{h_\lambda }} \right)\left| {{h_\lambda },{M_{S}}} \right\rangle } \label{eq:5}$$ Thus, there are one-hole spin doublet states, $C_{{2N_\lambda}}^{1} = {2N_\lambda}$, where $C^k_n$ is the number of combinations. Besides, there are $N_\mu=N_S+3N_T=C^2_{2N_\lambda}$ of the spin singlets $N_S=C_{N_\lambda}^{2}+N_\lambda$ and triplets $N_T=C_{N_\lambda}^{2}$: $$\left| ({N_+,{M'_{S'}}})_{\mu} \right\rangle = \sum\limits_{\lambda\lambda'} {{B_\mu}\left( {{h_\lambda},{h_{\lambda'}}} \right)\left| {{h_\lambda},{h_{\lambda'}},{M'_{S'}}} \right\rangle } \label{eq:6}$$ in the two-hole sector (Fig.\[fig:1\]) in the ${N_\lambda}$ -orbital approach. Using the intracell Hamiltonian $H_0$ in the cell function representation the configuration weights $\beta_\mu(h_\lambda)$ and $B_\tau(h_\lambda,h_{\lambda'})$ can be obtained by the exact diagonalization procedure for the matrices $(\hat{H_0})_{\lambda\lambda'}$ and $(\hat{H_0})^{{\lambda\lambda'}}_{\lambda''\lambda'''}$ in the $E_i({N_{h},{M_S}})$-eigenvalue problem in different sectors $N_h$. [@Ovchinnikov_etal2012] The sum (\[eq:2\]) over all the $r$-th excited states with $l\neq l_1$ in the sector $N_0$ can not be omitted because of the light pumping. These excited states must be considered along with the $\mu$ - excited states in the nearest $N_+$ sector. The superexchange interaction appears at the second order of the cell perturbation theory with respect to hoppings.[@Jefferson_etal1992] That corresponds to virtual excitations from the occupied singlet and triplet bands through the insulating gap to the conduction band and back. These perturbations are described by the off-diagonal elements $t_{fg}^{rr'}$ with $r=\{0l\}$ and $r'=\{l\mu\}$ in expression (\[eq:2\]). In the Hubbard model, there is only one such element $t^{01}$, which describes the hoppings between the lower and upper Hubbard bands. In order to extract them, we generalize the projection operator method proposed by Chao et al  [@Chao_etal1977] to the multiorbital GTB approach. Since the diagonal Hubbard operators are projection operators, the $X$-representation allows us to construct the set of projection operators. The total number of diagonal operators $X_f^{ii'}$ is equal to $N_\mu+N_l+1$ and the sequence indexes $l$ and $\mu$ ($1 \le l \le N_l$, $1 \le \mu \le N_\mu$) runs over all electron states in the configuration spaces in Fig.\[fig:1\]. Using a set of generalized operators $${p_0} = \left( {X_i^{00} + \sum\limits_l {X_i^{ll}} } \right)\left( {X_j^{00} + \sum\limits_l {X_j^{ll}} } \right), \label{eq:7}$$ and $${p_\mu } = X_i^{\mu \mu } + X_j^{\mu \mu } - X_i^{\mu \mu }\sum\limits_\nu {X_j^{\nu \nu }} \label{eq:8}$$ with $\mu(\nu) =1,2,...N_\mu$ we can identify the contribution to the superexchange from the interband transitions. As will be seen below, a generalized approach with the operators (\[eq:7\]) and (\[eq:8\]) differs from the work [@Chao_etal1977] just in details. It can be checked that each of operator ${p_0 }$ and ${p_\mu }$ is a projection operator $p_0^2=p_0$ and $(p_\mu ^2 = {p_\mu })$. These operators also form a complete and orthogonal system, $p_0+\sum\limits_{\mu = 1}^{N_\mu} {{p_\mu }} = 1$, ${p_0 }{p_\mu } =0$ and ${p_\mu }{p_\nu } = {\delta _{\mu \nu }}{p_\mu }$. We highlight the diagonal and off-diagonal matrix elements in expression: $$\begin{aligned} &H& =(H_0+H_1^{in})+H_1^{out}, \label{eq:9}\end{aligned}$$ According to the work we introduce a Hamiltonian of the exchange-coupled $(ij)$-th pair: $h=(h_0+h_1^{in})+h_1^{out}=H_{ij}$, where $H=\sum\limits_{ij}H_{ij}$ and $$\begin{aligned} h_0+h_1^{in} ={p_0}h{p_0}+\sum\limits_{\mu \nu } {{p_\mu}h{p_\nu}} \label{eq:10}\end{aligned}$$ and $$\begin{aligned} h_1^{out} ={p_0}h\left(\sum\limits_{\mu }{p_\mu}\right)+\left(\sum\limits_{\mu} {p_\mu}\right)h{p_0} \label{eq:11}\end{aligned}$$ are intra- and inter- band contributions in $H_1$ respectively. We perform the standard unitary transformation to project out the interband hopping and to derive superexchange interaction $${\tilde{h}} = {e^{G}}h{e^{-G}}, \label{eq:12}$$ where the matrix $\hat G$ satisfies to the equation $$\begin{aligned} {p_0}{h}\left(\sum\limits_{\mu}{p_\mu }\right)&+&\left(\sum\limits_{\mu}{p_\mu }\right){h}{p_0}+\nonumber\\ &+&{\left[ {{G,\left({p_0}h{p_0}+\sum\limits_{\mu \nu } {{p_\mu}h{p_\nu}}\right)}} \right] } = 0, \label{eq:13}\end{aligned}$$ and transformed Hamiltonian are given by $$\begin{aligned} \tilde{h}&\approx&\left({p_0}h{p_0}+\sum\limits_{\mu \nu } {{p_\mu}h{p_\nu}}\right)+ \nonumber\\ &+&\frac{1}{2}\left[ {{G,\left({p_0}{h}\sum\limits_{\mu}{p_\mu }+\sum\limits_{\mu}{p_\mu }{h}{p_0}\right)}} \right] \label{eq:14}\end{aligned}$$ where the contributions from inter-band transitions can be calculated as: $${p_0}{h}\left(\sum\limits_{\mu}{p_\mu }\right) =\sum\limits_{ll'\mu} {t_{ij}^{l0,l'\mu }X_i^{l0}X_j^{l'\mu }} \label{eq:15}$$ $$\left(\sum\limits_\mu {{p_\mu }}\right) {h}{p_0} = {\sum\limits_{ll'\mu } {t_{ij}^{\mu l',0l}X_i^{\mu l'}X_j^{0l}} } \nonumber$$ Note, due to the absence of additivity over $l$-number of the excited state in the projective operator $p_0$, the solution of Eq.(\[eq:13\]) has the form $$G = \sum\limits_\mu { {\sum\limits_{ll'} {\frac{{t_{ij}^{l0,l'\mu }}}{{{\Delta _{ll'\mu }}}}} \left( {X_i^{\mu l'}X_j^{0l} - X_i^{l0}X_j^{l'\mu }} \right)} } \label{eq:16}$$ where $\Delta_{ll'\mu}=\varepsilon _0 + \varepsilon _\mu - \left( \varepsilon _l + \varepsilon _{l'}\right)$, and the commutator in (\[eq:14\]) can be represented as $$\begin{aligned} \delta h=\frac{1}{2}&&\sum\limits_{\mu \nu } {\left\{ {\left[ {{G_\nu },\left( {{p_0}{h}{p_\mu } +{p_\mu }{h}{p_0}} \right)} \right]} \right\}}= \nonumber \\ && = \frac{1}{2}\sum\limits_{\mu \nu } {\left\{ {\left[ {\sum\limits_{ll'} {\frac{{t_{ij}^{l0,l'\nu }}}{{{\Delta _{ll'\nu }}}}\left( {X_i^{\mu l'}X_j^{0l} - X_i^{l0}X_j^{l'\mu }} \right)} ,\sum\limits_{kk'} {t_{ij}^{k0,k'\mu }\left( {X_i^{0k}X_j^{\mu k'} + h.c.} \right)} } \right]} \right\}}. \label{eq:17}\end{aligned}$$ The right part of the exprexion (\[eq:14\]) for effective Hamiltonian $\tilde h$ can now be derived explicity. Calculating commutator in the above expression (\[eq:17\]) hence we obtain the effective Hamiltonian for the exchange-coupled $(ij)$-th pair as $$\begin{aligned} \delta \tilde{h} = \sum\limits_{ll'kk'} {\sum\limits_{\mu \nu } {\left( {\frac{{t_{ij}^{l0,l'\nu }t_{ij}^{k0,k'\mu }}}{{{\Delta _{ll'\nu }}}}} \right){\delta _{\mu \nu }}\left\{ {X_i^{l \uparrow ,k \uparrow }X_j^{l' \downarrow ,k' \downarrow } + X_i^{l \downarrow ,k \downarrow }X_j^{l' \uparrow ,k' \uparrow } - \left( {X_i^{l \uparrow ,k \downarrow }X_j^{l' \downarrow ,k' \uparrow } + X_i^{l \downarrow ,k \uparrow }X_j^{l' \uparrow ,k' \downarrow }} \right)} \right\}} } &+& \nonumber \\ +{\sum\limits_{ll'kk'}\sum\limits_{\mu \nu } {\left( {\frac{{t_{ij}^{l0,l'\nu }t_{ij}^{k0,k'\mu }}}{{{\Delta _{ll'\nu }}}}} \right) {{\delta _{kl}}{\delta _{k'l'}}\left( {X_i^{00}X_j^{\mu \nu } + X_i^{\mu \nu }X_j^{00}} \right) } } }=\delta {\tilde{h}_{s-ex}}+ \delta {\tilde{h}_\rho }&&,\nonumber \\ \label{eq:18}\end{aligned}$$ and only a first contribution includes superexchange interaction $\delta {H_{s - ex}} = \sum\limits_{ij}\tilde{h}_{s-ex}$: $$\begin{aligned} \delta {H_{s - ex}} = \sum\limits_{ij} {\sum\limits_{ll'kk'} {\sum\limits_\mu {\frac{2{\left( {t_{ij}^{l0,l'\mu }t_{ij}^{k0,k'\mu }} \right)}}{{{\Delta _{ll'\mu }}}}\left\{ {\left( {{\delta _{{l_1}k}}Z_{il}^ + + {\delta _{{l_1}l}}Z_{ik}^ + + {\delta _{lk}}{{\hat S}_{il}}} \right)\left( {{\delta _{{l_1}k'}}Z_{jl'}^ + + {\delta _{{l_1}l'}}Z_{jk'}^ + + {\delta _{l'k'}}{{\hat S}_{jl'}}} \right)} \right.} } }&& - \nonumber \\ \left. { - \frac{1}{4}\left( {{\delta _{{l_1}k}}y_{il}^ + + {\delta _{{l_1}l}}y_{ik}^ - + {\delta _{lk}}{n_{il}}} \right)\left( {{\delta _{{l_1}k'}}y_{jl'}^ + + {\delta _{{l_1}l'}}y_{jk'}^ - + {\delta _{l'k'}}{n_{jl'}}} \right)} \right\}&&,\nonumber\\ \label{eq:19}\end{aligned}$$ where $S_{il}^ + = X_i^{l\uparrow,l\downarrow }$, $2S_{il}^z = \sum\limits_\sigma {\eta \left( \sigma \right)X_i^{{l}\sigma ,{l}\sigma }}$, $Z_l^+= \hat S_{il_1}X_i^{l_1l}$ and $y_{il}^ + = {\hat n_{i{l_1}}}X_i^{{l_1}l}$ are a spin–exciton and electron-exciton operators at the $i$-th cell. For simplicity, we assumed that $X_i^{l\sigma,l'\sigma}=X_i^{l\bar{\sigma},l'\bar{\sigma}}=X_i^{l,l'}$. Note that the contribution in Eq.(\[eq:19\]) at $l=k$ and $l'=k'$ $$\begin{aligned} \delta H_s= \sum\limits_{ij} {\sum\limits_{ll'} {\sum\limits_\mu {\frac{{{{2\left( {t_{ij}^{l0,l'\mu }} \right)}^2}}}{{{\Delta _{ll'\mu }}}}\left({S_{il}}{S_{jl'}}-\frac{1}{4}n_ln_{l'}\right)} } }, \nonumber \\ \label{eq:20} \end{aligned}$$ where $ {\hat S_{il}}{\hat S_{jl'}} = \frac{1}{2}\sum\limits_\sigma {\left( {X_i^{l\sigma l\bar \sigma }X_j^{l'\bar \sigma l'\sigma } - X_i^{l\sigma l\sigma }X_j^{l'\bar \sigma l'\bar \sigma }} \right)} $, is an analogue of the conventional superexchange with exchange constant $J_{ij}^{ll'} = 2\sum\limits_\mu {{{{{\left( {t_{ij}^{l0,l'\mu }} \right)}^2}} \mathord{\left/ {\vphantom {{{{\left( {t_{ij}^{l0,l'\mu }} \right)}^2}} {{\Delta _{ll'\mu }}}}} \right. \kern-\nulldelimiterspace} {{\Delta _{ll'\mu }}}}} $. An exciton energy can not exceed the semiconductor gap ${E_g} = \left[ {{\varepsilon _{{\mu _0}}} + {\varepsilon _0} - 2{\varepsilon _{{l_1}}}} \right]$, because of the divergence of superexchange contributions $\delta H_{s-ex}\rightarrow\infty$ at $\delta_{ll_1}\rightarrow E_g$. At $\delta _{l{l_1}}>E_g$ the exciton cell state decays into an electron-hole pair state. Therefore photocarriers are generated under light pumping with a frequency $h\nu_q$ higher than the absorption edge, and the superexchange on the photoexcited intracell states can be calculated in approach (\[eq:19\]) only at the light pumping with the frequency in the transparency region of the material. It’s partly colored magnetic nondoped materials.  Let’s obtain the contribution (\[eq:20\]) to the exchange energy of the system in the framework of mean-field approximation. $$\begin{aligned} \delta {H_{s - ex}} = - \frac{1}{2}\sum\limits_{ij} {\sum\limits_{ll'} {J_{ij}^{ll'}} \left\langle {X_i^{l\sigma l\sigma }} \right\rangle \left\langle {X_j^{l'\bar \sigma l'\bar \sigma }} \right\rangle } \approx - \frac{{zN}}{2}\left[ {J_{\left\langle {ij} \right\rangle }^{{l_1}{l_1}}p_{{l_1}}^2 + 2\sum\limits_{l \ne {l_1}} {J_{\left\langle {ij} \right\rangle }^{l{l_1}}{p_l}{p_{{l_1}}} + } \sum\limits_{l,l' \ne {l_1}} {J_{\left\langle {ij} \right\rangle }^{ll'}{p_l}{p_{l'}}} } \right] \label{eq:21}\end{aligned}$$ where ${p_{{l_1}}} = 1-{\left(\sum\limits_{l \ne {l_1}}{p_l} {} \right)} $ and ${p_l} = \left\langle {X_i^{l \uparrow l \uparrow }} \right\rangle = \left\langle {X_i^{l \downarrow l \downarrow }} \right\rangle $ is a probability to detect a cell in the excited state $\left| {{{\left( {{N_0},{M_S}} \right)}_l}} \right\rangle $. Thus the light pumping effects in superexchange are frequency selective and linear on the amplitude pumping. In compound La214 the ground cell state is formed by a single hole $b_1$ orbital, the $a_1$ orbital may be excited by the external pumping (Fig.\[fig:2\]).The standard mechanism of the superexchange in the ground state is shown in Fig.\[fig:2\]b, while the superexchange via optically excited term is shown in Fig.\[fig:2\]a. the formation of spin-exciton interaction that is beyond the Heisenberg model is shown in Fig.\[fig:2\]c. \[sec:III\] Results for copper oxide $La214$ ============================================ We test the approach on the high-$T_c$ parent materal La214. At the LDA parameters of Hamiltonian taken from  [@Korshunov_etal2005] $ J_{bb}\approx$0.15 $eV$, $\delta_{ll_1}=\delta_{ab}$=1.78 $eV$, $E_g=$2.00 $eV$, and the $r=\{^2b_1,A_1\}$ - band index  [@Feiner_etal1996; @Gavrichkov_etal2001] corresponds to $\{l_1,\mu=1\}$ first removal electron state. Using the exact diagonalization procedure with LDA parameters, one obtains the weights $\alpha_l$, $\beta_l$ and $A_\mu$, $B_\mu$ at the doublet and singlet, triplet states: $$\left| ({N_{0},{M_{S}}})_{l=1} \right\rangle = {\left| {{}^2{b_1}} \right\rangle } = \sum\limits_{\lambda = {d_z},{p_z},a} {{\beta _{{l=1}}}\left( {{h_\lambda }} \right)\left| {{h_\lambda },{\sigma _{\tfrac{1}{2}}}} \right\rangle } ; \left| ({N_{0},{M_{S}}})_{l=2} \right\rangle ={\left| {{}^2{a_1}} \right\rangle } = \sum\limits_{\lambda = {d_z},{p_z},a} {{\alpha _{l=2}}\left( {{h_a}} \right)\left| {{h_a},{\sigma _{\tfrac{1}{2}}}} \right\rangle }, \label{eq:23}$$ $$\begin{aligned} \left| ({N_+,{M'_{S'}}})_{\mu=1} \right\rangle &=& {\left| {{A_1}} \right\rangle } = \sum\limits_{\lambda ,\lambda ' = b,{d_x},a,{p_{z,}}{d_z}} {A_{\mu=1} }\left( {{h_\lambda },{h_{\lambda '}}} \right)\left| {{h_\lambda },{h_{\lambda '}},{0}} \right\rangle , \nonumber \\ \left| ({N_+,{M'_{S'}}})_{\mu=2} \right\rangle &=& {\left| {{}^3{B_1}} \right\rangle } = \sum\limits_{\lambda = b,{d_x}} {\sum\limits_{\lambda ' = a,{p_z},{d_z}} {{B_{\mu=2} }\left( {{h_\lambda },{h_{\lambda '}}} \right)\left| {{h_\lambda },{h_{\lambda '}},{M_1}} \right\rangle } }, \label{eq:24}\end{aligned}$$ where $h_b$ and $h_{d_x}$ are the holes in the $b$-symmetrized cell states of oxygen and $d_{x^2-y^2}$ cooper states of the CuO$_2$ layer, respectively. Because of $ \delta_{ab}<E_g$, just two contributions from the doublets ${\left| {{}^2{a_1}} \right\rangle }$ and ${\left| {{}^2{b_1}} \right\rangle } $ are available in the sum (\[eq:18\]) over $l$. Due to the symmetry CuO$_2$ layer $\gamma_\lambda(\{^2a_1,A_{1}\})=0$ at any $\lambda$ and therefore $t_{ij}^{{b}0,aA }=t_{ij}^{{a}0,aA }=0$. Thus we evaluate the contribution (\[eq:23\]) like the next: $$\begin{aligned} \left\langle {\delta {H_{s - ex}}} \right\rangle = - \frac{{zN}}{2}\sum\limits_\mu {\left[ {\frac{{{{\left( {t_{}^{b0,b\mu }} \right)}^2}}}{{{\Delta _{b\mu }}}}p_b^2 + 2\left( {\frac{{{{\left( {t_{}^{b0,a\mu }} \right)}^2}}}{{{\Delta _{ba\mu }}}} + \frac{{{{\left( {t_{}^{a0,b\mu }} \right)}^2}}}{{{\Delta _{ba\mu }}}}} \right){p_a}{p_b} + \frac{{{{\left( {t_{}^{a0,a\mu }} \right)}^2}}}{{{\Delta _{b\mu }}}}p_a^2} \right]} \sim &&\nonumber \\ \sim - \frac{{zN}}{2} {\left[ {0.15(eV)\cdot p_b^2 +2 \frac{{{{\left( {t_{}^{a0,bA_1}} \right)}^2}}}{{{\Delta _{baA_1}}}}{p_a}{p_b}} \right]}&& \label{eq:25}\end{aligned}$$ ![\[fig:3\] A dependence of the antiferromagnetic contribution (\[eq:25\]) on the occupation $p_a$ of excited state $|^2a_1>$](Fig_3_exchange_under_light) Without external irradiation $p_b=1$, $p_a=0$, and Eq.(\[eq:25\]) results in the exchange interaction $J_{bb}$ (the first term in the right side of Eq.(\[eq:25\])) in the ground state obtained earlier in the work. [@Chao_etal1977] What are the modifications of the exchange interaction that we can observe in L214 under resonance light pumping? The answer to this question depends on the ratio of the exchange interaction in the ground and excited states. Depletion of the ground state $p_b=1-x$ decreases $J_{bb}$, and a new contribution $J_{ba}$ via excited orbital $a_1$ appears (see Fig.\[fig:2\]). Using LDA parameters, and summing over all $\mu$ in the second term in Eq.(\[eq:25\]), we finally obtain the result shown in Fig.\[fig:3\]. So most likely superexchange contribution (\[eq:25\]) will increase at any small population of excited states in La214 by a factor of $ \sim~ 4\cdot10^{-3} eV(\%)^{-1}$ \[sec:IV\] Conclusion ===================== In summary, we would like to emphasize that optical pumping results in the occupation of some high energy multielectron states with different overlapping of the excited wavefunctions between neighboring ions vs the ground state orbitals. It is evident that this pumping results in the modification of the exchange interaction. Nevertheless an accurate calculation of a large number of contributions from different multielectron excited states is not a trivial theoretical problem. The gain of the Hubbard operators approach is the ability to control each excited state and its contribution to the ionic spin and orbital moment. Our approach to the exchange interaction via excited states is just a straightforward generalization of the previously developed projection technique for the Hubbard model [@Chao_etal1977] and for the ground state of La214 within the realistic multiband $pd$ model. [@Gavrichkov_etal2008] The obtained effective Hamiltonian (\[eq:19\]) contains not only spin-spin interactions via excited states but also more complicated exchange interactions accompanied with exciton or bi-exciton that are beyond standard Heisenberg model. For undoped insulating cuprates the theory results in a prediction of the antiferromagnetic coupling strengthening proportional to the concentration of the excited states At the concentration of excited states 1% an increased exchange interaction is estimated by the magnitude $\sim 40$K. For simplicity we have assumed stationary pumping with resonance absorbtion. Then the spectral dependence of the modified exchange coupling should coincide with the $d-d$ absorption spectrum. Due to the short time of the local electronic excitations $\leq$ 1 (fs) a dynamics of exchange interaction for the time intervals more then 10 (fs) probably can be also treated in our approach. It is evidently that the spin-exciton effects found here may be important in the dynamical regimes. This work was supported by RFFI grants 16-02-00273, No.14-02-00186.

ArXiv

--- abstract: 'A new parametrization of the reionization history is presented to facilitate robust comparisons between different observations and with theory. The evolution of the ionization fraction with redshift can be effectively captured by specifying the midpoint, duration, and asymmetry parameters. Lagrange interpolating functions are then used to construct analytical curves that exactly fit corresponding ionization points. The shape parametrizations are excellent matches to theoretical results from radiation-hydrodynamic simulations. The comparative differences for reionization observables are: ionization fraction $|\Delta x_\text{i}| \lesssim 0.03$, 21cm brightness temperature $|\Delta T_\text{b}| \lesssim 0.7\, \text{mK}$, Thomson optical depth $|\Delta \tau| \lesssim 0.001$, and patchy kinetic Sunyaev-Zel’dovich angular power $|\Delta D_\ell | \lesssim 0.1\, \mu\text{K}^2$. This accurate and flexible approach will allow parameter-space studies and self-consistent constraints on the reionization history from 21cm, CMB, and high-redshift galaxies and quasars.' author: - Hy Trac title: 'Parametrizing the Reionization History with the Redshift Midpoint, Duration, and Asymmetry' --- Introduction {#sec:intro} ============ The reionization of hydrogen by the first stars, galaxies, and quasars is a milestone event in the first billion years. Ionizing radiation from luminous sources convert the cold and neutral gas into a warm and highly ionized medium [@2013fgu..book.....L for a review]. Recent observations suggest that the epoch of reionization (EoR) was already in significant progress by redshift $z \sim 8$ and must have ended by $z \sim 6$ . Upcoming observations will better constrain the reionization history, as well as the abundance and properties of the radiation sources. The reionization history is quantified by the evolution of the ionization fraction $x_\text{i}(z)$. It is used in the calculation of EoR observables such as the 21cm brightness temperature [e.g. @1997ApJ...475..429M; @2006PhR...433..181F], Thomson optical depth, cosmic microwave background (CMB) temperature and polarization anisotropies [e.g. @1987MNRAS.226..655B; @1997PhRvD..55.1822Z], and kinetic Sunyaev-Zel’dovich (KSZ) effect . Thus, it is important to establish a standard parametrization of the reionization history to facilitate robust comparisons between different observations and with theory. The ionization fraction is often parametrized using a $\tanh$ function with two free parameters which set the redshift midpoint and width [@2008PhRvD..78b3002L]. However, the width parameter does not clearly define the duration of the EoR and the simple functional form does not allow for possible asymmetry. Redshift-asymmetric parameterizations using polynomials, exponentials, and power-laws have recently been proposed . Generalized logistic functions [@doi:10.1093/jxb/10.2.290] can have asymmetric sigmoid shapes, but the physical interpretation of some of the free parameters for reionization is not straightforward. In this Letter, I present an accurate parametrization of the reionization history in terms of the redshift midpoint, duration, and asymmetry. Lagrange interpolating functions are used to construct analytical curves that exactly fit corresponding ionization points. The shape parametrizations are then compared against radiation-hydrodynamic simulations from the Simulations and Constructions of the Reionization of Cosmic Hydrogen (SCORCH) project [@2015ApJ...813...54T; @2017arXiv171204464D]. The adopted cosmological parameters are: $\Omega_{\rm m} = 0.3$, $\Omega_\Lambda = 0.7$, $\Omega_{\rm b} = 0.045$, $h = 0.7$, $\sigma_8 = 0.8$, $n_{\rm s} = 0.96$, $Y_\text{He} = 0.24$, and $T_\text{CMB} = 2.725$ K. Methods {#sec:method} ======= Midpoint, Duration, and Asymmetry --------------------------------- The reionization history is quantified by the ionized hydrogen fraction, which can be mass-weighted or volume-weighted. I will work with the mass-weighted version $x_\text{i,M}$ as the volume-averaged ionized hydrogen number density is given by $$\bar{n}_\text{HII,V} = x_\text{i,M} \bar{n}_\text{H,V} .$$ From here on, the mass-weighted and volume-averaged subscripts will be dropped to simplify the notation. Also, let $z_\text{x}$ denote the redshift corresponding to the ionization factor $\text{x}=100x_\text{i}$. In @2017arXiv171204464D, we choose the redshift midpoint as $z_{50}$ and present two practical choices for defining the duration $\Delta_\text{z}$ and asymmetry $A_\text{z}$ parameters. In the first case: $$\begin{aligned} \Delta_\text{z50} & \equiv z_{25} - z_{75} , \nonumber \\ A_\text{z50} & \equiv \frac{z_{25} - z_{50}}{z_{50} - z_{75}} ,\end{aligned}$$ the redshifts correspond to quartile ionization fractions ($x_\text{i} = 0.25, 0.50, 0.75$) and $\Delta_\text{z50}$ is analogous to a full width half max. In the second case: $$\begin{aligned} \Delta_\text{z90} & \equiv z_{05} - z_{95} , \nonumber \\ A_\text{z90} & \equiv \frac{z_{05} - z_{50}}{z_{50} - z_{95}} ,\end{aligned}$$ the redshifts correspond to early- and late-ionization fractions ($x_\text{i} = 0.05, 0.95$) and $\Delta_\text{z90}$ effectively quantifies the full extent of the EoR. While other definitions (e.g. $\Delta_\text{z68}, \Delta_\text{z95}$) can be adopted, extreme choices (e.g. $\Delta_\text{z99}$) are not recommended because the start and end of the EoR are difficult to determine precisely. Lagrange Interpolating Functions -------------------------------- Given the midpoint, duration, and asymmetry parameters, the relevant redshifts are uniquely specified and given by $$\begin{aligned} z_{25} & = z_{50} + \frac{\Delta_\text{z50} A_\text{z50}}{1 + A_\text{z50}} , \nonumber \\ z_{75} & = z_{25} - \Delta_\text{z50},\end{aligned}$$ or $$\begin{aligned} z_{05} & = z_{50} + \frac{\Delta_\text{z90} A_\text{z90}}{1 + A_\text{z90}} , \nonumber \\ z_{95} & = z_{05} - \Delta_\text{z90} .\end{aligned}$$ An analytical function that exactly passes through a given set of ionization points and therefore satisfies the chosen midpoint, duration, and asymmetry parameters can be constructed using the method of Lagrange interpolation. In practice, a straightforward interpolation of $x_\text{i}$ in terms of $z$ or $1+z$ can have oscillations, but the approach works better with a simple change of variables, $$\begin{aligned} u & = \ln(1+z) , \nonumber \\ v & = \ln x_\text{i} .\end{aligned}$$ For $N$ points, a polynomial $v(u)$ of degree $N-1$ can be constructed as $$\begin{aligned} v(u) & = \sum_{j=1}^N p_j(u) , \nonumber \\ p_j(u) & = v_j\prod_{\substack{k = 1 \\ k \ne j}}^N \frac{u - u_k}{u_j - u_k} .\end{aligned}$$ The ionization fraction $x_\text{i}(z) = \exp[v(u)]$ has the advantages of being continuous, differentiable, integrable, and invertible. At higher redshifts toward the start of the EoR, $x_\text{i}$ asymptotically goes to zero as required. At lower redshifts after the end of reionization, a physical maximum limit of unity should be imposed in practice. Radiation-Hydrodynamic Simulations ---------------------------------- [lCCCCC]{}\[t!\] Sim 0 & 7.95 & 1.87 & 4.68 & 1.63 & 2.90\ Sim 1 & 7.91 & 2.27 & 5.45 & 1.59 & 2.69\ Sim 2 & 7.83 & 2.89 & 6.54 & 1.49 & 2.33 To test the accuracy of the analytical parametrizations, I compare them against simulation results from the SCORCH project. In @2017arXiv171204464D, we present three reionization simulations with the same galaxy luminosity functions, but with different radiation escape fraction $f_{\rm esc}(z)$ models. The simulations are designed to have fixed Thomson optical depth $\tau \approx 0.06$, consistent with recent CMB observations . The simulations are run with the RadHydro code, which combines N-body and hydrodynamic algorithms [@2004NewA....9..443T] with an adaptive raytracing algorithm [@2007ApJ...671....1T] to directly and simultaneously solve collisionless dark matter dynamics, collisional gas dynamics, and radiative transfer of ionizing photons. Each RadHydro simulation has $2048^3$ dark matter particles, $2048^3$ gas cells, and up to 12 billion adaptive rays in a comoving box of side length $50\ h^{-1}$Mpc. Table \[tab:sims\] lists the midpoint, duration, and asymmetry parameters for the three SCORCH simulations. The index in the model name reflects the power-law slope in the evolution of the radiation escape fraction with $1+z$. Sim 0 has constant $f_{\rm esc}$ and reionization starts latest, but ends earliest out of the three models. Sim 1 has $f_{\rm esc}(z)$ varying linearly and is an intermediate model. Sim 2 has $f_{\rm esc}(z)$ varying quadratically and reionization starts earliest, but ends latest. Results ======= Ionization Fraction ------------------- ![[**Top:**]{} The evolution of the mass-weighted ionization fraction with redshift. The analytical parametrizations accurately capture the redshift-asymmetric form of the simulation curves. [**Bottom:**]{} The typical differences in ionization fractions are $|\Delta x_\text{i}| \lesssim 0.02$, while the maximum differences of $|\Delta x_\text{i}| \lesssim 0.03$ are found near the start and end of the EoR, which are uncertain in the simulations.[]{data-label="fig:ionization"}](fig1.pdf){width="\hsize"} Figure \[fig:ionization\] shows the evolution of the ionization fraction for the redshift range $5.5 < z < 20$. Only Sim 0 and Sim 2 are shown for clarity as Sim 1 gives intermediate results. The analytical parametrizations are excellent matches to the simulation results and the typical differences are only $|\Delta x_\text{i}| \lesssim 0.02$. The maximum differences of $|\Delta x_\text{i}| \lesssim 0.03$ are found near the start and end of the EoR, which are also not accurately captured in reionization simulations and semi-analytical models. The shape parametrizations using $\Delta_\text{z50}$ and $A_\text{z50}$ produce more accurate results near the midpoint, while those with $\Delta_\text{z90}$ and $A_\text{z90}$ produce smaller differences near the start and end of the EoR as expected. More accurate fits to simulation results can be obtained by combining both cases and using five rather than three ionization points. However, for parameter-space studies and constraining reionization histories from different observations, it is preferable to use a smaller number of free parameters to reduce degeneracies. 21cm Brightness Temperature --------------------------- The global 21cm brightness temperature [e.g. @1997ApJ...475..429M] in units of mK is given by $$\delta T_\text{b} \approx 28 x_\text{HI}\left(1-\frac{T_\gamma}{T_\text{s}}\right)\left(\frac{\Omega_\text{b}h^2}{0.022}\right)\left[\left(\frac{0.15}{\Omega_\text{m}h^2}\frac{1+z}{10}\right)\right]^{1/2} ,$$ where $x_\text{HI} = 1 - x_\text{i}$ is the neutral hydrogen fraction, $T_\gamma$ is the radiation temperature, and $T_\text{s}$ is the spin temperature. The standard approximation $T_\text{s} \gg T_\gamma$ is used, which is a valid assumption except in the early stages of reionization [e.g. @2008ApJ...689....1S]. ![[**Top:**]{} The global 21cm brightness temperature for the redshift range excluding the early stages of reionization. [**Bottom:**]{} The differences in brightness temperatures correspond to those in the ionization fractions, but have opposite signs. The typical differences are only $|\Delta T_\text{b}| \lesssim 0.7$ mK and are small compared to current observational sensitivities.[]{data-label="fig:21cm"}](fig2.pdf){width="\hsize"} Figure \[fig:21cm\] shows the evolution of the brightness temperature for the ranges $5.5 < z < 12$ and $x_\text{HI} \lesssim 0.8$. The analytical parametrizations are excellent matches to the simulation results as expected. The differences in the brightness temperatures correspond to those in the ionization fractions, but have opposite signs. The typical differences are only $|\Delta T_\text{b}| \lesssim 0.7$ mK and are small compared to current observational sensitivities. Global 21cm experiments such as EDGES [@2008ApJ...676....1B], SCI-HI [@2014ApJ...782L...9V], SARAS [@2017ApJ...845L..12S], and PRIZM (Philip et al. in prep) that observe up to a frequency of 200 MHz will probe the reionization of hydrogen. To model and interpret their signals, they can use this analytical parameterization and explore the parameter space to put constraints on the redshift midpoint and duration, and possibly weaker bounds on the asymmetry. Thomson Optical Depth --------------------- The Thomson optical depth integrated from redshift 0 to $z$ is given as $$\tau(z) = \sigma_\text{T}\int_0^z \bar{n}_\text{e}(z) \left|\frac{cdt}{dz}\right| dz , \label{eqn:tau}$$ where the volume-averaged free electron number density, $$\bar{n}_\text{e} = x_\text{HII}\bar{n}_\text{H} + x_\text{HeII}\bar{n}_\text{He} + 2x_\text{HeIII}\bar{n}_\text{He} \label{eqn:ne}$$ is related to the mean number densities ($\bar{n}_\text{H}, \bar{n}_\text{He}$) and mass-weighted ionization fractions ($x_\text{HII}, x_\text{HeII}, x_\text{HeIII}$) for hydrogen and helium. HI and HeI are jointly ionized during the EoR [e.g. @2007ApJ...671....1T]. While HeII reionization is also extended [e.g. @2009ApJ...694..842M; @2017ApJ...841...87L], the simple approximation of an instantaneous transition at $z \approx 3$ is sufficiently accurate for calculating the optical depth. Figure \[fig:tau\] shows that the analytical parametrizations accurately reproduce the integrated optical depth $\tau$ from the simulations with typical differences of $|\Delta \tau| \lesssim 0.001\ (\lesssim 2\%)$. The shape parametrizations using $\Delta_\text{z90}$ and $A_\text{z90}$ produce very small differences of $|\Delta \tau| \lesssim 2\times10^{-4}$ because the differences in the ionization fraction $\Delta x_\text{i}$ have both positive and negative values that average to nearly zero over the EoR redshift range. For integrated statistics that are linear in $x_\text{i}$, I recommend using the parameterizations $\Delta_\text{z90}$ and $A_\text{z90}$. ![[**Top:**]{} The Thomson optical depth integrated up to redshift $z$. The SCORCH sims are designed to have fixed Thomson optical depth $\tau \approx 0.06$, consistent with recent CMB observations. [**Bottom:**]{} The differences in optical depths are smaller for parametrizations using $\Delta_\text{z90}$ and $A_\text{z90}$, which more effectively quantify the full extent of the EoR.[]{data-label="fig:tau"}](fig3.pdf){width="\hsize"} Planck will soon provide an update on their current constraint of $\tau = 0.058 \pm 0.012$ from measurements of the CMB temperature and polarization angular power spectra . Since the location and amplitude of the reionization bump is precisely set by the redshift midpoint and optical depth, using an analytical parametrization that exactly matches a given $z_{50}$ and accurately produces a desired $\tau$ is highly advantageous. The current tanh function in CAMB [@2008PhRvD..78b3002L] can be replaced with this more accurate and flexible parametrization. Patchy KSZ Effect ----------------- The KSZ temperature distortion integrated along the direction ${\bf \hat{n}}$ is given by $$\frac{\Delta T}{T}({\bf \hat{n}}) = -\frac{\sigma_\text{T}}{c}\int n_\text{e}({\bf v \cdot \hat{n}})e^{-\tau} \left|\frac{cdt}{dz}\right| dz ,$$ where the electron number density $n_\text{e}$, peculiar velocity ${\bf v}$, and optical depth $\tau$ are all dependent on ${\bf \hat{n}}$ and $z$. In @2013ApJ...776...83B, we choose to integrate over the redshift range $5.5 \leq z \leq 20$ for the patchy KSZ component since some models can have late end to reionization, like in Sim 2 here. To quantify the impact of small differences in $x_\text{i}(z)$ on the patchy KSZ effect, I use a new and fast semi-numerical method of modeling reionization on large scales. In Holst et al. (in prep), we develop a novel approach that uses abundance matching to exactly satisfy a given $x_\text{i}(z)$. Density and velocity fields are constructed using 2nd-order Lagrangian perturbation theory [@1998MNRAS.299.1097S] with $2048^3$ particles in a periodic comoving box of side length $1\ h^{-1}$Gpc. Full-sky HEALPix [@2005ApJ...622..759G] maps with $N_\text{side} = 4096$ are then constructed by ray tracing through the simulated light cones. ![[**Top:**]{} The patchy KSZ angular power spectrum for temperature fluctuations integrated over the redshift range $5.5 < z < 20$. The overall amplitude increases for longer duration. [**Bottom:**]{} The differences in angular power have similar absolute values for both parametrizations, but there are larger differences for cases with larger $|\Delta x_\text{i}|$ and longer durations.[]{data-label="fig:ksz"}](fig4.pdf){width="\hsize"} Figure \[fig:ksz\] shows the angular power spectrum $D_\ell \equiv \ell(\ell + 1)C_\ell/(2\pi)$ in units of $\mu\text{K}^2$. The overall amplitude is expected to increase with both the redshift midpoint and duration [e.g. @2012ApJ...756...65Z; @2013ApJ...776...83B]. The shape parametrizations using $\Delta_\text{z50}$ and $A_\text{z50}$ overpredict, while the those using $\Delta_\text{z90}$ and $A_\text{z90}$ underpredict compared to the simulations. These trends correspond to the differences in the ionization fractions at higher redshifts $z > z_{50}$. There are also larger absolute differences for Sim 2 than Sim 0 because of the larger differences in $|\Delta x_\text{i}|$ and the longer duration. The maximum differences of $\Delta D_\ell \lesssim 0.1\ \mu\text{K}^2$ ($\lesssim 5\%$) are expected to be atypical since Sim 2 has a rather long duration and late end to reionization at $z \approx 5.5$. In upcoming work, I will explore the dependence of the patchy KSZ effect on the midpoint, duration, and asymmetry parameters. combined their $\tau$ constraints with South Pole Telescope measurements of the KSZ angular power at $\ell = 3000$ [@2015ApJ...799..177G], along with our KSZ theoretical models [@2013ApJ...776...83B] to infer a duration $\Delta_\text{zCMB} \equiv z_{10} - z_{99} < 2.9$ (95% confidence interval). In @2017arXiv171204464D, we find that the upper limit on the duration is in tension with our radiation-hydrodynamic simulations, all of which have longer durations. The current discrepancy most likely is due to assumptions made in the analyses, models, and simulations. Other contributing factors could be inconsistencies in parametrizing the reionization history and ambiguity in mass-weighted and volume-weighted ionization fractions. Conclusions =========== I present an accurate parametrization of the reionization history in terms of the redshift midpoint, duration, and asymmetry. Lagrange interpolating functions are used to construct analytical curves that exactly fit corresponding ionization points. I recommend using the shape parameters $\Delta_\text{z90}$ and $A_\text{z90}$ and caution against extreme choices (e.g. $\Delta_\text{z99}$) since the start and end of the EoR are difficult to determine precisely. More accurate fits to simulation results can be obtained by using more ionization points, but a smaller number of free parameters is preferable for fitting observations. This accurate and flexible approach will allow parameter-space studies and self-consistent constraints on the reionization history from 21cm, CMB, and high-redshift galaxies and quasars. I thank Nick Gnedin for the original motivation for this work. Thanks also go to Marcelo Alvarez, Nick Battaglia, Aristide Doussot, Ian Holst, and Sasha Kourov for helpful discussions. This work is supported by STScI grant HST-AR-15013.002-A. natexlab\#1[\#1]{}\[1\][[\#1](#1)]{} \[1\][doi: [](http://doi.org/#1)]{} \[1\][[](http://ascl.net/#1)]{} \[1\][[](https://arxiv.org/abs/#1)]{} , N., [Natarajan]{}, A., [Trac]{}, H., [Cen]{}, R., & [Loeb]{}, A. 2013, , 776, 83, , J. R., & [Efstathiou]{}, G. 1987, , 226, 655, , J. D., [Rogers]{}, A. E. E., & [Hewitt]{}, J. N. 2008, , 676, 1, , M., [Aghanim]{}, N., [Ili[ć]{}]{}, S., & [Langer]{}, M. 2015, , 580, L4, , A., [Trac]{}, H., & [Cen]{}, R. 2017, ArXiv e-prints. , S. R., [Oh]{}, S. P., & [Briggs]{}, F. H. 2006, , 433, 181, , E. M., [Reichardt]{}, C. L., [Aird]{}, K. A., [et al.]{} 2015, , 799, 177, , K. M., [Hivon]{}, E., [Banday]{}, A. J., [et al.]{} 2005, , 622, 759, , P., [Trac]{}, H., [Croft]{}, R., & [Cen]{}, R. 2017, , 841, 87, , A. 2008, , 78, 023002, , A., & [Furlanetto]{}, S. R. 2013, [The First Galaxies in the Universe]{} , P., [Meiksin]{}, A., & [Rees]{}, M. J. 1997, , 475, 429, , M., [Lidz]{}, A., [Zaldarriaga]{}, M., [et al.]{} 2009, , 694, 842, , J. P., & [Vishniac]{}, E. T. 1986, , 306, L51, , [Adam]{}, R., [Aghanim]{}, N., [et al.]{} 2016, , 596, A108, , [Aghanim]{}, N., [Ashdown]{}, M., [et al.]{} 2016, , 596, A107, , F. J. 1959, Journal of Experimental Botany, 10, 290, , M. G., [Amblard]{}, A., [Pritchard]{}, J., [et al.]{} 2008, , 689, 1, , R. 1998, , 299, 1097, , S., [Subrahmanyan]{}, R., [Udaya Shankar]{}, N., [et al.]{} 2017, , 845, L12, , R. A., & [Zeldovich]{}, Y. B. 1970, , 7, 3, , H., & [Cen]{}, R. 2007, , 671, 1, , H., [Cen]{}, R., & [Mansfield]{}, P. 2015, , 813, 54, , H., & [Pen]{}, U.-L. 2004, New Astronomy, 9, 443, , T. C., [Natarajan]{}, A., [J[á]{}uregui Garc[í]{}a]{}, J. M., [Peterson]{}, J. B., & [L[ó]{}pez-Cruz]{}, O. 2014, , 782, L9, , O., [Reichardt]{}, C. L., [Shaw]{}, L., [et al.]{} 2012, , 756, 65, , M. 1997, , 55, 1822,

ArXiv

--- abstract: 'Spectra of broad-lined Type Ic supernovae (SN Ic-BL), the only kind of SN observed at the locations of long-duration gamma-ray bursts (LGRBs), exhibit wide features indicative of high ejecta velocities ($\sim0.1c$). We study the host galaxies of a sample of 245 low-redshift ($z<0.2$) core-collapse SN, including 17 SN Ic-BL, discovered by galaxy-untargeted searches, and 15 optically luminous and dust-obscured $z<1.2$ LGRBs. We show that, in comparison with SDSS galaxies having similar stellar masses, the hosts of low-redshift SN Ic-BL and $z<1.2$ LGRBs have high stellar-mass and star-formation-rate densities. Core-collapse SN having typical ejecta velocities, in contrast, show no preference for such galaxies. Moreover, we find that the hosts of SN Ic-BL, unlike those of SN Ib/Ic and SN II, exhibit high gas velocity dispersions for their stellar masses. The patterns likely reflect variations among star-forming environments, and suggest that LGRBs can be used as probes of conditions in high-redshift galaxies. They may be caused by efficient formation of massive binary progenitors systems in densely star-forming regions, or, less probably, a higher fraction of stars created with the initial masses required for a SN Ic-BL or LGRB. Finally, we show that the preference of SN Ic-BL and LGRBs for galaxies with high stellar-mass and star-formation-rate densities cannot be attributed to a preference for low metal abundances but must reflect the influence of a separate environmental factor.' author: - 'Patrick L. Kelly' - 'Alexei V. Filippenko' - Maryam Modjaz - Daniel Kocevski title: 'The Host Galaxies of Fast-Ejecta Core-Collapse Supernovae' --- Introduction ============ In the cases of at least ten nearby (redshift $z\lesssim0.5$) LGRBs, observations have revealed a SN Ic-BL spectrum superimposed on the power-law continuum of the fading optical afterglow (@ga98 [@ma03; @st03; @hj03]; see @woosleybloom06 and @modjaz11rev for reviews). The spectra are characterized by wide features consistent with high ejecta velocities ($\sim$20,000–30,000kms$^{-1}$), and an absence of hydrogen and helium. The other principal classes of core-collapse SN, in contrast, exhibit spectroscopic features consistent with more slowly moving ejecta. The most common core-collapse SN are Type II ($\sim 60$%; @lileaman11; @smartteldridge09), which exhibit hydrogen (H) in their spectra; they are the final eruptions of stars that have retained their outer H shell. When the progenitor sheds, transfers to a companion, or internally mixes its outer H envelope during pre-SN evolution, the explosion will produce an H-deficient SN Ib, or an H- and He-deficient SN Ic [e.g., @fil97 and references therein]. While the absence of H and He in the spectra of SN Ic-BL indicates that their progenitors have lost their envelopes prior to core collapse, simulations additionally suggest that the progenitors of SN Ic-BL with associated LGRBs may also have rapid speeds of rotation. In models, only quickly rotating stars without an H envelope produce the outflowing jets that yield $\gamma$-ray emission [@hirschimeynet05; @yoonlanger05; @langernorman06] after core collapse to a black hole [@wo93; @macfadyenwoosley99]. Close massive binary systems that experience mass transfer, common-envelope evolution, or a merger [@podsialowskiivanova10; @langer12] are possible progenitors of LGRBs, because they can likely produce the required rapidly rotating massive stars without H or He envelopes. Recent observations of Galactic O-type stars show that, in fact, $\gtrsim 70$% of massive stars experience mass transfer with a companion and $\sim30$% undergo a merger [@sanademink12]. Alternative progenitor candidates include quickly rotating, metal-poor stars that internally mix their outer envelopes [@yoonlanger05; @woosleyheger06]. Single stars with high abundances, however, are considered improbable progenitors, because their comparatively strong winds [@vinkdekoter01] are expected to reduce their angular momentum. Spectroscopy of $z \lesssim 0.3$ host galaxies shows, in fact, that nearby SN Ic-BL with an associated GRB prefer more metal-poor environments than nearby Type Ic-BL SN having no obvious $\gamma$-ray emission [@mod08; @kocevski09; @grahamfruchter13]. The latter, in turn, prefer more metal-poor [@kelkir12; @sanderssoderberg12] and blue [@kelkir12] environments than do SN Ic without broad features. In several cases, however, the positions of low- and moderate-redshift LGRBs are spatially coincident or closely associated with massive, metal-rich galaxies (e.g., @lev10highmet [@perleylevan13; @elliottkruhler13]). Absorption spectroscopy also finds evidence for high-redshift $z \gtrsim 2$ systems along the line of sight to LGRBs with metallicities exceeding the $z \gtrsim 2$ cosmic average [@prochaska07]. @savagliorau12 has inferred a supersolar abundance from $z = 3.57$ absorption features consistent with a pair galaxies in the GRB 090323 afterglow spectrum. GRB 130702A [@singercenko13] occurred, however, in a metal-poor faint satellite of a $z=0.145$ massive galaxy [@kelfil13], raising the possibilty of superpositions or associations for some LGRBs that would be difficult to resolve at high redshift. Here we measure the host-galaxy properties of nearby core-collapse SN explosions using the imaging and photometry of the Sloan Digital Sky Survey (SDSS). For $z<1.2$ LGRB hosts, we estimate these host properties from published photometry and archival [*HST*]{} images. We show that SN Ic-BL and LGRBs exhibit a strong preference for galaxies that have high stellar-mass density and star-formation-rate density for their stellar mass. We also use SDSS spectra to show that the gas kinematics of SN Ic-BL hosts are exceptional. In §\[sec:data\], we describe the core-collapse SN and LGRB samples as well as the SDSS and [*HST*]{} galaxy data that we use in this analysis. Section \[sec:methods\] presents our techniques to analyze the galaxy imaging and spectroscopy, and the statistical methods we employ. In §\[sec:results\], we describe the results of our analysis, while §\[sec:discussion\] discusses the interpretation of the observed patterns. Our conclusions are presented in §\[sec:conclusions\]. Data {#sec:data} ==== We study the host galaxies of both nearby ($z<0.2$) core-collapse SN discovered by “galaxy-untargeted” transient searches (e.g., the Palomar Transient Factory; PTF; @raukulkarni09 [@lawkulkarni09]) which do not target specific potential hosts, and $z<1.2$ LGRBs detected by $\gamma$-ray satellites. We use the SDSS spectroscopic sample to build a control sample of low-redshift star-forming galaxies, and SDSS photometry and spectroscopy to measure properties of both the sample of low-redshift star-forming galaxies and the host galaxies of the nearby SN. For the host galaxies of $z < 1.2$ LGRBs, we estimate host properties using published photometry and [*HST*]{} imaging. SDSS DR10 Photometry and Spectroscopy ------------------------------------- The SDSS galaxy photometry and fiber spectra are from Data Release 10 [DR10; @sdssdrten13], and they were collected with the 2.5m telescope at Apache Point, New Mexico. The imaging survey, which spans 14,555 square degrees, consists of 53.9s integrations through the SDSS $ugriz$ filters, and the typical limiting $r$-band AB magnitude is 22.2. The typical sensitivity of available SDSS imaging makes possible detection of $z=0.1$ galaxies with absolute Vega magnitudes $M_B$ or $M_V$ brighter than about $-15.2$. Each Sloan $2048 \times 1498$ pixel CCD array records a $13.5' \times 9.9'$ field of view. The SDSS spectroscopic survey acquired approximately 45min of total integration in clear conditions, split into a series of three successive exposures. The spectrograph comprises $3''$ ($2''$ for BOSS) diameter fiber-optic cables placed at the positions of targets on the focal plane. Adjacent fibers can be no closer than $55''$ in a single fiber mask because of engineering constraints [@stra02], and the SDSS spectrographs record light with wavelengths 3800–9200Å. The targets selected for SDSS fiber spectroscopy consist of three primary “Legacy” categories of objects, and were also taken from several dozen ancillary programs, some of which were limited to specific parts of the survey (e.g., Stripe 82). Objects detected with $5\sigma$ significance in the imaging survey, with an extended light distribution and having $r$-band magnitude brighter than 17.77, as well as QSO candidates and luminous red galaxies (LRGs), formed the Legacy samples [@baldry05]. More limited special-program science targets included, for example, a $u$-band galaxy sample selected to investigate the properties of blue, faint galaxies. Table \[tab:selection\] shows the construction of our sample of SDSS galaxy spectra. Transient and Host-Galaxy Samples --------------------------------- ### Nearby SN Sample The core-collapse SN sample is constructed from $z < 0.2$ discoveries by SN surveys that do not target specific galaxies. The “galaxy-untargeted” SN search technique is akin to that of wide-field $\gamma$-ray satellite (e.g., [*Swift*]{}) searches for LGRBs. The SN in our sample consist of discoveries by galaxy-untargeted searches reported to the International Astronomical Union (IAU), as well as those published [@arcavi10] or reported via Astronomical Telegrams by the PTF, from 1990 January 1 through 2012 May 10. The SN searches that we considered to be galaxy-untargeted are identical those listed by @kelkir12. Our sample consists of only those nearby SN whose host galaxies have SDSS $ugriz$ imaging and, when appropriate, a fiber spectrum. The probability of detecting a SN in an image depends on the limiting magnitude of the transient search, as well as on the SN distance, luminosity, light-curve shape, and dust attenuation along the line of sight. Comparative analysis of the redshifts of core-collapse SN discovered by galaxy-untargeted searches found no significant evidence that variation among the principal species strongly affects their discovery rate [@kelkir12]. @kel08 showed that SN Ic are more closely associated with the highest surface brightness regions of their host galaxies than SN Ib, but here we combine these two SN types to assemble a larger sample of stripped-envelope SN that do not show high ejecta velocities. We also show, for comparison, the host of PTF12gzk, a peculiar SN Ic that did not exhibit broad spectroscopic features [@benamigalyam12] but in which radio observations found evidence for $\sim0.3c$ ejecta speeds [@horeshkulkarni13]. Lists of the nearby SN and the LGRBs are presented in Tables 2 and 3 (full tables are available in the electronic version). ### LGRB Sample Our LGRB sample includes both objects with a detected optical afterglow, and “dark” bursts without a luminous optical afterglow [@taylorfrail98] whose position was determined using their X-ray flux [@cenkokeleman09; @perleycenko09]. Analysis of the host galaxies of dark GRBs shows that they are both more massive [@perleylevan13] and dust-obscured [@djorgovskifrail01; @klosehenden03; @perleylevan13] than the hosts of GRBs with detected optical afterglows. The fraction of dark bursts in our sample () is approximately representative of the fraction of dust-obscured GRBs below the $z = 1.2$ redshift upper limit of our LGRB sample reported by @perleylevan13. We assemble $z<1.2$ LGRB host galaxies having archival [*HST*]{} images from the unobscured LGRBs assembled by @savaglio09 and dust-obscured LGRBs assembled by @perleylevan13, after rejecting several datasets that showed evidence of residual LGRB light. In many cases, [*HST*]{} images were acquired at least a year after the explosion; these are the data we use when available to measure galaxy sizes. When only images taken closer to the time of the GRB are available, we inspect the host galaxy to determine whether evidence for a point source at the explosion site exists. LGRBs exhibit a strong association with the brightest pixels of their hosts [@fru06], so a potential concern is that we could reject data where the LGRB coincided with a bright star-forming region. In practice, possible confusion was minimal for the data taken within $\sim4$ months after the GRB trigger. While we do not expect any significant contamination, here we are interested in the integrated properties of the host galaxy and not the specific region where the LGRB occurred. Even when an optical afterglow can be detected, identifying a coincident SN in the fading afterglow light curve, or by identifying a SN spectrum superimposed on the power-law afterglow continuum, requires high signal-to-noise ratio (S/N) data. Sufficient follow-up observations have only been possible for LGRBs with $z \lesssim 0.6$. Our GRB sample includes events with and without a detected optical afterglow or coincident SN. Since a representative fraction of the LGRBs in our sample have no optical counterpart, our findings should be robust to the effects of dust obscuration on LGRB detection. [lc]{} (1) Full Catalog & 948,205\ (2) Reliable Line Measurements & 910,532\ (3) Star Forming (Low or High S/N) & 377,763\ (4) H$\alpha$ S/N $>$ 20 &302,865\ (5) Fiber Offset $< 0.2\,R_{\rm P}$ & 289,474 \[tab:selection\] Methods {#sec:methods} ======= Measurements of Galaxy Properties --------------------------------- For both the $z<0.2$ core-collapse SN and $z<1.2$ LGRB samples, as well as the SDSS star-forming population, we estimate host-galaxy stellar masses $M$ and photometric star-formation rates (SFRs) by fitting PEGASE2 [@fi99] stellar population synthesis models to broadband photometry; see @kel10 for detailed information on the star-formation histories and initial-mass functions (IMFs) used. For nearby core-collapse SN host galaxies and the SDSS star-forming population, we fit Sloan $ugriz$ magnitudes. The multi-band photometry of the host galaxies of LGRBs was assembled from the GHostS database[^1]. We describe a second, complementary set of SFR estimates in the following section that is available only for the SDSS star-forming spectroscopic sample and which uses both the fiber spectrum and broadband $ugriz$ photometry. When comparing among samples, however, we only compare SFR values estimated using the same method. ### Analyses of SDSS Galaxy Spectra Several teams have performed and made available detailed measurements of SDSS galaxy properties from the Sloan photometry and spectroscopy. For the $z < 0.2$ samples of core-collapse SN host galaxies and the SDSS star-forming population, we use SFRs estimated from fitting both spectra and photometry, and gas velocity dispersions $\sigma_{\rm gas}$ measured from the H$\alpha$ emission-line profile. We use the star-forming classifications for SDSS galaxies and hybrid spectroscopic and photometric SFR estimates made available by a collaboration that was both at the Max Planck Institute for Astronomy (MPA) and Johns Hopkins University (JHU) (S. Charlot, G. Kauffmann, S. White, T. Heckman, C. Tremonti, and J. Brinchmann; MPA-JHU). SDSS fiber apertures generally do not cover the entire light distribution of each target galaxy. The MPA-JHU team therefore estimates the total SFR of each galaxy as the sum of the SFR within the fiber aperture determined from fitting the spectrum, and from a fit to the $ugriz$ light outside the fiber opening. We adopt the gas velocity dispersions estimated from SDSS emission-line profiles by the Portsmouth group [@thomassteele13], which apply the public Penalized PiXel Fitting [@cappellariemsellem04] (pPXF) and the Gas and Absorption Line Fitting [@sarzifalconbarroso06] (GANDALF v1.5) codes. The velocity dispersion of the gas is estimated from the widths of emission lines (e.g., H$\alpha$, \[[O$\;$[<span style="font-variant:small-caps;">[iii]{}</span>]{}]{}\]), and here we take the dispersion for H$\alpha$. ### Host-Galaxy Sizes The SDSS Photo pipeline performs separate fits of a de Vaucouleurs $r^{1/4}$ law and an exponential profile to the light distribution of each extended object. The pipeline next finds the linear combination of the two models (holding all parameters except flux fixed) that minimizes the $\chi^2$ statistic. To obtain an estimate of the half-light radius $r_{50}$, we compute the weighted average of the two components’ $r_{50}$ parameters and weight each by its fractional contribution to the total model flux. We use the GALFIT [@pe02] program to perform the same surface-brightness fitting analysis on archival [*HST*]{} images of LGRB host galaxies. We apply Source Extractor [SExtractor; @bert96] to drizzled and cosmic-ray-rejected images to estimate object positions, ellipticities, and magnitudes, and these are used as input GALFIT parameters. Extended sources except the LGRB hosts are modeled with Sersic profiles, and the instrument point-spread function (PSF) is used to model stars. The host-galaxy $r_{50}$ estimates show good agreement with existing estimates of $r_{50}$ from profile fitting [@conselicevreeswijk05], and with published $r_{80}$ SExtractor measurements [@svensson10]. The host galaxy of GRB 020903 is part of a complex association of interacting clumps, so the galaxy is not well approximated by a simple surface-brightness model. An additional complication is that much of the host-galaxy photometry was taken from the ground, where the galaxy components cannot be resolved. We therefore have excluded this host galaxy from the analysis. The angular diameter of ESO184-G82, the host galaxy of the nearby GRB 980425 ($z=0.0087$), approximately spans the available [*HST*]{} images. The physical resolution of the [*HST*]{} data is also substantially higher than that of the images of the SDSS SN host galaxies, or other LGRB hosts. Measurement of the host effective radius from model fitting requires significant imaging area without galaxy light to be able to fit robustly for the background level. We therefore do not include GRB 980425, which does not have SDSS exposures, in the LGRB sample. The host galaxy of GRB 051022 is also a system with two peaks that may possibly correspond to two strongly interacting galaxies, or else may instead be a galaxy with irregular morphology. We expect that the SDSS Photo pipeline would be most likely to model the host as a single system, so we construct a GALFIT model that consists of a single galaxy. ### Stellar-Mass and Star-Formation Densities Estimates We calculate the projected stellar-mass density $\Sigma_{M}$ and the projected star-formation density $\Sigma_{\rm SFR}$ from the galaxy stellar mass $M$ and star-formation rate SFR, respectively, and the parameters of the model $r$-band isophotal ellipse that encloses half of the galaxy light. We compute, for example, $\Sigma_{M} = {\rm log}_{10}(M \mathbin{/} 2 \mathbin{/} \pi A B)$, where $M$ is the stellar mass (in $M_{\odot}$), while $A$ and $B$ are the semimajor and semiminor axes (in kpc) of the isophotal ellipse that contains half of the galaxy $r$-band flux, determined from fitting its surface-brightness distribution. Comparison of Host-Galaxy Properties ------------------------------------ ### Median Relationship for Photometric Host Properties Photometric magnitudes measured through an appropriate aperture designed to enclose a specific percentage of the host light (e.g., a Petrosian aperture; [-@petrosian76]) can be expected to sample an approximately consistent fraction of galaxy light with increasing distance to sources. Physical properties estimated from broadband fluxes (e.g., stellar mass $M$) should therefore not exhibit strong aperture biases with increasing redshift in our sample, although the effects of surface-brightness dimming may become important at high redshifts. To study the properties of galaxies derived from photometric measurements, we therefore find the best-fitting $M$–$\Sigma_{M}$, $M$–$\Sigma_{\rm SFR}$, $M$–$r_{50}$, and SFR–$r_{50}$ relations for the complete $0.03 < z < 0.1$ SDSS star-forming catalog. We fit a second-degree polynomial to the median ordinate value (e.g., $\Sigma_{\rm SFR}$) across at least five bins in galaxy stellar mass. We include an additional point at (log $M$, $r_{50}$) = (0,0) and (log SFR, $r_{50}$) = ($-10$,0), respectively, when fitting for the SDSS $M$–SFR relation so that $r_{50}$ approaches zero for galaxies with negligible stellar mass $M$ or SFR. We use the Main SDSS spectroscopic sample Legacy “GALAXY” targets (i.e., where the 64 bit of the primTarget bitmask was set) to measure median relations of the properties of the low-redshift galaxy population. ### Spectroscopic Host Properties The fixed angular size of SDSS $3''$ fibers (or $2''$ for the Baryon Acoustic Oscillation Survey; BOSS), in contrast to (for example) a Petrosian aperture, samples a percentage of the light of each extended target that depends on the redshift and intrinsic size of the galaxy. To reach conclusions about spectroscopic measurements that are not sensitive to aperture effects, we apply a separate predictive algorithm that we have developed that yields robust comparisons. To predict the expected value of an observable (e.g., $\sigma_{\rm gas}$) given other host-galaxy properties (e.g., mass $M$), we perform, for each $z < 0.2$ SN host galaxy, a locally weighted multiple linear least-squares fit to the SDSS catalog. We model the observable of interest $O_j$ as $$O_j = \sum\limits_{i = 0}^{n} A_i \, x_{ij},$$ where $j$ indexes the SDSS galaxies, and each $x_{ij}$ is a galaxy property, a combination (e.g., multiplicative product) of galaxy properties, or unity. For example, we model the gas velocity dispersion $\sigma_{\rm gas}$ as $$\sigma_{{\rm gas},j} = A_0 + A_1{\rm log}_{10}M_j + A_2({\rm log}_{10}M_j)^2 + A_3\epsilon_j,$$ where $M$ is stellar mass and $\epsilon$ is galaxy ellipticity. Each row $j$ of the design matrix $A_i^j$ and $O^j$ is multiplied by the weight $W_j = e^{-u_j}$, where $u_j$ is $$u_j = \frac{(z_j - z^{\rm SN})^2}{(0.01)^2} + {\rm log}_{10}\left(\frac{M_j}{M^{\rm SN}}\right)^2 + \frac{(F_{{\rm aper},j} - F_{\rm aper}^{\rm SN})^2}{(0.1)^2}$$ and $z$ is galaxy redshift. To be able to predict a specific property $O_{\rm pred}$ of each host galaxy (e.g., $\sigma_{\rm gas}$) independent of its observed value, $O_{\rm obs}$, we exclude the host galaxy itself when fitting the predictive model to the properties of SDSS galaxies. This approach assigns greater weight to galaxies with similar observables, and mitigates any possible selection effect with galaxy redshift or fiber fraction. The best-fit parameters $A_i$ for each fit are used to compute a prediction for the value of the galaxy property, $O_{\rm pred}$, which can then be compared to the observed value, $O_{\rm obs}$, as a residual, $$\Delta O \equiv O_{\rm obs} - O_{\rm pred}. \label{eqn:residual}$$ Here the fitting analysis requires only modest numbers of nearby datapoints from the SDSS spectroscopic catalog. Results {#sec:results} ======= Host-Galaxy $\Sigma_{M}$ and $\Sigma_{\rm SFR}$ ----------------------------------------------- In Figures \[fig:massmassdensity\] and \[fig:masssfrdensity\], we plot galaxy stellar-mass density $\Sigma_{M}$ and star-formation density $\Sigma_{\rm SFR}$, estimated from broadband magnitudes, against stellar mass $M$. These show that $z<0.2$ SN Ic-BL and $z < 1.2$ LGRB host galaxies have high stellar-mass density and star-formation density for their stellar masses, compared with the low-redshift ($z<0.2$) SDSS star-forming galaxy population. In contrast, SN Ib/Ic (with slower ejecta speeds) and SN II show no preference for galaxies that have high stellar-mass density or high star-formation density for their stellar masses. We compute the residuals of each host galaxy from the SDSS $M$–$\Sigma_{M}$ and $M$–$\Sigma_{\rm SFR}$ relations. Our statistical method is to determine the probability that each pair of residual distributions is identical using the Kolmogorov-Smirnov (KS) two-sample test. We find significant evidence for differences between the SN Ic-BL ($n=$ ) and the SN II ($n=$ ;  and  for $\Sigma_{M}$ and $\Sigma_{\rm SFR}$, respectively) host distributions. Comparison between the SN Ic-BL and SN Ib/Ic ($n=$ ) host distributions likewise finds evidence for distinct underlying distributions ( and ). The residuals of LGRB ($n=$ ) host galaxies from the SDSS $M$–$\Sigma_{M}$ and $M$–$\Sigma_{\rm SFR}$ relations are more positive than (i.e., $\Sigma_{M}$ and $\Sigma_{\rm SFR}$ are greater than) those of SN Ib/Ic ( and , respectively) and SN II ( and ) hosts, while we find no statistically significant difference with the SN Ic-BL host residual distribution ( and ). The galaxy $M$–$\Sigma_{M}$ relation shows no significant change with increasing redshift to $z\approx1.1$ (e.g., @bardenrix05), so comparisons involving the LGRB host $M$–$\Sigma_{M}$ relation should not be strongly affected by evolution in the galaxy population with redshift. Preference for Overdense Galaxies --------------------------------- We study the $z<0.2$ SDSS star-forming population to investigate whether galaxies with relatively high stellar-mass and star-formation densities have comparatively low chemical abundances. In Figure \[fig:sdsspop\], we show the average @tre04 oxygen abundance of $z<0.2$ SDSS star-forming galaxies as a function of $M$ and $\Sigma_{M}$ as well as $\Sigma_{\rm SFR}$. The star-forming SDSS galaxies with the highest stellar-mass or star-formation-rate densities (in each stellar-mass bin) are not, on average, comparatively metal poor. The galaxies with the highest stellar-mass densities are, on average, more metal-rich by $\sim$0.2 dex. An important question is whether the high stellar-mass and star-formation-rate densities of SN Ic-BL and $z < 1.2$ LGRB host galaxies can be attributed entirely to the preference for low metal abundance observed among low-redshift SN Ic-BL [@kelkir12; @sanderssoderberg12] and LGRBs [@mod08; @grahamfruchter13]. We can represent such a preference as a progenitor formation efficiency $\eta$ per unit stellar mass created that depends solely on metallicity $Z$ and diminishes with increasing $Z$. If $\eta$ is a function of only metallicity $Z$, then the rate $r$ of fast-ejecta transients in a galaxy with a given SFR is $$r \propto \eta(Z) \times {\rm SFR}.$$ Half of the star formation in each galaxy mass $M$ bin plotted in Figure \[fig:sdsspop\] occurs in galaxies below the $M-\Sigma_{M}$ and $M-\Sigma_{\rm SFR}$ relations marked by thick blue lines. A transient population whose $\eta$ has no environmental dependence would be expected to explode equally in galaxies above and below these SFR-weighted relations. If $\eta$ instead increases at low abundance (and depends only on $Z$), we would expect greater numbers of SN Ic-BL and $z < 1.2$ LGRBs on the side of the relation, either above or below, that is comparatively metal poor. Inspection of Figure \[fig:sdsspop\] shows that, on average, higher stellar-mass density corresponds to higher metallicity, while metallicity does not vary strongly with star-formation density among galaxies in each bin in stellar mass. Consequently, if $\eta$ increases at low abundance (and depends only on $Z$), we would expect a plurality of SN Ic-BL and $z < 1.2$ LGRBs to be below the SFR-weighted $M-\Sigma_{M}$ relation, and approximately equal numbers above and below the SFR-weighted $M-\Sigma_{\rm SFR}$ relation. Since this prediction is very different from the pattern we observe, the production of SN Ic-BL and LGRB progenitors much be enhanced by an additional factor other than low metallicity that varies with $\Sigma_{M}$ as well as $\Sigma_{\rm SFR}$. Host-Galaxy Gaseous Velocity Dispersions ---------------------------------------- High gas dispersion velocities provide additional, independent evidence for distinct physical conditions in the host galaxies of low-redshift SN Ic-BL. In Figure \[fig:halphaveldisp\], the SN Ic-BL hosts describe an $M$–$\sigma_{\rm gas}$ relation with an offset to high gas velocity dispersion measured from the H$\alpha$ emission lines in the SDSS fiber spectra. The SN Ic-BL host distribution is significantly different from the SN Ib/Ic () and the SN II () host distributions. The comparatively high gas velocity dispersions of SN Ic-BL host galaxies may have one or more physical explanations. Rotationally supported galaxies having more compact and dense mass configurations are expected to have higher velocity dispersions. Alternatively, recent analysis of the gas kinematics of a sample of $z \approx 0.1$–0.3 compact, highly star-forming galaxies finds complex profiles for the strong emission lines, with several narrow ($\sigma_{\rm gas} \approx 10$–120kms$^{-1}$) and broad ($\sigma_{\rm gas} \approx 100$–250kms$^{-1}$) components [@amorinvilchez12] that may be due to strong stellar winds, or emission from expanding SN remnants. In Figures \[fig:icblmosaic\] and \[fig:othermosaic\], we show images of core-collapse SN host galaxies and the measured H$\alpha$ velocity dispersions juxtaposed with control samples of SDSS star-forming galaxies having similar parameters. These demonstrate the high velocity dispersions of SN Ic-BL host galaxies. Host-Galaxy Sizes ----------------- In Figures \[fig:masshl\] and \[fig:sfrhl\], we plot galaxy half-light radius $r_{50}$ (in kpc) against $M$ and SFR estimated from their broadband magnitudes. These show that $z < 0.2$ SN Ic-BL and $z < 1.2$ LGRB host galaxies are compact for their stellar masses and SFRs, compared with the low-redshift ($z < 0.2$) SDSS star-forming galaxy population. In contrast, SN Ib/Ic (with slower ejecta speeds) and SN II show no preference for galaxies that are relatively compact in size. Hybrid $\Sigma_{\rm SFR}$ Estimates ----------------------------------- In Figure \[fig:sfrspecdensity\], we plot $\Sigma_{\rm SFR}$ against $M$ for the sample of nearby ($z < 0.2$) core-collapse SN galaxies with SDSS spectra. Here we calculate $\Sigma_{\rm SFR}$ using the SFR estimated by the MPA-JHU group from the fiber spectrum and by modeling the galaxy [*ugriz*]{} light outside of the 3$''$ fiber. These SFR measurements may be expected to be more accurate and precise than the SFR estimates that are based on galaxy broadband photometry alone, and provide evidence that the preference among SN Ic-BL and LGRBs for high $\Sigma_{\rm SFR}$ seen in Figure \[fig:masssfrdensity\] galaxies is robust. The SN Ic-BL ($n=$ ) and SN II ($n=$ ) distributions show significantly different ($p=$ ) residual distributions from their predicted $\Sigma_{\rm SFR}$ values. $M$–SFR Relation ---------------- Figure \[fig:msfr\] shows the relationship between $M$ and SFR for the core-collapse host galaxies and the SDSS star-forming population. This shows that host galaxies of SN Ic-BL are not substantially more strongly star forming for their stellar masses than other core-collapse hosts, and suggests that relatively compact host sizes may primarily account for their high star-formation-rate densities. $M$–$Z$ Relation ---------------- Given the existing evidence that $z \lesssim 0.3$ SN Ic-BL [@kelkir12; @sanderssoderberg12] and LGRBs [@mod08; @grahamfruchter13] prefer metal-poor environments, a reasonable question is whether the high $\Sigma_{M}$ and $\Sigma_{\rm SFR}$ host galaxies in our sample having SDSS spectra describe a metal-poor mass-metallicity ($M$–$Z$) relation. Kelly et al. (2014, in prep.) report no significant evidence that the SN Ic-BL offset distribution ($n=$ ) differs from the SN Ib/Ic (; $n=$ ) or the SN II (; $n=$ ) distributions. Discussion {#sec:discussion} ========== We have found that low-redshift SN Ic-BL and $z<1.2$ LGRB host galaxies show stellar-mass and star-formation densities high compared with those of low-redshift galaxies having similar stellar masses. Core-collapse SN with more slowly expanding ejecta, however, exhibit no preference for galaxies having overdense stellar-mass distributions and star formation. SN Ic-BL host galaxies exhibit high gas velocity dispersions for their stellar masses, providing evidence for exceptional conditions in their hosts. Given the effect of stellar metallicity to strengthen line-driven winds and remove the angular momentum of massive stars, studies of SN and LGRB host galaxies have often attempted to explain environmental patterns in terms of chemical abundance. However, across the stellar-mass range populated by SN Ic-BL and LGRB hosts, SDSS star-forming galaxies with high stellar-mass and star-formation-rate densities are not, on average, more metal poor than less dense galaxies having similar stellar masses (see Figure \[fig:sdsspop\]). Therefore, a preference other than for low metallicity environments must be responsible for the overrepresentation of SN and LGRBs in galaxies with high stellar-mass and star-formation-rate densities. Furthermore, the $M$–$Z$ relation of SN Ic-BL host galaxies, which have high stellar-mass and star-formation densities, is not significantly metal poor compared to the SDSS $M$–$Z$ relation (Kelly et al. 2014, in prep.). In fact, a preference of high-velocity explosions for rapidly star-forming or overdense regions may help explain the observed association of LGRBs with the brightest regions of their host galaxies [@fru06]. An initial suggestion was that a strong association with the brightest regions would be expected for LGRBs if they occur preferentially in low-mass, metal-poor galaxies. If the most massive stars form in OB associations, then these will be more likely to correspond to the peaks of the light distribution of low-mass, metal-poor galaxies. OB associations in more massive, metal-rich spirals may instead be outranked in brightness by, for example, the nucleus. While such a low-metallicity preference may contribute, we suggest that an affinity to high star-formation density is important. A plausible explanation for the association of LGRBs with regions of higher star-formation density is the formation efficiency of young, bound star clusters. Observations of extragalactic star clusters have found evidence that bound-cluster formation efficiency increases with the star-formation density [@goddard10; @silvavillaadamo13]. Binary systems may be created more frequently in bound clusters, and those that form are expected to become progressively tighter through dynamical interactions with other members of the cluster [@heggie75; @hutmcmillan92]. Therefore, massive stars that form in dense star clusters are more likely to be in tight binary systems. Interacting massive binaries are candidate progenitor systems for SN Ic-BL and LGRBs, because mass transfer or common-envelope evolution leading (in some cases) to a merger can yield a rapidly rotating star whose outer envelope is not composed of H. If SN Ic-BL and LGRBs explode from stars that are more massive than the progenitors of SN Ib/Ic and SN II, an IMF that becomes top-heavy in dense, highly star-forming regions provides an alternative explanation for the patterns we observe. Indirect evidence from the absorption features of $\lesssim0.3\,{\rm M}_{\odot}$ stars suggests instead that the IMF may be bottom-heavy in elliptical galaxies with high stellar velocity dispersions and \[Mg/Fe\], which are thought to have had high star formation densities during their formation epoch [@vandokkumconroy10; @conroyvandokkum12]. This suggests that the IMF may be bottom heavy in dense, highly star-forming regions, while the opposite trend would be required to explain our observations. An improved census of massive stars across low-redshift environments may help to address more directly whether the patterns we have found can be explained by variation in the upper end of the IMF. The median redshift ($z_{\rm m}=\lgrbmedian$) of the $z < 1.2$ LGRB sample is substantially higher than those of the core-collapse SN sample ($z_{\rm m} \approx $ ) and the SDSS sample of galaxy spectra ($z_{\rm m} \approx 0.08$). The and relations among star-forming galaxies may evolve toward lower $\Sigma_{M}$ and $\Sigma_{\rm SFR}$ densities from $z \approx \lgrbmedian$ to the low-redshift universe. As we have shown, however, the low-redshift SN Ic-BL host galaxy population exhibits the same preference for high stellar-mass and star-formation density galaxies as LGRBs. Additionally, if stellar-mass and star-formation-rate density are proxies for the star-forming conditions that promote the formation of SN Ic-BL progenitors, then we expect that these same physical conditions exist in high-redshift LGRB host galaxies having similar positions in the and planes. Given the evidence that the progenitors of high-velocity ejecta core-collapse explosions form more efficiently in galaxies that have high stellar-mass and star-formation densities, a reasonable expectation is that LGRBs will be improved tracers of star formation at earlier epochs where galaxies have higher densities of stellar mass and ongoing star formation than in the low-redshift universe [@trujilloforster06]. LGRBs may be the most powerful probes of star formation at the highest redshifts. The extreme luminosities of LGRBs allow detection at the earliest epochs of star formation ($z \gtrsim 8$), where optical and infrared instruments are not yet able to detect sub-$L_{*}$ galaxies. The evidence we have found that the formation of LGRB progenitors depends sensitively on star-forming conditions suggests that they will be critical tools for investigating early star formation in detail. Conclusions {#sec:conclusions} =========== We have studied the host galaxies of a sample of 245 low-redshift core-collapse SN, including 17 SN Ic-BL discovered by galaxy-untargeted searches and 15 optically luminous and dust-obscured $z<1.2$ LGRBs. We have used the uniform fiber spectra and photometry of the SDSS to measure the properties of the core-collapse SN host-galaxy sample, as well as characterize the low-redshift star-forming galaxy population. Published multi-band photometry as well as [*HST*]{} imaging were used to study the $z < 1.2$ LGRB host galaxies. The outflowing ejecta of SN Ic-BL, from the wide features of their spectra, and of LGRBs, through their association with SN Ic-BL and $\gamma$-ray emission, are inferred to expand with high velocities ($\sim$20,000–30,000kms$^{-1}$). We have found that these core-collapse explosions in which a significant fraction of the ejecta is moving at high velocity prefer galaxies with high stellar-mass and star-formation densities, when compared to galaxies having similar stellar masses. The core-collapse SN in our sample having typical velocities of their ejecta, in contrast, are found approximately equally in galaxies with comparatively low and high stellar-mass and star-formation densities. From the widths of H$\alpha$ emission lines, we find that the hosts of SN Ic-BL have exceptionally high gas velocity dispersions when compared with star-forming galaxies having similar stellar masses. While early LGRB host environment analyses helped to establish the connection between LGRBs and massive stars (e.g., @bloom02), subsequent analysis has focused on the effect of metal abundance on the creation of their progenitors (e.g., @mod08), or sought to explain environmental patterns in terms of a preference for metal-poor galaxies (e.g., @fru06; @svensson10; @grahamfruchter13). Here we have shown that the preference of low-redshift SN Ic-BL and $z < 1.2$ LGRBs for galaxies with high stellar-mass and star-formation-rate densities, compared to star-forming galaxies of similar stellar mass, is separate from a preference for low metal abundances. Observations of extragalactic star formation have suggested that dense stellar clusters may form with greater efficiency in regions of dense star formation [@goddard10; @silvavillaadamo13]. A prospective explanation for the host-galaxy patterns we find is the efficient formation of tight massive binary progenitor systems in such densely star-forming environments. Alternatively, if SN Ic-BL and LGRB progenitors have greater stellar masses than those of core-collapse SN with typical ejecta velocities, a top-heavy IMF in galaxies with high stellar-mass density and star-formation density could account for these same patterns. However, such a top-heavy IMF would need to be reconciled with evidence instead for a bottom-heavy IMF in massive galaxies that were densely star forming while assembling their mass [@vandokkumconroy10; @conroyvandokkum12]. The construction of ground-based telescopes with large aperture (e.g., the Thirty Meter Telescope; Giant Magellan Telescope) and the James Webb Space Telescope will make possible observations of galaxies formed shortly after reionization, but the faintest galaxies will remain beyond their sensitivity. LGRBs, whose $\gamma$-ray emission is currently visible to $z \approx 8$ [@tanvir09], may provide an effective approach to probing the detailed star-forming conditions in these early galaxies. We thank Sandra Savaglio who, as referee, provided insightful comments and suggestions. We additionally acknowledge useful discussions about measurements and host galaxies with Jarle Brinchmann, Ori Fox, David Elbaz, John Graham, Matt Lehnert, Steven Stahler, Paul Crowther, and Josh Bloom. A.V.F.’s group at UC Berkeley has received generous financial assistance from the Christopher R. Redlich Fund, the TABASGO Foundation, Weldon Wood, and NSF grant AST-1211916, as well as from NASA/[*HST*]{} grant AR-12850 from the Space Telescope Science Institute, which is operated by the Association of Universities for Research in Astronomy (AURA), Inc., under NASA contract NAS5-26555. This research has made use of the GHostS database (www.grbhosts.org), which is partly funded by NASA/[*Spitzer*]{} RSA Agreement No. 1287913. natexlab\#1[\#1]{} , C. P., [Alexandroff]{}, R., [Allende Prieto]{}, C., [et al.]{} 2013, ArXiv e-prints , R., [V[í]{}lchez]{}, J. M., [H[ä]{}gele]{}, G. F., [et al.]{} 2012, , 754, L22 , I., [Gal-Yam]{}, A., [Kasliwal]{}, M. M., [et al.]{} 2010, , 721, 777 , I. K., [Glazebrook]{}, K., [Budav[á]{}ri]{}, T., [et al.]{} 2005, , 358, 441 , J. A., [Phillips]{}, M. M., & [Terlevich]{}, R. 1981, , 93, 5 , M., [Rix]{}, H.-W., [Somerville]{}, R. S., [et al.]{} 2005, , 635, 959 , S., [Gal-Yam]{}, A., [Filippenko]{}, A. V., [et al.]{} 2012, , 760, L33 , E., & [Arnouts]{}, S. 1996, AJ, 117, 393 , J. S., [Kulkarni]{}, S. R., & [Djorgovski]{}, S. G. 2002, , 123, 1111 , M., & [Emsellem]{}, E. 2004, , 116, 138 , S. B., [Kelemen]{}, J., [Harrison]{}, F. A., [et al.]{} 2009, , 693, 1484 , C., & [van Dokkum]{}, P. G. 2012, , 760, 71 , C. J., [Vreeswijk]{}, P. M., [Fruchter]{}, A. S., [et al.]{} 2005, , 633, 29 , S. G., [Frail]{}, D. A., [Kulkarni]{}, S. R., [et al.]{} 2001, , 562, 654 , J., [Kr[ü]{}hler]{}, T., [Greiner]{}, J., [et al.]{} 2013, , 556, A23 , A. V. 1997, , 35, 309 , M., & [Rocca-Volmerange]{}, B. 1999, ArXiv e-prints , A. S., [Levan]{}, A. J., [Strolger]{}, L., [et al.]{} 2006, , 441, 463 , T. J., [Vreeswijk]{}, P. M., [van Paradijs]{}, J., [et al.]{} 1998, Nature, 395, 670 , Q. E., [Bastian]{}, N., & [Kennicutt]{}, R. C. 2010, , 405, 857 , J. F., & [Fruchter]{}, A. S. 2013, , 774, 119 , D. C. 1975, , 173, 729 , R., [Meynet]{}, G., & [Maeder]{}, A. 2005, , 443, 581 , J., [Sollerman]{}, J., [M[ø]{}ller]{}, P., [et al.]{} 2003, Nature, 423, 847 , A., [Kulkarni]{}, S. R., [Corsi]{}, A., [et al.]{} 2013, ArXiv e-prints, arXiv:1306.5755 , P., [McMillan]{}, S., [Goodman]{}, J., [et al.]{} 1992, , 104, 981 , P. L., [Filippenko]{}, A. V., [Fox]{}, O. D., [Zheng]{}, W., & [Clubb]{}, K. I. 2013, , 775, L5 , P. L., [Hicken]{}, M., [Burke]{}, D. L., [Mandel]{}, K. S., & [Kirshner]{}, R. P. 2010, , 715, 743 , P. L., & [Kirshner]{}, R. P. 2012, , 759, 107 , P. L., [Kirshner]{}, R. P., & [Pahre]{}, M. 2008, , 687, 1201 , S., [Henden]{}, A. A., [Greiner]{}, J., [et al.]{} 2003, , 592, 1025 , D., [West]{}, A. A., & [Modjaz]{}, M. 2009, , 702, 377 , N. 2012, , 50, 107 , N., & [Norman]{}, C. A. 2006, , 638, L63 , N. M., [Kulkarni]{}, S. R., [Dekany]{}, R. G., [et al.]{} 2009, , 121, 1395 , E. M., [Kewley]{}, L. J., [Graham]{}, J. F., & [Fruchter]{}, A. S. 2010, , 712, L26 , W., [Leaman]{}, J., [Chornock]{}, R., [et al.]{} 2011, , 412, 1441 , A. I., & [Woosley]{}, S. E. 1999, , 524, 262 , T., [Garnavich]{}, P. M., [Stanek]{}, K. Z., [et al.]{} 2003, ApJ, 599, 394 , M. 2011, Astronomische Nachrichten, 332, 434 , M., [Kewley]{}, L., [Kirshner]{}, R. P., [et al.]{} 2008, AJ, 135, 1136 , C. Y., [Ho]{}, L. C., [Impey]{}, C. D., & [Rix]{}, H.-W. 2002, , 124, 266 , D. A., [Cenko]{}, S. B., [Bloom]{}, J. S., [et al.]{} 2009, , 138, 1690 , D. A., [Levan]{}, A. J., [Tanvir]{}, N. R., [et al.]{} 2013, ArXiv e-prints, arXiv:1301.5903 , V. 1976, , 209, L1 , P., [Ivanova]{}, N., [Justham]{}, S., & [Rappaport]{}, S. 2010, , 406, 840 , J. X., [Chen]{}, H.-W., [Dessauges-Zavadsky]{}, M., & [Bloom]{}, J. S. 2007, , 666, 267 , A., [Kulkarni]{}, S. R., [Law]{}, N. M., [et al.]{} 2009, , 121, 1334 , H., [de Mink]{}, S. E., [de Koter]{}, A., [et al.]{} 2012, Science, 337, 444 , N. E., [Soderberg]{}, A. M., [Levesque]{}, E. M., [et al.]{} 2012, , 758, 132 , M., [Falc[ó]{}n-Barroso]{}, J., [Davies]{}, R. L., [et al.]{} 2006, , 366, 1151 , S., [Glazebrook]{}, K., & [Le Borgne]{}, D. 2009, , 691, 182 , S., [Rau]{}, A., [Greiner]{}, J., [et al.]{} 2012, , 420, 627 , E., [Adamo]{}, A., & [Bastian]{}, N. 2013, ArXiv e-prints , L. P., [Cenko]{}, S. B., [Kasliwal]{}, M. M., [et al.]{} 2013, , 776, L34 , S. J., [Eldridge]{}, J. J., [Crockett]{}, R. M., & [Maund]{}, J. R. 2009, , 395, 1409 , K. Z., [Matheson]{}, T., [Garnavich]{}, P. M., [et al.]{} 2003, ApJ, 591, L17 , M. A., [Weinberg]{}, D. H., [Lupton]{}, R. H., [et al.]{} 2002, , 124, 1810 , K. M., [Levan]{}, A. J., [Tanvir]{}, N. R., [Fruchter]{}, A. S., & [Strolger]{}, L.-G. 2010, , 405, 57 , N. R., [Fox]{}, D. B., [Levan]{}, A. J., [et al.]{} 2009, , 461, 1254 , G. B., [Frail]{}, D. A., [Kulkarni]{}, S. R., [et al.]{} 1998, , 502, L115 , D., [Steele]{}, O., [Maraston]{}, C., [et al.]{} 2013, , 431, 1383 , C. A., [Heckman]{}, T. M., [Kauffmann]{}, G., [et al.]{} 2004, , 613, 898 , I., [F[ö]{}rster Schreiber]{}, N. M., [Rudnick]{}, G., [et al.]{} 2006, , 650, 18 , P. G., & [Conroy]{}, C. 2010, , 468, 940 , J. S., [de Koter]{}, A., & [Lamers]{}, H. J. G. L. M. 2001, , 369, 574 , S. E., & [Bloom]{}, J. S. 2006, , 44, 507 , S. E., & [Heger]{}, A. 2006, , 637, 914 , S. E., [Langer]{}, N., & [Weaver]{}, T. A. 1993, ApJ, 411, 823 , S., & [Langer]{}, N. 2005, , 443, 643 ------------ --------- ------- ------------------- ----------------------------- ------------------------------ ---------------------------------------- ---------------- Name [*z*]{} Type Mass SFR (phot) $\Sigma_{M}$ $\Sigma_{\rm SFR}$ $r_{50}$ (log M$_{\odot}$) (log M$_{\odot}$ yr$^{-1}$) (log M$_{\odot}$ kpc$^{-2}$) (log M$_{\odot}$ yr$^{-1}$ kpc$^{-2}$) (kpc) GRB010921 0.451 GRB 9.69$\pm$0.09 0.35$\pm$0.19 8.53$\pm$0.09 -0.81$\pm$0.19 2.05$\pm$0.02 GRB011121 0.362 GRB 9.66$\pm$0.17 0.32$\pm$0.18 6.86$\pm$0.17 -2.48$\pm$0.18 11.04$\pm$0.64 GRB021211 1.006 GRB 8.62$\pm$0.54 -0.59$\pm$0.20 7.94$\pm$0.54 -1.26$\pm$0.20 0.93$\pm$0.02 GRB030329 0.168 GRB 8.03$\pm$0.12 -0.91$\pm$0.20 7.99$\pm$0.12 -0.94$\pm$0.20 0.67$\pm$0.03 GRB050416A 0.653 GRB 9.19$\pm$0.19 0.18$\pm$0.16 7.99$\pm$0.19 -1.02$\pm$0.16 2.17$\pm$0.12 GRB051022 0.807 GRB 9.72$\pm$0.05 1.25$\pm$0.09 8.01$\pm$0.05 -0.45$\pm$0.09 4.88$\pm$0.44 GRB060218 0.034 GRB 7.72$\pm$0.19 -1.11$\pm$0.24 7.96$\pm$0.19 -0.87$\pm$0.24 0.31$\pm$0.00 GRB090417B 0.345 GRB 10.14$\pm$0.14 -0.27$\pm$0.29 8.41$\pm$0.14 -2.00$\pm$0.29 3.45$\pm$0.06 GRB970508 0.835 GRB 8.62$\pm$0.24 -0.26$\pm$0.17 8.46$\pm$0.24 -0.42$\pm$0.17 0.68$\pm$0.01 GRB970828 0.960 GRB 9.62$\pm$0.52 0.23$\pm$0.25 8.65$\pm$0.52 -0.73$\pm$0.25 1.66$\pm$0.80 GRB980613 1.097 GRB 9.00$\pm$0.34 0.68$\pm$0.18 8.42$\pm$0.34 0.10$\pm$0.18 1.03$\pm$0.19 GRB980703 0.966 GRB 10.00$\pm$0.15 0.98$\pm$0.16 9.06$\pm$0.15 0.03$\pm$0.16 1.37$\pm$0.04 GRB990712 0.433 GRB 9.37$\pm$0.05 -0.22$\pm$0.09 8.37$\pm$0.05 -1.21$\pm$0.09 2.25$\pm$0.08 GRB991208 0.706 GRB 8.91$\pm$0.25 -0.12$\pm$0.17 9.09$\pm$0.25 0.06$\pm$0.17 0.33$\pm$0.02 PTF 09awk 0.062 Ib 9.53$\pm$0.07 0.16$\pm$0.12 8.56$\pm$0.07 -0.81$\pm$0.12 1.61$\pm$0.05 PTF 09axc 0.115 II 9.99$\pm$0.06 -0.19$\pm$0.18 8.87$\pm$0.06 -1.32$\pm$0.18 1.51$\pm$0.07 PTF 09axi 0.064 II 9.33$\pm$0.09 -0.28$\pm$0.15 7.42$\pm$0.09 -2.18$\pm$0.15 4.10$\pm$0.15 PTF 09bce 0.023 II 10.95$\pm$0.03 0.70$\pm$0.07 9.17$\pm$0.03 -1.08$\pm$0.07 4.46$\pm$0.03 PTF 09bw 0.150 II 9.55$\pm$0.17 -0.26$\pm$0.36 7.90$\pm$0.17 -1.90$\pm$0.36 2.99$\pm$0.40 PTF 09cjq 0.019 II 10.40$\pm$0.03 0.30$\pm$0.07 8.10$\pm$0.03 -2.00$\pm$0.07 7.32$\pm$0.03 PTF 09ct 0.150 II 9.95$\pm$0.14 0.17$\pm$0.26 8.46$\pm$0.14 -1.33$\pm$0.26 2.76$\pm$0.38 PTF 09cu 0.057 II 10.47$\pm$0.04 0.68$\pm$0.08 8.19$\pm$0.04 -1.60$\pm$0.08 6.63$\pm$0.06 PTF 09dfk 0.016 Ib 8.96$\pm$0.31 -0.82$\pm$0.34 8.49$\pm$0.31 -1.29$\pm$0.34 0.76$\pm$0.02 PTF 09djl 0.184 II 10.09$\pm$0.15 -0.04$\pm$0.36 9.47$\pm$0.15 -0.65$\pm$0.36 1.03$\pm$0.21 PTF 09dra 0.077 II 10.57$\pm$0.07 0.81$\pm$0.08 7.80$\pm$0.07 -1.96$\pm$0.08 15.03$\pm$0.22 PTF 09due 0.029 II 10.25$\pm$0.03 0.70$\pm$0.07 7.93$\pm$0.03 -1.62$\pm$0.07 7.30$\pm$0.03 PTF 09dzt 0.087 Ic 9.98$\pm$0.06 0.44$\pm$0.16 7.18$\pm$0.06 -2.37$\pm$0.16 12.35$\pm$0.46 PTF 09ebq 0.024 II 9.75$\pm$0.04 0.02$\pm$0.08 8.51$\pm$0.04 -1.22$\pm$0.08 2.04$\pm$0.02 PTF 09fbf 0.021 II 10.05$\pm$0.03 0.60$\pm$0.07 7.97$\pm$0.03 -1.48$\pm$0.07 8.30$\pm$0.04 PTF 09fmk 0.063 II 5.24$\pm$0.14 PTF 09foy 0.060 II 10.20$\pm$0.09 0.52$\pm$0.13 7.67$\pm$0.09 -2.01$\pm$0.13 7.46$\pm$0.09 PTF 09g 0.040 II 9.55$\pm$0.04 0.18$\pm$0.08 7.84$\pm$0.04 -1.53$\pm$0.08 2.98$\pm$0.02 PTF 09gof 0.103 II 10.06$\pm$0.06 0.60$\pm$0.20 7.56$\pm$0.06 -1.91$\pm$0.20 8.63$\pm$0.24 PTF 09hdo 0.047 II 10.75$\pm$0.03 0.70$\pm$0.07 8.61$\pm$0.03 -1.44$\pm$0.07 5.63$\pm$0.03 PTF 09hzg 0.028 II 10.55$\pm$0.03 -0.40$\pm$0.07 8.38$\pm$0.03 -2.57$\pm$0.07 9.06$\pm$0.05 PTF 09iex 0.020 II 8.36$\pm$0.47 -1.10$\pm$0.50 6.98$\pm$0.47 -2.48$\pm$0.50 2.42$\pm$0.14 PTF 09ige 0.064 II 9.76$\pm$0.07 0.32$\pm$0.20 7.61$\pm$0.07 -1.83$\pm$0.20 5.51$\pm$0.08 PTF 09igz 0.086 II 9.40$\pm$0.08 -0.16$\pm$0.15 7.72$\pm$0.08 -1.84$\pm$0.15 2.89$\pm$0.17 PTF 09ism 0.029 II 9.56$\pm$0.19 -0.16$\pm$0.19 7.83$\pm$0.19 -1.89$\pm$0.19 3.34$\pm$0.07 PTF 09ps 0.106 Ic 9.29$\pm$0.08 -0.18$\pm$0.15 8.20$\pm$0.08 -1.27$\pm$0.15 2.80$\pm$0.14 PTF 09q 0.090 Ic 10.83$\pm$0.09 0.86$\pm$0.12 8.23$\pm$0.09 -1.74$\pm$0.12 8.95$\pm$0.11 PTF 09r 0.027 II 9.01$\pm$0.23 -1.16$\pm$0.46 8.45$\pm$0.23 -1.72$\pm$0.46 1.49$\pm$0.02 PTF 09sh 0.038 II 9.95$\pm$0.03 0.40$\pm$0.07 7.97$\pm$0.03 -1.58$\pm$0.07 4.76$\pm$0.08 PTF 09sk 0.035 Ic-BL 8.85$\pm$0.14 -0.38$\pm$0.23 7.79$\pm$0.14 -1.45$\pm$0.23 2.36$\pm$0.05 PTF 09t 0.039 II 9.60$\pm$0.03 0.34$\pm$0.10 7.69$\pm$0.03 -1.57$\pm$0.10 5.14$\pm$0.04 PTF 09tm 0.035 II 10.35$\pm$0.03 0.30$\pm$0.07 8.62$\pm$0.03 -1.43$\pm$0.07 4.80$\pm$0.02 PTF 09uj 0.065 II 9.78$\pm$0.08 0.20$\pm$0.19 7.69$\pm$0.08 -1.89$\pm$0.19 5.75$\pm$0.10 PTF 10aavz 0.062 Ic-BL 9.09$\pm$0.11 -0.55$\pm$0.17 8.00$\pm$0.11 -1.64$\pm$0.17 2.33$\pm$0.13 PTF 10bau 0.026 II 10.35$\pm$0.03 0.40$\pm$0.07 8.32$\pm$0.03 -1.63$\pm$0.07 4.62$\pm$0.02 PTF 10bhu 0.036 Ic 9.44$\pm$0.14 -0.05$\pm$0.17 7.78$\pm$0.14 -1.71$\pm$0.17 3.63$\pm$0.04 PTF 10bip 0.051 Ic 9.10$\pm$0.10 -0.43$\pm$0.21 8.10$\pm$0.10 -1.43$\pm$0.21 2.01$\pm$0.06 PTF 10con 0.033 II 0.37$\pm$62.69 PTF 10cqh 0.041 II 10.77$\pm$0.04 0.97$\pm$0.08 8.51$\pm$0.04 -1.29$\pm$0.08 6.80$\pm$0.04 PTF 10cwx 0.073 II 9.65$\pm$0.08 0.08$\pm$0.18 7.80$\pm$0.08 -1.78$\pm$0.18 3.68$\pm$0.30 PTF 10cxq 0.047 II 8.93$\pm$0.11 -0.13$\pm$0.23 7.38$\pm$0.11 -1.68$\pm$0.23 5.86$\pm$0.14 ------------ --------- ------- ------------------- ----------------------------- ------------------------------ ---------------------------------------- ---------------- ----------- --------- ------- ------------------- ----------------------------- ------------------------------ ---------------------------------------- ---------------- Name [*z*]{} Type Mass SFR (phot) $\Sigma_{M}$ $\Sigma_{\rm SFR}$ $r_{50}$ (log M$_{\odot}$) (log M$_{\odot}$ yr$^{-1}$) (log M$_{\odot}$ kpc$^{-2}$) (log M$_{\odot}$ yr$^{-1}$ kpc$^{-2}$) (kpc) PTF 10cxx 0.034 II 10.27$\pm$0.13 0.24$\pm$0.17 8.88$\pm$0.13 -1.15$\pm$0.17 2.62$\pm$0.02 PTF 10czn 0.045 II 10.45$\pm$0.04 0.96$\pm$0.09 7.74$\pm$0.04 -1.75$\pm$0.09 9.71$\pm$0.05 PTF 10dk 0.074 II 8.54$\pm$0.19 -1.06$\pm$0.18 7.64$\pm$0.19 -1.96$\pm$0.18 1.47$\pm$0.51 PTF 10hv 0.052 II 11.22$\pm$0.04 1.60$\pm$0.07 9.17$\pm$0.04 -0.46$\pm$0.07 6.85$\pm$0.11 PTF 10qts 0.091 Ic-BL 9.60$\pm$0.09 -0.91$\pm$0.30 8.83$\pm$0.09 -1.68$\pm$0.30 1.17$\pm$0.36 PTF 10s 0.051 II 9.62$\pm$0.09 -0.09$\pm$0.15 7.98$\pm$0.09 -1.73$\pm$0.15 2.97$\pm$0.06 PTF 11cgx 0.033 II 9.92$\pm$0.13 0.41$\pm$0.17 8.19$\pm$0.13 -1.32$\pm$0.17 3.48$\pm$0.02 PTF 11cwi 0.056 II 10.56$\pm$0.04 0.51$\pm$0.08 8.54$\pm$0.04 -1.51$\pm$0.08 4.63$\pm$0.08 PTF 11dad 0.072 II 9.97$\pm$0.08 0.33$\pm$0.15 7.93$\pm$0.08 -1.71$\pm$0.15 7.48$\pm$0.15 PTF 11dqk 0.036 II 9.85$\pm$0.03 0.45$\pm$0.13 7.95$\pm$0.03 -1.45$\pm$0.13 3.93$\pm$0.02 PTF 11dqr 0.082 II 9.34$\pm$0.10 -0.21$\pm$0.19 6.64$\pm$0.10 -2.91$\pm$0.19 10.25$\pm$0.98 PTF 11dsb 0.190 II 9.70$\pm$0.08 0.23$\pm$0.16 7.28$\pm$0.08 -2.18$\pm$0.16 7.42$\pm$0.57 PTF 11dtd 0.040 II 10.25$\pm$0.03 0.80$\pm$0.07 7.75$\pm$0.03 -1.70$\pm$0.07 9.01$\pm$0.06 PTF 11ecp 0.034 II 10.00$\pm$0.03 0.60$\pm$0.07 8.05$\pm$0.03 -1.35$\pm$0.07 6.36$\pm$0.04 PTF 11ekj 0.043 II 9.92$\pm$0.12 -0.35$\pm$0.15 10.10$\pm$0.12 -0.18$\pm$0.15 0.43$\pm$0.08 PTF 11emc 0.082 II 9.22$\pm$0.09 -0.38$\pm$0.18 7.68$\pm$0.09 -1.92$\pm$0.18 3.97$\pm$0.23 PTF 11epi 0.032 II 11.00$\pm$0.03 0.10$\pm$0.07 8.48$\pm$0.03 -2.42$\pm$0.07 11.21$\pm$0.06 PTF 11ftr 0.018 II 7.80$\pm$0.32 -0.84$\pm$0.36 7.31$\pm$0.32 -1.33$\pm$0.36 0.84$\pm$0.01 PTF 11fuu 0.097 IIn 10.34$\pm$0.05 0.47$\pm$0.09 8.56$\pm$0.05 -1.30$\pm$0.09 4.02$\pm$0.08 PTF 11fuv 0.030 II 11.15$\pm$0.03 1.20$\pm$0.07 8.76$\pm$0.03 -1.19$\pm$0.07 7.24$\pm$0.04 PTF 11gdz 0.013 II 10.39$\pm$0.04 0.38$\pm$0.08 9.40$\pm$0.04 -0.61$\pm$0.08 1.29$\pm$0.01 PTF 11hyg 0.030 Ic 10.84$\pm$0.04 1.02$\pm$0.07 8.32$\pm$0.04 -1.50$\pm$0.07 7.81$\pm$0.03 PTF 11iqb 0.013 IIn 11.59$\pm$0.32 1.58$\pm$0.35 9.34$\pm$0.32 -0.68$\pm$0.35 5.93$\pm$0.03 PTF 11ixk 0.021 Ic 10.26$\pm$0.04 0.60$\pm$0.07 8.19$\pm$0.04 -1.46$\pm$0.07 5.48$\pm$0.02 PTF 11izq 0.062 Ib 9.23$\pm$0.09 -0.33$\pm$0.17 7.24$\pm$0.09 -2.31$\pm$0.17 4.66$\pm$0.18 PTF 11jgp 0.070 II 9.91$\pm$0.04 0.54$\pm$0.10 7.95$\pm$0.04 -1.42$\pm$0.10 4.78$\pm$0.07 PTF 11kjk 0.067 II 9.12$\pm$0.10 -0.17$\pm$0.20 7.44$\pm$0.10 -1.85$\pm$0.20 5.67$\pm$0.25 PTF 11klg 0.027 Ic 10.62$\pm$0.19 0.44$\pm$0.22 8.56$\pm$0.19 -1.62$\pm$0.22 5.15$\pm$0.04 PTF 11kqn 0.066 II 11.26$\pm$0.05 1.09$\pm$0.07 8.89$\pm$0.05 -1.29$\pm$0.07 8.06$\pm$0.08 PTF 11ktr 0.106 II 10.45$\pm$0.04 0.80$\pm$0.07 8.03$\pm$0.04 -1.63$\pm$0.07 8.32$\pm$0.18 PTF 11mhh 0.039 II 9.96$\pm$0.15 0.22$\pm$0.15 7.95$\pm$0.15 -1.79$\pm$0.15 4.45$\pm$0.07 PTF 11mmk 0.049 II 9.55$\pm$0.11 -0.16$\pm$0.14 8.10$\pm$0.11 -1.60$\pm$0.14 3.75$\pm$0.07 PTF 11mpv 0.043 II 9.26$\pm$0.09 -0.14$\pm$0.12 8.18$\pm$0.09 -1.22$\pm$0.12 2.58$\pm$0.04 PTF 11pdj 0.024 II 10.85$\pm$0.03 -0.20$\pm$0.07 8.96$\pm$0.03 -2.09$\pm$0.07 9.74$\pm$0.07 PTF 11qcc 0.043 II 9.73$\pm$0.12 -0.11$\pm$0.15 8.19$\pm$0.12 -1.65$\pm$0.15 6.11$\pm$0.12 PTF 11qcm 0.050 II 10.60$\pm$0.19 0.61$\pm$0.19 8.43$\pm$0.19 -1.57$\pm$0.19 7.71$\pm$0.05 PTF 11qgw 0.027 II 8.98$\pm$0.19 -0.27$\pm$0.29 7.16$\pm$0.19 -2.09$\pm$0.29 6.88$\pm$0.10 PTF 11qju 0.028 II 9.12$\pm$0.18 -0.41$\pm$0.23 10.24$\pm$0.18 0.71$\pm$0.23 0.15$\pm$0.34 PTF 11qux 0.041 II 0.66$\pm$0.01 PTF 12boj 0.037 II 10.58$\pm$0.16 0.57$\pm$0.17 8.53$\pm$0.16 -1.48$\pm$0.17 4.47$\pm$0.02 PTF 12bpy 0.060 II 9.36$\pm$0.14 -0.15$\pm$0.22 8.70$\pm$0.14 -0.80$\pm$0.22 0.96$\pm$0.04 PTF 12bwq 0.040 Ib 9.53$\pm$0.22 -0.11$\pm$0.27 7.07$\pm$0.22 -2.57$\pm$0.27 9.62$\pm$0.21 PTF 12cdc 0.070 II 10.78$\pm$0.11 0.77$\pm$0.14 7.77$\pm$0.11 -2.25$\pm$0.14 15.31$\pm$0.16 PTF 12cde 0.013 Ib/c 8.36$\pm$0.32 -1.17$\pm$0.38 7.03$\pm$0.32 -2.50$\pm$0.38 3.61$\pm$0.12 PTF 12cgb 0.026 II 9.27$\pm$0.04 -0.14$\pm$0.12 7.99$\pm$0.04 -1.42$\pm$0.12 2.61$\pm$0.01 PTF 12dke 0.067 II 9.91$\pm$0.08 0.46$\pm$0.15 7.40$\pm$0.08 -2.05$\pm$0.15 11.27$\pm$0.20 PTF 12eje 0.078 II 9.92$\pm$0.07 0.49$\pm$0.18 7.82$\pm$0.07 -1.61$\pm$0.18 5.60$\pm$0.07 PTF 12fes 0.036 Ib 10.55$\pm$0.04 0.99$\pm$0.07 7.98$\pm$0.04 -1.58$\pm$0.07 8.51$\pm$0.07 PTF 12gcx 0.045 II 9.74$\pm$0.04 0.48$\pm$0.08 7.42$\pm$0.04 -1.84$\pm$0.08 7.16$\pm$0.09 PTF 12gvr 0.056 Ib/c 10.60$\pm$0.03 1.08$\pm$0.08 8.07$\pm$0.03 -1.46$\pm$0.08 9.24$\pm$0.04 PTF 12gzk 0.014 12gzk 7.47$\pm$0.11 -1.46$\pm$0.38 7.43$\pm$0.11 -1.50$\pm$0.38 0.54$\pm$0.05 PTF 12jje 0.042 II 10.01$\pm$0.12 0.27$\pm$0.16 7.89$\pm$0.12 -1.84$\pm$0.16 5.37$\pm$0.08 PTF 12ne 0.033 II 9.91$\pm$0.13 -0.06$\pm$0.17 8.62$\pm$0.13 -1.35$\pm$0.17 1.82$\pm$0.02 PTF 13bvn 0.004 Ib 11.10$\pm$0.03 1.00$\pm$0.07 9.66$\pm$0.03 -0.44$\pm$0.07 2.68$\pm$0.02 PTF 13c 0.011 II 9.04$\pm$0.04 -0.70$\pm$0.07 8.00$\pm$0.04 -1.74$\pm$0.07 1.56$\pm$0.01 ----------- --------- ------- ------------------- ----------------------------- ------------------------------ ---------------------------------------- ---------------- ----------- --------- -------- ------------------- ----------------------------- ------------------------------ ---------------------------------------- ---------------- Name [*z*]{} Type Mass SFR (phot) $\Sigma_{M}$ $\Sigma_{\rm SFR}$ $r_{50}$ (log M$_{\odot}$) (log M$_{\odot}$ yr$^{-1}$) (log M$_{\odot}$ kpc$^{-2}$) (log M$_{\odot}$ yr$^{-1}$ kpc$^{-2}$) (kpc) PTF 13cab 0.030 Ib 9.32$\pm$0.30 -0.28$\pm$0.37 7.72$\pm$0.30 -1.88$\pm$0.37 4.39$\pm$0.06 PTF 13cac 0.030 II 8.91$\pm$0.31 -0.87$\pm$0.37 7.38$\pm$0.31 -2.40$\pm$0.37 3.97$\pm$0.13 PTF 13cbf 0.039 Ic 9.40$\pm$0.03 -0.00$\pm$0.07 7.72$\pm$0.03 -1.68$\pm$0.07 5.78$\pm$0.06 PTF 13d 0.024 II 10.25$\pm$0.04 0.60$\pm$0.07 8.07$\pm$0.04 -1.58$\pm$0.07 6.27$\pm$0.02 SN 1999ap 0.040 II 9.35$\pm$0.12 -0.04$\pm$0.18 7.33$\pm$0.12 -2.06$\pm$0.18 5.25$\pm$0.10 SN 1999as 0.127 Ic pec 9.02$\pm$1.05 SN 1999bc 0.021 Ic 10.46$\pm$0.04 0.62$\pm$0.08 8.18$\pm$0.04 -1.66$\pm$0.08 6.74$\pm$0.03 SN 1999bd 0.151 II 9.66$\pm$0.09 0.07$\pm$0.16 8.58$\pm$0.09 -1.01$\pm$0.16 2.86$\pm$0.28 SN 2001bk 0.043 II 8.82$\pm$0.13 -0.79$\pm$0.19 7.58$\pm$0.13 -2.02$\pm$0.19 3.13$\pm$0.15 SN 2001fb 0.032 II 9.95$\pm$0.03 0.60$\pm$0.07 8.53$\pm$0.03 -0.82$\pm$0.07 3.46$\pm$0.02 SN 2001ij 0.038 II P 10.20$\pm$0.03 0.40$\pm$0.07 8.22$\pm$0.03 -1.59$\pm$0.07 5.55$\pm$0.04 SN 2002dg 0.047 Ib 9.10$\pm$0.11 0.06$\pm$0.16 7.10$\pm$0.11 -1.94$\pm$0.16 4.31$\pm$0.07 SN 2002ew 0.030 II 9.36$\pm$0.04 -0.08$\pm$0.11 7.95$\pm$0.04 -1.49$\pm$0.11 2.76$\pm$0.02 SN 2002fa 0.060 II 10.12$\pm$0.09 0.59$\pm$0.20 7.43$\pm$0.09 -2.10$\pm$0.20 9.16$\pm$0.11 SN 2002fu 0.091 II 9.86$\pm$0.07 0.25$\pm$0.14 7.90$\pm$0.07 -1.72$\pm$0.14 9.06$\pm$0.21 SN 2002hj 0.024 II 9.60$\pm$0.20 -0.02$\pm$0.26 7.75$\pm$0.20 -1.87$\pm$0.26 4.04$\pm$0.05 SN 2002ik 0.032 II P 10.55$\pm$0.03 0.70$\pm$0.07 8.52$\pm$0.03 -1.33$\pm$0.07 5.92$\pm$0.03 SN 2002in 0.076 II 9.11$\pm$0.09 -0.48$\pm$0.19 7.64$\pm$0.09 -1.96$\pm$0.19 4.15$\pm$0.33 SN 2002ip 0.079 II 8.90$\pm$0.10 -0.61$\pm$0.19 6.76$\pm$0.10 -2.75$\pm$0.19 5.55$\pm$0.80 SN 2002iq 0.056 II 9.38$\pm$0.06 0.13$\pm$0.14 7.81$\pm$0.06 -1.44$\pm$0.14 3.98$\pm$0.18 SN 2002jl 0.064 II 8.01$\pm$0.23 -1.23$\pm$0.22 7.30$\pm$0.23 -1.93$\pm$0.22 1.37$\pm$0.33 SN 2003cv 0.028 II pec 8.31$\pm$0.20 -0.96$\pm$0.27 7.40$\pm$0.20 -1.86$\pm$0.27 1.90$\pm$0.05 SN 2003dq 0.046 II 9.01$\pm$0.12 -0.43$\pm$0.16 7.59$\pm$0.12 -1.85$\pm$0.16 2.95$\pm$0.12 SN 2003kj 0.100 II 8.77$\pm$0.18 -0.79$\pm$0.18 7.14$\pm$0.18 -2.43$\pm$0.18 3.04$\pm$0.78 SN 2004cm 0.004 II P 9.65$\pm$0.03 0.00$\pm$0.07 8.51$\pm$0.03 -1.14$\pm$0.07 1.75$\pm$0.01 SN 2004gy 0.027 II 8.26$\pm$0.20 -1.02$\pm$0.26 7.44$\pm$0.20 -1.83$\pm$0.26 1.63$\pm$0.05 SN 2004ht 0.067 II 10.30$\pm$0.07 0.60$\pm$0.09 7.97$\pm$0.07 -1.72$\pm$0.09 7.58$\pm$0.09 SN 2004hv 0.061 II 8.82$\pm$0.13 -0.70$\pm$0.19 6.90$\pm$0.13 -2.62$\pm$0.19 5.27$\pm$0.98 SN 2004hx 0.014 II 8.69$\pm$0.31 -1.11$\pm$0.35 7.72$\pm$0.31 -2.08$\pm$0.35 1.63$\pm$0.02 SN 2004ic 0.093 II 10.92$\pm$0.06 1.03$\pm$0.08 7.80$\pm$0.06 -2.09$\pm$0.08 15.04$\pm$0.30 SN 2005bn 0.028 II 9.14$\pm$0.20 -0.44$\pm$0.28 8.30$\pm$0.20 -1.28$\pm$0.28 1.38$\pm$0.01 SN 2005fq 0.140 II 8.83$\pm$0.14 -0.54$\pm$0.18 7.46$\pm$0.14 -1.91$\pm$0.18 3.35$\pm$1.08 SN 2005gi 0.050 II 9.12$\pm$0.14 -0.39$\pm$0.16 7.10$\pm$0.14 -2.41$\pm$0.16 4.74$\pm$0.26 SN 2005hl 0.020 Ib 10.40$\pm$0.04 0.40$\pm$0.07 8.91$\pm$0.04 -1.09$\pm$0.07 2.94$\pm$0.01 SN 2005hm 0.030 Ib 8.61$\pm$0.30 -1.53$\pm$0.34 8.28$\pm$0.30 -1.87$\pm$0.34 0.94$\pm$0.19 SN 2005kb 0.015 II 9.35$\pm$0.31 -0.43$\pm$0.39 8.19$\pm$0.31 -1.59$\pm$0.39 4.32$\pm$0.05 SN 2005kr 0.130 Ic-BL 8.63$\pm$0.16 -0.62$\pm$0.20 8.36$\pm$0.16 -0.88$\pm$0.20 0.57$\pm$0.30 SN 2005ks 0.100 Ic-BL 9.88$\pm$0.08 0.27$\pm$0.14 8.44$\pm$0.08 -1.17$\pm$0.14 3.70$\pm$0.08 SN 2005lb 0.030 II 8.75$\pm$0.30 -0.94$\pm$0.32 7.47$\pm$0.30 -2.22$\pm$0.32 3.95$\pm$0.37 SN 2005lc 0.010 II 8.21$\pm$0.33 -1.34$\pm$0.39 7.41$\pm$0.33 -2.13$\pm$0.39 1.32$\pm$0.06 SN 2005lm 0.080 II 9.23$\pm$0.08 -0.24$\pm$0.16 8.01$\pm$0.08 -1.46$\pm$0.16 1.83$\pm$0.07 SN 2005mk 0.150 II 9.44$\pm$0.09 -0.16$\pm$0.17 7.50$\pm$0.09 -2.09$\pm$0.17 4.13$\pm$0.64 SN 2005mn 0.050 Ib 9.61$\pm$0.16 0.17$\pm$0.22 7.46$\pm$0.16 -1.97$\pm$0.22 11.24$\pm$0.26 SN 2006L 0.039 IIn 8.28$\pm$0.16 -1.19$\pm$0.19 7.63$\pm$0.16 -1.84$\pm$0.19 1.18$\pm$0.08 SN 2006M 0.015 IIn 9.05$\pm$0.03 -0.42$\pm$0.08 7.73$\pm$0.03 -1.74$\pm$0.08 1.93$\pm$0.01 SN 2006ad 0.030 II 8.97$\pm$0.05 -1.27$\pm$0.10 11.66$\pm$0.05 1.42$\pm$0.10 0.02$\pm$0.02 SN 2006ag 0.035 IIn 8.05$\pm$0.16 -0.63$\pm$0.18 6.91$\pm$0.16 -1.77$\pm$0.18 1.74$\pm$0.08 SN 2006aj 0.033 Ic-BL 8.08$\pm$0.17 -1.31$\pm$0.25 8.01$\pm$0.17 -1.38$\pm$0.25 0.46$\pm$0.10 SN 2006bj 0.038 II 9.38$\pm$0.15 -0.08$\pm$0.19 7.32$\pm$0.15 -2.15$\pm$0.19 4.80$\pm$0.07 SN 2006cu 0.029 IIn 10.15$\pm$0.03 0.50$\pm$0.07 7.79$\pm$0.03 -1.86$\pm$0.07 7.15$\pm$0.02 SN 2006cv 0.100 IIn 10.33$\pm$0.05 -0.60$\pm$0.07 8.61$\pm$0.05 -2.32$\pm$0.07 3.18$\pm$0.22 SN 2006cw 0.065 II 10.02$\pm$0.06 0.50$\pm$0.09 7.82$\pm$0.06 -1.70$\pm$0.09 6.72$\pm$0.11 SN 2006cy 0.036 IIn 10.21$\pm$0.04 0.69$\pm$0.08 7.88$\pm$0.04 -1.63$\pm$0.08 6.07$\pm$0.03 SN 2006db 0.023 IIn 8.69$\pm$0.19 -0.59$\pm$0.24 7.40$\pm$0.19 -1.88$\pm$0.24 2.39$\pm$0.04 SN 2006fg 0.030 II 7.99$\pm$0.30 -1.53$\pm$0.32 8.17$\pm$0.30 -1.35$\pm$0.32 0.34$\pm$0.03 ----------- --------- -------- ------------------- ----------------------------- ------------------------------ ---------------------------------------- ---------------- ----------- --------- ------- ------------------- ----------------------------- ------------------------------ ---------------------------------------- ---------------- Name [*z*]{} Type Mass SFR (phot) $\Sigma_{M}$ $\Sigma_{\rm SFR}$ $r_{50}$ (log M$_{\odot}$) (log M$_{\odot}$ yr$^{-1}$) (log M$_{\odot}$ kpc$^{-2}$) (log M$_{\odot}$ yr$^{-1}$ kpc$^{-2}$) (kpc) SN 2006fo 0.021 Ib 10.25$\pm$0.03 0.50$\pm$0.07 8.53$\pm$0.03 -1.22$\pm$0.07 3.18$\pm$0.01 SN 2006fq 0.070 II P 9.96$\pm$0.04 0.50$\pm$0.07 8.13$\pm$0.04 -1.34$\pm$0.07 4.61$\pm$0.04 SN 2006gd 0.150 II P 10.97$\pm$0.07 0.86$\pm$0.16 8.23$\pm$0.07 -1.89$\pm$0.16 15.45$\pm$0.42 SN 2006gy 0.019 IIn 11.17$\pm$0.32 0.54$\pm$0.35 9.36$\pm$0.32 -1.27$\pm$0.35 4.74$\pm$0.03 SN 2006ho 0.110 II 11.25$\pm$0.03 0.80$\pm$0.07 9.01$\pm$0.03 -1.44$\pm$0.07 6.09$\pm$0.09 SN 2006ic 0.040 II 10.45$\pm$0.04 0.70$\pm$0.07 8.17$\pm$0.04 -1.58$\pm$0.07 5.95$\pm$0.04 SN 2006ih 0.130 II 8.95$\pm$0.18 -0.30$\pm$0.20 9.55$\pm$0.18 0.30$\pm$0.20 0.24$\pm$0.09 SN 2006ii 0.030 II 9.44$\pm$0.30 -0.32$\pm$0.31 7.51$\pm$0.30 -2.25$\pm$0.31 4.11$\pm$0.07 SN 2006ij 0.040 II 9.81$\pm$0.11 0.28$\pm$0.16 7.74$\pm$0.11 -1.79$\pm$0.16 6.70$\pm$0.06 SN 2006ip 0.030 Ic 9.64$\pm$0.19 0.15$\pm$0.19 7.87$\pm$0.19 -1.61$\pm$0.19 3.61$\pm$0.03 SN 2006ir 0.020 Ic 8.77$\pm$0.32 -0.85$\pm$0.35 7.11$\pm$0.32 -2.51$\pm$0.35 3.49$\pm$0.05 SN 2006iw 0.030 II 9.67$\pm$0.29 -0.02$\pm$0.31 8.06$\pm$0.29 -1.63$\pm$0.31 4.19$\pm$0.09 SN 2006ix 0.080 II 9.65$\pm$0.11 -0.02$\pm$0.19 7.79$\pm$0.11 -1.88$\pm$0.19 7.91$\pm$0.23 SN 2006kh 0.060 II 9.46$\pm$0.08 -0.23$\pm$0.13 8.97$\pm$0.08 -0.72$\pm$0.13 0.81$\pm$0.02 SN 2006kn 0.120 II 9.51$\pm$0.07 0.10$\pm$0.19 7.30$\pm$0.07 -2.12$\pm$0.19 10.94$\pm$0.58 SN 2006lc 0.016 Ib 10.65$\pm$0.03 0.40$\pm$0.07 8.59$\pm$0.03 -1.66$\pm$0.07 5.27$\pm$0.02 SN 2006lh 0.032 II 7.95$\pm$0.18 -1.41$\pm$0.22 7.71$\pm$0.18 -1.65$\pm$0.22 0.62$\pm$0.06 SN 2006ls 0.140 I pec 9.88$\pm$0.08 0.25$\pm$0.16 7.33$\pm$0.08 -2.30$\pm$0.16 11.10$\pm$0.56 SN 2006lt 0.015 Ib 8.82$\pm$0.34 -1.01$\pm$0.35 9.60$\pm$0.34 -0.23$\pm$0.35 0.18$\pm$0.08 SN 2006nq 0.025 II 9.38$\pm$0.19 -0.11$\pm$0.19 7.40$\pm$0.19 -2.09$\pm$0.19 4.86$\pm$0.05 SN 2006ns 0.120 II 9.81$\pm$0.04 0.54$\pm$0.15 7.77$\pm$0.04 -1.50$\pm$0.15 6.99$\pm$0.14 SN 2006nx 0.050 Ic-BL 8.57$\pm$0.17 -0.98$\pm$0.25 7.40$\pm$0.17 -2.16$\pm$0.25 1.79$\pm$0.20 SN 2006ny 0.080 II P 10.22$\pm$0.05 0.18$\pm$0.09 8.41$\pm$0.05 -1.63$\pm$0.09 7.10$\pm$0.15 SN 2006qk 0.060 Ic-BL 9.61$\pm$0.13 -0.31$\pm$0.17 8.89$\pm$0.13 -1.03$\pm$0.17 1.71$\pm$0.05 SN 2006rc 0.080 IIn 9.88$\pm$0.11 0.25$\pm$0.17 7.70$\pm$0.11 -1.93$\pm$0.17 6.60$\pm$0.15 SN 2006rq 0.070 II 10.86$\pm$0.04 0.02$\pm$0.07 8.67$\pm$0.04 -2.17$\pm$0.07 8.67$\pm$0.08 SN 2006ru 0.020 II 11.15$\pm$0.03 0.10$\pm$0.07 10.25$\pm$0.03 -0.80$\pm$0.07 1.63$\pm$0.00 SN 2006ry 0.060 II 10.95$\pm$0.03 9.51$\pm$0.03 4.05$\pm$0.07 SN 2006tf 0.074 IIn 8.16$\pm$0.17 -0.97$\pm$0.21 8.84$\pm$0.17 -0.29$\pm$0.21 0.61$\pm$1.85 SN 2006th 0.140 II 4.24$\pm$0.97 SN 2007I 0.022 Ic-BL 8.92$\pm$0.21 -0.75$\pm$0.26 7.78$\pm$0.21 -1.88$\pm$0.26 1.92$\pm$0.05 SN 2007bg 0.034 Ic-BL 7.99$\pm$0.23 -2.03$\pm$0.33 8.21$\pm$0.23 -1.81$\pm$0.33 0.39$\pm$0.16 SN 2007bo 0.040 II 9.13$\pm$0.21 -0.33$\pm$0.28 7.35$\pm$0.21 -2.10$\pm$0.28 3.61$\pm$0.08 SN 2007bp 0.030 II 10.80$\pm$0.14 1.00$\pm$0.22 7.99$\pm$0.14 -1.81$\pm$0.22 10.71$\pm$0.06 SN 2007bt 0.040 IIn 9.65$\pm$0.22 -0.03$\pm$0.29 7.62$\pm$0.22 -2.07$\pm$0.29 4.85$\pm$0.06 SN 2007bu 0.030 II 8.46$\pm$0.30 -1.26$\pm$0.34 7.67$\pm$0.30 -2.05$\pm$0.34 1.54$\pm$0.10 SN 2007bv 0.050 II 11.34$\pm$0.09 1.10$\pm$0.15 8.65$\pm$0.09 -1.59$\pm$0.15 11.25$\pm$0.09 SN 2007bw 0.140 IIn 9.66$\pm$0.08 0.40$\pm$0.15 7.45$\pm$0.08 -1.81$\pm$0.15 5.43$\pm$0.28 SN 2007bx 0.020 II 8.16$\pm$0.48 -1.24$\pm$0.51 7.19$\pm$0.48 -2.21$\pm$0.51 2.16$\pm$0.14 SN 2007by 0.040 II 10.23$\pm$0.13 0.42$\pm$0.19 8.00$\pm$0.13 -1.81$\pm$0.19 6.11$\pm$0.04 SN 2007ce 0.046 Ic-BL 8.08$\pm$0.13 -0.60$\pm$0.16 8.03$\pm$0.13 -0.66$\pm$0.16 0.45$\pm$0.19 SN 2007dp 0.030 II 9.14$\pm$0.28 -0.52$\pm$0.33 8.03$\pm$0.28 -1.63$\pm$0.33 1.77$\pm$0.02 SN 2007dq 0.050 II 8.71$\pm$0.20 -1.01$\pm$0.28 7.14$\pm$0.20 -2.58$\pm$0.28 3.42$\pm$0.39 SN 2007dw 0.050 II 10.07$\pm$0.06 0.59$\pm$0.09 7.80$\pm$0.06 -1.69$\pm$0.09 6.10$\pm$0.04 SN 2007dy 0.040 Ib 9.21$\pm$0.22 -0.32$\pm$0.27 7.29$\pm$0.22 -2.24$\pm$0.27 3.71$\pm$0.12 SN 2007eb 0.040 Ic-BL 8.50$\pm$0.22 -0.78$\pm$0.30 7.22$\pm$0.22 -2.06$\pm$0.30 3.31$\pm$0.16 SN 2007ed 0.070 II 10.69$\pm$0.11 0.54$\pm$0.15 8.71$\pm$0.11 -1.44$\pm$0.15 4.87$\pm$0.08 SN 2007eh 0.010 II 8.04$\pm$0.32 -1.32$\pm$0.39 9.05$\pm$0.32 -0.32$\pm$0.39 0.15$\pm$0.06 SN 2007el 0.030 II 9.53$\pm$0.30 -0.13$\pm$0.34 7.21$\pm$0.30 -2.45$\pm$0.34 6.23$\pm$0.17 SN 2007em 0.030 II 7.63$\pm$0.31 -1.54$\pm$0.36 7.00$\pm$0.31 -2.17$\pm$0.36 1.21$\pm$0.25 SN 2007eq 0.030 Ib/c 8.49$\pm$0.30 -1.14$\pm$0.32 7.47$\pm$0.30 -2.17$\pm$0.32 2.80$\pm$0.16 SN 2007er 0.070 II 9.62$\pm$0.12 0.08$\pm$0.16 7.32$\pm$0.12 -2.22$\pm$0.16 6.69$\pm$0.19 SN 2007es 0.030 II 10.94$\pm$0.04 0.99$\pm$0.08 8.56$\pm$0.04 -1.39$\pm$0.08 7.46$\pm$0.08 SN 2007et 0.040 II 10.00$\pm$0.04 0.41$\pm$0.08 7.71$\pm$0.04 -1.89$\pm$0.08 6.38$\pm$0.03 SN 2007eu 0.040 II 9.33$\pm$0.05 -0.03$\pm$0.13 7.40$\pm$0.05 -1.96$\pm$0.13 3.92$\pm$0.06 ----------- --------- ------- ------------------- ----------------------------- ------------------------------ ---------------------------------------- ---------------- ----------- --------- -------- ------------------- ----------------------------- ------------------------------ ---------------------------------------- ---------------- Name [*z*]{} Type Mass SFR (phot) $\Sigma_{M}$ $\Sigma_{\rm SFR}$ $r_{50}$ (log M$_{\odot}$) (log M$_{\odot}$ yr$^{-1}$) (log M$_{\odot}$ kpc$^{-2}$) (log M$_{\odot}$ yr$^{-1}$ kpc$^{-2}$) (kpc) SN 2007ew 0.030 II 9.38$\pm$0.30 -0.40$\pm$0.36 7.54$\pm$0.30 -2.24$\pm$0.36 4.78$\pm$0.14 SN 2007fa 0.060 II 11.07$\pm$0.09 1.05$\pm$0.10 8.54$\pm$0.09 -1.48$\pm$0.10 8.70$\pm$0.10 SN 2007fe 0.030 II 9.39$\pm$0.21 -0.38$\pm$0.23 7.89$\pm$0.21 -1.88$\pm$0.23 2.63$\pm$0.02 SN 2007ff 0.050 Ic 10.45$\pm$0.13 0.64$\pm$0.18 8.09$\pm$0.13 -1.72$\pm$0.18 6.45$\pm$0.06 SN 2007fg 0.030 II 8.87$\pm$0.20 -0.35$\pm$0.29 7.12$\pm$0.20 -2.10$\pm$0.29 3.81$\pm$0.08 SN 2007fk 0.040 IIn 8.61$\pm$0.22 -1.05$\pm$0.41 7.04$\pm$0.22 -2.62$\pm$0.41 3.64$\pm$0.20 SN 2007fw 0.050 IIn 9.05$\pm$0.18 -0.52$\pm$0.22 7.74$\pm$0.18 -1.83$\pm$0.22 3.91$\pm$0.15 SN 2007fy 0.050 II 9.96$\pm$0.04 0.61$\pm$0.07 8.00$\pm$0.04 -1.35$\pm$0.07 5.62$\pm$0.04 SN 2007fz 0.014 II 8.44$\pm$0.31 -1.22$\pm$0.32 7.85$\pm$0.31 -1.81$\pm$0.32 1.54$\pm$0.01 SN 2007gh 0.020 II 9.45$\pm$0.30 -0.30$\pm$0.33 8.70$\pm$0.30 -1.05$\pm$0.33 1.10$\pm$0.01 SN 2007gl 0.030 Ic 9.90$\pm$0.17 0.20$\pm$0.17 7.39$\pm$0.17 -2.31$\pm$0.17 7.37$\pm$0.07 SN 2007gm 0.030 II 9.13$\pm$0.31 -0.51$\pm$0.43 8.05$\pm$0.31 -1.59$\pm$0.43 1.48$\pm$0.03 SN 2007gs 0.040 II 9.57$\pm$0.13 -0.01$\pm$0.12 7.40$\pm$0.13 -2.18$\pm$0.12 6.47$\pm$0.12 SN 2007gy 0.040 IIn 9.34$\pm$0.22 -0.22$\pm$0.23 7.26$\pm$0.22 -2.29$\pm$0.23 4.40$\pm$0.18 SN 2007gz 0.050 II 9.80$\pm$0.18 0.01$\pm$0.19 9.01$\pm$0.18 -0.78$\pm$0.19 1.01$\pm$0.01 SN 2007hb 0.022 Ic 10.55$\pm$0.03 0.80$\pm$0.07 8.51$\pm$0.03 -1.24$\pm$0.07 4.87$\pm$0.01 SN 2007hi 0.070 II 8.11$\pm$0.35 -1.49$\pm$0.26 10.20$\pm$0.35 0.59$\pm$0.26 0.06$\pm$0.12 SN 2007hn 0.030 Ic 9.86$\pm$0.30 -0.03$\pm$0.30 8.29$\pm$0.30 -1.60$\pm$0.30 3.30$\pm$0.04 SN 2007hs 0.070 II 8.68$\pm$0.20 -1.11$\pm$0.28 7.65$\pm$0.20 -2.15$\pm$0.28 2.50$\pm$0.75 SN 2007hw 0.080 II 10.85$\pm$0.03 1.00$\pm$0.07 8.55$\pm$0.03 -1.30$\pm$0.07 7.15$\pm$0.06 SN 2007ib 0.030 II 9.52$\pm$0.18 -0.02$\pm$0.30 7.69$\pm$0.18 -1.86$\pm$0.30 4.31$\pm$0.03 SN 2007iu 0.090 II 9.31$\pm$0.13 -0.35$\pm$0.15 6.59$\pm$0.13 -3.07$\pm$0.15 9.77$\pm$0.24 SN 2007ja 0.090 II P 10.65$\pm$0.03 1.00$\pm$0.07 8.22$\pm$0.03 -1.43$\pm$0.07 7.78$\pm$0.08 SN 2007jf 0.070 II P 9.65$\pm$0.13 -0.03$\pm$0.21 7.78$\pm$0.13 -1.90$\pm$0.21 5.06$\pm$0.19 SN 2007jm 0.090 II n 9.77$\pm$0.10 0.10$\pm$0.16 8.34$\pm$0.10 -1.34$\pm$0.16 4.36$\pm$0.19 SN 2007jn 0.060 II 8.99$\pm$0.14 -0.38$\pm$0.21 7.54$\pm$0.14 -1.84$\pm$0.21 4.79$\pm$0.24 SN 2007kw 0.070 II 10.82$\pm$0.06 0.72$\pm$0.08 8.67$\pm$0.06 -1.44$\pm$0.08 6.24$\pm$0.07 SN 2007ky 0.070 II 10.85$\pm$0.03 0.90$\pm$0.07 8.02$\pm$0.03 -1.93$\pm$0.07 14.62$\pm$0.12 SN 2007kz 0.130 II 11.20$\pm$0.05 1.36$\pm$0.09 8.14$\pm$0.05 -1.70$\pm$0.09 16.25$\pm$0.21 SN 2007lb 0.060 II 1.90$\pm$0.67 SN 2007ld 0.030 II 7.96$\pm$0.30 -1.35$\pm$0.34 6.54$\pm$0.30 -2.77$\pm$0.34 2.72$\pm$0.43 SN 2007lj 0.040 II 8.06$\pm$0.24 -1.27$\pm$0.28 8.43$\pm$0.24 -0.90$\pm$0.28 0.51$\pm$0.90 SN 2007lz 0.090 II 9.30$\pm$0.12 -0.39$\pm$0.20 7.18$\pm$0.12 -2.51$\pm$0.20 5.65$\pm$0.49 SN 2007md 0.050 II 10.93$\pm$0.10 0.97$\pm$0.14 8.57$\pm$0.10 -1.40$\pm$0.14 9.00$\pm$0.04 SN 2007ms 0.040 II pec 8.50$\pm$0.22 -0.57$\pm$0.27 7.70$\pm$0.22 -1.37$\pm$0.27 1.70$\pm$0.09 SN 2007nm 0.046 Ic 8.47$\pm$0.22 -1.72$\pm$0.35 8.87$\pm$0.22 -1.32$\pm$0.35 0.36$\pm$0.22 SN 2007nr 0.140 II P 9.46$\pm$0.07 0.00$\pm$0.18 7.67$\pm$0.07 -1.79$\pm$0.18 3.54$\pm$0.26 SN 2007nw 0.060 II P 10.26$\pm$0.05 0.42$\pm$0.08 8.19$\pm$0.05 -1.65$\pm$0.08 5.68$\pm$0.09 SN 2007ny 0.140 II P 9.11$\pm$0.18 -0.66$\pm$0.24 7.52$\pm$0.18 -2.25$\pm$0.24 6.39$\pm$3.82 SN 2007qb 0.080 II 10.27$\pm$0.07 0.73$\pm$0.15 7.53$\pm$0.07 -2.01$\pm$0.15 10.93$\pm$0.18 SN 2007qv 0.100 II 10.51$\pm$0.10 0.57$\pm$0.12 8.32$\pm$0.10 -1.63$\pm$0.12 6.88$\pm$0.17 SN 2007qw 0.150 Ic-BL 9.40$\pm$0.05 0.13$\pm$0.13 8.31$\pm$0.05 -0.96$\pm$0.13 1.84$\pm$0.22 SN 2007qx 0.060 Ib 10.03$\pm$0.11 0.34$\pm$0.15 7.90$\pm$0.11 -1.80$\pm$0.15 6.87$\pm$0.08 SN 2007sd 0.090 II P 8.95$\pm$0.11 -0.48$\pm$0.19 7.38$\pm$0.11 -2.06$\pm$0.19 4.73$\pm$0.61 SN 2007sj 0.040 Ib/c 10.45$\pm$0.03 0.80$\pm$0.07 8.01$\pm$0.03 -1.64$\pm$0.07 7.09$\pm$0.03 SN 2007sx 0.120 II 10.96$\pm$0.04 1.05$\pm$0.10 7.84$\pm$0.04 -2.07$\pm$0.10 16.58$\pm$0.29 SN 2007sz 0.020 II 8.56$\pm$0.31 -0.98$\pm$0.31 7.21$\pm$0.31 -2.34$\pm$0.31 2.20$\pm$0.04 SN 2007tn 0.050 II 10.27$\pm$0.16 0.36$\pm$0.16 8.18$\pm$0.16 -1.73$\pm$0.16 4.56$\pm$0.06 SN 2008bj 0.019 II 8.54$\pm$0.28 -0.95$\pm$0.28 6.92$\pm$0.28 -2.57$\pm$0.28 3.24$\pm$0.04 SN 2008fm 0.039 IIn 11.29$\pm$0.13 1.28$\pm$0.14 8.30$\pm$0.13 -1.71$\pm$0.14 17.13$\pm$0.10 SN 2008fn 0.030 Ib/c 9.86$\pm$0.20 -0.56$\pm$0.30 8.33$\pm$0.20 -2.09$\pm$0.30 2.63$\pm$0.03 SN 2008fo 0.030 Ic 9.50$\pm$0.04 -0.07$\pm$0.09 7.70$\pm$0.04 -1.87$\pm$0.09 4.20$\pm$0.02 SN 2008fs 0.039 Ib/c 10.19$\pm$0.05 0.25$\pm$0.11 8.29$\pm$0.05 -1.65$\pm$0.11 4.15$\pm$0.04 SN 2008fz 0.133 IIn 9.93$\pm$0.05 1.25$\pm$0.10 13.12$\pm$0.05 4.44$\pm$0.10 0.01$\pm$0.08 SN 2008gd 0.059 II 9.89$\pm$0.09 0.47$\pm$0.14 7.42$\pm$0.09 -2.00$\pm$0.14 7.67$\pm$0.10 ----------- --------- -------- ------------------- ----------------------------- ------------------------------ ---------------------------------------- ---------------- ----------- --------- -------- ------------------- ----------------------------- ------------------------------ ---------------------------------------- ---------------- Name [*z*]{} Type Mass SFR (phot) $\Sigma_{M}$ $\Sigma_{\rm SFR}$ $r_{50}$ (log M$_{\odot}$) (log M$_{\odot}$ yr$^{-1}$) (log M$_{\odot}$ kpc$^{-2}$) (log M$_{\odot}$ yr$^{-1}$ kpc$^{-2}$) (kpc) SN 2008iu 0.130 Ic-BL 8.22$\pm$0.28 -0.72$\pm$0.23 8.26$\pm$0.28 -0.68$\pm$0.23 0.63$\pm$1.21 SN 2008iy 0.041 IIn 10.20$\pm$0.03 -0.10$\pm$0.07 18.89$\pm$0.03 8.59$\pm$0.07 0.00$\pm$0.00 SN 2008ja 0.069 IIn 8.38$\pm$0.20 -1.21$\pm$0.21 7.85$\pm$0.20 -1.74$\pm$0.21 1.35$\pm$0.37 SN 2009W 0.017 II P 8.92$\pm$0.32 -0.69$\pm$0.34 7.23$\pm$0.32 -2.39$\pm$0.34 3.06$\pm$0.12 SN 2009bh 0.090 Ic 10.83$\pm$0.10 0.86$\pm$0.12 8.23$\pm$0.10 -1.74$\pm$0.12 8.95$\pm$0.11 SN 2009bj 0.027 II 9.01$\pm$0.23 -1.16$\pm$0.46 8.45$\pm$0.23 -1.72$\pm$0.46 1.49$\pm$0.02 SN 2009bk 0.039 II 9.60$\pm$0.03 0.33$\pm$0.09 7.69$\pm$0.03 -1.58$\pm$0.09 5.14$\pm$0.04 SN 2009bl 0.040 II 9.55$\pm$0.04 0.18$\pm$0.08 7.84$\pm$0.04 -1.53$\pm$0.08 2.98$\pm$0.02 SN 2009ct 0.060 II 10.47$\pm$0.06 0.69$\pm$0.09 8.15$\pm$0.06 -1.64$\pm$0.09 6.98$\pm$0.06 SN 2009dh 0.060 II P 7.99$\pm$0.19 -1.54$\pm$0.25 7.80$\pm$0.19 -1.73$\pm$0.25 0.61$\pm$0.33 SN 2009di 0.130 Ic 8.57$\pm$0.24 -0.47$\pm$0.22 6.98$\pm$0.24 -2.07$\pm$0.22 3.54$\pm$1.26 SN 2009dw 0.042 II P 7.97$\pm$0.27 -1.45$\pm$0.23 6.35$\pm$0.27 -3.07$\pm$0.23 3.51$\pm$1.06 SN 2009fe 0.047 II 10.80$\pm$0.03 -0.09$\pm$0.08 9.59$\pm$0.03 -1.30$\pm$0.08 1.82$\pm$0.02 SN 2009jd 0.025 II 9.40$\pm$0.19 -0.08$\pm$0.20 7.16$\pm$0.19 -2.31$\pm$0.20 5.67$\pm$0.07 SN 2009kf 0.182 II P 9.66$\pm$0.09 0.14$\pm$0.19 7.98$\pm$0.09 -1.54$\pm$0.19 4.10$\pm$0.55 SN 2009lx 0.027 II P 10.45$\pm$0.03 0.51$\pm$0.07 8.37$\pm$0.03 -1.57$\pm$0.07 4.79$\pm$0.03 SN 2009nn 0.046 IIn 9.76$\pm$0.05 0.21$\pm$0.08 7.79$\pm$0.05 -1.76$\pm$0.08 5.20$\pm$0.06 SN 2009nu 0.040 II 9.24$\pm$0.22 -0.56$\pm$0.29 7.02$\pm$0.22 -2.77$\pm$0.29 5.84$\pm$0.28 SN 2010K 0.020 II 8.36$\pm$0.47 -1.10$\pm$0.50 6.98$\pm$0.47 -2.48$\pm$0.50 2.42$\pm$0.14 SN 2010Q 0.055 Ic 7.46$\pm$0.20 -1.18$\pm$0.22 6.80$\pm$0.20 -1.83$\pm$0.22 1.03$\pm$0.40 SN 2010ah 0.050 Ic-BL 8.82$\pm$0.13 -0.85$\pm$0.20 7.67$\pm$0.13 -2.00$\pm$0.20 4.43$\pm$0.27 SN 2010ay 0.067 Ic-BL 8.58$\pm$0.09 0.03$\pm$0.11 8.92$\pm$0.09 0.37$\pm$0.11 0.34$\pm$0.02 SN 2010gq 0.018 II 10.50$\pm$0.03 0.50$\pm$0.07 8.63$\pm$0.03 -1.38$\pm$0.07 4.64$\pm$0.02 SN 2010jc 0.024 II P 10.75$\pm$0.03 0.70$\pm$0.07 7.92$\pm$0.03 -2.13$\pm$0.07 11.07$\pm$0.08 SN 2010jy 0.042 IIn 9.11$\pm$0.12 -0.50$\pm$0.18 7.58$\pm$0.12 -2.04$\pm$0.18 2.73$\pm$0.11 SN 2010mb 0.133 Ic 9.63$\pm$0.07 0.14$\pm$0.16 7.56$\pm$0.07 -1.93$\pm$0.16 8.76$\pm$0.32 SN 2011ak 0.027 II P 10.45$\pm$0.03 0.50$\pm$0.07 7.95$\pm$0.03 -2.00$\pm$0.07 7.73$\pm$0.05 SN 2011an 0.016 IIn 9.61$\pm$0.04 -0.02$\pm$0.07 7.23$\pm$0.04 -2.39$\pm$0.07 7.18$\pm$0.04 SN 2011aw 0.055 Ib/c 1.67$\pm$1.37 SN 2011bm 0.022 Ic 9.45$\pm$0.03 0.10$\pm$0.07 7.88$\pm$0.03 -1.47$\pm$0.07 2.67$\pm$0.01 SN 2011bn 0.031 II 11.20$\pm$0.03 1.20$\pm$0.07 8.80$\pm$0.03 -1.20$\pm$0.07 8.83$\pm$0.06 SN 2011bs 0.036 II 7.44$\pm$0.28 -2.12$\pm$0.26 8.43$\pm$0.28 -1.12$\pm$0.26 0.56$\pm$0.18 SN 2011cl 0.025 II P 8.76$\pm$0.07 SN 2011cq 0.017 II pec 9.69$\pm$0.31 0.13$\pm$0.34 7.41$\pm$0.31 -2.15$\pm$0.34 5.51$\pm$0.03 SN 2011cw 0.040 IIn 0.59$\pm$0.28 SN 2011cz 0.060 II P 8.24$\pm$0.22 -1.24$\pm$0.25 7.32$\pm$0.22 -2.16$\pm$0.25 3.03$\pm$2.16 SN 2011db 0.025 II 9.68$\pm$0.19 0.14$\pm$0.27 7.50$\pm$0.19 -2.04$\pm$0.27 5.19$\pm$0.03 SN 2011en 0.020 II P 8.70$\pm$0.46 -0.48$\pm$0.42 7.29$\pm$0.46 -1.89$\pm$0.42 2.46$\pm$0.05 SN 2011eo 0.030 II P 8.02$\pm$0.30 -1.57$\pm$0.33 7.80$\pm$0.30 -1.78$\pm$0.33 0.62$\pm$0.07 SN 2011eu 0.110 IIn 9.54$\pm$0.15 -0.38$\pm$0.34 9.76$\pm$0.15 -0.17$\pm$0.34 1.39$\pm$0.98 SN 2011ev 0.030 II P 8.99$\pm$0.27 -0.23$\pm$0.33 7.08$\pm$0.27 -2.14$\pm$0.33 7.65$\pm$0.11 SN 2011ew 0.070 II P 10.19$\pm$0.04 -0.80$\pm$0.07 9.92$\pm$0.04 -1.07$\pm$0.07 0.98$\pm$0.78 SN 2011fa 0.060 II P 7.85$\pm$0.23 -1.61$\pm$0.25 7.69$\pm$0.23 -1.76$\pm$0.25 0.57$\pm$0.25 SN 2011fz 0.016 Ib/c 10.55$\pm$0.03 0.60$\pm$0.07 8.00$\pm$0.03 -1.95$\pm$0.07 7.92$\pm$0.06 SN 2011hn 0.014 II P 10.02$\pm$0.31 0.15$\pm$0.33 8.20$\pm$0.31 -1.67$\pm$0.33 6.02$\pm$0.02 SN 2011iw 0.023 IIn 7.92$\pm$0.22 -1.64$\pm$0.23 7.97$\pm$0.22 -1.59$\pm$0.23 0.48$\pm$0.05 SN 2011jb 0.084 IIn 9.20$\pm$0.09 -0.46$\pm$0.18 8.48$\pm$0.09 -1.18$\pm$0.18 1.24$\pm$0.11 SN 2011jj 0.045 II P 10.99$\pm$0.04 1.10$\pm$0.07 8.05$\pm$0.04 -1.84$\pm$0.07 13.98$\pm$0.09 SN 2011ke 0.143 Ic 8.91$\pm$0.15 -0.42$\pm$0.19 6.86$\pm$0.15 -2.47$\pm$0.19 6.77$\pm$1.61 SN 2012D 0.026 II P 9.60$\pm$0.04 0.01$\pm$0.07 8.12$\pm$0.04 -1.48$\pm$0.07 2.57$\pm$0.01 SN 2012F 0.030 Ib 7.86$\pm$0.30 -1.63$\pm$0.36 7.74$\pm$0.30 -1.75$\pm$0.36 0.64$\pm$0.06 SN 2012W 0.018 II 10.05$\pm$0.03 0.30$\pm$0.07 7.70$\pm$0.03 -2.05$\pm$0.07 7.34$\pm$0.03 SN 2012al 0.040 IIn 9.77$\pm$0.17 0.01$\pm$0.26 7.62$\pm$0.17 -2.13$\pm$0.26 6.66$\pm$0.11 SN 2012bg 0.033 II P 8.29$\pm$0.16 -1.30$\pm$0.20 7.95$\pm$0.16 -1.64$\pm$0.20 0.89$\pm$0.06 SN 2012br 0.019 II P 7.40$\pm$0.34 -1.96$\pm$0.37 7.13$\pm$0.34 -2.23$\pm$0.37 0.72$\pm$0.05 ----------- --------- -------- ------------------- ----------------------------- ------------------------------ ---------------------------------------- ---------------- ----------- --------- ------ ------------------- ----------------------------- ------------------------------ ---------------------------------------- --------------- Name [*z*]{} Type Mass SFR (phot) $\Sigma_{M}$ $\Sigma_{\rm SFR}$ $r_{50}$ (log M$_{\odot}$) (log M$_{\odot}$ yr$^{-1}$) (log M$_{\odot}$ kpc$^{-2}$) (log M$_{\odot}$ yr$^{-1}$ kpc$^{-2}$) (kpc) SN 2012ch 0.009 II P 8.30$\pm$0.04 -1.20$\pm$0.07 7.72$\pm$0.04 -1.78$\pm$0.07 1.52$\pm$0.02 SN 2012cr 0.010 II 9.95$\pm$0.03 0.00$\pm$0.07 8.45$\pm$0.03 -1.50$\pm$0.07 3.82$\pm$0.01 SN 2012cz 0.036 IIn 10.35$\pm$0.03 0.30$\pm$0.07 8.66$\pm$0.03 -1.39$\pm$0.07 4.85$\pm$0.02 SN 2012dp 0.036 Ib 10.59$\pm$0.04 0.82$\pm$0.12 8.00$\pm$0.04 -1.76$\pm$0.12 8.58$\pm$0.07 SN 2012ed 0.015 II 7.32$\pm$0.17 -2.01$\pm$0.42 7.32$\pm$0.17 -2.01$\pm$0.42 0.80$\pm$0.12 SN 2012ex 0.023 Ib 10.45$\pm$0.03 0.70$\pm$0.07 8.48$\pm$0.03 -1.27$\pm$0.07 4.07$\pm$0.01 SN 2012fc 0.023 II P 9.80$\pm$0.03 0.29$\pm$0.07 8.19$\pm$0.03 -1.32$\pm$0.07 3.82$\pm$0.02 SN 2012hw 0.038 II P 9.90$\pm$0.11 0.19$\pm$0.15 7.49$\pm$0.11 -2.22$\pm$0.15 6.59$\pm$0.05 SN 2012il 0.175 Ic 9.03$\pm$0.16 -0.59$\pm$0.20 8.18$\pm$0.16 -1.44$\pm$0.20 1.93$\pm$0.63 SN 2013an 0.014 II 0.32$\pm$0.25 SN 2013aw 0.027 II 9.45$\pm$0.19 -0.30$\pm$0.24 7.80$\pm$0.19 -1.95$\pm$0.24 6.63$\pm$0.06 SN 2013bn 0.054 Ic 0.78$\pm$0.13 SN 2013br 0.074 II 9.15$\pm$0.10 -0.47$\pm$0.19 8.14$\pm$0.10 -1.48$\pm$0.19 1.73$\pm$0.22 SN 2013bw 0.038 II P 10.60$\pm$0.03 1.10$\pm$0.07 8.15$\pm$0.03 -1.35$\pm$0.07 7.61$\pm$0.03 SN 2013dn 0.056 IIn 10.70$\pm$0.03 1.20$\pm$0.07 8.59$\pm$0.03 -0.91$\pm$0.07 5.50$\pm$0.06 ----------- --------- ------ ------------------- ----------------------------- ------------------------------ ---------------------------------------- --------------- ----------- --------- ------- ------------------- -------------------- ----------------------------- ---------------------------------------- --------------- Name [*z*]{} Type Mass $\sigma_{\rm vel}$ SFR (spec) $\Sigma_{\rm SFR}$ Fraction deV. (log M$_{\odot}$) (km s$^{-1}$) (log M$_{\odot}$ yr$^{-1}$) (log M$_{\odot}$ yr$^{-1}$ kpc$^{-2}$) PTF 09awk 0.062 Ib 9.54$\pm$0.06 63.7$\pm$0.7 -0.03$\pm$0.14 -1.00$\pm$0.14 1.0 PTF 09dra 0.077 II 10.50$\pm$0.08 77.1$\pm$2.6 0.30$\pm$0.33 -2.46$\pm$0.33 0.5 PTF 09ige 0.064 II 9.73$\pm$0.06 44.2$\pm$1.2 0.21$\pm$0.25 -1.93$\pm$0.25 0.2 PTF 09ism 0.029 II 9.12$\pm$0.21 33.2$\pm$3.1 -0.68$\pm$0.40 -2.42$\pm$0.40 0.2 PTF 09sk 0.036 Ic-BL 8.93$\pm$0.15 51.4$\pm$0.7 -0.40$\pm$0.22 -1.48$\pm$0.22 0.5 PTF 09uj 0.065 II 9.75$\pm$0.08 52.7$\pm$3.4 -0.08$\pm$0.27 -2.17$\pm$0.27 0.0 PTF 10bau 0.026 II 10.75$\pm$0.03 80.8$\pm$1.1 0.41$\pm$0.22 -1.62$\pm$0.22 0.2 PTF 10bhu 0.036 Ic 9.43$\pm$0.14 51.0$\pm$2.1 -0.27$\pm$0.29 -1.92$\pm$0.29 0.0 PTF 10con 0.033 II 9.68$\pm$0.16 68.6$\pm$2.3 -0.60$\pm$0.30 -1.96$\pm$0.30 0.1 PTF 10cxx 0.034 II 10.03$\pm$0.13 68.2$\pm$1.0 -0.07$\pm$0.18 -1.46$\pm$0.18 0.2 PTF 10s 0.051 II 9.66$\pm$0.09 38.7$\pm$1.3 -0.25$\pm$0.24 -1.90$\pm$0.24 0.0 PTF 11cgx 0.034 II 9.95$\pm$0.04 57.6$\pm$1.6 -0.00$\pm$0.26 -1.75$\pm$0.26 0.0 PTF 11cwi 0.056 II 10.58$\pm$0.09 110.9$\pm$1.2 0.70$\pm$0.13 -1.33$\pm$0.13 1.0 PTF 11dqk 0.036 II 9.81$\pm$0.04 42.5$\pm$0.8 0.35$\pm$0.23 -1.56$\pm$0.23 0.1 PTF 11dtd 0.040 II 10.38$\pm$0.04 69.0$\pm$3.6 -0.39$\pm$0.78 -2.89$\pm$0.78 0.0 PTF 11ecp 0.034 II 10.15$\pm$0.12 53.9$\pm$1.3 0.25$\pm$0.30 -1.71$\pm$0.30 0.0 PTF 11gdz 0.013 II 9.89$\pm$0.29 68.0$\pm$0.5 -0.34$\pm$0.13 -1.36$\pm$0.13 0.1 PTF 11jgp 0.072 II 10.02$\pm$0.06 55.0$\pm$1.3 0.28$\pm$0.25 -1.70$\pm$0.25 0.0 PTF 11mpv 0.043 II 9.27$\pm$0.11 41.1$\pm$1.4 -0.47$\pm$0.22 -1.56$\pm$0.22 0.4 PTF 11qcm 0.051 II 10.65$\pm$0.09 83.7$\pm$2.2 0.01$\pm$0.32 -2.19$\pm$0.32 0.0 PTF 11qju 0.028 II 9.35$\pm$0.19 37.6$\pm$1.3 -0.32$\pm$0.23 -2.22$\pm$0.23 0.0 PTF 11qux 0.041 II 9.58$\pm$0.11 48.0$\pm$0.7 -0.14$\pm$0.13 -1.23$\pm$0.13 0.0 PTF 12cgb 0.026 II 9.29$\pm$0.20 49.3$\pm$0.8 -0.30$\pm$0.27 -1.58$\pm$0.27 0.1 PTF 12dke 0.067 II 9.83$\pm$0.09 57.7$\pm$3.0 0.18$\pm$0.30 -2.32$\pm$0.30 0.1 PTF 12eje 0.078 II 9.86$\pm$0.09 49.8$\pm$1.5 -0.64$\pm$0.16 -2.75$\pm$0.16 0.0 PTF 12gcx 0.045 II 9.67$\pm$0.16 39.4$\pm$1.2 -0.96$\pm$0.35 -3.28$\pm$0.35 0.1 PTF 12gzk 0.014 12gzk 7.29$\pm$0.09 59.0$\pm$0.7 -2.40$\pm$0.09 -2.45$\pm$0.09 1.0 PTF 13c 0.011 II 8.68$\pm$0.30 49.4$\pm$1.4 -0.68$\pm$0.33 -1.72$\pm$0.33 0.0 PTF 13cbf 0.040 Ic 9.51$\pm$0.14 47.0$\pm$0.4 0.26$\pm$0.20 -1.44$\pm$0.20 0.2 SN 2004hy 0.058 II 9.69$\pm$0.11 46.4$\pm$4.4 -0.16$\pm$0.31 -2.54$\pm$0.31 0.2 SN 2005hl 0.023 Ib 10.43$\pm$0.18 63.8$\pm$1.1 0.16$\pm$0.22 -1.46$\pm$0.22 0.0 SN 2005kb 0.015 II 9.14$\pm$0.38 28.6$\pm$1.4 -1.00$\pm$0.30 -2.16$\pm$0.30 0.0 SN 2005ks 0.099 Ic-BL 9.89$\pm$0.07 70.4$\pm$1.4 -0.10$\pm$0.24 -1.53$\pm$0.24 0.0 SN 2005lc 0.014 II 8.50$\pm$0.34 34.8$\pm$3.1 -1.53$\pm$0.27 -2.59$\pm$0.27 0.1 SN 2005lm 0.085 II 9.33$\pm$0.08 43.7$\pm$1.1 -0.60$\pm$0.08 -1.87$\pm$0.08 0.3 SN 2005mn 0.047 Ib 9.52$\pm$0.13 44.2$\pm$2.6 -0.17$\pm$0.30 -2.27$\pm$0.30 0.2 SN 2006M 0.015 IIn 8.86$\pm$0.29 43.9$\pm$2.1 -0.64$\pm$0.24 -1.96$\pm$0.24 0.1 SN 2006bj 0.038 II 9.40$\pm$0.14 43.9$\pm$2.9 -1.57$\pm$0.23 -3.63$\pm$0.23 0.2 SN 2006cw 0.061 II 9.87$\pm$0.08 49.7$\pm$1.6 0.13$\pm$0.27 -2.02$\pm$0.27 0.0 SN 2006db 0.023 IIn 8.69$\pm$0.21 46.4$\pm$3.0 -0.92$\pm$0.27 -2.21$\pm$0.27 0.2 SN 2006fo 0.021 Ib 10.65$\pm$0.03 61.3$\pm$0.8 0.05$\pm$0.24 -1.67$\pm$0.24 0.1 SN 2006fq 0.068 II P 10.04$\pm$0.04 46.1$\pm$0.4 0.32$\pm$0.25 -1.48$\pm$0.25 0.1 SN 2006gd 0.155 II P 11.02$\pm$0.06 101.9$\pm$7.8 -0.33$\pm$1.00 -3.10$\pm$1.00 0.6 SN 2006iw 0.031 II 9.73$\pm$0.15 34.2$\pm$2.0 -0.52$\pm$0.36 -2.15$\pm$0.36 0.0 SN 2006ix 0.076 II 9.57$\pm$0.08 48.3$\pm$2.5 -1.01$\pm$0.15 -2.81$\pm$0.15 0.1 SN 2006kh 0.060 II 9.47$\pm$0.10 65.5$\pm$1.0 -0.70$\pm$0.10 -1.18$\pm$0.10 0.3 SN 2006kn 0.120 II 10.08$\pm$0.16 51.6$\pm$3.4 -0.30$\pm$0.39 -2.52$\pm$0.39 0.0 SN 2006ns 0.120 II 9.88$\pm$0.06 43.4$\pm$1.8 0.33$\pm$0.26 -1.71$\pm$0.26 0.2 SN 2006qk 0.058 Ic-BL 9.55$\pm$0.11 57.4$\pm$1.4 -0.60$\pm$0.13 -1.30$\pm$0.13 0.1 SN 2007I 0.022 Ic-BL 8.82$\pm$0.20 42.7$\pm$2.9 -1.07$\pm$0.29 -2.20$\pm$0.29 0.0 SN 2007bo 0.044 II 9.11$\pm$0.12 41.6$\pm$2.5 -0.44$\pm$0.35 -2.29$\pm$0.35 0.1 SN 2007bp 0.028 II 10.60$\pm$0.04 46.0$\pm$4.3 -1.61$\pm$1.11 -4.36$\pm$1.11 0.4 SN 2007dp 0.033 II 9.19$\pm$0.14 43.8$\pm$1.6 -0.62$\pm$0.27 -1.82$\pm$0.27 0.0 SN 2007fe 0.033 II 9.59$\pm$0.17 53.1$\pm$1.2 -0.33$\pm$0.23 -1.92$\pm$0.23 0.3 SN 2007fg 0.026 II 8.84$\pm$0.19 37.7$\pm$0.8 -0.60$\pm$0.21 -2.22$\pm$0.21 0.0 ----------- --------- ------- ------------------- -------------------- ----------------------------- ---------------------------------------- --------------- ----------- --------- -------- ------------------- -------------------- ----------------------------- ---------------------------------------- --------------- Name [*z*]{} Type Mass $\sigma_{\rm vel}$ SFR (spec) $\Sigma_{\rm SFR}$ Fraction deV. (log M$_{\odot}$) (km s$^{-1}$) (log M$_{\odot}$ yr$^{-1}$) (log M$_{\odot}$ yr$^{-1}$ kpc$^{-2}$) SN 2007fy 0.045 II 10.03$\pm$0.11 59.2$\pm$0.8 0.27$\pm$0.25 -1.61$\pm$0.25 0.1 SN 2007ib 0.034 II 10.00$\pm$0.14 57.1$\pm$0.9 0.12$\pm$0.25 -1.84$\pm$0.25 0.1 SN 2007jf 0.070 II P 9.56$\pm$0.10 44.6$\pm$2.2 -1.07$\pm$0.12 -2.94$\pm$0.12 0.4 SN 2007jm 0.091 II n 9.70$\pm$0.10 56.9$\pm$1.8 -0.55$\pm$0.13 -2.00$\pm$0.13 0.3 SN 2007ky 0.074 II 11.03$\pm$0.07 79.4$\pm$4.4 0.47$\pm$0.45 -2.41$\pm$0.45 0.6 SN 2007lx 0.058 II 10.72$\pm$0.10 69.2$\pm$2.7 0.79$\pm$0.33 -1.46$\pm$0.33 0.8 SN 2007nw 0.057 II P 10.26$\pm$0.10 74.9$\pm$1.8 0.09$\pm$0.33 -1.93$\pm$0.33 0.6 SN 2007qb 0.079 II 10.27$\pm$0.05 70.2$\pm$1.3 0.67$\pm$0.26 -2.06$\pm$0.26 0.6 SN 2007qw 0.151 Ic-BL 9.36$\pm$0.08 49.1$\pm$1.0 -0.09$\pm$0.08 -1.18$\pm$0.08 0.7 SN 2008bj 0.019 II 8.49$\pm$0.32 27.3$\pm$2.5 -0.80$\pm$0.23 -2.42$\pm$0.23 0.0 SN 2008fo 0.030 Ic 9.49$\pm$0.20 52.2$\pm$0.8 0.03$\pm$0.22 -1.76$\pm$0.22 0.0 SN 2008gd 0.059 II 9.89$\pm$0.09 41.3$\pm$3.4 -0.01$\pm$0.34 -2.48$\pm$0.34 0.2 SN 2009bk 0.039 II 9.80$\pm$0.15 44.8$\pm$1.4 0.16$\pm$0.31 -1.74$\pm$0.31 0.0 SN 2009bl 0.040 II 9.85$\pm$0.04 54.4$\pm$2.2 0.14$\pm$0.27 -1.58$\pm$0.27 0.0 SN 2009ct 0.057 II 10.58$\pm$0.08 97.4$\pm$3.1 0.42$\pm$0.34 -1.85$\pm$0.34 0.6 SN 2010ay 0.067 Ic-BL 8.55$\pm$0.09 61.3$\pm$0.6 -0.01$\pm$0.12 0.32$\pm$0.12 1.0 SN 2011bm 0.022 Ic 9.85$\pm$0.04 57.1$\pm$0.8 -0.04$\pm$0.23 -1.63$\pm$0.23 0.3 SN 2011cq 0.017 II pec 9.97$\pm$0.33 67.9$\pm$2.4 -0.65$\pm$0.42 -2.94$\pm$0.42 0.4 SN 2011en 0.018 II P 8.51$\pm$0.32 12.7$\pm$7.5 -1.07$\pm$0.24 -2.37$\pm$0.24 0.1 SN 2011hn 0.014 II P 9.95$\pm$0.04 12.6$\pm$2.8 -0.52$\pm$0.41 -2.35$\pm$0.41 0.2 SN 2011jm 0.003 Ic 8.82$\pm$0.06 31.9$\pm$0.3 -2.51$\pm$0.15 -3.23$\pm$0.15 0.4 SN 2012D 0.026 II P 9.65$\pm$0.03 47.4$\pm$0.5 0.14$\pm$0.21 -1.34$\pm$0.21 0.0 SN 2012al 0.038 IIn 9.55$\pm$0.17 34.8$\pm$3.9 -1.49$\pm$1.01 -3.59$\pm$1.01 0.0 SN 2012dp 0.036 Ib 10.55$\pm$0.03 85.4$\pm$0.8 0.56$\pm$0.17 -2.02$\pm$0.17 0.9 SN 2012ex 0.023 Ib 8.77$\pm$0.37 21.2$\pm$0.8 -1.49$\pm$0.11 -3.58$\pm$0.11 0.3 SN 2012hw 0.038 II P 10.06$\pm$0.13 39.4$\pm$2.4 0.05$\pm$0.31 -2.36$\pm$0.31 0.0 SN 2013aw 0.027 II 9.37$\pm$0.21 40.3$\pm$1.2 -0.57$\pm$0.26 -2.21$\pm$0.26 0.1 ----------- --------- -------- ------------------- -------------------- ----------------------------- ---------------------------------------- --------------- [^1]: http://www.grbhosts.org

ArXiv

--- abstract: 'Many machine learning models have important structural tuning parameters that cannot be directly estimated from the data. The common tactic for setting these parameters is to use resampling methods, such as cross–validation or the bootstrap, to evaluate a candidate set of values and choose the best based on some pre–defined criterion. Unfortunately, this process can be time consuming. However, the model tuning process can be streamlined by adaptively resampling candidate values so that settings that are clearly sub-optimal can be discarded. The notion of [*futility analysis*]{} is introduced in this context. An example is shown that illustrates how adaptive resampling can be used to reduce training time. Simulation studies are used to understand how the potential speed–up is affected by parallel processing techniques.' author: - | Max Kuhn ([max.kuhn@pfizer.com]{})\ Nonclinical Statistics\ Pfizer Global R$\&$D\ Groton, CT 06340, USA bibliography: - 'adaptive.bib' title: 'Futility Analysis in the Cross–Validation of Machine Learning Models' --- [**Keywords**]{}: Predictive Modeling, Adaptive Resampling, Bootstrapping, Support Vector Machine, Neural Networks, Parallel Computations Introduction {#S:intro} ============ Machine learning [@Bishop2007] uses past data to make accurate predictions of future events or samples. In comparison to traditional inferential statistical techniques, machine learning models tend to be more complex black–box models that are created to maximize predictive accuracy [@Breiman:2001wl]. The process to develop these models is usually very data–driven and focuses on performance statistics calculated from external data sources. The effectiveness of a model is commonly measured using a single statistic which we will refer to here as [*model fitness*]{}. For models predicting a numeric outcome, the fitness statistic might be the root mean squared error (RMSE) or the coefficient of determination ($R^2$). For classification, where a categorical outcome is being predicted, the error rate might be an appropriate measure. When creating machine learning models, there are often [*tuning parameters*]{} or [*hyper parameters*]{} to optimize. These are typically associated with structural components of a model that cannot be directly estimated from the data. For example: - A $K$–nearest neighbor classifies new samples by first finding the $K$ closest samples in the training set and determines the predicted value based on the known outcomes of the nearest neighbors. In this model, $K$ (and possibly the distance metric) are tuning parameters. - In classification and regression trees, the depth of the tree must be determined. Many tree–based models first grow a tree to the maximum size then prune tree back to avoid over–fitting. @cart use cost–complexity pruning for this purpose and parameterize this setting in terms of the complexity value $c_p$. @quinlan1993c4 uses an alternative pruning strategy based on uncertainty estimates of the error rate. In this case, the confidence factor (CF) is a tuning parameter that determines the depth of the tree. - Partial least squares models [@wold2001pls] utilize the data in terms of latent variables called PLS components. The number of components must be determined before the final PLS model can be used. In many cases, the values of the tuning parameters can have a profound effect on model efficacy. @Delgado14 describe a study where a large number of classifiers were evaluated over a wide variety of machine learning benchmark data sets. Their findings validate the importance of model tuning. Despite their importance, it is rare that reasonable values of these parameters are known [*a priori*]{}. To determine appropriate values of the tuning parameters, one approach is to use some form of resampling to estimate how well the model performs on the training set [@apm]. Cross–validation, the bootstrap or variations of these are commonly used for this purpose. A single iteration of resampling involves determining a subset of training set points that are used to fit the model and a separate “holdout” set of samples to estimate model fitness. This process is repeated many times and the performance estimates from each holdout set are averaged into a final overall estimate of model efficacy. There are different types of resampling methods, the most common types being: $k$–fold cross–validation, repeated $k$–fold cross–validation, leave–one–out cross–validation, Monte Carlo cross–validation and the bootstrap [@apm]. Each resampling scheme has its own variance and bias properties. For example, the bootstrap tends to have small variance but substantial bias while traditional $k$–fold cross–validation has small bias but high uncertainty (depending on $k$). Recent research [@Molinaro:2005p47; @Kim:2009im] suggests that repeating $k$–fold cross–validation is advisable based on having acceptable variance and bias in comparison to the other approaches. Denote the training data as $D$. If there are $B$ resamples, denote the resampled version of the data as $R_{i}$ and the holdout set induced by resampling as $T_{i}$, where $i = 1\ldots B$. The complete set of tuning parameters is symbolized by $\Theta$ and an individual candidate set of tuning parameters by $\theta_j$ with $j=1\ldots p$. Some models have multiple tuning parameters and, in these cases, $\theta$ is vector valued. From each resample and candidate parameter set, let the fitted model be $\hat{f}_{ij}(R_{i};\theta_j)$ and the resulting estimate of model fitness be denoted as $Q_{ij}$. The resampled estimate of fitness for each parameter set is $\hat{Q}_j = 1/B\sum_{i=1}^B Q_{ij}$. The grid search strategy outlined in Algorithm \[A:Resamp\] is one possible approach for optimizing the tuning parameters. First, the set of candidate values $\Theta$ is determined along with the type of resampling and the number of data splits. For each parameter combination, the model fitness is estimated via resampling and the relationship between the tuning parameters and model performance is characterized. From this, a rule for choosing $\theta_{opt}$, based on the resampling profile, is needed. The choice can be made based on the empirically best result or by some other process, such as the one–standard error rule of @cart. After choosing $\theta_{opt}$, one final model $\hat{f}(D;\theta_{opt})$ is created using the optimized settings and the entire training set $D$. Grid search is not the only approach that can be used to optimize tuning parameters. For example, @Ustun:2005ew used evolutionary search procedures to optimize the performance of a support vector machine regression model. \[A:Resamp\] Define parameter set $\Theta$ Calculate $\hat{Q}_1\ldots \hat{Q}_p$ Determine $\theta_{opt}$ Fit the final model $\hat{f}(D; \theta_{opt})$ @krstajic2014cross discuss resampling in the context of model tuning and describe potential pitfalls. They review historical publications and also differentiate between [*cross–validatory choice*]{} and [*cross–validatory assessment*]{}. The goal of the former is to choose between sub–models (e.g. a 3–nearest neighbor versus 5–nearest neighbor model). The latter is focused on an accurate assessment of a single model. In some cases, both are important and this manuscript focuses on choosing a model and estimating performance to an acceptable level of precision. To illustrate the process of model tuning, a data set for predicting whether a chemical compound will damage an organism’s genetic material, otherwise known as [*mutagenicity*]{}, was used [@Kazius:2005up]. They labeled 4335 compounds as either a mutagen or non–mutagen. We generated 830 descriptors of molecular structure [@Leach:2007tq] for each compound and used these as predictors of mutagenicity. Examples of the descriptors used in these analyses are atom counts, molecular weight, surface area and other measures of size and charge. Using a predictive model, future compounds can be assessed for their potential toxicity based on these properties. A support vector machine (SVM) classification model with a radial basis function kernel [@vapnik2010nature] is used to illustrate parameter tuning. There are two tuning parameters: the radial basis function scale parameter $\sigma$ and the cost value associated with the support vectors. However, @Caputo describe an analytical formula to estimate $\sigma$ from the training set and this method was used to eliminate $\sigma$ from the tuning grid. As a result, $\Theta$ is one dimensional and the candidate set of cost values consisted of 21 settings ranging from 0.25 to 256 on the log$_2$ scale, i.e. $\Theta = \left\{2^{-2}, 2^{-1.5} \ldots, 2^{8}\right\}$. Simple bootstrap resampling [@Efron:1983ul] was used to tune the model where, on average, the number of samples held out at each iteration of resampling was 1598. To evaluate how well the model performed within each resampling iteration, an ROC curve [@Altman:1994uv; @Fawcett:2006gr; @Brown:2006wp] is created by applying $\hat{f}_{ij}(R_i, \theta_j)$ to $T_i$. The area under the ROC curve is then used to quantify model fitness. The results of this process are shown in Figure \[F:qsar\_svm\_cv\]. When the cost value is small, the model has poor performance due to under–fitting. After a peak in performance is reached, the model begins to become too complex and over–fit. A simple “pick–the–winner” strategy would select a model with a cost value of $2^{1.5}$. This sub–model is associated with an area under the curve of 0.901. With the complexity of the model now determined, the final SVM model is created with this value, the estimate of $\sigma$ and the entire training set. ![The relationship between the SVM tuning parameter and the cross–validated estimate of the area under the ROC curve. Each point is the average of 50 estimates from the resampled ROC curves. Using a “pick–the–winner” strategy, $\theta_{opt}=2^{1.5}$.[]{data-label="F:qsar_svm_cv"}](qsar_full_svm){width=".8\textwidth"} One issue with this approach is that all model parameters are treated with equal priority even after substantial evidence is available. For example, for these data, it is very clear that $\theta=2^{8}$ is highly unlikely to yield optimal results. Despite this, all $B\times p$ models are created before the the relationship between the parameters and the outcome are considered. This leads to excessive computations and, for large data sets and/or computationally expensive models, this can drastically slow down the tuning process. Additionally, some feature selection techniques are [*wrappers*]{} around machine learning models [@Kohavi:1997wca] that search for small combinations of predictors that optimize performance. If a model has important tuning parameters, this resampling process might occur at each step in the subset selection search. In this situation, increasing the computational efficiency can have a major impact on the overall time to develop a model. The remainder of the manuscript outlines an adaptive resampling scheme that can be used to find acceptable values of the tuning parameters with fewer model fits. The mutagenicity data are further analyzed with the proposed methodology. Simulation studies are also used to characterize the efficacy and efficiency of these approaches. Finally, the effect of parallel computations on adaptive resampling techniques are studied. Adaptive Resampling via Futility Analysis ========================================= As resampling progresses, there may be some parameter values that are unlikely to be chosen as the optimal settings. Our goal is to identify these as early as possible so that unneeded computations can be avoided. The situation is somewhat similar to futility analysis in clinical trials [@Lachin:2005uu]. Whereas futility in clinical investigations is defined as “the inability of the trial to achieve its objectives” [@Snapinn:2006ih], we might consider a tuning parameter as futile if it is unlikely to have optimal performance. However, there are several differences between classical futility analysis and parameter tuning. Most clinical trials involve a small number of pre–planned comparisons and the trial is usually designed to have substantial power to detect pre–specified effect sizes. In our case, there may be a large number of parameter settings and thus many comparisons. Also, prior to model building, there may be little [*a priori*]{} knowledge of how well the model may perform. For this reason, the understanding of a meaningful difference in the model fitness values may not be known. Given these differences, it is unlikely that existing statistical techniques for clinical futility analysis can be employed for this particular problem. However, the general concept of futility is still applicable. Algorithm \[A:adaptive\] outlines how a futility assessment can be incorporated into the model tuning process. The nominal tuning process is used for the first $B_{min}$ iterations of resampling. At $B_{min}$, the fitness values $Q_{ij}$ are used to assess futility for each tuning parameter. Based on these results, a pre–defined rule is used to determine which values in $\Theta$ are unlikely to be the best and these parameter settings are removed from further consideration (and no longer resampled). The number of tuning parameter settings at each stage of resampling is denoted as $p_i$. If $p_i > 1$, the resampling process continues until either $p_i = 1$ or the maximum number of resamples is reached ($i = B$). In the former case, the resampling process would continue only for $\theta_{opt}$ to measure the fitness value to the maximum precision. In the latter cases, the nominal selection process is used to determine $\theta_{opt}$ from the parameters still under consideration (i.e. line 11 in Algorithm \[A:Resamp\]). \[A:adaptive\] Define parameter set $\Theta$ Fit the final model $\hat{f}(D; \theta_{opt})$ The pivotal detail in Algorithm \[A:adaptive\] is the method for estimating futility. In the next sub–sections, two approaches are considered in detail. One technique uses the fitness values $Q_{ij}$ to measure futility while the other uses dichotomized “scores”. In these descriptions we assume, without loss of generality, that larger values of the fitness statistic are better. Measuring Futility via Linear Models {#S:HT} ------------------------------------ @Shen:2011vi describe a method for assessing futility during resampling. At resampling iteration $i$, they contrast sub–models by treating the performance estimates resulting from each split as a blocked experiment and fit a linear model $$Q_{kj} = \mu + \tau_j + \beta_k + \epsilon_{kj}$$ where $\mu$ is the grand mean, $\beta_k$ is the effect of the $k^{th}$ resample ($k = 1\ldots i$), $\tau_j$ is the effect of the sub–model ($j = 1\ldots p_i$) and $\epsilon_{kj}$ are the errors, assumed to be iid $N(0, \sigma^2)$. The interest is in comparing sub–models via statistical hypothesis tests on the $\tau_j$. There is a strong likelihood that the resampled fitness values have appreciable [*within–resample*]{} correlations where fitness values resulting from one data split tend to have a higher correlation with one another when compared to fitness values generated using a different split of the data. If this factor is not taken into account, it is likely that any inferential statements made about different parameter values may be underpowered or inaccurate. Rather than estimating the within–resample correlation, the block parameter in their model is used to account for the resample–to–resample effect in the data. @Shen:2011vi focused on testing $H_0: \tau_j = \tau_{j'}$ versus $H_1: \tau_j \ne \tau_{j'}$ for all $j \ne j'$ and suggests removing all “dominated models” from further evaluation. They also used multiple comparison corrections to account for repeated testing. The confidence level $\alpha$ is a tuning parameter for Algorithm \[A:adaptive\] and can be used to control the greediness of the adaptive procedure. In this manuscript, a variation of this approach is proposed. First, we suggest using one–sided hypotheses where the current best setting is determined and the other sub–models are compared to this setting. This should improve the power of the comparisons. Secondly, our approach does not attempt to compensate for multiple testing. The confidence level controls the aggressiveness that the algorithm will eliminate sub–models. The family–wise error rate guards against any false positive findings which, in this context, has limited relevance. Additionally, @Shen:2011vi cast doubt that, after correction, the nominal significance level under the null hypothesis is really $\alpha$ when being used in this manner. Finally, instead of blocking on the splits, our model directly estimates the within–resample correlation. Specifically, at iteration $k$, we model $$\label{E:delta} Q_{kj} = \mu + \tau_j + \epsilon_{kj}$$ where the errors are assumed to have a normal distribution with mean zero and a block diagonal covariance matrix $\Sigma$. The blocks are of size $p_i \times p_i$ and defined as $\Sigma_k = \sigma^2 (1-\rho) I_{p_i} + \sigma^2\rho J_{p_i}$. This is an exchangeable (or compound–symmetric) covariance structure where $$Cov[\epsilon_{kj}, \epsilon_{k'j}] = \begin{cases} 0 & \text{if } k \ne k' \\ \sigma^2_r & \text{if } k = k' \end{cases}$$ In this way, the within–resample correlation is estimated to be $\rho = \sigma^2/(\sigma^2+\sigma^2_r)$. This model can be fit via generalized least squares [@Vones:Chinc:97] to estimate the effects of the tuning parameters via the $\tau_j$ and the two variance parameters. The model in Equation \[E:delta\] contains $p_k-1$ slope parameters $\tau_j$. The “reference cell” in this model is the current best condition at the $k^{th}$ iteration and, using this parameterization, the $\tau_j$ estimate the loss of performance for parameter setting $\theta_j$ from the current numerically optimal condition. One–sided $(1 - \alpha)$$\%$ confidence intervals can be constructed for the $\tau_j$. If the interval contains zero, this would be equivalent to failing to reject $H_0: \tau_j = 0$ versus $H_1: \tau_j > 0$. Rejection of this null indicates that the average performance of model $j$ is statistically worse than the current best model. At each iteration of resampling, any tuning parameter setting whose interval does not contain zero is removed and is not evaluated on subsequent iterations. Like Shen’s model, the confidence level controls the rate at which sub–models are discarded. Returning to the mutagenicity example, a first evaluation of the SVM sub–models occurred after $i = 10$ splits. At the this point, the SVM model with the largest mean area under the ROC curve was $\theta = 2^{ 2 }$. The performance profile after 10 resamples was very similar to the one shown in Figure \[F:qsar\_svm\_cv\]. Fixing this sub–model as the reference, $\Delta_{ij}$ values were computed for ($i \leq 10$) and these values were used in Equation \[E:delta\]. From this model, $\widehat{\rho} = 0.34$ and $\widehat{\sigma} = 0.0043$. Using $\alpha = 0.01$, there were 6 sub–models whose intervals included zero: $2^{ 0 }$, $2^{ 0.5 }$, $2^{ 1 }$, $2^{ 1.5 }$, $2^{ 2.5 }$, $2^{ 3 }$. Therefore, the next iteration of resampling would only evaluate 7 sub–models. After this evaluation, several more models were removed: at $i = 10$, 15 models were removed, at $i = 11$, 2 models were removed and a single models were removed at $i = 13$ and $i = 14$. There were 2 surviving values of $\theta$ at $B = 50$: $2^{ 1.5 }$ and $2^{ 2 }$. The usual pick–the–winner strategy was used here to select $\theta = 2^{ 1.5 }$. The potential advantage of adaptive sampling can quantified by the [*speed–up*]{}, calculated as the total execution time for tuning the model with the complete set of resamples divided by the execution time of the adaptive procedure. For example, a speed–up of 1.5 is a fifty percent decrease in the execution time when using the adaptive procedure. For this approach and these data, a speed–up of 3.5 was achieved. This adaptive process fit 299 SVM models or 28.5$\%$ of the number required for the full set of resamples and resulted in the same choice of the SVM cost parameter. Note that, for some models, there can be a multiple tuning parameters and this can lead to a large number of distinct combinations. The data used to conduct the futility analysis is driven by the number of model parameters ($p_i$) and the current number of resamples ($i$). It is possible that, when $p_i$ is large and $i$ is small, the generalized linear model will be over–determined and/or inestimable. Also, the assumption of normality of the residuals may be unrealistic since a highly accurate machine learning model might generate resampled performance estimates that are skewed. For example, as the area under the ROC curve approaches unity, its resampling distribution can become significantly left–skewed. Since the generalized linear model is estimating multiple variance parameters, non–normal residuals can have a profoundly adverse affect on those estimates. Bradley–Terry Models to Estimate Futility {#S:BT} ----------------------------------------- The approach shown here to measure futility is based on @Consensus, who developed models to create consensus rankings of different models based on resampled performance statistics. We modify their method to characterize the differences of tuning parameters [*within a model*]{}. More recently, @Eugster:2013hl used similar methods to characterize differences between models and across different data sets. For our purposes, the resampling data generated during the tuning process can be decomposed into a set of win/loss/tie comparisons based on the resampled performance estimates. To compare settings $\theta_j$ and $\theta_{j'}$ at resampling iteration $i$, the number of wins for $\theta_j$ is the sum of the resamples where $Q_{ij} > Q_{ij'}$. The converse is also true for the number of wins for $\theta_{j'}$. Ties are handled as a half–win for each team. Given a set of pair–wise win/loss/tie statistics, the Bradley–Terry model [@bradley1952rank] is a logistic regression model where the outcome is $$logit[Pr(Q_{ij} > Q_{ij'})] = \lambda_j - \lambda_{j'}.$$ The $\lambda$ parameters are estimated via maximum likelihood estimation in the usual manner. In the context of evaluating tuning parameter combinations, the estimated contrasts $\widehat{\lambda}_j - \widehat{\lambda}_{j'}$ can be interpreted as the log–odds that tuning parameter $\theta_j$ has a better [*ability*]{} to win compared to the reference setting of $\theta_{j'}$. Our approach is to use the parameter associated with the best average fitness value as the reference setting. The consequence of this choice is that most of the differences $\widehat{\lambda}_j - \widehat{\lambda}_{j'}$ will be negative and larger values indicate performance that is closer to the current best setting. It may be possible that one or more sub–models have no wins against any other sub–model. The consequence of this situation is that the ability estimates become degenerate and their associated standard errors can be orders of magnitude larger than is reasonable. To avoid this, these cases are removed from the data prior to fitting the logistic regression model and the corresponding sub–models are not considered in the remaining iterations of resampling. To use this approach for comparing tuning parameters, the win/loss/tie data can be used to fit the Bradley–Terry model. Similar to the approach in the previous section, one–sided (asymptotic) confidence bounds for the ability values can be calculated and used to winnow values in $\Theta$. These intervals are asymptotic and use a normal quantile $\Phi^{-1}(1-\alpha)$. The intervals produced by generalized least squares are not asymptotic and use a similar quantile of a $t$–distribution. Analogous to the linear model approach, any tuning parameter settings whose upper bound is not greater than zero would be eliminated from further consideration. This process is repeated at each resampling iteration after $B_{min}$ until either a single parameter setting remains or the maximum number of resamples is reached. For the previously trained SVM model, suppose the first futility analysis was also conducted at $B_{min} = 10$. From these data, $ \binom{21}{2} = 210$ sets of ten “tournaments” were played between pairs of competing sub–models. The area under the ROC curve is used to compute win/loss/tie scores. For example, the model $\theta=2^{2}$ has 8 wins and 2 losses against $\theta=2^{3}$. Again, $\theta = 2^{ 2 }$ was designated as the reference model, the Bradley–Terry model was computed. Figure \[F:qsar\_bt\] shows the estimates of the ability scores for each setting along with their corresponding 95$\%$ one–sided intervals. Note that the model with the best average area under the ROC curve did not have the highest ability, illustrating the difference in average fitness values and dichotomized “competitions” between models. Including the reference sub–model, the conditions that survived the filtering process were $\theta = \{2^{ 0.5 }, 2^{ 1 }, 2^{ 1.5 }, 2^{ 2 }, 2^{ 2.5 }\}$. Similar to the other adaptive procedure, additional models were removed in subsequent iterations: at $i = 10$, 16 models were removed and single models were eliminated at $i = 13$, $i = 24$ and $i = 34$. There were the same 2 surviving values of $\theta$ at $B = 50$: $2^{ 1.5 }$ and $2^{ 2 }$. The same sub–model was selected as the previous two analyses. When the futility was estimated by means of the Bradley–Terry model, a speed–up of 3.2 was achieved. As before, this was largely due to fitting only 331 SVM models (31.5$\%$ of the full set of resamples). ![Ability estimates ($\widehat{\lambda}_j - \widehat{\lambda}_{j'}$) from the Bradley–Terry model after $B_{min} = 10$ resamples. $\theta = 2^{ 2 }$ was used as the reference model and the error bars correspond to asymptotic 95$\%$ confidence intervals. []{data-label="F:qsar_bt"}](qsar_bt){width=".8\textwidth"} Since the resampled performance measures are coerced to binary values, this may have the effect of desensitizing the adaptive procedure and result in a potential loss of inferential power. As previously noted, in situations where the fitness values are close to their boundaries (e.g. an area under the ROC curve near unity or RMSE near zero), the distribution of the $Q_{ij}$ can become significantly skewed and the linear model’s assumption of normality of the residuals may not hold. However, the dichotomization process does protect against skewness or influence of aberrantly extreme values of $Q_{ij}$. Also, the effective sample size used for the Bradley–Terry model is based on the number of pair–wise competitions. At iteration $i$ of resampling, the number of win/loss/tie statistics is $i\times\binom{p_i}{2}$. When $p_i$ is large relative to $i$, the Bradley–Terry model may be estimable where the generalized linear model may not. Simulation Studies ================== To understand the potential benefits and pitfalls of these approaches, simulation studies were used. @Friedman:1991p109 described a system for simulating nonlinear regression models. Four independent uniform random variables $X_1 \ldots X_4$ were used with the following regression structure: $$Y = atan(((X_2 X_3 - (1/(X_2 X_4)))/X_1) + \epsilon$$ where $\epsilon \sim N(0, 0.1)$. To make the simulation more realistic, 46 independent, non–informative random normal predictors were added to the training and test sets. This system was used to evaluate the efficacy of the proposed algorithm. For each simulated data set, an artificial neural network model [@Bishop95] was used to model and predict the data. A single layer feed–forward architecture was used and candidate sub–models included between 1 and 10 hidden units. Also, the model was tuned over three amounts of weight decay: 0, 10$^{-3}$ and 10$^{-2}$. The full set of 30 sub–models was evaluated for each neural network model. Repeated 10–fold cross–validation was used as the resampling method. The total number of resamples varied between 20 and 100 in the simulations along with: - The training set size was varied: 200, 400 and 600. - Two settings for the minimum number of resamples ($B_{min}$) were evaluated: 10 and 20. - The confidence values for the confidence intervals were evaluated over three values for both adaptive procedures: $\alpha \in \{0.001, 0.01, 0.1\}$. For each simulation setting, a minimum 100 data sets were created and analyzed (the final number was affected by hardware failures). The simulated test set ($n$ = 100,000) root mean squared error was calculated for models corresponding to the complete set of resamples and adaptive strategies. The efficacy of the two adaptive procedures were quantified by the speed–up and the difference in RMSE between the full set of resamples and the adaptive procedures. The computations were conducted in using a modified version of the package [@caret]. To illustrate the relationship between performance and the tuning parameters, one simulated model with 1000 training set points was tuned with six repeats of 10–fold cross–validation. The resulting profile can be seem in Figure \[F:sim\_nnet\_profile\]. Based on these data, the optimal parameter settings are a single hidden unit and a weight decay value of 0.01. The other settings that are most competitive with this condition have one or two hidden units and many of the other settings are likely to be eliminated quickly, depending on how much uncertainty is associated with the $Q_{kj}$. ![An example of a simulated resampling profile for a neural network model using $B=100$ resamples generated with repeated 10–fold cross–validation.[]{data-label="F:sim_nnet_profile"}](sim_nnet_profile){width=".8\textwidth"} In general, the concordance between the two adaptive procedures and the nominal approach with the full set of resamples was good. Using linear models, 81.9$\%$ of the final parameters matched the settings found with the nominal procedure (across all simulated data sets and conditions). Similarly, 82$\%$ of the model settings match found via the Bradley–Terry model matched. However, the full resampling process is not infallible may not yield the best test set results. Figure \[F:speedup\_seq\] shows the percentages of simulations where the adaptive procedures selected the same sub–model as the nominal resampling scheme [*or*]{} the adaptive resampling chose a model with [*better*]{} test set results. Based on this revised criterion, the overall percentage of models at least as good as the matching fully resampled model increased to 88.9$\%$ and 88.2$\%$ for the linear models and Bradley–Terry approach, respectively. From the plots in Figure \[F:speedup\_seq\], there are several patterns. First, the training set size has the most significant affect on the probability of a good model. The smallest training set size ($n=200$) has the worst efficacy and the two larger sizes had roughly comparable findings in terms of choosing a good model. This is most likely due to the quality of the estimated values of $Q_{ij}$ since larger training set sizes lead to larger holdout sets. As the accuracy in the $Q_{ij}$ increases, the likelihood of discarding a quality value of $\theta_j$ decreases. In these simulations, $B_{min}$ and the confidence level did not appear to have a major impact on the quality of the model within the ranges that were studied here. For values of $B > 20$, the number of resamples did not appear to have much of an effect. Finally, the method of computing futility showed comparable value. The speed–up of the procedures are also shown in these figures. The median speed-up over all simulations were 2.9 and 3.1 for the generalized least squares and Bradley–Terry methods, respectively. Similarly, the best case speed–up, where the training set size and the number of resamples are large, were 26.3 and 33.1, respectively. There were a small number of simulations (0.5% of the total) that took longer with the adaptive procedures and were more likely to occur using generalized linear models. This occurred with small training sets and fewer resamples. Overall, the speed–ups were driven by the number of resamples ($B$) and the training set size. This makes intuitive sense as these two factors are surrogates for the total computational cost of model tuning. ![The median speed–up versus the percentage of simulations where the adaptive procedure selected a model at least as good as the full resampling process. The legend indicates the training set size (either 200, 400 or 600) and the value of $\alpha$ (either 0.001, 0.01 or 0.1).[]{data-label="F:speedup_seq"}](speedup_gls_seq){width=".9\textwidth"} ![The median speed–up versus the percentage of simulations where the adaptive procedure selected a model at least as good as the full resampling process. The legend indicates the training set size (either 200, 400 or 600) and the value of $\alpha$ (either 0.001, 0.01 or 0.1).[]{data-label="F:speedup_seq"}](speedup_bt_seq){width=".9\textwidth"} The Effect of Parallel Processing ================================= Parallel processing has become more common in scientific computing. Many computers are currently configured with multicore architectures and open–source software is widely available to run computations in parallel [@Schmidberger:2009tx; @mccallum2011parallel]. The model tuning process is “embarrassingly parallel”. In Algorithm \[A:Resamp\], the two [for]{} loops (lines 2 and 4) are not serial, meaning the computations inside of the loops are independent. For the SVM model described in Section \[S:intro\], a total of 1050 models were fit across different sub–model configurations and resamples. There is no logical barrier to running these computations in parallel and doing so leads to substantial speed–up [@apm]. For the non–adaptive approach, the result of parallel processing with six [*worker*]{} processes was a speed–up of 3–fold (relative to sequentially processing the full set of resamples). Reasonable questions would be “can I get the same time reduction using parallel processing and the full set of resamples? or “does the adaptive procedure still offer advantages in parallel?” The process shown in Algorithm \[A:adaptive\] can also benefit from parallel processing. First, all of the computations across models (i.e. those within lines 4 and 6 of Algorithm \[A:adaptive\]) can be conducted in parallel. The only situation where the resamples cannot be run in parallel are when $i > B_{min}$ and $p_i > 1$. However, there is some slow–down associated with the additional computations required to conduct the futility analysis. For the SVM model, the adaptive techniques were also evaluated with six worker processes. When compared to the full resampling approach run in parallel, the speed–up for the linear model and Bradley–Terry approaches were 3.6 and 3.5, respectively. This indicates that there is benefit to the adaptive procedures above and beyond those imparted using parallel processing. The simulation studies were repeated with parallel processing. In this study, multicore forking of calculations [@Schmidberger:2009tx; @Eugster:tr] is used to run combinations of models and resampled data sets using more than one processor on the same machine. Version 0.1.7 of ’s package was used. In these simulations, the computational tasks were split over six sub–processes as previously described. Figure \[F:speedup\_par\] shows the average speed–ups for the sequential and parallel computations. Under these conditions, the median speed-up over all simulations were 3.2 and 3.5 for the generalized least squares and Bradley–Terry methods, respectively. For both adaptive procedures, there was a high degree of correlation in the median speed–ups. This indicate that, for this simulation, parallel processing did not eliminate the benefits of adaptively removing tuning parameter values. Since the speed–ups are comparable between different technologies, this indicates that, independent of the technology, adaptive methods are faster. For example, in the generalized least squares simulations with 600 points in the training set, $B = 60$, $B_{min} = 20$ and $\alpha = 0.01$, the median time to get the full set of resamples sequentially was 40.9 hours. Using parallel processing only, the median time was reduced to 11.3 hours. Without parallel processing, adaptive resampling would have reduced the training time to 13.8 hours. However, the biggest savings occurred with adaptive resampling in parallel; here the median time was 3.9 hours. ![The median speed–ups for the sequential and parallel computations.[]{data-label="F:speedup_par"}](speedup_gls_par){width=".9\textwidth"} ![The median speed–ups for the sequential and parallel computations.[]{data-label="F:speedup_par"}](speedup_bt_par){width=".9\textwidth"} Discussion ========== In this manuscript, a resampling scheme was described that is effective at finding reasonable values of tuning parameters in a more computationally efficient manner. The efficacy and efficiencies of the procedures are best when the training set size is not small and computational cost of fitting the sub–models is moderate to high. The computational gains afforded by parallel processing technologies do not obviate the gains in the proposed methodologies. There are several possible improvements to the adaptive procedures that were not explored here. For some models, there is the possibility that several values of $\theta$ will produce identical models. For example, some decision trees are tuned over their [*maximum*]{} possible depth but, after pruning, the same tree may result. As another example, multivariate adaptive regression spline [@Friedman:1991p109] models conduct feature selection to remove model terms. Because of this, it is possible for multiple values in $\Theta$ to generate the same model predictions and thus have identical fitness values. The consequence of this is that the futility analysis procedure will never be able to differentiate between these models and may conduct unnecessary iterations of resampling. To mitigate these risks, a one–time filter could be used to remove values of $\theta$ that are identical (within floating point precision) when $i \ge B_{min}$. Also, if the modeling goal was strictly [*cross–validatory choice*]{} and the adaptive procedure determines eliminates all but a single setting, there may be no need to estimate the performance for the model using all $B$ resamples. In such cases, the speed–up values would substantially increase. Finally, if there is a string desire to minimize the possibility that the adaptive procedure might choose a sub–optimal model, a more conservative approach could be used. For example, this manuscript used a static confidence level. Alternatively, a dynamic approach where the likelihood of discarding settings is small at the beginning of the adaptive procedure, but increases at each iteration, may be more appropriate. 0.2in

ArXiv

--- abstract: 'The aim of this paper is to unveil an unexpected relationship between the normal form of a polynomial with respect to a polynomial ideal and the more geometric concept of orthogonality. We present a new way to calculate the normal form of a polynomial with respect to a polynomial ideal I in the ring of multivariate polynomials over a field K, provided the field K is finite and the ideal I is a vanishing ideal. In order to use the concept of orthogonality, we introduce a symmetric bilinear form on a vector space over a finite field.' author: - 'Edgar Delgado-Eckert[^1] [^2] [^3]' bibliography: - 'MathRef.bib' title: Canonical representatives for residue classes of a polynomial ideal and orthogonality --- Polynomial algebras, polynomial ideals, Gröbner bases, inner products, normal form 13P10; 15A63 Introduction ============ A well known result of B. Buchberger is the existence of the normal form of a polynomial with respect to a polynomial ideal $I$ in the ring of multivariate polynomials over a field $K$. This result follows from the existence of so called Gröbner bases for polynomial ideals. For a given fixed term ordering, this normal form is unique [@806353], [MR0463136]{}, [@MR0268178]. In this paper we present a new way to calculate this normal form, provided the field $K$ is finite and the ideal $% I $ is a vanishing ideal, i.e. $I$ is equal to the set of polynomials which vanish in a given set of points $X$. Our method doesn’t pursue establishing a new, especially efficient, algorithm for the computation of such a normal form. Rather, the aim of this paper is to unveil an interesting way to look at this issue based on the concept of orthogonality. For orthogonality to apply, we introduce a symmetric bilinear form on a vector space (see, for instance, [@SCHARLAU]). A symmetric bilinear form can be seen as a generalized inner product. Some authors have explored vector spaces endowed with generalized forms of inner products. For example, we refer to the following papers: [@MR0133024], [@MR0385527],[@MR0377485], [@MR0482441], [@MR578592], [MR586525]{}, [@MR2064794]. Having defined a symmetric bilinear form, we are able to introduce the notion of orthogonality and orthonormality. Then we consider the orthogonal solution of a solvable inhomogeneous under-determined linear operator equation. If one thinks of an inhomogeneous under-determined system of linear equations in an Euclidean space, the orthogonal solution is simply the solution that is perpendicular to the affine subspace associated with the system. After going through existence and uniqueness considerations, we come to the main statement of this paper, namely, that the above mentioned normal form can be obtained as the orthogonal solution of a system of linear equations. That system of equations arises as a linear formulation of the multivariate polynomial interpolation problem. Based on our literature research, we believe that the study of polynomial algebras in the framework of symmetric bilinear spaces (vector spaces endowed with a symmetric bilinear form) represents a novel approach. Suitable extensions of our method to more general fields (i.e. infinite fields) could open new possibilities for studying problems in the areas of polynomial algebra, computational algebra and algebraic geometry using functional analytic or linear algebraic techniques. The concept of orthogonal solution is not limited by monomial orders, as it is the case for Gröbner bases calculations. In this sense, our method reveals a wider class of normal forms (with respect to vanishing ideals) in which the normal forms à la Buchberger appear as special cases. Another application that we will describe in detail elsewhere is the problem of choosing a particular interpolant among all possible solutions of a highly under-determined multivariate interpolation problem. This is related to the study of the performance of so called “reverse engineering” algorithms such as the one presented in [@MR2086931]. The organization of this article is the following: Section 2 is devoted to the general definition of *symmetric bilinear spaces* and *orthogonal solutions* of an inhomogeneous linear operator equation. Subsection 2.1 covers basic definitions and properties of symmetric bilinear spaces, in particular, the concepts of *orthogonality* and *orthonormality* are introduced. Subsection 2.2 introduces the notion of orthogonal solution of a solvable under-determined linear operator equation. Existence and uniqueness of orthogonal solutions are proved and some issues regarding the existence of orthonormal bases are discussed. Section 3 deals with the vector space of functions $F:K^{n}\rightarrow K,$ where $K$ is a finite field and $n\in %TCIMACRO{\U{2115} }% %BeginExpansion \mathbb{N} %EndExpansion .$ In subsection 3.1 we paraphrase the known result that all the functions in that space are polynomial functions. Subsection 3.2 introduces a linear operator called *evaluation epimorphism* and formulates the multivariate polynomial interpolation problem in a linear algebraic fashion. Section 4 covers the more technical aspect of constructing special symmetric bilinear forms. Using that type of symmetric bilinear form will allow us to prove the main result of this article in section 5. Section 5 is devoted to the statement and proof of our main result. Namely, that the canonical normal form of an arbitrary polynomial $f$ with respect to a vanishing ideal $I(X)$ in the ring of multivariate polynomials over a finite field $K$ can be calculated as the orthogonal solution of a linear operator equation involving the evaluation epimorphism. For standard terminology, notation and well known results in computational algebraic geometry and commutative algebra we refer to [@MR1417938] and [@MR1213453]. Symmetric bilinear vector spaces and orthogonal solutions of inhomogeneous systems of linear equations ====================================================================================================== Basic definitions ----------------- In this subsection we will introduce the concept of a symmetric bilinear form in a vector space. With this concept it will be possible to define symmetric bilinear vector spaces and orthonormality. Furthermore, some basic properties are briefly reviewed (cf. [@SCHARLAU]) Let $V$ be a vector space over a field $K.$ A* *symmetric and bilinear mapping$$\left\langle \cdot ,\cdot \right\rangle :V\times V\rightarrow K$$is called *symmetric bilinear form* on $V.$ Let be $n\,,m\in %TCIMACRO{\U{2115} }% %BeginExpansion \mathbb{N} %EndExpansion $ natural numbers and $K$ a field. The set of all $m\times n$ matrices ($m$* rows and* $n$* columns*) with entries in $K$ is denoted by $% M(m\times n;$ $K).$ \[MatrixDefOfBilinForms\]Let $V$ be a finite dimensional vector space over a field $K$. After fixing a basis $(u_{1},...,u_{d})$ of $V,$ it is a well known result, that there is a one-to-one correspondence between the set of all symmetric bilinear forms on $V$ and the set of all $d\times d$ symmetric matrices with entries in $K$ seen as representing matrices with respect to the basis $(u_{1},...,u_{d}).$ A vector space $V$ over a field $K$ endowed with a symmetric bilinear form$$\left\langle \cdot ,\cdot \right\rangle :V\times V\rightarrow K$$is called a *symmetric bilinear space*. Every (real) Euclidean space is due to the positive definiteness of its inner product a symmetric bilinear space. Given a symmetric bilinear space $V$ over a field $K$, *orthogonality* and *orthonormality* of two vectors $v,w\in V$ as well as the concept of *orthonormal basis* are defined exactly as in the Euclidean case. Similarly, the *orthogonal complement* $W^{\perp }:=\left\{ v\in V\mid v\perp w\text{ }\forall \text{ }w\in W\right\} $ of a subspace $W\subseteq V$  is a subspace of $V.$ Furthermore, if $\left( w_{1},...w_{d}\right) $ is an orthonormal basis of $V,$ then for every vector $v\in V$ holds$$v=\tsum_{k=1}^{d}\left\langle v,w_{k}\right\rangle w_{k}$$where the field elements $\left\langle v,w_{i}\right\rangle \in K,$ $% i=1,...,d$ are the well known *Fourier coefficients*. Contrary to the case of Euclidean or unitary vector spaces, in symmetric bilinear spaces orthonormal bases don’t always exist. \[Const.Of.Orthonorm.Basis\]Let $d\in %TCIMACRO{\U{2115} }% %BeginExpansion \mathbb{N} %EndExpansion $ be a natural number and $V$ a $d$-dimensional vector space over a field $% K. $ Furthermore let $\left( u_{1},...u_{d}\right) $ be a basis of $V.$ Then one can construct a symmetric bilinear form on $V$ by setting$$\left\langle u_{i},u_{j}\right\rangle :=\delta _{ij}\text{ }\forall \text{ }% i,j\in \{1,...,d\}$$(see also Remark \[MatrixDefOfBilinForms\].) Here the basis $\left( u_{1},...u_{d}\right) $ is obviously orthonormal. Orthogonal solutions of inhomogeneous linear operator equations --------------------------------------------------------------- Let $d\in %TCIMACRO{\U{2115} }% %BeginExpansion \mathbb{N} %EndExpansion $ be a natural number and $V$ a $d$-dimensional symmetric bilinear space over a field $K.$ Furthermore, let $W$ be an arbitrary vector space over the field $K$, $T:V\rightarrow W$ a non-injective linear operator and $w\in W$ a vector with the property$$w\in T(V)$$Now let $m:=$nullity$(T)\in %TCIMACRO{\U{2115} }% %BeginExpansion \mathbb{N} %EndExpansion $ be the dimension of the kernel of $T.$ A solution $v^{\ast }\in V$ of the equation$$Tv=w$$is called *orthogonal solution*, if for an arbitrary basis $% (u_{1},...,u_{m})$ of $\ker (T)$ the following orthogonality conditions hold$$\left\langle u_{i},v^{\ast }\right\rangle =0\text{ }\forall \text{ }i\in \{1,...,m\}$$ \[Orth.Sol.LiesOnOrth.Comp.\]Let $(u_{1},...,u_{m})$ be a basis of $\ker (T).$ Then each arbitrary vector $u\in \ker (T)$ can be written in the form$$u=\tsum_{i=1}^{m}\lambda _{i}u_{i}$$with suitable field elements $\lambda _{i}\in K.$ If the orthogonality conditions$$\left\langle u_{i},v^{\ast }\right\rangle =0\text{ }\forall \text{ }i\in \{1,...,m\}$$hold for the basis $(u_{1},...,u_{m}),$ then we have$$\left\langle u,v^{\ast }\right\rangle =\left\langle \tsum_{i=1}^{m}\lambda _{i}u_{i},v^{\ast }\right\rangle =\tsum_{i=1}^{m}\lambda _{i}\left\langle u_{i},v^{\ast }\right\rangle =0$$and that means$$v^{\ast }\in \ker (T)^{\perp }$$In particular, for any other different basis $(w_{1},...,w_{m})$ of $\ker (T) $ it holds$$\left\langle w_{j},v^{\ast }\right\rangle =0\text{ }\forall \text{ }j\in \{1,...,m\}$$ Let $d\in %TCIMACRO{\U{2115} }% %BeginExpansion \mathbb{N} %EndExpansion $ be a natural number and $V$ a $d$-dimensional symmetric bilinear space over a field $K.$ Furthermore, let $W$ be an arbitrary vector space over the field $K$, $T:V\rightarrow W$ a non-injective linear operator and $w\in W$ a vector with the property$$w\in T(V)$$If $\ker (T)$ has an *orthonormal basis*, then the equation$$Tv=w$$has always a unique orthogonal solution $v^{\ast }\in V.$ Let $m:=$nullity$(T)=\dim (\ker (T))\in %TCIMACRO{\U{2115} }% %BeginExpansion \mathbb{N} %EndExpansion $ be the dimension of the null space of $T$ and $(u_{1},...,u_{m})$ an orthonormal basis of $\ker (T).$ Since $w\in T(V),$ there must exist a solution $\widehat{\xi }\in V$ of $Tv=w.$ For any other solution $\xi \in V$ we have$$T(\xi -\widehat{\xi })=T(\xi )-T(\widehat{\xi })=0$$and therefore$$\xi -\widehat{\xi }\in \ker (T)$$That means that all solutions $\xi \in V$ of $Tv=w$ can be written in the form$$\xi =\widehat{\xi }+\tsum_{i=1}^{m}\lambda _{i}u_{i}$$with the $\lambda _{i}\in K,$ $i=1,...,m$ running over all $K.$ In particular, we can construct a very specific solution by choosing the parameters $\lambda _{i}\in K,$ $i=1,...,m$ in the following manner$$\lambda _{i}:=-\left\langle u_{i},\widehat{\xi }\right\rangle ,\text{ }% i=1,...,m$$For this solution$$v^{\ast }:=\widehat{\xi }+\tsum_{i=1}^{m}-\left\langle u_{i},\widehat{\xi }% \right\rangle u_{i}$$and for every $j\in \{1,...,m\}$ it holds$$\begin{aligned} \left\langle u_{j},v^{\ast }\right\rangle &=&\left\langle u_{j},\widehat{\xi }+\tsum_{i=1}^{m}-\left\langle u_{i},\widehat{\xi }\right\rangle u_{i}\right\rangle =\left\langle u_{j},\widehat{\xi }\right\rangle +\tsum_{i=1}^{m}-\left\langle u_{i},\widehat{\xi }\right\rangle \left\langle u_{j},u_{i}\right\rangle \\ &=&\left\langle u_{j},\widehat{\xi }\right\rangle +\tsum_{i=1}^{m}-\left\langle u_{i},\widehat{\xi }\right\rangle \delta _{ji}=\left\langle u_{j},\widehat{\xi }\right\rangle -\left\langle u_{j},% \widehat{\xi }\right\rangle =0\end{aligned}$$This shows the existence of an orthogonal solution of $Tv=w.$ Now let $% \widetilde{v}\in V$ be another orthogonal solution of $Tv=w.$ Again, since$$T(v^{\ast }-\widetilde{v})=T(v^{\ast })-T(\widetilde{v})=0$$we can write$$v^{\ast }=\widetilde{v}+\tsum_{i=1}^{m}\alpha _{i}u_{i}$$with suitable $\alpha _{i}\in K.$ From the orthogonality conditions for $% v^{\ast }$ and $\widetilde{v}$ we have $\forall $ $j\in \{1,...,m\}$$$\begin{aligned} 0 &=&\left\langle u_{i},v^{\ast }\right\rangle =\left\langle u_{i},% \widetilde{v}+\tsum_{i=1}^{m}\alpha _{i}u_{i}\right\rangle =\left\langle u_{j},\widetilde{v}\right\rangle +\left\langle u_{j},\tsum_{i=1}^{m}\alpha _{i}u_{i}\right\rangle \\ &=&\tsum_{i=1}^{m}\alpha _{i}\left\langle u_{j},u_{i}\right\rangle =\tsum_{i=1}^{m}\alpha _{i}\delta _{ji}=\alpha _{j}\end{aligned}$$and that means $v^{\ast }=\widetilde{v}.$ The existence of an orthonormal basis of $\ker (T)$ is crucial for the proof of this theorem. It is important to notice that in a symmetric bilinear space over a general field $K,$ the Gram-Schmidt orthonormalization only works if the norm$$\left\Vert v\right\Vert :=\sqrt{\left\langle v,v\right\rangle }$$of the vectors used in the Gram-Schmidt process exists in the field $K\ $and is not equal to the zero element. In general terms, the existence of square roots would be assured in a field $K$ which satisfies$$\forall \text{ }x\in K\text{ }\exists \text{ }y\in K\text{ such that }y^{2}=x \label{ExistenceOfSquareRoots}$$Now, if $K$ is finite, then (\[ExistenceOfSquareRoots\]) holds if and only if $Char(K)=2.$After fixing a basis $(u_{1},...,u_{d})$ for the vector space $V,$ the question whether $\left\langle v,v\right\rangle =0$ for $v\neq 0$ is equivalent to the nontrivial solvability in $K^{d}$ of the following quadratic form$$\vec{x}^{t}A\vec{x}=0 \label{QuadraticForm}$$where $A$ is the representing matrix of $\left\langle .,.\right\rangle $with respect to the basis * *$(u_{1},...,u_{d})$ (see Remark [MatrixDefOfBilinForms]{}). In chapter 3, §2 of [@MR1429394] explicit formulas for the exact number of solutions in $K^{n}$ of equations of the type (\[QuadraticForm\]), where $A$ is a $n\times n$ symmetric matrix with entries in a finite field $K$, can be found. \[ZeroSol\]Let $K,$ $d,$ $V,$ $W$ and $T$ be as in the theorem above. If $\ker (T)$ has an *orthonormal basis*, then the equation$$Tv=0$$has always the unique orthogonal solution $0\in V.$ The vector space of functions $\mathbf{F}_{q}^{n}\rightarrow \mathbf{F}_{q}$ ============================================================= In the next subsection we review the well known result that any function $% F:K^{n}\rightarrow K,$ where $K$ is a finite field and $n\in %TCIMACRO{\U{2115} }% %BeginExpansion \mathbb{N} %EndExpansion $, is a polynomial function. Furthermore, we introduce the family of fundamental monomial functions. The ring of polynomial functions in $n$ variables over $\mathbf{F% }_{q}$ and the vector space of functions $\mathbf{F}_{q}^{n}\rightarrow \mathbf{F}_{q}$ ------------------------------------------------------------------------ We will denote a finite field with $\mathbf{F}_{q}$, where $q$ stands for the number of elements of the field ($q$ is a power of the prime characteristic of the field). We call a commutative Ring $(R,+,\cdot )$ with multiplicative identity $% 1\neq 0$ and the binary operations $\cdot $ and $+$ just Ring $R$. The following three results are well known: Let $R$ be a ring and $n\in %TCIMACRO{\U{2115} }% %BeginExpansion \mathbb{N} %EndExpansion $ a natural number. The set $$PF_{n}(R):=\{g\text{ }|\text{ }g:R^{n}\rightarrow R\text{ is polynomial}\}$$together with the common operations $+$ and $\cdot $ of addition and multiplication of mappings is a ring. This ring is called *ring of all polynomial functions over* $R$* in* $n$* *$R$*-valued variables.* Let $K$ be an arbitrary field and $n\in %TCIMACRO{\U{2115} }% %BeginExpansion \mathbb{N} %EndExpansion $ a natural number. The set of all functions$$f:K^{n}\rightarrow K$$together with the common operations of addition of mappings and scalar multiplication is a vector space over $K$. We denote this vector space with $% F_{n}(K).$ \[ThmFuncIsPolFunc\]Let $\mathbf{F}_{q}$ be a finite field. Then for the *sets* $F_{n}(\mathbf{F}_{q})$ and $PF_{n}(\mathbf{F}_{q})$ it holds$$F_{n}(\mathbf{F}_{q})=PF_{n}(\mathbf{F}_{q})$$ This result is proved in Chapter 7, Section 5 of [@MR1429394]. Let $n,q\in %TCIMACRO{\U{2115} }% %BeginExpansion \mathbb{N} %EndExpansion $ be natural numbers. Further let $>$ be a total ordering on $\left( %TCIMACRO{\U{2115} }% %BeginExpansion \mathbb{N} %EndExpansion _{0}\right) ^{n}.$ The according to $>$ decreasingly ordered set $$M_{q}^{n}:=\left\{ \alpha \in \left( %TCIMACRO{\U{2115} }% %BeginExpansion \mathbb{N} %EndExpansion _{0}\right) ^{n}\mid \alpha _{j}<q\text{ }\forall \text{ }j\in \{1,...,n\}\right\}$$of all $n$-tuples with entries smaller than $q$ is denoted by $% M_{q}^{n}\subset \left( %TCIMACRO{\U{2115} }% %BeginExpansion \mathbb{N} %EndExpansion _{0}\right) ^{n}.$ \[KardinalitaetVonMp\]In order to avoid a too complicated notation, we skip the appearance of the order relation $>$ in the symbol for this set. It is easy to prove, that $M_{q}^{n}$ contains exactly $q^{n}$ $n$-tuples. We will index the $n$-tuples in $M_{q}^{n}$ starting with the biggest and ending with the smallest:$$\alpha _{1}>\alpha _{2}>...>\alpha _{q^{n}}$$ For any fixed natural numbers $n,q\in %TCIMACRO{\U{2115} }% %BeginExpansion \mathbb{N} %EndExpansion $ and for each multi index $\alpha \in M_{q}^{n}$ consider the monomial function$$\begin{aligned} g_{nq\alpha } &:&K^{n}\rightarrow K \\ \vec{x} &\mapsto &g_{nq\alpha }(\overrightarrow{x}):=\overrightarrow{x}% ^{\alpha }\end{aligned}$$All these monomial functions $g_{nq\alpha },$ $\alpha \in M_{q}^{n}$ are called *fundamental monomial functions*. The following result is elementary. Its easy induction proof is left to the reader: \[Fund.Mon.Fct.AreBasis\]A basis for the vector space $F_{n}(\mathbf{F}% _{q})$ is given by the fundamental monomial functions$$(g_{nq\alpha })_{\alpha \in M_{q}^{n}}$$ The basis elements in the basis $(g_{nq\alpha })_{\alpha \in M_{q}^{n}}$ are ordered according to the order relation $>$ used to order the $n$-tuples in the set $M_{q}^{n}.$ That means (see Remark \[KardinalitaetVonMp\])$$(g_{nq\alpha })_{\alpha \in M_{q}^{n}}=(g_{nq\alpha _{i}})_{i\in \{1,...,q^{n}\}}$$ Solving the polynomial interpolation problem in $PF_{n}(\mathbf{F% }_{q})$ ------------------------------------------------------------------ In this subsection we define the *evaluation epimorphism* of a tuple $(% \vec{x}_{1},...,\vec{x}_{m})\in (\mathbf{F}_{q}^{n})^{m}$ of points in the space $\mathbf{F}_{q}^{n}.$ The evaluation epimorphism allows for a linear algebraic formulation of the multivariate polynomial interpolation problem. Let $\mathbf{F}_{q}$ be a finite field and $n,m\in %TCIMACRO{\U{2115} }% %BeginExpansion \mathbb{N} %EndExpansion $ natural numbers with$m\leq q^{n}$. Further let$$\vec{X}:=(\vec{x}_{1},...,\vec{x}_{m})\in (\mathbf{F}_{q}^{n})^{m}$$be a tuple of $m$ **different** $n$-tuples with entries in the field $% \mathbf{F}_{q}.$ Then the mapping$$\begin{aligned} \Phi _{\vec{X}} &:&F_{n}(\mathbf{F}_{q})\rightarrow \mathbf{F}_{q}^{m} \\ f &\mapsto &\Phi _{\vec{X}}(f):=(f(\vec{x}_{1}),...,f(\vec{x}_{m}))^{t}\end{aligned}$$is a surjective linear operator. $\Phi _{\vec{X}}$ is called the *evaluation epimorphism* *of the tuple* $\vec{X}.$ The proof of the linearity is left to the reader. Now let $\vec{b}\in \mathbf{F}_{q}^{m}$ be an arbitrary vector. Since $m\leq q^{n}$ we can construct a function$$g\in F_{n}(\mathbf{F}_{q})$$with the property$$g(\vec{x}_{i})=b_{i}\text{ }\forall \text{ }i\in \{1,...,m\}$$and that means exactly$$\Phi _{\vec{X}}(g)=\vec{b}\text{ \ \ }\endproof$$ \[FullRank\]Since a basis of $F_{n}(\mathbf{F}_{q})$ is given by the fundamental monomial functions $(g_{nq\alpha })_{\alpha \in M_{q}^{n}},$ the matrix$$A:=(\Phi _{\vec{X}}(g_{nq\alpha }))_{\alpha \in M_{q}^{n}}\in M(m\times q^{n};\mathbf{F}_{q})$$representing the evaluation epimorphism $\Phi _{\vec{X}}$ of the tuple $\vec{X}$ with respect to the basis $(g_{nq\alpha })_{\alpha \in M_{q}^{n}}$ of $F_{n}(\mathbf{F}_{q})$ and the canonical basis of $\mathbf{F}_{q}^{m}$ has always the full rank $m=\min (m,q^{n}).$ That also means, that the dimension of the $\ker (\Phi _{\vec{X}})$ is$$\dim (\ker (\Phi _{\vec{X}}))=\dim (F_{n}(\mathbf{F}_{q}))-m=q^{n}-m$$ Let $\mathbf{F}_{q}$ be a finite field and $n,m\in \mathbb{N} $ natural numbers with $m\leq q^{n}$. Further let$$\vec{X}:=(\vec{x}_{1},...,\vec{x}_{m})\in (\mathbf{F}_{q}^{n})^{m}$$be a tuple of $m$ different $n$-tuples with entries in the field $\mathbf{F}_{q}$ and $\vec{b}\in \mathbf{F}_{q}^{m}$ a vector. Then the interpolation problem of finding a polynomial function $f\in PF_{n}(\mathbf{F}_{q})$ with the property$$f(\vec{x}_{i})=b_{i}\text{ }\forall \text{ }i\in \{1,...,m\}$$can be solved by solving the system of linear equations$$A\vec{y}=\vec{b} \label{Int.Cond.}$$where$$A:=(\Phi _{\vec{X}}(g_{nq\alpha }))_{\alpha \in M_{q}^{n}}$$is the matrix representing the evaluation epimorphism $\Phi _{\vec{X}}$ of the tuple $\vec{X}$ with respect to the basis $(g_{nq\alpha })_{\alpha \in M_{q}^{n}}$ of $F_{n}(\mathbf{F}_{q})$ and the canonical basis of $\mathbf{F}_{q}^{m}$. The entries of a solution vector of the equations (\[Int.Cond.\]) are the coefficients of the solution with respect to the basis $(g_{nq\alpha })_{\alpha \in M_{q}^{n}}.$ Since $F_{n}(\mathbf{F}_{q})=PF_{n}(\mathbf{F}_{q}),$ a solution of the interpolation problem can be found by solving the equation$$\Phi _{\vec{X}}(g)=\vec{b} \label{OpEq}$$for $g,$ where $\Phi _{\vec{X}}$ is the surjective linear operator$$\begin{aligned} \Phi _{\vec{X}} &:&F_{n}(\mathbf{F}_{q})\rightarrow \mathbf{F}_{q}^{m} \\ f &\mapsto &\Phi _{\vec{X}}(f):=(f(\vec{x}_{1}),...,f(\vec{x}_{m}))^{t}\end{aligned}$$of the above theorem. After fixing the basis $(g_{nq\alpha })_{\alpha \in M_{q}^{n}}$ of $F_{n}(\mathbf{F}_{q})$ and the canonical basis of $\mathbf{F}_{q}^{m},$ equation (\[OpEq\]) implies the following system of linear equations for the coefficients of the solutions with respect to the basis $(g_{nq\alpha })_{\alpha \in M_{q}^{n}}$$$A\vec{y}=\vec{b}$$where$$A:=(\Phi _{\vec{X}}(g_{nq\alpha }))_{\alpha \in M_{q}^{n}}$$is the matrix representing the map $\Phi _{\vec{X}}$ with respect to the basis $(g_{nq\alpha })_{\alpha \in M_{q}^{n}}$ of $F_{n}(\mathbf{F}_{q})$ and the canonical basis of $\mathbf{F}_{q}^{m}$. According to Remark [FullRank]{}, the matrix $A$ has full rank and therefore a solution of $A\vec{y}=\vec{b}$ always exists. Construction of special purpose symmetric bilinear forms ======================================================== Let $\mathbf{F}_{q}$ be a finite field and $n,m\in \mathbb{N} $ natural numbers with $m<q^{n}$. Further let$$\vec{X}:=(\vec{x}_{1},...,\vec{x}_{m})\in \left( \mathbf{F}_{q}^{n}\right) ^{m}$$be a tuple of $m$ different $n$-tuples with entries in the field $\mathbf{F}_{q}$ and $d:=\dim (F_{n}(\mathbf{F}_{q})).$ Now consider the evaluation epimorphism $\Phi _{\vec{X}}$ of the tuple $\vec{X}.$ By Remark [FullRank]{} and due to the fact $m<q^{n},$ the nullity of $\Phi _{\vec{X}}$ is given by$$s:=\dim (\ker (\Phi _{\vec{X}}))=\dim (F_{n}(\mathbf{F}_{q}))-m=q^{n}-m>0$$Now let $(u_{1},...,u_{s})$ be a basis of $\ker (\Phi _{\vec{X}})\subseteq F_{n}(\mathbf{F}_{q}).$ By the basis extension theorem, we can extend the basis $(u_{1},...,u_{s})$ to a basis$$(u_{1},...,u_{s},u_{s+1},...,u_{d})$$of the whole space $F_{n}(\mathbf{F}_{q}).$ As in example [Const.Of.Orthonorm.Basis]{}, we can construct a symmetric bilinear form on $F_{n}(\mathbf{F}_{q})$ by setting$$\left\langle u_{i},u_{j}\right\rangle :=\delta _{ij}\text{ }\forall \text{ }i,j\in \{1,...,d\}$$Here the basis $\left( u_{1},...u_{d}\right) $ is orthonormal and the vectors $(u_{s+1},...,u_{d})$ are a basis of the orthogonal complement $\ker (\Phi _{\vec{X}})^{\perp }$ of $\ker (\Phi _{\vec{X}}).$ In general, the way we extend the basis $(u_{1},...,u_{s})$ of $\ker (\Phi _{\vec{X}})$ to a basis$$(u_{1},...,u_{s},u_{s+1},...,u_{d})$$of the whole space $F_{n}(\mathbf{F}_{q})$ determines crucially the symmetric bilinear form we get by setting $\left\langle u_{i},u_{j}\right\rangle :=\delta _{ij}$ $\forall $ $i,j\in \{1,...,d\}.$ Consequently, the orthogonal solution of $\Phi _{\vec{X}}(g)=\vec{b}$ may vary according to the chosen extension $u_{s+1},...,u_{d}\in F_{n}(\mathbf{F}_{q}).$ One systematic way to get a basis of the whole space $F_{n}(\mathbf{F}_{q})$ starting with a basis $(u_{1},...,u_{s})$ of $\ker (\Phi _{\vec{X}})$ is the following: let$$\left( \vec{y}_{1},...,\vec{y}_{s}\right) ^{t} \label{MatrixU}$$be the matrix whose rows are the coordinate vectors $\vec{y}_{1},...,\vec{y}_{s}\in K^{d}$ of $(u_{1},...,u_{s})$ with respect to the basis $(g_{nq\alpha })_{\alpha \in M_{q}^{n}}$ of $F_{n}(\mathbf{F}_{q}).$ Now we perform Gauss-Jordan elimination on the matrix (\[MatrixU\]), obtaining the matrix $R.$ Now consider the set $B:=\{\vec{e}_{1},...,\vec{e}_{d}\}$ of canonical unit vectors of the space $\mathbf{F}_{q}^{d}.$ For every pivot element $r_{ij}$ used during the Gauss-Jordan elimination performed on ([MatrixU]{}), eliminate the canonical unit vector $\vec{e}_{j}$ from the set $B.$ This yields the set $\tilde{B}.$ The coordinate vectors for a basis for the whole space $F_{n}(\mathbf{F}_{q})$ are now given by the the rows of $R$ and the vectors in the set $\tilde{B}.$ We call this way of construction of the orthonormal basis for the space $F_{n}(\mathbf{F}_{q})$ the *standard orthonormalization.* We illustrate the algorithm using an example: Suppose $q=3$, $\mathbf{F}_{3}=\mathbb{Z} _{3},$ $m=4$, $d=3^{2}=9,$ $s=5$ and that after performing Gauss-Jordan elimination on (\[MatrixU\]) we get the following matrix$$R:=\left( \begin{array}{ccccccccc} 1 & 0 & z_{1,3} & 0 & 0 & z_{1,6} & 0 & z_{1,8} & z_{1,9} \\ 0 & 1 & z_{2,3} & 0 & 0 & z_{2,6} & 0 & z_{2,8} & z_{2,9} \\ 0 & 0 & 0 & 1 & 0 & z_{3,6} & 0 & z_{3,8} & z_{3,9} \\ 0 & 0 & 0 & 0 & 1 & z_{4,6} & 0 & z_{4,8} & z_{4,9} \\ 0 & 0 & 0 & 0 & 0 & 0 & 1 & z_{5,8} & z_{5,9}\end{array}\right) \label{ReducedMat}$$(The $z_{i,j}\in \mathbf{F}_{q}$ stand for unspecified field elements). Then for the extension of the basis we choose the following canonical basis vectors$$\vec{e}_{3},\vec{e}_{6},\vec{e}_{8},\vec{e}_{9}\in \mathbb{Z} _{3}^{9}$$Now we substitute coordinate vectors $\left( \vec{y}_{1},...,\vec{y}_{5}\right) $ of the basis $(u_{1},...,u_{5})$ by the rows in the reduced matrix \[ReducedMat\] (this step is not strictly necessary, but it will be needed to prove the theorems below) and get the following coordinate vectors for a basis for the whole space $F_{2}(\mathbb{Z} _{3})$$$(\widetilde{\vec{y}_{1}},...,\widetilde{\vec{y}_{s}},\vec{y}_{s+1},...,\vec{y}_{d}):=\left( R^{t},\vec{e}_{3},\vec{e}_{6},\vec{e}_{8},\vec{e}_{9}\right)$$In this specific example we use the standard lexicographic ordering on $\left( \mathbb{N} _{0}\right) ^{2}$ and so we have$$M_{3}^{2}=\{(2,2),(2,1),(2,0),(1,2),(1,1),(1,0),(0,2),(0,1),(0,0)\}$$and$$(g_{23\alpha }(\vec{x}))_{\alpha \in M_{3}^{2}}=\left( x_{2}^{2}x_{1}^{2},x_{2}^{2}x_{1},x_{2}^{2},x_{2}x_{1}^{2},x_{2}x_{1},x_{2},x_{1}^{2},x_{1},1\right)$$Thus the orthonormal basis $(\widetilde{u_{1}},...,\widetilde{u_{s}},u_{s+1},...,u_{d})$ of $F_{2}(\mathbb{Z} _{3})$ evaluated at the point $\vec{x}\in \mathbb{Z} _{3}^{2}$ would be$$\left( \begin{array}{c} x_{2}^{2}x_{1}^{2}+z_{1,3}x_{2}^{2}+z_{1,6}x_{2}+z_{1,8}x_{1}+z_{1,9} \\ x_{2}x_{1}^{2}+z_{2,3}x_{2}^{2}+z_{2,6}x_{2}+z_{2,8}x_{1}+z_{2,9} \\ x_{2}x_{1}^{2}+z_{3,6}x_{2}+z_{3,8}x_{1}+z_{3,9} \\ x_{2}x_{1}+z_{4,6}x_{2}+z_{4,8}x_{1}+z_{4,9} \\ x_{1}^{2}+z_{5,8}x_{1}+z_{5,9} \\ x_{2}^{2} \\ x_{2} \\ x_{1} \\ 1\end{array}\right) ^{t}$$and the orthogonal solution of $\Phi _{\vec{X}}(g)=\vec{b}$ is a vector in $Span(x_{2}^{2}$ $,$ $x_{2}$ $,$ $x_{1}$ $,$ $1).$ In the next section, we will establish the exact relationship between the orthogonal solution of $\Phi _{\vec{X}}(g)=\vec{b}$ (using the symmetric bilinear form defined above) and the normal form with respect to the vanishing ideal $I(X).$ This relationship can be established if the order relation $>$ used to order the $n$-tuples in the set $M_{q}^{n}$ is a *monomial ordering*. If, more generally, total orderings on $\left( \mathbb{N} _{0}\right) ^{n}$ are used to order the set $M_{q}^{n},$ the set of possible orthogonal solutions of $\Phi _{\vec{X}}(g)=\vec{b}$ can be seen as a wider class of normal forms (with respect to vanishing ideals) in which the “classical” normal forms (attached to monomial orderings) appear as special cases. Orthogonal solutions of $\Phi _{\vec{X}}(g)=\vec{b}$ and the normal form with respect to the vanishing ideal $I(X)$ =================================================================================================================== In this section we will show the main result of this article: Given a set of points $X\subset K^{n}$, an arbitrary polynomial $f\in K[\tau _{1},...,\tau _{n}]$ and a monomial order $>,$ the normal form of $f$ with respect to the vanishing ideal $I(X)\subseteq K[\tau _{1},...,\tau _{n}]$ can be calculated as the orthogonal solution of$$\Phi _{\vec{X}}(g)=\vec{b}$$where $\vec{b}$ is given by$$b_{i}:=\widetilde{f}(\vec{x}_{i}),\text{ }i=1,...,m$$The yet undefined notation $\widetilde{f}$ suggests that a mapping between the ring $K[\tau _{1},...,\tau _{n}]$ of polynomials and the vector space of functions $F_{n}(\mathbf{F}_{q})$ is needed. That mapping will be defined and characterized in the first lemma and theorem of this section. After introducing some notation we arrive at an important preliminary result in Theorem \[GroebnerBasis\], which states how a (particular) basis of $\ker (\Phi _{\vec{X}})$ can be extended to a Gröbner basis of $I(X).$ With that result our goal can be easily reached. Please note that through this section a more technical result stated and proved in the appendix is used. Let $K$ be a field, $n,q\in \mathbb{N} $ natural numbers and $K[\tau _{1},...,\tau _{n}]$ the polynomial ring in $n$ indeterminates over $K.$ Then the set of all polynomials of the form$$\sum_{\alpha \in M_{q}^{n}}a_{\alpha }\tau _{1}^{\alpha _{1}}...\tau _{n}^{\alpha _{n}}\in K[\tau _{1},...,\tau _{n}]$$with coefficients $a_{\alpha }\in K$ is a vector space over $K.$ We denote this set with $P_{q}^{n}(K)\subset K[\tau _{1},...,\tau _{n}].$ The easy proof is left to the reader. Let $\mathbf{F}_{q}$ be a finite field and $n\in \mathbb{N} $ a natural number. Then the vector spaces $P_{q}^{n}(\mathbf{F}_{q})$ and $F_{n}(\mathbf{F}_{q})$ are isomorphic. After defining the linear mapping$$\begin{aligned} \varphi &:&P_{q}^{n}(\mathbf{F}_{q})\rightarrow F_{n}(\mathbf{F}_{q}) \\ g &=&\sum_{\alpha \in M_{q}^{n}}a_{\alpha }\tau _{1}^{\alpha _{1}}...\tau _{n}^{\alpha _{n}}\mapsto \varphi (g)(\vec{x}):=\sum_{\alpha \in M_{q}^{n}}a_{\alpha }\overrightarrow{x}^{\alpha }\end{aligned}$$the claim follows easily. The mapping $\varphi $ is defined on the set $P_{q}^{n}(K)\subset K[\tau _{1},...,\tau _{n}],$ but of course it can naturally be extended to $K[\tau _{1},...,\tau _{n}]$ as$$\begin{aligned} \varphi &:&K[\tau _{1},...,\tau _{n}]\rightarrow F_{n}(\mathbf{F}_{q}) \\ g &=&\sum_{\alpha \in \Gamma }a_{\alpha }\tau _{1}^{\alpha _{1}}...\tau _{n}^{\alpha _{n}}\mapsto \varphi (g)(\vec{x}):=\sum_{\alpha \in \Gamma }a_{\alpha }\overrightarrow{x}^{\alpha }\end{aligned}$$where $\Gamma $ is a finite set of multi indexes. We denote the image under $\varphi :K[\tau _{1},...,\tau _{n}]\rightarrow F_{n}(\mathbf{F}_{q})$ of a polynomial $g\in K[\tau _{1},...,\tau _{n}]$ with$$\widetilde{g}:=\varphi (g)\in F_{n}(\mathbf{F}_{q})$$ Let $d\in \mathbb{N} $ be a natural number, $V$ a $d$-dimensional vector space over a field $K$ and $F$ a basis of $V.$ Furthermore, let $U\subset V$ be an arbitrary *proper* subspace of $V.$ Now let $s:=\dim (U)\in \mathbb{N} .$ A basis $(u_{1},...,u_{s})$ of $U$ is called a *cleaned kernel basis with respect to the basis* $F$ if the matrix $\left( \vec{y}_{1},...,\vec{y}_{s}\right) ^{t}$ whose rows are the coordinate vectors $\vec{y}_{1},...,\vec{y}_{s}\in K^{d}$ of $(u_{1},...,u_{s})$ with respect to the basis $F$ is in reduced row echelon form. For a tuple $\vec{x}=(x_{1},...,x_{n})$ we write $x:=\{x_{1},...,x_{n}\}$ for the set containing all the entries in the tuple $\vec{x}.$ \[GroebnerBasis\]Let $\mathbf{F}_{q}$ be a finite field, $n,m\in \mathbb{N} $ natural numbers with $m<q^{n}$ and $>$ a fixed monomial order. Further let$$\vec{X}:=(\vec{x}_{1},...,\vec{x}_{m})\in \left( \mathbf{F}_{q}^{n}\right) ^{m}$$be a tuple of $m$ different $n$-tuples with entries in the field $\mathbf{F}_{q}$ and $s:=\dim (\ker (\Phi _{\vec{X}})).$ In addition, let $(u_{1},...,u_{s})$ be a cleaned kernel basis of $\ker (\Phi _{\vec{X}})\subseteq F_{n}(\mathbf{F}_{q})$ with respect to the basis $(g_{nq\alpha })_{\alpha \in M_{q}^{n}}$. Then the family of polynomials$$\left( \tau _{1}^{q}-\tau _{1},\tau _{2}^{q}-\tau _{2},...,\tau _{n}^{q}-\tau _{n},\varphi ^{-1}(u_{1}),...,\varphi ^{-1}(u_{s})\right)$$is a Gröbner basis of the vanishing ideal $I(X)\subseteq \mathbf{F}_{q}[\tau _{1},...,\tau _{n}]$ with respect to the monomial order $>.$ The idea of the proof is to show that$$U:=\left( \tau _{1}^{q}-\tau _{1},\tau _{2}^{q}-\tau _{2},...,\tau _{n}^{q}-\tau _{n},\varphi ^{-1}(u_{1}),...,\varphi ^{-1}(u_{s})\right)$$generates the ideal $I(X)$ and that for any polynomial $g\in I(X)$ the remainder on division of $g$ by $U$ is zero. According to a well known fact about Gröbner bases (see proposition 5.38 of [@MR1213453]) this is equivalent to $U$ being a Gröbner basis for $I(X).$ For this proof, remember that the fundamental monomial functions $(g_{nq\alpha })_{\alpha \in M_{q}^{n}}$ are ordered decreasingly with respect to the order $>.$Now let $g\in I(X)\subseteq \mathbf{F}_{q}[\tau _{1},...,\tau _{n}]$ be an arbitrary polynomial in the vanishing ideal of $X.$ Since$$\left( \tau _{1}^{q}-\tau _{1},\tau _{2}^{q}-\tau _{2},...,\tau _{n}^{q}-\tau _{n}\right)$$is a universal Gröbner basis for $I(\mathbf{F}_{q}^{n})$ (see Theorem \[Th.UniversalBasis\] in the appendix), there is a unique$r\in \mathbf{F}_{q}[\tau _{1},...,\tau _{n}]$ with the properties 1. No term of $r$ is divisible by any of $LT(\tau _{1}^{q}-\tau _{1})=\tau _{1}^{q},LT(\tau _{2}^{q}-\tau _{2})=\tau _{2}^{q},...,LT(\tau _{n}^{q}-\tau _{n})=\tau _{n}^{q}.$ That means in particular $r\in P_{q}^{n}(\mathbf{F}_{q}).$ 2. There is a $q\in I(\mathbf{F}_{q}^{n})$ such that $g=q+r$ This means that when we start to divide $g$ by the (ordered) family $U$ we get the intermediate result$$g=q+r$$where the remainder $r\in P_{q}^{n}(\mathbf{F}_{q})$ and $q\in \left\langle \tau _{1}^{q}-\tau _{1},\tau _{2}^{q}-\tau _{2},...,\tau _{n}^{q}-\tau _{n}\right\rangle =I(\mathbf{F}_{q}^{n}).$ If $r=0,$ then we are done and the remainder $\bar{g}^{U}$ on division of $g$ by $U$ is zero. If $r\neq 0,$ then we know from$$r=g-q$$that $r\in I(X)$ ($q\in I(\mathbf{F}_{q}^{n})\subseteq I(X)$) and this is equivalent to$$\widetilde{r}(\vec{x})=\varphi (r)(\vec{x})=0\text{ }\forall \text{ }\vec{x}\in \mathbf{F}_{q}^{n}\Leftrightarrow \widetilde{r}\in \ker (\Phi _{\vec{X}})$$Since $(u_{1},...,u_{s})$ is a basis for $\ker (\Phi _{\vec{X}}),$ there are unique $\lambda _{i}\in \mathbf{F}_{q},$ $i=1,...,s$ with$$\widetilde{r}=\sum_{i=1}^{s}\lambda _{i}u_{i}$$Applying the vector space isomorphism $\varphi ^{-1}:F_{n}(\mathbf{F}_{q})\rightarrow P_{q}^{n}(\mathbf{F}_{q})$ to this equation yields$$r=\sum_{i=1}^{s}\lambda _{i}\varphi ^{-1}(u_{i})$$From the requirement on $(u_{1},...,u_{s})$ to be a cleaned kernel basis of $\ker (\Phi _{\vec{X}})$ now follows for each $j\in \{1,...,s\},$ that the leading term$$LT(\varphi ^{-1}(u_{j}))$$doesn’t appear in the polynomials $\varphi ^{-1}(u_{i}),$ $i\in \{1,...,s\}\backslash \{j\}.$ Consequently, in the expression$$\sum_{i=1}^{s}\lambda _{i}\varphi ^{-1}(u_{i})$$no cancellation of the leading terms $LT(\varphi ^{-1}(u_{i})),$ $i=1,...,s$ can occur. Therefore, the division of $r=\sum_{i=1}^{s}\lambda _{i}\varphi ^{-1}(u_{i})$ by $\left( \varphi ^{-1}(u_{1}),...,\varphi ^{-1}(u_{s})\right) $ must yield$$r=\sum_{i=1}^{s}\lambda _{i}\varphi ^{-1}(u_{i})+0$$and the remainder $\bar{g}^{U}$ on division of $g$ by $U$ is zero. As a consequence,$$g\in \left\langle \tau _{1}^{q}-\tau _{1},\tau _{2}^{q}-\tau _{2},...,\tau _{n}^{q}-\tau _{n},\varphi ^{-1}(u_{1}),...,\varphi ^{-1}(u_{s})\right\rangle$$and since $g\in I(X)$ was arbitrary$$I(X)\subseteq \left\langle \tau _{1}^{q}-\tau _{1},\tau _{2}^{q}-\tau _{2},...,\tau _{n}^{q}-\tau _{n},\varphi ^{-1}(u_{1}),...,\varphi ^{-1}(u_{s})\right\rangle$$The inclusion$$\left\langle \tau _{1}^{q}-\tau _{1},\tau _{2}^{q}-\tau _{2},...,\tau _{n}^{q}-\tau _{n},\varphi ^{-1}(u_{1}),...,\varphi ^{-1}(u_{s})\right\rangle \subseteq I(X)$$is given by the fact $u_{1},...,u_{s}\in \ker (\Phi _{\vec{X}})$ and Theorem \[Th.UniversalBasis\]. Summarizing we can say$$\left\langle \tau _{1}^{q}-\tau _{1},\tau _{2}^{q}-\tau _{2},...,\tau _{n}^{q}-\tau _{n},\varphi ^{-1}(u_{1}),...,\varphi ^{-1}(u_{s})\right\rangle =I(X)$$and for every $g\in I(X)$ the remainder $\bar{g}^{U}$ on division of $g$ by $U$ is zero. Now proposition 5.38 of [@MR1213453] (see also the remarks after corollary 2, chapter 2, § 6 of [@MR1417938]) proves the claim. Let $\mathbf{F}_{q}$ be a finite field, $n,m\in \mathbb{N} $ natural numbers with $m<q^{n}$ and $>$ a fixed monomial order. Further let$$\vec{X}:=(\vec{x}_{1},...,\vec{x}_{m})\in (\mathbf{F}_{q}^{n})^{m}$$be a tuple of $m$ different $n$-tuples with entries in the field $\mathbf{F}_{q}$, $\vec{b}\in \mathbf{F}_{q}^{m}$ a vector, $d:=\dim (F_{n}(\mathbf{F}_{q}))$ and $s:=\dim (\ker (\Phi _{\vec{X}})).$ In addition, let $(u_{1},...,u_{s})$ be a *cleaned* kernel basis of $\ker (\Phi _{\vec{X}})\subseteq F_{n}(\mathbf{F}_{q})$ with respect to the basis $(g_{nq\alpha })_{\alpha \in M_{q}^{n}}$, $(u_{1},...,u_{s},u_{s+1},...,u_{d})$ an orthonormal basis of $F_{n}(\mathbf{F}_{q})$ constructed *using the standard orthonormalization* and $f\in \mathbf{F}_{q}[\tau _{1},...,\tau _{n}]$ a polynomial satisfying the interpolation conditions$$\widetilde{f}(\vec{x}_{j})=b_{j}\text{ }\forall \text{ }j\in \{1,...,m\}$$Furthermore, let $U\subseteq I(X)$ be an arbitrary Gröbner basis of the vanishing ideal $I(X)$ with respect to the monomial order $>$ and $v^{\ast }$ the orthogonal solution of $\Phi _{\vec{X}}(g)=\vec{b}$. Then$$\varphi ^{-1}(v^{\ast })=\overline{f}^{U}$$ If $\varphi ^{-1}(v^{\ast })=0$ then $v^{\ast }=0$ and$$\vec{b}=\Phi _{\vec{X}}(v^{\ast })=\Phi _{\vec{X}}(0)=\vec{0}$$In this case we also have$$\overline{f}^{U}=0$$and therefore$$\varphi ^{-1}(v^{\ast })=\overline{f}^{U}$$Assume $\varphi ^{-1}(v^{\ast })\neq 0.$ Since the remainder on division by a Gröbner basis is independent of which Gröbner basis we use (for a fixed monomial order), the idea of the proof is to show that $\varphi ^{-1}(v^{\ast })$ is the unique remainder on division by the Gröbner basis$$\left( \tau _{1}^{q}-\tau _{1},\tau _{2}^{q}-\tau _{2},...,\tau _{n}^{q}-\tau _{n},\varphi ^{-1}(u_{1}),...,\varphi ^{-1}(u_{s})\right)$$(see Theorem \[GroebnerBasis\]). Now, since $\varphi ^{-1}(v^{\ast })\in P_{q}^{n}(\mathbf{F}_{q}),$ no term of $\varphi ^{-1}(v^{\ast })$ is divisible by any of the$$LT(\tau _{1}^{q}-\tau _{1})=\tau _{1}^{q},LT(\tau _{2}^{q}-\tau _{2})=\tau _{2}^{q},...,LT(\tau _{n}^{q}-\tau _{n})=\tau _{n}^{q}$$If terms of $\varphi ^{-1}(v^{\ast })$ would be divisible by$$LT(\varphi ^{-1}(u_{1})),...,LT(\varphi ^{-1}(u_{s}))$$then after division by the family$$\left( \tau _{1}^{q}-\tau _{1},\tau _{2}^{q}-\tau _{2},...,\tau _{n}^{q}-\tau _{n},\varphi ^{-1}(u_{1}),...,\varphi ^{-1}(u_{s})\right)$$we would have$$\varphi ^{-1}(v^{\ast })=\sum_{i=1}^{s}h_{i}\varphi ^{-1}(u_{i})+r \label{Relation}$$where $h_{i},r\in \mathbf{F}_{q}[\tau _{1},...,\tau _{n}],$ $i=1,...,s$ and either $r=0$ or no term of $r$ is divisible by the$$LT(\tau _{1}^{q}-\tau _{1}),...,LT(\tau _{n}^{q}-\tau _{n}),LT(\varphi ^{-1}(u_{1})),...,LT(\varphi ^{-1}(u_{s}))$$If $r=0$, then$$\varphi ^{-1}(v^{\ast })=\sum_{i=1}^{s}h_{i}\varphi ^{-1}(u_{i})$$and the polynomial $\varphi ^{-1}(v^{\ast })$ vanishes on the set $X,$ that is$$\varphi (\varphi ^{-1}(v^{\ast }))(\vec{x})=v^{\ast }(\vec{x})=0\text{ }\forall \text{ }\vec{x}\in X$$Consequently$$\vec{b}=\Phi _{\vec{X}}(v^{\ast })=\vec{0}$$and due to the uniqueness of the orthogonal solution$$v^{\ast }=0$$But this is a contradiction to our assumption $\varphi ^{-1}(v^{\ast })\neq 0.$Now if $r\neq 0,$ since no term of $r$ is divisible by $LT(\tau _{1}^{q}-\tau _{1}),...,LT(\tau _{n}^{q}-\tau _{n}),$ then in particular $r\in P_{q}^{n}(\mathbf{F}_{q}).$ Due to the fact, that $(u_{1},...,u_{s},u_{s+1},...,u_{d})$ is a basis for $F_{n}(\mathbf{F}_{q}),$ we can write$$\widetilde{r}=\varphi (r)=\sum_{j=1}^{d}\lambda _{j}u_{j}$$with unique $\lambda _{j}\in \mathbf{F}_{q},$ $j=1,...,d.$ Applying the vector space isomorphism $\varphi ^{-1}:F_{n}(\mathbf{F}_{q})\rightarrow P_{q}^{n}(\mathbf{F}_{q})$ to this equation yields$$r=\sum_{j=1}^{d}\lambda _{j}\varphi ^{-1}(u_{j})$$From the requirement on $(u_{1},...,u_{s})$ to be a cleaned kernel basis of $\ker (\Phi _{\vec{X}})$ with respect to the basis $(g_{nq\alpha })_{\alpha \in M_{q}^{n}}$ and since the basis extension $(u_{1},...,u_{s},u_{s+1},...,u_{d})$ has been constructed using the standard orthonormalization, in the expression$$\sum_{j=1}^{d}\lambda _{j}\varphi ^{-1}(u_{j})$$no cancellation of the leading terms $LT(\varphi ^{-1}(u_{k})),$ $k=1,...,s$ can occur. But $r$ is not divisible by $LT(\varphi ^{-1}(u_{1})),...,LT(\varphi ^{-1}(u_{s}))$ and that forces$$\lambda _{k}=0,\text{ }\forall \text{ }k\in \{1,...,s\}$$In other words$$r=\sum_{j=s+1}^{d}\lambda _{j}\varphi ^{-1}(u_{j})\Leftrightarrow \widetilde{r}=\varphi (r)=\sum_{j=s+1}^{d}\lambda _{j}u_{j}$$which is equivalent to$$\widetilde{r}\in \ker (\Phi _{\vec{X}})^{\perp } \label{Ortho}$$From the equation (\[Relation\]) we know that$$r=\varphi ^{-1}(v^{\ast })-\sum_{i=1}^{s}h_{i}\varphi ^{-1}(u_{i})$$and that means$$\widetilde{r}(\vec{x})=v^{\ast }(\vec{x})\text{ }\forall \text{ }\vec{x}\in X$$In other words$$\Phi _{\vec{X}}(\widetilde{r})=\vec{b}$$This together with (\[Ortho\]) says that $\widetilde{r}$ is an orthogonal solution of $\Phi _{\vec{X}}(g)=\vec{b}.$ From the uniqueness now follows$$v^{\ast }=\widetilde{r}\Leftrightarrow \varphi ^{-1}(v^{\ast })=r$$Consequently, no term of the polynomial $\varphi ^{-1}(v^{\ast })$ is divisible by any of the leading terms of the elements of the Gröbner basis (see Theorem \[GroebnerBasis\])$$G:=\left( \tau _{1}^{q}-\tau _{1},\tau _{2}^{q}-\tau _{2},...,\tau _{n}^{q}-\tau _{n},\varphi ^{-1}(u_{1}),...,\varphi ^{-1}(u_{s})\right)$$for the vanishing ideal $I(X).$ Now we define the polynomial$$h:=f-\varphi ^{-1}(v^{\ast })$$Since $v^{\ast }$ is a solution of $\Phi _{\vec{X}}(g)=\vec{b}$ and $f$ satisfies the interpolation conditions$$\widetilde{f}(\vec{x}_{j})=b_{j}\text{ }\forall \text{ }j\in \{1,...,m\}$$we have$$\widetilde{h}(\vec{x})=\widetilde{f}(\vec{x})-v^{\ast }(\vec{x})=0\text{ }\forall \text{ }\vec{x}\in X\Leftrightarrow h\in I(X)$$So we have a polynomial $h\in I(X)$ such that$$f=h+\varphi ^{-1}(v^{\ast })$$By proposition 1, chapter 2, §6 in [@MR1417938], $\varphi ^{-1}(v^{\ast })$ is the unique remainder on division by the Gröbner basis $G.$ It is a well known fact, that the remainder on division by a Gröbner basis is independent of which Gröbner basis we use, as long as we use one fixed particular monomial order. Therefore$$\overline{f}^{U}=\overline{f}^{G}=\varphi ^{-1}(v^{\ast })\text{ \ \ }{}$$ Let $\mathbf{F}_{q}$ be a finite field, $n,m\in \mathbb{N} $ natural numbers with $m<q^{n}$ and $>$ a fixed monomial order. Further let$$\vec{X}:=(\vec{x}_{1},...,\vec{x}_{m})\in (\mathbf{F}_{q}^{n})^{m}$$be a tuple of $m$ different $n$-tuples with entries in the field $\mathbf{F}_{q}$, $U\subseteq I(X)$ an arbitrary Gröbner basis of the vanishing ideal $I(X)$ and $f\in \mathbf{F}_{q}[\tau _{1},...,\tau _{n}]$ an *arbitrary* polynomial. Then$$\overline{f}^{U}=\varphi ^{-1}(v^{\ast })$$where $v^{\ast }$ is the orthogonal solution of $\Phi _{\vec{X}}(g)=\vec{b}$ and $\vec{b}$ is given by$$b_{i}:=\widetilde{f}(\vec{x}_{i}),\text{ }i=1,...,m$$ \[MatrixCalculation\]Let$$A:=(\Phi _{\vec{X}}(g_{nq\alpha }))_{\alpha \in M_{q}^{n}}\in M(m\times q^{n};\mathbf{F}_{q})$$be the matrix representing the evaluation epimorphism $\Phi _{\vec{X}}$ of the tuple $\vec{X}$ with respect to the basis $(g_{nq\alpha })_{\alpha \in M_{q}^{n}}$ of $F_{n}(\mathbf{F}_{q})$ and the canonical basis of $\mathbf{F}_{q}^{m}$ and $S$ the matrix$$S_{ij}:=\left\langle g_{nq\alpha _{i}},g_{nq\alpha _{j}}\right\rangle ,\text{ }i,j\in \{1,...,q^{n}\}$$representing the symmetric bilinear form with respect to the basis $(g_{nq\alpha })_{\alpha \in M_{q}^{n}}$. Further let$\vec{y}_{1},...,\vec{y}_{s}\in \mathbf{F}_{q}^{d}$ be the coordinate vectors of $(u_{1},...,u_{s})$ with respect to the basis $(g_{nq\alpha })_{\alpha \in M_{q}^{n}}$. Then the above result states that the normal form $\overline{f}^{U}$ of $f$ with respect to the Gröbner basis $U\subseteq I(X)$ can be calculated by solving the following system of inhomogeneous linear equations$$\begin{aligned} A\vec{z} &=&\vec{b} \\ \vec{y}_{i}^{t}S\vec{z} &=&0,\text{ }i=1,...,s\end{aligned}$$ Acknowledgements ================ We would like to thank Dr. Gretchen Matthews, Dr. Michael Shapiro and Dr. Michael Stillman for very helpful comments and contributions for the content of this paper. Appendix ======== Let $K$ be a field, $n\in \mathbb{N} $ a natural number, $K[\tau _{1},...,\tau _{n}]$ the polynomial ring in $n$ indeterminates over $K$ and $>$ an arbitrary monomial order. Then for each natural number $m\in \mathbb{N} $ and each $i\in \{1,...,n\}$ it holds$$\tau _{i}^{m}>\tau _{i}^{m-1}>...>\tau _{i}>\tau _{i}^{0} \label{Inequa.}$$ The claim follows from the well-ordering, the translation invariance and transitivity of $>.$ Let $\mathbf{F}_{q}$ be a finite field and $n\in \mathbb{N} $ a natural number. Then the family of polynomials$$\left( \tau _{1}^{q}-\tau _{1},\tau _{2}^{q}-\tau _{2},...,\tau _{n}^{q}-\tau _{n}\right)$$is a basis for the vanishing ideal$$I(\mathbf{F}_{q}^{n})\subseteq \mathbf{F}_{q}[\tau _{1},...\tau _{n}]$$ The proof of this well-known result can be found after Lemma 3.1 of [@Germundsson]. \[Th.UniversalBasis\]Let $\mathbf{F}_{q}$ be a finite field and $n\in \mathbb{N} $ a natural number. Then the family of polynomials$$\left( \tau _{1}^{q}-\tau _{1},\tau _{2}^{q}-\tau _{2},...,\tau _{n}^{q}-\tau _{n}\right)$$is a universal Gröbner basis for the vanishing ideal$$I(\mathbf{F}_{q}^{n})\subseteq \mathbf{F}_{q}[\tau _{1},...\tau _{n}]$$ From the inequalities \[Inequa.\] it follows in particular for all possible monomial orders$$LM(\tau _{i}^{q}-\tau _{i})=\tau _{i}^{q}\text{ }\forall \text{ }i\in \{1,...,n\}$$As a consequence, for the least common multiple ($LCM$) of $LM(\tau _{j}^{q}-\tau _{j})$ and $LM(\tau _{i}^{q}-\tau _{i}),$ $i\neq j$ holds$$LCM(LM(\tau _{j}^{q}-\tau _{j}),LM(\tau _{i}^{q}-\tau _{i}))=LCM(\tau _{j}^{q},\tau _{i}^{q})=\tau _{j}^{q}\tau _{i}^{q}\text{ }\forall \text{ }i,j\in \{1,...,n\}\text{ with }i\neq j$$and for the $S$-polynomial of $\tau _{j}^{q}-\tau _{j}$ and $\tau _{i}^{q}-\tau _{i},$ $i\neq j$ we have$$S(\tau _{j}^{q}-\tau _{j},\tau _{i}^{q}-\tau _{i})=\tau _{i}^{q}(\tau _{j}^{q}-\tau _{j})-\tau _{j}^{q}(\tau _{i}^{q}-\tau _{i})=\tau _{j}^{q}\tau _{i}-\tau _{i}^{q}\tau _{j}\text{ }\forall \text{ }i,j\in \{1,...,n\}\text{ with }i\neq j$$Now let’s divide $S(\tau _{j}^{q}-\tau _{j},\tau _{i}^{q}-\tau _{i})=\tau _{j}^{q}\tau _{i}-\tau _{i}^{q}\tau _{j}$ by $\left( \tau _{1}^{q}-\tau _{1},\tau _{2}^{q}-\tau _{2},...,\tau _{n}^{q}-\tau _{n}\right) .$ Without loss of generality let$$\tau _{j}^{q}\tau _{i}>\tau _{i}^{q}\tau _{j}$$(which is equivalent to $LT(\tau _{j}^{q}\tau _{i}-\tau _{i}^{q}\tau _{j})=\tau _{j}^{q}\tau _{i}$). Then, after the first division step, we get the remainder$$-\tau _{i}^{q}\tau _{j}+\tau _{i}\tau _{j}$$Now we know from the inequalities (\[Inequa.\]) after translation by $\tau _{j}$$$\tau _{i}^{q}\tau _{j}>\tau _{i}\tau _{j}\Rightarrow LT(-\tau _{i}^{q}\tau _{j}+\tau _{i}\tau _{j})=-\tau _{i}^{q}\tau _{j}$$so we can continue the division process and we get the remainder$$-\tau _{i}^{q}\tau _{j}+\tau _{i}\tau _{j}-(-\tau _{j})(\tau _{i}^{q}-\tau _{i})=0$$By the theorem above$$I(\mathbf{F}_{q}^{n})=\left\langle \tau _{1}^{q}-\tau _{1},\tau _{2}^{q}-\tau _{2},...,\tau _{n}^{q}-\tau _{n}\right\rangle$$And so, according to Buchberger’s $S$-pair criterion (see Theorem 6 of chapter 2, §6 in [@MR1417938]),$$\left( \tau _{1}^{q}-\tau _{1},\tau _{2}^{q}-\tau _{2},...,\tau _{n}^{q}-\tau _{n}\right)$$is a universal Gröbner Basis for $I(\mathbf{F}_{q}^{n}).$ [^1]: Centre for Mathematical Sciences, Munich University of Technology, Boltzmannstr.3, 85747 Garching, Germany. [^2]: Pathology Department, Tufts University, 150 Harrison Av., Boston, MA 02111, USA (correspondence address). [^3]: The author acknowledges support by a Public Health Service grant (RO1 AI062989) to David Thorley-Lawson at Tufts University, Boston, MA.

ArXiv

--- abstract: 'In 1999 Berry and Keating showed that a regularization of the 1D classical Hamiltonian $H = xp$ gives semiclassically the smooth counting function of the Riemann zeros. In this paper we first generalize this result by considering a phase space delimited by two boundary functions in position and momenta, which induce a fluctuation term in the counting of energy levels. We next quantize the $x p$ Hamiltonian, adding an interaction term that depends on two wave functions associated to the classical boundaries in phase space. The general model is solved exactly, obtaining a continuum spectrum with discrete bound states embbeded in it. We find the boundary wave functions, associated to the Berry-Keating regularization, for which the average Riemann zeros become resonances. A spectral realization of the Riemann zeros is achieved exploiting the symmetry of the model under the exchange of position and momenta which is related to the duality symmetry of the zeta function. The boundary wave functions, giving rise to the Riemann zeros, are found using the Riemann-Siegel formula of the zeta function. Other Dirichlet L-functions are shown to find a natural realization in the model.' author: - Germán Sierra title: A quantum mechanical model of the Riemann zeros --- =cmss12 =cmu10 scaled1 [H]{} 0.2cm | $${\left[} \def$$[\]]{} ${\left(} \def\BL{\Bigr(} \def$[)]{} \#1\#2[[\#1 \#2]{}]{} \#1[[1 \#1]{}]{} \#1[[\#1]{}]{} \#1\#2[[\#1 \#2]{}]{} \#1[\#1 ]{} \#1[ | \#1 ]{} pi \#1[[e]{}\^[\^[\#1]{}]{}]{} \#1[\_[[\#1]{}]{}]{} \#1\#2[  \#1 , \#2 ]{} pi \#1[[e]{}\^[\^[\#1]{}]{}]{} \#1[\_[[\#1]{}]{}]{} =cmss12 =cmu10 scaled1 Ł[[L]{}]{} Introduction ============ At the beginning of the XX century Polya and Hilbert made the bold conjecture that the imaginary part of the Riemann zeros could be the oscillation frequencies of a physical system. If true this suggestion would imply a proof of the celebrated Riemann hypothesis (RH). The importance of this conjecture lies in its connection with the prime numbers. If the RH is true then the statistical distribution of the primes will be constrained in the most favorable way [@Edwards; @Titchmarsh2]. Otherwise, in the words of Bombieri, the failure of the RH would create havoc in the distribution of the prime numbers [@Bombieri] (see also [@Sarnak; @Conrey; @Watkins; @Rosu; @Elizalde] for reviews on the RH). After the advent of Quantum Mechanics, the Polya-Hilbert conjecture was formulated as the existence of a self-adjoint operator whose spectrum contains the imaginary part of the Riemann zeros. This conjecture was for a long time regarded as a wild speculation until the works of Selberg in the 50’s and those of Montgomery in the 70’s. Selberg found a remarkable duality between the length of geodesics on a Riemann surface and the eigenvalues of the Laplacian operator defined on it [@Selberg]. This duality is encapsulated in the so called Selberg trace formula, which has a strong similarity with the Riemann explicit formula relating the zeros and the prime numbers. The Riemann zeros would correspond to the eigenvalues, and the primes to the geodesics. This classical versus quantum version of the primes and the zeros is also at the heart of the so called Quantum Chaos approach to the RH. Quite independently of Selberg«s work, Montgomery showed that the Riemann zeros are distributed randomly and obeying locally the statistical law of the Random Matrix Theory (RMT) [@Mont]. The RMT was originally proposed to explain the chaotic behaviour of the spectra of nuclei but it has applications in another branches of Physics, specially in Condensed Matter [@Mehta]. There are several universality classes of random matrices, and it turns out that the one related to the Riemann zeros is the gaussian unitary ensemble (GUE) associated to random hermitean matrices. Montgomery analytical results found an impressive numerical confirmation in the works of Odlyzko in the 80’s, so that the GUE law, as applied to the Riemann zeros is nowadays called the Montgomery-Odlyzko law [@Odl]. An important hint suggested by this law is that the Polya-Hilbert Hamiltonian $H$ must break the time reversal symmetry. The reason being that the GUE statistics describes random Hamiltonians where this symmetry is broken. A simple example is provided by materials with impurities subject to an external magnetic field, as in the Quantum Hall effect. A further step in the Polya-Hilbert-Montgomery-Odlyzko pathway was taken by Berry [@B-chaos; @Berry1]. who noticed a similarity between the formula yielding the fluctuations of the number of zeros, around its average position $E_n \sim 2 \pi n/ \log n$, and a formula giving the fluctuations of the energy levels of a Hamiltonian obtained by the quantization of a classical chaotic system [@Gutzwiller]. The comparison between these two formulas suggests that the prime numbers $p$ correspond to the isolated periodic orbits whose period is $\log p$. In the Quantum Chaos scenario the prime numbers appear as classical objects, while the Riemann zeros are quantal. This classical/quantum interpretation of the primes/zeros is certainly reminiscent of the one underlying the Selberg trace formula mentioned earlier. A success of the Quantum Chaos approach is that it explains the deviations from the GUE law of the zeros found numerically by Odlyzko. The similarity between the fluctuation formulas described above, while rather appealing, has a serious drawback observed by Connes which has to do with an overall sign difference between them [@Connes]. It is as if the periodic orbits were missing in the underlying classical chaotic dynamics, a fact that is difficult to understand physically. This and other observations lead Connes to propose an abstract approach to the RH based on discrete mathematical objects known as adeles [@Connes]. The final outcome of Connes work is a trace formula whose proof, not yet found, amounts to that of a generalized version of the RH. In Connes approach there is an operator, which plays the role of the Hamiltonian, whose spectrum is a continuum with missing spectral lines corresponding to the Riemann zeros. We are thus confronted with two possible physical realizations of the Riemann zeros, either as point like spectra or as missing spectra in a continuum. Later on we shall see that both pictures can be reconciled in a QM model having a discrete spectra embedded in a continuum. The next step within the Polya-Hilbert framework came in 1999 when Berry and Keating [@BK1; @BK2] on one hand and Connes [@Connes] on the other, proposed that the classical Hamiltonian $H = x p$, where $x$ and $p$ are the position and momenta of a 1D particle, is closely related to the Riemann zeros. This striking suggestion was based on a semiclassical analysis of $H = x p$, which led these authors to reach quite opposite conclusions regarding the possible spectral interpretation of the Riemann zeros. The origin of the disagreement is due to the choice of different regularizations of $H = xp$. Berry and Keating choosed a Planck cell regularization in which case the smooth part of the Riemann zeros appears semiclassically as discrete energy levels. Connes, on the other hand choosed an upper cutoff for the position and momenta which gives semiclassically a continuum spectrum where the smooth zeros are missing. All these semiclassical results are heuristic and lack so far of a consistent quantum version. It is the aim of this paper to provide such a quantum version in the hope that it will sed new light concerning the spectral realization of the Riemann zeros. The organization of the paper is as follows. In section II we review the semiclassical approaches to $H = xp$ due to Berry, Keating and Connes which give an heuristic derivation of the asymptotic behaviour of the smooth part of the Riemann zeros. Then, we generalize the semiclassical Berry-Keating Planck cell regularization of $xp$ by means of two classical functions which define a [*wiggly* ]{} boundary for the allowed semiclassical region in phase space. This generalization allow us to explain semiclassically the fluctuation term in the spectrum. In section III we define the quantum Hamiltonian associated to the semiclassical approach introduced above. The Hamiltonian is given by the quantization of $H= xp$ plus an interaction term that depends on two generic boundary wave functions associated to the classical boundary functions of the semiclassical approach. In section IV we solve the Schroedinger equation finding the exact eigenfunctions and eigenenergies in terms of a function $\F(E)$ which plays the role of a Jost function for this model, and whose analyticity properties are studied in section V. In section VI we find the boundary wave functions that give rise to the quantum version of the semiclassical Berry-Keating model for the smooth zeros of the Riemann zeta function, which are common to all the even Dirichlet L-functions. We also find the boundary wave functions associated to the smooth approximation of the zeros of the odd Dirichlet L-functions. In section VII we quantize the relation between the fluctuation part of the spectrum and the semiclassical phase boundaries, obtaining the equations satisfied by the boundary wave functions, and we solve them explicitely. Finally, using the duality properties of these wave functions and the Riemann-Siegel formula of the zeta function we find a model whose Jost function is proportional to the zeta function. From this fact, and making some additional asumptions, we show that the Riemann zeros on the critical line are bound states of the model. However we cannot exclude the existence of zeros outside the critical line, which would imply a proof of the RH. We describe in an appendix the computation of the wave functions associated to the smooth and exact Riemann zeros. The present work is closely related to those in references [@Sierra1; @Sierra2; @Sierra3], where we studied an interacting version of the $xp$ Hamiltonian based on the relation of this model with the so called Russian doll model of superconductivity [@RD1; @RD2; @links]. For a field theoretical approach to the RH inspired by the latter works see reference [@Andre-RH]. We would like also to mention some important differences between the present paper and those of references [@Sierra1; @Sierra2; @Sierra3]. First of all, the position variable $x$ was choosen in [@Sierra1; @Sierra2; @Sierra3] to belong to the finite interval $(1,N)$ with $N \rightarrow \infty$, while in this paper we choose the half line $(0, \infty)$ which gives a more symmetric treatment between the position and momentum variables. Secondly, in the earlier references the interaction term was added to the inverse Hamiltonian $1/(xp)$, while in this paper we add the interaction directly to the Hamiltonian $xp$, which is more natural from a physical viewpoint. We have also tried to make an extensive use of the duality symmetry of the Riemann zeta function reflected in the functional relation it satisfies. Semiclassical approach ====================== The classical Berry-Keating-Connes (BKC) Hamiltonian [@BK1; @BK2; @Connes] H\^[cl]{}\_[0]{} = x p, \[s1\] has classical trayectories given by the hyperbolas (see fig.1a) x(t) = x\_0 e\^[t]{} , p(t) = p\_0 e\^[-t]{}. \[s2\] The dynamics is unbounded, so one should not expect a discrete spectrum even at the semiclassical level. To overcome this difficulty, Berry and Keating proposed in 1999 to restrict the phase space of the $xp$ model to those points $(x,p)$ where $|x| > l_x$ and $|p| > l_p$, with $l_x \, l_p = 2 \pi \hbar$. These constraints lead to a finite number of semiclassical states, $\CN(E)$, with energy between 0 and $E$ given by (E) = , \[s3\] where $A$ is the area of the allowed phase space region below the curve $E = x p$. The result, in units $\hbar = 1$, is \_[BK]{}(E) = ( -1 ) + 1 \[s4\] which agrees with the asymptotic limit of the smooth part of the formula giving the number of Riemann zeros whose imaginary part lies in the interval $(0,E)$, (E) \~ ( -1 ) + + O(E\^[-1]{}). \[s5\] The exact formula for the number of zeros, $\CN_R(E)$, due to Riemann, also contains a fluctuation term which depends on the zeta function [@Edwards] (see fig.\[n-formula\]), \_R(E) & = & (E) + \_[fl]{}(E) \[s6\]\ (E) & = & + 1\ \_[fl]{}(E) & = & ( + i E ) where $\theta(E)$ is the phase of the Riemann zeta function $\zeta(1/2 - i E)$, (E) = ( + E ) - \[s7\] whose asymptotic expansion (E) = ( ) - - + O(E\^[-1]{}) \[s7-1\] yields (\[s5\]). The function $\zeta(s)$, for $\Re \; s > 1$, can be related to the prime numbers $p$ thanks to the Euler product formula (s) = \_[p &gt; 1]{} , s &gt; 1 \[s7-2\] This expression diverges if $\Re \; s = 1/2$, however one can heuristically use it to write the fluctuation term in (\[s6\]) as \_[fl]{}(E) = - \_[p]{} \_[m=1]{}\^ ( m E p) \[s8\] which gives a reasonable result after truncating the sum over the primes. As observed by Berry, eq.(\[s8\]) resembles formally the fluctuation part of the spectrum of a classical 1D chaotic Hamiltonian with isolated periodic orbits \_[fl]{}(E) = \_[\_p]{} \_[m=1]{}\^ (S\_[cl]{}(E)) \[s9\] where $\gamma_p$ denotes the primitive periodic orbits, the label $m$ describes the windings of those orbits, $ \pm \lambda_p$ are the instability exponents and $S_{\rm cl}(E)$ is the classical action, which is equal to $m E T_{\gamma_p}$, with $ T_{\gamma_p}$ the period of $\gamma_p$. Comparing (\[s8\]) and (\[s9\]), Berry conjectured the existence of a classical chaotic Hamiltonian whose primitive periodic orbits would be labelled by the prime numbers $p=2,3, \dots$, with periods $T_p= \log p$ and instability exponents $\lambda_p = \pm \log p$ [@B-chaos; @Berry1]. Moreover, since each orbit is counted once, the Hamiltonian must break time reversal (otherwise there would be a factor $2/\pi$ in front of eq. (\[s8\]) instead of $1/\pi$). The quantization of this classical chaotic Hamiltonian would likely contain the Riemann zeros in its spectrum. This idea is the key of the Quantum Chaos approach to the Riemann hypothesis. Besides the fact that the earlier Hamiltonian has not yet been found there is the Connes criticism that the similarity between eqs.(\[s8\]) and (\[s9\]) fails in two issues. The first is the overall minus sign in (\[s8\]) as compared to (\[s9\]), and the second is that the term $2 \; {\rm sinh}(m \lambda_p/2)$ only becomes $p^{m/2}$ when $m \rightarrow \infty$. Connes relates the [*minus sign*]{} problem to an alternative interpretation of the Riemann zeros as missing spectral lines as opposed to the conventional one (we shall come back later to these conflicting interpretations). These two problems were the main Connes’s motivations to develop the adelic approach to the RH. As we saw above, the Quantum Chaos approach suggests that the fluctuation part of the spectrum of the yet unknown Riemann Hamiltonian has a classical origin related to the prime numbers. Taking into account the Berry-Keating heuristic derivation of the smooth part of the spectrum, it is tempting to extend the semiclassical approach in order to explain the fluctuation term in the Riemann formula for the zeros. The simplest idea is to generalize the allowed phase space of the $xp$ Hamiltonian replacing the boundaries $|x| = l_x$ and $|p| = l_p$ by two curves $x_\cl(p)$ and $p_\cl(x)$, such that (see fig 1b) x &gt; x\_(p), |p| &gt; p\_(x) \[s10\] where $x_\cl(p)$ and $p_\cl(x)$, are positive functions satisfying x\_(p) = x\_(-p)& &gt; 0, & p \[s11\]\ p\_(x) & &gt;0 & x \_+ These conditions split the allowed phase space into two disconnected regions in the first and forth quadrants of the $x p$ plane. Notice that $x$ is always positive while $p$ can be either positive or negative. The BK boundaries obviously correspond to the choice : x\_(p)= l\_x, p\_(x) = l\_p \[s12\] For the extended BC’s the minimal distance $l_x$ and minimal momentum $l_p$ can be defined as the intersection point of the curves, $x_\cl(p)$ and $p_\cl(x)$, which we shall assume to be unique, and satisfying x\_(l\_p)= l\_x, p\_(l\_x) = l\_p \[s13\] The classical $x p$ Hamiltonian together with the BK conditions have the exchange symmetry \[s14\] whose generalization to the extended model is = \[s15\] The counting of semiclassical states is based again on eq. (\[s3\]). The area below the curve $E = x p$ and bounded by the conditions (\[s10\]) is given by (see fig.1b) A & = & \_[l\_x]{}\^[x\_I]{} dx \_[p\_(x)]{}\^[l\_p x/l\_x]{} dp + \_[x\_I]{}\^[x\_M]{} d x \_[p\_(x)]{}\^[E/x]{} dp \[s16\]\ &+ & \_[l\_p]{}\^[p\_I]{} dp \_[x\_(p)]{}\^[l\_x p/l\_p]{} dx + \_[p\_I]{}\^[p\_M]{} dp \_[x\_(p)]{}\^[E/p]{} dx The quantities $x_M$, $p_M$ (resp. $x_I, p_I$) are the position and momenta of the points where the curve $E = x p$ intersects the boundaries $p_\cl(x)$, $x_\cl(p)$ (resp. the line $x/l_x = p/l_p$), and satisfy, E = x\_M p\_(x\_M) = x\_(p\_M) p\_M = x\_I p\_I, = \[s17\] The integration of (\[s16\]) yields A & = & E ( ) + E - l\_x l\_p \[s18\]\ & - & E ( ) - E ( )\ & - & \_[l\_x]{}\^[x\_M]{} dx p\_(x) - \_[l\_p]{}\^[p\_M]{} dp x\_(p) Partial integrating the last two terms in (\[s18\]) and dividing by $h = l_x l_p = 2 \pi (\hbar =1)$, the semiclassical value of $\CN(E)$ reads (E) & = & \_[BK]{}(E) \[s19\]\ & - & ( ) - ( )\ & + & \_[l\_x]{}\^[x\_M]{} x + \_[l\_p]{}\^[p\_M]{} p The BK conditions (\[s12\]) of course reproduce eq. (\[s4\]). More general boundary functions induce a fluctuation term in the counting formula of a form which recalls eq.(\[s6\]). Let us denote this term as n\_[fl]{} (E) & = & - ( ) - ( ) \[s20\]\ & + & \_[l\_x]{}\^[x\_M]{} x + \_[l\_p]{}\^[p\_M]{} p so that (E) = \_[BK]{}(E) + n\_(E) \[s20-1\] Taking the derivative of (\[s20\]) with respect to $E$, and using eqs.(\[s17\]) one gets = - ( ) - ( ) \[s21\] which implies that the boundary functions are related to the fluctuation part of the density of states. A further simplification is achieved imposing the $x p$ symmetry (\[s15\]) = , = \[s22\] which leads to = - ( ) = - ( ) \[s23\] Hence, $x p$-symmetric boundary functions $p_\cl(x_M)$ and $x_\cl(p_M)$ are completely fixed by the density of the fluctuations. To find $p_\cl(x)$, one combines (\[s23\]) and (\[s17\]) p\_(x\_M) = l\_p e\^[ - n’\_[fl]{}(E)]{} = , n’\_[fl]{}(E) = \[s24\] which gives $x_M$ as a function of $E$ x\_M = e\^[ n’\_[fl]{}(E)]{} \[s25\] If $n_{\rm fl}(E)=0$ , the latter equations reproduce the BK boundary conditions (\[s13\]). Eq.(\[s25\]) gives $x_M$ as a function of $E$ and it is monotonically increasing provided &gt; 0 1 + E &gt; 0 \[s26\] Under this condition we can expressed $E$ as a function of $x_M$ and replaced it in (\[s24\]), obtaining the boundary function $p_{x_M} = E(x_M)/x_M$. In this case the inverse problem of finding a Hamiltonian given the spectrum has a unique solution at the semiclassical level. If the fluctuations are strong enough at some energies, then condition (\[s26\]) could be violated implying that $E = E(x)$ as well as $p_\cl(x)$ will be multivalued functions. This gives rise to a manifold of boundary functions, each one having discontinuities at some values of $x$. From classical to quantum ========================= In this section we shall give a quantum version of the semiclassical results obtained above. The starting point is the quantization of the classical hamiltonian $H_0^\cl= xp$. Let us consider the usual normal ordered expression H\_0 = ( x p + p x) = - i ( x + ) \[t1\] where $p = - id/dx$. In references [@Sierra2; @Twamley] it was shown that $H_0$ becomes a self-adjoint operator in two cases where the domain of the $x$ variable are choosen as: 1) $ 0 < x < \infty $ or 2) $a < x < b$ with $a$ and $b$ finite. For the purposes of this paper we shall confine to the case 1. Case 2 was discussed at length in [@Sierra2]. Since $x > 0$ one can write (\[t1\]) as H\_0 = x\^[1/2]{} p x\^[1/2]{}, x &gt;0 \[t2\] The exact eigenfunctions of (\[t2\]) are given by \_E(x) = , E \[t3\] where the eigenenergies $E$ belong to the real line. The normalization of (\[t3\]) is the appropiate one for a continuum spectra, \_E| \_[E’]{} = \_0\^dx \_E\^\*(x) \_[E’]{}(x) = (E - E’). \[t4\] The quantum Hamiltonian associated to the semiclassical approach is H = H\_0 + i ( | \_a \_b | - | \_b \_a | ) \[t5\] where $\psi_a$ and $\psi_b$ are two wave functions associated to the boundary functions $p_\cl(x)$ and $x_\cl(p)$, respectively, i.e. \_a p\_(x), \_b x\_(p) \[t6\] We shall choose real functions $\psi_a(x)$ and $\psi_b(x)$ so that $H$ is an hermitean and antisymmetric operator, which implies that the eigenvalues appear in pairs ${E, -E}$. The interaction term in (\[t5\]) can be justified by the following heuristic argument. Let us consider a particle which at $t=0$ belong to the classical allowed region, i.e. $x_0 > x_\cl(p_0) $ and $p_0 > p_\cl(x_0)$. According to the classical evolution (\[s2\]), the position $x(t)$ increases while the momenta $p(t)$ decreases, i.e. (x\_0, p\_0) (e\^t x\_0, e\^[-t]{} p\_0) \[t7\] until a time $t_M$ where the particle hits the $p_\cl$-boundary. (e\^[t\_M]{} x\_0, e\^[-t\_M]{} p\_0) = (x\_M, p\_(x\_M)) \[t8\] The semiclassical approach suggests to transport this particle from the $p_\cl$-boundary to a point in the $x_\cl$-boundary, (see fig. \[teletransport\]) (x\_M, p\_(x\_M)) (x\_(p\_M), p\_M), \[t9\] while preserving the energy, E = x\_0 p\_0 = x\_M p\_(x\_M) = x\_(p\_M) p\_M \[t10\] Equation (\[t10\]) coincides with (\[s17\]) if we choose $(x_0, p_0) = (x_I, p_I)$. The transported particle at the $x_\cl$- boundary continues its classical evolution returning to the initial point $(x_0, p_0)$ after a time \_E = = \[t11\] This is also the period of the classical trayectory which has become a closed orbit thanks to the transport operation (\[t9\]). The semiclassical calculation of the previous section measures classical action associated to this periodic orbit. At the quantum level the free evolution of a state $\psi$ is given by the unitary transformation |(0) |(t) = e\^[- i t H\_0]{} |(0) \[t12\] The operator that performs the transport (\[t9\]) is given by one of the interacting terms in the Hamiltonian (\[t5\]), |-i |\_b\_a| \[t13\] which consists in the proyection of the state $\psi$ into the quantum state $\psi_a$, yielding the state $\psi_b$ as a result. The hermiticity of the Hamiltonian $H$ implies the existence of the inverse of the process (\[t13\]), i.e. |-i |\_a\_b| \[t14\] whose classical analogue is (see fig. \[teletransport\]b), (x\_(p\_M), p\_M), (x\_M, p\_(x\_M)) \[t15\] What is the physical meaning of this process? Let us take for a while a particle in the classical forbbiden region where $x_0 < x_\cl(p_0) $ but $p_0 > p_\cl(x_0)$. This particle will evolve freely according to eqs.(\[t7\]), until a time $t_M$ where it hits the $x_\cl$-boundary, i.e. (e\^[t\_M]{} x\_0, e\^[-t\_M]{} p\_0) = (x\_(p\_M), p\_M)) \[t16\] Then one can apply the inverse transport (\[t15\]) which carries the particle to the $p_\cl$-boundary where it continues its free and unbounded evolution : $x \rightarrow \infty$ and $p \rightarrow 0$. The phase space area traced by this trayectory is infinite which implies that the number of these kind of semiclassical states is infinite forming therefore a continuum. In summary, the transport operations between the two boundaries leads classically to closed periodic trayectories in the allowed phase space and to open trayectories in the forbbiden region. Semiclassically the closed periodic trayectories give rise to bound states while the open ones form a continuum. This is scenario that comes out from the solution of the quantum model, as we show in the next section. The existence of a semiclassical continuum in the $x p$ model was proposed by Connes in reference [@Connes]. Instead of the boundary conditions set by $l_x$ and $l_p$, Connes restricts the phase space of the model to be $|x| < \Lambda$, $|p| < \Lambda$, where $\Lambda$ is a cutoff which is sent to infinite at the end of the calculation. The number of semiclassical states is given now by \_C(E) = - ( -1 ) \[t17\] where the first term leads, in the limit $\Lambda \rightarrow \infty$, to a continuum while the second term coincides with minus the average position of the Riemann zeros (\[s4\]). A possible interpretation of these result is that the Riemann zeros, are missing spectral lines in a continuum, which is in apparent contradiction with the Berry-Keating interpretation of the zeros as bound states. As we shall show below both interpretations can be reconciled at the quantum level where the Riemann zeros appear as discrete spectra embbeded in a continuum of states. Exact solution of the Schroedinger equation =========================================== In this section we shall find explicitely the eigenstates and the eigenergies of the Hamiltonian (\[t5\]) for generic states $\psi_a$ and $\psi_b$. The method used is similar to the one employed in reference [@Sierra2], where instead of the Hamiltonian $xp$ we added an interaction to $1/xp$. The Schroedinger equation for an eigenstate $\psi_E(x)$ with energy $E$ is given by - i ( x + ) \_E(x) + i ( \_a(x) \_b| \_E - \_b(x) \_a| \_E = E \_E(x) ) \[u1\] Let us introduce the variable $q$ q= x, q \[u2\] and the overlap integrals A & = & \_a| \_E = \_0\^dx \_a(x) \_E(x) \[u3\]\ B & = & \_b| \_E = \_0\^dx \_b(x) \_E(x) which depend on $E$. Using these definitions eq.(\[u1\]) becomes - i ( + ) \_E(q) + i ( B \_a(q) - A \_b(q) ) = E \_E(q) \[u4\] The general solution of this equation is given by \_E(q) = e\^[- (1/2 - i E) q ]{} \[u5\] where $C_0$ is an integration constant. It is convenient to define the functions a(q) & = & e\^[q/2]{} \_a(q), \_a(x) = \[u6\]\ b(q) & = & e\^[q/2]{} \_b(q), \_b(x) = so that \_E(q) = e\^[- (1/2 - i E) q ]{} \[u7\] An alternative way to express (\[u7\]) is \_E(q) = e\^[- (1/2 - i E) q ]{} \[u8\] where $C_\infty$ is related to $C_0$ by C\_= C\_0 + B (-E) - A (-E) \[u9\] where (E) = \_[- ]{}\^dq e\^[i E q]{} f(q), f= a, b \[u10\] We shall assume that $a(q)$ and $b(q)$ satisfy & \_[q - ]{} \_[-]{}\^q d q’ e\^[- i E q’]{} f(q’) = 0, f = a, b & \[u11\]\ & \_[q ]{} \_q\^ d q’ e\^[- i E q’]{} f(q’) = 0, f = a, b & which implies that the asymptotic behaviour of $\psi_E(x)$ is dominated by $C_0, C_\infty$, i.e. \_[x 0]{} \_E(x) = , \_[x ]{} \_E(x) = , \[u12\] Plugging (\[u7\]) into (\[u3\]) yields the relation between the constants $A,B,C_0$, ( [cc]{} 1 + S\_[a,b]{} & - S\_[a,a]{}\ S\_[b,b]{} & 1 - S\_[b,a]{} ) ( [c]{} A\ B ) = C\_0 ( [c]{} (E)\ (E) ) \[u13\] where the functions $S_{f,g}(E)$ with $f,g = a,b$ are defined by [@s-function] S\_[f,g]{}(E) = \_[- ]{}\^dq e\^[i E q]{} f(q) \_[-]{}\^q dq’ e\^[-i E q’]{} g(q’) \[u14\] Similarly, introducing (\[u8\]) into (\[u3\]) yields ( [cc]{} 1 - \_[a,b]{} & \_[a,a]{}\ - \_[b,b]{} & 1 + \_[b,a]{} ) ( [c]{} A\ B ) = C\_ ( [c]{} (E)\ (E) ) \[u15\] where \_[f,g]{}(E) = \_[- ]{}\^dq e\^[i E q]{} f(q) \_q\^ dq’ e\^[-i E q’]{} g(q’) \[u16\] This function is related to $S_{f,g}$ in two ways, \_[f,g]{}(E) & = & - S\_[f,g]{}(E) + (E) (-E) \[u18\]\ \_[f,g]{}(E) & = & S\_[g,f]{}(-E) \[u19\] To derive these equations one makes a change of order in the integration. Combining (\[u18\]) and (\[u19\]) one obtains the [*shuffle*]{} relation S\_[f,g]{}(E) + S\_[g,f]{}(-E) = (E) (-E) \[u20\] The terminology is borrowed from the theory of multiple zeta functions where there is a similar relation between the two variable Euler-Zagier zeta function $\zeta(s_1, s_2)$, and the Riemann zeta function $\zeta(s)$ [@euler-zagier-1; @euler-zagier-2]. The solutions of the eqs.(\[u13\]) and (\[u15\]) depend on the determinant of the associated $2 \times 2$ matrices given by (E)& = & 1 + S\_[a,b]{} - S\_[b,a]{} + S\_[a,a]{} S\_[b,b]{} - S\_[a,b]{} S\_[b,a]{} \[u21\]\ (E)& = & 1 - \_[a,b]{} + \_[b,a]{} + \_[a,a]{} \_[b,b]{} - \_[a,b]{} \_[b,a]{} \[u22\] which are related by (\[u19\]) (E) = (-E) \[u23\] Moreover, since $a(x)$ and $b(x)$ are real functions one has S\_[f,g]{}\^\*(E) =S\_[f,g]{}(-E\^\*) \[u24\] which in turn implies \^\*(E) = (-E\^\*) \[u25\] After these observations we can return to the solution of (\[u13\]) and (\[u15\]). We shall distinguish two cases: 1) $\F(E) \neq 0$ and 2) $\F(E) = 0$, where $E$ is real since it is an eigenvalue of the Hamiltonian (\[t5\]). [**Case 1:**]{} $\F(E) \neq 0$ Eq.(\[u25\]) implies that $\F(-E) \neq 0$ and therefore $A$ and $B$ can be expressed in two different ways, A & = &\ & = & , \[u26\]\ B & = &\ & = & Now using eq.(\[u18\]), these eqs. reduce to = \[u27\] which by eq.(\[u25\]) is a pure phase for $E$ real. Hence, up to an overall factor, the integration constants for this solution can be choosen as C\_0 & = & (E)\ C\_& = & (-E) \[u28\]\ A & = & ( 1 - S\_[b,a]{}) (E) + S\_[a,a]{} (E)\ B & = & - S\_[b,b]{} (E) + ( 1 + S\_[a,b]{}) (E). Since the constants $C_0, C_\infty$ do not vanish, the wave function is non normalizable near the origin and infinity (recall eq. (\[u12\])) and therefore they correspond to scattering states. Of course they will be normalizable in the distributional sense. [**Case 2:**]{} $\F(E) = 0$. The integration constants can be choosen as C\_0 & = & 0\ C\_& = & 0 \[u29\]\ A & = & S\_[a,a]{}\ B & = & ( 1 + S\_[a,b]{}) which solves eqs. (\[u13\]) and (\[u15\]). Since $C_0= C_\infty = 0 $, the leading term of the behaviour of $\psi_E(x)$ vanish near the origin and infinity and under appropiate conditions on $\psi_{a,b}$, the state $\psi_E$ will be normalizable corresponding to a bound state. In the appendix we compute the norm of these states. Hence the generic spectrum of the Hamiltonian (\[t5\]) consist of a continuum covering the whole real line with, eventually, some isolated bound states embedded in it, whenever $\CF(E) = 0$. This structure also arises in the Hamiltonian studied in reference [@Sierra2]. The function $\CF(E)$ plays the role of the Jost function since its zeros gives the position of the bound states and its phase gives the scattering phase shift according to eq.(\[u27\]). Before we continue with the general formalism it is worth to study a simple case which illustrates the results obtained so far. An example: a quantum trap {#an-example-a-quantum-trap .unnumbered} -------------------------- Let us start with the classical version of a trap where a particle is restricted to the region $x_b < x < x_a$. The semiclassical number of states is given by the area formula (\[s3\]), n = = \_[x\_b]{}\^[x\_a]{} = \[u30\] which yields the eigenenergies E\_n = ,n . \[u31\] The quantum version of this model is realized by two boundary states $\psi_{a,b}(x)$ proportional to delta functions, i.e. \_a(x) = a\_0 x\_a\^[1/2]{} (x - x\_a), \_b(x) = b\_0 x\_b\^[1/2]{} (x - x\_b). \[u32-1\] The associated potentials $a(q), b(q)$ are & a(q) = a\_0 (q - q\_a), b(q) = b\_0 (q - q\_b), & \[u32\]\ & q\_[a]{} = x\_[a]{}, q\_[b]{} = x\_[b]{} & The various quantities defined above are readily computed obtaining & = & a\_0 e\^[i E q\_a]{}, = b\_0 e\^[i E q\_b]{}\ S\_[a,a]{} & = & , S\_[b,b]{} = \[u33\]\ S\_[a,b]{} & = & a\_0 b\_0 e\^[i E q\_[a,b]{}]{} S\_[b,a]{} = 0 where $q_{a,b} = q_a - q_b = \log(x_a/ x_b)$. Plugging these eqs. into (\[u21\]) yields (E) = 1 + ( )\^2 + a\_0 b\_0 e\^[i E q\_[a,b]{}]{} \[u34\] For generic values of $a_0, b_0$, the Jost function (\[u34\]) never vanishes obtaining a spectrum which is continuous. However, $\CF(E)$ vanishes provided the following condition holds = 1 (E) = 2 ( 1 + e\^[i E q\_[a,b]{}]{}) \[u35\] in which cases the spectrum contains bound states embbeded in the continuum with energies & = & 1 E\_n = n \[u36\]\ [If]{} & = & - 1 E\_n = n that agree with the semiclassical energies (\[u31\]) for $n >> 1$. The unnormalized wave function of the bound states, i.e. $\CF(E) = 0$, can be computed from eq. (\[u7\]) \_[E]{}(x) = { [ll]{} 1, & x\_b &lt; x &lt; x\_a\ 0, & x &lt; x\_b x &gt; x\_a . \[u37\] which shows that they are confined to the region $(x_b, x_a)$. The wave functions when $\CF(E) \neq 0$ can be similarly found obtaining \_[E]{}(x) = { [ll]{} (E), & 0 &lt; x &lt; x\_b\ 1 - ( )\^2, & x\_b &lt; x &lt; x\_a\ (-E), & x\_a &lt; x &lt;\ . \[u38\] Hence if (\[u35\]) holds, these wave functions vanishes in the region $(x_b, x_a)$ which contains the trapped particles (\[u37\]). In this example the mechanism responsible for the existence of bound states is the transport of the particles from the position $x_a$ to the position $x_b$. At the quantum level the confinement requires the fine tuning of the couplings (see eq. (\[u35\])), which introduces periodic or antiperiodic boundary conditions depending on the sign of $\ep$. When $|\ep| \neq 1 $ the particle can scape the trap and the bound states become resonances. Analyticity properties of $\CF(E)$ ================================== As in ordinary Quantum Mechanics, the Jost function $\CF(E)$ satisfy certain analyticity properties reflecting the causal structure of the dynamics. In our case these properties follows from those of the function $S_{f,g}$ (eq. (\[u14\])) and the definition (\[u21\]). Indeed, let us express $S_{f,g}(E)$ in terms of the Fourier transforms of the functions $f,g$. First we replace $g(q)$ by its inverse Fourier transform g(q’) = \_[- ]{}\^ e\^[i E q’]{} (-E’) \[a1\] back into eq.(\[u14\]), obtaining S\_[f,g]{}(E) = \_[- ]{}\^ (-E’) \_[- ]{}\^dq e\^[i E q]{} f(q) \_[- ]{}\^q dq’ e\^[i (E’ - E) q’]{}. \[a2\] The last integral is given by the distribution \_[- ]{}\^q dq’ e\^[i (E’ - E) q’]{} = e\^[i q (E’ - E)]{} \[a3\] where $P$ denotes the Cauchy principal part. Plugging (\[a3\]) into (\[a2\]) and using the Fourier transform of $f$ gives, S\_[f,g]{}(E) = \[a4\] Alternatively, one can write (\[a4\]) as S\_[f,g]{}(E) = \_[-]{}\^ \[a5\] with $\ep > 0$ an infinitesimal. Eq. (\[a5\]) shows that the poles of $S_{f,g}(E)$ are located in the lower half of the complex energy plane. Thus for well behave functions $\hf, \hg$, the function $S_{f,g}(E)$ will be analytic in the complex upper-half plane. These properties also apply to $\CF(E)$ which is the product of $S_{f,g}$ functions with $f,g = a,b$. Another important property of the Jost function $\CF(E)$ is that its zeros lie either on the real axis or below it, i.e. (E) = 0 E 0 \[a6\] The proof of this equation is similar to the one done in reference [@Sierra2], being convenient to regularize the interval $x \in (0, \infty )$ as $(N^{-1}, N)$ with $N \rightarrow \infty$. In the appendix we use the results obtained in this section to compute the norm of the eigenstates. The quantum version of the Berry-Keating model ============================================== Let us consider the BK constraints $x > l_x$ and $|p| > l_p$. It is rather natural to associate constraint $x > l_x$ with the wave function \_b(x) = b\_0 l\_x\^[1/2]{} (x- l\_x) \[r1\] which is localized at the boundary $x = l_x$. The factor $\l_x^{1/2}$ gives the correct dimensionality to $\psi_b(x)$, with $b_0$ a dimensionaless parameter. On the other hand the constraint $|p| > l_p$ admits two possible quantum versions, { [c]{} \_a\^+ (x)\ \_a\^-(x)\ . = 2 a\_0 ( )\^[1/2]{} { [c]{} ( l\_p x)\ ( l\_p x) . \[r2\] Due to the fact that $\psi_a$ has to be real, one cannot choose a pure plane wave $e^{i l_p x}$. The boundary wave functions (\[r1\]) and (\[r2\]) are the cosine and sine Fourier transform of each other, namely { [c]{} \_a\^+ (x)\ \_a\^-(x)\ . = ( )\^[1/2]{} \_0\^dy \_b(y) { [c]{} ( l\_p x y/l\_x)\ ( l\_p x y/l\_x)\ . \[r3\] Indeed, extending the domain of $\psi_b(x)$ according to the parity of $\psi_a^\eta \; (\eta = \pm) $ one gets \_b(-x) = \_b (x) \_a\^(x) = ( )\^[1/2]{} e\^[i (-1)]{} \_b ( ) \[r4\] which are the quantum analogue of the classical equations (\[s15\]). Later on, we shall consider more general wave functions $\psi_{a,b}$ to account for the fluctuations in the Riemann formula, imposing again eq.(\[r3\]). The relation (\[r3\]) between $\psi_a^{\pm}$ and $\psi_b$ must imply a close link between their Mellin transforms $\ha_\pm(E)$ and $\hb(E)$. To derive it, let us write \_(E) = \_0\^x\^[-1/2 + i E]{} \_a\^(x) = ( )\^[1/2]{} \_0\^dx x\^[-1/2 + i E]{} \_0\^dy \_b(y) { [c]{} ( l\_p x y/l\_x)\ ( l\_p x y/l\_x)\ . \[r5\] The basic integrals one needs are \_0\^dx x\^[- + i E]{} { [c]{} ( p x)\ (p x)\ . = ( )\^[ + i E]{} { [c]{} e\^[2 i \_+(E)]{}\ e\^[2 i \_-(E)]{}\ . \[r6\] where e\^[2 i \_(E)]{} = { [lll]{} \^[- i E]{} , & & = +\ \^[- i E]{} , & & = -\ . \[r7\] The function $\theta_+(E)$ coincides with the phase of the Riemann zeta function (\[s7\]), and more generally of the even Dirichlet L-functions, while $\theta_-(E)$ is the phase factor of the odd Dirichlet L-functions. These phases appear in the functional relation of even and odd $L$ functions, and they arise in our context from the two possible relations between the boundary functions $\psi_a^\pm$ and $\psi_b$. Plugging eq.(\[r6\]) into (\[r5\]) yields \_(E) = ( )\^[ i E]{} e\^[2 i \_(E)]{} \_0\^dy \_b(y) y\^[- - i E]{} \[r8\] where the integral is nothing but $\hb(-E)$, thus \_(E) = ( )\^[ i E]{} e\^[2 i \_(E)]{} (-E) \[r9\] This important equation reflects the relation (\[r3\]) which in turn is the quantum version of the $x p$ symmetry between boundaries. In the BK case, the Mellin transforms of the associated wave functions (\[r1\]) and (\[r2\]) are \_(E) = a\_0 ( )\^[ i E]{} e\^[2 i \_(E)]{}, (E) = b\_0 l\_x\^[ i E]{} \[r10\] which are pure phases, up to overall constants. The $S_{f,g}$ functions can be readily computed using eq.(\[a4\]). To do so, we first consider the products & \_(E) \_(-E) = a\_0\^2, &\ & (E) (-E) = b\_0\^2 & \[r11\]\ & \_(E) (-E) = a\_0 b\_0 e\^[2 i \_(E)]{} &\ & (E) \_(-E) = a\_0 b\_0 e\^[- 2 i \_(E)]{} & where we used $\l_x l_p = 2 \pi$ and that $\theta_\pm(-E) = - \theta_\pm(E)$. The diagonal terms of $S_{f,g}$ are given simply by S\_[a\_, a\_]{}(E) = , S\_[b,b]{}(E) = \[r12\] since the Hilbert transform of a constant is zero, i.e. P \_[- ]{}\^ = 0, E \[r13\] The computation of $S_{a_\pm,b}$ and $S_{b,a_\pm}$ uses the analytic properties of $e^{2 i \theta_\pm(E)}$. Let us focus on the case of $e^{2 i \theta_+(E)} = e^{2 i \theta(E)}$. This function converges rapidly to zero as $|E| \rightarrow \infty$ in the upper half plane, and it has poles at $E_n = i ( 2 n + 1/2) \;\; (n=0, 1, \dots)$ where it behaves like e\^[2 i (E)]{}\~ \[r14\] We can split $e^{2 i \theta(E)}$ into the sum e\^[2 i (E)]{} & = & \_+(E) + \_-(E) \[r15\]\ \_-(E)& = & \_[n=0]{}\^ where $\Omega_+(E)$ is analytic in the upper half plane and goes to zero at $+ i \infty$, while $\Omega_- (E)$ has poles in the upper half plane and behaves as $1/E$ at infinity. The function $\Omega_-(E)$ can also be written as \_-(E) & = & 2 \_0\^1 dx x\^[- 1/2 + i E]{} ( 2 x) \[r16\]\ & = & \_1F\_2 ( + i , , + i , - \^2) where $_1F_2$ is a hypergeometric function of the type $(1,2)$. From the analyticity properties of $\Omega_\pm$ one gets inmediately their Hilbert transform P \_[- ]{}\^ = \_(E), E \[r17\] Hence $S_{a_+,b} \equiv S_{a,b}$, as given by eq.(\[a4\]), becomes S\_[a,b]{}(E) & = & \[r18\]\ & = &\ & = & a\_0 b\_0 \_+(E) Similarly one finds S\_[b,a]{}(E) = a\_0 b\_0 \_-(-E) \[r19\] Notice that both functions are analytic in the upper half plane. The Jost function finally reads (E)& = & 1 + a\_0 b\_0 (\_+(E) - \_-(-E)) + ( )\^2\ & & - (a\_0 b\_0)\^2 \_+(E) \_-(-E) \[r20\] In the asymptotic limit $|E| >> 1$ \_-(E) \~ \_+(E) = e\^[2 i (E)]{} + O() \[r21\] which implies (E) = 1 + a\_0 b\_0 e\^[2 i (E)]{} + ( )\^2 + O() \[r22\] This Jost function has zeros on the real axis, up to order $1/E$, provided = = 1 (E) = 2( 1 + e\^[2 i (E)]{}) + O() \[r23\] The choice $\ep = - 1$ reproduces the smooth part of the Riemann formula (\[s6\]) since, = - 1 1 - e\^[2 i (E)]{} = 1 - e\^[2 i (E) ]{} = 0 \[r24\] where $E$ is the average position of the zeros. On the other hand the choice $\ep = 1$ leads to = 1 1 + e\^[2 i (E)]{} = 0 (E) = 0 \[r25\] so that the number of zeros in the interval $(0,E)$ is given by \_(E) = + \[r26\] which gives a better numerical approximation than the term $\langle \CN(E) \rangle$ that appears in the exact Riemann formula (\[s6\]) (see also fig.2). In the case of the sine boundary function (\[r2\]) one similarly obtains the smooth part of the zeros of the odd Dirichlet L-functions. In summary, we have shown that the semiclassical BK boundary conditions have a quantum counterpart in terms of the boundary wave functions $\psi_{a,b}$, and that the average Riemann zeros become asymptotically bound states of the model or more appropiately resonances. The quantum model of the Riemann zeros ====================================== In section II we showed how to incorporate the fluctuations of the energy levels in the heuristic $xp$ model by means of the functions $p_\cl(x)$ and $x_\cl(p)$ which define the boundaries of the allowed phase space. These functions are given by eq.(\[s23\]) in terms of the density of the fluctuation part of the energy levels. In the quantum model the functions $p_\cl(x)$ and $x_\cl(p)$ are represented by the wave functions $\psi_a$ and $\psi_b$. Hence it is natural to impose the following conditions ( + n’\_[fl]{}(H\_0) ) |\_a = & 0 & \[q1\]\ ( + n’\_[fl]{}(H\_0) ) |\_b = & 0 & \[q2\] where $n'_{\rm fl}(E) = d n_{\rm fl}(E)/dE$ and $H_0$ is the no interacting Hamiltonian (\[t1\]). The hat over $x$ and $p$ stress the fact that they are operators. Eqs.(\[q1\]) and (\[q2\]) can be taken as the definition of the boundary wave functions. To solve these eqs. let us write them as ( || + \_p + n’\_[fl]{}(H\_0) ) |\_a =& 0 &, \[q3\]\ ( + \_x + n’\_[fl]{}(H\_0) ) |\_b = & 0 &, \[q4\]\ \_p = - l\_p, \_x = - l\_x & & \[q5\] It is convenient to expand the states $|\psi_{a,b} \rangle$ in the basis (\[t3\]) |\_[a,b]{} = \_[- ]{}\^dE \_[a,b]{}(E) |\_E , x | \_E = \[q6\] Let us first consider eq.(\[q4\]) which in the basis (\[q6\]) becomes \_[- ]{}\^dE’ \_E| | \_[E’]{} \_b(E’) + ( \_x + n’\_[fl]{}(E) ) \_b(E) = 0 \[q7\] The matrix elements of the operator $\log \hx$ can be readily computed, \_E| | \_[E’]{} = - i ’(E’ - E) \[q8\] which replaced in (\[q7\]) and upon integration yields i + ( \_x + n’\_[fl]{}(E) ) \_b(E) = 0 \[q9\] The solution of (\[q9\]) is simply \_b(E) = \_[b,0]{} e\^[ i ( \_x E + n\_[fl]{}(E) ) ]{} \[q10\] where $\psi_{b,0}$ is an integration constant. The $x$-space representation of $\psi_b$ follows from (\[q10\]) and (\[q6\]) \_b(x) = \_[- ]{}\^dE \_b(E) \_E(x) = \_[b,0]{} \_[- ]{}\^ e\^[ i ( \_x E + n\_[fl]{}(E) ) ]{} x\^[-1/2 + i E]{} \[q11\] Recalling that $ \psi_b(x) = b(x)/\sqrt{x}$ one gets b(x) = \_[b,0]{} \_[- ]{}\^ e\^[ i ( \_x E + n\_[fl]{}(E) ) ]{} x\^[ i E]{} \[q12\] Observing that $b(x)$ is related to its Fourier transform $\hb(E)$, as b(x) = \_[- ]{}\^ (E) x\^[- i E]{} \[q13\] one finally obtains (E) = \_[b,0]{} e\^[ - i ( \_x E + n\_[fl]{}(E))]{} \[q14\] where we assumed that $n_{\rm fl}(E)$ is an odd function of $E$. If $n_{\rm fl}(E)=0$, eq.(\[q14\]) reproduces (\[r10\]), i.e. n\_[fl]{}(E) = 0 (E) = \_[b,0]{} l\_x\^[i E]{} = b\_0 l\_x\^[ i E]{} \[q15\] To simplify the notations we shall write (\[q14\]) as (E) = b\_0 l\_x\^[ i E]{} e\^[ - i n\_[fl]{}(E)]{} \[q16\] Let us now solve the condition (\[q3\]) for the wave function $\psi_a$. We first need to define the operator $\log |\hp|$ acting in the Hilbert space expanded by the functions $\phi_E \; (E \in \Rmath)$. In this respect it is worth to remember that the operator $\hp = - i d/dx$ is self-adjoint in the real line $(- \infty, \infty)$ and in the finite intervals $(a,b)$, but not in the half-line $(0, \infty)$ [@self]. However, the operator $\hp^2$ admits infinitely many self-adjoint extensions in the half-line provide the wave functions satisfy the boundary condition ’(0) = (0) \[q17\] where $\kappa \in \Rmath \cup \infty$. We shall confine ourselves to the cases where $\kappa = 0$ and $\infty$, which correspond to the von Neumann and Dirichlet BC’s respectively, & = & 0 ’ (0) = 0, \[q18\]\ & = & (0) = 0 The corresponding eigenstates of the operator $\hp^2$ with eigenvalues $p^2$ read { [l]{} \_p\^+\ \_p\^-\ . = { [ll]{} ( p x) & (p&gt;0)\ ( p x) & (p&gt;0)\ . \[q19\] These basis are complete in the space of functions defined in $(x >0)$, i.e. \_0\^dp (\_p\^(x))\^\* \_p\^(x’) = (x - x’), x,x’ &gt;0, = \[q20\] The operator $\log | \hp |$ will be defined as $\frac{1}{2} \log \hp^2$, and therefore admits the same self-adjoint extensions as $\hp^2$. The analogue of eq.(\[q7\]) reads now \_[- ]{}\^dE’ \_E| || | \_[E’]{} \_a(E’) + ( \_p + n’\_[fl]{}(E) ) \_a(E) = 0 \[q21\] The matrix elements of $\log |\hp|$ can be computed introducing the resolution of the identity in the basis (\[q19\]), \_E| || | \_[E’]{} = \_[0]{}\^dp p \_E| \_p\^\_p\^| \_[E’]{} \[q22\] where the overlap of the eigenstates of $\hp^2$ and $H_0$ are \_p\^| \_[E]{} = \_0\^ x\^[- + i E]{} { [c]{} ( p x)\ (p x)\ . \[q23\] These integrals were already computed in eq.(\[r6\]), and the result is \_p\^| \_[E]{} = e\^[2 i \_(E)]{} \[q24\] Plugging this eq. into (\[q22\]), and performing the integral gives \_E| || | \_[E’]{} = i ’(E’ - E) ( 2 )\^[i (E’ - E)]{} e\^[2 i (\_(E’) - \_(E))]{} \[q25\] which introduced in (\[q21\]) yields a differential equation whose solution is \_[a\_]{}(E) = \_[a,0]{} (2 )\^[- i E]{} e\^[- i ( \_p E + n\_[fl]{}(E) + 2 \_(E) ) ]{} \[q26\] The function $\psi_a(x)$ reads \_[a\_]{}(x) = \_[- ]{}\^dE \_[a\_]{}(E) \_E(x) = \_[a,0]{} \_[- ]{}\^ (2 )\^[- i E]{} e\^[ - i ( \_p E + n\_[fl]{}(E) + 2 \_(E) ) ]{} x\^[-1/2 + i E]{} \[q27\] while a\_(x) = \_[a,0]{} \_[- ]{}\^ (2 )\^[- i E]{} e\^[ - i ( \_p E + n\_[fl]{}(E) + 2 \_(E) ) ]{} x\^[i E]{} \[q28\] whose Fourier transform is \_(E) = \_[a,0]{} (2 )\^[ 1/2 + i E]{} e\^[ i ( \_p E + n\_[fl]{}(E) + 2 \_(E) ) ]{} \[q29\] If there are no fluctuations, eq.(\[q29\]) reduces to n\_[fl]{}(E) = 0 \_(E) = \_[a,0]{} ( )\^[i E]{} e\^[ 2 i \_(E) ) ]{} \[q30\] which coincides with eq.(\[r10\]). To simplify notations we shall write (\[q29\]) as \_(E) = a\_0 ( )\^[ i E]{} e\^[ i ( n\_[fl]{}(E) + 2 \_(E) ) ]{} \[q31\] The two solutions (\[q16\]) and (\[q31\]) satisfy the duality relation (\[r9\]) and hence the wave functions $\psi_{a_\pm}(x)$ is the cosine or sine Fourier transform of $\psi_b(x)$ ( see eq. (\[r3\])). Having found the boundary wave functions for generic fluctuations we turn into the computation of the corresponding Jost function. The basic products of the $\ha$ and $\hb$ functions needed to find the $S_{f,g}$ functions are similar to eqs.(\[r11\]), & \_(E) \_(-E) = a\_0\^2, &\ & (E) (-E) = b\_0\^2 & \[q32\]\ & \_(E) (-E) = a\_0 b\_0 e\^[2 i ( \_(E) + n\_[fl]{}(E)) ]{} &\ & (E) \_(-E) = a\_0 b\_0 e\^[- 2 i (\_(E) + n\_[fl]{}(E)]{} & The diagonal terms of $S_{f,g}$ are the same as in eq.(\[r12\]), i.e. S\_[a\_, a\_]{}(E) = , S\_[b,b]{}(E) = \[q33\] while the evaluation of the off-diagonal terms depends on the analytic properties of the function $e^{ 2 \pi i \; n_\pm(E)}$ where n\_(E) + n\_[fl]{}(E)) \[q34\] This definition is strongly reminiscent of the Riemann formula (\[s6\]), with $n_\pm(E)$ playing the role of $\CN_R(E)$, and $n_{\rm fl}(E)$ that of $\CN_{\rm fl}(E)$. However, we must keep in mind that $\CN_R(E)$ is a step function while we expect $n_\pm(E)$ to be a continuous interpolating function between the zeros. The value of $S_{a_\pm, b}$ is given by the integral S\_[a\_,b]{}(E) = \[q35\] We shall make the asumption that $e^{2 \pi i n_\pm(E)}$ is an analytic function in the upper half plane which goes to zero as $|E| \rightarrow \infty$. In this case the Cauchy integral on the RHS of (\[q35\]) is equal to $e^{2 \pi i n_\pm(E)}$ and one finds S\_[a\_,b]{}(E) = a\_0 b\_0 e\^[2 i n\_(E)]{} \[q36\] Similarly $S_{b, a_\pm}$ vanishes so that the Jost function reduces to (E) = 1 + a\_0 b\_0 e\^[2 i n\_(E)]{} + ( )\^2 \[q37\] and under the usual choice = = 1 (E) = 2( 1 + e\^[2 i n\_(E)]{}) \[q38\] When $n_{\rm fl}=0$ the results of the previous subsection showed that $\ep = 1$ gives a better numerical estimate to the smooth part of the zeros. In the sequel we shall also make that choice which implies that the number of zeros of $\CF(E)$ in the interval $(0,E)$ is \_(E) =\_(E) + n\_(E) = n\_(E) + \[q39\] where $\CN_\sm(E)$ was defined in (\[r26\]) for the particular case of the zeta function $\zeta(s)$, which corresponds to $n_+(E)$. Equation (\[q39\]) agrees asymptotically with the semiclassical formula (\[s20-1\]), which confirms the ansatz made for the states $\psi_a$ and $\psi_b$. The connection with the Riemann-Siegel formula {#the-connection-with-the-riemann-siegel-formula .unnumbered} ---------------------------------------------- The next problem is to find the function $n_\fl(E)$, and therefore $\CN_\QM(E)$, which gives the exact location of the Riemann zeros. Let us consider the case of the zeta function with the following choices of parameters = +, = 1, a\_0 = b\_0 = , l\_x = 1, l\_p = 2 \[q40\] which correspond to the potentials (recall (\[q31\]) and (\[q16\])) (t) & = & e\^[i ( 2 (t) + n\_(t) )]{} = e\^[i ( (t) + n(t) )]{} \[q41\]\ (t) & = & e\^[-i n\_(t) )]{} = e\^[i ((t) - n(t) )]{} where we skip a common factor $\sqrt{2}$ and denote $n(E) \equiv n_+(E)$. These two functions are interchanged under the transformation (t) & & e\^[2 i (t)]{} (-t) = (t) \[q42\]\ (t) & & e\^[2 i (t)]{} (-t) = (t) so that their sum is left invariant, (t) + (t) e\^[2 i (t)]{} ((-t) + (-t)) = (t) + (t) \[q43\] The functional relation satisfied by the zeta function implies (1/2 - i t) e\^[2 i (t)]{} (1/2 + i t) = (1/2 - i t) \[q44\] which suggests to relate $\ha + \hb$ and $\zeta$ as (1/2 - i t) = (t) ((t) + (t)) \[q45\] where $\rho(t)$ is a proportionally factor. Using eqs.(\[q42\]) into (\[q45\]) yields (1/2 - i t) = 2 (t) e\^[i (t)]{} (n(t)) \[q46\] This formula can be compared with the parametrization of the zeta function in terms of the Riemann-Siegel zeta function $Z(t)$ and its phase $\theta(t)$, (1/2 - i t) = Z(t) e\^[i (t)]{} \[q47\] which leads to, Z(t) = 2 (t) (n(t)) \[q48\] This equation is rather interesting since it implies that the zeros of $\cos(\pi n(t))$, which give the bound states of the QM model, are also zeros of $Z(t)$, of course if $\rho(t)$ does not have poles at those values. Viceversa, the zeros of $Z(t)$ can be zeros either of $\cos(\pi n(t))$, or of $\rho(t)$, or both. The latter possibility would be absent if the Rieman zeros are simple, as it is expected to be the case. A first hint on the structure of the functions $\rho(t)$ and $\cos(\pi n(t))$ can be obtained using the Riemann-Siegel formula for $Z(t)$, Z(t) = 2 \_[n=1]{}\^[(t)]{} n\^[-1/2]{} ( (t) - t n) + R(t), (t) = \[q49\] where $[x]$ the integer part of $x$ and $R(t)$ is a reminder of order $t^{-1/4}$. Combining the last two equations one finds Z(t) & = & 2 (t) \[q50\]\ & \~& 2 which suggests the following identifications (t) ( n\_(t) ) & \~& \_[n=1]{}\^[(t)]{} \[q51\]\ (t) ( n\_(t) ) & \~& - \_[n=1]{}\^[(t)]{} that can be combined into f(t) (t) e\^[i n\_(t)]{} \~ \_[n=1]{}\^[(t)]{} \[q52\] The fluctuation function $n_\fl(t)$ is then given by the phase of $f(t)$, i.e. n\_(t) = f(t) \[q53\] In fig. \[nqm\] we plot the values of $\CN_{QM}(t)$ that correspond to the approximate formula (\[q52\]), which shows an excelent agreement with the Riemann formula (\[s6\]). This is expected from the fact that the main term of the Riemann-Siegel formula already gives accurate results for the lowest Riemann zeros. For higher zeros one has to compute more terms of the reminder $R(t)$ depending on the desired accuracy. Observe that $\CN_{QM}(t)$ is a smooth function, except for some jumps at higher values of $t$ (not shown in fig. \[nqm\]) due to the approximation made, unlike $\CN_R(t)$, which is a step function. In fig. \[nfl\] we plot the values of (\[q53\]) together with those of the fluctuation part of the Riemann formula (\[s6\]), i.e. \_[fl]{}(t) = ( + i t ) \[q54\] The jumps in $\CN_{\rm fl}(t)$ correspond to the Riemann zeros, while those of $n_\fl(t)$ correspond, either to jumps of the function $\nu(t)$ appearing in the Riemann Siegel formula (\[q49\]), or to those points where the curve $f(t)$ cuts the negative real axis in the complex plane. We gave in section II a formal expression of eq.(\[q54\]) in terms of prime numbers, eq. (\[s8\]), which resembles the fluctuation part (\[s9\]) of a quantum chaotic system. Eq.(\[s8\]) is based on the Euler product formula (\[s7-2\]) which is not valid in the case where $s = 1/2 + it$, since $\Re \; s > 1$ for convergence of the infinite product. The Euler product formula does not apply to the truncated sum (\[q52\]), however we shall naively try to establish a relationship. Let us denote by $p_n$ the $n^{\rm th}$-prime number, e.g. $p_1=2, p_2 = 3$, etc, and by $\Pi(x)$ the number of primes less or equal to $x$. The sum (\[q52\]) involves all integers up to $\nu(t)$, which can be expressed as products of the first $\mu(t)$ prime numbers where (t) = ( (t)), p\_[(t)]{} = [inf]{} { p } &lt; (t) \[q55\] Using these functions we define a truncated Euler product as \_(1/2 + i t) \_[n = 1]{}\^[(t)]{} \[q56\] It is easy to see that $\zeta_\E(1/2 + i t)$ is not equal to $f(t)$, for there are terms in (\[q56\]) which do not appear in (\[q52\]), although all the terms appearing in the latter sum also appear in the former product. The point is that a numerical comparison of these two functions shows a qualitative agreement as depicted in fig. \[eulerRS\]. Indeed, the minima and maxima of their absolute value are located around the same points, and the same happens for the zeros of their arguments. The conclusion we draw from these heuristic considerations is that the function $f(t)$ contains some sort of information related to the primes numbers although not in the form of an Euler product formula as is the case of $\zeta_\E(1/2 + i t)$. It would be interesting to investigate the consequences of this results from the point of view of Quantum Chaos. The Berry-Keating formula of $Z(t)$ {#the-berry-keating-formula-of-zt .unnumbered} ----------------------------------- The main term of the Riemann-Siegel formula (\[q49\]) is not analytic in $t$ due to the discontinuity in the main sum. This problem was solved by Berry and Keating who found an alternative expression for $Z(t)$ [@BK3]. The formula is Z(t) = \_[n=1]{}\^( T\_n(t) + T\_n(-t)) \[bk1\] where T\_n(t)& = & T\_n\^\*(-t)= \_n(t) \[bk2\]\ \_n(t)& = & \_[C\_-]{} e\^[- z\^2 K\^2/(2 |t|)]{} e\^[ i \[ (z+t) - (t) - z n\]]{} and $C_-$ is an integration contour in the lower half plane with $\Im \; < -1/2$ that avoids a cut starting at the brach point $z = - t - i/2$. The constant $K$ in (\[bk2\]) can be choosen at will and it is related to the number of terms of the RS formula that has been smoothed for large values of $t$. Using eq.(\[bk1\]) one can write the zeta function as (1/2 - i t) = e\^[2 i (t)]{} \_[n=1]{}\^ + \_[n=1]{}\^ \[bk3\] which can be compared with (\[q45\]) obtaining f(t) = (t) e\^[ i n\_(t)]{} = \_[n=1]{}\^ \[bk4\] so that (\[bk3\]) can be written as (1/2 - i t) = e\^[2 i (t)]{} f(t) + f(-t) \[bk3b\] Eq.(\[bk4\]) gives an exact expression of $f(t)$, which is in fact a smooth version of (\[q52\]). Berry and Keating also found a series for $Z(t)$ which improves the RS series. The first term of that series corresponds to the following value of the $\beta_n(t)$ functions \_n\^[(0)]{}(t)& = & ( ) \[bk5\]\ (n,t) & = & n - ’(t), Q\^2(K,t) = K\^2 - i t ”(t) where $Erfc$ is the complementary error function. Using these formulas one can find a better numerical evaluation of the functions $\CN_{QM}(t)$ and $n_\fl(t)$. It is perhaps worth to mention that eq.(\[bk3b\]), with the approximate value of $f(t)$ given by (\[q52\]), is a particular case of the so called aproximate functional relation due to Hardy and Littlewood [@Edwards; @Titchmarsh2] (s) = \_[n x]{} n\^[-s]{} + \^[s- 1/2]{} \_[n y]{} n\^[1-s]{} + O(x\^[ - ]{}) + O( |t|\^[1/2 - ]{} y\^[-1]{}) \[bk3c\] where $s = \sigma + i t$, $|t| = 2 \pi x y$, $0 < \sigma < 1$. Recalling that in our model $t$ is the energy $E$, then equation $|t| = 2 \pi x y$ becomes the hyperbola $|E| = x p$ with $p = 2 \pi y = l_p y$ so that the sums in (\[bk3c\]) run over the integer values of the positions and momenta in units of $l_x$ and $l_p$ respectively. Eq.(\[bk3c\]) also suggests that the case where $\sigma \neq 1/2$ could be related to the non hermitean Hamiltonian $H_0 = (xp + p x)/2 - i (\sigma-1/2)$ whose right (resp. left) eigenfunctions are given by $1/x^{\sigma - i E}$ ( resp. $1/x^{1- \sigma - i E}$). On more general grounds, we would like to mention two important points. First is that one still needs to show that the function $n(t)$, defined in eq. (\[q34\]), is such that $e^{2 \pi i n(t)}$ is analytic in the upper-half plane and that it goes to zero as $|t| \rightarrow \infty$, so that the Jost function is indeed given by eq.(\[q39\]), as we have assumed so far. Second, and related to the latter point, is that that the function $n_\fl(t)$ is well defined provided $f(t)$ does not vanish for $t$ real, in which case (\[bk3b\]) reads also (1/2 - i t) = f(-t) ( 1 + e\^[2 i (t)]{} ) = f(-t) (t) \[bk6\] which shows that our construction of a QM model of the Riemann zeros relies on the absence of zeros of the function $f(t)$ on the critical line. These zeros were investigated by Bombieri long ago in an attempt to improve the existing lower bounds for the number of Riemann zeros on the critical line [@Bombieri2]. In this regard our results give further support, but not a proof, to the RH. As suggested in [@Sierra2; @Sierra3] that proof would follow if the zeta function $\zeta(1/2 - i t)$ can be realized as the Jost function of a QM model of the sort discussed so far, due to its special analyticity properties. Eq.(\[bk6\]) gives a partial realization of this idea but the function $f(t)$ lacks of a physical interpretation so far. The latter approach is analogue to the ones proposed in the past by several authors where the zeta function gives the scattering phase shift of some quantum mechanical model, particularly on the line $\Re \; s = 1$ [@Faddeev; @Lax; @G2; @Joffily; @BKL]. Another important question is: where are the prime numbers in our construction? As suggested by the Quantum Chaos scenario, the prime numbers may well be classical objects hidden in the quantum model, so the next question is: what is the classical limit of the Hamiltonian?. The free part is of course given by $xp$, but the interacting part is an antisymmetric matrix with no obvious classical version. The existence of such a classical Hamiltonian may help to answer the [*prime*]{} question but it may also lead to a real physical realization of the model. Work along this direction is under progress [@SP]. Acknowledgments {#acknowledgments .unnumbered} =============== I wish to thank for discussions M. Asorey, M. Berry, L.J. Boya, J. García-Esteve, J. Keating, J.I. Latorre, A. LeClair, J. Links, M.A. Martín-Delgado, G. Mussardo, J. Rodríguez-Laguna and P.K. Townsend. This work was supported by the CICYT of Spain under the contracts FIS2004-04885. I also acknowledge ESF Science Programme INSTANS 2005-2010. Appendix A: Wave functions and norms ==================================== In this appendix we shall derive alternative expressions of the eigenfunctions of the model and compute their norm. Let us start from eq.(\[u7\]) for the eigenfunctions of the Hamiltonian (\[t5\]), \_E(q) = e\^[- (1/2 - i E) q ]{} \[A1\] Replacing $a(q)$ and $b(q)$ by their Fourier transform, and using eq.(\[a3\]) one finds \_[-]{}\^q dq’ e\^[- i E q’]{} a(q’) = + e\^[- i q E]{} \_[-]{}\^ \[A2\] and a similar expression for the integral of $b(q)$. All the singular integrals appearing in this appendix must be understood in the Cauchy sense. Plugging the latter expressions into (\[A1\]) yields \_E(q) = e\^[- (1/2 - i E) q ]{} \[A3\] Using eqs.(\[u9\]), (\[u28\]) and (\[u29\]), the first term in the RHS becomes C\_0 + (B (-E) - A (-E) ) = = (E) \[A4\] so that $\psi(x)$ is given by \_E(x) = + \_[-]{}\^ x\^[-1/2 + i ]{} \[A5\] where $A(E)$ and $B(E)$ are given by the eqs.(\[u28\]) and (\[u29\]). The function (\[A5\]) can also be expanded in the basis (\[t3\]) of eigenfunctions of $H_0$, i.e. |\_E = \_[- ]{}\^ d \_E() |\_ \[A6\] namely \_E (x) = \_[- ]{}\^ d \_E() \[A7\] The result is \_E() = (E-) (E) + \[A8\] which shows that the delocalized states, i.e. $\CF(E) \neq 0$, have to be normalized in the distributional sense, while the localized states, i.e. $\CF(E_m) = 0$, have a norm given by \_[E\_m]{} | \_[E\_m]{} = \_[- ]{}\^ \[A9\] In the examples discussed throughout the paper the functions $\ha(t), \hb(t)$ are phase factors, up to overall constants. Moreover, if the function $\ha(t) \hb(-t)$ is analytic in the upper half-plane and vanishes when $|t| \rightarrow \infty, \ Re \; t > 0$, then the $S$-functions and the associated Jost function take a particular simple form if we allow for the existence of bound states, S\_[a,a]{} = S\_[b,b]{} = 1, S\_[a,b]{} = (t) (-t), S\_[b,a]{} = 0 (t) = 2 + (t) (-t) \[A10\] The integration constants $A,B$, corresponding to a bound state, can be choosen as A(E\_m) = - B(E\_m) = -1 \[A11\] which differ with respect to (\[u29\]) in an unimportant overall sign. The wave function (\[A5\]) also simplifies \_[E\_m]{} (x) = \_[-]{}\^ x\^[-1/2 + i ]{} \[A12\] and scalar product of two bound state wave functions becomes \_[E\_[m\_1]{} ]{}| \_[E\_[m\_2]{}]{} = \_[- ]{}\^ \[A13\] The analiticity of the Jost function $\F(E)$ in the upper-half plane implies the dispersion relation (E) = \_+ \_[-]{}\^ \[A14\] where $\F_\infty$ is the value of $\F(E)$ at $E = + i \infty$. From this equation, and the fact that $\F(E_{m_1} ) = \F(E_{m_2} )=0$, one can show that $\psi_{E_{m_1} }$ and $\psi_{E_{m_1} }$ are orthogonal. Furthermore, eq.(\[A14\]) yields also a simple expression for the norm of $\psi_{E_{m} }$ \_[E\_[m]{} ]{}| \_[E\_[m]{}]{} = \_[- ]{}\^ = - ’(E\_m) \[A15\] Finally, writing $\F(E)$ as in eq.(\[q38\]), i.e. (E) = 2 ( 1 + e\^[2 i n(E)]{} ) \[A16\] where $n(E)$ is the number of states, up to a constant, one derives that the norm of $\psi_{E_{m} }$ is proportional to the density of states at $E_m$, \_[E\_[m]{} ]{}| \_[E\_[m]{}]{} = 4 n’(E\_m) \[A17\] Wave functions associated to the smooth and exact Riemann zeros --------------------------------------------------------------- The Mellin transforms of the boundary wave functions associated to the smooth Riemann zeros were given in eq.(\[r10\]). Choosing $l_x = 1, l_p = 2 \pi, a_0=b_0= \sqrt{2}$ we have (t) = e\^[2 i (t)]{}, (t) = 1 \[A18\] The wave functions (\[A12\]) become in this case, \_[E\_m]{} (x) = \_[-]{}\^ x\^[-1/2 + i ]{} \[A19\] The integrals can be performed using the residue theorem obtaining \_[E\_m]{} (x) = + \_1F\_2( - ; , - , - \^2 x\^2) \[A20\] where $H(x-1) = 1$ if $x>1$ and 0 if $0<x<1$. One can show that $\sqrt{x} \psi_{E_m} \rightarrow 0$ as $x \rightarrow \infty$, if $1 + e^{2 i \theta(E_m)} =0$. In fig.\[psi\_bk\] we plot the absolute values of (\[A20\]) for those energies that correspond to the three lowest Riemann zeros. Notice that the functions are very small in the classical forbidden region $0 < x < 1$. The amplitude has a high frequency component common to the three waves plus a low frequency one that depends on the level. The wave functions associated to the exact Riemann zeros can be computed from eq.(\[A12\]) with $\ha(t)$ and $\hb(t)$ given by eq. (\[q41\]). We do not have an analytic expression for this integral, however a numerical estimate can be obtained truncating (\[A12\]) as \_[E\_m]{} (x) \~ \_[E\_m - ]{}\^[E\_m + ]{} x\^[-1/2 + i ]{} \[A21\] In fig.\[psi\_r\] we plot the result for the lowests Riemann zeros. The wave functions have some common features with those of fig. \[psi\_bk\], but they also exhibit a random behaviour. [999]{} H.M. Edwards, “Riemann’s Zeta Function”, Academic Press, New York, 1974. E.C. Titchmarsh, “The Theory of the Riemann Zeta-Function”, 2nd ed., Oxford University Press 1999, Oxford. E. Bombieri, “Problems of the Millenium: the Riemann hypothesis”, Clay Mathematics Institute (2000). http://www.claymath.org/millennium/Riemann-Hypothesis/ P. Sarnak, “Problems of the Millenium: the Riemann hypothesis (2004)”, Clay Mathematics Institute (2004). J.B. Conrey, “The Riemann Hypothesis.” Not. Amer. Math. Soc. 50, 341-353, 2003. See M. Watkins at http://secamlocal.ex.ac.uk/$\sim$mwatkins /zeta/physics.htm for a comprehensive review on several approaches to the RH. H.C. Rosu, “Quantum hamiltonians and prime numbers”, Mod. Phys. Lett. [**A18**]{} (2003) 1205; quant-ph/0304139. E. Elizalde, V. Moretti, S. Zerbini, “On recent strategies proposed for proving the Riemann hypothesis”, Int.J.Mod.Phys. [**A18**]{} (2003) 2189-2196; math-ph/0109006. A. Selberg, “Harmonic analysis and discontinuous groups in weakly symmetric Riemannian spaces with applications to Dirichlet series”, Journal of the Indian Mathematical Society 20 (1956) 47-87. H. Montgomery, “The pair correlation of zeros of the zeta function”, Analytic Number Theory, AMS (1973). M.L. Mehta, “Random matrices”, Elsevier Academic Press, 2004, Amsterdam. A. Odlyzko, “On the distribution of spacings between zeros of zeta functions”, Math. Comp. [**48**]{}, 273 (1987). M.V. Berry, in [*Quantum Chaos and Statistical Nuclear Physics*]{}. Eds. T.H. Seligman and H. Nishioka, Lecture Notes in Physics, No. 263, Springer Verlag, New York, 1986. M.V. Berry, “Quantum Chaology”, Proc. R. Soc. Lond. A 413, 183 (1987). M. C. Gutzwiller “Periodic orbits and classical quantization conditions”, J. Math. Phys. 12 no. 3 (1971). A. Connes, “Trace formula in noncommutative geometry and the zeros of the Riemann zeta function”, Selecta Mathematica (New Series) 5 (1999) 29; math.NT/9811068. M.V. Berry and J.P. Keating, “H=xp and the Riemann zeros”, in [*Supersymmetry and Trace Formulae: Chaos and Disorder*]{}, ed. J.P. Keating, D.E. Khmelnitskii and I. V. Lerner, Kluwer 1999. M. V. Berry and J. P. Keating, “The Riemann zeros and eigenvalue asymptotics”, SIAM REVIEW [**41**]{} (2) 236, 1999. G. Sierra, “The Riemann zeros and the Cyclic Renormalization Group”, J.Stat.Mech. 0512 (2005) P006; math.NT/0510572. G. Sierra,”H=xp with interaction and the Riemann zeros”, Nucl. Phys. [**B 776**]{}, (2007) 327; math-ph/0702034. G. Sierra,”Quantum reconstruction of the Riemann zeta function”, J. Phys. A: Math. Theor. [**40**]{} (2007) 1; math-ph/0711.1063. A. LeClair, J.M. Román and G. Sierra, “Russian doll Renormalization Group and Superconductivity”, Phys. Rev. [**B69**]{} (2004) 20505; cond-mat/0211338. A. Anfossi, A. LeClair, G. Sierra, “The elementary excitations of the exactly solvable Russian doll BCS model of superconductivity”, J. Stat. Mech. (2005) P05011; cond-mat/0503014. C. Dunning and J. Links,“Integrability of the Russian doll BCS model”,Nucl. Phys. [**B702**]{} (2004) 481, cond-mat/0406234. A. LeClair, “Interacting Bose and Fermi gases in low dimensions and the Riemann hypothesis”; math-ph/0611043. J. Twamley and G. J. Milburn, “The quantum Mellin transform”, New J. Phys. 8 (2006) 328; quant-ph/0702107. The $S_{f, g}(z)$ differs in a sign respect to the one considered in references [@Sierra2; @Sierra3]. S. Akiyama and Y. Tanigawa, “Multiple zeta values at non-positive integers”, [*Ramanujan J.*]{} [**5**]{} (2001), 327-351. L. Guo and B. Zhang, “Renormalization of Multiple zeta values”, math.NT/0606076. G. Bonneau, J. Faraut, G. Valent, “Self-adjoint extensions of operators and the teaching of quantum mechanics”, Am.J.Phys. 69 (2001) 322 quant-ph/0103153. M.V. Berry and J.P. Keating, “A new asymptotic representation for $\zeta(1/2 + i t)$ and quantum spectral determinants”, Proc. R. Soc. Lond. A (1992) [**437**]{} 151. E. Bombieri, “A lower bound for the zeros of Riemann«s zeta function on the critical line”, Séminaire N. Bourbaki, 1974-75, exp. no. 465, p. 176-182. B.S. Pavlov and L.D. Faddeev, “Scattering theory and automorphic functions”, Sov. Math. 3, 522 (1975), Plenum Publishing Corp. translation, N.Y; Lax and R.S. Phillips, [*Scattering Theory for Automorphic Functions*]{}, Princeton University Press, Princeton, 1976. M.C. Gutzwiller, “Stochastic behaviour in Quantum Scattering”, Physica [**D7**]{}, 341 (1983). S. Joffily, “Jost function, prime numbers and Riemann zeta function”, math-ph/0303014 R.K. Bhaduri, Avinash Khare, and J. Law, “Phase of the Riemann zeta function and the inverted harmonic oscillator”, Physical Review E 52 no. 1 (1995) 486-491; chao-dyn/9406006. G. Sierra and P.K. Townsend, work in preparation.

ArXiv

--- abstract: '[**[Abstract.]{}**]{} We investigate the intersection problem for finite semigroups, which asks for a given set of regular languages, represented by recognizing morphisms to finite semigroups, whether there exists a word contained in their intersection. We introduce compressibility measures as a useful tool to classify the intersection problem for certain classes of finite semigroups into circuit complexity classes and Turing machine complexity classes. Using this framework, we obtain a new and simple proof that for groups and commutative semigroups, the problem is contained in ${\ensuremath{\mathsf{NP}}}$. We uncover certain structural and non-structural properties determining the complexity of the intersection problem for varieties of semigroups containing only trivial submonoids. More specifically, we prove ${\ensuremath{\mathsf{NP}}}$-hardness for classes of semigroups having a property called *unbounded order* and for the class of all nilpotent semigroups of *bounded order*. On the contrary, we show that bounded order and commutativity imply containment in the circuit complexity class ${\ensuremath{\mathsf{qAC}}}^k$ (for some $k \in {\mathbb{N}}$) and decidability in quasi-polynomial time. We also establish connections to the monoid variant of the problem.' author: - Lukas Fleischer date: | FMI, University of Stuttgart[^1]\ Universitätsstraße 38, 70569 Stuttgart, Germany\ `fleischer@fmi.uni-stuttgart.de` title: The Intersection Problem for Finite Semigroups --- Introduction ============ A careful analysis of the complexity of decision problems for regular languages has triggered renewed interest in the classical intersection non-emptiness problem (called *intersection problem* in the following), as first described by Kozen in 1977 [@koz77:short], and in the closely related *membership problem for transformation monoids* [@BabaiLS87:short; @Beaudry88thesis; @BeaudryMT92; @FurstHopcroftLuks80:short; @Sims1967]. The connection between these two problems stems from the observation that a set of deterministic finite automata over a common alphabet can be considered as transformations on the (disjoint) union of their states. Both problems are well-known to be ${\ensuremath{\mathsf{PSPACE}}}$-complete in the general case but become easier when the inputs are restricted to have certain structural properties. These properties are often expressed in terms of membership to a certain *variety of finite monoids*; in the automaton setting, one considers the transition monoids of the automata. For example, for the variety of ${\mathcal{R}}$-trivial monoids, usually denoted by ${\mathbf{R}}$, both problems were shown to be decidable in non-deterministic polynomial time [@BeaudryMT92]. On the other hand, it is known that ${\ensuremath{\mathsf{PSPACE}}}$-completeness already holds for any variety not contained within ${\mathbf{DS}}$, the variety of all finite monoids whose regular ${\mathcal{D}}$-classes form subsemigroups. However, for many subvarieties of ${\mathbf{DS}}$, such as ${\mathbf{L}}$ (the left-right dual of ${\mathbf{R}}$) or ${\mathbf{DA}}$ (all aperiodic monoids from ${\mathbf{DS}}$), the problems are only known to be ${\ensuremath{\mathsf{NP}}}$-hard and to be contained within ${\ensuremath{\mathsf{PSPACE}}}$. The problem of determining the exact complexity for varieties in this interval has been open for more than 25 years [@BeaudryMT92; @tt02:short]. Recently, Kufleitner and the author suggested considering the algebraic variant of the problem, where the languages in the input are represented by finite monoids instead of automata [@FleischerK18:short]. Formally, it is defined as follows: [p[1.5cm]{}X]{} ------------------------------------------------------------------------ height 1pt @a xhline ${\ensuremath{\textsc{MonIsect}}}({\mathbf{C}})$\ : & Morphisms $h_i \colon A^* \to M_i \in {\mathbf{C}}$ and sets $P_i \subseteq M_i$ with $1 {{\leqslant}}i {{\leqslant}}k$\ : & Is $h_1^{-1}(P_1) {\mathbin{\cap}}\cdots {\mathbin{\cap}}h_k^{-1}(P_k) \ne \emptyset$?\ Here, ${\mathbf{C}}$ is some fixed class of finite monoids and the monoids themselves are given as multiplication tables. Transitioning to the algebraic setting allowed for making some substantial progress in understanding the complexity of the problem: Kufleitner and the author proved ${\ensuremath{\mathsf{NP}}}$-completeness of ${\ensuremath{\textsc{MonIsect}}}({\mathbf{DO}})$ where ${\mathbf{DO}}$ is a quite large subvariety of ${\mathbf{DS}}$ including both ${\mathbf{L}}$ and ${\mathbf{DA}}$. Still, even for the monoid variant, ${\ensuremath{\mathsf{PSPACE}}}$-completeness is only known to hold for varieties not contained within ${\mathbf{DS}}$, a proper superset of ${\mathbf{DO}}$. Attempts to progress further in understanding the complexity of ${\ensuremath{\textsc{MonIsect}}}$ led to the investigation of classes of semigroups ${\mathbf{C}}$ instead of monoids: [p[1.5cm]{}X]{} ------------------------------------------------------------------------ height 1pt @a xhline ${\ensuremath{\textsc{SgpIsect}}}({\mathbf{C}})$\ : & Morphisms $h_i \colon A^+ \to S_i \in {\mathbf{C}}$ and sets $P_i \subseteq S_i$ with $1 {{\leqslant}}i {{\leqslant}}k$\ : & Is $h_1^{-1}(P_1) {\mathbin{\cap}}\cdots {\mathbin{\cap}}h_k^{-1}(P_k) \ne \emptyset$?\ As in the monoid variant, the semigroups are assumed to be given as multiplication tables. While making this distinction between monoids and semigroups may sound subtle at first sight, it has a significant impact on complexity questions and is expected to yield new insights. For example, all known ${\ensuremath{\mathsf{PSPACE}}}$-hardness results rely heavily on the existence of neutral letters. We mainly investigate the intersection problem for *varieties of finite semigroups*. In [@FleischerK18:short], ${\ensuremath{\textsc{MonIsect}}}({\mathbf{V}})$ was shown to be ${\ensuremath{\mathsf{NP}}}$-hard for every non-trivial variety of finite monoids ${\mathbf{V}}$. Thus, in this work, we focus on the intersection problem for varieties of finite semigroups containing only trivial submonoids. We describe an infinite sequence of varieties ${\mathbf{V}}_1 \subseteq {\mathbf{V}}_2 \subseteq \cdots$ such that ${\ensuremath{\textsc{SgpIsect}}}({\mathbf{V}}_i) \in {\ensuremath{\mathsf{AC}}}^0$ for each $i {{\geqslant}}1$ but the intersection problem for its limit ${\ensuremath{\textsc{SgpIsect}}}({\mathbf{V}}_\infty)$ (where ${\mathbf{V}}_\infty = \bigcup_{i \in {\mathbb{N}}} {\mathbf{V}}_i$) is ${\ensuremath{\mathsf{NP}}}$-complete. This is surprising for the following reason: for the automaton and monoid variants, all known hardness results are tied to purely structural properties. ${\ensuremath{\mathsf{NP}}}$-hardness of ${\ensuremath{\textsc{MonIsect}}}$ comes from the fact that the problem is ${\ensuremath{\mathsf{NP}}}$-hard even for the monoid $U_1$ and for the cyclic group ${\mathbb{Z}}/ 2{\mathbb{Z}}$, and ${\ensuremath{\mathsf{PSPACE}}}$-hardness comes from the fact that even ${\ensuremath{\textsc{MonIsect}}}(B_2^1)$ is ${\ensuremath{\mathsf{PSPACE}}}$-hard [@FleischerK18:short]. Since every semigroup from ${\mathbf{V}}_\infty$ is contained in infinitely many varieties ${\mathbf{V}}_k$ in the sequence above, the existence of such a pattern cannot be the sole reason for ${\ensuremath{\mathsf{NP}}}$-hardness in the semigroup setting. It is open whether a similar situation occurs below ${\mathbf{DS}}$ in the monoid or in the automaton setting. To investigate other parameters with an impact on the complexity of the problem, we introduce a versatile framework based on the notion of *product circuits properties*. These properties are a measure of compressibility of witnesses for intersection non-emptiness. Using this framework, we obtain a new and easy proof that both ${\ensuremath{\textsc{SgpIsect}}}({\mathbf{G}})$ and ${\ensuremath{\textsc{SgpIsect}}}({\mathbf{Com}})$ are contained in ${\ensuremath{\mathsf{NP}}}$. We prove ${\ensuremath{\mathsf{NP}}}$-completeness of ${\ensuremath{\textsc{SgpIsect}}}$ for classes having a property we call *unbounded order* (this includes the class of all nilpotent and commutative semigroups) and for the class of all nilpotent semigroups of *bounded order*. On the contrary, we show that for every commutative variety with bounded order, the intersection problem is contained in some uniform version of a circuit complexity class ${\ensuremath{\mathsf{qAC}}}^k$ and thus decidable in quasi-polynomial time. As problems decidable in quasi-polynomial time cannot be ${\ensuremath{\mathsf{NP}}}$-hard unless the exponential time hypothesis fails, this suggests that an interplay of structural properties and non-structural properties determines the complexity of the problem. We also suggest a way to transfer complexity results from the monoid setting to the semigroup setting. Preliminaries {#sec:prelim} ============= #### Algebra. A *semigroup* is a non-empty set equipped with an associative binary operation, often also referred to as *multiplication*. A semigroup $M$ with a *neutral element*, [i.e., ]{}an element $e \in M$ such that $ex = x = xe$ for all $x \in M$, is called *monoid*. The neutral element is unique and usually denoted by $1$. An element $x \in S$ is *idempotent* if $x^2 = x$ and the set of all idempotent elements of a semigroup $S$ is denoted by $E(S)$. A *zero* element $z$ of a finite semigroup $S$ satisfies $zx = z = xz$ for all $x \in S$. Each semigroup contains at most one zero element and a semigroup is *nilpotent* if its only idempotent element is a zero element. The set of all finite words $A^*$ (resp. all non-empty finite words $A^+$) forms a monoid (resp. semigroup) with concatenation as multiplication. A *subsemigroup* (resp. *submonoid*) of a semigroup (resp. monoid) is a subset closed under multiplication. Let $S$ and $T$ be semigroups and let $M$ and $N$ be monoids. The *direct product* of $S$ and $T$ is the Cartesian product $S \times T$ with componentwise multiplication. A *semigroup morphism* from $S$ to $T$ is a mapping $h \colon S \to T$ such that $h(s)h(t) = h(st)$ for all $s, t \in S$. A *monoid morphism* from $M$ to $N$ is a semigroup morphism $h \colon M \to N$ which additionally satisfies $h(1) = 1$. The semigroup $T$ is a *divisor* of $S$ if there exists a surjective semigroup morphism from a subsemigroup of $S$ onto $T$. The monoid $N$ is a *divisor* of $M$ if there exists a surjective monoid morphism from a submonoid of $M$ onto $N$. We often use the term *morphism* to refer to both semigroup and monoid morphisms if the reference is clear from the context. A morphism $h \colon A^+ \to S$ to a finite semigroup $S$ *recognizes* a language $L \subseteq A^+$ if $h^{-1}(P) = L$ for some set $P \subseteq S$. The set $P$ is often called the *accepting set* for $L$. #### Varieties. A *variety of finite semigroups* is a class of finite semigroups which is closed under taking (semigroup) divisors and direct products. A *variety of finite monoids* is a class of finite monoids closed under taking (monoid) divisors and direct products. The class ${\mathbf{G}}$ of all finite groups and the class ${\mathbf{I}}$ containing only the trivial semigroup ${\ensuremath{\left\{1\right\}}}$ are both varieties of finite semigroups and varieties of finite monoids. We also consider the following varieties of finite semigroups: - ${\mathbf{Com}}$, the variety of all finite commutative semigroups, - ${\mathbf{N}}$, the variety of all finite nilpotent semigroups, - ${\mathbf{A}}_2 \cap {\mathbf{N}}$, the variety of all finite semigroups where $x^2 y = x^2 = y x^2$ for all $x, y \in S$, - ${\mathbf{LI}}_k$ (for $k \in {\mathbb{N}}$), the variety of all finite semigroups $S$ which satisfy the equation $x_1 \cdots x_k z y_k \cdots y_1 = x_1 \cdots x_k y_k \cdots y_1$ for all $x_1, \dots, x_k, y_1, \dots, y_k, z \in S$. Note that each of the varieties in this list contains semigroups which are not monoids. Hence, they do not form varieties of finite monoids. We will also briefly refer to the varieties ${\mathbf{DS}}$ and ${\mathbf{DO}}$ but their formal definitions are not needed. For a variety of finite semigroups ${\mathbf{V}}$, we denote by ${\mathbf{V}}_{\mathbf{M}}$ the class of all finite monoids which, when viewed as semigroups, belong to ${\mathbf{V}}$. It is easy to check that ${\mathbf{V}}_{\mathbf{M}}$ forms a variety of finite monoids. For each semigroup $S$ and each idempotent element $e \in E(S)$, the set $eSe$ forms a monoid with the multiplication induced by $S$ and with neutral element $e$, called the *local monoid* at $e$. For a variety of finite monoids ${\mathbf{V}}$, we denote by ${\mathbf{LV}}$ the variety of finite semigroups whose local monoids belong to ${\mathbf{V}}$. The operators ${(\cdot)}_{\mathbf{M}}$ and ${\mathbf{L}}(\cdot)$ are closely related. Let ${\mathbf{V}}$ be a variety of finite monoids and let ${\mathbf{W}}$ be a variety of finite semigroups. Then ${\mathbf{W}}_{\mathbf{M}}\subseteq {\mathbf{V}}$ if and only if ${\mathbf{W}}\subseteq {\mathbf{LV}}$. In particular, ${\mathbf{W}}\subseteq {\mathbf{LW}}_{\mathbf{M}}$.\[prop:vm-lm\] Suppose that ${\mathbf{W}}_{\mathbf{M}}\subseteq {\mathbf{V}}$ and let $S$ be an arbitrary semigroup from ${\mathbf{W}}$. For every idempotent element $e \in E(S)$, the monoid $eSe$ is a subsemigroup of $S$. By closure of ${\mathbf{W}}$ under taking subsemigroups, we obtain $eSe \in {\mathbf{W}}$. Since $eSe$ is a monoid, we obtain $eSe \in {\mathbf{W}}_{\mathbf{M}}$ and by assumption, we have $eSe \in {\mathbf{V}}$, as desired. Conversely, suppose that ${\mathbf{W}}\subseteq {\mathbf{LV}}$ and let $M$ be a monoid from ${\mathbf{W}}$. Let $e$ be the identity element of $M$. Since $M \in {\mathbf{LV}}$, we obtain $M = eMe \in {\mathbf{V}}$. As a direct consequence, ${\mathbf{LI}}$ is the largest variety of finite semigroups not containing any non-trivial monoids. The following proposition connects ${\mathbf{LI}}$ with the hierarchy of varieties $({\mathbf{LI}}_k)_{k \in {\mathbb{N}}}$ defined above. Let $S$ be a finite semigroup of cardinality $n$. Then $S \in {\mathbf{LI}}$ if and only if $S \in {\mathbf{LI}}_{n + 1}$. \[prop:li-nerb\] Suppose that $S \in {\mathbf{LI}}$ and let $x_1, \dots, x_{n+1}, y_1, \dots, y_{n+1}, z \in S$. By the pigeon hole principle, there exist indices $i, i' \in {\ensuremath{\left\{1, \dots, n+1\right\}}}$ such that $i < i'$ and $x_1 \cdots x_i = x_1 \cdots x_{i'}$. Thus, $x_1 \cdots x_i e = x_1 \cdots x_i$ for $e = (x_{i+1}\cdots x_{i'})^\omega$ and for every $\omega \in {\mathbb{N}}$. In particular, we may choose $\omega$ such that $e$ is idempotent. Analogously, there exists some index $j \in {\ensuremath{\left\{1, \dots, n+1\right\}}}$ and some idempotent element $f$ such that $f y_j \cdots y_1 = y_j \cdots y_1$. Since $S \in {\mathbf{LI}}$, we have $exf = ex(fef) = (exfe)f = ef = (eyfe)f = ey(fef) = eyf$ for all $x, y \in S$. This yields $$\begin{aligned} x_1 \cdots x_{n+1} z y_{n+1} \cdots y_1 & = x_1 \cdots x_i e x_{i+1} \cdots x_{n+1} z y_{n+1} \cdots y_{j+1} f y_j \cdots y_1 \\ & = x_1 \cdots x_i e x_{i+1} \cdots x_{n+1} y_{n+1} \cdots y_{j+1} f y_j \cdots y_1 \\ & = x_1 \cdots x_{n+1} y_{n+1} \cdots y_1, \end{aligned}$$ which shows that $S \in {\mathbf{LI}}_{n+1}$. Conversely, let $S$ be contained in ${\mathbf{LI}}_{n+1}$. For all $e \in E(S)$ and for all $x \in S$, we have $exe = e^{n+1} x e^{n+1} = e^{n+1} e^{n+1} = e$ where only the second equality uses $S \in {\mathbf{LI}}_{n+1}$. Thus, every local monoid $eSe$ is trivial, and $S \in {\mathbf{LI}}$. #### Complexity. We assume familiarity with standard definitions from circuit complexity. A function has *quasi-polynomial* growth if it is contained in $2^{{\mathcal{O}}(\log^c n)} = n^{{\mathcal{O}}(\log^{c-1} n)}$ for some fixed $c \in {\mathbb{N}}$. Throughout the paper, we denote by ${\ensuremath{\mathsf{AC}}}^k$ (resp. ${\ensuremath{\mathsf{qAC}}}^k$) the class of languages decidable by circuit families of depth ${\mathcal{O}}(\log^k n)$ and polynomial size (resp. quasi-polynomial size); see [@Barrington92; @straubing94; @Vollmer99] for details. We allow NOT gates but do not count them when measuring the depth or the size of a circuit. We will also refer to the standard complexity classes ${\ensuremath{\mathsf{NP}}}$ and ${\ensuremath{\mathsf{PSPACE}}}$. The *exponential time hypothesis* states that a deterministic Turing machine cannot decide ${\ensuremath{\textsc{3-Sat}}}$ in subexponential time. If the hypothesis is true, ${\ensuremath{\mathsf{NP}}}$-complete problems cannot be decided in quasi-polynomial time; see [e.g. ]{}[@ImpagliazzoP99:short]. #### Straight-Line Programs. A *straight-line program* (*SLP*) is a tuple $G = (V, A, P, X_s)$ where $V$ is a finite set of *variables*, $A$ is a finite set of *letters*, $P \colon V \to (V {\mathbin{\cup}}A)^*$ is a mapping and $X_s \in V$ is the so-called *start variable* such that the relation $${\ensuremath{\left\{(X, Y) \mid P(X) \in (V {\mathbin{\cup}}A)^* Y (V {\mathbin{\cup}}A)^*\right\}}}$$ is acyclic. For a variable $X \in V$, the word $P(X)$ is the *right-hand side* of $X$. Starting with some word $\alpha \in (V {\mathbin{\cup}}A)^*$ and repeatedly replacing variables $X \in V$ by $P(X)$ yields a word from $A^*$, the so called *evaluation of $\alpha$*, denoted by ${\mathsf{val}}(\alpha)$. The word *produced by $G$* is ${\mathsf{val}}(G) = {\mathsf{val}}(X_s)$. If the reference to $A$ and $V$ is clear, we will often use the notation $h(\alpha)$ instead of $h({\mathsf{val}}(\alpha))$ for the image of the evaluation of a word $\alpha \in (A {\mathbin{\cup}}V)^*$ under a morphism $h \colon A^+ \to S$. Analogously, we write $h(G)$ instead of $h({\mathsf{val}}(G))$. The *size* of $G$ is ${\ensuremath\left|G\right|} = \sum_{X \in V} {\ensuremath\left|P(X)\right|}$. Each variable $X$ of an SLP $G$ can be viewed as an SLP itself by making $X$ the start variable of $G$. The *canonical SLP of a word $w \in A^+$* is $G = (V, A, P, X_s)$ with $V = {\ensuremath{\left\{X_s\right\}}}$ and $P(X_s) = w$. The following simple lemma illustrates how SLPs can be used for compression; see [e.g. ]{}[@CharikarLLPPSS05; @FleischerK18:short] for a proof. Let $G = (V, A, P, X_s)$ be an SLP and let $e \in {\mathbb{N}}$. Then there exists an SLP $H$ of size ${\ensuremath\left|H\right|} {{\leqslant}}{\ensuremath\left|G\right|} + 4 \log(e)$ such that ${\mathsf{val}}(H) = ({\mathsf{val}}(G))^e$. \[lem:slp-intro\] Product Circuits Properties {#sec:pcp} =========================== Let ${\mathbf{C}}$ be a class of finite semigroups and let $f \colon {\mathbb{N}}\to {\mathbb{N}}$ be a monotonically increasing function. We say that ${\mathbf{C}}$ has the *$f(n)$ circuits property* if for each morphism $h_i \colon A^+ \to S$ to a finite semigroup $S \in {\mathbf{C}}$ and for each $w \in A^+$, there exists an SLP $G$ of size at most $f({\ensuremath\left|S\right|})$ such that $h(G) = h(w)$. We say that ${\mathbf{C}}$ has the *$f(n)$ product circuits property* if for each set of morphisms $h_i \colon A^+ \to S_i$ to finite semigroups $S_1, \dots, S_k \in {\mathbf{C}}$ and for each $w \in A^+$, there exists an SLP $G$ of size at most $f({\ensuremath\left|S_1\right|} + \dots + {\ensuremath\left|S_k\right|})$ such that $h_i(G) = h_i(w)$ for all $i \in {\ensuremath{\left\{1, \dots, k\right\}}}$. For a class of functions $\mathcal{C}$, we say that ${\mathbf{C}}$ has the *$\mathcal{C}$ circuits property* (resp. *$\mathcal{C}$ product circuits property*) if ${\mathbf{C}}$ has the $f(n)$ circuits property (resp. $f(n)$ product circuits property) for some $f \in \mathcal{C}$. Let us introduce some abbreviations for commonly used classes of functions. We will use the terms - *constant circuits property* and *constant product circuits property* (*[$\mathrm{const}$[CP]{}]{}* and *[$\mathrm{const}$[PCP]{}]{}*, in short) for the class of constant functions, [i.e., ]{}the class of all functions of the form $f(n) = c$ for some $c \in {\mathbb{N}}$, - *poly-logarithmic circuits property* and *poly-logarithmic product circuits property* (*[$\mathrm{polylog}$[CP]{}]{}* and *[$\mathrm{polylog}$[PCP]{}]{}*, in short) for the class of poly-logarithmic functions, [i.e., ]{}the class of all functions $f(n) = \log^c n$ for some $c \in {\mathbb{N}}$, and - *polynomial circuits property* and *polynomial product circuits property* (*[$\mathrm{poly}$[CP]{}]{}* and *[$\mathrm{poly}$[PCP]{}]{}*, in short) for the class of polynomials, [i.e., ]{}the class of all functions of the form $f(n) = n^c$ for some $c \in {\mathbb{N}}$. The intuition behind these concepts is as follows. The $f(n)$ circuits property is a compressibility measure for witnesses of non-emptiness of a language given by a recognizing morphism. The $f(n)$ product circuits property is a compressibility measure for witnesses of non-emptiness of intersections of languages given by recognizing morphisms. The terminology is inspired by the *poly-logarithmic circuits property* which was introduced in [@FleischerCCC18:short]: having the $f(n)$ circuits property is equivalent to requiring every element of a subsemigroup $S$ of a semigroup from the class to be computable by an *algebraic circuit* of size $f(n)$ over any set of generators of $S$. Analogously, having the $f(n)$ product circuits property can be expressed in terms algebraic circuits with multiplication gates for the direct product of semigroups. It is clear that the $f(n)$ product circuits property implies the $f(n)$ circuits property. For the other direction, a weaker statement holds. Let ${\mathbf{C}}$ be a class of finite semigroups which is closed under taking direct products and has the $f(n)$ circuits property. Then ${\mathbf{C}}$ has the ${f(n^n)}$ product circuits property. \[prop:cp-pcp\] Suppose we are given morphisms $h_i \colon A^+ \to S_i$ to finite semigroups $S_1, \dots, S_k \in {\mathbf{C}}$ and a word $w \in A^+$. Let $N = {\ensuremath\left|S_1\right|} + \cdots + {\ensuremath\left|S_k\right|}$. Every semigroup contains at least one element, so $N^N {{\geqslant}}N^k$ is an upper bound for the product ${\ensuremath\left|S_1\right|} \cdots {\ensuremath\left|S_k\right|}$. Let $S$ be the direct product $S_1 \times \cdots \times S_k$ and let $h \colon A^+ \to S$ be the morphism defined by $h(a) = (h_1(a), \dots, h_k(a))$ for all $a \in A$. By closure of ${\mathbf{C}}$ under taking direct products, we have $S \in {\mathbf{C}}$. Since ${\mathbf{C}}$ has the $f(n)$ circuits property, there exists some SLP $G$ of size at most $f({\ensuremath\left|S\right|}) = f({\ensuremath\left|S_1\right|} \cdots {\ensuremath\left|S_k\right|}) {{\leqslant}}f(N^N)$ such that $h(G) = h(w)$. By construction, $h_i(G) = h_i(w)$ for all $i \in {\ensuremath{\left\{1, \dots, k\right\}}}$. An essential ingredient in the proof of ${\ensuremath{\textsc{MonIsect}}}({\mathbf{DO}}) \in {\ensuremath{\mathsf{NP}}}$ is that the variety of finite groups ${\mathbf{G}}$ has the [$\mathrm{poly}$[PCP]{}]{}. In [@FleischerK18:short], this was verified by analyzing a variant of the Schreier-Sims algorithm. Using the previous proposition, we obtain a much simpler proof: it is well known—and easy to show—that ${\mathbf{G}}$ has the [[$\mathrm{polylog}$[CP]{}]{}]{}, a result often called the *Babai-Szemer[é]{}di Reachability Lemma* [@BabaiS84]. The statement then follows from the following corollary of Proposition \[prop:cp-pcp\]. Let ${\mathbf{C}}$ be a class of finite semigroups which is closed under taking direct products and has the [[$\mathrm{polylog}$[CP]{}]{}]{}. Then ${\mathbf{C}}$ has the [[$\mathrm{poly}$[PCP]{}]{}]{}. \[crl:plcp-pcp\] The corollary also implies that the variety of all commutative semigroups, which was shown to have the [$\mathrm{polylog}$[CP]{}]{}in [@FleischerCCC18:short], has the [$\mathrm{poly}$[PCP]{}]{}. Circuits properties and product circuits properties have a big impact on the complexity of the so-called *Cayley semigroup membership problem* and the intersection problem for a given class. The remainder of this section is devoted to establishing this link for product circuits properties. Let $h \colon A^+ \to S$ be a morphism to a finite semigroup $S$ of size $N$ and let $G$ be an SLP of size $m$ over $A$. Then there exists an unbounded fan-in Boolean circuit of size $m(N^2 + {\ensuremath\left|A\right|} + 2) {\ensuremath\left\lceil\log N\right\rceil}$ and depth $2m + 2$ which computes $h(G)$. Given the SLP, this circuit can be computed by a deterministic Turing machine in time polynomial in the circuit size. \[lem:slp\] Single multiplications can be performed by circuits of size $(N^2 + 1) {\ensuremath\left\lceil\log N\right\rceil}$ with one layer of AND gates and one layer of OR gates: to perform a multiplication of two elements $x$ and $y$, we need to extract the ${\ensuremath\left\lceil\log N\right\rceil}$-bit entry of the multiplication table in row $x$ and column $y$. We create a layer of $N^2 {\ensuremath\left\lceil\log N\right\rceil}$ AND gates, followed by a layer of ${\ensuremath\left\lceil\log N\right\rceil}$ OR gates. Each AND gate is connected to one bit of the multiplication table in the input and to all bits of the values $x$ and $y$. Some of the incoming wires corresponding to the values $x$ and $y$ are negated such that the AND gate copies the bit of the multiplication table if it belongs to the corresponding entry $(x, y)$ and evaluates to $0$ otherwise. In the second layer, there are ${\ensuremath\left\lceil\log N\right\rceil}$ OR gates. The $k$-th of these OR gates is fed with the outputs of all AND gates corresponding to the $k$-th bit of some multiplication table entry. Thus, there are $N^2$ incoming wires to each OR gate. Since, for given input values $x$ and $y$, at most one of the incoming wires to each OR gate evaluates to $1$, the result of the product $x \cdot y$ then clearly appears as output value of the OR gates. A very similar layout is used to lookup the image of a letter $a \in A$ under the morphism $h \colon A^+ \to S$. First, ${\ensuremath\left|A\right|} {\ensuremath\left\lceil\log N\right\rceil}$ AND-gates are used to zero out the images of all letters except for the image of the letter $a$. Then, we use ${\ensuremath\left\lceil\log N\right\rceil}$ OR gates to perform a bitwise OR of all these preprocessed images. Since all images except $h(a)$ are zeroed out, the result is $h(a)$, as desired. We evaluate the image of each of the variables bottom-up: for all letters $a \in A$ occurring in $G$ we first compute the image $h(a)$. Then, if $P(X) = \gamma_1 \cdots \gamma_\ell$ for some $\gamma_1, \dots, \gamma_\ell \in V {\mathbin{\cup}}A$ and the images $h(\gamma_1), \dots, h(\gamma_\ell)$ have already been computed, we compute $h(X) = h(\gamma_1) \cdots h(\gamma_\ell)$ by performing $\ell-1$ multiplications. Clearly, each “lookup gadget”, each multiplication gadget and the wires connecting these components can be computed by a deterministic Turing machine in time polynomial in the size of the resulting circuit. We are now able to prove the main result of this section. Let ${\mathbf{C}}$ be a class of finite semigroups with the $f(n)$ product circuits property. Then ${\ensuremath{\textsc{SgpIsect}}}({\mathbf{C}})$ is decidable by a family of unbounded fan-in Boolean circuits of size ${\mathcal{O}}((f(n) + n)^{(f(n))^2} {\hspace*{0.5pt}}f(n) {\hspace*{0.5pt}}n^3 \log n)$ and depth ${\mathcal{O}}(f(n))$. For each input size $n \in {\mathbb{N}}$, the corresponding circuit can be computed by a deterministic Turing machine in time polynomial in the size of the resulting circuit. \[thm:circuits\] Suppose we are given morphisms $h_i \colon A^+ \to S_i$ to finite semigroups $S_i \in {\mathbf{C}}$ and sets $P_i \subseteq S_i$ where $1 {{\leqslant}}i {{\leqslant}}k$ for some $k \in {\mathbb{N}}$. We let $N = {\ensuremath\left|S_1\right|} + \dots + {\ensuremath\left|S_k\right|}$. Note that if $n$ denotes the input size of the ${\ensuremath{\textsc{SgpIsect}}}$ instance, we have $N {{\leqslant}}n$ and ${\ensuremath\left|A\right|} {{\leqslant}}n$. Since ${\mathbf{C}}$ has the $f(n)$ product circuits property, we know that if there exists a word $w \in A^+$ such that $h_i(w) \in P_i$ for all $i \in {\ensuremath{\left\{1, \dots, k\right\}}}$, then this word is generated by some SLP of size at most $f(N) {{\leqslant}}f(n)$. First, note that for a given fixed SLP of size $f(n)$, we can compute the image of the word generated by the SLP under each of the morphisms by an unbounded fan-in Boolean circuit of size ${\mathcal{O}}(n {\hspace*{0.5pt}}f(n) {\hspace*{0.5pt}}n^2 \log n)$ and depth ${\mathcal{O}}(f(n))$ by Lemma \[lem:slp\]. Since there are at most $((f(n)+n)^{f(n)})^{f(n)}$ different SLPs of size $f(n)$—at most $f(n)$ variables and at most $(f(n)+n)^{f(n)}$ possible right-hand sides per variable—we can do this evaluation for each of the SLPs in parallel, check whether any of them produces a witness for intersection non-emptiness and feed the outcomes of all the circuits into a single OR gate. It is clear that an enumeration of all SLPs of size at most $f(n)$ can be realized by a deterministic Turing machine in time polynomial in the output size. For classes with the [$\mathrm{const}$[PCP]{}]{}and classes with the [$\mathrm{polylog}$[PCP]{}]{}, efficient decidability of ${\ensuremath{\textsc{SgpIsect}}}$ is an immediate consequence. Let ${\mathbf{C}}$ be a class of finite semigroups with the [$\mathrm{const}$[PCP]{}]{}. Then the decision problem ${\ensuremath{\textsc{SgpIsect}}}({\mathbf{C}})$ is contained in ${\ensuremath{\mathsf{AC}}}^0$. Let ${\mathbf{C}}$ be a class of finite semigroups with the [$\mathrm{polylog}$[PCP]{}]{}. Then the decision problem ${\ensuremath{\textsc{SgpIsect}}}({\mathbf{C}})$ is contained in ${\ensuremath{\mathsf{qAC}}}^k$ for some $k \in {\mathbb{N}}$. Moreover, it is decidable in quasi-polynomial time and thus not ${\ensuremath{\mathsf{NP}}}$-hard, unless the exponential time hypothesis fails. \[crl:plpcp\] Containment in ${\ensuremath{\mathsf{qAC}}}^k$ is an immediate consequence of Theorem \[thm:circuits\]. For decidability in quasi-polynomial time, we can use a Turing machine that first computes and then evaluates the circuit. The circuit evaluation is done by computing the output value of a gate whenever all its inputs are available. For the [$\mathrm{poly}$[PCP]{}]{}, the statement of Theorem \[thm:circuits\] only yields exponential-size circuits. We restate a more useful complexity result on [$\mathrm{poly}$[PCP]{}]{}classes from [@FleischerK18:short]. Let ${\mathbf{C}}$ be a class of finite semigroups with the [$\mathrm{poly}$[PCP]{}]{}. Then ${\ensuremath{\textsc{SgpIsect}}}({\mathbf{C}})$ is contained in ${\ensuremath{\mathsf{NP}}}$. \[thm:ppcp\] We proceed as in the proof of Theorem \[thm:circuits\] but instead of generating a circuit evaluating all SLPs of polynomial size in parallel, we non-deterministically guess only one such SLP. We then evaluate the corresponding circuit in polynomial time as described in Lemma \[lem:slp\]. Together with the observations above, we obtain an easy proof of containment of both ${\ensuremath{\textsc{SgpIsect}}}({\mathbf{G}})$ and ${\ensuremath{\textsc{SgpIsect}}}({\mathbf{Com}})$ in ${\ensuremath{\mathsf{NP}}}$. Even though product circuits properties are a powerful tool, in some cases, it is sufficient to consider short witnesses without compression. This is particularly true for varieties not containing any subgroups which we shall mostly be concerned with in the following section. Moreover, for the [$\mathrm{const}$[PCP]{}]{}, compressibility and the existence of short (non-compressed) witnesses are actually equivalent. A class of finite semigroups ${\mathbf{C}}$ has the [$\mathrm{const}$[PCP]{}]{}if and only if there exists some constant $\ell \in {\mathbb{N}}$ such that every non-empty intersection of languages recognized by semigroups from ${\mathbf{C}}$ contains a word of length at most $\ell$. \[prop:cpcp\] The direction from right to left is trivial: every word $w$ of length at most $\ell$ can be represented by its canonical SLP, which then has size at most $\ell$ as well. For the converse direction, suppose that there exists some $s \in {\mathbb{N}}$ such that every non-empty intersection contains a word generated by an SLP of size at most $s$. It is easy to see that the length of such a word is at most $s^s$: there are at most $s$ variables and the right-hand side of every variable has length at most $s$; the claim now follows by induction. Thus, we obtain the desired statement by setting $\ell = s^s$. The Intersection Problem for Locally Finite Semigroups {#sec:complexity} ====================================================== Before presenting any algorithms and hardness results for ${\ensuremath{\textsc{SgpIsect}}}$, let us first describe how to transfer existing results to the semigroup setting. Let ${\mathbf{V}}$ be a variety of finite semigroups. If ${\mathbf{V}}\not\subseteq {\mathbf{LI}}$, then ${\ensuremath{\textsc{SgpIsect}}}({\mathbf{V}})$ is ${\ensuremath{\mathsf{NP}}}$-hard. If ${\mathbf{V}}\not\subseteq {\mathbf{LDS}}$, then ${\ensuremath{\textsc{SgpIsect}}}({\mathbf{V}})$ is ${\ensuremath{\mathsf{PSPACE}}}$-hard. \[prop:hardness\] If ${\mathbf{V}}\not\subseteq {\mathbf{LI}}$, then ${\mathbf{V}}_{\mathbf{M}}\not\subseteq {\mathbf{I}}$. Therefore, by [@FleischerK18:short Theorem 8], ${\ensuremath{\textsc{MonIsect}}}({\mathbf{V}}_{\mathbf{M}})$ is ${\ensuremath{\mathsf{NP}}}$-hard. The claim now follows from the fact that ${\ensuremath{\textsc{MonIsect}}}({\mathbf{V}}_{\mathbf{M}})$ is trivially ${\ensuremath{\mathsf{AC}}}^0$-reducible to ${\ensuremath{\textsc{SgpIsect}}}({\mathbf{V}})$. The same technique allows lifting ${\ensuremath{\mathsf{PSPACE}}}$-hardness of ${\ensuremath{\textsc{MonIsect}}}({\mathbf{V}}_{\mathbf{M}})$ in the case ${\mathbf{V}}_{\mathbf{M}}\not\subseteq {\mathbf{DS}}$ [@FleischerK18:short Theorem 11]. It seems plausible that the ${\mathbf{L}}(\cdot)$ operator can be used to lift complexity results from ${\ensuremath{\textsc{MonIsect}}}$ to ${\ensuremath{\textsc{SgpIsect}}}$ in a more general way. We thus conjecture: If ${\ensuremath{\textsc{MonIsect}}}({\mathbf{V}})$ is in ${\ensuremath{\mathsf{NP}}}$, then ${\ensuremath{\textsc{SgpIsect}}}({\mathbf{LV}})$ is in ${\ensuremath{\mathsf{NP}}}$. \[con:transfer\] By [@FleischerK18:short], a proof of this conjecture would immediately yield that ${\ensuremath{\textsc{SgpIsect}}}({\mathbf{LDO}})$ is contained in ${\ensuremath{\mathsf{NP}}}$. A possible approach is making use of the fact that for a local variety of finite monoids ${\mathbf{V}}$, we have ${\mathbf{LV}}= {\mathbf{V}}* {\mathbf{D}}$; see [e.g. ]{}[@str85jpaa] for details. However, one also needs to account for the size of semigroups from ${\mathbf{V}}* {\mathbf{D}}$. Surprisingly, even lifting the group case is much harder than one would expect. Our attempts to adapt the group algorithm from [@FleischerK18:short] failed and it is known from [@FleischerCCC18:short] that ${\mathbf{LG}}$ does not have the [$\mathrm{polylog}$[CP]{}]{}, so we cannot use Corollary \[crl:plcp-pcp\] as in the group case. To summarize, up to this point, the complexity landscape of ${\ensuremath{\textsc{SgpIsect}}}$ looks as follows. By Proposition \[prop:hardness\], the problem is ${\ensuremath{\mathsf{NP}}}$-hard for every variety ${\mathbf{V}}\not\subseteq {\mathbf{LI}}$. Using the ${\mathbf{DO}}$-algorithm from [@FleischerK18:short], we know that the problem is ${\ensuremath{\mathsf{NP}}}$-complete for every variety ${\mathbf{V}}\subseteq {\mathbf{DO}}$ not contained within ${\mathbf{LI}}$. For ${\mathbf{V}}\not\subseteq {\mathbf{LDS}}$, the problem is ${\ensuremath{\mathsf{PSPACE}}}$-complete. This leaves two classes of varieties for further investigation: 1. For ${\mathbf{V}}\not\subseteq {\mathbf{DO}}$ and ${\mathbf{V}}\subseteq {\mathbf{LDS}}$, we do not know whether the problem is always ${\ensuremath{\mathsf{NP}}}$-complete, whether it becomes ${\ensuremath{\mathsf{PSPACE}}}$-complete for varieties contained within ${\mathbf{LDS}}$ already and whether any other classes inside ${\mathbf{LDS}}$ are connected to natural complexity classes, such as the polynomial hierarchy. 2. Thus far, we do not have any hardness results for ${\mathbf{V}}\subseteq {\mathbf{LI}}$. The remainder of this section is devoted to the second class of varieties. On one hand, it is not difficult to see that ${\ensuremath{\textsc{SgpIsect}}}({\mathbf{LI}})$ is contained in ${\ensuremath{\mathsf{NP}}}$. On the other hand, ${\ensuremath{\mathsf{NP}}}$-hardness holds only for some subvarieties of ${\mathbf{LI}}$ but not for others. Containment in ${\ensuremath{\mathsf{NP}}}$ actually already follows from ${\mathbf{LI}}\subseteq {\mathbf{DO}}$ but it also is an immediate consequence of the following result. For each $k {{\geqslant}}1$, the variety ${\mathbf{LI}}_k$ has the $2k$ product circuits property. In particular, ${\ensuremath{\textsc{SgpIsect}}}({\mathbf{LI}}_k)$ is contained in ${\ensuremath{\mathsf{AC}}}^0$. \[thm:nerb\] It suffices to show that for each $k \in {\mathbb{N}}$ and for each finite semigroup $S \in {\mathbf{LI}}_k$, each morphism $h \colon A^+ \to S$ and each $u = a_1 \cdots a_\ell \in A^+$ with $\ell > 2k$, the word $v = a_1 \cdots a_k a_{\ell-k+1} \cdots a_\ell$ satisfies $h(v) = h(u)$. To see this, note that $$\begin{aligned} h(v) & = h(a_1 \cdots a_k a_{\ell-k+1} \cdots a_\ell) = h(a_1) \cdots h(a_k) h(a_{\ell-k+1}) \cdots h(a_\ell) \\ & = h(a_1) \cdots h(a_k) h(a_{k+1} \cdots a_{\ell-k}) h(a_{\ell-k+1}) \cdots h(a_\ell) = h(a_1 \cdots a_\ell) = h(u) \end{aligned}$$ where the third equality holds by the definition of ${\mathbf{LI}}_k$. The length of $v$ is ${\ensuremath\left|v\right|} = k + (\ell - (\ell - k)) = 2k$. Since the word $v$ does not depend on $h$ or on $S$, the canonical SLP of $v$ yields the desired product circuits property. Combining Theorem \[thm:nerb\] with Proposition \[prop:li-nerb\], we immediately obtain that ${\mathbf{LI}}$ has the $2n + 2$ product circuits property: each of the semigroups $S_1, \dots, S_k$ in the input has cardinality at most $N = {\ensuremath\left|S_1\right|} + \cdots + {\ensuremath\left|S_k\right|}$. Hence, all semigroups $S_i$ belong to the variety ${\mathbf{LI}}_{N+1}$ and there exists a witness of size at most $2N+2$. The variety ${\mathbf{LI}}$ has the [[$\mathrm{poly}$[PCP]{}]{}]{}. In particular, ${\ensuremath{\textsc{SgpIsect}}}({\mathbf{LI}})$ is contained in ${\ensuremath{\mathsf{NP}}}$. \[crl:li\] Another consequence of Proposition \[prop:li-nerb\] is ${\mathop{\bigcup}}_{k \in {\mathbb{N}}} {\mathbf{LI}}_k = {\mathbf{LI}}$. For each variety in the infinite sequence ${\mathbf{LI}}_1 \subseteq {\mathbf{LI}}_2 \subseteq \cdots$, the intersection problem is in ${\ensuremath{\mathsf{AC}}}^0$ but for its limit ${\mathbf{LI}}$, the problem is only contained in ${\ensuremath{\mathsf{NP}}}$—and it is actually ${\ensuremath{\mathsf{NP}}}$-complete, as we shall see later. Therefore, in contrast to previously obtained hardness results which relied on purely structural properties, other parameters interfere with the complexity of ${\ensuremath{\textsc{SgpIsect}}}$ below ${\mathbf{LI}}$. We will investigate this phenomenon more carefully. A semigroup is *monogenic* if it is generated by a single element. The *order* of a class ${\mathbf{C}}$ of finite semigroups is the supremum of the cardinalities of all monogenic subsemigroups contained in ${\mathbf{C}}$. If the order is $\infty$, the class is said to have *unbounded order*. The following observation will be used implicitly several times later. Let $S$ be a finite semigroup from ${\mathbf{LI}}$ and let $s \in S$. Then there exists some integer $n \in {\mathbb{N}}$ such that for all $i \in {\mathbb{N}}$, we have $s^{n+i} = s^n$. This integer is the order of the monogenic subsemigroup generated by $s$. Moreover, if $S$ is nilpotent, then $s^n$ is the zero element. \[lem:order-li\] Since $S$ is finite, there exist $n \in {\mathbb{N}}$ and $p {{\geqslant}}1$ with $s^n = s^{n+p}$. Let $n$ and $p$ be minimal with this property. If $p > 1$, then $s^{np+1}$ generates a non-trivial subgroup of $S$, a contradiction to the assumption that $S \in {\mathbf{LI}}$. Thus $p = 1$, yielding the first part of the statement. It is clear that $s^{2n} = s^{n+n} = s^n$, thus $s^n$ is idempotent. Since in a nilpotent semigroup, the only idempotent element is a zero element, we obtain the desired statement. In follow-up results, we will use reductions from ${\ensuremath{\textsc{3-Sat}}}$ to prove ${\ensuremath{\mathsf{NP}}}$-hardness of ${\ensuremath{\textsc{SgpIsect}}}$ for varieties of semigroups with certain properties. To simplify notation, let us introduce some definitions. For a set of *variables* $X = {\ensuremath{\left\{x_1, \dots, x_k\right\}}}$, we let $\overline X = {\ensuremath{\left\{\overline x \mid x \in X\right\}}}$ where each $\overline x$ is a new symbol. The set of *literals* over $X$ is $X {\mathbin{\cup}}\overline X$ and a set of literals is a *clause*. An *assignment* $\mathcal{A} \colon X \to {\ensuremath{\left\{0, 1\right\}}}$ of truth values to the variables $X$ can be extended to all literals over $X$ by letting $\mathcal{A}(\overline x) = 1 - \mathcal{A}(x)$ and to clauses $C \subseteq X {\mathbin{\cup}}\overline X$ by letting $\mathcal{A}(C) = \max{\ensuremath{\left\{\mathcal{A}(\ell) \mid \ell \in C\right\}}}$. An assignment $\mathcal{A}$ *satisfies* a set of clauses ${\ensuremath{\left\{C_1, \dots, C_n\right\}}}$ if $\mathcal{A}(C_j) = 1$ for all $j \in {\ensuremath{\left\{1, \dots, n\right\}}}$. For a word $w \in (X {\mathbin{\cup}}\overline X)^+$, the mapping $\mathcal{A}_w \colon X \to {\ensuremath{\left\{0, 1\right\}}}$ defined by $\mathcal{A}_w(\ell) = 1$ if and only if $w \in (X {\mathbin{\cup}}\overline X)^* \ell (X {\mathbin{\cup}}\overline X)^*$ for all $\ell \in X {\mathbin{\cup}}\overline X$ is called the *assignment induced by $w$*. Note that this assignment is well-defined whenever ${\ensuremath{\left\{w\right\}}} {\mathbin{\cap}}(X {\mathbin{\cup}}\overline X)^* x_i (X {\mathbin{\cup}}\overline X)^* {\mathbin{\cap}}(X {\mathbin{\cup}}\overline X)^* \overline{x_i} (X {\mathbin{\cup}}\overline X)^* = \emptyset$ for all $i \in {\ensuremath{\left\{1, \dots, k\right\}}}$. Conversely, for a given assignment $\mathcal{A} \colon X \to {\ensuremath{\left\{0, 1\right\}}}$, we call $w_\mathcal{A} = \ell_1 \cdots \ell_k$, where $\ell_i = x_i$ if $\mathcal{A}(x_i) = 1$ and $\ell_i = \overline{x_i}$ otherwise, the *word induced by $\mathcal{A}$*. If ${\mathbf{V}}$ is a variety of finite semigroups with unbounded order, then the decision problem ${\ensuremath{\textsc{SgpIsect}}}({\mathbf{V}})$ is ${\ensuremath{\mathsf{NP}}}$-hard. \[thm:unbounded\] We may assume ${\mathbf{V}}\subseteq {\mathbf{LI}}$, otherwise ${\ensuremath{\textsc{SgpIsect}}}({\mathbf{V}})$ is ${\ensuremath{\mathsf{NP}}}$-hard by Proposition \[prop:hardness\]. For each $k \in {\mathbb{N}}$ the semigroup $S_k = {\ensuremath{\left\{1, \dots, k\right\}}}$ with the binary operation $i \circ j = \min{\ensuremath{\left\{i+j,k\right\}}}$ is contained in ${\mathbf{V}}$. To see this, take some arbitrary $k \in {\mathbb{N}}$. Since ${\mathbf{V}}$ has unbounded order, some monogenic semigroup $T$ of cardinality $m {{\geqslant}}k$ appears as a subsemigroup in ${\mathbf{V}}$. Let $s$ be a generator of $T$. By Lemma \[lem:order-li\] and since $m {{\geqslant}}k$, the mapping $h \colon T \to S_k$ defined by $h(s) = 1$ is a surjective morphism. By closure of ${\mathbf{V}}$ under divisors, the semigroup $S_k$ itself belongs to ${\mathbf{V}}$. We now reduce ${\ensuremath{\textsc{3-Sat}}}$ to ${\ensuremath{\textsc{SgpIsect}}}({\mathbf{V}})$. Suppose we are given a set of variables $X = {\ensuremath{\left\{x_1, \dots, x_k\right\}}}$ and a set of clauses ${\ensuremath{\left\{C_1, \dots, C_n\right\}}}$ where $C_j = {\ensuremath{\left\{\ell_{j1}, \ell_{j2}, \ell_{j3}\right\}}}$ for each $j \in {\ensuremath{\left\{1, \dots, n\right\}}}$ and for literals $\ell_{j1}, \ell_{j2}, \ell_{j3}$ over $X$. We let $S = S_{k+2}$ be the monogenic semigroup of cardinality $k + 2$ defined above. We introduce morphisms $g_0, \dots, g_k, h_1, \dots, h_n \colon (X {\mathbin{\cup}}\overline X)^+ \to S$ defined by $$g_i(\ell) = \begin{cases} 2 & \text{if $i > 0$ and $\ell \in {\ensuremath{\left\{x_i, \overline{x_i}\right\}}}$}, \\ 1 & \text{otherwise}, \end{cases} \qquad h_j(\ell) = \begin{cases} 2 & \text{if $\ell \in C_j$}, \\ 1 & \text{otherwise}. \end{cases}$$ for $0 {{\leqslant}}i {{\leqslant}}k$ and $1 {{\leqslant}}j {{\leqslant}}n$. We let $P_0 = {\ensuremath{\left\{k\right\}}}$, $P_1 = \dots = P_k = {\ensuremath{\left\{k+1\right\}}}$ and $Q_1 = \dots = Q_n = {\ensuremath{\left\{k+1, k+2\right\}}}$. It is easy to check that the intersection $$L = \bigcap_{i = 0}^n g_i^{-1}(P_i) {\mathbin{\cap}}\bigcap_{j = 1}^k h_j^{-1}(Q_j)$$ is non-empty if and only if there exists a satisfying assignment. To see this, the following three observations are crucial: 1. $g_0^{-1}(P_0)$ contains all words over $(X {\mathbin{\cup}}\overline X)$ with exactly $k$ letters, 2. $g_i^{-1}(P_i) {\mathbin{\cap}}g_0^{-1}(P_0)$ contains all words from the set $(X {\mathbin{\cup}}\overline X)^k$ with exactly one occurrence of $x_i$ or exactly one occurrence of $\overline{x_i}$ (but not both), and 3. $h_j^{-1}(Q_j) {\mathbin{\cap}}g_0^{-1}(P_0)$ contains all words from the set $(X {\mathbin{\cup}}\overline X)^k$ with at least one occurrence of any of the literals $\ell_{j1}, \ell_{j2}, \ell_{j3}$. By the first two properties, all words from $L$ are of the form $\ell_1 \cdots \ell_k \in (X {\mathbin{\cup}}\overline X)^k$ with ${\ensuremath\left|{\ensuremath{\left\{\ell_1, \dots, \ell_k\right\}}} {\mathbin{\cap}}{\ensuremath{\left\{x_i, \overline{x_i}\right\}}}\right|} = 1$ for all $i \in {\ensuremath{\left\{1, \dots, k\right\}}}$. Thus, for each $w \in L$, the assignment $\mathcal{A}_w$ induced by $w$ is well-defined. Now, if $w \in L$, by the third property, we have $\mathcal{A}_w(\ell_{j1}) = 1$ or $\mathcal{A}_w(\ell_{j2}) = 1$ or $\mathcal{A}_w(\ell_{j3}) = 1$ for each $j \in {\ensuremath{\left\{1, \dots, n\right\}}}$. Thus, $\mathcal{A}_w$ is satisfying. Conversely, if there exists a satisfying assignment $\mathcal{A} \colon X \to {\ensuremath{\left\{0, 1\right\}}}$, the word induced by $\mathcal{A}$ is contained in $L$. It is obvious that the reduction can be performed in polynomial time. A more careful analysis shows that the reduction can even be carried out by a ${\ensuremath{\mathsf{AC}}}^0$ circuit family. To complement the previous result, let us now consider a very restricted variety of order $2$ (one can show that all varieties ${\mathbf{V}}\subseteq {\mathbf{LI}}$ of order $1$ are so-called *rectangular bands* and contained in ${\mathbf{LI}}_1$ already). ${\ensuremath{\textsc{SgpIsect}}}({\mathbf{A}}_2 \cap {\mathbf{N}})$ is ${\ensuremath{\mathsf{NP}}}$-complete. \[thm:nil-bounded\] As in the previous proof, we reduce ${\ensuremath{\textsc{3-Sat}}}$ to ${\ensuremath{\textsc{SgpIsect}}}({\mathbf{A}}_2 \cap {\mathbf{N}})$. Containment in ${\ensuremath{\mathsf{NP}}}$ follows from Corollary \[crl:li\] and from ${\mathbf{A}}_2 \cap {\mathbf{N}}\subseteq {\mathbf{N}}\subseteq {\mathbf{LI}}$. Suppose we are given a set of variables $X = {\ensuremath{\left\{x_1, \dots, x_k\right\}}}$ as well as a set of clauses ${\ensuremath{\left\{C_1, \dots, C_n\right\}}}$ where $C_j = {\ensuremath{\left\{\ell_{j1}, \ell_{j2}, \ell_{j3}\right\}}}$ for each $j \in {\ensuremath{\left\{1, \dots, n\right\}}}$ and literals $\ell_{j1}, \ell_{j2}, \ell_{j3}$ over $X$. Let $S$ be the finite semigroup ${\ensuremath{\left\{(i, j) \mid 1 {{\leqslant}}i {{\leqslant}}j {{\leqslant}}k\right\}}} {\mathbin{\cup}}{\ensuremath{\left\{0\right\}}}$ defined by the multiplication $$(i, j)(k, \ell) = \begin{cases} (i, \ell) & \text{if $k = j+1$}, \\ 0 & \text{otherwise}. \end{cases}$$ The element $0$ is a zero element. Let $g, h_1, \dots, h_n \colon (X {\mathbin{\cup}}\overline X)^+ \to S$ be the morphisms defined by $g(x_i) = g(\overline{x_i}) = (i, i)$ and by $$h_j(x_i) = \begin{cases} (i, i) & \text{if $x_i \not\in C_j$}, \\ 0 & \text{otherwise}, \end{cases} \qquad h_j(\overline{x_i}) = \begin{cases} (i, i) & \text{if $\overline{x_i} \not\in C_j$}, \\ 0 & \text{otherwise}. \end{cases}$$ for $1 {{\leqslant}}i {{\leqslant}}k$ and $1 {{\leqslant}}j {{\leqslant}}n$. As accepting sets, we choose $P = {\ensuremath{\left\{(1, k)\right\}}}$ for $g$ and $Q_1 = \dots = Q_n = {\ensuremath{\left\{0\right\}}}$ for $h_1, \dots, h_n$. Again, we would like to show that the intersection $$L = g^{-1}(P) {\mathbin{\cap}}\bigcap_{j = 1}^k h_j^{-1}(Q_j)$$ is non-empty if and only if there exists a satisfying assignment for ${\ensuremath{\left\{C_1, \dots, C_n\right\}}}$. The following two properties hold: 1. $g^{-1}(P)$ contains all words of the form $\ell_1 \cdots \ell_k$ with $\ell_i \in {\ensuremath{\left\{x_i, \overline{x_i}\right\}}}$ for $1 {{\leqslant}}i {{\leqslant}}k$, 2. $g^{-1}(P) {\mathbin{\cap}}h_j^{-1}(Q_j)$ contains all words of this form containing at least one of the letters $\ell_{j1}, \ell_{j2}, \ell_{j3}$. Let $w \in A^+$ be a word with $g(w) \in P$ and $h_j(w) \in Q_j$ for all $j \in {\ensuremath{\left\{1, \dots, n\right\}}}$. Then, by the first property above, the assignment $\mathcal{A}_w$ induced by $w$ is well-defined. Moreover, by the second property, we have $\mathcal{A}_w(C_1) = \dots = \mathcal{A}_w(C_n) = 1$ and thus, $\mathcal{A}_w$ satisfies ${\ensuremath{\left\{C_1, \dots, C_n\right\}}}$. Conversely, it is easy to see that each word induced by a satisfying assignment is contained in $L$. Note that the constructed semigroup belongs to ${\mathbf{A}}_2 \cap {\mathbf{N}}$ since by definition, we have $(i, j)(i, j) = 0$ for all $(i, j) \in S$. It is obvious that the reduction can be performed in polynomial time. In view of the previous theorems, the following result might be surprising. For the class of all commutative semigroups within ${\mathbf{LI}}$, the semigroup intersection problem is ${\ensuremath{\mathsf{NP}}}$-hard by Theorem \[thm:unbounded\]. The variety ${\mathbf{A}}_2 \cap {\mathbf{N}}$ has order $2$ and its semigroup intersection problem is ${\ensuremath{\mathsf{NP}}}$-hard by Theorem \[thm:nil-bounded\]. However, if we combine commutativity and bounded order, the problem becomes easier. If ${\mathbf{V}}\subseteq {\mathbf{Com}}{\mathbin{\cap}}{\mathbf{LI}}$ is a variety of finite semigroups with bounded order, then ${\mathbf{V}}$ has the [[$\mathrm{polylog}$[PCP]{}]{}]{}. Thus, there exists some $k \in {\mathbb{N}}$ such that ${\ensuremath{\textsc{SgpIsect}}}({\mathbf{V}}) \in {\ensuremath{\mathsf{qAC}}}^k$ and ${\ensuremath{\textsc{SgpIsect}}}({\mathbf{V}})$ is decidable in quasi-polynomial time. \[thm:bounded-com\] We show that if every monogenic subsemigroup of $S \in {\mathbf{Com}}{\mathbin{\cap}}{\mathbf{LI}}$ has size at most $k$, then every product of at least $k (\log{\ensuremath\left|S\right|} + 1)$ elements is the zero element. Thus, every non-empty intersection of languages recognized by multiple morphisms to such semigroups contains a witness of logarithmic size. Note that ${\mathbf{Com}}{\mathbin{\cap}}{\mathbf{LI}}\subseteq {\mathbf{N}}$, so the $k$-fold power of any element in $S$ is the zero element. Assume, for the sake of contradiction, that there exists a product of at least $k (\log{\ensuremath\left|S\right|} + 1)$ elements which is not the zero element. By reordering elements, we can rewrite this product as $s_1^{i_1} \cdots s_m^{i_m}$ with $s_i \ne s_j$ for $1 {{\leqslant}}i < j {{\leqslant}}m$. We proceed by induction on $m$. If $m {{\leqslant}}\log{\ensuremath\left|S\right|} + 1$, then there exists some $r \in {\ensuremath{\left\{1, \dots, m\right\}}}$ with $i_r {{\geqslant}}k$. Since each monogenic subsemigroup of $S$ has size at most $k$, the element $s_r^{i_r}$ then is a zero element, a contradiction. Suppose now that $m > \log{\ensuremath\left|S\right|} + 1$. The set $T = \mathcal{P}({\ensuremath{\left\{1, \dots, m\right\}}}) \setminus {\ensuremath{\left\{\emptyset\right\}}}$ forms a semigroup with union as binary operation. Let $h \colon T \to S$ be the morphism defined by $h(r) = s_r^{i_r}$ for $1 {{\leqslant}}r {{\leqslant}}m$. We have ${\ensuremath\left|T\right|} = 2^m - 1 {{\geqslant}}2^{m-1} > 2^{\log{\ensuremath\left|S\right|}} = {\ensuremath\left|S\right|}$. Thus, by the pigeon hole principle, there exist two sets $K_1, K_2 \subseteq {\ensuremath{\left\{1, \dots, m\right\}}}$ with $K_1 \ne K_2$ and $h(K_1) = h(K_2)$. If $K_1 \subsetneq K_2$, then multiplying the product by $h(K_2 \setminus K_1)$ does not change its value and $k$-fold multiplication shows that the product is zero, a contradiction. The case $K_2 \subsetneq K_1$ is symmetric. Thus, we may assume that neither $K_1 \subseteq K_2$ nor $K_2 \subseteq K_1$. The *length* of a set $K \subseteq {\ensuremath{\left\{1, \dots, m\right\}}}$ is the sum of all $i_r$ with $r \in K$. By symmetry, we may assume that the length of $K_1$ is at most the length of $K_2$. We replace the factor $h(K_1)$ of the product by $h(K_2)$ and obtain the statement by induction on the number $m$ of different elements in the product—the length of this new product $h(K_2) h({\ensuremath{\left\{1, \dots, m\right\}}} \setminus K_1)$ is at least the length of the original product and the number of different elements decreases since $K_1 \setminus K_2 \ne \emptyset$. Open Problems ============= It remains open whether the observation that hardness is not always caused by purely structural properties also applies to varieties between ${\mathbf{LI}}$ and ${\mathbf{LDS}}$ in the semigroup setting, between ${\mathbf{DO}}$ and ${\mathbf{DS}}$ in the monoid setting or between ${\mathbf{R}}$ and ${\mathbf{DS}}$ in the automaton setting. Another major challenge is obtaining algebraic characterizations of all classes of finite semigroups with the [$\mathrm{poly}$[PCP]{}]{}. As a first step, we suggest proving (or disproving) that the variety ${\mathbf{LG}}$ has the [$\mathrm{poly}$[PCP]{}]{}. #### Acknowledgements. I would like to thank the anonymous referees of the conference version of this paper for providing helpful comments. [10]{} L. Babai, E. M. Luks, and [Á]{}. Seress. Permutation groups in [NC]{}. In [*STOC 1987, Proceedings*]{}, pages 409–420, 1987. L. Babai and E. Szemeredi. On the complexity of matrix group problems [I]{}. In [*25th Annual Symposium on Foundations of Computer Science*]{}, pages 229–240, Oct 1984. D. A. M. Barrington. Quasipolynomial size circuit classes. In [*Proceedings of the Seventh Annual Structure in Complexity Theory Conference*]{}, pages 86–93, Jun 1992. M. Beaudry. . PhD thesis, McGill University, Montreal, Quebec, 1988. M. Beaudry, P. McKenzie, and D. Th[é]{}rien. The membership problem in aperiodic transformation monoids. , 39(3):599–616, 1992. M. Charikar, E. Lehman, D. Liu, R. Panigrahy, M. Prabhakaran, A. Sahai, and A. Shelat. The smallest grammar problem. , 51(7):2554–2576, July 2005. L. Fleischer. . In [*CCC 2018, Proceedings*]{}, pages 25:1–25:12. Dagstuhl Publishing, 2018. L. Fleischer and M. Kufleitner. The intersection problem for finite monoids. In [*STACS 2018, Proceedings*]{}, pages 30:1–30:14. Dagstuhl Publishing, 2018. M. Furst, J. Hopcroft, and E. Luks. Polynomial-time algorithms for permutation groups. In [*SFCS 1980, Proceedings*]{}, pages 36–41, Oct 1980. R. Impagliazzo and R. Paturi. Complexity of k-[SAT]{}. In [*CCC 1999, Proceedings*]{}, pages 237–240, 1999. D. Kozen. Lower bounds for natural proof systems. In [*FOCS 1977, Proceedings*]{}, pages 254–266, Providence, Rhode Island, 1977. IEEE Computer Society Press. C. C. Sims. Computational methods in the study of permutation groups. In [*Proceedings of the Conference on Computational Problems in Abstract Algebra 1967, Oxford, United Kingdom*]{}, pages 169–183, New York, 1968. Pergamon. H. Straubing. Finite semigroup varieties of the form [$\mathbf{V}\ast \mathbf{D}$]{}. , 36(1):53–94, 1985. H. Straubing. . Birkh[ä]{}user, Boston, Basel and Berlin, 1994. P. Tesson and D. Th[é]{}rien. Diamonds are forever: [T]{}he variety $\mathrm{DA}$. In [*Semigroups, Algorithms, Automata and Languages 2001, Proceedings*]{}, pages 475–500. World Scientific, 2002. H. Vollmer. . Springer, Berlin, 1999. [^1]: Supported by the German Research Foundation (DFG) under grant DI 435/5–2.

ArXiv

--- author: - 'David R. Dalton[^1], Michael A. Slawinski [^2]' date: 'January 12, 2016' title: On commutativity of Backus and Gazis averages --- Abstract {#abstract .unnumbered} ======== We show that the Backus (1962) equivalent-medium average, which is an average over a spatial variable, and the Gazis et al. (1963) effective-medium average, which is an average over a symmetry group, do not commute, in general. They commute in special cases, which we exemplify. Introduction ============ Hookean solids are defined by their mechanical property relating linearly the stress tensor, $\sigma$, and the strain tensor, $\varepsilon$, $$\sigma_{ij}=\sum_{k=1}^3\sum_{\ell=1}^3c_{ijk\ell}\varepsilon_{k\ell}\,,\qquad i,j=1,2,3 \,.$$ The elasticity tensor, $c$, belongs to one of eight material-symmetry classes shown in Figure \[fig:orderrelation\]. ![[]{data-label="fig:orderrelation"}](FigPartialOrder.pdf) The Backus (1962) moving average allows us to quantify the response of a wave propagating through a series of parallel layers whose thicknesses are much smaller than the wavelength. Each layer is a Hookean solid exhibiting a given material symmetry with given elasticity parameters. The average is a Hookean solid whose elasticity parameters—and, hence, its material symmetry—allow us to model a long-wavelength response. This material symmetry of the resulting medium, to which we refer as [*equivalent*]{}, is a consequence of symmetries exhibited by the averaged layers. The long-wave-equivalent medium to a stack of isotropic or transversely isotropic layers with thicknesses much less than the signal wavelength was shown by Backus (1962) to be a homogeneous or nearly homogeneous transversely isotropic medium, where a [*nearly*]{} homogeneous medium is a consequence of a [*moving*]{} average. Backus (1962) formulation is reviewed by Slawinski (2016) and Bos et al. (2016), where formulations for generally anisotropic, monoclinic, and orthotropic thin layers are also derived. Bos et al. (2016) examine the underlying assumptions and approximations behind the Backus (1962) formulation, which is derived by expressing rapidly varying stresses and strains in terms of products of algebraic combinations of rapidly varying elasticity parameters with slowly varying stresses and strains. The only mathematical approximation in the formulation is that the average of a product of a rapidly varying function and a slowly varying function is approximately equal to the product of the averages of the two functions. According to Backus (1962), the average of $f(x_3)$ of “width” $\ell'$ is $$\label{eq:BackusOne} \overline f(x_3):=\int\limits_{-\infty}^\infty w(\zeta-x_3)f(\zeta)\,{\rm d}\zeta \,,$$ where $w(x_3)$ is the weight function with the following properties: $$w(x_3)\geqslant0\,, \quad w(\pm\infty)=0\,, \quad \int\limits_{-\infty}^\infty w(x_3)\,{\rm d}x_3=1\,, \quad \int\limits_{-\infty}^\infty x_3w(x_3)\,{\rm d}x_3=0\,, \quad \int\limits_{-\infty}^\infty x_3^2w(x_3)\,{\rm d}x_3=(\ell')^2\,.$$ These properties define $w(x_3)$ as a probability-density function with mean $0$ and standard deviation $\ell'$, explaining the use of the term “width” for $\ell'$. Gazis et al. (1963) average allows us to obtain the closest symmetric counterpart—in the Frobenius sense—of a chosen material symmetry to a generally anisotropic Hookean solid. The average is a Hookean solid, to which we refer as [*effective*]{}, whose elasticity parameters correspond to the symmetry chosen [*a priori*]{}. Gazis average is a projection given by $$\widetilde c^{\,\,\rm sym}:=\intop_{G^{\rm sym}}(g\circ c)\,\mathrm{d}\mu(g) \,, \label{eq:proj}$$ where the integration is over the symmetry group, $G^{\rm sym}$, whose elements are $g$, with respect to the invariant measure, $\mu$, normalized so that $\mu(G^{\rm sym})=1$; $\widetilde c^{\,\,\rm sym}$ is the orthogonal projection of $c$, in the sense of the Frobenius norm, on the linear space containing all tensors of that symmetry, which are $ c^{\,\,\rm sym}$. Integral (\[eq:proj\]) reduces to a finite sum for the classes whose symmetry groups are finite, which are all classes except isotropy and transverse isotropy. The Gazis et al.  (1963) approach is reviewed and extended by Danek et al. (2013, 2015) in the context of random errors. Therein, elasticity tensors are not constrained to the same—or even different but known—orientation of the coordinate system. Concluding this introduction, let us emphasize that the fundamental distinction between the two averages is their domain of operation. The Gazis et al. (1963) average is an average over symmetry groups at a point and the Backus (1962) average is a spatial average over a distance. Both averages can be used, separately or together, in quantitative seismology. Hence, an examination of their commutativity might provide us with an insight into their physical meaning and into allowable mathematical operations. Generally anisotropic layers and monoclinic medium ================================================== Let us consider a stack of generally anisotropic layers to obtain a monoclinic medium. To examine the commutativity between the Backus and Gazis averages, let us study the following diagram, $$\label{eq:CD2} \begin{CD} \rm{aniso}@>\rm{B}>>\rm{aniso}\\ @V\mathrm{G}VV @VV\rm{G}V\\ \rm{mono}@>>\rm{B}>\rm{mono} \end{CD}$$ and Proposition \[thm:One\], below, \[thm:One\] In general, the Backus and Gazis averages do not commute. To prove this proposition and in view of Diagram \[eq:CD2\], let us begin with the following corollary. For the generally anisotropic and monoclinic symmetries, the Backus and Gazis averages do not commute. To understand this corollary, we invoke the following lemma, whose proof is in \[AppOne1\]. \[lem:Mono\] For the effective monoclinic symmetry, the result of the Gazis average is tantamount to replacing each $c_{ijk\ell}$, in a generally anisotropic tensor, by its corresponding $c_{ijk\ell}$ of the monoclinic tensor, expressed in the natural coordinate system, including replacements of the anisotropic-tensor components by the zeros of the corresponding monoclinic components. Let us first examine the counterclockwise path of Diagram \[eq:CD2\]. Lemma \[lem:Mono\] entails a corollary. \[col:Mono\] For the effective monoclinic symmetry, given a generally anisotropic tensor, $C$, $$\label{eq:GazisMono} \widetilde{C}^{\,\rm mono}=C^{\,\rm mono} \,;$$ where $\widetilde{C}^{\,\rm mono}$ is the Gazis average of $C$, and $C^{\,\rm mono}$ is a monoclinic tensor whose nonzero entries are the same as for $C$. According to Corollary \[col:Mono\], the effective monoclinic tensor is obtained simply by setting to zero—in the generally anisotropic tensor—the components that are zero for the monoclinic tensor. Then, the second counterclockwise branch of Diagram \[eq:CD2\] is performed as follows. Applying the Backus average, we obtain (Bos et al., 2015) $$\langle c_{3333}\rangle=\overline{\left(\frac{1}{c_{3333}}\right)}^{\,\,-1}\,, \qquad \langle c_{2323}\rangle=\frac{\overline{\left(\frac{c_{2323}}{D}\right)}}{2D_2}\,,$$ $$\langle c_{1313}\rangle=\frac{\overline{\left(\frac{c_{1313}}{D}\right)}}{2D_2}\,, \qquad \langle c_{2313}\rangle=\frac{\overline{\left(\frac{c_{2313}}{D}\right)}}{2D_2}\,,$$ where $D\equiv 2(c_{2323}c_{1313}-c_{2313}^2)$ and $D_2\equiv (\overline{c_{1313}/D})(\overline{c_{2323}/D})-(\overline{c_{2313}/D})^2$. We also obtain $$\langle c_{1133}\rangle= \overline{\left(\frac{1}{c_{3333}}\right)}^{\,\,-1} \overline{\left(\frac{c_{1133}}{c_{3333}}\right)}\,, \quad \langle c_{2233}\rangle= \overline{\left(\frac{1}{c_{3333}}\right)}^{\,\,-1} \overline{\left(\frac{c_{2233}}{c_{3333}}\right)}\,, \quad \langle c_{3312}\rangle= \overline{\left(\frac{1}{c_{3333}}\right)}^{\,\,-1} \overline{\left(\frac{c_{3312}}{c_{3333}}\right)}\,,$$ $$\langle c_{1111}\rangle= \overline{c_{1111}}-\overline{\left(\frac{c_{1133}^2}{c_{3333}}\right)}+ \overline{\left(\frac{1}{c_{3333}}\right)}^{\,\,-1} \overline{\left(\frac{c_{1133}}{c_{3333}}\right)}^{\,2}\,,$$ $$\langle c_{1122}\rangle= \overline{c_{1122}}-\overline{\left(\frac{c_{1133}\,c_{2233}}{c_{3333}}\right)}+ \overline{\left(\frac{1}{c_{3333}}\right)}^{\,\,-1} \overline{\left(\frac{c_{1133}}{c_{3333}}\right)}\,\, \overline{\left(\frac{c_{2233}}{c_{3333}}\right)}\,,$$ $$\langle c_{2222}\rangle= \overline{c_{2222}}-\overline{\left(\frac{c_{2233}^2}{c_{3333}}\right)}+ \overline{\left(\frac{1}{c_{3333}}\right)}^{\,\,-1} \overline{\left(\frac{c_{2233}}{c_{3333}}\right)}^{\,2}\,,$$ $$\langle c_{1212}\rangle= \overline{c_{1212}}-\overline{\left(\frac{c_{3312}^2}{c_{3333}}\right)}+ \overline{\left(\frac{1}{c_{3333}}\right)}^{\,\,-1} \overline{\left(\frac{c_{3312}}{c_{3333}}\right)}^{\,2}\,,$$ $$\langle c_{1112}\rangle= \overline{c_{1112}}-\overline{\left(\frac{c_{3312}\,c_{1133}}{c_{3333}}\right)}+ \overline{\left(\frac{1}{c_{3333}}\right)}^{\,\,-1} \overline{\left(\frac{c_{1133}}{c_{3333}}\right)}\,\, \overline{\left(\frac{c_{3312}}{c_{3333}}\right)}$$ and $$\langle c_{2212}\rangle= \overline{c_{2212}}-\overline{\left(\frac{c_{3312}\,c_{2233}}{c_{3333}}\right)}+ \overline{\left(\frac{1}{c_{3333}}\right)}^{\,\,-1} \overline{\left(\frac{c_{2233}}{c_{3333}}\right)}\,\, \overline{\left(\frac{c_{3312}}{c_{3333}}\right)}\,,$$ where angle brackets denote the equivalent-medium elasticity parameters. The other equivalent-medium elasticity parameters are zero. Following the clockwise path of Diagram \[eq:CD2\], the upper branch is derived in matrix form in Bos et al. (2015). Then, from Bos et al. (2015) the result of the right-hand branch is derived by setting entries in the generally anisotropic tensor that are zero for the monoclinic tensor to zero. The nonzero entries, which are too complicated to display explicitly, are—in general—not the same as the result of the counterclockwise path. Hence, for generally anisotropic and monoclinic symmetries, the Backus and Gazis averages do not commute. Higher symmetries ================= Monoclinic layers and orthotropic medium {#sec:mono} ---------------------------------------- Proposition \[thm:One\] remains valid for layers exhibiting higher material symmetries, and simpler expressions of the corresponding elasticity tensors allow us to examine special cases that result in commutativity. Let us consider the following corollary of Proposition \[thm:One\]. \[thm:Two\] For the monoclinic and orthotropic symmetries, the Backus and Gazis averages do not commute. To study this corollary, let us consider the following diagram, $$\label{eq:CD} \begin{CD} \rm{mono}@>\rm{B}>>\rm{mono}\\ @V\mathrm{G}VV @VV\rm{G}V\\ \rm{ortho}@>>\rm{B}>\rm{ortho} \end{CD}$$ and the lemma, whose proof is in \[AppOne2\]. \[lem:Ortho\] For the effective orthotropic symmetry, the result of the Gazis average is tantamount to replacing each $c_{ijk\ell}$, in a generally anisotropic—or monoclinic—tensor, by its corresponding $c_{ijk\ell}$ of the orthotropic tensor, expressed in the natural coordinate system, including the replacements by the corresponding zeros. Lemma \[lem:Ortho\] entails a corollary. \[col:Ortho\] For the effective orthotropic symmetry, given a generally anisotropic—or monoclinic—tensor, $C$, $$\label{eq:GazisOrtho} \widetilde{C}^{\,\rm ortho}=C^{\,\rm ortho} \,.$$ where $\widetilde{C}^{\,\rm ortho}$ is the Gazis average of $C$, and $C^{\,\rm ortho}$ is an orthotropic tensor whose nonzero entries are the same as for $C$. Let us consider a monoclinic tensor and proceed counterclockwise along the first branch of Diagram \[eq:CD\]. Using the fact that the monoclinic symmetry is a special case of general anisotropy, we invoke Corollary \[col:Ortho\] to conclude that $\widetilde{C}^{\,\rm ortho}=C^{\,\rm ortho}$, which is equivalent to setting $c_{1112}$, $c_{2212}$, $c_{3312}$ and $c_{2313}$ to zero in the monoclinic tensor. We perform the upper branch of Diagram \[eq:CD\], which is the averaging of a stack of monoclinic layers to get a monoclinic equivalent medium, as in the case of the lower branch of Diagram \[eq:CD2\]. Thus, following the clockwise path, we obtain $$c_{1212}^\circlearrowright= \overline{c_{1212}}-\overline{\left(\frac{c_{3312}^2}{c_{3333}}\right)}+ \overline{\left(\frac{1}{c_{3333}}\right)}^{\,\,-1} \overline{\left(\frac{c_{3312}}{c_{3333}}\right)}^{\,2}\,,$$ $$c_{1313}^\circlearrowright=\overline{\left(\frac{c_{1313}}{D}\right)}/(2D_2)\,,\qquad c_{2323}^\circlearrowright=\overline{\left(\frac{c_{2323}}{D}\right)}/(2D_2)$$ Following the counterclockwise path, we obtain $$c_{1212}^\circlearrowleft=\overline{c_{1212}}\,,\quad c_{1313}^\circlearrowleft=\overline{\left(\frac{1}{c_{1313}}\right)}^{\,\,-1}\,,\quad c_{2323}^\circlearrowleft=\overline{\left(\frac{1}{c_{2323}}\right)}^{\,\,-1}\,.$$ The other entries are the same for both paths. In conclusion, the results of the clockwise and counterclockwise paths are the same if $c_{2313}=c_{3312}=0$, which is a special case of monoclinic symmetry. Thus, the Backus average and Gazis average commute for that case, but not in general. Orthotropic layers and tetragonal medium {#sec:ortho} ---------------------------------------- In a manner analogous to Diagram \[eq:CD\], but proceeding from the the upper-left-hand corner orthotropic tensor to lower-right-hand corner tetragonal tensor by the counterclockwise path, $$\label{eq:CD3} \begin{CD} \rm{ortho}@>\rm{B}>>\rm{ortho}\\ @V\mathrm{G}VV @VV\rm{G}V\\ \rm{tetra}@>>\rm{B}>\rm{tetra} \end{CD}$$ we obtain $$c_{1111}^\circlearrowleft=\overline{\frac{c_{1111}+c_{2222}}{2}- \frac{\left(\frac{c_{1111}+c_{2222}}{2}\right)^2}{c_{3333}}}+ \overline{\left(\frac{c_{1111}+c_{2222}}{2c_{3333}}\right)}^2 \overline{\left(\frac{1}{c_{3333}}\right)}^{\,\,-1} \,.$$ Following the clockwise path, we obtain $$c_{1111}^\circlearrowright=\overline{\frac{c_{1111}+c_{2222}}{2}- \frac{c_{1133}^2+c_{2233}^2}{2c_{3333}}}+ \frac{1}{2}\left[\overline{\left(\frac{c_{1133}}{c_{3333}}\right)}^2+ \overline{\left(\frac{c_{2233}}{c_{3333}}\right)}^2\right] \overline{\left(\frac{1}{c_{3333}}\right)}^{\,\,-1}\,.$$ These results are not equal to one another, unless $c_{1133}=c_{2233}$, which is a special case of orthotropic symmetry. Also $c_{2323}$ must equal $c_{1313}$ for $c_{2323}^\circlearrowright=c_{2323}^\circlearrowleft$. The other entries are the same for both paths. Thus, the Backus average and Gazis average do commute for $c_{1133}=c_{2233}$ and $c_{2323}=c_{1313}$, which is a special case of orthotropic symmetry, but not in general. Let us also consider the case of monoclinic layers and a tetragonal medium to examine the process of combining the Gazis averages, which is tantamount to combining Diagrams (\[eq:CD\]) and (\[eq:CD3\]), $$\begin{CD} \label{eq:CD4} \rm{mono}@>\rm{B}>>\rm{mono}\\ @V\mathrm{G}VV @VV\rm{G}V\\ \rm{ortho}@>>\rm{B}>\rm{ortho}\\ @V\mathrm{G}VV @VV\rm{G}V\\ \rm{tetra}@>>\rm{B}>\rm{tetra} \end{CD}$$ In accordance with Proposition \[thm:One\], there is—in general—no commutativity. However, the outcomes are the same as for the corresponding steps in Sections \[sec:mono\] and \[sec:ortho\]. In general, for the Gazis average, proceeding directly, $\rm{aniso}\xrightarrow{\rm{G}}\rm{iso}$, is tantamount to proceeding along arrows in Figure \[fig:orderrelation\], $\rm{aniso}\xrightarrow{\rm{G}}\cdots\xrightarrow{\rm{G}}\rm{iso}$. No such combining of the Backus averages is possible, since, for each step, layers become a homogeneous medium. Transversely isotropic layers ----------------------------- Lack of commutativity can also be exemplified by the case of transversely isotropic layers. Following the clockwise path of Diagram \[eq:CD\], the Backus average results in a transversely isotropic medium, whose Gazis average—in accordance with Figure \[fig:orderrelation\]—is isotropic. Following the counterclockwise path, Gazis average results in an isotropic medium, whose Backus average, however, is transverse isotropy. Thus, not only the elasticity parameters, but even the resulting material-symmetry classes differ. Also, we could—in a manner analogous to the one illustrated in Diagram \[eq:CD4\]—begin with generally anisotropic layers and obtain isotropy by the clockwise path and transverse isotropy by the counterclockwise path, which again illustrates noncommutativity. Discussion ========== Herein, we assume that all tensors are expressed in the same orientation of their coordinate systems. Otherwise, the process of averaging become more complicated, as discussed—for the Gazis average—by Kochetov and Slawinski (2009a, 2009b) and as mentioned—for the Backus average—by Bos et al. (2016). Mathematically, the noncommutativity of two distinct averages is shown by Proposition \[thm:One\], and exemplified for several material symmetries. We do not see a physical justification for special cases in which—given the same orientation of coordinate systems—these averages commute. This behaviour might support the view that a mathematical realm, which allows for fruitful analogies with the physical world, has no causal connection with it. Acknowledgments {#acknowledgments .unnumbered} =============== We wish to acknowledge discussions with Theodore Stanoev. This research was performed in the context of The Geomechanics Project supported by Husky Energy. Also, this research was partially supported by the Natural Sciences and Engineering Research Council of Canada, grant 238416-2013. References {#references .unnumbered} ========== =0.4in=1 Backus, G.E., Long-wave elastic anisotropy produced by horizontal layering, [*J. Geophys. Res.*]{}, [**67**]{}, 11, 4427–4440, 1962. =0.4in=1 Bóna, A., I. Bucataru and M.A. Slawinski, Space of $SO(3)$-orbits of elasticity tensors, [*Archives of Mechanics*]{}, [**60**]{}, 2, 121–136, 2008 =0.4in=1 Bos, L, D.R. Dalton, M.A. Slawinski and T. Stanoev, On Backus average for generally anisotropic layers, [*arXiv*]{}, 2016. =0.4in=1 Chapman, C. H., [*Fundamentals of seismic wave propagation*]{}, Cambridge University Press, 2004. =0.4in=1 Danek, T., M. Kochetov and M.A. Slawinski, Uncertainty analysis of effective elasticity tensors using quaternion-based global optimization and Monte-Carlo method, [*The Quarterly Journal of Mechanics and Applied Mathematics*]{}, [**66**]{}, 2, pp. 253–272, 2013. =0.4in=1 Danek, T., M. Kochetov and M.A. Slawinski, Effective elasticity tensors in the context of random errors, [*Journal of Elasticity*]{}, 2015. =0.4in=1 Gazis, D.C., I. Tadjbakhsh and R.A. Toupin, The elastic tensor of given symmetry nearest to an anisotropic elastic tensor, [*Acta Crystallographica*]{}, [**16**]{}, 9, 917–922, 1963. =0.4in=1 Kochetov, M. and M.A. Slawinski, On obtaining effective orthotropic elasticity tensors, [*The Quarterly Journal of Mechanics and Applied Mathematics*]{}, [**62**]{}, 2, pp. 149-Ð166, 2009a. =0.4in=1 Kochetov, M. and M.A. Slawinski, On obtaining effective transversely isotropic elasticity tensors, [*Journal of Elasticity*]{}, [**94**]{}, 1Ð-13., 2009b. =0.4in=1 Slawinski, M.A. [*Wavefronts and rays in seismology: Answers to unasked questions*]{}, World Scientific, 2016. =0.4in=1 Slawinski, M.A., [*Waves and rays in elastic continua*]{}, World Scientific, 2015. =0.4in=1 Thomson, W., [*Mathematical and physical papers: Elasticity, heat, electromagnetism*]{}, Cambridge University Press, 1890 {#AppOne1} Let us prove Lemma \[lem:Mono\]. For discrete symmetries, we can write integral (\[eq:proj\]) as a sum, $$\label{eq:AverageDisc} \widetilde C^{\,\rm sym}=\frac{1}{n}\left(\tilde{A}_1^{\rm sym}\,C\,\tilde{A}_1^{\rm sym}\,{}^{^T}+\ldots+\tilde{A}_n^{\rm sym}\,C\,\tilde{A}_n^{\rm sym}\,{}^{^T}\right) \,,$$ where $\widetilde C^{\rm sym}$ is expressed in Kelvin’s notation, in view of Thomson (1890, p. 110) as discussed in Chapman (2004, Section 4.4.2). To write the elements of the monoclinic symmetry group as $6\times 6$ matrices, we must consider orthogonal transformations in $\mathbb{R}^3$. Transformation $A\in SO(3)$ of $c_{ijk\ell}$ corresponds to transformation of $C$ given by $${\footnotesize \tilde{A}=\left[\begin{array}{cccccc} A_{11}^{2} & A_{12}^{2} & A_{13}^{2} & \sqrt{2}A_{12}A_{13} & \sqrt{2}A_{11}A_{13} & \sqrt{2}A_{11}A_{12}\\ A_{21}^{2} & A_{22}^{2} & A_{23}^{2} & \sqrt{2}A_{22}A_{23} & \sqrt{2}A_{21}A_{23} & \sqrt{2}A_{21}A_{22}\\ A_{31}^{2} & A_{32}^{2} & A_{33}^{2} & \sqrt{2}A_{32}A_{33} & \sqrt{2}A_{31}A_{33} & \sqrt{2}A_{31}A_{32}\\ \sqrt{2}A_{21}A_{31} & \sqrt{2}A_{22}A_{32} & \sqrt{2}A_{23}A_{33} & A_{23}A_{32}+A_{22}A_{33} & A_{23}A_{31}+A_{21}A_{33} & A_{22}A_{31}+A_{21}A_{32}\\ \sqrt{2}A_{11}A_{31} & \sqrt{2}A_{12}A_{32} & \sqrt{2}A_{13}A_{33} & A_{13}A_{32}+A_{12}A_{33} & A_{13}A_{31}+A_{11}A_{33} & A_{12}A_{31}+A_{11}A_{32}\\ \sqrt{2}A_{11}A_{21} & \sqrt{2}A_{12}A_{22} & \sqrt{2}A_{13}A_{23} & A_{13}A_{22}+A_{12}A_{23} & A_{13}A_{21}+A_{11}A_{23} & A_{12}A_{21}+A_{11}A_{22}\end{array}\right]} \,, \label{eq:ATildeQ}$$ which is an orthogonal matrix, $\tilde{A}\in SO(6)$ (Slawinski (2015), Section 5.2.5).[^3] The required symmetry-group elements are $$A_1^{\rm mono}= \left[ \begin{array}{ccc} 1 & 0 & 0\\ 0 & 1 & 0\\ 0 & 0 & 1\end{array}\right] \mapsto \left[\begin{array}{cccccc} 1 & 0 & 0 & 0 & 0 & 0\\ 0 & 1 & 0 & 0 & 0 & 0\\ 0 & 0 & 1 & 0 & 0 & 0\\ 0 & 0 & 0 & 1 & 0 & 0\\ 0 & 0 & 0 & 0 & 1 & 0\\ 0 & 0 & 0 & 0 & 0 & 1\end{array}\right] =\tilde{A}_1^{\rm mono}$$ $$A_2^{\rm mono}= \left[ \begin{array}{ccc} -1 & 0 & 0\\ 0 & -1 & 0\\ 0 & 0 & 1\end{array}\right] \mapsto \left[\begin{array}{cccccc} 1 & 0 & 0 & 0 & 0 & 0\\ 0 & 1 & 0 & 0 & 0 & 0\\ 0 & 0 & 1 & 0 & 0 & 0\\ 0 & 0 & 0 & -1 & 0 & 0\\ 0 & 0 & 0 & 0 & -1 & 0\\ 0 & 0 & 0 & 0 & 0 & 1\end{array}\right] =\tilde{A}_2^{\rm mono} \,.$$ For the monoclinic case, expression (\[eq:AverageDisc\]) can be stated explicitly as $$\widetilde C^{\rm mono}= \frac{\left(\tilde{A}_1^{\rm mono}\right)\,C\,\left(\tilde{A}_1^{\rm mono}\right)^T+\left(\tilde{A}_2^{\rm mono}\right)\,C\,\left(\tilde{A}_2^{\rm mono}\right)^T}{2} \,.$$ Performing matrix operations, we obtain $$\widetilde C^{\rm mono} =\left[\begin{array}{cccccc} c_{1111} & c_{1122} & c_{1133} & 0 & 0 & \sqrt{2}c_{1112}\\ c_{1122} & c_{2222} & c_{2233} & 0 & 0 & \sqrt{2}c_{2212}\\ c_{1133} & c_{2233} & c_{3333} & 0 & 0 & \sqrt{2}c_{3312}\\ 0 & 0 & 0 & 2c_{2323} & 2c_{2313} & 0\\ 0 & 0 & 0 & 2c_{2313} & 2c_{1313} & 0\\ \sqrt{2}c_{1112} & \sqrt{2}c_{2212} & \sqrt{2}c_{3312} & 0 & 0 & 2c_{1212} \end{array}\right] \,, \label{eq:MonoExplicitRef}$$ which exhibits the form of the monoclinic tensor in its natural coordinate system. In other words, $\widetilde{C}^{\rm mono}=C^{\rm mono}$, in accordance with Corollary \[col:Mono\]. {#AppOne2} Let us prove Lemma \[lem:Ortho\]. For orthotropic symmetry, $\tilde{A}_1^{\rm ortho}=\tilde{A}_1^{\rm mono}$ and $\tilde{A}_2^{\rm ortho}=\tilde{A}_2^{\rm mono}$ and $$A_3^{\rm ortho}= \left[ \begin{array}{ccc} -1 & 0 & 0\\ 0 & 1 & 0\\ 0 & 0 & -1\end{array}\right] \mapsto \left[\begin{array}{cccccc} 1 & 0 & 0 & 0 & 0 & 0\\ 0 & 1 & 0 & 0 & 0 & 0\\ 0 & 0 & 1 & 0 & 0 & 0\\ 0 & 0 & 0 & -1 & 0 & 0\\ 0 & 0 & 0 & 0 & 1 & 0\\ 0 & 0 & 0 & 0 & 0 & -1\end{array}\right] =\tilde{A}_3^{\rm ortho} \,,$$ $$A_4^{\rm ortho}= \left[ \begin{array}{ccc} 1 & 0 & 0\\ 0 & -1 & 0\\ 0 & 0 & -1\end{array}\right] \mapsto \left[\begin{array}{cccccc} 1 & 0 & 0 & 0 & 0 & 0\\ 0 & 1 & 0 & 0 & 0 & 0\\ 0 & 0 & 1 & 0 & 0 & 0\\ 0 & 0 & 0 & 1 & 0 & 0\\ 0 & 0 & 0 & 0 & -1 & 0\\ 0 & 0 & 0 & 0 & 0 & -1\end{array}\right] =\tilde{A}_4^{\rm ortho} \,.$$ For the orthotropic case, expression (\[eq:AverageDisc\]) can be stated explicitly as [$$\widetilde C^{\rm ortho}= \frac{\left(\tilde{A}_1^{\rm ortho}\right)\,C\,\left(\tilde{A}_1^{\rm ortho}\right)^T+\left(\tilde{A}_2^{\rm ortho}\right)\,C\,\left(\tilde{A}_2^{\rm ortho}\right)^T +\left(\tilde{A}_3^{\rm ortho}\right)\,C\,\left(\tilde{A}_3^{\rm ortho}\right)^T+\left(\tilde{A}_4^{\rm ortho}\right)\,C\,\left(\tilde{A}_4^{\rm ortho}\right)^T } {4} \,.$$]{} Performing matrix operations, we obtain $$\widetilde C^{\rm ortho} =\left[\begin{array}{cccccc} c_{1111} & c_{1122} & c_{1133} & 0 & 0 & 0\\ c_{1122} & c_{2222} & c_{2233} & 0 & 0 & 0\\ c_{1133} & c_{2233} & c_{3333} & 0 & 0 &0\\ 0 & 0 & 0 & 2c_{2323} & 0 & 0\\ 0 & 0 & 0 & 0 & 2c_{1313} & 0\\ 0& 0 & 0 & 0 & 0 & 2c_{1212} \end{array}\right] \,, \label{eq:OrthoExplicitRef}$$ which exhibits the form of the orthotropic tensor in its natural coordinate system. In other words, $\widetilde{C}^{\rm ortho}=C^{\rm ortho}$, in accordance with Corollary \[col:Ortho\]. [^1]: Department of Earth Sciences, Memorial University of Newfoundland, [dalton.nfld@gmail.com]{} [^2]: Department of Earth Sciences, Memorial University of Newfoundland, [mslawins@mac.com]{} [^3]: Readers interested in formulation of matrix (\[eq:ATildeQ\]) might refer to Bóna et al. (2008).

ArXiv

--- abstract: 'We give a geometric construct of a modular functor for any simple Lie-algebra and any level by twisting the constructions in [@TUY] and [@Ue2] by a certain fractional power of the abelian theory first considered in [@KNTY] and further studied in [@AU1].' address: - | Department of Mathematics\ University of Aarhus\ DK-8000, Denmark - | Department of Mathematics\ Faculty of Science, Kyoto University\ Kyoto, 606-01 Japan author: - 'J[ø]{}rgen Ellegaard Andersen' - Kenji Ueno title: Geometric construction of modular functors from Conformal Field Theory --- [^1] Introduction ============ This is the second paper in a series of three papers ([@AU1] and [@AU3]) in which we provide a geometric construction of modular functors and topological quantum field theories from conformal field theory building on the constructions in [@TUY] and [@Ue2] and [@KNTY]. In this paper we provide the geometric construction of a modular functor $V^{{\mathop{\fam0 {\mathfrak g}}\nolimits}}_{\ell}$ for each simple Lie algebra ${\mathop{\fam0 {\mathfrak g}}\nolimits}$ and a positive integer $\ell$ (the [*level*]{}). In our third paper [@AU3] in this series we give an explicit isomorphism of the modular functor underlying the Reshetikhin-Turaev TQFT for $U_q({\mbox{\sf sl}}(n))$ and the one constructed in this paper for the Lie algebra ${\mbox{\sf sl}}(n)$. This uses the Skein theory approach to the Reshetikhin-Turaev TQFT of Blanchet, Habegger, Masbaum and Vogel [@BHMV1], [@BHMV2] and [@Bl1]. In particular we use Blanchet’s [@Bl1] constructions of the Hecke-category and its associated modular tensor categories. This construction is really a generation of the BHMV-construction of the $U_q(sl_2({\mathop{\fam0 {\mathbb C}^{}}\nolimits} ))$-Reshetikhin-Turaev TQFT [@BHMV2] to the $U_q(sl_n({\mathop{\fam0 {\mathbb C}^{}}\nolimits} ))$-case. As a consequence of this, we construct a duality and a unitary structure on our modular functor in the case of ${\mathop{\fam0 {\mathfrak g}}\nolimits}= {\mbox{\sf sl}}(n)$. By a very general construction any modular functor with duality induces a topological quantum field theory in dimension $2+1$ by the work of Kontsevich [@Kont1] and Walker [@Walker] and also Grove [@Grove]. By applying this to the modular functor $V^{{\mathop{\fam0 {\mathfrak g}}\nolimits}}_{\ell}$, for ${\mathop{\fam0 {\mathfrak g}}\nolimits}= {\mbox{\sf sl}}(n)$, we get a TQFT for each $\ell$. We also prove in [@AU3] that this TQFT is isomorphic to the Reshetikhin-Turaev TQFT for $U_q({\mbox{\sf sl}}(n))$ at level $\ell$. Let us now describe our construction. Fix a simple Lie algebra ${\mathop{\fam0 {\mathfrak g}}\nolimits}$ and normalize the invariant inner product on it by requiring the highest root to have length squared equal to $2$. Let $\ell$ be a positive integer and consider the finite (label) set $P_{\ell}$ of integrable highest weight representations at level $\ell$ of the affine Lie algebra of ${\mathop{\fam0 {\mathfrak g}}\nolimits}$. By the usual highest weight vector representations, this finite set $P_{\ell}$ is naturally identified with a subset of the dominant integrable weights of ${\mathop{\fam0 {\mathfrak g}}\nolimits}$ (see formula (\[labelset\])). The main idea is to construct a modular functor $V^{{\mathop{\fam0 {\mathfrak g}}\nolimits}}_{\ell}$ by associating to each labeled marked surface the [*space of vacua*]{} using the given labels for the Lie algebra ${\mathop{\fam0 {\mathfrak g}}\nolimits}$ at level $\ell$ for some complex structure on the marked surface, as defined in [@TUY] [@Ue] and [@Ue2]. In order to make this construction independent of the complex structure it must be understood in terms of bundles with connections over the hole of Teichmüller space of the surface, relying on parallel transport to provide the required identifications between the different spaces of vacua. The consistencies of these identifications translates to flatness requirements on these connections. However, the [*sheaf of vacua*]{} construction in [@TUY] and [@Ue2] gives a bundles with a connections, which is only projectively flat, over Teichmüller space of the surface. By tensoring this bundle with a line bundle with a connection with the opposite curvature, we get a flat bundle over Teichmüller space and the vector space we associate to the labeled marked surface is the vector space of covariant constant sections of this resulting bundle. This line bundle is constructed as a fractional power of a certain rank $1$ [abelian]{}sheaf of vacua, which we considered in the first paper in this series [@AU1] from the same point of view as [@TUY] and [@Ue2]. The extraction of this fractional power brings in central extensions of mapping classes as the natural morphisms on which the resulting functor is defined. The construction and properties of this flat bundle primarily rely on the complex algebraic constructions and results of [@TUY] and [@Ue2] on the the sheaf of vacua construction yielding a conformal field theory for each simple Lie algebra ${\mathop{\fam0 {\mathfrak g}}\nolimits}$ and level $\ell$. For the $1$-dimensional correction theory, we draw on the work [@AU1], which in turn relies on [@KNTY]. The definition of the functor $V^{{\mathop{\fam0 {\mathfrak g}}\nolimits}}_{\ell}$ requires only considerations of smooth families of Riemann surfaces (with some extra structure which is specified in section \[New1\]) over smooth complex manifolds. This is the setting for section \[New1\] through to section \[construction\]. However, in order to define the glueing isomorphism, which a modular functor is required to have, we need to discuss certain very simple families of stable curves, which contains so call nodal curves. These are described and considered in section \[shofvacandglue\] and in the Appendix to this paper. The paper is organized as follows. In section \[AxiomsMF\] we give the axioms for a modular functor. We introduce the notion of a [*marked*]{} surface, which is a closed smooth oriented surface, with a finite subset of points with projective tangent vectors and a Lagrangian subspace of the first integer homology of the surface. These form a category on which there is the operation of disjoint union and the operation of orientation reversal. There is also the process of glueing on this category. If we have a finite set, we can label the finite set of points on a marked surface by elements from this finite [*label*]{} set and get the category of labeled marked surfaces. A modular functor based on some finite label set, is a functor from the category of labeled marked surfaces to the category of finite dimensional complex vector spaces, which takes the disjoint union operation to the tensor product operation and which takes the glueing process to a certain direct sum construction, such that some compatibility holds, as described in details in Definition \[DefMF\]. A modular functor is said to be with duality if further the operation of orientation reversal is taken to the operation of taking the dual vector space. In sections \[New1\] to \[New5\] we describe in detail how any simple Lie algebra and a level $\ell$, via the sheaf of vacua constructions in [@Ue2] yields a holomorphic vector bundles with a projectively flat connection over Teichmüller spaces of pointed surfaces equipped with symplectic basis of the first homology. In sections \[New3ab\] to \[prefsec\] we describe in detail how the abelian sheaf of vacua constructions in [@AU1] yields a holomorphic line bundles with a projectively flat connection and a preferred non-vanishing section over Teichmüller spaces of pointed surfaces equipped with a symplectic basis of the first homology. In section \[construction\] we describe our global geometric construction of a modular functor for any simple Lie algebra and a level $\ell$. Theorem \[mainconstT\] and \[mainconstTab\] summarizes the constructions from sections \[New1\] to \[prefsec\]. The preferred section of the abelian theory allows us to construct a certain fractional power of this line bundle as stated in Theorem \[fracpowerab\], which we tensor onto this holomorphic vector bundle, so as to obtain a holomorphic vector bundle with a [*flat*]{} connection over Teichmüller space. The modular functor is then defined (Definition \[def.main\]) by taking covariant constant sections of this flat bundle. The section ends with the construction of the disjoint union isomorphism. The glueing isomorphism is constructed in section \[shofvacandglue\], where we also prove the needed properties of glueing. In section \[verification\] we establish all the axioms of a modular functor is satisfied based on the main results of the preceding sections. We have included an Appendix, which recalls the nessessary definitions regarding nodal curves, families of stable curves and the glueing construction. The axioms for a modular functor {#AxiomsMF} ================================ We shall in this section give the axioms for a modular functor. These are due to G. Segal and appeared first in [@Se]. We present them here in a topological form, which is due to K. Walker [@Walker]. See also [@Grove]. We note that similar, but different, axioms for a modular functor are given in [@Tu] and in [@BB]. It is however not clear if these definitions of a modular functor is equivalent to ours. Let us start by fixing a bit of notation. By a closed surface we mean a smooth real two dimensional manifold. For a closed oriented surface ${\Sigma}$ of genus $g$ we have the non-degenerate skew-symmetric intersection pairing $$(\cdot,\cdot) : H_1({\Sigma},{{\mathbb Z}}) \times H_1({\Sigma},{{\mathbb Z}}) {\mathop{\fam0 \rightarrow}\nolimits}{{\mathbb Z}}.$$ Suppose ${\Sigma}$ is connected. In this case a Lagrangian subspace $L\subset H_1({\Sigma},{{\mathbb Z}})$ is by definition a subspace, which is maximally isotropic with respect to the intersection pairing. - A ${{\mathbb Z}}$-basis $(\vec \alpha, \vec \beta) = (\alpha_1,\ldots, \alpha_g,\beta_1, \ldots \beta_g)$ for $H_1({\Sigma},{{\mathbb Z}})$ is called a symplectic basis if $$(\alpha_i,\beta_j) = \delta_{ij}, \quad (\alpha_i,\alpha_j) = (\beta_i,\beta_j) = 0,$$ for all $i,j = 1, \ldots, g$. If ${\Sigma}$ is not connected, then $H_1({\Sigma},{{\mathbb Z}}) = \oplus_i H_1({\Sigma}_i,{{\mathbb Z}})$, where ${\Sigma}_i$ are the connected components of ${\Sigma}$. By definition a Lagrangian subspace is in this paper a subspace of the form $L = \oplus_i L_i$, where $L_i\subset H_1({\Sigma}_i,{{\mathbb Z}})$ is Lagrangian. Likewise a symplectic basis for $H_1({\Sigma},{{\mathbb Z}})$ is a ${{\mathbb Z}}$-basis of the form $((\vec \alpha^i, \vec \beta^i))$, where $(\vec \alpha^i, \vec \beta^i)$ is a symplectic basis for $H_1({\Sigma}_i,{{\mathbb Z}})$. For any real vector space $V$, we define $PV = (V-\{0\})/{\mathop{\fam0 {\mathbb R}^{}}\nolimits}_+.$ \[DefPointS\] A [*pointed surface*]{} $({\Sigma},P)$ is an oriented closed surface ${\Sigma}$ with a finite set $P\subset {\Sigma}$ of points. A pointed surface is called [*[stable]{}*]{}if the Euler characteristic of each component of the complement of the points $P$ is negative. A pointed surface is called [*[saturated]{}*]{}if each component of ${\Sigma}$ contains at least one point from $P$. \[DefMorPointS\] A [*morphism of pointed surfaces*]{} $f :({\Sigma}_1,P_1) {\mathop{\fam0 \rightarrow}\nolimits}({\Sigma}_2,P_2)$ is an isotopy class of orientation preserving diffeomorphisms which maps $P_1$ to $P_2$. Here the isotopy is required not to change the induced map of the first order Jet at $P_1$ to the first order Jet at $P_2$. \[msurface\] A [*marked surface*]{} $ {{{\mathbf \Sigma}}} = ({\Sigma}, P, V, L)$ is an oriented closed smooth surface ${\Sigma}$ with a finite subset $P \subset {\Sigma}$ of points with projective tangent vectors $V\in \sqcup_{p \in P}PT_{p}{\Sigma}$ and a Lagrangian subspace $L \subset H_1({\Sigma},{{\mathbb Z}})$. [*The notions of [stable]{}and [saturated]{}marked surfaces are defined just like for pointed surfaces.* ]{} \[mmorphism\] A [*morphism*]{} ${\mathbf f} : {{{\mathbf \Sigma}}}_1 \to {{{\mathbf \Sigma}}}_2$ of marked surfaces ${{{\mathbf \Sigma}}}_i = ({\Sigma}_i,P_i,V_i,L_i)$ is an isotopy class of orientation preserving diffeomorphisms $f : {\Sigma}_1 \to {\Sigma}_2$ that maps $(P_1,V_1)$ to $(P_2,V_2)$ together with an integer $s$. Hence we write ${\mathbf f} = (f,s)$. [*Any marked surface has an underlying pointed surface, but a morphism of marked surfaces does not quit induce a morphism of pointed surfaces, since we only require that the isotopies preserve the induced maps on the projective tangent spaces.* ]{} Let $\sigma$ be Wall’s signature cocycle for triples of Lagrangian subspaces of $H_1({\Sigma},{\mathop{\fam0 {\mathbb R}^{}}\nolimits})$ (See [@Wall]). \[composition\] Let ${\mathbf f}_1 = (f_1,s_1) : {{{\mathbf \Sigma}}}_1 \to {{{\mathbf \Sigma}}}_2$ and ${\mathbf f}_2 = (f_2,s_2) : {{{\mathbf \Sigma}}}_2 \to {{{\mathbf \Sigma}}}_3$ be morphisms of marked surfaces ${{{\mathbf \Sigma}}}_i = ({\Sigma}_i,P_i,V_i,L_i)$ then the [*composition*]{} of ${\mathbf f}_1$ and ${\mathbf f}_2$ is $${\mathbf f}_2 {\mathbf f}_1 = (f_2 f_1, s_2 + s_1 - \sigma((f_2f_1)_*L_1, f_{2*}L_2,L_3)).$$ With the objects being marked surfaces and the morphism and their composition being defined as in the above definition, we have constructed the category of marked surfaces. The mapping class group $\Gamma({{{\mathbf \Sigma}}})$ of a marked surface ${{{\mathbf \Sigma}}} = ({\Sigma},L)$ is the group of automorphisms of ${{{\mathbf \Sigma}}}$. One can prove that $\Gamma({{{\mathbf \Sigma}}})$ is a central extension of the mapping class group $\Gamma({\Sigma})$ of the surface ${\Sigma}$ defined by the 2-cocycle $c : \Gamma({{{\mathbf \Sigma}}}) \to \mathbb Z$, $c(f_1,f_2) = \sigma((f_1f_2)_*L,f_{1*}L,L)$. One can also prove that this cocycle is equivalent to the cocycle obtained by considering two-framings on mapping cylinders (see [@At1] and [@A]). Notice also that for any morphism $(f,s) : {{\mathbf \Sigma}}_1 \to {{\mathbf \Sigma}}_2$, one can factor $$\begin{aligned} (f,s) &=& \left(({\mathop{\fam0 Id}\nolimits},s') : {{\mathbf \Sigma}}_2 \to {{\mathbf \Sigma}}_2\right) \circ (f,s-s')\\ &=& (f,s-s') \circ \left(({\mathop{\fam0 Id}\nolimits},s') : {{\mathbf \Sigma}}_1 \to {{\mathbf \Sigma}}_1\right).\end{aligned}$$ In particular $({\mathop{\fam0 Id}\nolimits},s) : {{{\mathbf \Sigma}}} \to {{{\mathbf \Sigma}}}$ is $({\mathop{\fam0 Id}\nolimits},1)^s$. \[disjunion\] The operation of [*disjoint union of marked surfaces*]{} is $$({\Sigma}_1,P_1,V_1,L_1) \sqcup ({\Sigma}_2,P_2,V_2,L_2) = ({\Sigma}_1 \sqcup {\Sigma}_2,P_1 \sqcup P_2,V_1\sqcup V_2,L_1 \oplus L_2).$$ Morphisms on disjoint unions are accordingly $(f_1,s_1) \sqcup (f_2,s_2) = (f_1 \sqcup f_2,s_1 + s_2)$. We see that disjoint union is an operation on the category of marked surfaces. \[or\] Let ${{{\mathbf \Sigma}}}$ be a marked surface. We denote by $- {{{\mathbf \Sigma}}}$ the marked surface obtained from ${{{\mathbf \Sigma}}}$ by the [*operation of reversal of the orientation*]{}. For a morphism ${\mathbf f} = (f,s) : {{{\mathbf \Sigma}}}_1 \to {{{\mathbf \Sigma}}}_2$ we let the orientation reversed morphism be given by $- {\mathbf f} = (f,-s) : -{{{\mathbf \Sigma}}}_1 \to -{{{\mathbf \Sigma}}}_2$. We also see that orientation reversal is an operation on the category of marked surfaces. Let us now consider glueing of marked surfaces. Let $({\Sigma}, \{p_-,p_+\}\sqcup P,\{v_-,v_+\}\sqcup V,L)$ be a marked surface, where we have selected an ordered pair of marked points with projective tangent vectors $((p_-,v_-),(p_+,v_+))$, at which we will perform the glueing. Let $c : P(T_{p_-}{\Sigma}) {\mathop{\fam0 \rightarrow}\nolimits}P(T_{p_+}{\Sigma})$ be an orientation reversing projective linear isomorphism such that $c(v_-) = v_+$. Such a $c$ is called a [*glueing map*]{} for ${\Sigma}$. Let $\tilde{{\Sigma}}$ be the oriented surface with boundary obtained from ${\Sigma}$ by blowing up $p_-$ and $p_+$, i.e. $$\tilde{{\Sigma}} = ({\Sigma}-\{p_-,p_+\})\sqcup P(T_{p_-}{\Sigma})\sqcup P(T_{p_+}{\Sigma}),$$ with the natural smooth structure induced from ${\Sigma}$. Let now ${\Sigma}_c$ be the closed oriented surface obtained from $\tilde{{\Sigma}}$ by using $c$ to glue the boundary components of $\tilde{{\Sigma}}$. We call ${\Sigma}_c$ the glueing of ${\Sigma}$ at the ordered pair $((p_-,v_-),(p_+,v_+))$ with respect to $c$. Let now ${\Sigma}'$ be the topological space obtained from ${\Sigma}$ by identifying $p_-$ and $p_+$. We then have natural continuous maps $q : {\Sigma}_c {\mathop{\fam0 \rightarrow}\nolimits}{\Sigma}'$ and $n : {\Sigma}{\mathop{\fam0 \rightarrow}\nolimits}{\Sigma}'$. On the first homology group $n$ induces an injection and $q$ a surjection, so we can define a Lagrangian subspace $L_c \subset H_1({\Sigma}_c,{{\mathbb Z}})$ by $L_c = q_*^{-1}(n_*(L))$. We note that the image of $P(T_{p_-}{\Sigma})$ (with the orientation induced from $\tilde{{\Sigma}}$) induces naturally an element in $H_1({\Sigma}_c,{{\mathbb Z}})$ and as such it is contained in $L_c$. [*\[remarkglue2\] If we have two glueing maps $c_i : P(T_{p_-}{\Sigma}) {\mathop{\fam0 \rightarrow}\nolimits}P(T_{p_+}{\Sigma}),$ $i=1,2,$ we note that there is a diffeomorphism $f$ of ${\Sigma}$ inducing the identity on $(p_-,v_-)\sqcup(p_+,v_+)\sqcup(P,V)$ which is isotopic to the identity among such maps, such that $(df_{p_+})^{-1} c_2 df_{p_-} = c_1$. In particular $f$ induces a diffeomorphism $f : {\Sigma}_{c_1} {\mathop{\fam0 \rightarrow}\nolimits}{\Sigma}_{c_2}$ compatible with $f : {\Sigma}{\mathop{\fam0 \rightarrow}\nolimits}{\Sigma}$, which maps $L_{c_1}$ to $L_{c_2}$. Any two such diffeomorphisms of ${\Sigma}$ induces isotopic diffeomorphisms from ${\Sigma}_1$ to ${\Sigma}_2$.*]{} \[glueing\] Let ${{{\mathbf \Sigma}}} = ({\Sigma}, \{p_-,p_+\}\sqcup P,\{v_-,v_+\}\sqcup V,L)$ be a marked surface. Let $$c : P(T_{p_-}{\Sigma}) {\mathop{\fam0 \rightarrow}\nolimits}P(T_{p_+}{\Sigma})$$ be a glueing map and ${\Sigma}_c$ the glueing of ${\Sigma}$ at the ordered pair $((p_-,v_-),(p_+,v_+))$ with respect to $c$. Let $L_c \subset H_1({\Sigma}_c,{{\mathbb Z}})$ be the Lagrangian subspace constructed above from $L$. Then the marked surface ${{{\mathbf \Sigma}}}_c = ({\Sigma}_c,P,V,L_c)$ is defined to be the [*glueing*]{} of ${{{\mathbf \Sigma}}}$ at the ordered pair $((p_-,v_-),(p_+,v_+))$ with respect to $c$. We observe that glueing also extends to morphisms of marked surfaces which preserves the ordered pair $((p_-,v_-),(p_+,v_+))$, by using glueing maps which are compatible with the morphism in question. We can now give the axioms for a 2 dimensional modular functor. \[DefLS\] A [*label set*]{} $\L$ is a finite set furnished with an involution $\l \mapsto \hat \l$ and a trivial element $1$ such that $\hat 1 = 1$. \[lmsurface\] Let $\L$ be a label set. The category of [*$\L$-labeled marked surfaces*]{} consists of marked surfaces with an element of $\L$ assigned to each of the marked point and morphisms of labeled marked surfaces are required to preserve the labelings. An assignment of elements of $\L$ to the marked points of ${{{\mathbf \Sigma}}}$ is called a labeling of ${{{\mathbf \Sigma}}}$ and we denote the labeled marked surface by $({{{\mathbf \Sigma}}},\l)$, where $\l$ is the labeling. We define a labeled pointed surface similarly. [*The operation of disjoint union clearly extends to labeled marked surfaces. When we extend the operation of orientation reversal to labeled marked surfaces, we also apply the involution $\hat \cdot$ to all the labels.* ]{} \[DefMF\] A [*modular functor*]{} based on the label set $\L$ is a functor $V$ from the category of labeled marked surfaces to the category of finite dimensional complex vector spaces satisfying the axioms MF1 to MF5 below. ### MF1 {#mf1 .unnumbered} [*Disjoint union axiom*]{}: The operation of disjoint union of labeled marked surfaces is taken to the operation of tensor product, i.e. for any pair of labeled marked surfaces there is an isomorphism $$V(({{{\mathbf \Sigma}}}_1,\l_1) \sqcup ({{{\mathbf \Sigma}}}_2,\l_2)) ) \cong V({{{\mathbf \Sigma}}}_1,\l_1) \otimes V({{{\mathbf \Sigma}}}_2,\l_2).$$ The identification is associative. ### MF2 {#mf2 .unnumbered} [*Glueing axiom*]{}: Let ${{{\mathbf \Sigma}}} $ and ${{{\mathbf \Sigma}}}_c$ be marked surfaces such that ${{{\mathbf \Sigma}}}_c$ is obtained from ${{{\mathbf \Sigma}}} $ by glueing at an ordered pair of points and projective tangent vectors with respect to a glueing map $c$. Then there is an isomorphism $$V({{{\mathbf \Sigma}}}_c,\lambda) \cong \bigoplus_{\m \in \L} V({{{\mathbf \Sigma}}},\m,\hat \m,\l),$$ which is associative, compatible with glueing of morphisms, disjoint unions and it is independent of the choice of the glueing map in the obvious way (see remark \[remarkglue2\]). ### MF3 {#mf3 .unnumbered} [*Empty surface axiom*]{}: Let $\emptyset$ denote the empty labeled marked surface. Then $$\dim V(\emptyset) = 1.$$ ### MF4 {#mf4 .unnumbered} [*Once punctured sphere axiom*]{}: Let ${{\mathbf \Sigma}}= (S^2, \{p\},\{v\},0)$ be a marked sphere with one marked point. Then $$\dim V({{\mathbf \Sigma}},\l) = \left\{ \begin{array}{ll} 1,\qquad &\l = 1\\ 0,\qquad & \l \ne 1.\end{array}\right.$$ ### MF5 {#mf5 .unnumbered} [*Twice punctured sphere axiom*]{}: Let ${{\mathbf \Sigma}}= (S^2, \{p_1,p_2\},\{v_1,v_2\},\{0\})$ be a marked sphere with two marked points. Then $$\dim V({{\mathbf \Sigma}},(\l,\mu)) = \left\{ \begin{array}{ll} 1, \qquad &\l = \hat \mu\\ 0,\qquad &\l \ne \hat \mu.\end{array}\right.$$ In addition to the above axioms one may has extra properties, namely ### MF-D {#mf-d .unnumbered} [*Orientation reversal axiom*]{}: The operation of orientation reversal of labeled marked surfaces is taken to the operation of taking the dual vector space, i.e for any labeled marked surface $({{{\mathbf \Sigma}}},\l)$ there is a pairing $$\langle \cdot,\cdot\rangle : V({{{\mathbf \Sigma}}},\l) \otimes V(-{{{\mathbf \Sigma}}},\hat \l) {\mathop{\fam0 \rightarrow}\nolimits}{\mathop{\fam0 {\mathbb C}^{}}\nolimits},$$ compatible with disjoint unions, glueings and orientation reversals (in the sense that the induced isomorphisms $ V({{{\mathbf \Sigma}}},\l) \cong V(-{{{\mathbf \Sigma}}},\hat \l)^*$ and $V(-{{{\mathbf \Sigma}}},\hat \l) \cong V({{{\mathbf \Sigma}}},\l)^*$ are adjoints). .4cm and ### MF-U {#mf-u .unnumbered} [*Unitarity axiom*]{} Every vector space $V({{{\mathbf \Sigma}}},\l)$ is furnished with a hermitian inner product $$( \cdot,\cdot ) : V({{{\mathbf \Sigma}}},\l) \otimes \overline{V({{{\mathbf \Sigma}}},\l)} \to {\mathbb C}$$ so that morphisms induces unitary transformation. The hermitian structure must be compatible with disjoint union and glueing. If we have the orientation reversal property, then compatibility with the unitary structure means that we have a commutative diagrams $$\begin{CD} V({{{\mathbf \Sigma}}},\l) @>>\cong> V(-{{{\mathbf \Sigma}}},\hat \l)^*\\ @VV\cong V @V\cong VV\\ \overline{V({{{\mathbf \Sigma}}},\l)^*} @>\cong>> \overline{V(-{{{\mathbf \Sigma}}},\hat \l)}, \end{CD}$$ where the vertical identifications come from the hermitian structure and the horizontal from the duality. The rest of the paper is concerned with the detailed geometric construction of modular functors using conformal field theory. However, we shall assume the reader is familiar with [@Ue2] and [@AU1] and freely use the notations of these two papers in this paper. Teichmüller space and families of pointed Riemann Surfaces with formal neighbourhoods {#New1} ===================================================================================== Let us first review some basic Teichmüller theory. Let ${\Sigma}$ be a closed oriented smooth surface and let $P$ be finite set of points on ${\Sigma}$. \[mc\] A [*marked Riemann surface*]{} ${\mathbf C}$ is a Riemann surface $C$ with a finite set of marked points $Q$ and non-zero tangent vectors $W \in T_{Q}C= \bigsqcup_{q\in Q} T_{q}C$. \[morphmc\] A [*morphism*]{} between marked Riemann surface is a biholomorphism of the underlying Riemann surface which induces a bijection between the two sets of marked points and tangent vectors at the marked points. The notions of [stable]{}and [saturated]{}is defined just like for pointed surfaces. \[cs\] A [*complex structure*]{} on $({\Sigma},P)$ is a marked Riemann surface ${\mathbf C} = (C, Q, W)$ together with an orientation preserving diffeomorphism $\phi : {\Sigma}{\mathop{\fam0 \rightarrow}\nolimits}C$ mapping the points $P$ onto the points $Q$. Two such complex structures $\phi_{j} : ({\Sigma},P) {\mathop{\fam0 \rightarrow}\nolimits}{\mathbf C}_{j} = (C_{j},Q_{j},W_{j})$ are [*equivalent*]{} if there exists a morphism of marked Riemann surfaces $$\Phi : {\mathbf C}_{1} {\mathop{\fam0 \rightarrow}\nolimits}{\mathbf C}_{2}$$ such that $\phi_{2}^{-1} \Phi \phi_{1} : ({\Sigma},P) {\mathop{\fam0 \rightarrow}\nolimits}({\Sigma},P)$ is isotopic to the identity through maps inducing the identity on the first order neighbourhood of $P$. We shall often in our notation suppress the diffeomorphism, when we denote a complex structure on a surface. \[Teichmsp\] The [*Teichmüller space*]{} ${\mathop{\fam0 {\mathcal T}}\nolimits}_{({\Sigma},P)}$ of the pointed surface $({\Sigma},P)$ is by definition the set of equivalence classes of complex structures on $({\Sigma},P)$. We note there is a natural projection map from ${\mathop{\fam0 {\mathcal T}}\nolimits}_{({\Sigma},P)}$ to $T_P{\Sigma}= \sqcup_{p\in P} T_{p}{\Sigma},$ which we call $\pi_P$. There is a natural structure of a finite dimensional complex analytic manifold on Teichmüller space ${\mathop{\fam0 {\mathcal T}}\nolimits}_{({\Sigma},P)}$. Associated to any morphism of pointed surfaces $f: ({\Sigma}_1,P_1) {\mathop{\fam0 \rightarrow}\nolimits}({\Sigma}_2,P_2)$ there is a biholomorphism $f^* : {\mathop{\fam0 {\mathcal T}}\nolimits}_{({\Sigma}_1,P_1)}{\mathop{\fam0 \rightarrow}\nolimits}{\mathop{\fam0 {\mathcal T}}\nolimits}_{({\Sigma}_2,P_2)}$ which is induced by mapping a complex structure ${\mathbf C} = (C, Q,W)$, $\phi : {\Sigma}_1 {\mathop{\fam0 \rightarrow}\nolimits}C$ to $\phi \circ f^{-1} : {\Sigma}_2 {\mathop{\fam0 \rightarrow}\nolimits}C$. Moreover, compositions of morphisms go to compositions of induced biholomorphisms. There is an action of ${\mathop{\fam0 {\mathbb R}_+^{P}}\nolimits}$ on ${\mathop{\fam0 {\mathcal T}}\nolimits}_{({\Sigma},P)}$ given by scaling the tangent vectors. This action is free and the quotient ${\mathop{\fam0 {\mathcal T}^{(r)}}\nolimits}_{({\Sigma},P)} = {\mathop{\fam0 {\mathcal T}}\nolimits}_{({\Sigma},P)}/{\mathop{\fam0 {\mathbb R}_+^{P}}\nolimits}$ is a smooth manifold, which we call the [*reduced*]{} Teichmüller space of the pointed surface $({\Sigma}, P)$. Moreover the projection map $\pi_P$ descend to a smooth projection map from ${\mathop{\fam0 {\mathcal T}^{(r)}}\nolimits}_{({\Sigma},P)}$ to $\sqcup_{p\in P} P(T_{p}{\Sigma})$, which we denote $\pi^{(r)}_P$. We denote the fiber of this map over $V\in \sqcup_{p\in P} P(T_{p}{\Sigma})$ by ${\mathop{\fam0 {\mathcal T}}\nolimits}_{({\Sigma},P,V)}$. Teichmüller space of a marked surface ${{\mathbf \Sigma}}= ({\Sigma}, P, V, L)$ is by definition ${\mathop{\fam0 {\mathcal T}}\nolimits}_{{{\mathbf \Sigma}}}={\mathop{\fam0 {\mathcal T}}\nolimits}_{({\Sigma},P,V)}$, which we call the Teichmüller space of the marked surface. Morphisms of marked surfaces induce diffeomorphism of the corresponding Teichmüller spaces of marked surfaces, which of course also behaves well under composition. We observe that the self-morphism $({\mathop{\fam0 Id}\nolimits}, s)$ of a marked surface acts trivially on the associated Teichmüller space for all integers $s$. General Teichmüller theory implies that \[contractT\] The Teichmüller space ${\mathop{\fam0 {\mathcal T}}\nolimits}_{{{\mathbf \Sigma}}}$ of any marked surface ${{\mathbf \Sigma}}$ is contractible. Now let us recall the definition of a formal neighbourhood of a point on a Riemann surface. \[curvwfn\] Let $C$ be a Riemann surface and $q$ a point on $C$. Let $\mathcal O_{C,q}$ be the stalk of $\mathcal O_C$ at $q$ and let $\mathfrak{m}_q$ the maximal ideal in $\mathcal O_{C,q}$. We note that $\mathfrak{m}_q^n$, $n=0,1,2, \ldots$, gives a filtration of $\mathcal O_{C,q}$. A formal $n$’th-order neighbourhood at $q$ is a filtration preserving isomorphism $$\mathcal O_{C,q}/\mathfrak{m}_q^{n+1} \cong \mathbb C [[ \xi ]] / (\xi^{n+1}).$$ Let $\hat{\mathcal O}_{C,q} = \lim_{n{\mathop{\fam0 \rightarrow}\nolimits}\infty}\mathcal O_{C,q}/\mathfrak{m}_q^n$ be the completion of $\mathcal O_{C,q}$ with respect to the filtration. A [*formal neighbourhood*]{} (or [*formal coordinate*]{}) at $q$ is a filtration preserving isomorphism $$\eta : \hat{\mathcal O}_{C,q}\cong \mathbb C[[\xi]].$$ We note that we have a canonical isomorphism $$\begin{aligned} {\Ocal}_{C,q}/{\mathfrak m}_q^{2}& \simeq & {\mathop{\fam0 {\mathbb C}^{}}\nolimits}\oplus T^*_{q}C,\\ f &\mapsto& (f(q), df_q).\end{aligned}$$ Hence a formal $1$’st order neighbourhood induces and is determined by an isomorphism of $T^*_q C$ with ${\mathop{\fam0 {\mathbb C}^{}}\nolimits}$. Hence a formal $1$’st order neighbourhood determines and is determined by a non-zero vector in $T^*_q C$, specified by the property that it maps to $1\in {\mathop{\fam0 {\mathbb C}^{}}\nolimits}$ or equivalently a vector in $T_q C$ pairing to unity with this vector. A pointed Riemann surface with formal neighbourhoods $${{{\mathfrak X}}}= (C; q_1,\ldots, q_N; \eta_1, \ldots, \eta_N)$$ is the following data: A Riemann surface $C$, an ordered $N$-tuple of $N$ distinct points $(q_1,\ldots, q_N)$ on $C$ together with formal neighbourhoods $$\eta_j : \hat{\mathcal O}_{C,q_j}\cong \mathbb C[[\xi_j]]$$ for $j=1,\ldots, N$. We remark that a pointed Riemann surface with formal neighbourhoods is an “$N$-pointed smooth curve with formal neighbourhoods” in the sense of Definition 1.1.3. of [@Ue2]. \[v1order\] For a pointed Riemann surface with formal neighbourhoods ${{{\mathfrak X}}}$, we denote by $c({{{\mathfrak X}}})$ the underlying marked Riemann surface. For a labeled pointed Riemann surface with formal neighbourhoods $({{{\mathfrak X}}}, \vec \l)$, we denote by $c({{{\mathfrak X}}},\vec \l) = (c({{{\mathfrak X}}}),\l)$ the underlying labeled marked Riemann surface. Here $\l$ denotes the labeling of the marked points of $c({{{\mathfrak X}}})$ induced by $\vec \l$. A family of pointed Riemann Surfaces with formal neighbourhoods $${{{\mathfrak F}}}= ( \pi \: \mathcal C \rightarrow \mathcal B; \vs; \veta )$$ is the following date: - Connected complex manifolds $\mathcal C$ and $\mathcal B$, such that ${\operatornamewithlimits{dim}}_{\mathbb C} C = {\operatornamewithlimits{dim}}_{\mathbb C} \mathcal B + 1$. - A holomorphic submersion $\pi : \mathcal C {\mathop{\fam0 \rightarrow}\nolimits}\mathcal B$. - Holomorphic sections $s_j$, $j = 1,\ldots, N$ of $\pi$. - Filtered $\mathcal O_{\mathcal B}$-algebra isomorphisms $$\eta_j : \widehat{\mathcal{O} }_{/s_j} = \varprojlim_{n \to \infty} \mathcal{O} _Y/I_{j}^{n} \simeq \mathcal{O} _{\mathcal B}[[\xi]],$$ where $I_{j}$ is the defining ideal of $s_j({\mathcal B})$ in ${\mathcal C}$, $j = 1, \ldots, N$. Note that a family of pointed Riemann Surfaces with formal neighbourhoods is a “Family of $N$-pointed smooth curves with formal neighbourhoods” as in Definition 1.2.1 in [@Ue2]. See also the Appendix at the end of this paper. Let ${\Sigma}$ be a closed oriented smooth surface and let $P$ be finite set of $N$ marked points on ${\Sigma}$, i.e. $({\Sigma},P)$ is a pointed surface. For a connected smooth complex manifold $\mathcal B$ let $Y = {\Sigma}\times \mathcal B$. Let ${{{\mathfrak F}}}= ( \pi \: \mathcal C \rightarrow \mathcal B; \vs; \veta )$ be a family of pointed Riemann surfaces with formal neighbourhoods and assume we have a smooth fiber preserving diffeomorphism $\Phi_{{{\mathfrak F}}}$ from $Y$ to $\mathcal C$ taking the marked points to the sections $\vs$ and inducing the identity on $\mathcal B$. This data induces a unique holomorphic map $\Psi_{{{\mathfrak F}}}$ from $\mathcal B$ to the Teichmüller space ${\mathop{\fam0 {\mathcal T}}\nolimits}_{({\Sigma},P)}$ of the surface $({\Sigma},P)$ by the universal property of Teichmüller space. \[famonsurf\] The pair $({{{\mathfrak F}}},\Phi_{{{\mathfrak F}}})$ is called a family of pointed Riemann surfaces with formal neighbourhoods [*on*]{} $({\Sigma},P)$. If $P'\subset P$ is a strict subset, we say that $({{{\mathfrak F}}},\Phi_{{{\mathfrak F}}})$ is called a family of pointed Riemann surfaces with formal neighbourhoods [*over*]{} $({\Sigma},P')$. Often we will suppress $\Phi_{{{\mathfrak F}}}$ in our notation and just write ${{{\mathfrak F}}}$ is a family of pointed Riemann surfaces with formal neighbourhoods on $({\Sigma},P)$. \[goodfamily\] If a family ${{{\mathfrak F}}}= ( \pi \: \mathcal C \rightarrow \mathcal B; \vs; \veta )$ of pointed Riemann surfaces with formal neighbourhoods on $({\Sigma},P)$, as above, has the properties, that the base $\mathcal B$ is biholomorphic to an open ball and that the induced map $\Psi_{{{\mathfrak F}}}$ is a biholomorphism onto an open subset of Teichmüller space ${\mathop{\fam0 {\mathcal T}}\nolimits}_{({\Sigma},P)}$ then the family is said to be [good.]{} Note that if a family of pointed Riemann surfaces with formal neighbourhoods on $({\Sigma},P)$ is [*versal*]{} around some point $b\in B$, in the sense of Definition 1.2.2 in [@Ue2], then there is a open ball around $b$ in $B$, such that the restriction of the family to this neighbourhood is good. \[coverT\] For a stable and saturated pointed surface $(\Sigma,P)$ the Teichmüller space ${\mathop{\fam0 {\mathcal T}}\nolimits}_{({\Sigma},P)}$ can be covered by images of such [good]{}families. This follows from Theorem 1.2.9 in [@Ue2]. Suppose now that we have two [stable]{}and [saturated]{}families ${{{\mathfrak F}}}_i$, $i=1,2$ with the property that they have the same image $\Psi_{{{{\mathfrak F}}}_1}({\mathcal B}_1) = \Psi_{{{{\mathfrak F}}}_2}({\mathcal B}_2)$ in Teichmüller space ${\mathop{\fam0 {\mathcal T}}\nolimits}_{({\Sigma},P)}$ and that ${{{\mathfrak F}}}_2$ is a [good]{}family. \[famequivalence\] For such a pair of families there exists a unique fiber preserving biholomorphism $\Phi : \mathcal C_1{\mathop{\fam0 \rightarrow}\nolimits}\mathcal C_2$ covering $\Psi^{-1}_{{{{\mathfrak F}}}_2}\Psi_{{{{\mathfrak F}}}_1}$ such that $\Phi^{-1}_{{{{\mathfrak F}}}_2} \Phi \Phi_{{{{\mathfrak F}}}_1} : (Y, P) {\mathop{\fam0 \rightarrow}\nolimits}(Y,P)$ is isotopic to $\Psi^{-1}_{{{{\mathfrak F}}}_2}\Psi_{{{{\mathfrak F}}}_1}\times {\mathop{\fam0 Id}\nolimits}$ through such fiber preserving maps inducing the identity on the first order neighbourhood of $P$. This follows from uniqueness of the $\Phi$ in Definition \[cs\]. We note that there is some permutation $S$ of $\{1,\ldots, N\}$ such that $(S \Phi^* (\veta_2))^{(1)} = (\veta_1)^{(1)}$, i.e. $S \Phi^* (\veta_2)$ induce the same first order formal neighbourhoods as $\veta_1$ does. Suppose now $f$ is an orientation preserving diffeomorphism from $({\Sigma}_1, P_1)$ to $({\Sigma}_2, P_2)$. Let ${{{\mathfrak F}}}_1$ be a family of pointed Riemann surfaces with formal neighbourhoods of $({\Sigma}_1, P_1)$. By composing $\Phi_{{{{\mathfrak F}}}_1}$ with $f^{-1}\times {\mathop{\fam0 Id}\nolimits}$ we get a family of pointed Riemann surfaces with formal neighbourhoods of $({\Sigma}_2, P_2)$. We note that $\Psi_{{{{\mathfrak F}}}_2}= f^* \circ \Psi_{{{{\mathfrak F}}}_1}$, where $f^*$ is the induced map between the Teichmüller spaces. This operation on families clearly behaves well under compositions of diffeomorphisms. The space of vacua associated to a labeled marked Riemann Surface ================================================================= Affine Lie algebras and integrable highest weight modules --------------------------------------------------------- In this section we recall the basic facts about integrable highest weight representations of affine Lie algebras. For the details of integrable highest weight representations of affine Lie algebras we refer the reader to Kac’s book \[Ka\]. Let ${\mathfrak g}$ be a simple Lie algebra over the complex numbers $\mathbb C$, which we fix throughout the paper. Let ${\mathfrak h}$ be its Cartan subalgebra. By $\Delta$ we denote the root system of $({\mathfrak g} , {\mathfrak h})$. We have the root space decomposition $${\mathfrak g} = {\mathfrak h} \oplus \displaystyle{\sum_{\alpha \in \Delta}} {\mathfrak g}_\alpha .$$ Let ${{\mathfrak h}}_{\mathbf R}^{*}$ be the linear span of $\Delta$ over $\mathbf R$. Fix a choice of positive roots $\Delta_+$. Let $(\phantom{X}, \phantom{X} )$ be a constant multiple of the Cartan-Killing form of the simple Lie algebra ${\mathfrak g}$. For each element of $\lambda \in {\mathfrak h}$, there exists a unique element $H_\lambda \in {\mathfrak h}^*$ such that $$\lambda(H) = ( H_\lambda, H)$$ for all $H \in {\mathfrak h}$. For $\alpha \in \Delta$, $H_\alpha$ is called the [*root vector*]{} corresponding to the root $\alpha$. On ${\mathfrak h}^*$ we introduce an inner product by $$(\lambda, \mu) = ( H_\lambda, H_\mu ). \label{2.1-1}$$ Let us normalize the inner product $(\phantom{X}, \phantom{X})$ by requiring that $\theta$, the highest (or longest) root, has length squared $$(\theta ,\, \theta) = 2. \label{2.1-2}$$ Let $V_\lambda$ be the irreducible left ${\mathfrak g}$-module of highest weight $\lambda$. It is well-known that a finite dimensional irreducible left ${\mathfrak g}$-module is a highest weight module and two irreducible left ${\mathfrak g}$-modules are isomorphic if and only if they have the same highest weight. A weight $\lambda \in {\mathfrak h}_{\mathbf R}^*$ is called an integral weight, if $$2(\lambda, \alpha)/(\alpha, \alpha) \in \mathbf Z$$ for any $\alpha \in \Delta$. A weight $\lambda \in {\mathfrak h}_{\mathbf R}^*$ is called a dominant weight, if $$w(\lambda) \leq \lambda$$ for any element $w$ of the Weyl group $W$ of ${\mathfrak g}$. By $P_+$ we denote the set of dominant integral weights of ${\mathfrak g}$. A weight $\lambda$ is the highest weight of an irreducible left ${\mathfrak g}$-module if and only if $\lambda \in P_+$. Let $w$ be longest element of $W$. Then we define an involution $\dagger$ on $P_+$ by $$\label{involution} \lambda^\dagger = - w(\lambda).$$ One has that the opposite of dual of the left-${\mathfrak g}$-module $V_\lambda$ is isomorphic to left-${\mathfrak g}$-module $V_{\lambda^\dagger}$, meaning there exists a non-degenerate ${\mathfrak g}$-invariant perfect pairing $$(\phantom{X}, \phantom{X}) : V_\lambda \otimes V_{\lambda^\dagger} {\mathop{\fam0 \rightarrow}\nolimits}{\mathbb C}.$$ As mentioned in the introduction, we will need to fix $|0\rangle\in V_0 \setminus \{0\}$ and $$|0_{\lambda,\lambda^\dagger}\rangle\in (V_\lambda\otimes V_{\lambda^\dagger})^{{\mathfrak g}} \setminus \{0\},$$ where we put $|0_{\lambda \lambda^\dagger }\rangle= |0\rangle \otimes |0\rangle$ for $\lambda=0$ . Fixing such a vector is of course equivalent to fixing the above mentioned pairing. By $\mathbf C[[ \xi ]]$ and $\mathbf C((\xi))$ we mean the ring of formal power series in $\xi$ and the field of formal Laurent power series in $\xi$, respectively. \[D2.1.2\] [The affine Lie algebra $\widehat{{\mathfrak g}}$ over $\mathbf C((\xi))$ associated with ${\mathfrak g}$ is defined to be $$\widehat{{\mathfrak g}} = {\mathfrak g} \otimes {\mathbf C((\xi))} \oplus \mathbf C c$$ where $c$ is an element of the center of $\widehat{{\mathfrak g}}$ and the Lie algebra structure is given by $$\begin{aligned} [X\otimes f(\xi), \, Y\otimes g(\xi)] = [X, Y]\otimes f(\xi)g(\xi) + c\cdot (X,Y) {\operatornamewithlimits{Res}}_{\xi = 0} (g(\xi)df(\xi))\end{aligned}$$ for]{} $$X, \, Y \in {\mathfrak g}, \, f(\xi),\, g(\xi) \in \mathbf C((\xi)).$$ Put $$\widehat{{\mathfrak g}}_+ = {\mathfrak g} \otimes \mathbf C[[\xi]]\xi, \quad \widehat{{\mathfrak g}}_- ={\mathfrak g} \otimes \mathbf C[\xi^{-1}]\xi^{-1}. \label{2.1-5}$$ We regard $\widehat{{\mathfrak g}}_+$ and $\widehat{{\mathfrak g}}_-$ as Lie subalgebras of $\widehat{{\mathfrak g}}$. We have a decomposition $$\widehat{{\mathfrak g}} = \widehat{{\mathfrak g}}_+ \oplus {\mathfrak g} \oplus \mathbf C c \oplus \widehat{{\mathfrak g}}_-. \label{2.1-6}$$ Let us fix a positive integer $\ell$ (called the [*level*]{}) and put $$\label{labelset} P_\ell = \{\, \lambda \in P_+ \, | \, 0\le(\theta, \lambda) \le\ell \, \}.$$ For all levels $\ell$ we observe that $\dagger$ takes $P_\ell$ to it self. For each element $\lambda \in P_\ell$ we shall define the Verma module $\mathcal M_\lambda$ as follows. Put $$\widehat{{\mathfrak p}}_+ := \widehat{{\mathfrak g}}_+ \oplus {\mathfrak g} \oplus \mathbf C \cdot c.$$ Then $\widehat{{\mathfrak p}}_+$ is a Lie subalgebra of $\widehat{{\mathfrak g}}$. Let $V_\lambda$ is the irreducible left ${\mathfrak g}$-module of highest weight $\lambda$. The action of $\widehat{{\mathfrak p}}_+$ on $V_\lambda$ is defined as $$\begin{aligned} c v & =& \ell v \quad \text{for all $v \in V_\lambda$}\\ a v &= &0 \quad \text{for all $a \in \widehat{{\mathfrak g}}_+$ and $v \in V_\lambda$}\end{aligned}$$ Put $$\mathcal M_\lambda := U( \widehat{{\mathfrak g}}) \otimes_{\widehat{ {\mathfrak p}}_+} V_\lambda. \label{2.1-14}$$ Then $\mathcal M_\lambda$ is a left $\widehat{{\mathfrak g}}$-module and is called a [*Verma module*]{}. The Verma module $\mathcal M_\lambda$ is not irreducible and contains the maximal proper submodule $\mathcal J_\lambda$. The quotient module $\mathcal H_\lambda := \mathcal M_\lambda/\mathcal J_\lambda$ has the following properties. \[T2.1.4\] For each $\lambda \in P_\ell$, the left $\widehat{{\mathfrak g}}$-module ${\mathcal H}_\lambda$ is the unique left $\widehat{{\mathfrak g}}$-module [(]{}called the [integrable highest weight $\widehat{{\mathfrak g}}$-module)]{} satisfying the following properties. - $V_\lambda = \{ \, |v\rangle \in {\mathcal H}_\lambda \, | \; \widehat{{\mathfrak g}}_+ |v\rangle = 0 \, \}$ is the irreducible left ${\mathfrak g}$-module with highest weight $\lambda$. - The central element $c$ acts on $\mathcal H_\lambda$ as $\ell \cdot \hbox{\rm id}$. - $\mathcal H_\lambda$ is generated by $V_\lambda$ over $\widehat{{\mathfrak g}} _-$ with only one relation $$(X_\theta \otimes \xi^{-1} ) ^{\ell - (\theta, \lambda) + 1} | \lambda\rangle = 0 \label{2.1-15}$$ where $X_\theta \in {\mathfrak g}$ is the element corresponding to the maximal root $\theta$ and $|\lambda\rangle \in V_\lambda$ is a highest weight vector. The theorem says that the maximal proper submodule $\mathcal J_\lambda$ is given by $$\mathcal J_\lambda = U(\widehat{{\mathfrak p}}_- )|J_\lambda \rangle \label{2.1-16}$$ where we put $$|J_\lambda \rangle = (X_\theta \otimes \xi ^{-1} ) ^{\ell - (\theta, \lambda) + 1} | \lambda\rangle . \label{2.1-17}$$ For the details see (10.4.6) in [@Ka]. Similarly we have the integrable lowest weight right $\widehat{{\mathfrak g}}$-module ${\mathcal H}_{\lambda}^\dagger$ which will be discussed below. The Segal-Sugawara construction ------------------------------- We use the following notation $$\begin{aligned} X(n) & = & X \otimes \xi^n, \quad X \in {\mathfrak g} \\ X(z) & = &\sum_{n \in {\mathbf Z}} X(n) z^{-n-1}\end{aligned}$$ where $z$ is a variable. The normal ordering ${\,\lower.8ex \hbox{$\circ$} \llap{\raise.8ex\hbox{$\circ$}} \,}\phantom{X}{\,\lower.8ex \hbox{$\circ$} \llap{\raise.8ex\hbox{$\circ$}} \,}$ is defined by $${\,\lower.8ex \hbox{$\circ$} \llap{\raise.8ex\hbox{$\circ$}} \,}X(n) Y(m){\,\lower.8ex \hbox{$\circ$} \llap{\raise.8ex\hbox{$\circ$}} \,}= \begin{cases} X(n) Y(m), &n<m, \\ \frac{1}{2}(X(n)Y(m) + Y(m)X(n)) & n=m, \\ Y(m) X(n) & n > m. \end{cases}$$ Note that, if $n>m$ and $X=Y$, we have $${\,\lower.8ex \hbox{$\circ$} \llap{\raise.8ex\hbox{$\circ$}} \,}X(n) X(m) {\,\lower.8ex \hbox{$\circ$} \llap{\raise.8ex\hbox{$\circ$}} \,}= X(n) X(m) - n \delta_{n+m,0}(X,X) \cdot c. \label{2.2-1}$$ \[EM\] The [*energy-momentum*]{} tensor $T(z)$ of level $\ell$ is defined by $$T(z) = \frac{1}{2(g^{*} + \ell)} \sum_{a=1}^{\dim {\mathfrak g}} {\,\lower.8ex \hbox{$\circ$} \llap{\raise.8ex\hbox{$\circ$}} \,}J^a(z)J^a(z){\,\lower.8ex \hbox{$\circ$} \llap{\raise.8ex\hbox{$\circ$}} \,}$$ where $\{J^1,J^2,\ldots, J^{\dim {\mathfrak g}}\}$ is an orthonormal basis of ${\mathfrak g}$ with respect to the Cartan-Killing form $(\phantom{X},\phantom{X})$ and $g^{*}$ is the dual Coxeter number of ${\mathfrak g}$. Put $$L_n = \frac{1}{2(g^{*} + \ell)}\displaystyle{\sum_{m \in {\mathbf Z}} \sum_{a=1}^{\dim {\mathfrak g}}} {\,\lower.8ex \hbox{$\circ$} \llap{\raise.8ex\hbox{$\circ$}} \,}J^a(m) J^a(n-m){\,\lower.8ex \hbox{$\circ$} \llap{\raise.8ex\hbox{$\circ$}} \,}. \label{2.2-2}$$ Then we have the expansion $$T(z) = \sum_{n \in {\mathbf Z}}L_n z^{-n-2}.$$ The operator $L_n$ is called the $n$’th Virasoro operator and it acts on $\mathcal H_\lambda $. For $X \in {\mathfrak g}$, $f= f(z) \in {{\mathbf C}}(( z))$ and $\underline{\ell} = \ell(z)\displaystyle{\frac{d}{dz}}\in {{\mathbf C}}(( z ))\displaystyle{\frac{d}{dz}}$ we use the following notation. $$\begin{aligned} X[f] &= &{\operatornamewithlimits{Res}}_{z=0}(X(z)f(z)dz) \\ T[\underline{\ell}] &=& {\operatornamewithlimits{Res}}_{z=0}(T(z)\ell(z)dz).\end{aligned}$$ In particular, we have that $$L_0 = T[\xi \dfrac d{d \xi }]. \label{2.2-6}$$ To define a filtration $\{F_{\bullet}\}$ on $\mathcal H_\lambda $, we first define the subspace ${\mathcal H_\lambda }(d)$ of $\mathcal H_\lambda $ for a non-negative integer $d$ by $$\mathcal H_\lambda (d) = \{ \, |v\rangle \in \mathcal H_\lambda \,|\, \;\; L_0 |v\rangle = ( d + \Delta_\lambda)|v\rangle \, \} \label{2.2-8}$$ where $$\label{2.2.8a} \Delta_\lambda = \frac{(\lambda, \lambda) + 2(\lambda, \rho)} {2(g^{*} + \ell)}, \quad \rho = \frac{1}{2} \sum_{\alpha \in \Delta_+}\alpha.$$ The subspaces $\mathcal H_\lambda (d)$ are finite dimensional vector space and one has that $$\mathcal H_\lambda = \bigoplus_{d=0}^\infty \mathcal H_\lambda (d).$$ Now we define the filtration $\{ F_p\mathcal H_\lambda \}$ by $$F_p\mathcal H_\lambda = \sum_{d=0}^p \mathcal H_\lambda (d). \label{2.2-9}$$ Put $${\mathcal H}_\lambda^{\dag} (d) = {{\hbox{\rm Hom}}}_{\mathbf C}(\mathcal H_\lambda (d), {{\mathbf C}}). \label{2.2-10}$$ Then the dual space ${\mathcal H}_\lambda^{\dag}$ of ${\mathcal H}_\lambda$ is defined to be $${\mathcal H}_\lambda^{\dag} = {{\hbox{\rm Hom}}}_{{\mathbf C}}(\mathcal H_\lambda , {{\mathbf C}}) = \prod _{d=0}^\infty {\mathcal H}_\lambda^{\dag} (d) . \label{2.2-11}$$ By our definition ${\mathcal H}_\lambda^{\dag}$ is a right $\widehat{{{{\mathfrak g}}}}$-module. A decreasing filtration $\{F^p{\mathcal H}_\lambda^{\dag}\}$ is defined by $$F^p{\mathcal H}_\lambda^{\dag} = \prod_{d \ge p} {\mathcal H}_\lambda^{\dag} (d) . \label{2.2-12}$$ There is a unique canonical perfect bilinear pairing $$\langle \phantom{X}|\phantom{X}\rangle : {\mathcal H}_\lambda^{\dag} \times \mathcal H_\lambda \longrightarrow {{\mathbf C}}, \label{2.2-13}$$ given on $V_\lambda^\dagger \otimes V_\lambda$ by evaluation and which satisfies the following equality for each $a \in \widehat{{{{\mathfrak g}}}}$. $$\langle u|av\rangle = \langle ua|v\rangle, \quad \text{for all } \langle u| \in {\mathcal H}_\lambda^{\dag} \; \text{and } |v\rangle \in \mathcal H_\lambda \, .$$ Put $$V_\lambda^{\dag} = \{\,\langle v| \in {\mathcal H}_\lambda^{\dag}\, |\, \;\; \langle v|\widehat{{{{\mathfrak g}}}}_- = 0 \; \}.$$ It is easy to show that $V_\lambda^{\dag} = {\mathcal H}_\lambda^{\dag} (0)$ and $V_\lambda^{\dag}$ is the irreducible right ${\mathfrak g}$-module with lowest weight $\lambda$. The integrable highest weight right $\widehat{{{{\mathfrak g}}}}$-module with lowest weight $\lambda$ is generated by $V_\lambda^{\dag}$ over $\widehat{{{{\mathfrak g}}}}_+$ with only one relation $$\langle \lambda|(X_{-\theta} \otimes \xi)^{\ell -(\theta,\lambda) +1} = 0.$$ Now let us introduce the left ${\mathfrak g}$-module structure on $\mathcal H_\lambda^\dagger$ by $$X(n)\langle \Phi| := - \langle \Phi|X(-n) .$$ It is easy to check that this indeed defines the left ${\mathfrak g}$-module structure on $\mathcal H_\lambda^\dagger$. Now we give the relationship of the [*left*]{} ${\mathfrak g}$-module $\mathcal H_\lambda^\dagger$ and $\mathcal H_{\lambda^\dagger}$. \[L2.2.12\] There exists a unique canonical bilinear pairing $$(\phantom{X}|\phantom{X}) : \mathcal H_\lambda \times {\mathcal H}_{\lambda^\dagger} \rightarrow \mathbf C$$ such that we have $$(X(n)u|v) + (u|X(-n)v) = 0$$ for any $X \in {{{\mathfrak g}}}$, $n \in {\mathbf Z}$, $| u \rangle \in \mathcal H_\lambda$, $| v \rangle \in \mathcal H_{\lambda^\dagger}$, the pairing is zero on $\mathcal H_\lambda(d) \times \mathcal H_{\lambda^\dagger}(d')$, if $d \neq d'$ and it evaluates to $1$ on $|0_{\lambda,\lambda^\dagger}\rangle$. \[C2.2.13\] This pairing induces a canonical left ${\mathfrak g}$-module isomorphism $$\mathcal H_\lambda^\dagger \simeq \widehat{\mathcal H}_{\lambda^\dagger}$$ where $\widehat{\mathcal H}_{\lambda^\dagger}$ is the completion of ${\mathcal H}_{\lambda^\dagger}$ with respect to the filtration $\{F_p\}$. The space of vacua ------------------ \[D3.1.1\] The Lie algebra $\widehat{{\mathfrak g}}_N$ is defined as $$\widehat{{\mathfrak g}}_N = \bigoplus_{j=1}^N {\mathfrak g} \otimes_{{\mathbf C}}{{\mathbf C}}((\xi_j)) \oplus {{\mathbf C}}c$$ with the following commutation relations. $$[(X_j \otimes f_j), (Y_j \otimes g_j)] = ([X_j, Y_j] \otimes f_j g_j) + c \sum_{j=1}^N (X_j,Y_j) {\operatornamewithlimits{Res}}_{\xi_j = 0}(g_j df_j) \label{3.1-1}$$ where $(a_j)$ means $(a_1, a_2, \ldots, a_N)$ and $c$ belongs to the center of $\widehat{{\mathfrak g}}_N$. Let $\mathfrak X = (C; q_1,q_2, \ldots, q_N; \eta_1, \eta_2, \ldots, \eta_N)$ be a pointed saturated Riemann Surface with formal neighbourhoods and define $$\widehat{{\mathfrak g}}({\mathfrak X}) = {\mathfrak g} \otimes_{{\mathbf C}}H^0(C,\mathcal O_C (*{\sum_{j=1}^N}q_j)). \label{3.1-2}$$ We have the natural embedding $$t = \oplus t_i : H^0(C,\mathcal O_C (*{\sum_{j=1}^N}q_j)) \hookrightarrow \bigoplus_{j=1}^N{{\mathbf C}}((\xi_j))$$ given by Laurent expansion using the formal neighbourhoods. In the following we often regard $H^0(C,\mathcal O_C (*{\sum_{j=1}^N}q_j))$ as a subspace of $\displaystyle{\bigoplus_{j=1}^N}{{\mathbf C}}((\xi_j))$. One has by lemma 1.1.15 in [@Ue2], that $\widehat{{\mathfrak g}}({\mathfrak X})$ is a Lie subalgebra of $\widehat{{\mathfrak g}}_N$. Let us fix a non-negative integer $\ell$. For each $\vec \lambda = (\lambda_1, \ldots, \lambda_N) \in (P_\ell)^N$, the left $\widehat{{\mathfrak g}}_N$-module ${{{\Cal H}_{\vec \lambda}}}$ and a right $\widehat{{\mathfrak g}}_N$-module ${{\Cal H}_{\vec \lambda}^{\dagger}}$ are defined by $$\begin{aligned} {{{\Cal H}_{\vec \lambda}}}& = & \mathcal H_{\lambda_1} \otimes_{{\mathbf C}}\cdots \otimes_{{\mathbf C}}\mathcal H_{\lambda_N} \\ {{\Cal H}_{\vec \lambda}^{\dagger}}& = &{\mathcal H}_{\lambda_1}^\dagger \widehat{\otimes}_{{\mathbf C}}\cdots \widehat{\otimes}_{{\mathbf C}}{\mathcal H}_{\lambda_N}^\dagger. \end{aligned}$$ The hats over the tensor product means that the algebraic tensor product has been completed with respect to the induced filtration. For each element $X_j \in {\mathfrak g}$, $f(\xi_j) \in {{\mathbf C}}((\xi_j))$, the action $\rho_j$ of $X_j[f_j]$ on ${{{\Cal H}_{\vec \lambda}}}$ is given by $$\rho_j(X_j[f_j])|v_1 \otimes \cdots \otimes v_N\rangle = |v_1 \otimes \cdots \otimes v_{j-1} \otimes (X_j[f_j])v_j \otimes v_{j+1} \otimes \cdots v_N\rangle \label{3.1-3}$$ where $|v_1 \otimes \cdots \otimes v_N\rangle$ means $|v_1\rangle \otimes \cdots \otimes |v_N\rangle$, $|v_j\rangle \in \mathcal H_{\lambda_j}$. The left $\widehat{{\mathfrak g}}_N$-action is given by $$(X_1\otimes f_1, \ldots, X_N \otimes f_N)|v_1\otimes \cdots v_N\rangle = {\sum_{j=1}^N}\rho_j(X_j[f_j])|v_1 \otimes \cdots v_N\rangle. \label{3.1-4}$$ Similarly, the right $\widehat{{\mathfrak g}}_N$-action on ${{\Cal H}_{\vec \lambda}^{\dagger}}$ is defined by $$\langle u_1 \otimes \cdots u_N| (X_1 \otimes f_1, \ldots, X_N \otimes f_N) = {\sum_{j=1}^N}\langle u_1 \otimes \cdots u_N|\rho_j(X_j[f_j]). \label{3.1-5}$$ As a Lie subalgebra, $\widehat{{\mathfrak g}}({\mathfrak X})$ operates on ${{{\Cal H}_{\vec \lambda}}}$ and ${{\Cal H}_{\vec \lambda}^{\dagger}}$ as $$(X \otimes f)| v_1 \otimes \cdots \otimes v_N\rangle = {\sum_{j=1}^N}\rho_j(X \otimes t_j(f))|v_1 \otimes \cdots v_N\rangle \label{3.1-6}$$ and as $$\langle u_1 \otimes \cdots \otimes u_N| (X \otimes f) = {\sum_{j=1}^N}\langle u_1 \otimes \cdots \otimes u_N| \rho_j(X \otimes t_j(f)). \label{3.1-7}$$ The pairing $\langle \phantom{X}|\phantom{X}\rangle$ introduced in \[2.2-11\] induces a perfect bilinear pairing $$\langle \phantom{X}|\phantom{X}\rangle : {{\Cal H}_{\vec \lambda}^{\dagger}}\times {{{\Cal H}_{\vec \lambda}}}\rightarrow {{\mathbf C}}\label{3.1-8}$$ given by $$(\langle u_1 \otimes \ldots \otimes u_N|,|v_1\otimes \ldots \otimes u_N \rangle) \rightarrow \langle u_1|v_1\rangle\langle u_2|v_2\rangle \cdots \langle u_N|v_N\rangle$$ which is $\widehat{{\mathfrak g}}_N$-invariant: $$\langle\Psi(X_j\otimes f_j)|\Phi\rangle = \langle\Psi |(X_j\otimes f_j) \Phi\rangle.$$ Now we are ready to define the space of vacua attached to ${\mathfrak X}$. \[D3.1.3\] Assume that ${\mathfrak X}$ is saturated. Put $${{{\Cal V}_{\vec \lambda}}}({\mathfrak X}) = {{{\Cal H}_{\vec \lambda}}}/ \widehat{{\mathfrak g}}({\mathfrak X}){{{\Cal H}_{\vec \lambda}}}. \label{3.1-9}$$ The vector space ${{{\Cal V}_{\vec \lambda}}}({\mathfrak X})$ is called the [*space of covacua*]{} attached to ${\mathfrak X}$. The [*space of vacua*]{} attached to ${\mathfrak X}$ is defined as $${{\Cal V}_{\vec \lambda}^{\dagger}}({\mathfrak X}) = {{\hbox{\rm Hom}}}_{{\mathbf C}}({{{\Cal V}_{\vec \lambda}}}({\mathfrak X}), {{\mathbf C}}). \label{3.1-10}$$ One gets that $${{\Cal V}_{\vec \lambda}^{\dagger}}({\mathfrak X}) = \{ \; \langle \Psi | \in {{\Cal H}_{\vec \lambda}^{\dagger}}\;| \langle \Psi|\;\widehat {{{\mathfrak g}}}({{{\mathfrak X}}}) = 0 \;\}. \label{3.1-11}$$ Moreover, the pairing (\[3.1-8\]) induces a perfect pairing $$\langle\phantom{X}|\phantom{X}\rangle : {{\Cal V}_{\vec \lambda}^{\dagger}}({{{\mathfrak X}}}) \times {{{\Cal V}_{\vec \lambda}}}({{{\mathfrak X}}}) \rightarrow {{\mathbf C}}. \label{3.1-12}$$ The following theorem is proved in [@Ue2]. \[T3.1.5\] The vector spaces ${{{\Cal V}_{\vec \lambda}}}({{{\mathfrak X}}})$ and ${{\Cal V}_{\vec \lambda}^{\dagger}}({{{\mathfrak X}}})$ are finite-dimensional. Propagation of vacua -------------------- For a pointed Riemann Surface with formal neighbourhoods ${\mathfrak X} = (C;q_1,\ldots, q_N; \eta_1,\ldots, \eta_N)$ let $q_{N+1}$ be a point on $C \setminus \{q_1,\ldots, q_N\}$ and $\eta_{N+1}$ a formal neighbourhood of $C$ at $q_{N+1}$. Put $${\widetilde {{\mathfrak X}}} = (C; q_1, \ldots , q_N, q_{N+1}; \eta_1, \ldots , \eta_N, \eta_{N+1}).$$ Since there is a canonical inclusion $$\begin{aligned} {{{\Cal H}_{\vec \lambda}}}&\longrightarrow &{{{\Cal H}_{\vec \lambda}}}\otimes {\mathcal H}_0 \\ |v \rangle &\longrightarrow & |v \rangle \otimes |0 \rangle\end{aligned}$$ we have a canonical surjection $$\widehat{\iota}^* \; : \; {{\Cal H}_{\vec \lambda}^{\dagger}}\widehat{\otimes} {\mathcal H}_0^\dagger \longrightarrow {{\Cal H}_{\vec \lambda}^{\dagger}}\; .$$ \[T3.2.1\] The canonical surjection $\widehat{\iota}^*$ induces a canonical [*Propagation of vacua*]{} isomorphism $$P_{\widetilde{{\mathfrak X}},{\mathfrak X}} : \mathcal V_{{{\vec \lambda}},0}^\dagger(\widetilde{{\mathfrak X}}) {\mathop{\fam0 \rightarrow}\nolimits}\mathcal V_{{{\vec \lambda}}}^\dagger({\mathfrak X}).$$ Change of formal neighbourhoods ------------------------------- We let ${\mathcal D}$ be the automorphism group ${\mathop{\fam0 Aut}\nolimits}{\mathop{\fam0 {\mathbb C}^{}}\nolimits}((\xi))$ of the field ${\mathop{\fam0 {\mathbb C}^{}}\nolimits}((\xi))$ of formal Laurent series as a ${\mathop{\fam0 {\mathbb C}^{}}\nolimits}$-algebra. There is a natural isomorphism $$\begin{aligned} {\mathcal D} & \simeq & \{ \; \sum_{n=0}^\infty a_n \xi^{n+1} \; | \; a_0 \ne 0\; \}\\ h & \mapsto & h(\xi)\end{aligned}$$ where the composition $h \circ g$ of $h$, $g \in {\mathcal D}$ corresponds to the formal power series $h(g(\xi))$. Put $${\mathcal D}^p = \{ \;h \in {\mathcal D}\; | \; h(\xi) = \xi + a_p\xi^{p+1} + \cdots\;\}$$ for a positive integer $p$. Then we have a filtration $${\mathcal D}= {\mathcal D}^0 \supset {\mathcal D}^1 \supset {\mathcal D}^2 \supset \ldots$$ Also let $$\begin{aligned} \underline{d} &=& {\mathop{\fam0 {\mathbb C}^{}}\nolimits}[[\xi]]\xi\frac{d}{d\xi} \\ \underline{d}^p &=& {\mathop{\fam0 {\mathbb C}^{}}\nolimits}[[\xi]]\xi^{p+1}\frac{d}{d\xi} \quad p=0,1,2,\ldots\end{aligned}$$ Then, we have a filtration $$\underline{d} = \underline{d}^0 \supset \underline{d}^1 \supset \underline{d}^2 \supset \cdots$$ For any $\underline{l} \in \underline{d}$ and $f(\xi) \in{\mathop{\fam0 {\mathbb C}^{}}\nolimits}[[\xi]]$ define $\exp (\underline{l})(f(\xi))$ by $$\exp (\underline{l})(f(\xi)) = \sum_{k=0}^\infty \frac{1}{k!} (\underline{l}^k f(\xi)).$$ Set $$\begin{aligned} {\mathcal D}_+^0 &=& \{ \;h \in {\mathcal D}\; | \; h(\xi) = a\xi + a_1\xi^{2} + \cdots, \quad a>0 \;\} \\ \underline{d}_+^0 &=& \{ \; l(\xi)\frac{d}{d\xi}\;|\; l(\xi) = \alpha \xi + \alpha_1 \xi^2 +\cdots, \quad \alpha \in {\mathop{\fam0 {\mathbb R}^{}}\nolimits}\;\}\end{aligned}$$ Then, we have the following result. \[lem6.1\] The exponential map $$\begin{aligned} \exp \; : \; \underline{d} & \rightarrow & {\mathcal D} \\ \phantom{\exp \; : \;{}} \underline{l} & \mapsto & \exp(\underline{l}) \end{aligned}$$ \[exponential\] is surjective. Moreover, the exponential map induces an isomorphism $$\exp \; : \; \underline{d}_+^0 \simeq {\mathcal D}_+^0 .$$ Since, for any integer $n$, we have $$\exp(2\pi n \sqrt{-1} \xi \frac{d}{d\xi}) = id,$$ the exponential mapping is not injective on $\underline{d}$. For any element $\underline{l} \in {\mathcal D}_+^0$ we define $\exp(T[\underline{l}])$ by $$\exp(T[\underline{l}])= \sum_{k=0}^\infty \frac{1}{k!}T[\underline{l}]^k.$$ Then, $\exp(T[\underline{l}])$ operates on $ \mathcal{H}_\lambda$ from the left and on $ \mathcal{H}_\lambda^\dagger$ from the right. By Lemma \[exponential\], for any automorphism $h \in {\mathcal D}_+^0$, there exist a unique $\underline{l} \in \underline{d}_+^0$ with $\exp(\underline{l}) = h$. Now for $h \in {\mathcal D}_+^0$ define the operator $G[h]$ by $$G[h] = \exp (- T[\underline{l}])$$ where $\exp (\underline{l}) = h$. Then, $G[h]$ operates on $ \mathcal{H}_\lambda$ from the left and on $ \mathcal{H}_\lambda^\dagger$ from the right. Then the following important theorems hold. \[thm6.1\] For any $h \in \cD_+^0$, $f(\xi)d\xi \in \bC((\xi))d\xi$, $g(\xi) \in \bC((\xi))$ and $\underline{l} = l(\xi)\frac{d}{d\xi} \in \bC((\xi))\frac{d}{d\xi}$, we have the following equalities as operators on $\cF$ and $\cFd$. $$\begin{aligned} (1) &&G[h](\psi[f(\xi)d\xi])G[h]^{-1}= \psi[h^*(f(\xi)d\xi)] = \psi[f(h(\xi))h'(\xi)d\xi] \\ (2) && G[h](\ovpsi[g(\xi)])G[h]^{-1}= \ovpsi[h^*(g(\xi))] = \ovpsi[g(h(\xi))] \\ (3)&& G[h_1\circ h_2] = G[h_1]G[h_2] \\ (4)&& G[h]T[\underline{l}]G[h]^{-1} = T[{\mathop{\fam0 ad}\nolimits}(h)(\underline{l})]+ \frac16 {\operatornamewithlimits{Res}}_{\xi=0}\big(\{h(\xi);\xi\}l(\xi)d\xi\big) .\end{aligned}$$ where $\{f(\xi); \xi\}$ is the Schwarzian derivative. \[prop6.1\] For any $h_j \in {\mathcal D}_+^0$, $j=1,2, \ldots, N$ and a pointed Riemann surface with formal neighbourhoods $$\mathfrak{X}= (C; q_1,q_2, \ldots, q_N; \xi_1, \xi_2, \ldots, \xi_N)$$ put $$\mathfrak{X}_{(h)} =(C; q_1,q_2, \ldots, q_N; h_1(\xi_1), h_2(\xi_2), \ldots, h_N(\xi_N)).$$ Then, the isomorphism $G[h_1] \widehat{\otimes}\cdots \widehat{\otimes}G[h_N]$ $$\begin{aligned} \mathcal{H}_{\vec \lambda}^\dagger & \rightarrow & \mathcal{H}_{\vec \lambda}^\dagger \\ \langle \phi_1\widehat{\otimes} \cdots \widehat{\otimes}\phi_N| & \mapsto & \langle \phi_1G[h_1] \widehat{\otimes}\cdots \widehat{\otimes}\phi_NG[h_N]|\end{aligned}$$ induces the canonical isomorphism $$G[\vh] = G[h_1] \widehat{\otimes}\cdots \widehat{\otimes}G[h_N]: {{\Cal V}_{\vec \lambda}^{\dagger}}(\mathfrak{X}) \rightarrow {{\Cal V}_{\vec \lambda}^{\dagger}}(\mathfrak{X}_{(h)})$$ Let ${{{\mathfrak X}}}= (C;\vQ;\veta)$ be a Riemann surface with formal neighbourhoods and let $q_{N+1}$ be a further point on the curve $C$ and $\eta_{N+1}$ a formal neighbourhood of $C$ at $q_{N+1}$. Put $ \vtQ = (q_1, \dots , q_N, q_{N+1})$ and $ \vteta = ( \eta_1, \dots , \eta_N, \eta_{N+1}).$ Let $${\widetilde {{{{\mathfrak X}}}}} = (C;\vtQ;\vteta).$$ We have the canonical isomorphism $P_{{\widetilde {{{{\mathfrak X}}}}},{{{\mathfrak X}}}}$ from $\mathcal V_{{{\vec \lambda}},0}^\dagger(\widetilde{{{{\mathfrak X}}}})$ to $ \mathcal V_{{{\vec \lambda}}}^\dagger({{{\mathfrak X}}})$ as given in Theorem \[T3.2.1\]. Suppose now $\vxi$ is another formal neighbourhood at $\vQ$ and that $\xi_{N+1}$ is a formal neighbourhood at $q_{N+1}$. Let then $\vtxi= (\xi, \xi_{N+1})$, ${{{\mathfrak X}}}' = (C;\vQ;\vxi)$ and ${\widetilde {{{{\mathfrak X}}}}} = (C; \vtQ;\vtxi).$ Let $\vh$ be the formal coordinate change $\vxi = \vh(\veta)$ and $\widetilde{\vh}$ the formal coordinate change $\vtxi = \vth(\vteta)$. Assume now $c({{{\mathfrak X}}},\vec \l) = c({{{\mathfrak X}}}',\vec \l)$ and $c(\widetilde{{{{\mathfrak X}}}},\vec \l,0) = c(\widetilde{{{{\mathfrak X}}}}',\vec \l,0)$. Then $h_j\in \mathcal{D}^{p_j},$ $p_j\geq 1$ and ${\widetilde h}_j\in \mathcal{D}^{{\widetilde p}_j},$ ${\widetilde p}_j\geq 1$. We then get the following diagram: $$\begin{CD} \mathcal V_{{{\vec \lambda}},0}^\dagger(\widetilde{{{{\mathfrak X}}}}) @>P_{{\widetilde {{{\mathfrak X}}}},{{{\mathfrak X}}}}>> \mathcal V_{{{\vec \lambda}}}^\dagger({{{\mathfrak X}}}) \\ @V G[\vh] VV @VV G[\vth] V\\ \mathcal V_{{{\vec \lambda}},0}^\dagger(\widetilde{{{{\mathfrak X}}}}') @>P_{{\widetilde {{{\mathfrak X}}}}',{{{\mathfrak X}}}'}>> \mathcal V_{{{\vec \lambda}}}^\dagger({{{\mathfrak X}}}') \end{CD}\label{diagcocpro}$$ \[compcocpro\] The diagram (\[diagcocpro\]) is commutative. A simple explicit calculation shows that $$L_k |0\rangle = 0$$ if $k> 0$ and $$L_0 |0\rangle = \Delta_0 |0\rangle.$$ But by $\Delta_0=0 $. From this we immediately get that $$G[\tilde h_{N+1}] |0\rangle = |0\rangle,$$ which by the very construction of the propagation of vacua isomorphism makes the above diagram commute. $\square$ The definition of the space of vacua associated to a labeled marked Riemann surface. {#New3} ------------------------------------------------------------------------------------ Let ${\mathbf C} = (C,Q,W)$ be a pointed Riemann surface and let $\l$ be a labeling of ${\mathbf C}$. We shall now define the space of vacua attached to the pair $({\mathbf C}, \l)$. We do this by providing canonical isomorphisms between the spaces of vacua associated to all correspondingly label pointed Riemann surfaces with formal neighbourhoods over ${\mathbf C}$. First we treat the case where ${\mathbf C}$ is saturated. We notice that the definition of the space of vacua ${{\Cal V}_{\vec \lambda}^{\dagger}}({{{\mathfrak X}}})$ associated to a pointed Riemann surface with formal neighbourhoods ${{{\mathfrak X}}}= (C,\vQ,\veta)$ depends on the ordering of the marked points $\vQ = (q_1, \dots, q_N)$. Let $S$ be a permutation of $\{1, \ldots, N\}$. We then define ${{{\mathfrak X}}}_S = (C, S(vQ), S(\veta))$. The permutation $S$ acting from ${{{\Cal H}_{\vec \lambda}}}$ to ${{\mathcal H}_{S(\vec \lambda)}}$, induces an isomorphism $$S : {{\Cal V}_{\vec \lambda}^{\dagger}}({{{\mathfrak X}}}) {\mathop{\fam0 \rightarrow}\nolimits}{{\mathcal V}_{S(\vec \lambda)}^{\dagger}}({{{\mathfrak X}}}_S).$$ Clearly, compositions of permutations go to compositions of isomorphisms. Let ${{{\mathfrak X}}}' = (C,\vQ,\veta)$ and ${{{\mathfrak X}}}' = (C,\vQ',\veta')$ be two pointed Riemann surface with formal neighbourhoods such that $c({{{\mathfrak X}}}) = {\mathbf C} = c({{{\mathfrak X}}}')$. Let $S$ be such that $S(\vQ) = \vQ'$. Let $\vh$ be the formal change of coordinates from $S(\veta)$ to $\veta'$. Then as discussed above we have the isomorphism $$G[\vh]S : {{\Cal V}_{\vec \lambda}^{\dagger}}({{{\mathfrak X}}}) {\mathop{\fam0 \rightarrow}\nolimits}{{\mathcal V}_{S(\vec \lambda)}^{\dagger}}({{{\mathfrak X}}}')\label{isospov}$$ to identify the two spaces of vacua with. \[Defspofv\] Let ${\mathbf C} = (C,Q,W)$ be a [saturated]{}marked Riemann surface. Let $\l$ be a labeling of ${\mathbf C}$ using the set $P_\ell$. The space of vacua associated to the labeled marked curve $({\mathbf C}, \l)$ is by definition $${{{\Cal V}_{\vec \lambda}^{\dagger}}}({\mathbf C}) = \left. \coprod_{c({{{\mathfrak X}}},\vec \l) = ({\mathbf C},\l)}{{\Cal V}_{\vec \lambda}^{\dagger}}({{{\mathfrak X}}})\right/ \sim,$$ where the disjoint union is over all labeled curves with formal neighbourhoods with $({\mathbf C},\l)$ as the underlying labeled marked curve, $\vec \lambda$ is compatible with the labeling $\l$ and $\sim$ is the equivalence relation generated by the preferred isomorphisms (\[isospov\]). That the relation $\sim$ is an equivalence relation follows from (1) and (2) in Theorem \[thm6.1\]. Further it is clear that \[Spofviso\] The natural quotient map from ${{\Cal V}_{\vec \lambda}^{\dagger}}({{{\mathfrak X}}})$ to ${{{\Cal V}_{\vec \lambda}^{\dagger}}}({\mathbf C})$ is an isomorphism for all labeled Riemann surfaces with formal neighbourhoods $({{{\mathfrak X}}}, \vec \l)$ with $c({{{\mathfrak X}}},\vec \l) = ({\mathbf C},\l)$. Suppose $({\mathbf C}_i,\l_i)$ are labeled marked Riemann surfaces and $\Phi : ({\mathbf C}_1,\l_1) {\mathop{\fam0 \rightarrow}\nolimits}({\mathbf C}_2,\l_2)$ is a morphism of labeled marked Riemann Surfaces. Let $({{{\mathfrak X}}}_2,\vec \l)$ be a labeled pointed Riemann Surface with formal neighbourhoods such that $c({{{\mathfrak X}}}_2,\vec \l_2) = ({\mathbf C},\l_2)$. Let $\Phi^*{{{\mathfrak X}}}_2 = {{{\mathfrak X}}}_1$. Then $\Phi$ is a morphism of labeled marked Riemann surfaces with formal neighbourhoods. We obviously have that \[morphspofv\] The identity map on ${{\Cal H}_{\vec \lambda}^{\dagger}}$ induces a linear isomorphism from ${{\Cal V}_{\vec \lambda}^{\dagger}}({{{\mathfrak X}}}_1)$ to ${{\Cal V}_{\vec \lambda}^{\dagger}}({{{\mathfrak X}}}_2)$, which induces a well defined linear isomorphism $\Vdag(\Phi)$ from $\Spofvlamone({\mathbf C}_1)$ to $\Spofvlamtwo({\mathbf C}_2)$. Compositions of morphisms of labeled marked Riemann Surfaces go to compositions of the induced linear isomorphisms. Let $({\mathbf C}, \l)$ be a labeled marked Riemann surface which might not be [saturated.]{}Consider all [saturated]{}labeled marked Riemann surfaces $({\mathbf C}', \l')$ obtained from $({\mathbf C}, \l)$ by adding points labeled with the trivial label $0\in P_\ell$. \[Defspofvnc\] The space of vacua associated to the labeled marked Riemann surface $({\mathbf C}, \l)$ is by definition $${{{\Cal V}_{\vec \lambda}^{\dagger}}}({\mathbf C}) = \left. \coprod_{({\mathbf C}',\l')}{{\mathcal V}_{{\mathbf \lambda}'}^{\dagger}}({\mathbf C}')\right/ \sim,$$ where the disjoint union is over all labeled marked Riemann surfaces $({\mathbf C}', \l')$ discussed above and $\sim$ is the equivalence relation generated by the propagation of vacua isomorphisms given in Theorem \[T3.2.1\], the permutations of the order of the marked points and the change of formal coordinate isomorphisms given in Theorem \[prop6.1\]. By Theorem \[thm6.1\], Proposition \[compcocpro\] and 2, this $\sim$ is also an equivalence relation. We also remark that if we apply definition \[Defspofvnc\] to a labeled marked Riemann surface $({\mathbf C}, \l)$ which is already saturated, then we obtain a space of vacua which is naturally isomorphic to the space of vacua definition \[Defspofv\] produces. The bundle of vacua over Teichmüller space {#New4} ========================================== Definition of the sheaf of vacua {#Sheafofv} -------------------------------- Let $\mathfrak{F} = ( \pi : \mathcal{C} \rightarrow \mathcal{B} ; s_1, \ldots, s_N; \eta_1, \ldots,\eta_N)$ be a family of pointed saturated Riemann Surfaces of genus $g$ with formal neighbourhoods. The sheaf $\widehat{{{{\mathfrak g}}}}_N (\mathcal{B} )$ of affine Lie algebra over $\mathcal{B} $ is the sheaf of ${{{\Cal O}_{\Cal B}}}$-module $$\widehat{{{{\mathfrak g}}}}_N(\mathcal{B} )= {{{{\mathfrak g}}}}\otimes_{{{\mathbf C}}} ({\bigoplus_{j=1}^N}{{{\Cal O}_{\Cal B}}}((\xi_j))) \oplus {{{\Cal O}_{\Cal B}}}\cdot c$$ with the following commutation relation, which is $\mathcal{O} _{\mathcal{B} }$-bilinear. $$\begin{aligned} \lefteqn{[( X_1 \otimes f_1, \ldots, X_N \otimes f_N ), (Y_1 \otimes g_1, \ldots, Y_N \otimes g_N)] } \\ && = ([X_1,Y_1] \otimes (f_1g_1),\ldots, [X_N,Y_N] \otimes (f_Ng_N)) \oplus c \cdot {\sum_{j=1}^N}(X_j, Y_j) {\operatornamewithlimits{Res}}_{{\xi_j}=0}(g_jdf_j)\end{aligned}$$ where $ X_j, \, Y_j \in {\mathfrak{g} }, \quad f_j, \, g_j \in {{{\Cal O}_{\Cal B}}}(({\xi_j}))$ and we require $c$ to be central. Put $$\widehat{{{{\mathfrak g}}}}({{{\mathfrak F}}}) = {\mathfrak{g} } \otimes_{{{\mathbf C}}}\pi_{*} ({{{\Cal O}_{\Cal C}}}(*S))$$ where we define $$\begin{aligned} S & = {\sum_{j=1}^N}s_j({\mathcal{B} }) \\ \pi_{*} ({{{\Cal O}_{\Cal C}}}(*S)) & = {\mathop{\fam0 \stackrel{\textstyle Lim}{\textstyle \longleftarrow}}}_k \pi_{*} ({{{\Cal O}_{\Cal C}}}(kS))\,.\end{aligned}$$ Laurent expansion using the formal neighbourhoods $\eta_j$’s gives an inclusion: $$\tilde{t} : \pi_*({{{\Cal O}_{\Cal B}}}(*S)) \rightarrow {\bigoplus_{j=1}^N}{{{\Cal O}_{\Cal B}}}(({\xi_j}))$$ and we may regard $\widehat{{{{\mathfrak g}}}}({{{\mathfrak F}}})$ as a Lie subalgebra of $\widehat{{{{\mathfrak g}}}}_N(\mathcal{B} )$. For any ${{\vec \lambda}}= (\lambda_1, \ldots , \lambda_N) \in (P_\ell)^N$, put $$\begin{aligned} {{{\Cal H}_{\vec \lambda}}}(\mathcal{B} ) & = {{{\Cal O}_{\Cal B}}}\otimes_{{{\mathbf C}}}{\mathcal{H}}_{{\vec \lambda}}\, , \\ {{\Cal H}_{\vec \lambda}^{\dagger}}(\mathcal{B} ) & = \underline{\hbox{\rm Hom}}_{{{{\Cal O}_{\Cal B}}}}({{{\Cal H}_{\vec \lambda}}}(\mathcal{B} ), {{{\Cal O}_{\Cal B}}}) = \mathcal{O} _{\mathcal{B} } \otimes_{{{\mathbf C}}} {{\Cal H}_{\vec \lambda}^{\dagger}}.\end{aligned}$$ The pairing induces an ${{{\Cal O}_{\Cal B}}}$-bilinear pairing $$\label{perfectpairsheaf} \langle \phantom{X} | \phantom{X} \rangle : {{\Cal H}_{\vec \lambda}^{\dagger}}(\mathcal{B} ) \times {{{\Cal H}_{\vec \lambda}}}(\mathcal{B} ) \rightarrow {{{\Cal O}_{\Cal B}}}.$$ The sheaf of affine Lie algebra $\widehat{{{{\mathfrak g}}}}_N(\mathcal{B} )$ acts on ${{{\Cal H}_{\vec \lambda}}}(\mathcal{B} )$ and ${{\Cal H}_{\vec \lambda}^{\dagger}}(\mathcal{B} )$ by $$\begin{aligned} ( (X_1 \otimes \sum_{n \in {\mathbf{Z}}} a_n^{(1)}\xi_1^n), \ldots , (X_N & \otimes \sum_{n \in {\mathbf{Z}}} a_n^{(N)}\xi_N^n)) (F \otimes |\Psi \rangle ) \\ = & {\sum_{j=1}^N}\sum_{n \in {\mathbf{Z}}}(a_n^{(j)} F) \otimes \rho_j(X_j(n))|\Psi\rangle\end{aligned}$$ The action of $\widehat{\mathfrak g}_N(\mathcal{B} )$ on ${\mathcal{H}}_{{{\vec \lambda}}}^\dagger (\mathcal{B} )$ is the dual action of ${\mathcal{H}}_{{{\vec \lambda}}}(\mathcal{B} )$, that is, $$\langle \Psi a | \Phi \rangle = \langle \Psi| a \Phi \rangle \quad \text{for any } \; a \in \widehat{\mathfrak{g} }_N .$$ \[ShofVdef\] For the family ${{{\mathfrak F}}}$ of pointed Riemann surfaces with formal neighbourhoods, we define the sheaves of ${{{\Cal O}_{\Cal B}}}$-modules on $\mathcal{B}$ $$\begin{aligned} {{{\Cal V}_{\vec \lambda}}}({{{\mathfrak F}}}) & = {{{\Cal H}_{\vec \lambda}}}(\mathcal{B} ) / \widehat{{{{\mathfrak g}}}}({{{\mathfrak F}}}) {{{\Cal H}_{\vec \lambda}}}(\mathcal{B} ) \\ {{\Cal V}_{\vec \lambda}^{\dagger}}({{{\mathfrak F}}}) & = \underline{\hbox{\rm Hom}}_{{{{\Cal O}_{\Cal B}}}}({{{\Cal V}_{\vec \lambda}}}(\mathcal{B} ), {{{\Cal O}_{\Cal B}}}).\end{aligned}$$ These are the [*sheaf of covacua*]{} and the [*sheaf of vacua*]{} attached to the family ${{{\mathfrak F}}}$. Note that we have $${{\Cal V}_{\vec \lambda}^{\dagger}}({{{\mathfrak F}}}) = \{ \, \langle \Psi| \in {{\Cal H}_{\vec \lambda}^{\dagger}}({{{\mathfrak F}}}) \, | \;\; \langle \Psi | a = 0 \quad \text{for any }\; a \in \widehat{{{{\mathfrak g}}}}({{{\mathfrak F}}}) \, \} .$$ The pairing ) induces an ${{{\Cal O}_{\Cal B}}}$-bilinear pairing $$\langle \phantom{X} |\phantom{X} \rangle : {{\Cal V}_{\vec \lambda}^{\dagger}}({{{\mathfrak F}}}) \times {{{\Cal V}_{\vec \lambda}}}({{{\mathfrak F}}}) \rightarrow {{{\Cal O}_{\Cal B}}}.$$ By Theorem 4.1.6 in [@Ue2] and Corollary 4.2.4 in [@Ue2] we have the following highly non-trivial theorem. Note that the theorem is true even for a family of nodal curves (see Theorem \[stabllocalfreeness\]). \[localfreeness\] The sheaves $ {{\Cal V}_{\vec \lambda}^{\dagger}}({{{\mathfrak F}}})$ and ${{{\Cal V}_{\vec \lambda}}}({{{\mathfrak F}}})$ are locally free sheaves of ${{{\Cal O}_{\Cal B}}}$-modules of finite rank over $\mathcal B$. They are dual to each other. Hence ${{\Cal V}_{\vec \lambda}^{\dagger}}({{{\mathfrak F}}})$ is a holomorphic vector bundles over $\mathcal B$. Properties of the sheaf of vacua. {#New4Props} --------------------------------- To construct the bundle of vacua over Teichmüller space, we further need the following obvious property of the sheaf of vacua construction. \[Lfamiso\] Let ${{{\mathfrak F}}}_i$ be two families of saturated pointed Riemann surfaces with formal neighbourhoods over the same base $\mathcal B$. Let $\Phi : ({{{\mathfrak F}}}_1,\l_1) {\mathop{\fam0 \rightarrow}\nolimits}({{{\mathfrak F}}}_2,\l_2)$ be an isomorphism of labeled families, which induces the identity map on the base. Then the identity map on ${{\Cal H}_{\vec \lambda}^{\dagger}}(\mathcal B)$ induces a canonical isomorphism $$\Vdag(\Phi) : \Vdaglamone({{{\mathfrak F}}}_1) {\mathop{\fam0 \rightarrow}\nolimits}\Vdaglamtwo({{{\mathfrak F}}}_2).\label{famiso}$$ Suppose that we have two families of pointed Riemann surfaces with formal neighbourhoods ${{{\mathfrak F}}}_i$, $i=1,2$ on a pointed saturated and stable surface $({\Sigma},P)$, with the property that they have the same image $\Psi_{{{{\mathfrak F}}}_1}({\mathcal B}_1) = \Psi_{{{{\mathfrak F}}}_2}({\mathcal B}_2)$ in Teichmüller space ${\mathop{\fam0 {\mathcal T}}\nolimits}_{({\Sigma},P)}$ and that ${{{\mathfrak F}}}_2$ is a [good]{}family. For such a pair of families there exists by Proposition \[famequivalence\] a unique fiber preserving biholomorphism $\Phi_{12} : \mathcal C_1{\mathop{\fam0 \rightarrow}\nolimits}\mathcal C_2$ covering $\Psi^{-1}_{{{{\mathfrak F}}}_2}\Psi_{{{{\mathfrak F}}}_1}$ such that $\Phi^{-1}_{{{{\mathfrak F}}}_2} \Phi_{12} \Phi_{{{{\mathfrak F}}}_1} : (Y, P) {\mathop{\fam0 \rightarrow}\nolimits}(Y,P)$ is isotopic to $\Psi^{-1}_{{{{\mathfrak F}}}_2}\Psi_{{{{\mathfrak F}}}_1}\times {\mathop{\fam0 Id}\nolimits}$ through such fiber preserving maps inducing the identity on the first order neighbourhood of $P$. We note that $(\Phi_{12}^* (\veta_2))^{(1)} = (S\veta_1)^{(1)}$, where $S$ is some permutation of $\{1, \ldots, N\}$, i.e. $\Phi_{12}^* (\veta_2)$ induce the same first order formal neighbourhoods as $S\veta_1$ does. Let $\vh$ be the formal change of coordinates from $S\veta_1$ to $\Phi_{12}^* (\veta_2)$. Let $\widetilde{{{{\mathfrak F}}}_1} = \Phi_{12}^*({{{\mathfrak F}}}_2)$. Then $\Phi_{12}$ induces an isomorphism of families from $\widetilde{{{{\mathfrak F}}}_1}$ to $(\Psi^{-1}_{{{{\mathfrak F}}}_2}\Psi_{{{{\mathfrak F}}}_1})^{*}({{{\mathfrak F}}}_2)$. Choose labelings $\vec \l_1$ of ${{{\mathfrak F}}}_1$ and $\vec \l_2$ of ${{{\mathfrak F}}}_2$, which are compatible under the above isomorphism. \[vbtransf\] The action of $S$ on $\Hdaglamone(\mathcal B_1)$ induces an isomorphism from $\Vdaglamone({{{\mathfrak F}}}_1)$ to $\Vdaglamtwo(S{{{\mathfrak F}}}_1)$, where S acts on the family ${{{\mathfrak F}}}_1$ by permuting the numbering of the formal neighbourhoods and the sections. Furthermore $G[\vh] : \Hdaglamtwo(\mathcal B_1) \rightarrow \Hdaglamtwo(\mathcal B_1)$ induces an isomorphism from $\Vdaglamtwo(S{{{\mathfrak F}}}_1)$ to $\Vdaglamtwo(\widetilde{{{{\mathfrak F}}}_1})$. The natural pull back isomorphisms provided by Lemma 4.1.3 in [@Ue2] and the families isomorphism (\[famiso\]) provided by Lemma \[Lfamiso\] induces an isomorphism from $\Vdaglamtwo(\widetilde{{{{\mathfrak F}}}_1})$ to $(\Psi^{-1}_{{{{\mathfrak F}}}_2}\Psi_{{{{\mathfrak F}}}_1})^*\Vdaglamtwo({{{\mathfrak F}}}_2)$. The composite of these three isomorphism gives the transformation isomorphism $$G_{12} : \Vdaglamone({{{\mathfrak F}}}_1) {\mathop{\fam0 \rightarrow}\nolimits}(\Psi^{-1}_{{{{\mathfrak F}}}_2}\Psi_{{{{\mathfrak F}}}_1})^*\Vdaglamtwo({{{\mathfrak F}}}_2).\label{overlapiso}$$ The isomorphisms satisfies the cocycle condition $$G_{12} (\Psi^{-1}_{{{{\mathfrak F}}}_2}\Psi_{{{{\mathfrak F}}}_1})^*G_{23} = G_{13}.$$ $\square$ Let ${{{\mathfrak F}}}= ( \pi : \mathcal{C} \rightarrow \mathcal{B} ; s_1, \ldots, s_N; \eta_1, \ldots,\eta_N)$ be a family of pointed Riemann surfaces with formal neighbourhoods. Let $s_{N+1}$ be a further section of $\pi$ disjoint from the $s_i$ and $\eta_{N+1}$ a formal neighbourhood along $s_{N+1}$ and set $\widetilde{{{{\mathfrak F}}}} = ( \pi : \mathcal{C} \rightarrow \mathcal{B} ; s_1, \ldots, s_{N+1}; \eta_1, \ldots,\eta_{N+1})$ \[Povforfamilies\] The inclusion $$\begin{aligned} {{{\Cal H}_{\vec \lambda}}}(\mathcal B) &\longrightarrow &{{{\Cal H}_{\vec \lambda}}}(\mathcal B) \otimes {\mathcal H}_0(\mathcal B) \\ |v \rangle &\longrightarrow & |v \rangle \otimes |0 \rangle\end{aligned}$$ induces the propagation of vacua isomorphism $$P_{\widetilde{{\mathfrak F}},{\mathfrak F}} : \mathcal V_{{{\vec \lambda}},0}^\dagger(\widetilde{{\mathfrak F}}) {\mathop{\fam0 \rightarrow}\nolimits}\mathcal V_{{{\vec \lambda}}}^\dagger({\mathfrak F}).$$ The definition of the bundle of vacua over Teichmüller space {#New4D} ------------------------------------------------------------ Let ${\Sigma}$ be a closed oriented smooth surface and let $P$ be a finite set of marked points on ${\Sigma}$. Let $\l$ be a labeling of $({\Sigma}, P)$ and assume $({\Sigma}, P)$ is [stable]{}and [saturated.]{}We now define a holomorphic vector bundle ${{{\Cal V}_{\vec \lambda}^{\dagger}}}= {{{\Cal V}_{\vec \lambda}^{\dagger}}}({\Sigma},P)$ over Teichmüller space ${\mathop{\fam0 {\mathcal T}}\nolimits}_{({\Sigma},P)}$ using the cover $\{\Psi_{{{{\mathfrak F}}}}(\mathcal B)\}$, where ${{{\mathfrak F}}}$ runs over the good families of complex structures on $({\Sigma},P)$. \[dvbovac\] A holomorphic vector bundle ${{{\Cal V}_{\vec \lambda}^{\dagger}}}= {{{\Cal V}_{\vec \lambda}^{\dagger}}}({\Sigma},P)$ over Teichmüller space ${\mathop{\fam0 {\mathcal T}}\nolimits}_{({\Sigma},P)}$ is specify to be the bundle $(\Psi_{{{\mathfrak F}}}^{-1})^*{{\Cal V}_{\vec \lambda}^{\dagger}}({{{\mathfrak F}}})$ over $\Psi_{{{{\mathfrak F}}}}(\mathcal B)$ for any good families of complex structures on $({\Sigma},P)$. On overlaps of the image of two [good]{}families, we use the glueing isomorphism (\[overlapiso\]) to glue the corresponding bundles together. Proposition \[vbtransf\] implies that ${{{\Cal V}_{\vec \lambda}^{\dagger}}}({\Sigma},P)$ is a vector bundle over ${\mathop{\fam0 {\mathcal T}}\nolimits}_{({\Sigma},P)}$ with the following property. \[Teichpullback\] For any [stable]{}and [saturated]{}family ${{\tilde {\mathfrak F}}}$ of pointed curves with formal neighbourhoods on $({\Sigma},P)$ we have a preferred isomorphism $$\Upsilon_{{{\tilde {\mathfrak F}}}} : {{\Cal V}_{\vec \lambda}^{\dagger}}({{\tilde {\mathfrak F}}}) {\mathop{\fam0 \rightarrow}\nolimits}\Psi_{{{\tilde {\mathfrak F}}}}^*{{{\Cal V}_{\vec \lambda}^{\dagger}}}({\Sigma},P)$$ induced by the transformation isomorphism between ${{\Cal V}_{\vec \lambda}^{\dagger}}({{\tilde {\mathfrak F}}})$ and ${{\Cal V}_{\vec \lambda}^{\dagger}}({{{\mathfrak F}}})$, for [good]{}families ${{{\mathfrak F}}}$ of complex structures on $({\Sigma},P)$ such that $\Psi_{{{\mathfrak F}}}(\mathcal B)$ intersect $\Psi_{{{\tilde {\mathfrak F}}}}({\mathcal B}')$ nonempty. Let $({\Sigma},P,\lambda)$ be a general labeled pointed surface, i.e. $({\Sigma},P)$ might not be [stable]{}nor [saturated.]{}Let $({\Sigma},P_i,\lambda_i)$, $i=1,2$ be any labeled marked surfaces obtained from $({\Sigma},P,\lambda)$ by labeling further points by $0\in P_\ell$. Assume that $({\Sigma},P_i)$ are [stable]{}and [saturated]{}pointed surfaces. Let $\bar P = P_1 \cup P_2$ and $\bar \l$ be the induced labeling of $\bar P$. Note that $({\Sigma},\bar P)$ is also [stable]{}and [saturated.]{} We get holomorphic projection maps $\pi_i : {\mathop{\fam0 {\mathcal T}}\nolimits}_ {({\Sigma},\bar P)}{\mathop{\fam0 \rightarrow}\nolimits}{\mathop{\fam0 {\mathcal T}}\nolimits}_ {({\Sigma},P_i)}$. As a direct consequence of Proposition \[compcocpro\] we get the following. \[propvaciso\] Iterations of the propagation of vacua isomorphism given in Theorem \[Povforfamilies\] induces natural isomorphisms of bundles $$\Spofvlambar (\Sigma,\bar P) \cong \pi_1^*\Spofvlamone (\Sigma,P_1) \cong \pi_2^*\Spofvlamtwo (\Sigma,P_2),$$ which satisfies associativity. Suppose now $f : ({\Sigma}_1, P_1) {\mathop{\fam0 \rightarrow}\nolimits}({\Sigma}_2,P_2)$ is a morphism of [stable]{}and [saturated]{}pointed surfaces. Then of course $f$ induces a morphism $f^*$ from ${\mathop{\fam0 {\mathcal T}}\nolimits}_{({\Sigma}_1,P_1)}$ to ${\mathop{\fam0 {\mathcal T}}\nolimits}_{({\Sigma}_2,P_2)}$. Let $\l_1$ be a labeling of $({\Sigma}_1,P_1)$ and let $\l_2$ be the induced labeling on $({\Sigma}_2,P_2)$ such that $f : ({\Sigma}_1,P_1,\l_1) {\mathop{\fam0 \rightarrow}\nolimits}({\Sigma}_2,P_2,\l_2)$ is a morphism of labeled pointed surfaces. Let now ${{{\mathfrak F}}}_1$ be a [good]{}family of stable pointed curves with formal neighbourhoods of $({\Sigma}_1,P_1)$. Then by composing with $f^{-1}\times {\mathop{\fam0 Id}\nolimits}$ we get a [good]{}family ${{{\mathfrak F}}}_2$ of stable pointed curves with formal neighbourhoods on $({\Sigma}_2,P_2)$ over the same base $\mathcal B_1$. The identity morphism on ${{\Cal H}_{\vec \lambda}^{\dagger}}(\mathcal B_1)$ then induces a morphism $\Vdag(f) : {{\Cal V}_{\vec \lambda}^{\dagger}}({{{\mathfrak F}}}_1) {\mathop{\fam0 \rightarrow}\nolimits}{{\Cal V}_{\vec \lambda}^{\dagger}}({{{\mathfrak F}}}_2)$ which covers the identity on the base. This is precisely the morphism induced from the morphism of families $\Phi_f = f\times {\mathop{\fam0 Id}\nolimits}: {{{\mathfrak F}}}_1 {\mathop{\fam0 \rightarrow}\nolimits}{{{\mathfrak F}}}_2$ by Lemma \[Lfamiso\]. This intern then induces a morphism $\Vdag(f) : (\Psi_{{{{\mathfrak F}}}_1}^{-1})^*(\Vdaglamone({{{\mathfrak F}}}_1)) {\mathop{\fam0 \rightarrow}\nolimits}(\Psi_{{{{\mathfrak F}}}_2}^{-1})^*(\Vdaglamtwo({{{\mathfrak F}}}_2))$ which covers $f^* : \Psi_{{{{\mathfrak F}}}_1}(B_1) {\mathop{\fam0 \rightarrow}\nolimits}\Psi_{{{{\mathfrak F}}}_2}(B_1)$. \[comptransf\] The above construction provides a well defined lift of $f^* : {\mathop{\fam0 {\mathcal T}}\nolimits}_{({\Sigma}_2,P_2)} {\mathop{\fam0 \rightarrow}\nolimits}{\mathop{\fam0 {\mathcal T}}\nolimits}_{({\Sigma}_1,P_1)}$ to a morphism $\Spofv(f) : \Spofvlamone({\Sigma}_1,P_1) {\mathop{\fam0 \rightarrow}\nolimits}\Spofvlamtwo({\Sigma}_2,P_2)$ which behaves well under compositions. $\square$ Assume that $({\Sigma}_i,P_i, \l_i)$ are labeled pointed surfaces, which need not be neither [stable]{}nor [saturated.]{}Let $f : ({\Sigma}_1,P_1, \l_1) {\mathop{\fam0 \rightarrow}\nolimits}({\Sigma}_2,P_2,\l_2)$ be an orientation preserving diffeomorphism of labeled pointed surfaces. Let $({\Sigma}_i, P'_i,\l'_i)$ be labeled pointed surfaces obtained from $({\Sigma}_i,P_i, \l_i)$ by labeling further points by $0\in P_\ell$ such that $({\Sigma}_i, P'_i,\l'_i)$ are [stable]{}and [saturated]{}labeled pointed surfaces such that $f:({\Sigma}_1,P'_1,\l'_1) {\mathop{\fam0 \rightarrow}\nolimits}({\Sigma}_2, P'_2,\l'_2)$ is a morphism of labeled pointed surfaces. We obviously have the following result. \[morphcomp\] The lift of $f^* : {\mathop{\fam0 {\mathcal T}}\nolimits}_{({\Sigma}_1,P'_1)} {\mathop{\fam0 \rightarrow}\nolimits}{\mathop{\fam0 {\mathcal T}}\nolimits}_{({\Sigma}_2,P'_2)}$ to a morphism $\Spofv(f) : \Spofvlampone({\Sigma}_1,P'_1) {\mathop{\fam0 \rightarrow}\nolimits}\Spofvlamptwo({\Sigma}_2,P'_2)$ as given by Theorem \[comptransf\] is compatible with the isomorphisms given in Proposition \[propvaciso\]. The connection in the bundle of vacua over Teichmüller space {#New5} ============================================================ Twisted first order differential operators acting on the sheaf of vacua ----------------------------------------------------------------------- Let $\mathfrak{F} = (\pi : \mathcal{C} \rightarrow \mathcal{B} ; s_1, \ldots, s_N ; \eta_1, \ldots, \eta_N)$ is a family of saturated pointed Riemann surfaces of genus $g$ with formal neighbourhoods. We will assume that ${{{\mathfrak F}}}^{(0)}= (\pi : \mathcal{C} \rightarrow \mathcal{B} ; s_1, \ldots, s_N)$ be a [*versal family*]{} of pointed stable curves of genus $g$ in the sense of definition 1.2.2. in [@Ue2]. We consider the divisors $$S_j = s_j(\mathcal{B} ), \qquad S = \sum_{j=1}^N S_j.$$ There is an exact sequence $$0 \rightarrow \Theta_{\mathcal{C} /\mathcal{B} } \rightarrow \Theta_{\mathcal{C} } \overset{d\pi}{\rightarrow} \pi^*\Theta_{\mathcal{B} } \rightarrow 0$$ where $\Theta_{\mathcal{C} /\mathcal{B} }$ is a sheaf of vector fields tangent to the fibres of $\pi$. Put $$\Theta'_{\mathcal{C} ,\pi} = d\pi^{-1}(\pi^{-1}\Theta_{\mathcal{B} }).$$ Hence, $\Theta'_{\mathcal{C} ,\pi}$ is a sheaf of vector field on $\mathcal{C} $ whose vertical components are constant along the fibers of $\pi$. That is, in a neighbourhood of a point of a fiber $\Theta'_{\mathcal{C} ,\pi}$ consists of germs of holomorphic vector fields of the form $$a(z,u) \frac{\partial}{\partial z} + \sum_{i=1}^n b_i(u) \frac{\partial}{\partial u_i}$$ where $(z,u_1 ,\ldots, u_n)$ is a system of local coordinates such that the mapping $\pi$ is expressed as the projection $$\pi (z,u_1 ,\ldots, u_n) =(u_1, \ldots, u_n).$$ More generally, we can define a sheaf $\Theta'_{\mathcal{C} }(mS)_\pi$ as the one consisting of germs of meromorphic vector fields of the form $$A(z,u) \frac{\partial}{\partial z} + \sum_{i=1}^n B_i(u) \frac{\partial}{\partial u_i}$$ where $A(z,u)$ has the poles of order at most $m$ along $S$. We have an exact sequence $$0 \rightarrow \Theta_{\mathcal{C} /\mathcal{B} }(mS) \rightarrow \Theta'_{\mathcal{C} }(mS)_\pi \overset{d\pi}{\rightarrow } \pi^{-1}\Theta_{\mathcal{B} } \rightarrow 0 .$$ Note that $\Theta'_{\mathcal{C} }(mS)_\pi $ has the structure of a sheaf of Lie algebras by the usual bracket operation on vector fields and the above exact sequence is one of sheaves of Lie algebras. For $m > \dfrac1N(2g - 2)$ we have an exact sequence of $\mathcal{O} _\mathcal{B} $-modules. $$0 \rightarrow \pi_* \Theta_{\mathcal{C} /\mathcal{B} }(mS) \rightarrow \pi_* \Theta'_{\mathcal{C} }(mS)_\pi \overset{d\pi}{\rightarrow } \Theta_{\mathcal{B} } \rightarrow 0$$ which is also an exact sequence of sheaves of Lie algebras. Taking $m \rightarrow \infty$ we obtain the exact sequence $$0 \rightarrow \pi_* \Theta_{\mathcal{C} /\mathcal{B} }(*S) \rightarrow \pi_* \Theta'_{\mathcal{C} }(*S)_\pi \overset{d\pi}{\rightarrow } \Theta_{\mathcal{B} } \rightarrow 0 .$$ Recall that we have the following exact sequence of ${\mathcal{O} }_{\mathcal{B} }$-modules. $$0 \rightarrow \Theta_{\mathcal{C} /\mathcal{B} }( - S)) \rightarrow \Theta_{\mathcal{C} /\mathcal{B} }( mS)) \rightarrow \bigoplus_{j=1}^N \bigoplus_{k=0}^m {\mathcal{O} }_{\mathcal{B} }\xi_j^{-k} \displaystyle{\frac{d}{d\xi_j}} \rightarrow 0$$ Which for any positive integer $m \geq 4g -3$ gives the following exact sequence $$0 \rightarrow \pi_*(\Theta_{\mathcal{C} /\mathcal{B} }( mS)) \overset{b_m}{\rightarrow} \bigoplus_{j=1}^N \bigoplus_{k=0}^m {\mathcal{O} }_{\mathcal{B} }\xi_j^{-k} \displaystyle{\frac{d}{d\xi_j}} \overset{\vartheta_m}{\rightarrow} R^1\pi_* \Theta_{\mathcal{C} /\mathcal{B} }(-S) \rightarrow 0$$ From which we deduce the following exact sequence of ${{{\Cal O}_{\Cal B}}}$-modules $$0 \rightarrow \pi_*(\Theta_{\mathcal{C} /\mathcal{B} }( * S)) \overset{b}{\rightarrow} \bigoplus_{j=1}^N {\mathcal{O} }_{\mathcal{B} }[\,\xi_j^{-1}\,] \displaystyle{\frac{d}{d\xi_j}} \overset\vartheta{\rightarrow} R^1\pi_* \Theta_{\mathcal{C} /\mathcal{B} }(-S) \rightarrow 0.$$ Note that the mappings $b$ and $b_m$ correspond to the Laurent expansions with respect to $\xi_j$ up to zero-th order. By the Kodaira-Spencer theory (see [@Ue2] for details) we have the following commutative diagram. $$\begin{matrix} 0 \rightarrow & \pi_* \Theta_{\mathcal{C} /\mathcal{B} }(*S) & \rightarrow & \pi_* \Theta'_{\mathcal{C} }(*S)_\pi & \overset{d\pi}{\rightarrow }& \Theta_{\mathcal{B} } & \rightarrow 0 \\ & & & & &&\\ & \Vert & \phantom{p} & \;\;\downarrow \;p & \phantom{\rho} &\;\; \downarrow \;\rho & \\ & & & & &&\\ 0 \rightarrow & \pi_* \Theta_{\mathcal{C} /\mathcal{B} }(*S) & \rightarrow & \bigoplus_{j=1}^N \mathcal{O} _\mathcal{B} [\; \xi_j^{-1}] \dfrac d{d \xi_j} & \overset{\vartheta}{\rightarrow }& R^1\pi_* \Theta_{\mathcal{C} /\mathcal{B} }( -S) & \rightarrow 0 \end{matrix}$$ where $\rho$ is the Kodaira-Spencer mapping of the family $\mathfrak{F}^{(0)}$ and $p$ is given by taking the non-positive part of the $\dfrac d {d\xi_j}$ part of the Laurent expansions of the vector fields in $\pi_*\Theta_\mathcal{C} (mS)_\pi$ at $s_j(\mathcal{B})$ . Since our family $\mathfrak{F}^{(0)}$ is versal, it follows from Proposition 1.2.6. in [@Ue2], that the Kodaira-Spencer mapping $\rho$ is an isomorphism of $\mathcal{O}_\mathcal{B} $-modules. Therefore, $p$ is an isomorphism. Put $$\mathcal{L}(\mathfrak{F}) := \bigoplus_{j=1}^N \mathcal{O} _\mathcal{B} [\; \xi_j^{-1}] \frac d{d \xi_j}.$$ Then, we have the following exact sequence $$0 \rightarrow \pi_* \Theta_{\mathcal{C} /\mathcal{B} }(*S) \rightarrow \mathcal{L}(\mathfrak{F} ) \overset{\theta}{\rightarrow } \Theta_{\mathcal{B} } \rightarrow 0 \label{Lfs}$$ of $\mathcal{O} _\mathcal{B} $-modules. The Lie bracket $[\phantom{X}, \phantom{X}]_d$ on $\mathcal{L}(\mathfrak{F} )$ is determined by requiring $p$ to be a Lie algebra isomorphism. Thus, for $\vec \ell$, $\vec m \in \mathcal{L}(\mathfrak{F})$ we have $$\label{35f} [\vec \ell, \vec m]_d = [\vec \ell, \vec m]_0 + \theta(\vec \ell)(\vec m) - \theta(\vec m)(\vec \ell)$$ where $[\phantom{X},\phantom{X}]_0$ is the usual bracket of formal vector fields and the action of $\theta(\vec \ell)$ on $$\vec m = (m_1{d\over{d\xi_1}}, \ldots, m_N{d\over{d\xi_N}})$$ is defined by $$\theta(\vec \ell)(\vec m) = ( \theta(\vec \ell)(m_1){d\over{d\xi_1}}, \ldots, \theta(\vec \ell)(m_N){d\over{d\xi_N}} ).$$ Then, the exact sequence (\[Lfs\]) is an exact sequence of sheaves of Lie algebras. For $\vec \ell = (\underline l_1, \ldots, \underline l_N) \in \mathcal{L}({{{\mathfrak F}}})$, the action $D(\vec \ell)$ on $\mathcal{H}_{{\vec \lambda}}(\mathcal{B} )$ is defined by $$D(\vec \ell) (F \otimes |\Phi \rangle) = \theta(\vec \ell)(F) \otimes |\Phi\rangle - F \cdot (\sum_{j=1}^N \rho_j(T[\underline l_j] ) |\Phi\rangle$$ where $$F \in {\mathcal{O} }_{\mathcal{B} }, \quad |\Phi\rangle \in {\mathcal{H}}_{{{\vec \lambda}}}.$$ We have the following propositions. The action $D(\vec \ell)$ of $\vec \ell \in \mathcal{L}({{{\mathfrak F}}})$ on ${{{\Cal H}_{\vec \lambda}}}(\mathcal{B} )$ defined above has the following properties. [1)]{} For any $f \in {\mathcal{O} }_{\mathcal{B} }$ we have $$D(f\vec \ell) = f D(\vec \ell).$$ [2)]{} For $\vec \ell$, $\vec m \in \mathcal{L}({{{\mathfrak F}}})$ we have $$[\,D(\vec \ell), D(\vec m)\,] = D([\,\vec \ell, \vec m\,]_d) +\frac{c_v}{12} {\sum_{j=1}^N}{\operatornamewithlimits{Res}}_{\xi_j=0} \left( \frac{d^3 \ell_j}{d \xi_j^3} m_j d\xi_j \right)\cdot {\text id}.$$ [3)]{} For $f \in {\mathcal{O} }_{\mathcal{B} }$ and $|\phi\rangle \in {{{\Cal H}_{\vec \lambda}}}(\mathcal{B} )$ we have $$D(\vec \ell)(f|\phi\rangle) = (\theta(\vec \ell)(f))|\phi\rangle + f D(\vec \ell)|\phi \rangle.$$ Namely, $D(\vec \ell)$ is a first order differential operator, if $\theta(\vec \ell ) \neq 0$. We define the dual action of $\mathcal{L}({{{\mathfrak F}}})$ on ${{\Cal H}_{\vec \lambda}^{\dagger}}(\mathcal{B} )$ by $$D(\vec \ell) (F \otimes \langle \Psi|) = (\theta(\vec \ell)F) \otimes \langle \Psi| + {\sum_{j=1}^N}F \cdot \langle \Psi| \rho_j(T[\underline l_j]).$$ where $$F \in {{{\Cal O}_{\Cal B}}}, \qquad \langle \Psi | \in {{\Cal H}_{\vec \lambda}^{\dagger}}(\mathcal{B}).$$ Then, for any $|\widetilde \Phi\rangle \in {{{\Cal H}_{\vec \lambda}}}(\mathcal{B})$ and $\langle \widetilde\Psi| \in {{\Cal H}_{\vec \lambda}^{\dagger}}(\mathcal{B})$, we have $$\{D(\vec \ell) \langle \widetilde\Psi| \}|\widetilde\Phi\rangle + \langle \widetilde\Psi|\{D(\vec \ell)|\widetilde\Phi\rangle\} = \theta(\vec \ell)\langle \widetilde\Psi|\widetilde\Phi\rangle .$$ For any $\vec \ell \in \mathcal{L}({{{\mathfrak F}}})$ we have $$D(\vec \ell)(\widehat{\mathfrak{g} }(\mathfrak{F}){{{\Cal H}_{\vec \lambda}}}(\mathcal{B} )) \subset \widehat{\mathfrak{g} }(\mathfrak{F} ){{{\Cal H}_{\vec \lambda}}}(\mathcal{B} ).$$ Hence, $D(\vec \ell)$ operates on ${{{\Cal V}_{\vec \lambda}}}({{{\mathfrak F}}})$. Moreover, it is a first order differential operator, if $\theta(\vec \ell) \neq 0$. For each element $\vec \ell \in \mathcal{L}({{{\mathfrak F}}})$, $D(\vec \ell)$ acts on ${{\Cal V}_{\vec \lambda}^{\dagger}}({{{\mathfrak F}}})$. Moreover, if $\theta(\vec \ell) \neq 0$, then $D(\vec \ell)$ acts on ${{\Cal V}_{\vec \lambda}^{\dagger}}({{{\mathfrak F}}})$ as a first order differential operator. Note that for the natural bilinear pairing $\langle \phantom{X}|\phantom{X}\rangle : {{\Cal V}_{\vec \lambda}^{\dagger}}({{{\mathfrak F}}}) \times {{{\Cal V}_{\vec \lambda}}}({{{\mathfrak F}}}) \rightarrow {{{\Cal O}_{\Cal B}}}$, we have the equality $$\{ D(\vec \ell) \langle\Psi |\}|\Phi\rangle + \langle\Psi |\{D(\vec \ell)|\Phi\rangle \}= \theta (\vec \ell) (\langle\Psi |\Phi\rangle).$$ The projectively flat connections on the on the bundle of vacua --------------------------------------------------------------- Let ${\Sigma}$ be a closed oriented smooth surface and let $P$ be a finite set of marked points on ${\Sigma}$. Assume that $({\Sigma},P)$ is [stable]{}and [saturated]{}pointed surface. Let ${{{\mathfrak F}}}= ( \pi \: \mathcal C \rightarrow \mathcal B; \vs; \veta )$ be a family of pointed Riemann Surfaces with formal neighbourhoods on $({\Sigma}, P)$. Recall the discussion of symmetric bidifferentials from section 1.4 in [@Ue2]. We now introduce the notion of a normalized symmetric bidifferential. \[nsymbidif\] A symmetric bidifferential $\omega \in H^0(\mathcal C \times_{\mathcal B} \mathcal C, \omega_{\mathcal C \times_{\mathcal B} \mathcal C/\mathcal B}(2\Delta))$ with $$\omega= \left(\frac{1}{(x-y)^2} + \text{holomorphic}\right)dxdy$$ in a neighbourhood of the diagonal of ${\mathcal C \times_{\mathcal B} \mathcal C}$ is called a [*normalized*]{} symmetric bidifferential for the family ${{{\mathfrak F}}}$. For a Riemann surface $R$ we let $(\vec{\alpha}, \vec{\beta})= (\alpha_1, \ldots, \alpha_g,\beta_1,\ldots,\beta_g)$ be a symplectic basis of $H_1(R,{{\mathbb Z}}{})$. We can find a basis $\{\omega_1, \ldots, \omega_g\}$ of holomorphic one forms of $R$ with $$\label{betaone} \int_{\beta_i}\omega_j = \delta_{i j}, \quad 1 \le i,j \le g.$$ The matrix $$\label{period} \tau = (\tau_{ij}), \quad \tau_{ij} = \int_{\alpha_i}\omega_j$$ is then called the period matrix of the Riemann surface $R$. The complex torus $$J(R) = {\mathbb C}^g/(\tau, I_g)$$ is called a Jacobian variety. If we chose a point $P$ on $R$ we can define a holomorphic mapping $$\begin{aligned} {2} j : &R &\quad \rightarrow & \quad J(R) \\ & Q & \quad \mapsto &\left( \int_P^Q \omega_1, \ldots, \int_P^Q \omega_g \right).\end{aligned}$$ If a family of pointed Riemann surfaces is given, we can construct a family of Jacobian varieties and a family of holomorphic mappings. We have the following lemma as a consequence of the construction in Section 1.4 in [@Ue2]. \[nsymbidifexists\] For any family of pointed Riemann surfaces with formal neighbourhoods ${{{\mathfrak F}}}$ on $({\Sigma},P)$ and any symplectic basis $(\vec{\alpha}, \vec{\beta})=(\alpha_1, \ldots, \alpha_g,\beta_1,\ldots,\beta_g)$ of $H_1({\Sigma}, {{\mathbb Z}}{})$, there is a unique normalized symmetric bidifferential $\omega \in H^0(\mathcal C \times_{\mathcal B} \mathcal C, \omega_{\mathcal C \times_{\mathcal B} \mathcal C/\mathcal B}(2\Delta))$ determined by formula $$\label{primeform} \omega(x,y)dxdy = \frac{\partial^2 \log E(x,y)}{\partial x \partial y}dx dy$$ where $E(x,y)(\sqrt{dx})^{-1}(\sqrt{dy})^{-1}$ is the prime form associated to the symplectic basis $(\vec{\alpha}, \vec{\beta})$ for each Riemann surface. For a prime form see Chapter II of [@Fa]. Please do note that $\alpha$ and $\beta$ play the reverse roles in [@Ue2], but the same as in [@AU1]. Note that a prime form of a Riemann surface $R$ (hence, also a normalized symmetric bidifferential) is uniquely determined by a symplectic basis $(\vec{\alpha}, \vec{\beta})$ of $H_1(R,{{\mathbb Z}}{})$ . If $(\vec{\widehat{\alpha}}, \vec{\widehat{\beta}})= (\widehat{\alpha}_1, \ldots, \widehat{\alpha}_g, \widehat{\beta}_1,\ldots,\widehat{\beta}_g)$ is another symplectic basis , there exists a symplectic matrix $$\Lambda = \left( \begin{array}{cc} A&B\\C&D \end{array} \right) \in Sp(g, {{\mathbb Z}}{})$$ such that $$\label{action1} \left(\begin{array}{c} \widehat{\alpha}_1\\ \vdots\\ \widehat{\alpha}_g \\ \widehat{\beta}_1\\ \vdots \\ \widehat{\beta}_g \end{array}\right)= \left(\begin{array}{cc}A & B\\C & D\end{array}\right) \left(\begin{array}{c} \alpha_1\\ \vdots\\ \alpha_g \\ \beta_1\\ \vdots \\ \beta_g \end{array}\right).$$ Also for any element $\Lambda \in Sp(g, {{\mathbb Z}}{})$, by we can define a new symplectic basis $\Lambda (\vec{\alpha},\vec{\beta}) = (\widehat{\alpha}_1, \ldots, \widehat{\alpha}_g, \widehat{\beta}_1, \ldots, \widehat{\beta}_g)$. Then the normalized symmetric bidifferential $\widehat{\omega}(x,y)dx dy$ associated to the symplectic basis $\Lambda (\vec{\alpha},\vec{\beta})$ and the normalized symmetric bidifferential $\omega(x,y)dx dy$ have a relation: $$\begin{aligned} \label{bidiffrelation} \widehat{\omega}(x,y)dxdy &=& \omega(x,y)dxdy \\ && -\frac12 \sum_{i\le j} \left\{ v_i(x) v_j(y) +v_j(x)v_i(y)\right\} \frac{\partial}{\partial \tau_{ij}}\log \det(C \tau +D) \nonumber\end{aligned}$$ where $\{\omega_1 =v_1(x)dx, \ldots ,\omega_g=v_g(x)dx\}$ is a basis of holomorphic one-forms on the Riemann surface $R$ which satisfies , $\tau_{ij}$ is defined by and $$\Lambda = \left( \begin{array}{cc} A&B\\C&D \end{array} \right) .$$ For details see [@Fa], Chapter II. Let $\omega$ be a normalized symmetric bidifferential for ${{{\mathfrak F}}}$. In the formal neighbourhood $\eta_j$ we define the quadratic differential ([*projective connection*]{}) $$\label{projectiveconnection} S_{\omega}d\eta_j^2 = 6 \lim_{\overset{\eta {\mathop{\fam0 \rightarrow}\nolimits}\eta_j}{\xi {\mathop{\fam0 \rightarrow}\nolimits}\eta_j}} \left( \omega(\eta,\xi) - \frac{d\eta d\xi}{(\eta -\xi)^2}\right) .$$ We define $$a_\omega : \mathcal{L}({{{\mathfrak F}}}) \rightarrow {{{\Cal O}_{\Cal B}}}$$ as an ${{{\Cal O}_{\Cal B}}}$-module homomorphism by $$a_\omega(\vec \ell) = - \frac{c_v}{12} \sum_{j=1}^N {\operatornamewithlimits{Res}}_{\eta_j =0} (\ell_j(\eta_j) S_\omega(\eta_j) d\eta_j)$$ for all $\vec \ell \in \mathcal{L}({{{\mathfrak F}}})$. \[conomega\] For each $b\in \mathcal B$ and each element $X \in (\Theta_{\mathcal{B}})_b$, there is an element $\vec \ell \in \mathcal{L}({{{\mathfrak F}}})_b$ with $\theta(\vec \ell) = X$. Define an operator $\nabla_X^{(\omega)}$ acting on ${{\Cal V}_{\vec \lambda}^{\dagger}}({{{\mathfrak F}}})_b$ from the left by $$\nabla_X^{(\omega)}(\langle \Phi|) = D(\vec \ell)([\langle\Phi|) + a_\omega(\vec \ell)([\langle \Phi|).$$ \[holomorphicconnection\] $\nabla^{(\omega)}$ is a well-defined holomorphic connection in ${{\Cal V}_{\vec \lambda}^{\dagger}}({{{\mathfrak F}}})$. In [@Ue2] we introduced the holomorphic connection on ${{{\Cal V}_{\vec \lambda}}}({{{\mathfrak F}}})_b$ and our connection $\nabla^{(\omega)}$ is its dual connection (see Proposition 5.1.4 in [@Ue2]). We define a connection in the vector bundle ${{\Cal V}_{\vec \lambda}^{\dagger}}({{{\mathfrak F}}})$ over $\mathcal B$ by letting its $(1,0)$-part be given by the holomorphic connection $\nabla^{(\omega)}$ just defined and its $(0,1)$-part be given by the $\bar \partial$-operator determined by the holomorphic structure on ${{\Cal V}_{\vec \lambda}^{\dagger}}({{{\mathfrak F}}})$. We also denote this connection $\nabla^{(\omega)}$. \[curvaturecom\] The curvature of the connection $\nabla^{(\omega)}$ is given by $$R^\omega(X,Y) = \left\{ -a_\omega(\vec{n}) + X(a_\omega(\vec{m})) - Y(a_\omega(\vec{l})) - \frac{c_v}{12}\sum_{j=1}^N {\operatornamewithlimits{Res}}_{\xi_j=0}\left( \frac{d^3 l_j}{d\xi_j^3} m_jd\xi_j \right)\right\} \otimes {\mathop{\fam0 Id}\nolimits}$$ where $\vec \ell$ and $\vec m$ are liftings of $X$ and $Y$ to $\mathcal{L}({{{\mathfrak F}}})$, i.e. $\theta(\vec \ell) = X$ and $\theta(\vec m) = Y$, and $\vec{n} = [\vec{l}, \vec{m}]_d$ (see ). Hence we see that the connection is projectively flat and the curvature is of type $(2,0)$. It follows from the definition of the connection in the above definition that the $(1,1)$ and $(0,2)$-part of the curvature vanishes. $\square$ The definition of the connection in the bundle of vacua over Teichmüller space {#New5.3} ------------------------------------------------------------------------------ Suppose we have two [good]{}families ${{{\mathfrak F}}}_i$, $i=1,2$ of pointed Riemann Surfaces with for mal neighbourhoods, with the property that they have the same image $\Psi_{{{{\mathfrak F}}}_1} ({\mathcal B}_1) = \Psi_{{{{\mathfrak F}}}_2}({\mathcal B}_2)$ in Teichmüller space ${\mathop{\fam0 {\mathcal T}}\nolimits}_{({\Sigma},P)}$. \[contransf\] Let $\nabla^{(\omega)}_i$ be the connection in ${{\Cal V}_{\vec \lambda}^{\dagger}}({{{\mathfrak F}}}_i)$. Then we have that $$G_{12}^*(\nabla^{(\omega)}_2) = \nabla^{(\omega)}_1.$$ Since the connection is descended from the $\mathcal H$-level and $G_{12}$ is also descended from this level, we just need to check the transformation rule on this level. Up on the $\mathcal H$-level it follows straight from Theorem \[thm6.1\] (3). $\square$ \[conTeich\] Let $({\Sigma},P,\l)$ be a closed oriented stable and [saturated]{}marked surface and let $(\vec \alpha, \vec \beta) = (\alpha_1, \ldots, \alpha_g,\beta_1,\ldots,\beta_g)$ be a symplectic basis of $H_1({\Sigma}, {{\mathbb Z}}{})$. There is a unique connection $\nabla^{(\vec \alpha, \vec \beta)} = \nabla^{(\vec \alpha, \vec \beta)}({\Sigma},P)$ in the bundle ${{{\Cal V}_{\vec \lambda}^{\dagger}}}({\Sigma},P)$ over ${\mathop{\fam0 {\mathcal T}}\nolimits}_{({\Sigma},P)}$ with the property that for any [good]{}family ${{{\mathfrak F}}}$ of stable pointed Riemann surfaces with formal neighbourhoods on $({\Sigma},P)$ we have that $$\Psi_{{{{\mathfrak F}}}}^*(\nabla^{(\vec \alpha, \vec \beta)}) = \nabla^{(\omega)}.$$ In particular the connection is holomorphic and projectively flat with $(2,0)$-curvature as described in Theorem \[curvaturecom\]. If we act on the symplectic basis $(\vec \alpha,\vec \beta)$ by an element $\Lambda = \left(\begin{array}{cc}A & B\\C & D\end{array}\right) \in \text{Sp} (g,{{\mathbb Z}}{})$ by , then we have $$\nabla^{\Lambda(\vec \alpha, \vec \beta)} - \nabla^{(\vec \alpha, \vec \beta)} = - \frac{c_v}{2} \Pi^*( d \log \det (C \tau + D )),\label{contchbasis}$$ where $\Pi$ is the period mapping of holomorphic one-forms form the base space of $\mathfrak{F}$ to the Siegel upper-half plane of degree $g$. If $f : ({\Sigma}_1,P_1,\l_1) {\mathop{\fam0 \rightarrow}\nolimits}({\Sigma}_2,P_2,\l_1)$ is an orientation preserving diffeomorphism of labeled pointed surfaces which maps the symplectic basis $(\vec \alpha^{(1)},\vec \beta^{(1)})$ of $H_1({\Sigma}_1,{{\mathbb Z}}{})$ to the symplectic basis $(\vec \alpha^{(2)},\vec \beta^{(2)})$ of $H_1({\Sigma}_2,{{\mathbb Z}}{})$ then we have that $$\Spofv(f)^*(\nabla^{(\vec \alpha^{(2)},\vec \beta^{(2)})}) = \nabla^{(\vec \alpha^{(1)}, \vec \beta^{(1)})}.$$ The existence of the connection is a consequence of Lemma \[contransf\]. The transformation law (\[contchbasis\]) is proved in section 5.2 in [@Ue2]. $\square$ \[Teichpullbackcon\] For any [stable]{}and [saturated]{}family ${{\tilde {\mathfrak F}}}$ of pointed Riemann surfaces with formal neighbourhoods on $({\Sigma},P)$ the preferred isomorphism $$\Upsilon_{{{\tilde {\mathfrak F}}}} : {{\Cal V}_{\vec \lambda}^{\dagger}}({{\tilde {\mathfrak F}}}) {\mathop{\fam0 \rightarrow}\nolimits}\Psi_{{{\tilde {\mathfrak F}}}}^*{{{\Cal V}_{\vec \lambda}^{\dagger}}}({\Sigma},P)$$ given in Proposition \[Teichpullback\] preserves connections. This follows directly from Lemma \[contransf\]. $\square$ \[Rplusact\] The ${\mathop{\fam0 {\mathbb R}_+^{P}}\nolimits}$-action on ${\mathop{\fam0 {\mathcal T}}\nolimits}_{({\Sigma},P)}$ lifts by the use of the connection $\nabla^{(\vec \alpha, \vec \beta)}$ to an action ${{{\Cal V}_{\vec \lambda}^{\dagger}}}$ of ${\mathop{\fam0 {\mathbb R}_+^{P}}\nolimits}$ on ${{{\Cal V}_{\vec \lambda}^{\dagger}}}({\Sigma},P)$ which preserves the connection $\nabla^{(\vec \alpha, \vec \beta)}$. Since the ${\mathop{\fam0 {\mathbb R}_+^{P}}\nolimits}$ action on the formal coordinates is just obtained by scaling the coordinates by positive scalars, we get a well defined homomorphism from ${\mathop{\fam0 {\mathbb R}_+^{P}}\nolimits}$ to the group of formal coordinates changes for any family of $N$-pointed Riemann surfaces with formal coordinates, hence by composing with the group homomorphism $G$ we get an action of ${\mathop{\fam0 {\mathbb R}_+^{P}}\nolimits}$ on ${{{\Cal V}_{\vec \lambda}^{\dagger}}}({\Sigma},P)$. Note that we have here used Theorem 3.2.4. (2) for $p=0$ in [@Ue2], but only for these special real coordinates changes. This action preserved the connection by Theorem 3.2.4. (3) in [@Ue2]. $\square$ Let now ${{\mathbf \Sigma}}_i$, $i=1,2$ be marked surface and let $f_j : ({\Sigma}_1,P_1) {\mathop{\fam0 \rightarrow}\nolimits}({\Sigma}_2,P_2)$, $j=1,2$ be diffeomorphisms of pointed surfaces, which induce the same morphism of marked surfaces from ${{\mathbf \Sigma}}_1$ to ${{\mathbf \Sigma}}_2$. Then there exists a unique $v\in {\mathop{\fam0 {\mathbb R}_+^{P}}\nolimits}$ such that $v \cdot d_{P_1}f_1 = d_{P_2}f_2$. \[momswd\] We have a commutative diagram $$\begin{CD} \Spofvlamone({\Sigma}_1,P_1) @> \Spofv(f_1) >> \Spofvlamtwo({\Sigma}_2,P_2)\\ @V = VV @VV\Spofvlamtwo(v)V\\ \Spofvlamone({\Sigma}_1,P_1) @> \Spofv(f_2) >> \Spofvlamtwo({\Sigma}_2,P_2), \end{CD}$$ By the construction of $\Spofvlami({\Sigma}_i,P_i)$ we just need to check the commutativity on the $\mathcal H$-level, where this just amounts to $G$ being a homomorphism, which again is the content of Theorem 3.2.4. (2) in [@Ue2]. $\square$ Let $({\Sigma}, P,\l)$ be a general labeled pointed surface which might not be [stable]{}or [saturated.]{}Let now $(\tilde{P},\tilde{\l})$ be obtained from $(P,\l)$ by labeling further points not in $P$ by $0\in P_\ell$ such that $(\tilde{{\Sigma}},\tilde{P},\tilde{\l})$ is a [stable]{}and [saturated]{}labeled pointed surface. Let $\tilde{\pi} : {\mathop{\fam0 {\mathcal T}}\nolimits}_{({\Sigma},\tilde{P})}{\mathop{\fam0 \rightarrow}\nolimits}{\mathop{\fam0 {\mathcal T}}\nolimits}_{({\Sigma},P)}$ be the natural projection map. \[propvaccon\] The connection $\nabla^{(\vec \alpha, \vec \beta)} = \nabla^{(\vec \alpha, \vec \beta)}({\Sigma},P)$ is flat with trivial holonomy when restricted to any of the fibers of $\tilde{\pi} : {\mathop{\fam0 {\mathcal T}}\nolimits}_{({\Sigma},\tilde{P})}{\mathop{\fam0 \rightarrow}\nolimits}{\mathop{\fam0 {\mathcal T}}\nolimits}_{({\Sigma},P)}$. The propagation of vacua isomorphism applied along the fibers of $\tilde \pi$ is compatible with the connection. In the notation of chapter 5 of [@Ue2] this is seen by the following calculation. Let $\tilde N = |\tilde P|$. Let ${{{\mathfrak F}}}$ be a family of $\tilde N$-pointed curves with formal neighbourhoods on $(\Sigma,\tilde P)$ such that $\Psi_{{{{\mathfrak F}}}}(\mathcal B)$ is contained in a fiber of $\tilde \pi$. Further we can assume that when we consider ${{{\mathfrak F}}}$ over $(\Sigma,P)$, then it is simply a product family of a fixed $N$-pointed curve with formal neighbourhoods crossed with the base $\mathcal B$, i.e. we only vary the coordinates and the points in $\tilde P - P$. Then for any tangent field $X$ on $\mathcal B$, we choose a corresponding $\vec \ell = (\ell_1, \ldots, \ell_N, \ell_{N+1}, \ldots, \ell_{\tilde N})\in \mathcal L({{{\mathfrak F}}})$, such that $\ell_j = 0$, $j=1, \ldots, N$ and $\ell_j \in {\mathop{\fam0 {\mathbb C}^{}}\nolimits}[[\xi]]$, $j=N+1, \ldots, \tilde N$. Then for a section of $\mathcal H_{\tilde \lambda}(\mathcal B)$ of the form $F\otimes|\Phi\rangle\otimes |0\rangle\otimes \ldots \otimes |0\rangle$ we compute that $$\begin{aligned} \nabla^{(\omega)}_X(F |\Phi\rangle\otimes |0\rangle\otimes \ldots \otimes |0\rangle) &=& X(F) |\Phi\rangle\otimes |0\rangle\otimes \ldots \otimes |0\rangle\\ && - F \sum_{j=N+1}^{\tilde N} \rho_{j}(T[\ell_j]) (|\Phi\rangle\otimes |0\rangle\otimes \ldots \otimes |0\rangle) \\ && - a_\omega(\vec \l)F |\Phi\rangle\otimes |0\rangle\otimes \ldots \otimes |0\rangle\\ &=& X(F) |\Phi\rangle)\otimes |0\rangle\otimes \ldots \otimes |0\rangle.\end{aligned}$$ Here we have used that $\rho_{j}(T[\ell_j])(|0\rangle)=0$ because $\ell_j\in {\mathop{\fam0 {\mathbb C}^{}}\nolimits}[[\xi]]$ and $a_\omega(\vec \l)=0$ for the same reason. Hence we see that $\nabla^{(\omega)}$ is the trivial connection in $\mathcal H_{\tilde \lambda}(\mathcal B)$. Hence the connection is flat with trivial holonomy on the subbundle ${{\mathcal V}_{{\mathbf {\tilde \lambda}}}^{\dagger}}({{{\mathfrak F}}}) \subset {{\mathcal H}_{{\mathbf {\tilde \lambda}}}^{\dagger}}(\mathcal B)$. $\square$ Let $({\Sigma},P,\lambda)$ be a general labeled pointed surface, i.e. $({\Sigma},P)$ might not be [stable]{}nor [saturated.]{}Let $({\Sigma},P',\lambda')$ and $({\Sigma},P'',\lambda'')$ be labeled marked surfaces obtained from $({\Sigma},P,\lambda)$ by labeling further points not in $P$ by $0\in P_\ell$. Assume that $({\Sigma},P')$ and $({\Sigma},P'')$ are [stable]{}and [saturated]{}pointed surfaces. Let $\bar P = P' \cup P''$ and $\bar \l$ be the induced labeling of $\bar P$. Note that $({\Sigma},\bar P)$ is also [stable]{}and [saturated.]{} \[pullbackcon\] Let $(\vec \alpha, \vec \beta)$ be a symplectic basis of $H_1({\Sigma},{{\mathbb Z}}{})$. The isomorphisms given in Theorem \[propvaciso\] satisfies $$\nabla^{(\vec \alpha, \vec \beta)}({\Sigma},\bar P) = (\pi')^*\nabla^{(\vec \alpha, \vec \beta)}({\Sigma},P') = (\pi'')^*\nabla^{(\vec \alpha, \vec \beta)}({\Sigma},P'').$$ We only have to consider the case of adding one point to a [stable]{}and [saturated]{}curve, i.e. say $\bar P = P'$ and $P'$ is obtained from $P''$ by adding one more point. Let $\bar N$ be the number of points in $\bar P$ and ${{{\mathfrak F}}}$ be a family of $\bar N$-pointed curves with formal neighbourhoods on $(\Sigma, \bar P)$. For any tangent field $X$ on $\mathcal B$, we choose a corresponding $\vec \ell = (\ell_1, \ldots, \ell_{\bar N}, \ell_{\bar N+1})\in \mathcal L({{{\mathfrak F}}})$, such that $\ell_{\bar N+1} \in {\mathop{\fam0 {\mathbb C}^{}}\nolimits}[[\xi]]$. Then the same computation as above shows that $$\nabla^{(\omega)}(F |\Phi\rangle\otimes |0\rangle ) = \nabla^{(\omega)}(F |\Phi\rangle)\otimes |0\rangle.$$ The Proposition follows directly from this. $\square$ Let now $(\Sigma_i,P'_i,\lambda'_i)$ and $(\Sigma_i,P''_i,\lambda''_i)$ be [stable]{}and [saturated]{}labeled pointed surfaces, obtained from the labeled pointed surfaces $(\Sigma_i,P_i,\lambda_i)$, by labeling further points with the zero-label. Assume that $f' : ({\Sigma}_1,P_1',\l_1') {\mathop{\fam0 \rightarrow}\nolimits}({\Sigma}_2,P_2',\l_2')$ and $f'' : ({\Sigma}_1,P_1'',\l_1'') {\mathop{\fam0 \rightarrow}\nolimits}({\Sigma}_2,P_2'',\l_2'')$ are diffeomorphisms which induce isotopic maps from $({\Sigma}_1,P_1)$ to $({\Sigma}_2,P_2)$, where the isotopy is through maps which induces the same map from $PT_{P_1}{\Sigma}_1$ to $PT_{P_2}{\Sigma}_2$. \[propvmorph\] With respect to the propagation of vacua isomorphisms, we get that $$(\pi')^* \Spofv(f') = (\pi'')^* \Spofv(f'').$$ This follows directly from the way the morphisms for $f'$ and $f''$ are defined. Definition of the space of [abelian]{}vacua associated to a Riemann surface. {#New3ab} ============================================================================ Fermion Fock space ------------------ Let $\hZ$ be the set of all half integers. Namely $$\hZ = \{ n + 1/2\, | \, n \in \bZ \,\}.$$ Let $\cWd$ be an infinite-dimensional vector space over $\bC$ with a filtration $\{F^m\cWd\}_{m \in \bZ}$ which satisfies the following conditions. 1. The filtration $\{F^m\cWd\}$ is decreasing; 2. $\bigcup_{m \in \bZ} F^m\cWd = \cWd$, $\bigcap_{m \in \bZ} F^m\cWd = \{0\}$; 3. $\dim_\bC F^m\cWd/ F^{m+1}\cWd = 1$; 4. The vector space $\cWd$ is complete with respect to the uniform topology such that $\{F^m\cWd\}$ is a basis of open neighbourhoods of 0. We introduce a basis $\{e^\nu \}_{\nu \in \hZ}$ of $\cWd$ in such a way that $$e^{m+1/2} \in F^m\cWd \setminus F^{m+1}\cWd.$$ Then, each element $ u \in \cWd$ can uniquely be expressed in the form $$u= \sum_{\nu> n_0, \nu \in \hZ}^\infty a_\nu e^\nu$$ for some $n_0$ and with respect to this basis the filtration is given by $$F^m\cWd = \left\{ u \in \cWd \, \left| \, u= \sum_{\nu> m, \nu \in \hZ}^\infty a_\nu e^\nu\, \right. \right\}.$$ We fix the basis $\{e^\nu\}_{\nu \in \hZ}$ throughout the present paper. Let $\bC((\xi))$ be a field of formal Laurent series over the complex number field. Then the basis gives us a filtration preserving linear isomorphism $$\begin{aligned} \bC((\xi)) & \cong & \cWd \\ \xi^n &\mapsto &e^{n+1/2}.\end{aligned}$$ By mapping $\xi^n d\xi$ to $e^{n+1/2}$ we of course also get a filtration preserving linear isomorphism between $\bC((\xi))d\xi$ and $\cWd$. We let $\{\oe_\nu\}_{\nu \in \hZ}$ be the dual basis of $\{e^\nu\}_{\nu \in \hZ}$. Then, put $$\cW = \bigoplus_{\nu \in \hZ} \bC\oe_\nu$$ Then $\cW$ is the topological dual of the vector space $\cWd$. There is a natural pairing $(\phantom{X}|\phantom{X}) : \cWd\times \cW \rightarrow \bC$ defined by $$(e^\nu|\oe_\mu) = \delta^\nu_\mu.$$ In other word we have $$(u|v) = v(u).$$ Let us introduce the semi-infinite exterior product of the vector spaces $\cW$ and $\cWd$. For that purpose we first introduce the notion of a Maya diagram. A Maya diagram $M$ of the charge $p$, $p \in \bZ$ is a set $$M = \left\{ \mu(p-1/2), \mu(p-3/2), \mu(p-5/2), \ldots\right\},$$ where $\mu$ is an increasing function $$\mu : \hZ_{<p} = \{ \nu \in \hZ\, | \, \nu < p\,\} \rightarrow \hZ$$ such that there exists an integer $n_0$ such that $$\mu(\nu) = \nu$$ for all $\nu < n_0$. The function $\mu$ is called the [*characteristic function*]{} of the Maya diagram $M$. The set of Maya diagrams of charge $p$ is written as ${\mathcal M}_p$. For a Maya diagram M we have $\mu(\nu) = \nu$ for almost all $\nu$. Therefore the set $$\{ \mu(\nu) - \nu \,|\, \nu \in \hZ, \, \mu(\nu) - \nu >0\,\}$$ is finite and the number $$d(M) = \sum_{\nu \in \hZ} (\mu(\nu) - \nu)$$ is finite. The number $d(M)$ is also written as $d(\mu)$ and it is called the [*degree*]{} of the Maya diagram $M$ with characteristic function $\mu$. The finite set of Maya diagrams of degree $d$ and change $p$ is denoted ${\mathcal M}_p^d$. Clearly ${\mathcal M}_p = \coprod_{d} {\mathcal M}_p^d$. For a Maya diagram $M$ of charge $p$ we define two semi-infinite products $$\begin{aligned} |M\rangle &= & \oe_{\mu(p-1/2)}\wedge \oe_{\mu(p-3/2)} \wedge \oe_{\mu(p-5/2)} \wedge \cdots \\ \langle M| &=& \cdots \wedge e^{\mu(p-5/2)}\wedge e^{\mu(p-3/2)} \wedge e^{\mu(p-1/2)}\end{aligned}$$ Formally, these semi-infinite products is just another notation for the corresponding Maya diagram. This notation is particular convenient for the following discussion. However, by using the basis $e_\nu$, we clear indicate the relation to the vector spaces $\cW$ and $\cWd$. For any integer $p$ put $$\begin{aligned} |p \rangle &= & \oe_{p-1/2}\wedge \oe_{p-3/2} \wedge \oe_{p-5/2} \wedge \cdots \\ \langle p| &=& \cdots \wedge e^{p-5/2}\wedge e^{p-3/2} \wedge e^{p-1/2}\end{aligned}$$ Now the [*fermion Fock space*]{} $\cFd(p)$ of [*charge*]{} $p$ and the [*dual fermion Fock space*]{} $\cF(p)$ of [*charge*]{} $p$ are defined by $$\begin{aligned} \cF(p) &=& \bigoplus_{M \in {\mathcal M}_p} \bC |M\rangle \\ \cFd(p) &=& \prod_{M \in {\mathcal M}_p} \bC \langle M|\end{aligned}$$ We observe that $$\cF(p) = \bigoplus_{d\geq 0} \cF_d(p),$$ where $$\cF_d(p) = \bigoplus_{M\in {\mathcal M}_p^d} \bC |M\rangle.$$ The dual pairing $$\langle \cdot | \cdot \rangle : \cFd(p) \times \cF(p) \rightarrow \bC$$ is given by $$\langle M| N \rangle = \delta_{M,N}, \quad M, N \in {\mathcal M}_p$$ Put also $$\begin{aligned} \cF &=& \bigoplus_{p \in \bZ} \cF(p) \\ \cFd &=& \bigoplus_{p \in \bZ} \cFd(p)\end{aligned}$$ The vector space $\cFd$ is called the [*fermion Fock space*]{} and $\cF$ is called the [*dual fermion Fock space*]{}. These are the semi-infinite exterior products of the vector spaces $\cWd$ and $\cW$ respectively, which we shall be interested in. We only define the fermion Fock space by using the basis $e_\nu$, since we are fixing this basis throughout. The above pairing can be extended to the one on $\cFd \times \cF$ by assuming that the paring is zero on $\cFd(p) \times \cF(p')$ if $p \ne p'$. Let us introduce the [*fermion operators*]{} $\psi_\nu$ and $\ovpsi_\nu$ for all half integers $\nu \in \hZ$ which act on $\cF$ from the left and on $\cFd$ from the right. $$\begin{aligned} \hbox{\rm Left action on $\cF$} &\quad & \psi_\nu = i(\oe_\nu), \quad \ovpsi_\nu = \oe_{-\nu} \wedge \\ \hbox{\rm Right action on $\cFd$} &\quad & \psi_\nu = \wedge e^\nu, \quad \ovpsi_\nu = i(e^{-\nu})\end{aligned}$$ where $i(\cdot)$ is the interior product. For example we have $$\begin{aligned} \psi_{-3/2}|0\rangle &=& i(\oe_{-3/2}) \oe_{-1/2}\wedge \oe_{-3/2}\wedge \cdots = - \oe_{-1/2}\wedge\oe_{-5/2}\wedge \oe_{-7/2}\wedge \cdots , \\ \langle 0 | \ovpsi_{5/2} &=& \cdots \wedge e^{-5/2}\wedge e^{-3/2} \wedge e^{-1/2} i(e^{-5/2}) =\cdots \wedge e^{-7/2}\wedge e^{-3/2} \wedge e^{-1/2}\end{aligned}$$ Note that $\psi_\nu$ maps $\cF(p)$ to $\cF(p-1)$, hence decreases the charge by one, and $\ovpsi_\nu$ maps $\cF(p)$ to $\cF(p+1)$, hence increase the charge by one. Similarly the right action of $\psi_\nu$ maps $\cFd(p)$ to $\cFd(p+1)$ and $\ovpsi_\nu$ maps $\cFd(p)$ to $\cFd(p-1)$. It is easy to show that for any $\langle u| \in \cF$ and $|v\rangle \in \cFd$ we have $$\langle u | \psi_\nu v\rangle = \langle u \psi_\nu | v\rangle, \quad \langle u | \ovpsi_\nu v\rangle = \langle u \ovpsi_\nu | v\rangle .$$ The fermion operators have the following anti-commutation relations as operators on $\cF$ and $\cFd$. $$\begin{aligned} {[\psi_\nu, \psi_\mu ]}_+ &=& 0, \label{anticomm1}\\ {[\ovpsi_{\nu}, \ovpsi_{\mu} ]}_+ &=& 0, \label{anticomm2}\\ {[\psi_\nu, \ovpsi_{\mu} ]}_+ &=& \delta_{\nu +\mu, 0}, \label{anticomm3}\end{aligned}$$ where we define $${[A,B]}_+ = AB+BA.$$ Note that for each Maya diagram $M$ of charge $p$ we can find non-negative half integers $$\mu_1<\mu_2<\cdots<\mu_r<0, \quad \nu_1<\nu_2<\cdots<\nu_s<0, \quad r\ge 0, \,s\ge 0$$ with $r- s=p$ and $\mu_i \neq \nu_j$ such that $$\label{maya} |M\rangle = (-1)^{\sum_{i=1}^s\nu_i +s/2} \ovpsi_{\mu_1}\ovpsi_{\mu_2}\cdots\ovpsi_{\mu_r} \psi_{\nu_s}\psi_{\nu_{s-1}}\cdots\psi_{\nu_1}|0\rangle.$$ The negative half integers $\mu_i$’s and $\nu_j$’s are uniquely determined by the Maya diagram $M$. The [*normal ordering*]{} $\normalord\phantom{X}\normalord$ of the fermion operators are defined as follows. $$\normalord A_\nu B_\mu \normalord = \left\{ \begin{array}{ll} - B_\mu A_\nu & \hbox{\rm if $\mu <0$ and $\nu>0$}, \\ A_\nu B_\mu & \hbox{\rm otherwise,} \end{array} \right.$$ where $A$ and $B$ is $\psi$ or $\ovpsi$. By , and the normal ordering is non-trivial if and only if $\mu <0$ and $A_\nu ={\bar B}_{-\mu} $. The field operators $\psi(z)$ and $\ovpsi(z)$ are defined by $$\begin{aligned} \psi(z)&=& \sum_\mu \psi_\mu z^{-\mu -1/2} , \\ \overline{\psi}(z) &=& \sum_\mu \overline{\psi}_\mu z^{-\mu -1/2} .\end{aligned}$$ The current operator $J(z)$ is defined by $$J(z) = \normalord \overline{\psi}(z)\psi(z) \normalord = \sum_{n \in \bZ} J_n z^{-n-1}$$ Note that thanks to the normal ordering, the operator $J_n$ can operate on $\cF$ and $\cFd$ even though $J_n$ is an infinite sum of operators. The energy-momentum tensor $T(z)$ is defined by $$\label{emab} T(z) = \normalord \frac{d \psi(z)}{dz} \ovpsi(z) \normalord = \sum_{n \in \bZ} L_n z^{-n-2}.$$ Again due to the normal ordering, the coefficients $L_n$ operates on $\cF$ and $\cFd$. The set $\{L_n\}_{n \in \bZ}$ forms the [*Virasoro*]{} algebra with central charge $c=-2 $. The field operators $\psi(z)$ and $\ovpsi(z)$, the current operator $J(z)$ and the energy-momentum tensor $T(z)$ form the so-called spin $j=0$ $bc$-system or ghost system in the physics literature. Abelian Vacua ------------- For a positive integer $N$ put $$\begin{aligned} \cF_N &=& \bigoplus_{p_1, \ldots,p_N \in\bZ} \cF(p_1) \otimes \cdots \otimes \cF(p_N) ,\\ \cFd_N &=& \bigoplus_{p_1, \ldots, p_N \in \bZ} {\cFd}(p_1) \hat{\otimes} \cdots \hat{\otimes} {\cFd}(p_N),\end{aligned}$$ where $\hat{\otimes}$ means the complete tensor product. \[dfn3.1\] Let ${\mathfrak X} = (C; q_1,q_2, \ldots, q_N; \xi_1, \xi_2, \ldots, \xi_N)$ be a [saturated]{}pointed Riemann Surface with formal neighbourhoods . The abelian vacua (ghost vacua in \[2\]) ${{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}(\gX)$ of the spin $j=0$ ghost system associated to $\mathfrak{X}$ is a linear subspace of $\cFd_N$ consisting of elements $\langle \Phi|$ satisfying the following conditions: 1. For all $|v\rangle \in \cF_N$, there exists a meromorphic function $f \in H^0(C,\cO_C(*\sum_{j=1}^Nq_j))$ such that $\langle \Phi| \rho_j(\psi(\xi_j))|v\rangle$ is the Laurent expansion of $f$ at the point $q_j$ with respect to the formal coordinate $\xi_j$; 2. For all $|v\rangle \in \cF_N$, there exists a meromorphic one-form $\omega \in H^0(C,\omega_C(*\sum_{j=1}^Nq_j))$ such that $\langle \Phi| \rho_j(\ovpsi(\xi_j))|v\rangle d\xi_j$ is the Laurent expansion of $\omega$ at the point $q_j$ with respect to the coordinates $\xi_j$, where $\rho_j(A)$ means that the operator $A$ acts on the $j$-th component of $\cF_N$ as $$\rho_j(A)|u_1 \otimes u_2 \otimes \cdots \otimes u_N\rangle = (-1)^{p_1+\cdots+p_{j-1}} |u_1 \otimes \cdots \otimes u_{j-1}\otimes Au_j \otimes u_{j+1} \otimes \cdots \otimes u_N\rangle.$$ We will reformulate the above two conditions into gauge conditions. For that purpose we introduce the following notation. For a meromorphic one-form $\omega \in H^0(C,\omega_C(*\sum_{j=1}^Nq_jj))$ we let $$\omega_j = (\sum_{n=-n_0}^\infty a_n \xi_j^n )d\xi_j$$ be the Laurent expansion at $q_j$ with respect to the coordinate $\xi_j$. Then, for the field operator $\psi(z)$ let us define $\psi[\omega_j]$ by $$\psi[\omega_j]= {\operatornamewithlimits{Res}}_{\xi_j=0}(\psi(\xi_j)\omega_j) = \sum_{n= -n_0}^\infty a_n\psi_{n+1/2}.$$ Similarly we can define $\ovpsi[\omega_j]$. For a meromorphic function $f \in H^0(C,\cO_C(*\sum_{j=1}^Nq_j))$ we let $f_j(\xi_j)$ be the Laurent expansion of $f$ at $q_j$ with respect to the coordinate $\xi_j$. For the field operator $\psi(z)$ define $\psi[f_j]$ by $$\psi[f_j] = {\operatornamewithlimits{Res}}_{\xi_j=0}(\psi(\xi_j)f_j(\xi_j)d\xi_j)$$ Put $$\begin{aligned} \psi[\omega] = (\psi[\omega_1], \ldots, \psi[\omega_N]) , && \ovpsi[\omega] = (\ovpsi[\omega_1], \ldots, \ovpsi[\omega_N]) \\ \psi[f]= (\psi[f_1], \ldots, \psi[f_N]), && \ovpsi[f]= (\ovpsi[f_1], \ldots, \ovpsi[f_N]).\end{aligned}$$ Then, these operate on $\cF_N$ from the left and on $\cFd_N$ from the right. For example, $\ovpsi[f]$ operates on $\cF_N$ from the left by $$\begin{aligned} \ovpsi[f] |u_1\otimes \cdots \otimes u_N \rangle & = & \sum_{j=1}^N\rho_j(\ovpsi[f_j])|u_1\otimes \cdots \otimes u_N \rangle \\ &= & \sum_{j=1}^N(-1)^{p_1+\cdots+p_{j-1}}|u_1 \otimes \cdots \otimes u_{j-1} \otimes \ovpsi[f_j]u_j \otimes u_{j+1}\otimes \cdots \otimes u_N\rangle\end{aligned}$$ for $|u_j\rangle \in \cFd(p_j)$ and operates on $\cFd_N$ from the right by $$\begin{aligned} \langle v_N \otimes \cdots v_1|\ovpsi[f] & = & \sum_{j=1}^N \langle v_N \otimes \cdots v_1|\rho_j(\ovpsi[f_j]) \\ &= & \sum_{j=1}^N(-1)^{p_1+\cdots+p_{j-1}} \langle v_N \otimes \cdots \otimes v_{j+1} \otimes v_j \ovpsi[f_j] \otimes v_{j-1}\otimes \cdots \otimes v_1 |\end{aligned}$$ for $\langle v_j| \in \cFd(p_j)$. \[thm3.1\] The element $\langle \Phi| \in \cFd_N$ belongs to the space of abelian vacua ${{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}(\gX)$ of the $j=0$ ghost system if and only if $\langle \Phi|$ satisfies the following two conditions. 1. $\langle\Phi|\psi[\omega] =0$ for any meromorphic one-form $\omega \in H^0(C, \omega_C(* \sum_{j=1}^Nq_j))$. 2. $\langle\Phi|\ovpsi[f] =0$ for any meromorphic function $f \in H^0(C,\cO_C(*\sum_{j=1}^Nq_j))$. The first (resp. second) condition in the above theorem is called the first (resp. second) gauge condition. The first and second gauge conditions can be rewritten in the following form: 1. $\sum_{j=1}^N(-1)^{p_1+\cdots+p_{j-1}} \langle \Phi| u_1 \otimes \cdots \otimes \cdots \otimes u_{j-1} \otimes \psi[\omega_j]u_j \otimes u_{j+1}\otimes \cdots \otimes u_N\rangle =0$ for any $\omega \in H^0(C, \omega_C(*\sum_{j=1}^Nq_j))$ and $|u_j\rangle \in \cF(p_j)$, $j=1,2,\ldots,N$. 2. $\sum_{j=1}^N(-1)^{p_1+\cdots+p_{j-1}} \langle \Phi| u_1 \otimes \cdots \otimes u_{j-1} \otimes \ovpsi[f_j]u_j \otimes u_{j+1}\otimes \cdots \otimes u_N\rangle =0$ for any $f \in H^0(C, \cO_C(*\sum_{j=1}^Nq_j))$ and $|u_j\rangle \in \cF(p_j)$, $j=1,2,\ldots,N$. It is easy to show that the abelian vacua ${{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}(\gX)$ is a finite dimensional vector space. More strongly we can prove the following theorem. \[thm3.2\] For any pointed Riemann surface $\gX= (C;Q_1,\ldots,Q_N;\xi_1,\ldots \xi_N)$ with formal neighbourhoods we have $$\dim_\bC {{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}(\gX) =1.$$ In the later application we need to consider a disconnected Riemann surface. The following proposition is an immediate consequence of the definition. \[prop3.1\] Let $$\gX_1=(C_1;q_1, \ldots,q_M;\xi_1, \ldots, \xi_M)$$ and $$\gX_2=(C_2;q_{M+1}, \ldots,q_M;\xi_{M+1}, \ldots, \xi_N)$$ be [saturated]{}pointed Riemann surfaces with formal neighbourhoods. Let $C$ be the disjoint union $C_1 \sqcup C_2$ of the Riemann surfaces $C_1$, $C_2$. Put $$\gX=(C; q_1, \ldots, q_N; \xi_1, \ldots,\xi_N).$$ Then we have $${{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}(\gX) = {{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}(\gX_1) \otimes {{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}(\gX_2).$$ Now we can introduce the dual abelian vacua. \[dfn3.2\] Let $\cF_{{\mathop{\fam0 ab}\nolimits}}(\gX)$ be the subspace of $\cF_N$ spanned by $\psi[\omega] \cF_N$, $\omega \in H^0(C,\omega_C(*\sum_{j=1}^Nq_j))$ and $\ovpsi[f] \cF_N$, $f \in H^0(C,\cO(*\sum_{j=1}^Nq_j))$. Put $$\cV_{{\mathop{\fam0 ab}\nolimits}}(\gX) = \cF_N/\cF_{{\mathop{\fam0 ab}\nolimits}}(\gX).$$ The quotients space $\cV_{{\mathop{\fam0 ab}\nolimits}}(\gX)$ is called the space of [*dual abelian vacua*]{} or [*dual ghost vacua*]{} of the $j=0$ ghost system. Since ${{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}(\gX)$ is finite dimensional, $\cV_{{\mathop{\fam0 ab}\nolimits}}(\gX)$ is dual to ${{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}(\gX)$. \[thm3.3\] The space of abelian vacua ${{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}(\gX)$ is isomorphic to the determinant of the canonical bundle $\omega_C$ $${{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}(\gX) \cong \det(H^0(C,\omega_C)).$$ Let $\gX=(C;q_1,\ldots,q_N;\xi_1,\ldots,\xi_N)$ be an $N$-pointed Riemann surface with formal neighbourhoods. Let $q_{N+1}$ be a non-singular point and choose a formal coordinate $\xi_{N+1}$ of $C$ with center $q_{N+1}$. Put $$\widetilde{\gX} = (C;q_1,\ldots,q_N,q_{N+1} ;\xi_1,\ldots,\xi_N, \xi_{N+1}).$$ Then the canonical linear mapping $$\begin{aligned} \iota : \cF_N & \rightarrow & \cF_{N+1} \\ |v\rangle & \mapsto & |v\rangle \otimes |0\rangle\end{aligned}$$ induces the canonical mapping $$\iota^* :\cFd_{N+1} \rightarrow \cFd_N.$$ \[thm3.4\] The canonical mapping $\iota^*$ induces an isomorphism $${{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}(\widetilde{\gX}) \cong {{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}(\gX) .$$ This isomorphism is denoted the “Propagation of abelian vacua” isomorphism. Let us consider change of formal neighbourhoods. We use the same notation as in §4.5. For any automorphism $h \in {\mathcal D}_+^0$ we can define the action $G[h]$ on the fermion Fock space $\mathcal F$ by using the energy-momentum tensor as in §4.5. \[prop6.1a\] For any $h_j \in {\mathcal D}_+^0$, $j=1,2, \ldots, N$ and $N$-pointed curve $$\mathfrak{X}= (C; Q_1,Q_2, \ldots, Q_n; \xi_1, \xi_2, \ldots, \xi_N)$$ with formal neighbourhoods, put $$\mathfrak{X}_{(h)} =(C; Q_1,Q_2, \ldots, Q_N; h_1(\xi_1), h_2(\xi_2), \ldots, h_N(\xi_N)).$$ Then, the isomorphism $G[h_1] \widehat{\otimes}\cdots \widehat{\otimes}G[h_N]$ $$\begin{aligned} \cFd_N & \rightarrow & \cFd_N \\ \langle \phi_1\widehat{\otimes} \cdots \widehat{\otimes}\phi_N| & \mapsto & \langle \phi_1G[h_1] \widehat{\otimes}\cdots \widehat{\otimes}\phi_NG[h_N]|\end{aligned}$$ induces the canonical isomorphism $$G[h_1] \widehat{\otimes}\cdots \widehat{\otimes}G[h_N]: {{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}(\gX) \rightarrow {{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}(\mathfrak{X}_{(h)})$$ The space of [abelian]{}vacua associated to a Riemann surface ------------------------------------------------------------- Let ${\mathbf C}$ be a compact Riemann surface. For a [stable]{}and [saturated]{}pointed Riemann surface with formal neighbourhoods ${{{\mathfrak X}}}$ we denote by ${\mathop{\fam0 {\tilde c}}\nolimits}({{{\mathfrak X}}})$ the underlying Riemann surface. Suppose we now have two pointed Riemann surfaces with formal neighbourhoods ${{{\mathfrak X}}}_i$ such that ${\mathop{\fam0 {\tilde c}}\nolimits}({{{\mathfrak X}}}_1) = {\mathop{\fam0 {\tilde c}}\nolimits}({{{\mathfrak X}}}_2)={\mathbf C}$. Choose for each component of ${\mathbf C}$ a point with a formal neighbourhood, which is not a point with formal neighbourhoods of ${{{\mathfrak X}}}_i$, i=1,2. Let ${{{\mathfrak X}}}_0$ be the resulting [stable]{}and [saturated]{}pointed Riemann surface with formal neighbourhoods (if ${{{\mathfrak X}}}_0$ is not stable, then add further points with formal neighbourhoods). Then iterations of the propagation of vacua isomorphism determined by the inclusion of $\mathcal F_N$ into $\mathcal F_{N+1}$ given by $|v\rangle \mapsto |v\rangle\otimes |0\rangle$, induces isomorphisms from ${{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}({{{\mathfrak X}}}_0)$ to ${{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}({{{\mathfrak X}}}_i)$, $i=1,2$. It is elementary to check that the resulting isomorphism from ${{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}({{{\mathfrak X}}}_1)$ to ${{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}({{{\mathfrak X}}}_2)$ is independent of ${{{\mathfrak X}}}_0$. Furthermore, we get from the commutativity of the following diagram $$\begin{CD} {\mathcal F}_1 @> = >> {\mathcal F}_1\\ @V {\mathop{\fam0 Id}\nolimits}\otimes |0\rangle VV @V {\mathop{\fam0 Id}\nolimits}\otimes |0\rangle VV\\ {\mathcal F}_2 @> {\mathop{\fam0 Id}\nolimits}\otimes G[h] >> {\mathcal F}_2, \end{CD}$$ which follows from the fact that $G[h] |0\rangle = |0\rangle$, that these isomorphisms are also compatible with the change of formal coordinates isomorphism induced by $G[h]$. \[Defspofvab\] The space of [abelian]{}vacua associated to the Riemann surface ${\mathbf C}$ is by definition $${{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}({\mathbf C}) = \coprod_{{\mathop{\fam0 {\tilde c}}\nolimits}({{{\mathfrak X}}}) = {\mathbf C}}{{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}({{{\mathfrak X}}})/\sim,$$ where the disjoint union is over all Riemann surfaces with formal neighbourhoods with ${\mathbf C}$ as the underlying Riemann surface and $\sim$ is the equivalence relation generated by the isomorphisms discussed above. It is obvious that \[Spofvisoab\] The natural quotient map from ${{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}({{{\mathfrak X}}})$ to ${{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}({\mathbf C})$ is an isomorphism for all Riemann surfaces with formal neighbourhoods ${{{\mathfrak X}}}$ with ${\mathop{\fam0 {\tilde c}}\nolimits}({{{\mathfrak X}}}) = {\mathbf C}$. Suppose ${\mathbf C}_i$, $i=1,2$ are Riemann surfaces and $\Phi : {\mathbf C}_1 {\mathop{\fam0 \rightarrow}\nolimits}{\mathbf C}_2$ is a morphism of labeled marked Riemann surfaces. Let ${{{\mathfrak X}}}_2$ be a Riemann surface with formal neighbourhoods such that ${\mathop{\fam0 {\tilde c}}\nolimits}({{{\mathfrak X}}}_2) = {\mathbf C}_2$. Let $\Phi^*{{{\mathfrak X}}}_2 = {{{\mathfrak X}}}_1$. Then $\Phi$ is a morphism of Riemann surfaces with formal neighbourhoods. We clearly have that \[morphspofvab\] The identity map on $\mathcal F_N$ induces a linear isomorphism from ${{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}({{{\mathfrak X}}}_1)$ to ${{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}({{{\mathfrak X}}}_2)$, which induces a well defined linear isomorphism ${{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}(\Phi)$ from ${{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}({\mathbf C}_1)$ to ${{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}({\mathbf C}_2)$. Compositions of morphisms of labeled marked Riemann surfaces go to compositions of the induced linear isomorphisms. Definition of the line bundle of [abelian]{}vacua over Teichmüller space {#New4ab} ======================================================================== Sheaf of abelian vacua ---------------------- Let $$\gF = (\pi : \cC \rightarrow \cB; s_1, \ldots, s_N; \xi_1, \ldots, \xi_N)$$ be a family of $N$-pointed semi-stable curves with formal neighbourhoods. That is $\cC$ and $\cB$ are complex manifolds, $\pi$ is a proper holomorphic mapping, and for each point $b \in \cB$, $\gF(b)= (C_b =\pi^{-1}(b); s_1(b), \ldots, s_N(b); \xi_1, \ldots,\xi_N)$ is an $N$-pointed semi-stable curve with formal neighbourhoods. We let $\Sigma$ be the locus of double points of the fibers of $\gF$ and let $D$ be $\pi(\Sigma)$. Note that $\Sigma$ is a non-singular submanifold of codimension two in $\cC$, and $D$ is a divisor in $\cB$ whose irreducible components $D_i$, $i = 1, 2, \dots ,m'$ are non-singular. In this section we use the following notation freely. $$S_j = s_j(\cB), \quad S = \sum_{j=1}^N S_j.$$ Put $$\cF_N(\cB) = \cF_N \otimes_\bC \cO_\cB, \quad \cFd_N(\cB)= \cO_\cB \otimes_\bC \cFd_N.$$ \[dfn4.1\] We define the subsheaf ${{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}(\gF)$ of $\cFd_N(\cB)$ by the gauge conditions: $$\begin{aligned} &&\sum_{j=1}^N \langle \Phi | \psi[\omega_j] = 0 , \quad \hbox{\rm for all $\omega \in \pi_*(\omega_{\cC/\cB}(*S))$}, \\ &&\sum_{j=1}^N \langle \Phi | \ovpsi[f_j] = 0, \quad \hbox{\rm for all $f \in \pi_*\cO_\cC(*S)$} .\end{aligned}$$ where $\omega_j$ and $f_j$ are the Laurent expansion of $\omega$ and $f$ along $S_j$ with respect to the coordinate $\xi_j$. The sheaf ${{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}(\gF)$ is called the sheaf of ($j=0$) abelian vacua or the sheaf of abelian vacua of the family $\gF$. Similarly the sheaf $\cV_{{\mathop{\fam0 ab}\nolimits}}(\gF)$ of ($j=0$) dual abelian vacua of the family is defined by $$\cV_{{\mathop{\fam0 ab}\nolimits}}(\gF)= \cF_N(\cB)/\cF_{{\mathop{\fam0 ab}\nolimits}}(\gF).$$ where $\cF_{{\mathop{\fam0 ab}\nolimits}}(\gF)$ is the $\cO_\cB$-submodule of $\cF_N(\cB)$ given by $\cF_{{\mathop{\fam0 ab}\nolimits}}(\gF) = \cF^0_{{\mathop{\fam0 ab}\nolimits}}(\gF) + \cF^1_{{\mathop{\fam0 ab}\nolimits}}(\gF)$, where $\cF^0_{{\mathop{\fam0 ab}\nolimits}}(\gF)$ is the span of $\ovpsi[f]\cF_N(\cB)$ for all $f \in \pi_*\cO_\cC(*S)$ and $\cF^1_{{\mathop{\fam0 ab}\nolimits}}(\gF)$ is the span of $\psi[\omega]\cF_N(\cB)$ for all $\omega \in \pi_*\omega_{\cC/\cB}(*S)$. Note that we have $${{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}(\gF) = \underline{{{\hbox{\rm Hom}}}}_{\cO_\cB}(\cV_{{\mathop{\fam0 ab}\nolimits}}(\gF), \cO_\cB).$$ Moreover, by the right exactness of the tensor product we have that $$\label{4.1} \cV_{{\mathop{\fam0 ab}\nolimits}}(\gF)\otimes_{\cO_\cB} \cO_{\cB, b}/\mathfrak{m}_b \cong \cV_{{\mathop{\fam0 ab}\nolimits}}(\gF(b)).$$ \[thm5.2a\] The sheaves $\cV_{{\mathop{\fam0 ab}\nolimits}}(\gF)$ and ${{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}(\gF)$ are invertible $\cO_\cB$-modules. They are dual to each other. The line bundle of [abelian]{}vacua over Teichmüller space ---------------------------------------------------------- Let ${{{\mathfrak F}}}_i$, $i=1,2$ be two families of [stable]{}and [saturated]{}pointed Riemann surfaces with formal neighbourhoods. Assume we have a morphism of families (not necessarily preserving sections nor formal coordinates) $$\begin{CD} {\mathcal C}_1 @> \Phi >> {\mathcal C}_2\\ @V VV @VVV\\ {\mathcal B}_1 @> \Psi >> {\mathcal B}_2, \end{CD}$$ which is a fiberwise biholomorphism. Let now ${{{\mathfrak F}}}_0 = (\mathcal C_1 {\mathop{\fam0 \rightarrow}\nolimits}\mathcal B_1;\vec s_0,\vec \eta_0)$ be obtained from ${{{\mathfrak F}}}_1$, by replacing $(\vec s_1,\vec \eta_1)$ by $(\vec s_0,\vec \eta_0)$ such that $\vec s_0(\mathcal B_1)$ is disjoint from $\vec s_1(\mathcal B_1)$ and from $\vec {\tilde s}_2(\mathcal B_1)$, where $\Phi\vec {\tilde s}_2 = \vec s_2 \Psi$. The propagation of vacua isomorphism induces an isomorphism between ${{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}({{{\mathfrak F}}}_0)$ and ${{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}({{{\mathfrak F}}}_1)$. Furthermore the propagation of vacua induces an isomorphism between ${{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}({{{\mathfrak F}}}_0)$ and ${{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}(\tilde {{{\mathfrak F}}}_2)$, where $\tilde {{{\mathfrak F}}}_2 = (\mathcal C_1 {\mathop{\fam0 \rightarrow}\nolimits}\mathcal B_1; \vec {\tilde s}_2, \vec {\tilde \eta}_2)$ and $\Phi \vec {\tilde \eta}_2= \Psi^*\vec \eta_2$. The identity on $\mathcal F_N(\mathcal B_1)$ induces an isomorphism between ${{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}(\tilde{\mathcal F}_2)$ and ${{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}(\Psi^*(\mathcal F_2))$. Composing these with the pull back isomorphism just as in the non-abelian case, we arrive at the following proposition. \[Tfamisoab\] We get an induced bundle morphism $$\begin{CD} {{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}({{{\mathfrak F}}}_1) @> {{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}(\Phi) >> {{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}({{{\mathfrak F}}}_2)\\ @V VV @VVV\\ {\mathcal B}_1 @> \Psi >> {\mathcal B}_2, \end{CD}\label{}$$ determined as above. Moreover, composition of such family morphisms goes to composition of the induced bundle morphisms. Suppose now that we have two families ${{{\mathfrak F}}}_i$, $i=1,2$ over ${\Sigma}$ with the property that they have the same image ${\mathop{\fam0 {\tilde \Psi}}\nolimits}_{{{{\mathfrak F}}}_1}({\mathcal B}_1) = {\mathop{\fam0 {\tilde \Psi}}\nolimits}_{{{{\mathfrak F}}}_2}({\mathcal B}_2)$ in Teichmüller space ${\mathop{\fam0 {\mathcal T}}\nolimits}_{{\Sigma}}$ and that ${{{\mathfrak F}}}_2$ is a [good]{}family with respect to ${\Sigma}$. For such a pair of families there exists by Proposition \[famequivalence\] a unique fiber preserving biholomorphism $\Phi_{12} : \mathcal C_1{\mathop{\fam0 \rightarrow}\nolimits}\mathcal C_2$ covering $\Psi^{-1}_{{{{\mathfrak F}}}_2}\Psi_{{{{\mathfrak F}}}_1}$ such that $\Phi^{-1}_{{{{\mathfrak F}}}_2} \Phi_{12} \Phi_{{{{\mathfrak F}}}_1} : Y {\mathop{\fam0 \rightarrow}\nolimits}Y$ is isotopic to $\Psi^{-1}_{{{{\mathfrak F}}}_2}\Psi_{{{{\mathfrak F}}}_1}\times {\mathop{\fam0 Id}\nolimits}$ through such fiber preserving maps. By Theorem \[thm5.2a\] we have that ${{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}({{{\mathfrak F}}}_i)$ are holomorphic line bundles over $\mathcal B_i$. By Proposition \[Tfamisoab\] we get induced a glueing isomorphism $${{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}(\Phi_{12}) : {{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}({{{\mathfrak F}}}_1) {\mathop{\fam0 \rightarrow}\nolimits}{{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}({{{\mathfrak F}}}_2). \label{overlapisoab}$$ \[dvbovacab\] Let ${\Sigma}$ be a closed oriented smooth surface. We now define a line bundle ${{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}= {{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}({\Sigma},P)$ over Teichmüller space ${\mathop{\fam0 {\mathcal T}}\nolimits}_{{\Sigma}}$ using the cover $\{{\mathop{\fam0 {\tilde \Psi}}\nolimits}_{{{{\mathfrak F}}}}(\mathcal B)\}$, where ${{{\mathfrak F}}}$ runs over the [stable]{}and [saturated]{}[good]{}families of pointed Riemann surfaces with formal neighbourhoods over ${\Sigma}$. Over ${\mathop{\fam0 {\tilde \Psi}}\nolimits}_{{{{\mathfrak F}}}}(\mathcal B)$ we specify the line bundle as $({\mathop{\fam0 {\tilde \Psi}}\nolimits}_{{{\mathfrak F}}}^{-1})^*{{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}({{{\mathfrak F}}})$. On overlaps of the image of two [good]{}families, we use the glueing isomorphism ${{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}(\Phi_{12})$ to glue the corresponding bundles together. We obviously have the following \[Teichpullbackab\] For any [stable]{}and [saturated]{}family ${{\tilde {\mathfrak F}}}$ of pointed Riemann surfaces with formal neighbourhoods over ${\Sigma}$ we have a preferred isomorphism $${\tilde \Upsilon}_{{{\tilde {\mathfrak F}}}} : {{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}({{\tilde {\mathfrak F}}}) {\mathop{\fam0 \rightarrow}\nolimits}{\mathop{\fam0 {\tilde \Psi}}\nolimits}_{{{\tilde {\mathfrak F}}}}^*{{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}({\Sigma})$$ induced by the transformation isomorphism between ${{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}({{\tilde {\mathfrak F}}})$ and ${{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}({{{\mathfrak F}}})$, for [good]{}families ${{{\mathfrak F}}}$ of pointed Riemann surfaces with formal neighbourhoods over ${\Sigma}$ such that ${\mathop{\fam0 {\tilde \Psi}}\nolimits}_{{{\mathfrak F}}}(\mathcal B)$ intersect ${\mathop{\fam0 {\tilde \Psi}}\nolimits}_{{{\tilde {\mathfrak F}}}}({\mathcal B}')$ nonempty. Suppose now $f : {\Sigma}_1 {\mathop{\fam0 \rightarrow}\nolimits}{\Sigma}_2$ is a morphism of surfaces. Then of course $f$ induces a biholomorphism $f^*$ from ${\mathop{\fam0 {\mathcal T}}\nolimits}_{{\Sigma}_1}$ to ${\mathop{\fam0 {\mathcal T}}\nolimits}_{{\Sigma}_2}$. Let now ${{{\mathfrak F}}}_1$ be a [good]{}family of stable pointed Riemann surfaces with formal neighbourhoods over ${\Sigma}_1$. Then by composing with $f^{-1}\times {\mathop{\fam0 Id}\nolimits}$ we get a [good]{}family ${{{\mathfrak F}}}_2$ of stable pointed Riemann surfaces with formal neighbourhoods over ${\Sigma}_2$ over the same base $\mathcal B_1$. The identity morphism on $\mathcal F_N(\mathcal B_1)$ then induces a morphism ${{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}(f) : {{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}({{{\mathfrak F}}}_1) {\mathop{\fam0 \rightarrow}\nolimits}{{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}({{{\mathfrak F}}}_2)$ which covers the identity on the base. This is precisely the morphism induced from the morphism of families $\Phi_f = f\times {\mathop{\fam0 Id}\nolimits}: {{{\mathfrak F}}}_1 {\mathop{\fam0 \rightarrow}\nolimits}{{{\mathfrak F}}}_2$ by Proposition \[Tfamisoab\]. This in turn induces a morphism ${{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}(f) : ({\mathop{\fam0 {\tilde \Psi}}\nolimits}_{{{{\mathfrak F}}}_1}^{-1})^*({{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}({{{\mathfrak F}}}_1)) {\mathop{\fam0 \rightarrow}\nolimits}({\mathop{\fam0 {\tilde \Psi}}\nolimits}_{{{{\mathfrak F}}}_2}^{-1})^*({{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}({{{\mathfrak F}}}_2))$ which covers $f^* : {\mathop{\fam0 {\tilde \Psi}}\nolimits}_{{{{\mathfrak F}}}_1}(B_1) {\mathop{\fam0 \rightarrow}\nolimits}{\mathop{\fam0 {\tilde \Psi}}\nolimits}_{{{{\mathfrak F}}}_2}(B_1)$. \[comptransfab\] The above construction provides a well defined lift of $f^* : {\mathop{\fam0 {\mathcal T}}\nolimits}_{{\Sigma}_2} {\mathop{\fam0 \rightarrow}\nolimits}{\mathop{\fam0 {\mathcal T}}\nolimits}_{{\Sigma}_1}$ to a morphism ${{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}(f) : {{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}({\Sigma}_1) {\mathop{\fam0 \rightarrow}\nolimits}{{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}({\Sigma}_2)$ which behaves well under compositions. The proof is exactly the same as the proof of Proposition \[comptransf\]. The connection in the line bundle of [abelian]{}vacua over Teichmüller space. {#New5ab} ============================================================================= Let us use the same notation as in §6.1. Let ${{{\mathfrak F}}}$ be a family of stable and saturated pointed Riemann surfaces with formal neighbourhoods on $({\Sigma},P)$. For an element ${\vec \ell} = ({\ell}_1, \dots, {\ell}_N) $ in $$\mathcal{L}(\mathfrak{F}) := \bigoplus_{j=1}^N \mathcal{O} _\mathcal{B} [\; \xi_j^{-1}] \frac d{d \xi_j},$$ the action $D({\vec \ell})$ on ${{{\mathfrak F}}}$ is defined by $$\label{4.2.8} D(\vec \ell) (F \otimes | u \rangle) = \theta(\vec \ell)(F) \otimes | u \rangle - F \cdot \Bigl(\sum_{j=1}^N \rho_j(T[{\ell}_j] )\Bigr) | u \rangle,$$ where $$F \in {\cO}_{\cB}, \quad | u \rangle \in \cF_N,$$ and $$T[\ell] = {\operatornamewithlimits{Res}}_{z=0}(T(z) \ell(z) dz),.$$ Here $T(z)$ is the energy-momentum tensor T of spin $j=0$ bc ghost system. Then the action has the similar properties as those of Proposition 6.1. We define the dual action of $\cL(\gF)$ on $\cFd_N(\cB)$ by $$\label{4.2.9} D(\vec \ell) (F \otimes \langle \Phi|) = \theta(\vec \ell)(F) \otimes \langle \Phi| + \sum_{j=1}^N F \cdot \langle \Phi| \rho_j(T[{\ell}_j]).$$ where $$F \in \cO_{\cB}, \quad \langle \Phi | \in \cFd_N(\cB).$$ Then, for any $| u \rangle \in \cF_N(\cB)$ and $\langle\Phi | \in \cFd_N(\cB)$, we have $$\label{4.2.10} \{D(\vec \ell) \langle \Phi| \}|\widetilde{\Phi}\rangle + \langle \Phi|\{D(\vec \ell)|\widetilde{\Phi}\rangle\} = \theta(\vec \ell)\langle \Phi|\widetilde{\Phi}\rangle .$$ Now the operator $D(\vec \ell)$ acts on $\cV_{{\mathop{\fam0 ab}\nolimits}}(\gF)$. \[prop4.3a\] For any $\vec \ell \in \cL(\gF)$ we have $$D(\vec \ell)(\cF_{{\mathop{\fam0 ab}\nolimits}}(\gF)) \subset \cF_{{\mathop{\fam0 ab}\nolimits}}(\gF).$$ Hence, $D(\vec \ell)$ operates on $\cV_{{\mathop{\fam0 ab}\nolimits}}(\gF)$. Moreover, it is a first order differential operator, if $\theta(\vec \ell) \neq 0$. Now choose a meromorphic bidifferential $$\omega \in H^0(\cC\times_{\cB}\cC, \omega_{\cC\times_{\cB}\cC}(2\Delta))$$ defined by . Put $$\label{aomega} b_\omega(\vec{\ell}) =\sum_{j=1}^N {\operatornamewithlimits{Res}}_{\xi_j=0} \Big(\ell_j(\xi_j) S_\omega(\xi_j)d\xi_j\Bigr)$$ where $S_\omega$ is the projective connection defined by . Then this defines an $\cO_\cB$-module homomorphism $$b_\omega : \cL(\gF) \rightarrow \cO_\cB,$$ and if $\theta(\vec{\ell})=0$ then we have that $$\langle \Phi |\{D(\vec{\ell})| u \rangle \} =\frac16 b_\omega(\vec{\ell}) \langle \Phi | u \rangle .$$ For a vector field $X$ on $\cB$ choose $\vec{\ell} \in \cL(\gF)$ such that $\theta(\vec{\ell})=X$. Then the connection on ${{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}(\gF)$ is defined by $$\label{4.19a} \nabla_X^{(\omega)}(\langle\Phi|) = D(\vec{\ell})(\langle\Phi|) + \frac16 b_\omega(\vec{\ell}) (\langle\Phi|),$$ for $\langle\Phi| \in {{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}(\gF)$. This is well-defined and is independent of the choice of $\vec{\ell} \in\cL(\gF)$ with $\theta(\vec{\ell})=X$. Just like for the non-abelian conformal field theory (see for example [@Ue2], section 5) we can prove the following theorem. \[thm4.2a\] The operator $\nabla^{(\omega)}$ defines a projectively flat holomorphic connection of the sheaves $\cV_{{\mathop{\fam0 ab}\nolimits}}(\gF)$ and ${{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}(\gF)$. Moreover, the connection has a regular singularity along the locus $D \subset \cB$ which is the locus of the singular curves. The connection $\nabla^{(\omega)}$ depends on the choice of bidifferential $\omega$ and if we choose another bidifferential $\omega'$ then there exists a holomorphic one-form $\phi_{\omega,\omega'}$ on $\cB$ such that $$\label{4.20a} \nabla_X^{(\omega)} - \nabla_X^{(\omega')}= \frac16\langle \phi_{\omega,\omega'}, \, X\rangle.$$ Moreover, the curvature form $R$ of $\nabla_X^{(\omega)}$ is given by $$\label{4.21a} R(X,Y) = \frac16 \Big\{b_\omega(\vec{n}) - X(b_\omega(\vec{m}) ) +Y(b_\omega(\vec{\ell}) )- \sum_{j=1}^N {\operatornamewithlimits{Res}}_{\xi_j=0}\big(\frac{d^3\ell_j}{d\xi_j}m_jd \xi_j\bigr)\Bigr\},$$ where $X$,$Y \in \Theta_{\cC/\cB}(*S))$, $\vec{\ell}$, $\vec{m} \in \cL(\gF)$ with $X=\theta(\vec{\ell})$, $Y= \theta(\vec{m})$, and $\vec{n} \in \cL(\gF)$ is defined by $\vec{n} = [\vec{\ell}, \vec{m}]_d$ $($see $)$. \[comparison\] Let $\mathfrak{F}$ be a family of stable and saturated pointed Riemann surfaces with formal neighbourhoods on $({\Sigma}, P)$. If we use the same bidifferential $\omega$ to define the connections on the bundle of vacua and the line bundle of abelian vacua on $\mathfrak{F}$, then we have $$\label{relation} R^\omega(X,Y) = \frac{c_v}{2} R(X,Y)\otimes {\mathop{\fam0 Id}\nolimits}.$$ \[confamab\] Let ${{{\mathfrak F}}}$ be a family of stable and saturated pointed Riemann surfaces with formal neighbourhoods on $({\Sigma},P)$ and choose a symplectic basis $(\alpha_1, \ldots, \alpha_g,\beta_1,\ldots,\beta_g)$ of $H_1({\Sigma}, {{\mathbb Z}}{})$. Let $\omega\in H^0(\mathcal C \times_{\mathcal B} \mathcal C, \omega_{\mathcal C \times_{\mathcal B} \mathcal C/\mathcal B}(2\Delta))$ be the normalized symmetric bidifferential determined by this data. Then there is the connection ${\mathop{\fam0 {\tilde \nabla}}\nolimits}^{(\omega)}$ in the bundle ${{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}({{{\mathfrak F}}})$ whose $(1,0)$-part is given by formula (4.23) in [@AU1] and whose $(0,1)$-part is just the $\overline{\partial}$-operator in this holomorphic line-bundle. The curvature of this connection is given by the formula . $\square$ Suppose now that we have two [good]{}families ${{{\mathfrak F}}}_i$, $i=1,2$ with the property that they have the same image ${\mathop{\fam0 {\tilde \Psi}}\nolimits}_{{{{\mathfrak F}}}_1} ({\mathcal B}_1) = {\mathop{\fam0 {\tilde \Psi}}\nolimits}_{{{{\mathfrak F}}}_2}({\mathcal B}_2)$ in Teichmüller space ${\mathop{\fam0 {\mathcal T}}\nolimits}_{{\Sigma}}$. For such a pair of families there exists by Proposition \[famequivalence\] a unique fiber preserving biholomorphism $\Phi_{12} : \mathcal C_1{\mathop{\fam0 \rightarrow}\nolimits}\mathcal C_2$ covering $\Psi^{-1}_{{{{\mathfrak F}}}_2}\Psi_{{{{\mathfrak F}}}_1}$ such that $\Phi^{-1}_{{{{\mathfrak F}}}_2} \Phi_{12} \Phi_{{{{\mathfrak F}}}_1} : (Y, P) {\mathop{\fam0 \rightarrow}\nolimits}(Y,P)$ is isotopic to $\Psi^{-1}_{{{{\mathfrak F}}}_2}\Psi_{{{{\mathfrak F}}}_1}\times {\mathop{\fam0 Id}\nolimits}$. \[contransfab\] Let ${\mathop{\fam0 {\tilde \nabla}}\nolimits}^{(\omega)}_i$ be the connection in ${{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}({{{\mathfrak F}}}_i)$ described in Proposition \[confamab\]. Then we have that $${{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}(\Phi_{12})^*({\mathop{\fam0 {\tilde \nabla}}\nolimits}^{(\omega)}_2) = {\mathop{\fam0 {\tilde \nabla}}\nolimits}^{(\omega)}_1.$$ This follows from Theorem \[thm6.1\] above, by the same argument as in the non-abelian case. \[conTeichab\] Let ${\Sigma}$ be a closed oriented surface and let $(\vec \alpha, \vec \beta) = (\alpha_1, \ldots, \alpha_g,\beta_1,\ldots,\beta_g)$ be a symplectic basis of $H_1({\Sigma}, {{\mathbb Z}}{})$. There is a unique connection ${\mathop{\fam0 {\tilde \nabla}}\nolimits}^{(\vec \alpha, \vec \beta)} = {\mathop{\fam0 {\tilde \nabla}}\nolimits}^{(\vec \alpha, \vec \beta)}({\Sigma},P)$ in the bundle ${{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}({\Sigma},P)$ over ${\mathop{\fam0 {\mathcal T}}\nolimits}_{{\Sigma}}$ with the property that for any [good]{}family ${{{\mathfrak F}}}$ of stable pointed Riemann surfaces with formal neighbourhoods over ${\Sigma}$ we have that $${\mathop{\fam0 {\tilde \Psi}}\nolimits}_{{{{\mathfrak F}}}}^*({\mathop{\fam0 {\tilde \nabla}}\nolimits}^{(\vec \alpha, \vec \beta)}) = {\mathop{\fam0 {\tilde \nabla}}\nolimits}^{(\omega)}.$$ In particular the connection is compatible with the holomorphic line-bundle structure on ${{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}({\Sigma},P)$. The curvature is of type $(2,0)$ as stated in Proposition \[confamab\]. If we act on the symplectic basis $(\vec \alpha,\vec \beta)$ by an element $\Lambda = \left(\begin{array}{cc}A & B\\C & D\end{array}\right) \in \text{Sp} (g,{{\mathbb Z}}{})$ so as to obtain $\Lambda(\vec \alpha, \vec \beta)$, as defined by , then $${\mathop{\fam0 {\tilde \nabla}}\nolimits}^{\Lambda(\vec \alpha, \vec \beta)} - {\mathop{\fam0 {\tilde \nabla}}\nolimits}^{(\vec \alpha, \vec \beta)} = \frac{1}{2} \Pi^*( d \log \det (C \tau + D )),\label{contransformab}$$ where $\Pi$ is the period mapping of holomorphic one-forms form the base space of $\mathfrak{F}$ to the Siegel upper-half plane of degree $g$. If $f : {\Sigma}_1 {\mathop{\fam0 \rightarrow}\nolimits}{\Sigma}_2$ is an orientation preserving diffeomorphism of surfaces which maps the symplectic basis $(\vec \alpha^{(1)},\vec \beta^{(1)})$ of $H_1({\Sigma}_1,{{\mathbb Z}}{})$ to the symplectic basis $(\vec \alpha^{(2)},\vec \beta^{(2)})$ of $H_1({\Sigma}_2,{{\mathbb Z}}{})$ then we have that $${{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}(f)^*({\mathop{\fam0 {\tilde \nabla}}\nolimits}^{(\vec \alpha^{(2)},\vec \beta^{(2)})}) = {\mathop{\fam0 {\tilde \nabla}}\nolimits}^{(\vec \alpha^{(1)}, \vec \beta^{(1)})}.$$ $\square$ \[Teichpullbackconab\] For any [stable]{}and [saturated]{}family ${{\tilde {\mathfrak F}}}$ of pointed Riemann surfaces with formal neighbourhoods over ${\Sigma}$ the preferred isomorphism $${\tilde \Upsilon}_{{{\tilde {\mathfrak F}}}} : {{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}({{\tilde {\mathfrak F}}}) {\mathop{\fam0 \rightarrow}\nolimits}{\mathop{\fam0 {\tilde \Psi}}\nolimits}_{{{\tilde {\mathfrak F}}}}^*{{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}({\Sigma})$$ given by Proposition \[Teichpullbackab\] preserves connections and is compatible with the lift ${{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}(f)$. This follows directly from Lemma \[contransfab\] and Theorem \[conTeichab\]. The preferred non-vanishing section of the bundle of [abelian]{}vacua. {#prefsec} ====================================================================== Let $\gX=(C;Q;\xi)$ be a one-pointed smooth curve of genus $g$ with a formal neighbourhood. We shall show that if we fix a symplectic basis $(\vec{\alpha}, \vec{\beta})=(\alpha_1, \ldots, \alpha_g, \beta_1, \ldots, \beta_g)$ of $H_1(C, \bZ)$, then there is a canonical preferred non-zero vector $\langle \omega(\gX,(\{\alpha,\beta\})| \in {{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}(\gX)$. Let us choose a normalized basis $\{\omega_1, \ldots, \omega_g\}$ of holomorphic one-forms on $C$ which is characterized by $$\label{betaone2} \int_{\beta_i}\omega_j = \delta_{i j}, \quad 1 \le i,j \le g.$$ The period matrix is given by $$\tau = (\tau_{ij}), \quad \tau_{ij} = \int_{\alpha_i}\omega_j.$$ Now the numbers $I_n^i$, $n = 1,2, \ldots$, $i=1,\ldots g$ are defined by $$\omega_i = (\sum_{n=1}^\infty I_n^i \xi^{n-1})d\xi.$$ Note that the numbers $I_n^i$ depend on the symplectic basis $(\vec{\alpha}, \vec{\beta})$ and the formal neighbourhood $\xi$. For a positive integer $n \ge 1$ let $\omega_Q^{(n)}$ be a meromorphic one-form on $C$ which has a pole of order $n+1$ at $Q$ and holomorphic elsewhere such that $$\begin{aligned} \label{omegaQ1} \int_{\alpha_i} \omega_Q^{(n)} &= & -\frac{2 \pi \sqrt{-1}I_n^i}{n}, \quad \int_{\beta_i} \omega_Q^{(n)} = 0, \quad 1\le i \le g \\ \label{omegaQ} \omega_Q^{(n)} &=& \bigl( \frac{1}{\xi^{n+1}} + \sum_{m=1}^\infty q_{n,m}\xi^{m-1}\bigr)d\xi.\end{aligned}$$ These conditions uniquely determine $\omega_Q^{(n)}$. Note that the second equality of and imply the first equality of . The preferred element $\langle \omega(\gX,\{\alpha, \beta\})| \in {{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}(\gX)$ is defined by $$\langle \omega(\gX,\{\alpha, \beta\}) | = \cdots e(\omega_{g+2} )\wedge e(\omega_{g+1}) \wedge e(\omega_g) \wedge \cdots \wedge e(\omega_1),$$ where $$\omega_{g+n} = \omega_Q^{(n)}.$$ For details see Lemma 3.1 and its proof of [@AU1]. We call $\{\omega_n\}$, $n=1,2,\ldots$ a normalized basis for $\gX$. Note that the normalized basis depends on the choice of a symplectic basis of $H_1(C,\bZ)$ and the coordinate $\xi$. \[thm6.2\] For $h(\xi) \in \cD_+^0$ put $\gX_h = \{ C;Q;\eta= h(\xi)\}$. Then $$\langle \omega(\gX,\{\alpha, \beta\})| G[h] = \langle \omega(\gX_h,\{\alpha, \beta\})|,$$ where $G[h] : {{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}(\gX) \rightarrow {{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}(\gX_h)$ is the canonical isomorphism given in Proposition \[prop6.1a\] \[thm6.3\] Let $(\alpha_1, \ldots \alpha_g,\beta_1, \ldots, \beta_g )$ and $(\widetilde{\alpha}_1, \ldots \widetilde{\alpha}_g, \widetilde{\beta}_1, \ldots, \widetilde{\beta}_g )$ be symplectic bases of $H^1(C, \bZ)$ of the non-singular curve $C$. Assume that $\{\beta_1, \ldots, \beta_g\}$ and $\{\widetilde{\beta}_1, \ldots, \widetilde{\beta}_g \}$ span the same Lagrangian sublattice in $H^1(C, \bZ)$. Then $$\langle \omega(\gX, \{\alpha,\beta\}) | = \det U \langle \omega(\gX, \{\widetilde{\alpha}, \widetilde{\beta}\}) | ,$$ where $U \in GL(g,\bZ)$ is defined by $$\left( \begin{array}{c} \widetilde{\beta}_1 \\ \vdots \\ \widetilde{\beta}_g \end{array} \right) = U \left( \begin{array}{c} \beta_1 \\ \vdots \\ \beta_g \end{array} \right) .$$ Let $\{p, q\}$ be two smooth points on the curve $C$ with formal neighbourhoods $\xi$, $\eta$, respectively. Put $\gX_0 = (C; p,q; \xi , \eta)$, $\gX_1= (C; p; \xi)$, $\gX_2=(C;q;\eta)$. Then the natural imbeddings $$\begin{aligned} \iota_1&:& \cF \hookrightarrow \cF_2 \\ && |u\rangle \mapsto |u \rangle \otimes |0\rangle \\ \iota_2&:& \cF \hookrightarrow \cF_2 \\ && |u\rangle \mapsto |0\rangle \otimes |u \rangle\\\end{aligned}$$ induce canonical isomorphisms $$\begin{array}{ccccc} &&{{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}(\gX_0)& &\\ && && \\ & {}^{\iota_1^*} \swarrow&& \searrow^{\iota_2^*} & \\ && && \\ &{{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}(\gX_1) && {{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}(\gX_2) & \end{array}$$ by Theorem \[thm3.4\]. \[thm6.4\] Under the above notation we have $$\iota_2^*\circ (\iota^*_1)^{-1}(\langle \omega(\gX_1,\{\alpha, \beta\})| ) = \langle\omega(\gX_2,\{\alpha, \beta\})|.$$ Let ${\Sigma}$ be a closed oriented surface. Assume first that ${\Sigma}$ is connected. As described above, the choice of a symplectic basis gives a preferred section in the line bundle of abelian vacua associated to any family of stable and saturated pointed Riemann surfaces with formal neighbourhoods over ${\Sigma}$. We have that \[prefsecfam\] Let ${{{\mathfrak F}}}$ be a family of stable and saturated pointed Riemann surfaces with formal neighbourhoods over ${\Sigma}$ and choose a symplectic basis $(\vec \alpha, \vec \beta)$ of $H_1({\Sigma}, {{\mathbb Z}}{})$. Then there is a preferred non-vanishing holomorphic section $s^{(\vec \alpha, \vec \beta)}_{{{{\mathfrak F}}}}$ in the bundle ${{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}({{{\mathfrak F}}})^{\otimes 2}$ given by $$s^{(\vec \alpha, \vec \beta)}_{{{{\mathfrak F}}}}(t) = (\langle \omega(\gX_t, \{\alpha(t), \beta(t)\}|)^{\otimes 2}.$$ If we act on the symplectic basis $(\vec \alpha, \vec \beta)$ by an element $\Lambda = \left(\begin{array}{cc}(U^t)^{-1} & B\\0 & U\end{array}\right) \in \text{Sp} (g,{{\mathbb Z}}{})$ in order to obtain $\Lambda(\vec \alpha, \vec \beta)$ as described in , then $$s^{\Lambda(\vec \alpha, \vec \beta)} = s^{(\vec \alpha, \vec \beta)}.\label{sectiontransformfam}$$ This is clear from Theorem \[thm6.3\], since we have $\det U = \pm 1$. $\square$ Thus the section $s^{(\vec \alpha, \vec \beta)}_{{{{\mathfrak F}}}}$ only really depends on the Lagrangian subspace $L = {\mathop{\fam0 Span}\nolimits}\{\beta_i\}$ and we therefore denote it $s_{{{\mathfrak F}}}(L)$. Suppose now that ${\Sigma}$ is not connected and that ${\Sigma}=\coprod_i{\Sigma}_i$ is the decomposition of ${\Sigma}$ into its connected components ${\Sigma}_i$. Let ${{{\mathfrak F}}}$ be a family of stable pointed Riemann surfaces with formal neighbourhoods over ${\Sigma}$. Let ${{{\mathfrak F}}}_i$ be the restriction of ${{{\mathfrak F}}}$ to ${\Sigma}_i$. Let $N_i$ be the number of sections of ${{{\mathfrak F}}}_i$ and $N = \sum_i N_i$ the number of sections of ${{{\mathfrak F}}}$. We obviously have the following lemma. \[DJDlemma\] The isomorphism $\mathcal F_N \cong \otimes_i \mathcal F_{N_i}$ induces an isomorphism of holomorphic line bundles $${{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}({{{\mathfrak F}}}) \cong \otimes_i {{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}({{{\mathfrak F}}}_i),$$ which is compatible with the connections. Suppose now that $(\vec\alpha, \vec\beta) = ((\vec\alpha_i, \vec\beta_i))$ is a symplectic basis of $H_1({\Sigma},{{\mathbb Z}})$. We then define the preferred section to be $$s_{{{\mathfrak F}}}^{(\vec\alpha, \vec\beta)} = \otimes_i s_{{{{\mathfrak F}}}_i}^{(\vec\alpha_i, \vec\beta_i)}.$$ For the rest of this section ${\Sigma}$ is just any closed oriented surface. Suppose now that we have two [good]{}families ${{{\mathfrak F}}}_i$, $i=1,2$ with the property that they have the same image ${\mathop{\fam0 {\tilde \Psi}}\nolimits}_{{{{\mathfrak F}}}_1} ({\mathcal B}_1) = {\mathop{\fam0 {\tilde \Psi}}\nolimits}_{{{{\mathfrak F}}}_2}({\mathcal B}_2)$ in Teichmüller space ${\mathop{\fam0 {\mathcal T}}\nolimits}_{{\Sigma}}$. For such a pair of families there exists by Theorem \[famequivalence\] a unique fiber preserving biholomorphism $\Phi_{12} : \mathcal C_1{\mathop{\fam0 \rightarrow}\nolimits}\mathcal C_2$ covering $\Psi^{-1}_{{{{\mathfrak F}}}_2}\Psi_{{{{\mathfrak F}}}_1}$ such that $\Phi^{-1}_{{{{\mathfrak F}}}_2} \Phi_{12} \Phi_{{{{\mathfrak F}}}_1} : Y {\mathop{\fam0 \rightarrow}\nolimits}Y$ is isotopic to $\Psi^{-1}_{{{{\mathfrak F}}}_2}\Psi_{{{{\mathfrak F}}}_1}\times {\mathop{\fam0 Id}\nolimits}$ through fiber preserving diffeomorphisms. \[prefsecfamch\] Let $s^{(\vec \alpha, \vec \beta)}_{{{{\mathfrak F}}}_i}$ be the preferred sections of ${{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}({{{\mathfrak F}}}_i)^{\otimes 2}$ described in Theorem \[prefsecfam\]. Then we have that $${{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}(\Phi_{12})^{\otimes 2} (s^{(\vec \alpha, \vec \beta)}_{{{{\mathfrak F}}}_1}) = s^{(\vec \alpha, \vec \beta)}_{{{{\mathfrak F}}}_2}.$$ $\square$ \[secTeichab\] Let ${\Sigma}$ be a closed oriented surface and let $(\vec \alpha, \vec \beta) = (\alpha_1, \ldots, \alpha_g,\beta_1,\ldots,\beta_g)$ be a symplectic basis of $H_1({\Sigma}, {{\mathbb Z}}{})$. Then there is a unique non-vanishing holomorphic section $s^{(\vec \alpha, \vec \beta)}= s^{(\vec \alpha, \vec \beta)}_{{\Sigma}}$ in the bundle ${{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}({\Sigma})^{\otimes 2}$ over ${\mathop{\fam0 {\mathcal T}}\nolimits}_{{\Sigma}}$ with the property that for any [good]{}family ${{{\mathfrak F}}}$ of stable pointed Riemann surfaces with formal neighbourhoods over ${\Sigma}$ we have that $$({\mathop{\fam0 {\tilde \Psi}}\nolimits}_{{{{\mathfrak F}}}}^*)^{\otimes 2}(s^{(\vec \alpha, \vec \beta)}) = s^{(\vec \alpha, \vec \beta)}_{{{{\mathfrak F}}}}.$$ The sections transforms according to the transformation rule (\[sectiontransformfam\]. If $f : {\Sigma}_1 {\mathop{\fam0 \rightarrow}\nolimits}{\Sigma}_2$ is an orientation preserving diffeomorphism of surfaces which maps the symplectic basis $(\vec \alpha^{(1)},\vec \beta^{(1)})$ of $H_1({\Sigma}_1,{{\mathbb Z}}{})$ to the symplectic basis $(\vec \alpha^{(2)},\vec \beta^{(2)})$ of $H_1({\Sigma}_2,{{\mathbb Z}}{})$ then we have that $$({{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}(f)^*)^{\otimes 2}(s^{(\vec \alpha^{(2)},\vec \beta^{(2)})}_{{\Sigma}_2}) = s^{(\vec \alpha^{(1)}, \vec \beta^{(1)})}_{{\Sigma}_1}.$$ $\square$ Likewise, we see that the section only depends on the Lagrangian subspace and we denote it therefore $s(L) = s_{\Sigma}(L)$. \[Teichpullbackconabsec\] For any [stable]{}and [saturated]{}family ${{\tilde {\mathfrak F}}}$ of pointed Riemann surfaces with formal neighbourhoods over ${\Sigma}$ the preferred isomorphism $${\tilde \Upsilon}_{{{\tilde {\mathfrak F}}}}^{\otimes 2} : {{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}({{\tilde {\mathfrak F}}})^{\otimes 2} {\mathop{\fam0 \rightarrow}\nolimits}{\mathop{\fam0 {\tilde \Psi}}\nolimits}_{{{\tilde {\mathfrak F}}}}^*{{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}({\Sigma})^{\otimes 2}$$ given by Proposition \[Teichpullbackab\] preserves the preferred sections. This Follows from Lemma \[prefsecfamch\]. The geometric construction of the modular functor. {#construction} ================================================== For the convenience of the reader, let us summarize the results of the sheaf of vacua constructions over Teichmüller spaces of pointed surfaces obtained in non-abelian case in sections \[New3\] to \[New5\] and in the abelian case in sections \[New3ab\] to \[prefsec\]. \[mainconstT\] Let $({\Sigma}, P, \l)$ be a [stable]{}and [saturated]{}labeled pointed surface. - The sheaf of vacua construction (see Definition \[dvbovac\]) yields a vector bundle ${{{\Cal V}_{\vec \lambda}^{\dagger}}}= {{{\Cal V}_{\vec \lambda}^{\dagger}}}({\Sigma}, P)$ over the Teichmüller space ${\mathop{\fam0 {\mathcal T}}\nolimits}_{({\Sigma},P)}$ of $({\Sigma},P)$ whose fiber at a complex structure ${\mathbf C}$ on $({\Sigma}, P)$ is identified (via the isomorphism given in Proposition \[Teichpullback\]) with the space of vacua ${{{\Cal V}_{\vec \lambda}^{\dagger}}}({\mathbf C})$ as defined in Definition \[Defspofv\]. - For each symplectic basis of $H_1({\Sigma},{{\mathbb Z}})$, we get induced a connection in ${{{\Cal V}_{\vec \lambda}^{\dagger}}}$ over ${\mathop{\fam0 {\mathcal T}}\nolimits}_{({\Sigma},P)}$. Any two of these connections differ by a global scalar-value 1-form on ${\mathop{\fam0 {\mathcal T}}\nolimits}_{({\Sigma},P)}$. See Theorem \[conTeich\]. - Each of these connections is projectively flat and their curvature are described in details in Theorem \[curvaturecom\] and \[thm4.2a\]. - There is a natural lift of morphisms of pointed surfaces to these bundles covering induced biholomorphisms between Teichmüller spaces, which preserves compositions. See Proposition \[comptransf\]. - A morphism of pointed surfaces transforms these connections according to the way it transforms symplectic bases of the first homology. See Theorem \[contransf\]. \[remconbaslag\][*If we choose a Lagrangian subspace $L$ of $H_1({\Sigma},{{\mathbb Z}})$ and constrain the symplectic basis $(\alpha_i,\beta_i)$ of $H_1({\Sigma},{{\mathbb Z}})$ such that $L = {\mathop{\fam0 Span}\nolimits}\{\beta_i\}$ then we see from the transformation laws in Theorem \[contransf\], that we get a connection in ${{{\Cal V}_{\vec \lambda}^{\dagger}}}$ which depends only on $L$.*]{} Since the connections in the vector bundle ${{{\Cal V}_{\vec \lambda}^{\dagger}}}$ are only projectively flat, we need a 1-dimensional theory with connections, whose curvature after taking tensor products, can cancel this curvature and result in a bundle with a flat connection. There are obstructions to doing this mapping class group equivariantly, so we expect to see central extension of the mapping class groups occurring. As we shall see below, this is exactly what happens, when one extracts the necessary root of the abelian theory treated in [@AU1], so as to get the right scaling of the curvature. Again, the following theorem summarizes the results and constructions, now in the abelian case treated in section \[New3ab\] to \[prefsec\]. \[mainconstTab\] Let ${\Sigma}$ be a closed oriented surface. - The sheaf of abelian vacua construction (see Definition \[dvbovacab\]) yields a line bundle ${{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}= {{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}({\Sigma})$ over the Teichmüller space ${\mathop{\fam0 {\mathcal T}}\nolimits}_{{\Sigma}}$ of ${\Sigma}$, whose fiber at a complex structure ${\mathbf C}$ on ${\Sigma}$ is identified (via the isomorphism given in Proposition \[Teichpullbackab\]) with the space of abelian vacua ${{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}({\mathbf C})$ as defined in Definition \[Defspofvab\]. - For each symplectic basis of $H_1({\Sigma},{{\mathbb Z}})$, we get induced a holomorphic connection in ${{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}$ over ${\mathop{\fam0 {\mathcal T}}\nolimits}_{{\Sigma}}$ (see Theorem \[conTeichab\]). The difference between the connections associated to two different basis’s is the global scalar-value 1-form on ${\mathop{\fam0 {\mathcal T}}\nolimits}_{{\Sigma}}$ given in (\[contransformab\]). - The curvature of each of these connections are described in Proposition \[confamab\]. - For each symplectic basis of $H_1({\Sigma},{{\mathbb Z}})$, we also get a preferred non-vanishing section of $({{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}})^{\otimes 2}$ as specified in Theorem \[secTeichab\]. The transformation formula (\[sectiontransformfam\]) states how the preferred sections transforms under change of the symplectic basis of $H_1({\Sigma},{{\mathbb Z}})$. - There is a natural lift of morphisms of surfaces to these bundles covering induced biholomorphisms between Teichmüller space, which preserves compositions. See Proposition \[comptransfab\]. - A morphism of surfaces transforms these connections and the preferred sections according to the way it transforms symplectic bases of the first homology. See Theorem \[contransfab\] and \[secTeichab\]. [*If we choose a Lagrangian subspace $L$ of $H_1({\Sigma},{{\mathbb Z}})$ and constrain the symplectic basis $(\alpha_i,\beta_i)$ of $H_1({\Sigma},{{\mathbb Z}})$ such that $L = {\mathop{\fam0 Span}\nolimits}\{\beta_i\}$ then we see from the transformation laws in Theorem \[secTeichab\], that we get a preferred non-vanishing section $s=s(L)$ and a connection in $({{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}})^{\otimes 2}$ which only depends on $L$.* ]{} From the discussion of the curvatures of the connections in ${{{\Cal V}_{\vec \lambda}^{\dagger}}}$ and ${{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}$, i.e. by comparing the curvature formula , it is clear that the root of $({{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}})^{\otimes 2}$ we are seeking is $c_{{\mathop{\fam0 {\upsilon}}\nolimits}}$. The following theorem provided us with such a root. \[fracpowerab\] For any marked surface ${{\mathbf \Sigma}}= ({\Sigma}, L)$ there exists a line bundle, which we denoted $({{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}})^{-\frac{1}{2}c_{{\mathop{\fam0 {\upsilon}}\nolimits}}}(L)= ({{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}})^{-\frac{1}{2}c_{{\mathop{\fam0 {\upsilon}}\nolimits}}}({{\mathbf \Sigma}})$, over ${\mathop{\fam0 {\mathcal T}}\nolimits}_{\Sigma}$ that satisfies the following: - $({{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}})^{-\frac{1}{2}c_{{\mathop{\fam0 {\upsilon}}\nolimits}}}$ is a functor from the category of marked surfaces to the category of line bundles over Teichmüller spaces of closed oriented surfaces. - If we choose a symplectic basis of $H_1({\Sigma},{{\mathbb Z}})$ for a marked surface ${{\mathbf \Sigma}}$ then we get induced a connection in $({{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}})^{-\frac{1}{2}c_{{\mathop{\fam0 {\upsilon}}\nolimits}}}(L)$, whose curvature is $-\frac{1}{2}c_{{\mathop{\fam0 {\upsilon}}\nolimits}}$ times the curvature of the corresponding connection in ${{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}$. The difference between the connections associated to two different bases is $-\frac{1}{2}c_{{\mathop{\fam0 {\upsilon}}\nolimits}}$ times the global scalar-value 1-form on ${\mathop{\fam0 {\mathcal T}}\nolimits}_{{\Sigma}}$ given in (\[contransformab\]). Let $(({{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}})^{\otimes 2})^*$ be the complement of the zero section of $({{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}})^{\otimes 2}$. Let $\widetilde{({{{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}})^{\otimes 2}}$ be the fiberwise universal cover of $(({{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}})^{\otimes 2})^*$ based at the section $s(L)$. This is a completely functorial construction on pairs of line bundles and non-vanishing sections. There is a unique lift of the ${\mathop{\fam0 {\mathbb C}^{}}\nolimits}^*$-action on $(({{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}})^{\otimes 2})^*$ to a ${\mathop{\fam0 {\mathbb C}^{}}\nolimits}$-action on $\widetilde{({{{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}})^{\otimes 2}}$ with respect to the covering map $\exp$ from ${\mathop{\fam0 {\mathbb C}^{}}\nolimits}$ to ${\mathop{\fam0 {\mathbb C}^{}}\nolimits}^*$. For any $\alpha\in {\mathop{\fam0 {\mathbb C}^{}}\nolimits}^*$ we can now functorially define a line bundle $(({{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}})^{\otimes 2})^\alpha(L)$ as follows: $$(({{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}})^{\otimes 2})^\alpha(L) = \widetilde{({{{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}})^{\otimes 2}} \times_{\rho_\alpha}{\mathop{\fam0 {\mathbb C}^{}}\nolimits},$$ where $\rho_\alpha(z) : {\mathop{\fam0 {\mathbb C}^{}}\nolimits} {\mathop{\fam0 \rightarrow}\nolimits}{\mathop{\fam0 {\mathbb C}^{}}\nolimits}$ is the linear map given by multiplication by $\exp(\alpha z)$ for all $z\in {\mathop{\fam0 {\mathbb C}^{}}\nolimits}$. We emphasis the dependence of this bundle on the section $s(L)$ and hence on $L$ in the notation for this bundle. Here we choose $\alpha =- c_v/4$ to define $({{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}})^{-\frac{1}{2}c_{{\mathop{\fam0 {\upsilon}}\nolimits}}}$. It is clear from the construction of $(({{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}})^{\otimes 2})^\alpha(L)$, that a connection in ${{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}$ will induce a connection in $(({{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}})^{\otimes 2})^\alpha(L)$, whose curvature two-form is $\alpha/2$ times the curvature two-form of that connection in ${{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}$. For the construction of the functor on the morphisms of marked surfaces, we refer to [@Walker] and [@A]. $\square$ By pulling $({{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}})^{-\frac{1}{2}c_{{\mathop{\fam0 {\upsilon}}\nolimits}}}(L)$ with its connection back to ${\mathop{\fam0 {\mathcal T}}\nolimits}_{({\Sigma},P)}$ from ${\mathop{\fam0 {\mathcal T}}\nolimits}_{\Sigma}$, we get a line bundle with with a connection on ${\mathop{\fam0 {\mathcal T}}\nolimits}_{({\Sigma},P)}$, which we also denote $({{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}})^{-\frac{1}{2}c_{{\mathop{\fam0 {\upsilon}}\nolimits}}}(L)$. Let now $({{\mathbf \Sigma}}, \l) = ({\Sigma}, P,V,L, \l)$ be a [stable]{}and [saturated]{}labeled marked surface. From the above Theorems \[mainconstT\] and \[fracpowerab\], we see that there is a well defined flat connection in the vector bundle ${{{\Cal V}_{\vec \lambda}^{\dagger}}}\otimes ({{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}})^{-\frac{1}{2}c_{{\mathop{\fam0 {\upsilon}}\nolimits}}}(L)$ over ${\mathop{\fam0 {\mathcal T}}\nolimits}_{({\Sigma}, P)}$ gotten by taking the tensor product connection of the two connections induced by any symplectic basis $(\alpha_i,\beta_i)$ of $H_1({\Sigma},{{\mathbb Z}})$. Now ${\mathop{\fam0 {\mathcal T}}\nolimits}_{({\Sigma}, P)}$ forms a ${\mathop{\fam0 {\mathbb R}_+^{P}}\nolimits}$-principal bundle over the reduced Teichmüller space ${\mathop{\fam0 {\mathcal T}^{(r)}}\nolimits}_{({\Sigma}, P)}$. Hence we can use the flat connection to push forward this bundle to obtain a bundle with a flat connection over the reduced Teichmüller space. \[Flatbdloredteich\] For the [stable]{}and [saturated]{}labeled marked surface $({{\mathbf \Sigma}},\l)$ we define the vector bundle ${{{\Cal V}_{\vec \lambda}^{\dagger}}}({{\mathbf \Sigma}})$ with its flat connection $\nabla({{\mathbf \Sigma}},\l)$ as the push forward of the bundle ${{{\Cal V}_{\vec \lambda}^{\dagger}}}\otimes ({{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}})^{-\frac{1}{2}c_{{\mathop{\fam0 {\upsilon}}\nolimits}}}(L)$ to the reduced Teichmüller space ${\mathop{\fam0 {\mathcal T}^{(r)}}\nolimits}_{({\Sigma}, P)}$ followed by restriction to the fiber ${\mathop{\fam0 {\mathcal T}}\nolimits}_{{{\mathbf \Sigma}}}$. [*By Theorem \[mainconstT\] and \[mainconstTab\] and Lemma \[momswd\] we see that morphism of [stable]{}and [saturated]{}marked surfaces induces isomorphisms of flat vector bundles covering corresponding diffeomorphisms of Teichmüller spaces of the corresponding marked surfaces.*]{} However, for a labeled marked surface $({{\mathbf \Sigma}}, \l) = ({\Sigma}, P,V,L, \l)$, which is not [stable]{}or not [saturated]{}we need to say a little more. Namely, let $({{\mathbf \Sigma}}',\l')$ be obtained from $({{\mathbf \Sigma}},\l)$ by further labeling points not in $P$ by the zero label $0\in P_\ell$ and choose projective tangent vectors at these new labeled points, such that ${{\mathbf \Sigma}}'$ is both [stable]{}and [saturated.]{}. Let $\pi'$ be the projection map from ${\mathop{\fam0 {\mathcal T}}\nolimits}_{{{\mathbf \Sigma}}'}$ to ${\mathop{\fam0 {\mathcal T}}\nolimits}_{{{\mathbf \Sigma}}}$. \[nonstabsat\] The connection $\nabla({{\mathbf \Sigma}}',\l')$ has trivial holonomy along the fibers of the projection map $\pi'$. The connection $\nabla({{\mathbf \Sigma}}',\l')$ induces a flat connection in the bundle over ${\mathop{\fam0 {\mathcal T}}\nolimits}_{{{\mathbf \Sigma}}}$ obtained by push forward ${{\mathcal V}_{{\mathbf \lambda}'}^{\dagger}}({{\mathbf \Sigma}}')$ along $\pi'$ using $\nabla({{\mathbf \Sigma}}',\l')$. If $({{\mathbf \Sigma}}'',\l'')$ is another [stable]{}and [saturated]{}labeled marked surface obtained from $({{\mathbf \Sigma}},\l)$ in the same way by adding zero-labeled point to $P$, then iterations of the propagation of vacua isomorphisms given in Proposition \[propvaciso\] induces a connection preserving isomorphism between the corresponding pair of bundles over ${\mathop{\fam0 {\mathcal T}}\nolimits}_{{{\mathbf \Sigma}}}$. This proposition follows directly from Proposition \[pullbackcon\] and the definition of the flat vector bundle ${{\mathcal V}_{{\mathbf \lambda}'}^{\dagger}}({{\mathbf \Sigma}}',P')$. \[defnonstabsat\] We define the vector bundle with its flat connection $({{{\Cal V}_{\vec \lambda}^{\dagger}}}({{\mathbf \Sigma}}),\nabla({{\mathbf \Sigma}},\l))$ over ${\mathop{\fam0 {\mathcal T}}\nolimits}_{{{\mathbf \Sigma}}}$ to be $\pi'_*({{\mathcal V}_{{\mathbf \lambda}'}^{\dagger}}({{\mathbf \Sigma}}'), \nabla({{\mathbf \Sigma}}',\l'))$ for any [stable]{}and [saturated]{}labeled marked surface $({{\mathbf \Sigma}}',\l')$ obtained from $({{\mathbf \Sigma}},\l)$ by adding zero-labeled marked points to $P$. Suppose now $f : ({{\mathbf \Sigma}}_1,\l_1) {\mathop{\fam0 \rightarrow}\nolimits}({{\mathbf \Sigma}}_2,\l_2)$ is a morphism of labeled marked surfaces and that $({{\mathbf \Sigma}}'_i,\l'_i)$ is obtained as above from $({{\mathbf \Sigma}}_i,\l_i)$ by adding zero-labeled points and further that $f': ({{\mathbf \Sigma}}'_1,\l'_1) {\mathop{\fam0 \rightarrow}\nolimits}({{\mathbf \Sigma}}'_2,\l'_2)$ is any morphism of labeled marked surfaces, which induces $f$ when restricted to $({{\mathbf \Sigma}}_1,\l_1)$. We then have the following result as a direct consequence of Proposition \[propvmorph\]. \[mornonstabsat\] The induced morphism of flat vector bundles $\Spofv(f') : \Spofvlampone({{\mathbf \Sigma}}'_1) {\mathop{\fam0 \rightarrow}\nolimits}\Spofvlamptwo({{\mathbf \Sigma}}'_2)$ induces a morphism of flat vector bundles from $\Spofvlamone({{\mathbf \Sigma}}_1)$ to $\Spofvlamtwo({{\mathbf \Sigma}}_2)$ which only depends on $f$ and which behaves well under compositions of morphism of labeled marked surface. Let us now collect the thus fare obtained in the following theorem. The construction given above gives a functor from the category of labeled marked surfaces to the category of vector bundles with flat connections over Teichmüller spaces of marked surfaces. The modular functor we seek is now simply just obtained by composing with the functor which takes covariant constant sections of vector bundles with connections. \[def.main\] Let $\ell$ be a positive integer. Let $P_{\ell}$ be the finite set defined in (\[labelset\]) with the involution $\dagger$ as defined by (\[involution\]). Let $({{\mathbf \Sigma}}, \l) = ({\Sigma},P,V, L, \l)$ be a labeled marked surface using the label set $P_{\ell}$. The functor $V^{{\mathop{\fam0 {\mathfrak g}}\nolimits}}_{\ell}$ is by definition the composite of the functor, which assigns to $({{\mathbf \Sigma}}, \l)$ the flat vector bundle ${{{\Cal V}_{\vec \lambda}^{\dagger}}}({{\mathbf \Sigma}})$ over ${\mathop{\fam0 {\mathcal T}}\nolimits}_{{{\mathbf \Sigma}}}$, and the functor, which takes covariant constant sections. [*For a labeled marked surface $({{\mathbf \Sigma}},\l)$ and a complex structure ${\mathbf C}$ on it, we see that Proposition \[Teichpullback\] and \[Teichpullbackab\] give an isomorphism $$V^{{\mathop{\fam0 {\mathfrak g}}\nolimits}}_{\ell}({{\mathbf \Sigma}},\l) \cong {{{\Cal V}_{\vec \lambda}^{\dagger}}}({\mathbf C})\otimes ({{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}})^{-\frac{1}{2}c_{{\mathop{\fam0 {\upsilon}}\nolimits}}}(L)({\mathbf C}),$$ since ${\mathop{\fam0 {\mathcal T}}\nolimits}_{{{\mathbf \Sigma}}}$ is contractible. Moreover, if ${{\mathbf f}} : ({{\mathbf \Sigma}}_1,\l_1){\mathop{\fam0 \rightarrow}\nolimits}({{\mathbf \Sigma}}_2,\l_2)$ is a morphism of labeled marked Riemann surfaces, which is realized by a morphism of labeled marked Riemann surfaces $\Phi: {{\mathbf C}}_1 {\mathop{\fam0 \rightarrow}\nolimits}{{\mathbf C}}_2$, such that $\Phi^*(L_2) = L_1$, then we have the following commutative diagram $$\begin{CD} V^{{\mathop{\fam0 {\mathfrak g}}\nolimits}}_{\ell}({{\mathbf \Sigma}}_1,\l_1) @>\cong>> {{{\Cal V}_{\vec \lambda}^{\dagger}}}({\mathbf C}_1)\otimes ({{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}})^{-\frac{1}{2}c_{{\mathop{\fam0 {\upsilon}}\nolimits}}}(L_1)({\mathbf C}_1)\\ @V {V^{{\mathop{\fam0 {\mathfrak g}}\nolimits}}_{\ell}({\mathbf f})} VV @V\Vdag(\Phi)\otimes {{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}(\Phi)^{-\frac{1}{2}c_{{\mathop{\fam0 {\upsilon}}\nolimits}}} VV\\ V^{{\mathop{\fam0 {\mathfrak g}}\nolimits}}_{\ell}({{\mathbf \Sigma}}_2,\l_2) @>\cong>>{{{\Cal V}_{\vec \lambda}^{\dagger}}}({\mathbf C}_2)\otimes ({{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}})^{-\frac{1}{2}c_{{\mathop{\fam0 {\upsilon}}\nolimits}}}(L_2)({\mathbf C}_2). \end{CD}$$* ]{} [*Let $({{\mathbf \Sigma}},\l)$ be a labeled marked surface and suppose that $({{\mathbf \Sigma}}',\l')$ is obtained from $({{\mathbf \Sigma}},\l)$ by labeling further points by $0\in P_\ell$, then by Proposition \[nonstabsat\] we get induced an isomorphism $$V^{{\mathop{\fam0 {\mathfrak g}}\nolimits}}_{\ell}({{\mathbf \Sigma}},\l) \cong V^{{\mathop{\fam0 {\mathfrak g}}\nolimits}}_{\ell}({{\mathbf \Sigma}}',\l').$$* ]{} Let $({{\mathbf \Sigma}},\l)$ be a labeled marked surface. Let ${{{\mathfrak F}}}= (\pi : {{\mathcal C}}{\mathop{\fam0 \rightarrow}\nolimits}\mathcal B,\vec s,\vec \eta)$ be a family of stable and saturated Riemann surfaces with formal neighbourhoods over ${{\mathbf \Sigma}}$. We define $V^{{\mathop{\fam0 {\mathfrak g}}\nolimits}}_{\ell}({{{\mathfrak F}}},\l)$ to be the covariant constant sections of the flat bundles $ {{{\Cal V}_{\vec \lambda}^{\dagger}}}({{{\mathfrak F}}})\otimes ({{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}})^{-\frac{1}{2}c_{{\mathop{\fam0 {\upsilon}}\nolimits}}}(L)({{{\mathfrak F}}})$ over $\mathcal B$. Form this definition it is clear that we get an isomorphism $$I_{{{\mathfrak F}}}: V^{{\mathop{\fam0 {\mathfrak g}}\nolimits}}_{\ell}({{{\mathfrak F}}},\l) {\mathop{\fam0 \rightarrow}\nolimits}V^{{\mathop{\fam0 {\mathfrak g}}\nolimits}}_{\ell}({{\mathbf \Sigma}},\l).$$ In order for the functor $V^{{\mathop{\fam0 {\mathfrak g}}\nolimits}}_{\ell}$ to be modular, we need to further construct the disjoint union isomorphism and the glueing isomorphism and to check that the axioms of a modular functor is satisfied. First we construct the disjoint union isomorphism. The glueing isomorphism will be constructed in the following section. Let $({{\mathbf \Sigma}}_i,\l_i) = ({\Sigma}_i,P_i,V_i,L_i, \l_i)$, $i=1,2$, be two [stable]{}and [saturated]{}labeled marked surfaces and let $({{\mathbf \Sigma}},\l) = ({{\mathbf \Sigma}}_1,\l_1) \sqcup ({{\mathbf \Sigma}}_2,\l_2)$. Let $\l = \l_1 \sqcup \l_2$. We have that ${\mathop{\fam0 {\mathcal T}}\nolimits}_{{{\mathbf \Sigma}}} = {\mathop{\fam0 {\mathcal T}}\nolimits}_{{{\mathbf \Sigma}}_1}\times{\mathop{\fam0 {\mathcal T}}\nolimits}_{{{\mathbf \Sigma}}_2}$. Let $\pi_i : {\mathop{\fam0 {\mathcal T}}\nolimits}_{{{\mathbf \Sigma}}}{\mathop{\fam0 \rightarrow}\nolimits}{\mathop{\fam0 {\mathcal T}}\nolimits}_{{{\mathbf \Sigma}}_i}$ be the projection onto the $i$’th factor. We clearly have that \[disjointuniiso\] The natural isomorphism ${{{\Cal H}_{\vec \lambda}}}\cong{{{\mathcal H}_{{\vec \lambda}_1}}}\otimes {{{\mathcal H}_{{\vec \lambda}_2}}}$ (for any ordering $\vec \l, \vec \l_1$ and $\vec \l_2$ of $\l, \l_1$ and $\l_2$ respectively) induces an isomorphism of vector bundles with connections $${{{\Cal V}_{\vec \lambda}^{\dagger}}}({\Sigma},P) \cong \pi_1^*\Spofvlamone({\Sigma}_1,P_1) \otimes \pi_2^*\Spofvlamtwo({\Sigma}_2,P_2),\label{isodisj}$$ where we use the Lagrangian subspaces to fix the connections in all three bundles. The isomorphism is compatible with isomorphism induced by disjoint union of morphism of corresponding labeled marked surfaces. [*These disjoint union isomorphisms are clearly compatible with the propagation of vacua isomorphisms given in Proposition \[nonstabsat\].*]{} Further it is easy to see that \[disjointuniisoab\] The isomorphism given in Lemma \[DJDlemma\] induces an isomorphism of line bundles with connections $${{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}({\Sigma}) \cong \pi_1^*{{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}({\Sigma}_1) \otimes \pi_2^*{{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}({\Sigma}_2),\label{isodisjab}$$ where we use the Lagrangian subspaces to fix the connections in all three bundles. The isomorphism is compatible with isomorphism induced by disjoint union of morphism of corresponding labeled marked surfaces. Moreover the preferred sections of the squares of these bundles specified by the given Lagrangian subspaces are compatible with this isomorphism. From this proposition it then follows that we get the corresponding isomorphism of $-\frac{1}{2}c_{{\mathop{\fam0 {\upsilon}}\nolimits}}$-power of these bundles. Combining this with (\[isodisj\]) we now get induced a preferred isomorphism of flat vector bundles $${{{\Cal V}_{\vec \lambda}^{\dagger}}}({{\mathbf \Sigma}}) \cong \pi_1^*\Spofvlamone({{\mathbf \Sigma}}_1) \otimes \pi_2^*\Spofvlamtwo({{\mathbf \Sigma}}_2),$$ which intern induces the required isomorphism of the corresponding vector spaces of covariant constant sections: $$V^{{\mathop{\fam0 {\mathfrak g}}\nolimits}}_{\ell}({{\mathbf \Sigma}},\l) \cong V^{{\mathop{\fam0 {\mathfrak g}}\nolimits}}_{\ell}({{\mathbf \Sigma}}_1,\l_1) \otimes V^{{\mathop{\fam0 {\mathfrak g}}\nolimits}}_{\ell}({{\mathbf \Sigma}}_2,\l_2)$$ which is natural with respect to disjoint union of morphisms. Sheaf of vacua and gluing. {#shofvacandglue} ========================== Let ${{{\mathbf \Sigma}}} = ({\Sigma}, \{p_-,p_+\}\sqcup P,\{v_-,v_+\}\sqcup V,L)$ be a marked surface. Let $$c : P(T_{p_-}{\Sigma}) {\mathop{\fam0 \rightarrow}\nolimits}P(T_{p_+}{\Sigma})$$ be a glueing map and ${\Sigma}_c$ the glueing of ${\Sigma}$ at the ordered pair $((p_-,v_-),(p_+,v_+))$ with respect to $c$ as described in section \[AxiomsMF\]. We shall first assume that $({\Sigma}_c,P)$ is [stable]{}and [saturated.]{} Let ${{{\mathfrak F}}}= (\pi : {{\mathcal C}}{\mathop{\fam0 \rightarrow}\nolimits}\mathcal B; s_-,s_+,{\vec s}; \eta_-,\eta_+,\vec{\eta})$ be a family of pointed Riemann surfaces with formal neighbourhoods on ${{\mathbf \Sigma}}$ over a simply-connected base $\mathcal B$. Let $D$ be the unit disk in the complex plane. Assume we have holomorphic functions $x_\pm : U_\pm \subset {{\mathcal C}}{\mathop{\fam0 \rightarrow}\nolimits}D$ such that for each $b\in \mathcal B$ we have that $x_\pm\mid_{\pi^{-1}(b)} : U_\pm\cap \pi^{-1}(b){\mathop{\fam0 \rightarrow}\nolimits}D$ are local coordinates for $\pi^{-1}(b)$ centered at $p_\pm$ and further that $x_\pm = \eta_\pm$ as formal neighbourhoods. Further we assume that $d_{p_\pm}(x_\pm \mid_{\pi^{-1}(b)})(v_\pm) = 1$ and that $$c = P(d_{p_+}(x_+\mid_{\pi^{-1}(b)}))^{-1} \circ P({\overline{\cdot}}) \circ P(d_{p_-}(x_-\mid_{\pi^{-1}(b)})) : P(T_{p_-}{\Sigma}) {\mathop{\fam0 \rightarrow}\nolimits}P(T_{p_+}{\Sigma})$$ where $P(\overline{\cdot}) : P(T_0 D) {\mathop{\fam0 \rightarrow}\nolimits}P(T_0 D)$ is induced by the the real linear map $z \mapsto {\overline z}$. Assume that $P \subset {\Sigma}- (U_- \cup U_+)$. Set $\mathcal B_c = \mathcal B \times D$ and $\pi_D : \mathcal B_c {\mathop{\fam0 \rightarrow}\nolimits}D $ be the projection onto the second factor. Let us now construct a [stable]{}and [saturated]{}family of pointed curves with formal neighbourhoods ${{{\mathfrak F}}}_{c} = (\pi_c : \mathcal C_c {\mathop{\fam0 \rightarrow}\nolimits}\mathcal B_c,\vec s, \vec \eta)$, in the sense of Definition \[stablefamily\] in the Appendix below, by applying the glueing construction pointwise over $\mathcal B$ to ${{{\mathfrak F}}}$: Let $$\mathcal C^1 = \{(z,w,\tau)\in D^{\times 3} \mid zw = \tau\}$$ $$\mathcal C_c^1 = C^1 \times \mathcal B$$ and $$\mathcal C_c^2 = \{(y,\tau) \in \mathcal C \times D \mid y \in U_\pm \Rightarrow |x_\pm(y)| > |\tau|\}$$ Let then $$\mathcal C_c = \mathcal C_c^1 \cup_{\phi} \mathcal C_c^2,$$ where $$\phi : ((U_- - p_-)\times D \cup (U_+ - p_+) \times D) \cap \mathcal C_c^2 {\mathop{\fam0 \rightarrow}\nolimits}\mathcal C_c^1$$ is given by $$\phi(y,\tau) = \left\{ \begin{array}{ll} (x_-(y), \tau/x_-(y),\tau,\pi(y)), & y \in U_- - p_-\\ (\tau/x_+(y),x_+(y),\tau,\pi(y)), & y \in U_+ - p_+ \end{array}\right. .$$ One easily checks that $\mathcal C_c$ is a smooth complex manifold of dimension $\dim(\mathcal B) + 2$ and that we have an obvious holomorphic projection map $\pi_c : \mathcal C_c {\mathop{\fam0 \rightarrow}\nolimits}\mathcal B_c$. Let ${{{\mathfrak F}}}_{c} = (\pi_c : \mathcal C_c {\mathop{\fam0 \rightarrow}\nolimits}\mathcal B_c,\vec s, \vec \eta)$. The fibers over $\mathcal B \times \{0\}$ are nodal curves, hence ${{{\mathfrak F}}}_c$ is not a family of pointed Riemann surfaces with formal neighbourhoods, however it is a family of pointed stable curves with formal neighbourhoods in the sense of Definition \[stablefamily\] in the Appendix below. If $D^* = D \setminus \{0\}$, then the restricted family ${{{\mathfrak F}}}_c|_{\mathcal B \times D^*}$ is however a family of pointed Riemann surfaces with formal neighbourhoods. Set ${{\tilde D}}= \{ \zeta\in {\mathbb C} | {\mathop{\fam0 Im}\nolimits}(\zeta) > 0\}$. On ${{\tilde D}}$ we now consider the real coordinates $(r,\theta)$ given by $r(\zeta) = \exp(-2 \pi {\mathop{\fam0 Im}\nolimits}(\zeta))$ and $\theta(\zeta) = {\mathop{\fam0 Re}\nolimits}(\zeta)$. Let $(r_\pm,\theta_\pm)$ be $x_\pm$ composed with polar coordinates. Let ${\tilde {\mathcal B}}_c= \mathcal B \times {\tilde D}$ and $p_c : {\tilde {\mathcal B}}_c {\mathop{\fam0 \rightarrow}\nolimits}\mathcal B_c^* = \mathcal B\times D^*$ be given by $p_c(b,\zeta) = (b, \exp(2 \pi i \zeta))$. Then ${\tilde {\mathcal B}}_c$ is the universal cover of ${\mathcal B}_c$. Let ${\tilde {\mathcal C}}_c = p_c^*{\mathcal C}_c$, ${\mathcal C}'_c = {\mathcal C}_c\mid_{{\mathcal B}_c^*}$, ${{\tilde {\mathfrak F}}}_c = p^*_c{{{\mathfrak F}}}_c$, ${\tilde \pi}_c : {\tilde {\mathcal C}}_c {\mathop{\fam0 \rightarrow}\nolimits}{\tilde {\mathcal B}}_c$, ${\tilde \pi}_{\tilde D} : {\tilde {\mathcal B}}_c {\mathop{\fam0 \rightarrow}\nolimits}{\tilde D}$ and ${\tilde \pi}_{\mathcal B} : {\tilde {\mathcal B}}_c {\mathop{\fam0 \rightarrow}\nolimits}{\mathcal B}$. Let $$V_\pm = \Phi_{{{{\mathfrak F}}}}^{-1}(U_\pm)$$ and $$\tx_\pm : V_\pm {\mathop{\fam0 \rightarrow}\nolimits}{\mathcal B}\times D$$ be given by $$\tx_\pm = (\pi_{\mathcal B}, x_\pm \circ \Phi_{{{{\mathfrak F}}}}).$$ Let us now define a fiber preserving diffeomorphism $$f : {\tilde {\mathcal C}}_c {\mathop{\fam0 \rightarrow}\nolimits}{\Sigma}_c\times {\tilde B}_c$$ by $$f(y,r,\theta) = \left\{ \begin{array}{ll} (\tx_-^{-1}(\pi(y),\chi_r(r_-(y)),\theta_-(y) + \frac{1}{2}\frac{1-r_-(y)}{1 - r^{1/2}}\theta), r, \theta) & \mbox{if }y\in U_-, 1\geq r_-(y) \geq r^{1/2} \\ (\tx_+^{-1}(\pi(y),-\chi_r(r_-(y)),-\theta_-(y) - \frac{1}{2}\frac{r_-(y)-r}{ r^{1/2}-r}\theta), r, \theta) & \mbox{if }y\in U_-, r^{1/2}\geq r_-(y) \geq r, \end{array} \right.$$ and extend $f$ to all of ${\tilde {\mathcal C}}_c$ by the map $\Phi_{{{{\mathfrak F}}}}^{-1}\times {\mathop{\fam0 Id}\nolimits}$ on $({\mathcal C}-(U_+\cup U_-))\times {{\tilde D}}$. Here $\chi_r$ is a smooth family of diffeomorphisms $$\chi_r : [1,r] {\mathop{\fam0 \rightarrow}\nolimits}[1,-1], \mbox{ }r \in (0,1),$$ with the properties that $\chi_r$ is the identity near $1$, $\chi_r$ maps $\rho \mapsto -r/\rho$ near $r$ and $\chi_r(r^{1/2}) = 0$ for each $r\in (0,1)$. We will furthermore assume that for all $\rho \in (0,1)$ we have that $$\lim_{r{\mathop{\fam0 \rightarrow}\nolimits}0}\chi_r(\rho) = \rho\mbox{ and } \lim_{r{\mathop{\fam0 \rightarrow}\nolimits}0} \chi_r(r/\rho) = -\rho,$$ for all $\rho \in (0,1).$ The extra conditions on $\chi_r$ implies that the limit $\lim_{r {\mathop{\fam0 \rightarrow}\nolimits}0}q \circ f(\cdot,r,0) : {\Sigma}' {\mathop{\fam0 \rightarrow}\nolimits}{\Sigma}'$ exists and is equal to ${\mathop{\fam0 Id}\nolimits}: {\Sigma}' {\mathop{\fam0 \rightarrow}\nolimits}{\Sigma}'$. We observe that the monodromy $(f|_{\{b\}\times \pi^{-1}(\zeta+1)}) \circ (f|_{\{b\}\times \pi^{-1}(\zeta)})^{-1}$ is a Dehn twist in $P(T_{p_-}{\Sigma})$. Using $f^{-1} : {\Sigma}_c\times {\tilde {\mathcal B}}_c {\mathop{\fam0 \rightarrow}\nolimits}{\tilde {\mathcal C}}_c$ we see that ${{\tilde {\mathfrak F}}}_c$ is a family of [stable]{}and [saturated]{}curves with formal neighbourhoods on ${\Sigma}_c$. The inclusion $${{{\Cal H}_{\vec \lambda}}}(\mathcal B) \hookrightarrow \bigoplus_{\mu \in P_\ell} {\mathcal H}_{\mu, \mu^\dagger, \vec \lambda}(\mathcal B),$$ given by $$|\phi\rangle \mapsto \bigoplus_{\mu \in P_\ell} |0_{\mu, \mu^\dagger} \otimes \phi\rangle$$ induces an isomorphism of vector bundles $${{\Cal V}_{\vec \lambda}^{\dagger}}({{{\mathfrak F}}}_c|_{{\mathcal B}\times \{0\}}) \cong \bigoplus_{\mu \in P_\ell}{{\mathcal V}_{\mu, \mu^{\dagger}, \vec{\lambda}}^{\dagger}}({{{\mathfrak F}}}).\label{factfam}$$ This is the content of Theorem 4.4.9 in [@Ue2]. The [abelian]{}sheaf of vacua construction applied to the family ${{{\mathfrak F}}}_{c}$ gives a holomorphic line bundle ${{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}({{{\mathfrak F}}}_{c})$ over $\mathcal B_c$. This follows from Theorem \[thm5.2ab\] below. We get an isomorphism of vector bundles $${{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}({{{\mathfrak F}}}_c)|_{{\mathcal B}\times \{0\}} \cong {{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}({{{\mathfrak F}}})\label{factfamab}$$ induced by the isomorphism given in Theorem \[thm3.5\] below. The preferred section $s_{{{{\mathfrak F}}}_c}(L_c)$ is continuous over $\mathcal B_c$. Over $\pi^{-1}_D(0)$ it is mapped via the above isomorphism to the preferred section $s_{{{{\mathfrak F}}}}(L)$ of ${{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}({{{\mathfrak F}}})$. This follows by Theorem \[thm6.5\] and \[thm6.7\]. As discussed before the Lagrangian subspace $L$ determines connections in the bundles $\oplus_\mu{{\mathcal V}_{\mu, \mu^{\dagger}, \vec{\lambda}}^{\dagger}}({{{\mathfrak F}}})$ and ${{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}({{{\mathfrak F}}})$. \[nodalfambidif\] The Lagrangian subspace $L_c \subset H_1({\Sigma}_c,{{\mathbb Z}}{})$ determines a unique normalized symmetric bidifferential $\omega_c \in H^0({\mathcal C}_c\times_{{\mathcal B}_c}{\mathcal C}_c, \omega_{{\mathcal C}_c\times_{{\mathcal B}_c}{\mathcal C}_c}(2\Delta))$ specified by formula for any symplectic basis $(\vec \alpha,\vec \beta)$ of $H_1({\Sigma}_c,{{\mathbb Z}}{})$ such that $L_c = {\mathop{\fam0 Span}\nolimits}\{\beta_i\}$. $\square$ By Definition \[conomega\] and Theorem \[curvaturecom\] we get that $\omega_c$ determines a projectively flat connection in ${{\Cal V}_{\vec \lambda}^{\dagger}}({{{\mathfrak F}}}_c)|_{{\mathcal B}_c^*}$ and by Theorem \[thm4.2a\] a connection in ${{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}({{{\mathfrak F}}}_c)|_{{\mathcal B}_c^*}$. Let us now recall the conclusion of the glueing constructions on the sheaf of vacua both in the non-abelian and abelian case applied to the family ${{{\mathfrak F}}}_c$: The explicit formula and Theorem \[thm5.3.4\] below give an isomorphism between sections of ${{\Cal V}_{\vec \lambda}^{\dagger}}({{{\mathfrak F}}}_c)|_{{\mathcal B}\times \{0\}}$ and sections of ${{\Cal V}_{\vec \lambda}^{\dagger}}({{\tilde {\mathfrak F}}}_c)$, which are covariant constant along the fibers of $\tilde \pi_{{{{\tilde D}}}}$. The connection in ${{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}({{{\mathfrak F}}}_c)|_{{\mathcal B}^*_c}$ determined by $\omega_c$ extends to a connection on all of ${{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}({{{\mathfrak F}}}_c)$, hence we get by parallel transport along the fibers of $\pi_{{D}}$ an isomorphism between sections of ${{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}({{{\mathfrak F}}}_c)\mid_{{\mathcal B}\times \{0\}}$ and sections of ${{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}({{{\mathfrak F}}}_c)$, which are covariant constant along the fibers of $\pi_{{D}}$. This follows from Theorem \[glueabcova\] below and formula gives an explicit formula for this isomorphism. By applying the fractional power construction to the line bundle ${{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}({{{\mathfrak F}}}_c)^{\otimes 2}$ with the preferred section $s_{{{{\mathfrak F}}}_c}(L_c)$, we get a line bundle over $\mathcal B_c$, which we denote ${{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}({{{\mathfrak F}}}_c)^{-\frac{1}{2}c_v}(L_c)$. By the very construction of this bundle we see that ${{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}({{{\mathfrak F}}}_c)^{-\frac{1}{2}c_v}(L_c)\mid_{{\mathcal B}\times \{0\}}$ is identified with $({{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}})({{{\mathfrak F}}})^{-\frac{1}{2}c_v}(L)$. We get a connection in this bundle from its construction and an isomorphism from sections of $({{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}})({{{\mathfrak F}}})^{-\frac{1}{2}c_v}(L)$ to sections of ${{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}({{{\mathfrak F}}}_c)^{-\frac{1}{2}c_v}(L_c)$ over $\mathcal B_c$, which are covariant constant along the fibers of $\pi_{{D}}$. \[glueconsistent\] The tensor product of these two glueing constructions gives an isomorphism $I_c({{{\mathfrak F}}},x_\pm)$ from covariant constant sections of $\bigoplus_{\mu\in P_\ell} {{\mathcal V}_{\mu, \mu^{\dagger}, \vec{\lambda}}^{\dagger}}({{{\mathfrak F}}}) \otimes {{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}({{{\mathfrak F}}})^{-\frac{1}{2}c_v}(L)$ over $ {\mathcal B}$ to covariant constant sections of ${{\Cal V}_{\vec \lambda}^{\dagger}}({{\tilde {\mathfrak F}}}_c) \otimes {{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}({{\tilde {\mathfrak F}}}_c)^{-\frac{1}{2}c_v}(L_c)$ over ${\tilde {\mathcal B}}_c$: $$I_c({{{\mathfrak F}}},x_\pm) : \bigoplus_{\mu\in P_\ell} V^{{\mathop{\fam0 {\mathfrak g}}\nolimits}}_{\ell}({{{\mathfrak F}}},\mu,\mu^\dagger,\l) {\mathop{\fam0 \rightarrow}\nolimits}V^{{\mathop{\fam0 {\mathfrak g}}\nolimits}}_{\ell}({{\tilde {\mathfrak F}}}_c,\l).$$ $\square$ Let ${\mathbf C}^{(i)}$, $i=1,2,$ be two complex structures on ${\Sigma}$ and let $x_\pm^{(i)} : U_\pm^{(i)}{\mathop{\fam0 \rightarrow}\nolimits}D$ be coordinates around $p_\pm$ with $d_{p_\pm}x_\pm^{(i)}(v_\pm) = 1$ such that $c = P((d_{p_+}x^{(i)}_+)^{-1} \circ P({\overline{\cdot}}) \circ P(d_{p_-}x^{(i)}_-) : P(T_{p_-}{\Sigma}) {\mathop{\fam0 \rightarrow}\nolimits}P(T_{p_+}{\Sigma})$. Let $\eta_j^{(i)}$ be formal coordinates around $p_j\in C^{(i)}$. \[indepglue\] For such two pairs $({\mathbf C}^{(i)}, x^{(i)}_\pm)$, $i=1,2$, of complex structures and holomorphic coordinates on $({\Sigma}, \{p_+,p_-\}\cup P)$ we have that $$I_c({\mathbf C}^{(1)}, x^{(1)}_\pm) = I_c({\mathbf C}^{(2)}, x^{(2)}_\pm).$$ This follows straight from Theorem \[glueconsistent\], since we clearly have the following \[contfamily\] There exists a family of pointed Riemann surfaces with formal neighbourhoods ${{{\mathfrak F}}}= (\pi : \mathcal C {\mathop{\fam0 \rightarrow}\nolimits}\mathcal B, \vec s, \vec \eta)$ on $({\Sigma},P)$, holomorphic functions $x_\pm : U_\pm \subset \mathcal C {\mathop{\fam0 \rightarrow}\nolimits}D$ and $b_i \in \mathcal B$ $i=1,2$ such that the following holds - The base $\mathcal B$ is simply-connected. - Restriction to the fiber $$(\pi^{-1}(b_i),\vec \eta\mid_{\pi^{-1}(b_i)},x_\pm\mid_{\pi^{-1}(b_i)} : U_\pm \cap \pi^{-1}(b_i){\mathop{\fam0 \rightarrow}\nolimits}D)$$ over $b_i$, $i=1,2$, is the same complex structure on $({\Sigma},P)$ as $$({\mathbf C}^{(i)},\vec \eta^{(i)},x_\pm^{(i)} : U_\pm^{(i)} {\mathop{\fam0 \rightarrow}\nolimits}D)$$ with the same formal coordinates and the same coordinates around $p_\pm$. - For each $b\in \mathcal B$ we have that $x_\pm\mid_{U_\pm\cap\pi^{-1}(b)} : U_\pm\cap\pi^{-1}(b) {\mathop{\fam0 \rightarrow}\nolimits}D$ are holomorphic coordinates around $p_\pm\in \pi^{-1}(b)$. \[IC\] We define the glueing isomorphism $$I_c=I_c({{\mathbf \Sigma}},\l) : \bigoplus_{\mu \in P_\ell} V^{{\mathop{\fam0 {\mathfrak g}}\nolimits}}_{\ell}({{\mathbf \Sigma}},\mu,\mu^\dagger,\l) {\mathop{\fam0 \rightarrow}\nolimits}V^{{\mathop{\fam0 {\mathfrak g}}\nolimits}}_{\ell}({{\mathbf \Sigma}}_c,\l)$$ to be equal to $I_c({\mathbf C}, x_\pm)$ for any pair $({\mathbf C}, x_\pm)$ of a complex structure and holomorphic coordinates on $({\Sigma},\{p_+,p_-\}\cup P)$. Recall that it is assumed that $({\Sigma}_c,P)$ is [stable]{}and [saturated.]{}Let now $({{\mathbf \Sigma}}',\l')$ be a labeled marked surface obtained from $({{\mathbf \Sigma}},\l)$ by labeling further points by $0\in P_\ell$. \[probofvacglue\] We get the following commutative diagram of isomorphisms: $$\begin{CD} \oplus_{\mu\in P_\ell}V^{{\mathop{\fam0 {\mathfrak g}}\nolimits}}_{\ell}({{\mathbf \Sigma}},\mu,\mu^\dagger,\l) @>I_c({{\mathbf \Sigma}},\l)>> V^{{\mathop{\fam0 {\mathfrak g}}\nolimits}}_{\ell}({{\mathbf \Sigma}}_c,\l)\\ @V\cong VV @V\cong VV\\ \oplus_{\mu\in P_\ell}V^{{\mathop{\fam0 {\mathfrak g}}\nolimits}}_{\ell}({{\mathbf \Sigma}}',\mu,\mu^\dagger,\l') @>I_c({{\mathbf \Sigma}}',\l')>> V^{{\mathop{\fam0 {\mathfrak g}}\nolimits}}_{\ell}({{\mathbf \Sigma}}_c', \l'), \end{CD}\label{ccprobofvacglue}$$ where the vertical isomorphisms are the ones gotten from Theorem \[nonstabsat\]. $\square$ \[gluenonstab\] In the cases where $({{\mathbf \Sigma}},\l)$ is not [stable]{}or not [saturated,]{}we define the glueing isomorphism, to be the unique isomorphism $$I_c = I_c({{\mathbf \Sigma}},\l): \oplus_{\mu\in P_\ell}V^{{\mathop{\fam0 {\mathfrak g}}\nolimits}}_{\ell}({{\mathbf \Sigma}},\mu,\mu^\dagger,\l) {\mathop{\fam0 \rightarrow}\nolimits}V^{{\mathop{\fam0 {\mathfrak g}}\nolimits}}_{\ell}({{\mathbf \Sigma}}_c,\l)$$ which makes the diagram (\[ccprobofvacglue\]) commutative for any [stable]{}and [saturated]{}labeled marked surface $({{\mathbf \Sigma}}_c', \l')$ obtained from $({{\mathbf \Sigma}},\l)$ by labeling further points by $0\in P_\ell$. By the naturality of the glueing construction we have that \[gluemorph\] The glueing isomorphism are compatible with the isomorphisms induced by glueing morphisms of marked surfaces. That is suppose $${\mathbf f} : ({\Sigma}^1, \{p^1_-,p^1_+\}\sqcup P^1,\{v^1_-,v^1_+\}\sqcup V^1,L^1) {\mathop{\fam0 \rightarrow}\nolimits}({\Sigma}^2, \{p^2_-,p^2_+\}\sqcup P^2,\{v^2_-,v^2_+\}\sqcup V^2,L^2)$$ is a morphism of marked surfaces and that there are glueing maps $$c_i : P(T_{p^i_-}{\Sigma}^i) {\mathop{\fam0 \rightarrow}\nolimits}P(T_{p^i_+}{\Sigma}^i),$$ such that $(d_{p^1_+}f)^{-1} c_2 d_{p^1_-}f = c_1$, then we get the following commutative diagram $$\begin{CD} \oplus_{\mu\in P_\ell}V^{{\mathop{\fam0 {\mathfrak g}}\nolimits}}_{\ell}({{\mathbf \Sigma}}^1,\mu,\mu^\dagger,\l^1) @>I_{c_1}({{\mathbf \Sigma}}^1,\l^1)>> V^{{\mathop{\fam0 {\mathfrak g}}\nolimits}}_{\ell}({{\mathbf \Sigma}}^1_{c_1},\l^1)\\ @V {V^{{\mathop{\fam0 {\mathfrak g}}\nolimits}}_{\ell}({\mathbf f})} VV @V {V^{{\mathop{\fam0 {\mathfrak g}}\nolimits}}_{\ell}({\mathbf f})} VV\\ \oplus_{\mu\in P_\ell}V^{{\mathop{\fam0 {\mathfrak g}}\nolimits}}_{\ell}({{\mathbf \Sigma}}^2,\mu,\mu^\dagger,\l^2) @>I_{c_2}({{\mathbf \Sigma}}^2,\l^2)>> V^{{\mathop{\fam0 {\mathfrak g}}\nolimits}}_{\ell}({{\mathbf \Sigma}}^2_{c_2},\l^2), \end{CD}$$ for all labelings $\l^i$ of $P^i$ compatible with ${\mathbf f}$. We summarize the results on the glueing construction. \[Glueingiso\] There is an isomorphism $I_c$ from $\oplus_{\mu \in P_\ell} V^{{\mathop{\fam0 {\mathfrak g}}\nolimits}}_{\ell}({{\mathbf \Sigma}},\mu,\mu^\dagger,\l)$ to $V^{{\mathop{\fam0 {\mathfrak g}}\nolimits}}_{\ell}({{\mathbf \Sigma}}_c,\l)$ as specified in Definition \[IC\] and \[gluenonstab\], which is independent of the glueing map $c$ in the following sense: If $c_i : P(T_{p_-}{\Sigma}) {\mathop{\fam0 \rightarrow}\nolimits}P(T_{p_+}{\Sigma})$, $i=1,2$, are glueing maps and $f : {{\mathbf \Sigma}}_{c_1}{\mathop{\fam0 \rightarrow}\nolimits}{{\mathbf \Sigma}}_{c_2}$ is a diffeomorphism as described in Remark \[remarkglue2\], then we have that $$I_{c_2} = V^{{\mathop{\fam0 {\mathfrak g}}\nolimits}}_{\ell}(f) I_{c_1}$$ as isomorphisms from $\oplus_{\mu \in P_\ell} V^{{\mathop{\fam0 {\mathfrak g}}\nolimits}}_{\ell}({{\mathbf \Sigma}},\mu,\mu^\dagger,\l)$ to $V^{{\mathop{\fam0 {\mathfrak g}}\nolimits}}_{\ell}({{\mathbf \Sigma}}_{c_2},\l)$. Moreover the isomorphisms $I_c$ are compatible with glueing of morphisms of labeled marked surfaces. Let $({{\mathbf \Sigma}}',\l')$ be another labeled marked surface, which is [stable]{}and [saturated.]{}Let ${{\mathbf \Sigma}}''$ be the disjoint union of ${{\mathbf \Sigma}}$ and ${{\mathbf \Sigma}}'$. Let ${{\mathbf \Sigma}}_c''$ be the glueing of ${{\mathbf \Sigma}}''$ using the glueing map $c$. Then we clearly have that ${{\mathbf \Sigma}}_c'' = {{\mathbf \Sigma}}_c \sqcup {{\mathbf \Sigma}}'$. It is trivial to check that \[duiglue\] The glueing isomorphism is compatible with the disjoint union isomorphism, namely the following diagram is commutative $$\begin{CD} \bigoplus_{\mu\in P_\ell}V^{{\mathop{\fam0 {\mathfrak g}}\nolimits}}_{\ell}({{\mathbf \Sigma}}'',\mu,\mu^\dagger,\l,\l') @>>> \bigoplus_{\mu\in P_\ell}V^{{\mathop{\fam0 {\mathfrak g}}\nolimits}}_{\ell}({{\mathbf \Sigma}},\mu,\mu^\dagger,\l,)\otimes V^{{\mathop{\fam0 {\mathfrak g}}\nolimits}}_{\ell}({{\mathbf \Sigma}}'\l')\\ @VI_c({{\mathbf \Sigma}}'',\l,\l') VV @VI_c({{\mathbf \Sigma}},\l)\otimes {\mathop{\fam0 Id}\nolimits}VV\\ V^{{\mathop{\fam0 {\mathfrak g}}\nolimits}}_{\ell}({{\mathbf \Sigma}}_c'',\l,\l') @>>> V^{{\mathop{\fam0 {\mathfrak g}}\nolimits}}_{\ell}({{\mathbf \Sigma}}_c,\l)\otimes V^{{\mathop{\fam0 {\mathfrak g}}\nolimits}}_{\ell}({{\mathbf \Sigma}}',\l') . \end{CD}$$ Let now ${{{\mathbf \Sigma}}} = ({\Sigma}, \{p^{(1)}_-,p^{(1)}_+,p^{(2)}_-,p^{(2)}_+\} \sqcup P,\{v^{(1)}_-,v^{(1)}_+,v^{(2)}_-,v^{(2)}_+\}\sqcup V,L)$ be a marked surface. Let $$c^{(i)} : P(T_{p^{(i)}_-}{\Sigma}) {\mathop{\fam0 \rightarrow}\nolimits}P(T_{p^{(i)}_+}{\Sigma})$$ be glueing maps and ${\Sigma}_{c^{(i)}}$ the glueing of ${\Sigma}$ at the ordered pair $((p^{(i)}_-,v^{(i)}_-),(p^{(i)}_+,v^{(i)}_+))$ with respect to $c^{(i)}$. Let ${\Sigma}_{c^{(12)}}$ be the glueing with respect to $c^{(12)} = c^{(1)}\sqcup c^{(2)}$. \[glueglue\] The glueing isomorphisms commute, meaning the following diagram is commutative $$\begin{CD} \bigoplus_{\mu_1,\mu_2\in P_\ell}V^{{\mathop{\fam0 {\mathfrak g}}\nolimits}}_{\ell}({{\mathbf \Sigma}},\mu_1,\mu_1^\dagger,\mu_2,\mu_2^\dagger,\l) @>\bigoplus_{\mu_2\in P_\ell}I_{c^{(1)}}({{\mathbf \Sigma}},\mu_2,\mu_2^\dagger,\l)>> \bigoplus_{\mu_2\in P_\ell}V^{{\mathop{\fam0 {\mathfrak g}}\nolimits}}_{\ell}({{\mathbf \Sigma}}_{c^{(1)}},\mu_2,\mu_2^\dagger,\l)\\ @V\bigoplus_{\mu_1\in P_\ell}I_{c^{(2)}}({{\mathbf \Sigma}},\mu_1,\mu_1^\dagger,\l)VV @V I_{c^{(2)}}({{\mathbf \Sigma}}_{c^{(1)}},\l)VV\\ \bigoplus_{\mu_1\in P_\ell}V^{{\mathop{\fam0 {\mathfrak g}}\nolimits}}_{\ell}({{\mathbf \Sigma}}_{c^{(2)}},\mu_1,\mu_1^\dagger,\l) @>I_{c^{(1)}}({{\mathbf \Sigma}}_{c^{(2)}},\l)>> V^{{\mathop{\fam0 {\mathfrak g}}\nolimits}}_{\ell}({{\mathbf \Sigma}}_{c^{(12)}}, \l). \end{CD}$$ Choose a complex structure on ${\Sigma}$ and let ${\mathbf C}$ denote the resulting marked Riemann surface. The complex structure ${\mathbf C}$ gives a point in the Teichmüller space ${\mathop{\fam0 {\mathcal T}}\nolimits}_{({\Sigma},\{p^{(1)}_-,p^{(1)}_+,p^{(2)}_-,p^{(2)}_+\} \sqcup P)}$. Choose centered coordinates $x^{(i)}_\pm : U_\pm {\mathop{\fam0 \rightarrow}\nolimits}D$ around $p^{(i)}_\pm$ with $d_{p^{(i)}_\pm}x^{(i)}_\pm(v^{(i)}_\pm) = 1$ and such that $c^{(i)} = P(d_{p^{(i)}_+}x^{(i)}_+)^{-1} \circ P({\overline{\cdot}}) \circ P(d_{p^{(i)}_-}x^{(i)}_-) : P(T_{p^{(i)}_-}{\Sigma}) {\mathop{\fam0 \rightarrow}\nolimits}P(T_{p^{(i)}_+}{\Sigma})$. The following construction of a smooth $3$-dimensional complex manifold $\mathcal C$ with a holomorphic map $\pi : \mathcal C {\mathop{\fam0 \rightarrow}\nolimits}D\times D$ is the main ingredient in this proof : Let $$\mathcal C_1 = \{(z^{(i)}_1,w^{(i)}_2,\tau^{(1)},\tau^{(2)})\in ({\mathop{\fam0 {\mathbb C}^{2}}\nolimits}\sqcup {\mathop{\fam0 {\mathbb C}^{2}}\nolimits})\times D\times D \mid z^{(i)}w^{(i)} = \tau^{(i)}, |z^{(i)}| < 1, |w^{(i)}| < 1, |\tau^{(i)}| < 1, i=1,2\}$$ and $$\mathcal C_2 = \{(y,\tau) \in {\Sigma}\times D\times D \mid y \in U^{(i)}_\pm \Rightarrow |x^{(i)}_\pm(y)| > |\tau^{(i)}|\}$$ Let then $$\mathcal C = \mathcal C_1 \cup_{\phi} \mathcal C_2,$$ where $$\phi : ( (U^{(i)}_- - p^{(i)}_-)\times D\times D \cup (U^{(i)}_+ - p^{(i)}_+) \times D\times D )\cap \mathcal C_2 {\mathop{\fam0 \rightarrow}\nolimits}\mathcal C_1$$ is given by $$\phi(y,\tau) = \left\{ \begin{array}{ll} (x^{(i)}_-(y), \tau^{(i)}/x^{(i)}_-(y),\tau^{(1)},\tau^{(2)}), & y \in U^{(i)}_- - p^{(i)}_-\\ (\tau^{(i)}/x^{(i)}_+(y),x^{(i)}_+(y),\tau^{(1)},\tau^{(2)}), & y \in U^{(i)}_+ - p^{(i)}_+ \end{array}\right. .$$ One easily checks that $\mathcal C$ is a smooth complex manifold of dimension $3$ and that we have an obvious holomorphic projection map $\pi : \mathcal C {\mathop{\fam0 \rightarrow}\nolimits}D\times D$. Choose formal neighbourhoods $\vec \eta$ for the points $P$. We thus get a family of stable pointed curves with formal neighbourhoods ${{{\mathfrak F}}}_{c^{(12)}}$ over $D\times D$ obtained by applying the glueing construction at the two pairs $(p_-^{(1)}, p_+^{(1)})$ and $(p_-^{(2)}, p_+^{(2)})$. Let $p = p_1\times p_2 : {{\tilde D}}\times {{\tilde D}}{\mathop{\fam0 \rightarrow}\nolimits}D^*\times D^*$, be the projection and let $\tilde {{{\mathfrak F}}}_{c^{(12)}} = p^*({{{\mathfrak F}}}_{c^{(12)}}|_{D^*\times D^*})$. We denote the two projections onto the first factor by $p^{(1)} : {{\tilde D}}\times {{\tilde D}}{\mathop{\fam0 \rightarrow}\nolimits}{{\tilde D}}$ and onto the second by $p^{(2)} : {{\tilde D}}\times {{\tilde D}}{\mathop{\fam0 \rightarrow}\nolimits}{{\tilde D}}.$ Further let $\tilde {{{\mathfrak F}}}_{c^{(1)}}$ be the pull back under $p_1$ of the normalization of ${{{\mathfrak F}}}_{c^{(12)}}\mid_{D^*\times\{0\}}$ at $[p_-^{(2)}] = [ p_+^{(2)}]$ and $\tilde {{{\mathfrak F}}}_{c^{(2)}}$ be the pull back under $p_2$ of the normalization of ${{{\mathfrak F}}}_{c^{(12)}}\mid_{{\{0\}}\times D^*}$ at $[p_-^{(1)}] = [ p_+^{(1)}]$. Let ${{{\mathfrak X}}}= ({\mathbf C},\{p^{(1)}_-,p^{(1)}_+,p^{(2)}_-,p^{(2)}_+\} \sqcup P, \{x^{(1)}_-,x^{(1)}_+,x^{(2)}_-,x^{(2)}_+\} \sqcup \vec \eta)$. The glueing construction in the non-abelian case applied to ${{{\mathfrak F}}}_{c^{(1)}}$ (respectively to ${{{\mathfrak F}}}_{c^{(2)}}$) and then to ${{{\mathfrak F}}}_{c^{(12)}}$ results in two two-variable versions of for any element of ${\mathcal V}_{\mu_1,\mu_1^\dagger, \mu_2,\mu_2^\dagger, \nu}^{\dagger}({{{\mathfrak X}}})$. Explicitly for an element $\langle \Psi | \in {\mathcal V}_{\mu_1,\mu_1^\dagger, \mu_2,\mu_2^\dagger, \nu}^{\dagger}({{{\mathfrak X}}})$ we get a section $\langle \widehat{\Psi}^{(1)}|$ of $\mathcal{V}_{\mu_2, \mu_2^\dagger,\nu}^\dagger ({{{\mathfrak F}}}_{c^{(1)}})$ given by $$\langle \widehat{\Psi}^{(1)}| \Phi^{(1)}\rangle = \sum_{d=0}^\infty \left\{ \sum_{i=1}^{m_d} \langle \Psi|v^{(1)}_i(d)\otimes v^i_{(1)}(d) \otimes \Phi^{(1)}\rangle \right\}(\tau^{(1)})^{\Delta_{\mu_1} +d},$$ where $\{ v^{(1)}_1(d), \ldots, v^{(1)}_{m_d}(d)\}$ is a basis of $\mathcal{H}_{\mu_1}(d)$ and $\{ v^1_{(1)}(d), \ldots, v^{m_d}_{(1)}(d)\}$ is the dual basis of $\mathcal{H}_{\mu_1^\dagger}(d)$ and $| \Phi^{(1)}\rangle$ is any section of $\mathcal H_{\mu_2,\mu_2^\dagger,\nu}$. The holomorphic section $\langle \widehat{\Psi}^{(1)}|$ is covariant constant along the fibers of $p_1$. Next we construct a holomophic section $\langle \widehat{\Psi}^{(12)}|$ of $\mathcal{V}_\nu^\dagger ({{{\mathfrak F}}}_{c^{(12)}})$ determined by $$\langle \widehat{\Psi}^{(12)}| \Phi\rangle = \sum_{e=0}^\infty \left\{ \sum_{j=1}^{m_e} \langle \widehat{\Psi}^{(1)}|v_j^{(2)}(e)\otimes v^j_{(2)}(e) \otimes \Phi\rangle \right\}(\tau^{(2)})^{\Delta_{\mu_2} +e},$$ where $\{ v^{(2)}_1(e), \ldots, v^{(2)}_{m_e}(e)\}$ is a basis of $\mathcal{H}_{\mu_2}(e)$ and $\{ v^1_{(2)}(e), \ldots, v^{m_e}_{(2)}(e)\}$ is the dual basis of $\mathcal{H}_{\mu_2^\dagger}(e)$ and $| \Phi\rangle$ is any element in $\mathcal H_{\nu}$. This section is covariant constant along the fibers of $p^{(1)}$. Similarly, starting form $\langle \Psi | \in {\mathcal V}_{\mu_1,\mu_1^\dagger, \mu_2,\mu_2^\dagger, \nu}^{\dagger}({{{\mathfrak X}}})$ we can construct a holomorphic section of $\mathcal{V}_{\mu_1,\mu_1^\dagger,\nu}({{{\mathfrak F}}}_{c^{(2)}})$, covariant constant along the fibers of $p_2$, which is determined by $$\langle \widehat{\Psi}^{(2)}| \Phi\rangle = \sum_{e=0}^\infty \left\{ \sum_{j=1}^{m_e} \langle \Psi^{(2)}|v_j^{(2)}(e)\otimes v^j_{(2)}(e) \otimes \Phi\rangle \right\}(\tau^{(2)})^{\Delta_{\mu_2} +e}.$$ Then, starting form $\langle \widehat{\Psi}^{(2)}|$ we can construct a holomorphic section $\langle \widehat{\Psi}^{(21)}|$ of $\mathcal{V}_\nu^\dagger ({{{\mathfrak F}}}_{c^{(12)}})$ which is covariant constant along the fibers of $p^{(2)}$, and given by $$\langle \widehat{\Psi}^{(21)}| \Phi\rangle = \sum_{d=0}^\infty \left\{ \sum_{i=1}^{m_d} \langle \widehat{\Psi}^{(2)}|v^{(1)}_i(d)\otimes v^i_{(1)}(d) \otimes \Phi \rangle \right\}(\tau^{(1)})^{\Delta_{\mu_1} +d}.$$ Now $\langle \widehat{\Psi}^{(12)}|$ and $\langle \widehat{\Psi}^{(21)}|$ is given by the same power series $$\sum_{d=0}^\infty \sum_{i=1}^{m_d} \sum_{j=1}^{m_e} \langle \Psi|v^{(1)}_i(d) \otimes v^i_{(1)}(d) \otimes v^{(2)}_j(e) \otimes v^j_{(2)}(e) \otimes \Phi \rangle (\tau^{(1)})^{\Delta_{\mu_1} +d} (\tau^{(2)})^{\Delta_{\mu_2} +e}.$$ Hence, the two sections $\langle \widehat{\Psi}^{(12)}|$ and $\langle \widehat{\Psi}^{(21)}|$ of $\mathcal{V}_\nu^\dagger ({{{\mathfrak F}}}_{c^{(12)}})$ coincide, and in fact they are covariant constant. From this we in particular see that the connection in ${\mathcal V}_{ \nu}^{\dagger}(\tilde {{{\mathfrak F}}}_{c^{(12)}})$ is flat. By applying the same argument to the abelian theory, we get that the glueing construction is also independent of the order of the glueing in the abelian case and the connection in ${{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}({{{\mathfrak F}}}_{c^{(12)}})$ is flat over $D\times D$. The theorem now follows. $\square$ Verification of the axioms {#verification} ========================== It is now straight forward to check the axioms of a modular functor given the results obtained in the previous sections. \[Main\] The functor $V^{{\mathop{\fam0 {\mathfrak g}}\nolimits}}_{\ell}$ from the category of labeled marked surfaces to the category finite dimensional vector spaces is a modular functor. In order to check axiom [*MF1*]{}, we only need to check that the disjoint union isomorphisms satisfies associativity, but this follows from associativity of the isomorphisms between the corresponding ${\mathcal H}^\dagger$’s and ${\mathcal F}$’s. We have that the glueing isomorphism $I_c$ from Theorem \[Glueingiso\] is compatible with - The disjoint union isomorphisms: Proposition \[duiglue\]. - The glueing isomorphisms them self, i.e. the glueing isomorphisms should commute: Theorem \[glueglue\] Hence axiom [*MF2*]{} is checked. Axiom [*MF3*]{} is trivial, since we define $V^{{\mathop{\fam0 {\mathfrak g}}\nolimits}}_{\ell}(\emptyset) = {\mathop{\fam0 {\mathbb C}^{}}\nolimits}$. Axiom [*MF4*]{} and [*MF5*]{} follows from Corollary 3.5.2 (1) and (2) in [@Ue2]. $\square$ Appendix. Families of stable curves and glueing {#App} =============================================== In order to define the functor $V^{{\mathop{\fam0 {\mathfrak g}}\nolimits}}_{\ell}$, we only need to refer to Riemann Surface and families of such which form smooth complex manifolds. However, in defining the glueing morphisms we also need to consider the so called [*stable nodal curves*]{}. These are one dimensional algebraic sub-varieties (algebraic curves) of complex projective space, such that the singularities are locally analytically isomorphic to a neighbourhood of the origin of $xy=0$. A neighbourhood of a node is obtained by patching together at the origins of two small disks $D_1=\{\; x; |\; |x|< \varepsilon_1\;\}$ and $D_2=\{\; y; |\; |y|< \varepsilon_2\;\}$. By reversing the process, from a neighbourhood of a node we obtain two disconnected small disks. This process is called normalization or desingularization of the node. Thus, by a normalization of a nodal curve $C$ we obtain a compact Riemann surface $\widetilde{C}$ and holomorphic map $\nu : \widetilde{C} \rightarrow C$ such that for a node $P$, the inverse image $\nu^{-1}(P)$ consists of two distinct points $P_+$ and $P_-$. A curve which has only nodes as singularities is called a nodal curve. Stable curves ------------- We begin by introducing the notion of stable curves. \[stablecurve\] The data ${\mathfrak X} = (C;\, Q_1, Q_2, \ldots , Q_N)$ consisting of an algebraic curve $C$ and points $Q_1, \ldots, Q_N$ on $C$ are called an [*($N$-)pointed stable curve,*]{} if the following conditions are satisfied. 1. The curve $C$ is a nodal curve. 2. $Q_1,Q_2, \ldots, Q_N$ are non-singular points of the curve $C$. 3. If an irreducible component $C_i$ is a Riemann sphere ${\mathbf P}^1$ (resp. a rational curve with one double point, resp. an elliptic curve), the sum of the number of intersection points of $C_i$ and other components and the number of $Q_j$’s on $C_i$ is at least three (resp. one, resp. one). 4. $\dim_{{\mathbf C}}H^1(C,\mathcal O_C) = g$. Note that the condition (3) is equivalent to saying that the Euler characteristic of each component of the complement of the points $Q_j$ on it and the nodes is negative. A pointed stable curve with formal neighbourhoods is defined in an analogous way as a pointed Riemann surface with formal neighbourhoods (see Definition \[v1order\]). Also we can define a family of pointed stable curves with formal neighbourhoods. \[stablefamily\] The data ${{{\mathfrak F}}}= (\pi : \mathcal{C} \rightarrow \mathcal{B}; s_1,s_2,\ldots, s_N; {\eta}_1 , {\eta}_2 , \ldots, {\eta}_N )$ is called a family of ($N$-)pointed stable curves of genus $g$ with formal neighbourhoods, if the following conditions are satisfied. 1. Both $\mathcal{C}$ and $\mathcal{B}$ are connected complex manifolds, $\pi : \mathcal{C} \rightarrow \mathcal{B}$ is a proper flat holomorphic map and $s_1,s_2, \ldots,s_N$ are holomorphic sections of $\pi$. 2. For each point $b \in \mathcal{B}$ the data $(\mathcal{C}_b :=\pi^{-1}(b); s_1(b), s_2(b), \ldots, s_N(b))$ is an $N$-pointed stable curve of genus $g$. 3. For each $j$, ${\eta}_j $ is an $\mathcal O_{\mathcal{B}}$-algebra isomorphism $${\eta}_j : \widehat{\mathcal O}_{_{\mathcal{C}}/s_j} = \varprojlim_{n \to \infty} \mathcal O_{\mathcal{C}}/I_{j}^{n} \simeq \mathcal O_{\mathcal{B}}[[\xi]],$$ where $I_{j}$ is the defining ideal of $s_j(\mathcal{B})$ in $Y$. The only families of pointed stable curves we need in this paper are all constructed explicitly from families of pointed Riemann surfaces via the glueing process discussed in the begining of section \[shofvacandglue\]. Sheaf of vacua for a family of pointed stable curves ---------------------------------------------------- For a family of stable curves with formal neighbourhoods ${{{\mathfrak F}}}= ( \pi : \mathcal{C} \rightarrow \mathcal{B} ; \vec{s}; \vec{\eta})$ we can define the sheaf of vacua ${{\Cal V}_{\vec \lambda}^{\dagger}}({{{\mathfrak F}}})$ and the sheaf of covacua ${{{\Cal V}_{\vec \lambda}}}({{{\mathfrak F}}})$ just as in Definition \[ShofVdef\]. If the family contains smooth curves (Riemann surfaces), then we have the connection given by Definition \[conomega\] on the complement of the locus of nodal curves and the connection has a regular singularity along this locus (see §5.3 of [@Ue2]). Using this connection with regular singularities it is shown in [@Ue2], that Theorem \[localfreeness\] is also valid for a family of stable curves. \[stabllocalfreeness\] For a stable family of stable curves with formal neighbourhoods ${{{\mathfrak F}}}= ( \pi : \mathcal{C} \rightarrow \mathcal{B} ; \vec{s}; \vec{\eta})$, the sheaves $ {{\Cal V}_{\vec \lambda}^{\dagger}}({{{\mathfrak F}}})$ and ${{{\Cal V}_{\vec \lambda}}}({{{\mathfrak F}}})$ are locally free sheaves of ${{{\Cal O}_{\Cal B}}}$-modules of finite rank over $\mathcal B$. They are duals to of each other. We include this theorem here for completeness. We do not need this result for the constructions in this paper. The sheaf of abelian vacua associated to families of pointed stable curves -------------------------------------------------------------------------- For a family of stable curves with formal neighbourhoods ${{{\mathfrak F}}}= ( \pi : \mathcal{C} \rightarrow \mathcal{B} ; \vec{s}; \vec{\eta})$ we can define the sheaf of abelian vacua ${{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}({{{\mathfrak F}}})$ just as in Definition \[dfn4.1\]. We have following result. \[thm5.2ab\] The abelian vacua construction applied to ${{{\mathfrak F}}}$ gives a holomorphic line bundle ${{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}({{{\mathfrak F}}})$ over $\mathcal B$. Next let us consider a nodal curve $C$ with node $P$. Let $\widetilde{C}$ be the Riemann surface obtained by resolving the singularity at $P$ and let $\pi : \widetilde{C} \rightarrow C$ be the natural holomorphic mapping. Then $\pi^{-1}(P)$ consists of two points $P_+$ and $P_-$. Let $$\gX=(C;q_1,\ldots,q_N;\xi_1,\ldots,\xi_N)$$ be a pointed nodal curve with formal neighbourhoods and we let $$\widetilde{\gX} = (\widetilde{C};P_+,P_-,q_1,\ldots,q_N;z,w, \xi_1,\ldots,\xi_N)$$ be the associated pointed Riemann surface with formal neighbourhoods. Define an element $|0_{+,-}\rangle \in \cF\otimes \cF$ by $$\label{0+-} |0_{+,-}\rangle = |0\rangle \otimes |-1\rangle - |-1\rangle \otimes |0 \rangle .$$ The natural inclusion $$\begin{aligned} \cF_N & \hookrightarrow & \cF_{N+2} \\ |u \rangle & \mapsto & |0_{+,-}\rangle \otimes |u \rangle\end{aligned}$$ defines a natural linear mapping $$\iota_{+,-}^* : \cFd_{N+2} \rightarrow \cFd_N .$$ \[thm3.5\] The natural mapping $\iota^*_{+,-}$ induces a natural isomorphism $${{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}(\widetilde{\gX}) \cong {{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}(\gX).$$ Let us now consider the case where we have one marked point on $C$ $$\gX=(C;Q;\xi)$$ and we let $$\widetilde{\gX} = (\widetilde{C};P_+,P_-,Q;z,w, \xi)$$ be the associated 3-pointed curve with formal neighbourhoods. We can define $\langle \omega(\gX,\{\alpha, \beta\})|$ similar to the non-singular case, by choosing a basis $$(\vec{\alpha}, \vec{\beta})=(\alpha_1, \ldots, \alpha_{g-1},\alpha_g, \beta_1, \ldots, \beta_{g-1})$$ of $H_1(C,\bZ)$, in such a way that $\alpha_1$, $\alpha_2, \ldots,\alpha_{g-1}$ and $\beta_1$, $\beta_2, \ldots, \beta_{g-1}$ is the image of a symplectic basis of $H_1(\widetilde{C},\bZ)$ under natural map to $H_1(C,\bZ)$ and $\alpha_g$ corresponds to the invariant cycle of a flat deformation of the curve $C$. Then we can choose a basis $\{\omega_1, \ldots, \omega_{g-1}, \omega_g, \omega_{g+1}, \omega_{g+2}, \ldots \}$ of $H^0(C, \omega(*Q))$ such that $\{\pi^*\omega_1, \ldots, \pi^*\omega_{g-1}, \pi^*\omega_{g+1}, \pi^*\omega_{g+2}, \ldots \}$ is a normalized basis of $H^0(\widetilde{C}, \omega_{\widetilde{C}}(*Q))$ as in (\[betaone2\]), (\[omegaQ1\]) and (\[omegaQ\]) where we put $$\pi^*\omega_{g+n} = \omega_Q^{(n)}, \quad n=1,2, \ldots,$$ and $\pi^*\omega_g$ is a meromorphic one-form on $\widetilde{C}$ which has poles of order one at $P_+$ and $P_-$ with residue $-1$ and 1, respectively, is holomorphic outside $P_\pm$ and $$\int_{P_+}^{P_-}\pi^*\omega_g = 1.$$ Then put $$\langle \omega(\gX,\{\alpha, \beta\})| = \langle \cdots \wedge e(\omega_{m} )\wedge \cdots \wedge e(\omega_2)\wedge e(\omega_1) |.$$ The proof of Lemma 3.1 of [@AU1] applies also in this case and shows that $\langle \omega(\gX,\{\alpha, \beta\})|$ is an element of ${{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}(\gX)$. Let $$\widehat{\gX} = (\widetilde{C}, Q; \xi) .$$ Then, by applying Theorem \[thm3.4\] at the points $P_\pm$ we have a canonical isomorphism $$\iota^* : {{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}(\widetilde{\gX}) \cong {{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}(\widehat{\gX}) .$$ \[thm6.5\] Under the above assumptions and notation we have that $$\iota^* \circ (\iota_{+,-}^*)^{-1} (\langle \omega(\gX,\{\alpha, \beta\}) |) = (-1)^g \langle \omega(\widehat{\gX},\{\widehat{\alpha}, \widehat{\beta}\})|$$ where $\{\widehat{\alpha}, \widehat{\beta}\}= \{\alpha_1, \ldots, \alpha_{g-1}, \beta_1, \ldots, \beta_{g-1}\}$. Let $\gF=(\pi : \cC\rightarrow \cB, s_1, \ldots, s_N, \xi_1, \ldots, \xi_N)$ be a family of $N$-pointed Riemann surfaces. For any point $b \in \cB$ there exists an open neighbourhood $U_b$ such that $\pi^{-1}(U_b)$ is topologically trivial so that we can choose smoothly varying symplectic bases $$(\alpha_1(t), \ldots, \alpha_g(t), \beta_1(t), \ldots, \beta_g(t)), \quad t \in \cB.$$ Then we can define, using propagation of vacua, $$\langle \omega(\gX_t, \{\alpha(t), \beta(t)\})| \in {{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}(\gF)_t$$ where $\gX_t=(\pi^{-1}(t), s_1(t), \xi_1)$. \[thm6.6\] The section $\langle \omega(\gX_t, \{\alpha(t), \beta(t)\})|$ is a holomorphic section of ${{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}(\gF)$ over $U_b$. The glueing of vacua construction --------------------------------- Let us briefly recall the glueing of vacua. First let us consider the non-abelian case. We use freely the notation in §\[shofvacandglue\]. Let ${{{\mathfrak F}}}= (\pi : {{\mathcal C}}{\mathop{\fam0 \rightarrow}\nolimits}\mathcal B; s_-,s_+,{\vec s}; \eta_-,\eta_+,\vec{\eta})$ be a family of pointed Riemann surfaces with formal neighbourhoods on ${{\mathbf \Sigma}}$ over a simply-connected base $\mathcal B$ and we let ${{{\mathfrak F}}}_{c} = (\pi_c : \mathcal C_c {\mathop{\fam0 \rightarrow}\nolimits}\mathcal B_c,\vec s, \vec \eta)$ be a [stable]{}and [saturated]{}family of $N$-pointed curves with formal neighbourhoods obtained by applying the glueing construction pointwise over $\mathcal B$ to ${{{\mathfrak F}}}$ where $\mathcal{B}_c = \mathcal{B} \times D$ with the unit disk $D$. By the isomorphism holomorphic sections of $\oplus_{\mu}{{\mathcal V}_{\mu, \mu^{\dagger}, \vec{\lambda}}^{\dagger}}({{{\mathfrak F}}})$ over $\mathcal {B}$ may be regarded as holomorphic sections of ${{\Cal V}_{\vec \lambda}^{\dagger}}({{{\mathfrak F}}}_c|_{{\mathcal B}\times \{0\}})$ over $\mathcal{B} \times \{0\}$. The glueing of vacua is an isomorphism from holomorphic section of $\oplus_{\mu}{{\mathcal V}_{\mu, \mu^{\dagger}, \vec{\lambda}}^{\dagger}}({{{\mathfrak F}}})$ over $\mathcal{B}$ to a holomorphic sections of ${{\Cal V}_{\vec \lambda}^{\dagger}}({\tilde {{{\mathfrak F}}}}_c)$ over ${\tilde {\mathcal B}}_c$, which is covariant constant along the direction of ${{\tilde D}}$. The construction is as follows. By Lemma \[L2.2.12\] we can choose a basis $\{ v_1(d), \ldots, v_{m_d}(d)\}$ of $\mathcal{H}_\mu(d)$ and the dual basis $\{ v^1(d), \ldots, v^{m_d}(d)\}$ of $\mathcal{H}_{\mu^\dagger}(d)$ such that $$\label{nonabelianpair} (v^j(d)|v_k(d)) = \delta^j_k .$$ For a holomorphic section $ \langle \Psi| \in {{\mathcal V}_{\mu, \mu^{\dagger}, \vec{\lambda}}^{\dagger}}({{{\mathfrak F}}})$ over $\mathcal{B}$ we define a formal series $\langle \widehat{\Psi}|$ by $$\label{formalsolution1} \langle \widehat{\Psi}| \Phi\rangle = \sum_{d=0}^\infty \left\{ \sum_{i=1}^{m_d} \langle \Psi|v_i(d)\otimes v^i(d) \otimes \Phi\rangle \right\}\tau^{\Delta_\mu +d},$$ for all $| \Phi\rangle \in \mathcal H_{\vec \lambda}.$ Here the fractional power $\tau^{\Delta_\mu +d}$ is clearly well defined on ${{\tilde D}}$. This formal power series converges and defines in fact a holomorphic section of ${{\Cal V}_{\vec \lambda}^{\dagger}}({{{\mathfrak F}}}_c)$. Namely, we have the following theorem. \[thm5.3.4\] The formal power series $\langle \widehat{\Psi}|$ is a formal solution of the differential equation $$\label{fuchsian} \left(\tau \frac{d}{d\tau} - T[\vec{l}] + a(\vec{l})\right) \langle \widehat{\Psi}|=0,$$ of Fuchsian type where $$\vec{l}=\left( l_1(\xi_1)\frac{d}{d\xi_1}, \ldots, l_N(\xi_N) \frac{d}{d\xi_N} \right)$$ is an $N$-tuple of formal vector fields such that $\theta(l) =\tau \dfrac{d}{d\tau}$. Moreover, $\langle \widehat{\Psi}|$ converges and defines a holomorphic section of ${{\Cal V}_{\vec \lambda}^{\dagger}}(\tilde {{{\mathfrak F}}}_c)$ over ${\tilde {\mathcal B}}_c$. The differential equation gives the holomorphic connection along the direction of ${{\tilde D}}$. Hence, the above formal power series is a holomorphic section of ${{\Cal V}_{\vec \lambda}^{\dagger}}({{{\mathfrak F}}}_c)$ which is covariant constant along the direction of ${{\tilde D}}$. Let us now discuss the glueing in the abelain case. Recall we have a perfect pairing $$\{\phantom{X}|\phantom{X}\}_+ : \mathcal{F}_d(p) \times \mathcal{F}_d(-p-1) .$$ Let $\{v_i(d,p)\}_{i=1,\ldots,m_d}$ be a basis of $\cF_d(p)$ for any $p \in \bZ$ and $\{v^i(d,p)\}_{i=1,\ldots,m_d}$ be the dual basis of $\cF_d(-p-1)$ with respect to the pairing $\{\phantom{X}|\phantom{X}\}_+$. For a holomorphic section $\langle \psi|$ of ${{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}({{{\mathfrak F}}})$ define $\langle \widetilde{\psi} |$ by $$\label{formalsolution2} \langle \widetilde{\psi} | u \rangle = \sum_{p \in \bZ}\Bigl\{ \sum_{d=0}^\infty \sum_{i=1}^{m_d}(-1)^{p+d} \langle \psi |v_i(d,p)\otimes v^i(d,-p-1)\otimes u\rangle \Bigr\}\tau^{d+p(p+1)/2}.$$ Then the formal power series converges and defines a holomorphic section of $\langle \widetilde{\psi}| \in {{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}({{{\mathfrak F}}}_c)$. Moreover this section is covariant constant along the directions of $D$. \[glueabcova\] The glueing construction gives an isomorphism between sections of ${{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}({{{\mathfrak F}}})$ and sections of ${{\mathcal V}_{{\mathop{\fam0 ab}\nolimits}}^{\dagger}}({{{\mathfrak F}}}_c)$ which are covariant constant along the directions of $D$. This theorem follows directly form Theorem 5.3 in [@AU1]. Finally, we analyze the preferred section for the families. Suppose we have a continuous basis $(\alpha_i(t),\beta_i(t))$ of $H_1(\pi^{-1}(t),\bZ)$, $t\in (0,1)\subset D$, such that we get a well defined limit as $t$ goes to zero, which gives a symplectic basis, say $(\alpha_1(0),\ldots, \alpha_{g-1}(0),\alpha_{g}(0),\beta_{1}(0),\ldots,\beta_{g-1}(0))$ of $H_1(C_0,\bZ)$ as described above for nodal curves and $\beta_g(0) = 0$. Let $\gX_t=(\pi^{-1}(t), s_1(t), \ldots s_N(t), \xi_1, \ldots \xi_N)$. \[thm6.7\] We have that $$\langle \omega(\gX_0, \{\alpha(0), \beta(0)\})| = \lim_{t\to 0} \langle \omega(\gX_t, \{\alpha(t), \beta(t)\})|.$$ [99]{} J. E. Andersen, The Witten invariant of finite oder mapping tori I, To appear in [*Journal für Reine und Angewandte Matematik*]{}. J. E. Andersen & K. Ueno, Abelian Conformal Field Theory and Determinant Bundles, MPS-Preprint 2003 – 5, math.QA/0304135. J. E. Andersen & K. Ueno, Construction of the Reshetikhin-Turaev TQFT from conformal field theory, Preprint in preparation. M.F. Atiyah, On Framings of 3-manifolds, [*Topology*]{} 29 (1990) 1-7 B. Bakalov; A. Jr. Kirillov, Lectures on tensor categories and modular functors, [*University Lecture Series*]{}, 21. American Mathematical Society, Providence, RI, 2001. C. Blanchet, Hecke algebras, modular categories and $3$-manifolds quantum invariants, [*Topology*]{} 39 no. 1, 193–223, 2000. C. Blanchet; N. Habegger; G. Masbaum; P. Vogel, Three-manifold invariants derived from the Kauffman bracket, [*Topology*]{} 31 no. 4, 685–699, 1992. C. Blanchet; N. Habegger; G. Masbaum; P. Vogel, Topological quantum field theories derived from the Kauffman bracket, [*Topology*]{} 34 no. 4, 883–927, 1995. J. D. Fay, Theta functions on Riemann surfaces, [*Lecture Notes in Math.*]{}, 352, Springer-Verlag, 1973. G. Felder, Brst approach to minimal models, [*Nucl. Phys. B*]{}, 317, 215–236, 1989. J. Grove, Constructing TQFTs from modular functors, [*J. Knot Theory Ramifications*]{} 10 no. 8, 1085–1131, 2001. V. Kac, Infinite dimensional Lie algebras, third edition, [*Cambridge University Press*]{}, 1990. N. Kawamoto, Y. Namikawa, A. Tsuchiya and Y. Yamada, Geometric realization of conformal field theory on Riemann surfaces, [*Comm. Math. Phys.*]{}, 116, 247–308, 1988. M. Kontsevich, Rational conformal field theory and invariants of 3-manifolds, [*Preprint of Centre de Physique Theorique Marseille,*]{} CPT-88/p2189, 1988. G. Segal, The definition of conformal field theory, [*preprint*]{} 1992. A. Tsuchiya, K. Ueno and Y. Yamada", Conformal Field Theory on Universal Family of Stable Curves with Gauge Symmetries, [*Advanced Studies in Pure Mathematics*]{} 19, 459–566, 1989. V. Turaev, Quantum invariants of knots and 3-manifolds, W. de Gruyter, Berlin, 1994.. K. Ueno, On conformal field theory, [*London Math. Soc. Lecture Note*]{} 208, 283–345, 1995. K. Ueno, Introduction to conformal field theory with gauge symmetries, [*Geometry and physics (Aarhus, 1995), Lecture Notes in Pure and Appl. Math.*]{} 184, 603–745, Dekker, New York, 1997. K. Walker, On Witten’s 3-manifold invariants, [*Preliminary version \# 2, Preprint*]{} 1991. C.T.C. Wall, Non-additivity of the signature, [*Invent. Math.*]{} 7 (1969) 269-274 [^1]: This research was conducted partly by the first author for the Clay Mathematics Institute at University of California, Berkeley and for MaPhySto – A Network for Mathematical Physics and Stochastics, funded by The Danish National Research Foundation. The second author is partially supported by Grant in Aid for Scientific Research [no]{}. 14102001 of JSPS

ArXiv

--- abstract: 'In spatially periodic Hermitian systems, such as electronic systems in crystals, the band structure is described by the band theory in terms of the Bloch wave functions, which reproduces energy levels for large systems with open boundaries. In this paper, we establish a generalized Bloch band theory in one-dimensional spatially periodic tight-binding models. We show how to define the Brillouin zone in non-Hermitian systems. From this Brillouin zone, one can calculate continuum bands, which reproduce the band structure in an open chain. As an example we apply our theory to the non-Hermitian SSH model. We also show the bulk-edge correspondence between the winding number and existence of the topological edge states.' author: - Kazuki Yokomizo - Shuichi Murakami title: 'Bloch Band Theory for Non-Hermitian Systems' --- The band theory in crystals is fundamental for describing electronic structure [@Kittel1996]. By introducing the Bloch wave vector ${\bm k}$, the band structure calculated within a unit cell reproduces that of a large crystal with open boundaries. Here it is implicitly assumed that the electronic states are almost equivalent between a system with open boundaries and that with the periodic boundaries, represented by the Bloch wave function with real ${\bm k}$. It is because the electronic states extend over the system. Recently, non-Hermitian systems, which are described by non-Hermitian Hamiltonians have been attracting much attention. These systems have been both theoretically and experimentally studied in many fields of physics [@Gonzalez2017; @Kozii2017; @Zyuzin2018; @Shen2018v2; @Molina2018; @Yoshida2018; @Carlstrom2018; @Yoshida2018ex; @Philip2018; @Wu2018; @Chen2018; @Moors2019; @Okugawa2019; @Budich2019; @Yang2019; @San2016; @Zeng2016; @Li2018; @Kawabata2018s; @Ezawa2019; @Guo2009; @Ruter2010; @Feng2011; @Regensburger2012; @Feng2013; @Poli2015; @Zhen2015; @Zhao2015; @Ding2015; @Weimann2017; @Hodaei2017; @Chen2017; @St2017; @Bahari2017; @Wang2018; @Zhou2018; @Parto2018; @Zhao2018; @Harari2018; @Bandres2018; @Pan2018; @Jin2018; @Malzard2018; @Oztas2018; @Kremer2019; @Bliokh2019; @Wang2019nat; @Lee2014; @Li2016; @Xu2017; @Ashida2017; @Gong2018; @Nakagawa2018; @Takata2018; @Pan2019; @Li2019; @Liu2019; @Runder2009; @Zeuner2015; @Mochizuki2016; @Xiao2017; @Hatano1996; @Hatano1997; @Hatano1998; @Lourenco2018; @Gong2018; @Rosenthal2018; @Ezawa2018; @Luo2018; @Ezawa2018v2; @Wang2018v]. In particular, the bulk-edge correspondence has been intensively studied in topological systems. In contrast to Hermitian systems, it seems to be violated in some cases. The reasons for this violation have been under debate [@Hu2011; @Esaki2011; @Lee2016; @Leykam2017; @Martinez2018; @Ye2018; @Shen2018; @Yuce2018; @Yin2018; @Yuce2018v2; @Kunst2018; @Yao2018; @Gong2018; @Yao20182d; @Kawabata2018b; @Yuce2018v3; @Kawabata2019; @Jin2019; @Wang2019; @Ozcakmakli2019; @Edvardsson2019]. One of the controversy is that in many previous works, the Bloch wave vector has been treated as real in non-Hermitian systems similarly to Hermitian ones. In Ref. , it was proposed that in one-dimensional (1D) non-Hermitian systems, the wave number $k$ becomes complex. The value of $\beta\equiv{\rm e}^{ik}$ is confined on a loop on the complex plane, and this loop is a generalization of the Brillouin zone in Hermitian systems. In non-Hermitian systems, the wave functions in large systems with open boundaries do not necessarily extend over the bulk, but are localized at the either end of the chain unlike those in Hermitian systems. Thus far, how to obtain the generalized Brillouin zone has been known only for simple systems. In this paper, we establish a generalized Bloch band theory to calculate continuum bands in a 1D tight-binding model. First of all, we establish a way to determine the generalized Brillouin zone $C_\beta$ for $\beta\equiv{\rm e}^{ik}$, $k\in{\mathbb C}$. This determines continuum bands, which reproduce band structure for a large crystal with open boundaries. We introduce the “Bloch” Hamiltonian $H\left(k\right)$ and rewrite it in terms of $\beta$ as $H\left(\beta\right)$. Then the characteristic equation $f\left(\beta,E\right)=0$, which is an eigenvalue equation for $H\left(\beta\right)$, is an algebraic equation for $\beta$, and let $2M$ be the degree of the equation. We find that when the characteristic equation has solutions $\beta_i~(i=1,\cdots,2M)$ with $\left|\beta_1\right|\leq\left|\beta_2\right|\leq\cdots\leq\left|\beta_{2M-1}\right|\leq\left|\beta_{2M}\right|$, $C_\beta$ is given by the trajectory of $\beta_M$ and $\beta_{M+1}$ under a condition $\left|\beta_M\right|=\left|\beta_{M+1}\right|$. Furthermore, one can get continuum bands by diagonalizing $H\left(\beta\right)$ with $\beta\in~C_{\beta}$. We note that in Hermitian systems, this condition reduces to $C_\beta:\left|\beta\right|=1$, meaning that the Bloch wave number $k$ being real. In Ref. , the case with $M=1$ has been solved, and we give a nontrivial extension into general $M$. A byproduct of our theory is that one can prove the bulk-edge correspondence. The bulk-edge correspondence has been discussed, but in most cases it has not been shown rigorously but by observation on some particular cases, together with an analogy to Hermitian systems. It in fact shows that the bulk-edge correspondence for the real Bloch wave vector cannot be true in non-Hermitian systems. In this paper, we show the bulk-edge correspondence in the non-Hermitian SSH model with the generalized Brillouin zone, and discuss the relationship between a topological invariant in the bulk and existence of the edge states. We start with a 1D tight-binding model, with its Hamiltonian given by $$H=\sum_n\sum_{i=-N}^N\sum_{\mu,\nu=1}^qt_{i,\mu\nu}c_{n+i,\mu}^\dag c_{n,\nu}, \label{eq1}$$ where $N$ represents the range of the hopping and $q$ represents internal degrees of freedom per unit cell. This Hamiltonian can be non-Hermitian, meaning that $t_{i,\mu\nu}$ is not necessarily equal to $t_{-i,\nu\mu}^\ast$. Then one can write the real-space eigen-equation as $H\ket{\psi}=E\ket{\psi}$, where the eigenvector is written as $\ket{\psi}=\left(\psi_{1,1},\cdots,\psi_{1,q},\cdots,\psi_{L,1},\cdots,\psi_{L,q}\right)^{\rm T}$ in an open chain. Thanks to the spatial periodicity, one can write the eigenvector as a linear combination: $$\psi_{n,\mu}=\sum_j\phi_{n,\mu}^{\left(j\right)},~\phi_{n,\mu}^{\left(j\right)}=\left(\beta_j\right)^n\phi_\mu^{\left(j\right)},~(\mu=1,\cdots,q). \label{eq3}$$ By imposing that $\phi^{\left(j\right)}_{n,\mu}$ is a bulk eigenstate, one can obtain the characteristic equation for $\beta=\beta_j$ as $$f\left(\beta,E\right)=0. \label{eq4}$$ This characteristic equation is nothing but an eigenvalue equation for the generalized Bloch Hamiltonian $H\left(\beta\right)$, and therefore it can be obtained similarly to Hermitian cases. Equation (\[eq4\]) is an algebraic equation for $\beta$ with an even degree $2M$ in general cases. ![\[fig:band\]Schematic figure of the band structure (a) in a finite open chain with various system sizes $L$, and (b) of the generalized Bloch Hamiltonian. The vertical axis represents the distribution of the complex energy $E$.](band2.eps){width="8.5cm"} One can see from Eq. (\[eq3\]) that $\beta$ corresponds to the Bloch wave number $k\in{\mathbb R}$ via $\beta={\rm e}^{ik}$ in Hermitian systems. The bulk-band structure for reality of $k$ reproduces the band structure of a long open chain. When extending this idea to non-Hermitian systems, we should choose the values of $\beta$ such that the bands of the Hamiltonian $H\left(\beta\right)$ reproduce those of a long open chain (Fig. \[fig:band\]). The levels are discrete in a finite open chain, and as the system size becomes larger, the levels become dense and asymptotically form continuum bands (Fig. \[fig:band\]). Therefore in order to find the generalized Brillouin zone $C_\beta$, one should consider asymptotic behavior of level distributions in an open chain in the limit of a large system size. In Hermitian systems $\left|\beta\right|$ is equal to unity, meaning that the eigenstates extend over the bulk. On the other hand, in non-Hermitian systems, $\left|\beta\right|$ is not necessarily unity, and these states may be localized at an either end of the chain. Therefore these bands cannot be called bulk bands, but should be called continuum bands. These states are incompatible with the periodic boundaries. The continuum bands are formed by changing $\beta$ continuously along $C_\beta$ as we show later. We find how to determine the generalized Brillouin zone $C_\beta$, which determines the continuum bands. Here we number the solutions $\beta_i~(i=1,\cdots,2M)$ of Eq. (\[eq4\]) so as to satisfy $\left|\beta_1\right|\leq\left|\beta_2\right|\leq\cdots\leq\left|\beta_{2M-1}\right|\leq\left|\beta_{2M}\right|$. We find that the condition to get the continuum bands can be written as $$\left|\beta_M\right|=\left|\beta_{M+1}\right|, \label{eq6}$$ and the trajectory of $\beta_M$ and $\beta_{M+1}$ gives $C_\beta$. When $M=1$, the two solutions of the quadratic equation share same absolute values, as has been proposed in Ref. ; nonetheless, for general $M$, its condition has not been known so far. In Hermitian systems, we can prove that Eq. (\[eq6\]) becomes $\left|\beta_M\right|=\left|\beta_{M+1}\right|=1$ [@SM], and $C_\beta$ is a unit circle $\left|\beta\right|=1$. To get Eq. (\[eq6\]), we focus on boundary conditions in an open chain. Here we outline the discussion to show Eq. (\[eq6\]), and the detailed discussion is given by Ref. . We impose the wave function Eq. (\[eq3\]) to represent an eigenstate. Apart from the positions near the two ends, it leads to the characteristic equation Eq. (\[eq4\]). The boundary conditions give another constraint onto the values of $\beta_i~(i=1,\cdots,2M)$, in a form of an algebraic equation. We now suppose the system size $L$ to be quite large, and consider a condition to achieve a densely distributed levels (Fig. \[fig:band\]). The equation consists of terms of the form $\left(\beta_{i_1}\beta_{i_2}\cdots\beta_{i_M}\right)^{L+1}$. When $\left|\beta_M\right|\neq\left|\beta_{M+1}\right|$, there is only one leading term proportional to $\left(\beta_{M+1}\cdots\beta_{2M}\right)^{L+1}$, which does not allow continuum bands. Only when $\left|\beta_M\right|=\left|\beta_{M+1}\right|$, there are two leading terms proportional to $\left(\beta_{M}\beta_{M+2}\cdots\beta_{2M}\right)^{L+1}$ and to $\left(\beta_{M+1}\beta_{M+2}\cdots\beta_{2M}\right)^{L+1}$. In such a case, the relative phase between $\beta_M$ and $\beta_{M+1}$ can be changed almost continuously for a large $L$, producing the continuum bands. We note that our condition Eq. (\[eq6\]) is independent of any boundary conditions of an open chain. In Ref. , it was proposed that the continuum bands require $\left|\beta_i\right|=\left|\beta_j\right|$. Nonetheless, it is not sufficient; except for the case $\left|\beta_M\right|=\left|\beta_{M+1}\right|$, it does not allow the continuum bands.  Non-Hermitian SSH model. The dotted boxes indicate the unit cell. (b)-(d) Generalized Brillouin zone $C_\beta$ of this model. The values of the parameters are (b) $t_2=1,t_3=1/5,\gamma_1=4/3$, and $\gamma_2=0$; (b-1): $t_1=1.1$ and (b-2): $t_1=-1.1$, (c) $t_1=0.3,t_2=1.1,t_3=1/5$, and $\gamma_1=0$; (c-1): $\gamma_2=4/3$ and (c-2): $\gamma_2=-4/3$, and (d) $t_2=0.5,t_3=1/5,\gamma_1=5/3$, and $\gamma_2=1/3$; (d-1): $t_1=0.3$ and (d-2): $t_1=-0.3$.](SSH.eps){width="8.5cm"} We apply Eq. (\[eq6\]) to the non-Hermitian SSH model as shown in Fig. \[fig:SSH\] (a). It is given by $$\begin{aligned} H&=&\sum_n\left[\left(t_1+\frac{\gamma_1}{2}\right)c_{n,{\rm A}}^\dag c_{n,{\rm B}}+\left(t_1-\frac{\gamma_1}{2}\right)c_{n,{\rm B}}^\dag c_{n,{\rm A}}\right. \nonumber\\ &&+\left(t_2+\frac{\gamma_2}{2}\right)c_{n,{\rm B}}^\dag c_{n+1,{\rm A}}+\left(t_2-\frac{\gamma_2}{2}\right)c_{n+1,{\rm A}}^\dag c_{n,{\rm B}} \nonumber\\ &&\left.+t_3\left(c_{n,{\rm A}}^\dag c_{n+1,{\rm B}}+c_{n+1,{\rm B}}^\dag c_{n,{\rm A}}\right)\right], \label{eq10}\end{aligned}$$ where $t_1,t_2,t_3,\gamma_1$, and $\gamma_2$ are real. The generalized Bloch Hamiltonian $H\left(\beta\right)$ can be obtained by a replacement ${\rm e}^{ik}\rightarrow\beta$, similarly to Hermitian systems, as $$H\left(\beta\right)=R_+\left(\beta\right)\sigma_++R_-\left(\beta\right)\sigma_-, \label{eq13}$$ where $\sigma_\pm=\left(\sigma_x\pm i\sigma_y\right)/2$, and $R_\pm\left(\beta\right)$ are given by $$\begin{aligned} R_+\left(\beta\right)&=&\left(t_2-\frac{\gamma_2}{2}\right)\beta^{-1}+\left(t_1+\frac{\gamma_1}{2}\right)+t_3\beta, \nonumber\\ R_-\left(\beta\right)&=&t_3\beta^{-1}+\left(t_1-\frac{\gamma_1}{2}\right)+\left(t_2+\frac{\gamma_2}{2}\right)\beta. \label{eq14}\end{aligned}$$ Therefore the characteristic equation can be written as $R_+\left(\beta\right)R_-\left(\beta\right)=E^2$, which is a quartic equation for $\beta$, i.e. $M=2$, having four solutions $\beta_i~(i=1,\cdots,4)$. Then Eq. (\[eq6\]) is given by $\left|\beta_2\right|=\left|\beta_3\right|$ [@SM]. The trajectory of $\beta_2$ and $\beta_3$ satisfying the condition $\left|\beta_2\right|=\left|\beta_3\right|$ determines the generalized Brillouin zone $C_\beta$, and it is shown in Fig. \[fig:SSH\] (b)-(d) for various values of the parameters. It always forms a loop enclosing the origin on the complex plane. Nonetheless, we do not have a rigorous proof that $C_{\beta}$ is always a single loop encircling the origin. We find some features of $C_\beta$. Firstly, our result does not depend on whether $\left|\beta\right|$ is larger or smaller than unity, as opposed to the suggestions in the previous works [@Yao2018; @Liu2019]; in Fig. \[fig:SSH\] (d-2), $\left|\beta\right|$ takes both the values more than and less than 1. Secondly, $C_\beta$ can be a unit circle even for non-Hermitian cases, for example when $t_1=t_3=\gamma_2=0$. Finally, $C_\beta$ can have cusps, corresponding to the cases with three solutions share the same absolute value [@SM].  Phase diagram and bulk-edge correspondence with $t_3=1/5,\gamma_1=5/3$, and $\gamma_2=1/3$. (a) Phase diagram on the $t_1$-$t_2$ plane. The blue region represents that the winding number is $1$, and the orange region represents that the system has exceptional points. Along the black arrow in (a) with $t_2=1.4$, we show the results for (b) the winding number, (d) energy bands in a finite open chain, and (e) the continuum bands from the generalized Brillouin zone $C_\beta$. The edge states are shown in red in (d). (c) shows $\ell_+$ (red) and $\ell_-$ (blue) on the $\bm{R}$ plane with $t_1=1$ and $t_2=1.4$.](t1t2-2.eps){width="8.5cm"} We calculate the winding number $w$ for the Hamiltonian Eq. (\[eq13\]). Thanks to the chiral symmetry, $w$ can be defined as [@SM] $$w=-\frac{w_+-w_-}{2},~w_{\pm}=\frac{1}{2\pi}\left[\arg R_\pm\left(\beta\right)\right]_{C_\beta}, \label{eq15}$$ where $\left[\arg R_\pm\left(\beta\right)\right]_{C_\beta}$ means the change of the phase of $R_\pm\left(\beta\right)$ as $\beta$ goes along the generalized Brillouin zone $C_\beta$ in the counterclockwise way. It was proposed that $w$ corresponds to the presence or absence of the topological edge states [@Yao2018]. We show how the gap closes in our model. It closes when $E=0$, i.e. $R_+\left(\beta\right)=0$ or $R_-\left(\beta\right)=0$. Let $\beta=\beta_i^a~(i=1,2,~a=+,-)$ denote the solutions of the equation $R_{a}\left(\beta\right)=0$ with $\left|\beta_1^a\right|\leq\left|\beta_2^a\right|$. When $E=0$ is in the continuum bands, Eq. (\[eq6\]) should be satisfied for the four solutions $\beta_i^\pm~(i=1,2)$. It can be classified into two cases, (a) $\left|\beta_1^a\right|\leq\left|\beta_2^a\right|=\left|\beta_1^{-a}\right|\leq\left|\beta_2^{-a}\right|~(a=+,-)$, and (b) $\left|\beta_1^a\right|\leq\left|\beta_1^{-a}\right|=\left|\beta_2^{-a}\right|\leq\left|\beta_2^a\right|~(a=+,-)$. In the case (a), as one changes one parameter, the gap closes at $E=0$ and $w_+$ and $-w_-$ change by one at the same time, giving rise to the change of the winding number by unity. On the other hand, in the case (b), only one of the two coefficients $R_\pm\left(\beta\right)$ becomes zero, and it represents an exceptional point. We obtain the phase diagram on the $t_1$-$t_2$ plane in Fig. \[fig3\] (a) and one on the $\gamma_1$-$\gamma_2$ plane in Fig. \[fig4\] (a). In these phase diagrams, the winding number $w$ is $1$ in the blue region. By definition, $w$ changes only when $R_\pm\left(\beta\right)=0$ on the generalized Brillouin zone $C_\beta$, and the gap closes. The energy bands in a finite open chain calculated along the black arrow in Fig. \[fig3\] (a) are shown in Fig. \[fig3\] (d), and one can confirm that the edge states appear in the region where $w=1$. In addition, the continuum bands using $C_\beta$ (Fig. \[fig3\] (e)) agree with these energy bands. In Fig. \[fig4\] (b), we give the energy bands calculated along the green arrow in Fig. \[fig4\] (a), and the edge states appear similarly to Fig. \[fig3\] (d). On the other hand, the system has the exceptional points in the orange region. The phase with the exceptional points extends over a finite region [@SM]. ![\[fig4\]Phase diagram and bulk-edge correspondence with $t_1=0,t_2=1$, and $t_3=1/5$. (a) Phase diagram on the $\gamma_1$-$\gamma_2$ plane. The blue region represents that the winding number is $1$, and the orange region represents that the system has exceptional points. (b) Energy bands calculated along the green arrow in (a) with $\gamma_2=1.4$. Note that $\gamma_c\simeq1.89$. The edge states are shown in red. (c)-(d) Loops $\ell_+$ (red) and $\ell_-$ (blue) on the ${\bm R}$ plane. The values of the parameters are (c) $\gamma_1=-1$ and $\gamma_2=1.4$, and (d) $\gamma_1=2.1$ and $\gamma_2=1.4$. Note that $\ell_-$ passes the origin in (d), which corresponds to exceptional points.](g1g2-3.eps){width="8.5cm"} We discuss the bulk-edge correspondence in our model. The loops $\ell_{\pm}$ drawn by $R_{\pm}\left(\beta\right)$ on the ${\bm R}$ plane are shown in Fig. \[fig3\] (c) and Figs. \[fig4\] (c) and (d) for certain values of the parameters. Both in Fig. \[fig3\] (c) and in Fig. \[fig4\] (c), the system has the winding number $w=1$ since both $\ell_+$ and $\ell_-$ surround the origin ${\cal O}$, leading to $w_+=-1$ and $w_-=1$. In Fig. \[fig4\] (a), one can continuously change the values of the parameters to the Hermitian limit, $\gamma_1,\gamma_2\rightarrow0$, while keeping the gap open and $w=1$ remain. The same is true for Fig. \[fig3\] (a). Therefore by following the proof in Hermitian cases [@Ryu2002], one can prove the bulk-edge correspondence even for the non-Hermitian cases, and existence of zero-energy states in a finite open chain is derived [@SM]. On the other hand, $\ell_-$ passes ${\cal O}$ as shown in Fig. \[fig4\] (d), where the system has the exceptional points. We note that the winding number is not well-defined in this case. In summary, we establish a generalized Bloch band theory in 1D tight-binding systems, and obtain the condition for the continuum bands. We show the way to construct the generalized Brillouin zone $C_\beta$, which is fundamental for obtaining the continuum bands. Here the Bloch wave number $k$ takes complex values in non-Hermitian systems. Our conclusion, $\left|\beta_M\right|=\left|\beta_{M+1}\right|$, is physically reasonable in several aspects. First, it is independent of any boundary conditions. Thus for a long open chain, irrespective of any boundary conditions, the spectrum asymptotically approaches the same continuum bands calculated from $C_\beta$. Second, it reproduces the known result in the Hermitian limit, i.e. $\left|\beta\right|=1$. Third, the form of the condition is invariant under a replacement $\beta\rightarrow1/\beta$. Suppose the numbering of the sites is reversed by putting $n^\prime=L+1-n$ for the site index $n(=1,\cdots,L)$; then $\beta$ becomes $\beta^\prime=1/\beta$, but the form of the condition is invariant: $\left|\beta_M^\prime\right|=\left|\beta_{M+1}^\prime\right|$. Through this definition of the continuum bands, one can show the bulk-edge correspondence without ambiguity by defining the winding number $w$ from the generalized Brillouin zone in 1D systems with chiral symmetry. Indeed we showed that the zero-energy states appear in the non-Hermitian SSH model when $w$ takes non-zero values, and also revealed that these states correspond to topological edge states. It is left for future works how to calculate the continuum bands for systems with other symmetries. The construction of the generalized Brillouin zone can be extended to higher dimensions as well. In two-dimensional systems, we introduce the two parameters $\beta^x\left(={\rm e}^{ik_x}\right)$ and $\beta^y\left(={\rm e}^{ik_y}\right)$. Then the characteristic equation $f\left(\beta^x,\beta^y,E\right)=0$ is an algebraic equation for $\beta^x$ and $\beta^y$. If we fix $\beta^y$, this system can be regarded as a 1D system, and the criterion is given by $\left|\beta^x_{M_x}\right|=\left|\beta^x_{M_x+1}\right|$, where $2M_x$ is the degree of the characteristic equation for $\beta^x$. The same is true for $\beta^y$. Thus we can get the conditions for the continuum bands. Nevertheless, it is still an open question how to determine the generalized Brillouin zone in higher dimensions. Some previous works which assume reality of ${\bm k}$ require further investigations. For example, the tight-binding model in Ref.  is equivalent to our model with $t_3=\gamma_2=0$. Within our theory, it has neither exceptional points nor anomalous edge states, as opposed to Ref. . This work was supported by Grant-in-Aid for Scientific Research (Grants No. JP18H03678 and No. JP16J07354) by MEXT, Japan, by CREST, JST (No. JP-MJCR14F1), and by MEXT Elements Strategy Initiative to Form Core Research Center (TIES). Kazuki Yokomizo also was supported by JSPS KAKENHI (Grant No. 18J22113). \[1\]\[1\][\#1]{} [93]{}ifxundefined \[1\][ ifx[\#1]{} ]{}ifnum \[1\][ \#1firstoftwo secondoftwo ]{}ifx \[1\][ \#1firstoftwo secondoftwo ]{}““\#1””@noop \[0\][secondoftwo]{}sanitize@url \[0\][‘\ 12‘\$12 ‘&12‘\#12‘12‘\_12‘%12]{}@startlink\[1\]@endlink\[0\]@bib@innerbibempty @noop [**]{}, Vol.  (, ) [****,  ()](\doibase 10.1103/PhysRevB.96.045437) @noop [ ()]{} [****,  ()](\doibase 10.1103/PhysRevB.97.041203) [****, ()](\doibase 10.1103/PhysRevLett.121.026403) [****,  ()](\doibase 10.1103/PhysRevLett.120.146601) [****,  ()](\doibase 10.1103/PhysRevB.98.035141) [****,  ()](\doibase 10.1103/PhysRevA.98.042114) @noop [ ()]{} [****,  ()](\doibase 10.1103/PhysRevB.98.155430) @noop [ ()]{} [****,  ()](\doibase 10.1103/PhysRevB.98.245130) [****,  ()](\doibase 10.1103/PhysRevB.99.041116) [****,  ()](\doibase 10.1103/PhysRevB.99.041202) [****,  ()](\doibase 10.1103/PhysRevB.99.041406) [****,  ()](\doibase 10.1103/PhysRevB.99.081102) @noop [****,  ()]{} [****,  ()](\doibase 10.1103/PhysRevA.94.022119) [****,  ()](\doibase 10.1103/PhysRevB.97.115436) [****,  ()](\doibase 10.1103/PhysRevB.98.085116) @noop [ ()]{} [****,  ()](\doibase 10.1103/PhysRevLett.103.093902) @noop [****,  ()]{} @noop [****,  ()]{} @noop [****,  ()]{} @noop [****,  ()]{} @noop [****,  ()]{} @noop [****,  ()]{} @noop [****,  ()]{} [****,  ()](\doibase 10.1103/PhysRevB.92.235310) @noop [****,  ()]{} @noop [****,  ()]{} @noop [****,  ()]{} @noop [****,  ()]{} [****,  ()](\doibase 10.1126/science.aao4551) [****,  ()](\doibase 10.1103/PhysRevB.97.014428) [****,  ()](\doibase 10.1126/science.aap9859) [****,  ()](\doibase 10.1103/PhysRevLett.120.113901) @noop [****,  ()]{} [ (), 10.1126/science.aar4003](\doibase 10.1126/science.aar4003) [ (), 10.1126/science.aar4005](\doibase 10.1126/science.aar4005) @noop [****,  ()]{} [****, ()](\doibase 10.1103/PhysRevLett.121.073901) [****,  ()](\doibase 10.1103/PhysRevA.98.033807) [****,  ()](\doibase 10.1103/PhysRevA.98.042104) @noop [****,  ()]{} @noop [****,  ()]{} @noop [****,  ()]{} [****,  ()](\doibase 10.1103/PhysRevX.4.041001) @noop [ ()]{} [****,  ()](\doibase 10.1103/PhysRevLett.118.045701) @noop [****,  ()]{} [****,  ()](\doibase 10.1103/PhysRevX.8.031079) [****,  ()](\doibase 10.1103/PhysRevLett.121.203001) [****,  ()](\doibase 10.1103/PhysRevLett.121.213902) [****,  ()](\doibase 10.1103/PhysRevA.99.011601) @noop [****,  ()]{} [****,  ()](\doibase 10.1103/PhysRevLett.122.076801) [****,  ()](\doibase 10.1103/PhysRevLett.102.065703) [****,  ()](\doibase 10.1103/PhysRevLett.115.040402) [****,  ()](\doibase 10.1103/PhysRevA.93.062116) @noop [****,  ()]{} [****,  ()](\doibase 10.1103/PhysRevLett.77.570) [****, ()](\doibase 10.1103/PhysRevB.56.8651) [****, ()](\doibase 10.1103/PhysRevB.58.8384) [****,  ()](\doibase 10.1103/PhysRevB.98.085126) [****,  ()](\doibase 10.1103/PhysRevB.97.220301) @noop [ ()]{} @noop [ ()]{} @noop [ ()]{} [****,  ()](\doibase 10.1103/PhysRevLett.120.246601) [****,  ()](\doibase 10.1103/PhysRevB.84.153101) [****,  ()](\doibase 10.1103/PhysRevB.84.205128) [****,  ()](\doibase 10.1103/PhysRevLett.116.133903) [****,  ()](\doibase 10.1103/PhysRevLett.118.040401) [****,  ()](\doibase 10.1103/PhysRevB.97.121401) @noop [****,  ()]{} [****,  ()](\doibase 10.1103/PhysRevLett.120.146402) [****, ()](\doibase 10.1103/PhysRevA.97.042118) [****,  ()](\doibase 10.1103/PhysRevA.97.052115) [****, ()](\doibase 10.1103/PhysRevA.98.012111) [****,  ()](\doibase 10.1103/PhysRevLett.121.026808) [****, ()](\doibase 10.1103/PhysRevLett.121.086803) [****,  ()](\doibase 10.1103/PhysRevLett.121.136802) [****,  ()](\doibase 10.1103/PhysRevB.98.165148) @noop [****,  ()]{} @noop [****,  ()]{} [****,  ()](\doibase 10.1103/PhysRevB.99.081103) [****,  ()](\doibase 10.1103/PhysRevB.99.075130) [****, ()](\doibase 10.1103/PhysRevA.99.022127) [****,  ()](\doibase 10.1103/PhysRevB.99.081302) @noop [****, ()](\doibase 10.1103/PhysRevLett.89.077002)

ArXiv

--- author: - | Denise Ratasich ^1^, Faiq Khalid ^1^, Florian Geissler ^2^, Radu Grosu ^1^,\ Muhammad Shafique ^1^, Ezio Bartocci ^1^\ ^1^ Institute of Computer Engineering, TU Wien, Austria\ (e-mail: firstname.lastname@tuwien.ac.at)\ ^2^ CPS Dependability Research Lab, Intel Corporation, Germany\ (e-mail: florian.geissler@intel.com) bibliography: - 'bib/denise.bib' - 'bib/ezio.bib' - 'bib/faiq.bib' date: '2018-10-29' title: | A Roadmap Towards Resilient Internet of Things\ for Cyber-Physical Systems ---

ArXiv

Several direct and indirect (through-the-substrate) mechanisms can lead to an effective interaction between adsorbates on metal surfaces. The possible role of adsorbate-induced substrate relaxation was considered some time ago by Lau and Kohn [@Lau77; @Lau78], using an elastic continuum model of the surface. They concluded that the resulting interaction was repulsive between identical adsorbates, varied as $\rho^{-3}$ with separation $\rho$, and was inversely proportional to the shear modulus of the substrate. For adsorbates separated by a few atomic spacings, however, the continuum theory may not be valid. More recently, in a series of papers Kevan et al. [@Kevan98; @Skelton94; @Skelton97; @Wei95; @Wei97] have determined adsorbate–adsorbate interaction energies for CO at several metal surfaces, using a transfer-matrix analysis of thermal desorption spectra. They have also qualitatively discussed the adsorbate-induced strain as a possible mechanism of the adsorbate–adsorbate interaction at intermediate range, i.e., several substrate atoms apart. The potential energy of the atomic lattice of a solid in the harmonic approximation can be written as $$\label{harmonic} V = \sum_{i,j,\mu, \nu} x_\mu (i) D_{\mu \nu} (i,j) x_\nu (j) ,$$ where $x_\mu (i)$ is the $\mu$-th component of the displacement of the $i$-th atom from the equilibrium position. Now assume that an impurity is created by replacing one atom with a different species. (In the general discussion we talk about an “impurity”, although we are primarily interested in the chemisorption case, where the adsorbate also introduces additional degrees of freedom. The generalization to the latter case is straightforward.) The potential energy after the replacement can again be expressed in the form (\[harmonic\]), but the new equlibrium positions $x'_\mu (i)$ are in general different: $$\label{impurity} x'_\mu (i) = x_\mu (i) + \Delta x_\mu (i) .$$ The new dynamical matrix $D'_{\mu \nu} (i,j)$ is also different, and a constant term appears which shifts the energy minimum. Using (\[impurity\]), the potential energy of the system after the impurity has been introduced can be expressed in the coordinates $x_\mu (i)$: $$\label{harmonic2} V' = \sum_{i,j,\mu, \nu} x_\mu (i) D'_{\mu \nu} (i,j) x_\nu (j) + \sum_{i,\mu} F_\mu (i) x_\mu (i) + V'_0,$$ i.e., the dynamical matrix is modified, and linear (force) and constant (energy shift) terms appear. The foregoing considerations depend only upon the assumed stability of the solid, i.e., the existence of the minimum of the potential energy. For our application to chemisorption, we specifically assume the following properties: (a) the force terms $F_\mu (i)$ are nonzero only for a small number of substrate atoms around the adsorption site; (b) similarly, only a few elements of the dynamical matrix $D_{\mu \nu} (i,j)$ change, if any; (c) the effect is linear, so that the chemisorption of another molecule (and, consequently, of a third, a fourth, etc.) can be described by the same set of parameters, of course centered around the new adsorption site. The condition (c) is rather restrictive and excludes systems which reconstruct at large adsorbate coverage, as well as those where other interactions are important, such as the direct adsorbate–adsorbate repulsion, the electrostatic dipole–dipole interaction, or the “chemical” competition for the same electronic orbitals in the substrate. Estimates, however, show that such interactions are usually weak and short-ranged, while we are interested in medium-range interactions (second nearest neighbor and beyond) of nonionic adsorbates. In this paper we take into account the in-plane relaxation within the first atomic layer of the substrate induced by a chemisorbed species. The relaxation in the perpendicular direction can, of course, be equally strong, but it does not contribute much to the effective adsorbate–adsorbate interaction and to the adsorbate-induced surface stress. Also, we do not consider the coupling to internal adsorbate coordinates, which was discussed in Ref. [@Brako98] in an application to the damping of adsorbate vibrations. For a hexagonal layer of atoms, we write the potential energy as $$\begin{aligned} V & = & \frac{1}{2} \sum_{i} \sum_{j=1}^{6} \frac{1}{2}K_1 [\hat{{\bf{r}}}_{ij} \cdot ({\bf{r}}_i - {\bf{r}}_j )]^2 + \frac{1}{2} \sum_{i}K_2 {\bf{r}}_i^2, \label{potentialsubstrate}\end{aligned}$$ where $\bf{r}_i=(x_i,y_i)$ is the in-plane displacement from the equlibrium position of the $i$-th atom. The term $K_1$ describes a central atom–atom interaction, and the term $K_2$ binds atoms to their equilibrium positions, simulating the interaction to lower atomic layers. Without it, the model would be too “soft” to long-wavelength perturbations. The trade-off is that the lowest phonon frequency becomes finite, i.e., there are no true “acoustic” modes, but this has little influence on the relaxation energy and other static quantities calculated in this work. Now assume that an atom or a molecule chemisorbs on top of the atom $i = 0$, and that the induced change of the potential energy is $$\begin{aligned} V' &=& \sum_{j=1}^{6} \frac{1}{2} \Delta K_1 [\hat{{\bf{r}}}_{0j} \cdot ({\bf{r}}_0 - {\bf{r}}_j )]^2 \nonumber \\ && - \sum_{j=1}^6\frac{k_2}{\alpha} \hat{\bf{r}}_{0j}\cdot ({\bf{r}}_0 - {\bf{r}}_j) + V'_0, \label{potentialinteraction}\end{aligned}$$ where the first term describes the change of the force constant between the atom $0$ and the six surrounding atoms, and the second is a linear force term. In the following we drop the $\Delta K_1$ and $V'_0$ terms, which are beyond the level of accuracy assumed in this work, although, in general, the $\Delta K_1$ term ought to be included. The definition of the chemisorption-induced surface stress $\tau$ is [@Ibach97] $$\delta W = A \tau_x \delta\epsilon_x,$$ where $A$ is the surface area, $\delta\epsilon_x$ the strain, and $\delta W$ the difference of the work involved in straining a clean surface and a surface with adsorbates. In our model, we obtain $$\tau_x =- \frac{k_2}{a \alpha} 2 \sqrt{3}\ \theta, \label{stressontop}$$ where $\theta$ is the adsorbate coverage. (If $ \Delta K_1$ is different from zero, an additional factor of order unity appears in the above expression.) Irrespective of the sign of the force term $k_2/\alpha$, there is always an energy gain due to the relaxation of the surrounding atoms, i.e., the minimum of $V + V'$ is less than zero. If another molecule adsorbs on a nearby site, the forces induced by the two adsorbates act in opposite directions and the relaxation is less complete than with adsorbates far apart, which leads to an effective interaction. We calculate the interaction energies by comparing the minimum of the potential energy $V + V'$ for a single adsorbed molecule with the minimum for two molecules adsorbed on, in turn, second, third, fourth, and fifth nearest-neighbor sites, see Fig. \[fig:Pt\](a). (Throughout the paper we assume that there is a strong repulsion between first nearest neighbor adsorbates caused by “chemical” effects, and do not consider them.) =0.20 =0.20 We first consider the chemisorption of CO on a Pt(111) surface. There is a large amount of tensile stress in the first atomic layer of the clean Pt(111) surface [@Hohage95; @Boisvert97]. Although the surface does not reconstruct at room temperature, a reconstruction is observed at high temperatures in the presence of saturated Pt vapor [@Hohage95]. CO adsorbs initially into on-top sites of Pt(111), but the energy difference for the adsorption into bridge sites is obviously small, since some bridge adsorbates are found already at coverages $\theta$ above 0.15 [@Yoshinobu96]. Some authors report (Ref. [@Yeo97-King] and references therein) that a regular $(\sqrt{3} \times \sqrt{3}){\rm R}30^{\circ}$ structure at a coverage $\theta = 1/3$ is formed, Fig. \[fig:Pt\](b), but others claim that the densest structure of exclusively on-top adsorbates exists at $\theta = 0.29$ [@Tushaus87], and that further chemisorption occurs into bridge sites. The regular structure at $\theta = 0.5$ contains an equal number of on-top and bridge adsorbates. We have chosen the values of the parameters $K_1$, $K_2$, and $k_2/\alpha$ which give good agreement with experimental data on adsorbate-induced surface stress [@Ibach97] and with low-coverage interaction energies [@Skelton94], as shown in Table \[tab:energies\]. We had to choose a rather small value for the force constant $K_1$ between atoms in the first surface layer. (The value of $K_2$ has little effect on the results. The much larger value of $K_1$ used in a similar model in Ref. [@Brako98] was an overestimate.) The reduction from bulk values is characteristic of many close-packed noble-metal surfaces [@Bortolani90], but the reduction we find is larger than suggested in the surface phonon calculation in Ref. [@Bortolani89]. Consequently, the highest resulting vibrational frequency of two-dimensional phonons of the first surface layer is about 5 meV, by about a factor of two smaller than the frequency of surface phonons along the edge of the Brillouin zone of around 10 meV calculated in Ref. [@Bortolani89]. As discussed above, it is possible that part of the softening is localized around the adsorption site only, the term $\Delta K_1$ in Eq. (\[potentialinteraction\]). ------------ --------------- --------------- --------------- --------------- ---------- -------- $W_{2\rm NN}$ $W_{\rm 3NN}$ $W_{\rm 4NN}$ $W_{\rm 5NN}$ $E_0/6 $ $\tau$ Theory 194 314 108 119 231 1.24 Experiment 120 400 236 452 1.2 ------------ --------------- --------------- --------------- --------------- ---------- -------- : Interaction energies $W_{n{\rm NN}}$ (in K) and the induced surface stress $\tau$ (in N/m) for on-top adsorption of CO on Pt(111). $W_{n{\rm NN}}=2 E_0-E_{n{\rm NN}}$, where $E_0$ is the relaxation energy of a single CO molecule, $E_{n{\rm NN}}$ is the relaxation energy for two CO molecules adsorbed at the $n$-th nearest-neighbor sites, etc. Theoretical values are calculated using $K_1 = 4$ N/m, $K_2 = 2.5$ N/m, $k_2/\alpha = 0.3 \times 10^{-9} $ N. The nearest-neighbor distance is $a = 2.76$ Å. \[tab:energies\] The values in Table \[tab:energies\] show that the repulsive interaction is strong between CO adsorbed on sites lying along rows of substrate atoms, and weaker for adsorbates separated by hollows, even if they are less far apart. In our opinion, the rather large interaction energy between fourth nearest neighbor adsorbates in Ref. [@Skelton94] is influenced by the contributions from more distant sites, which were not included in their analysis. Clean nickel (111) surfaces do not reconstruct. Unlike some earlier claims, it is now accepted that at low temperature CO chemisorbs initially into threefold hollow sites [@Chen89; @Held98]. At room temperature, some bridge and on-top sites seem to be occupied even at low coverages [@Held98]. At a coverage $ \theta = 0.33$, CO forms a regular $(\sqrt{3}\times\sqrt{3}){\rm R30}^{\circ}$ structure [@Becker93], but it is not clear whether the molecules adsorb into fcc or hcp positions. At $\theta = 0.5$, a regular c(4$\times$2)–2CO structure is formed, in which an equal number of fcc and hcp sites is occupied [@Becker93; @Mapledoram94; @Sprunger95], Fig. \[fig:Ni\](b). The top layer of Ni atoms shows buckling in that nonequivalent atoms have different vertical relaxation. The CO molecules are slightly tilted from the direction perpendicular to the surface. =0.20 =0.20 The initial heat of adsorption at room temparature is 130 kJ/mol [@Stuckless93]. It decreases slowly at first, to around 122 kJ/mol at $\theta = 0.33$ and 112 kJ/mol at $\theta = 0.5$. However, the fact that not all adsorption sites are equivalent, and the rather large standard deviation of experimental data make the interpretation of the coverage dependence of the heat of adsorption uncertain. We describe the first layer of Ni atoms by the same potential as for Pt, Eq. (\[potentialsubstrate\]). The interaction terms are similar to Eq. (\[potentialinteraction\]), but the adsorbate is in a threefold hollow site and the sums run over the three surrounding Ni atoms (Fig. \[fig:Ni\]). The induced surface stress is $$\tau = - \frac{k_2}{\alpha a} \theta, \label{stresshollow}$$ where the nearest-neighbor Ni–Ni distance is $a = 2.49$ Å. From the experimental data $\tau = -0.55$ N/m at $\theta = 0.33$ [@Grossmann94] we have estimated $k_2/\alpha = 1 \times 10^{-10}$ N. In a LEED analysis of the c$(4 \times 2)$–2CO structure which forms at $\theta = 0.5$, Mapledoram et al. [@Mapledoram94] found that the lateral displacement of the Ni atoms next to an adsorbate, [*a*]{} and [*c*]{} in Fig. \[fig:Ni\](b), was 0.03 Å. We have reproduced this value by taking $K_1 = 6$ N/m and $K_2 = 2$ N/m. (As before, the value of $K_2$ is of lesser importance.) This is a considerable reduction from the bulk values, but not as large as for the Pt(111) surface, in agreement with the fact that the Ni(111) surface seems less prone to reconstruct than Pt(111). The energy gain per adsorbate is 22 K. Using these values of the parameters, the relaxation energy for a single adsorbate is only 42 K, which is 30 times smaller than that we have found for CO/Pt(111). (This reduction can be well estimated treating each surrounding Ni atom as an independent oscillator: The force $k_2/\alpha$ is three times smaller than in the CO/Pt(111) case, the Ni–Ni force constant is 50% larger, and there are 3 surrounding atoms instead of 6.) Furthermore, the displacements of Ni atoms around the hollow adsorption site are not along chains of atoms, and do not cause a large displacement of other atoms. The calculated effective interaction energies for second nearest neighbors and beyond are therefore only a few K, too small to be observable. In our opinion, the interaction energy of 100 K between second neighbor adsorbates suggested by Skelton et al. [@Skelton97] is either a combined result of other mechanisms or an artefact of their procedure. We note that an earlier study [@Gijzeman84] reported that there was essentially no interaction already between second neighbor adsorbates. Interaction energies have also been determined for CO chemisorbed on Rh [@Wei97; @Payne92] and Cu [@Wei95] surfaces. We discuss these systems only qualitatively, since the proposed values are still uncertain, and there is no quantitative data on other adsorbate-induced properties. Wei et al. [@Wei97] estimated that $W_2 = -100$ K and $W_3 = 150$ K for the on-top chemisorbed CO on Rh(111). An earlier measurement by Payne et al. [@Payne92] reported $W_2 = 170$ K and $W_3 = -85$ K. In our model, the relative magnitudes of interaction energies for on-top adsorbates on fcc (111) surfaces are always similar to those found for CO/Pt(111). In particular, we expect large repulsion between third nearest neighbor adsorbates which lie along a chain of substrate atoms. In this respect the values proposed in Ref. [@Wei97] seem more probable, although the origin of the attractive $W_2$ (if it is real) is not clear. For the on-top CO on Cu(111), the same authors found $W_2 = 107$ K, $W_3 > 800$ K, $W_4 = 155$ K [@Wei95]. The value of $W_3$ seems too large compared with the other two, but otherwise the results are quite similar to CO/Pt(111). The square lattice of the first layer of the (100) surface of the fcc bulk is not well described by a purely pairwise potential similar to Eq. (\[potentialsubstrate\]), since in the absence of angular force constants and of coupling to lower layers, only the [*ad hoc*]{} term $K_2$ ensures the stability. Nevertheless, the trends in the interaction energies between on-top adsorbates can be deduced by analogy with (111) surfaces. We expect a repulsive interaction between third nearest neighbor adsorbates which lie on the same chain of atoms, and no interaction between second nearest neighbors which lie diagonally on different chains, because the forces on adjacent substrate atoms are orthogonal. In the latter case, even a weak attraction is possible owing to partly collinear displacements induced by the two adsorbates on more distant surface atoms. Indeed, the values $W_2 = 0$ and $W_3 = 400$ K were found for Rh(100) [@Wei97], while the values suggested for Cu(100) were $W_2 = -33$ K and $W_3 = 13$ K [@Wei95]. It is interesting that the continuum elastic theory [@Lau78] also gives a strong repulsion in the $\langle 110 \rangle$ direction and possibly a weak attraction in the $\langle 100 \rangle$ direction between adsorbates on (100) surfaces of noble metals. We have shown that the adsorbate-induced substrate relaxation leads to an adsorbate–adsorbate interaction which is quite long-ranged and has a nonmonotonic distance dependence, being particularly large between molecules adsorbed in on-top positions along a chain of surface atoms. The same mechanism also leads to other observable effects, such as the adsorbate-induced surface stress and the relaxation displacement of substrate atoms. Quantitative agreement with experiment can be obtained for CO adsorbed on Pt(111) and some other surfaces using a simple model, assuming that force constants between atoms in the first surface layer are considerably weaker than in the bulk, which is a known property of many close-packed noble metal surfaces. The results emphasize the importance of allowing the full substrate relaxation in the first-principle calculations of chemisorption on metal surfaces. This work was supported by the Ministry of Science and Technology of the Republic of Croatia under contract No. 00980101. [99]{} K. H. Lau and W. Kohn, Surface. Sci. [**65**]{}, 607 (1977). K. H. Lau, Solid State Commun. [**28**]{}, 757 (1978). S. D. Kevan, J. Molecular Catalysis A: Chemical [**131**]{}, 19 (1998). D. C. Skelton, D. H. Wei, and S. D. Kevan, Surf. Sci. [**320**]{}, 77 (1994). D. C. Skelton, D.-H. Wei, and S. D. Kevan, Surf. Sci. [**370**]{} 64 (1997). D. H. Wei, D. C. Skelton, and S. D. Kevan, Surf. Sci. [**326**]{}, 167 (1995). D. H. Wei, D. C. Skelton, and S. D. Kevan, Surf. Sci. [**381**]{}, 49 (1997). R. Brako and D. Šokčevi' c, Surf. Sci. [**401**]{}, L388 (1998). H. Ibach, Surf. Sci. Rep. [**29**]{}, 193 (1997). M. Hohage, T. Michely, and G. Comsa, Surf. Sci. [**337**]{}, 249 (1995). G. Boisvert, L. J. Lewis, and M. Scheffler, Phys. Rev. B [**57**]{}, 1881 (1998). J. Yoshinobu and M. Kawai, Surf. Sci. [**363**]{}, 105 (1996). Y. Y. Yeo, L. Vattuone, and D. A. King, J. Chem. Phys. [**106**]{}, 392 (1997). M. Tüshaus, E. Schweizer, P. Hollins, and A. M. Bradshaw, J. Electron. Spectrosc. Related Phenomena [**44**]{}, 305 (1987). V. Bortolani, F. Ercolessi, E. Tosatti, A. Franchini, and G. Santoro, Europhys. Lett. [**12**]{}, 149 (1990). V. Bortolani, A. Franchimi, G. Santoro, J. P. Toennies, Ch. Wöll, and G. Zhang, Phys. Rev. B [**40**]{}, 3524 (1989). J. G. Chen, W. Erley, and H. Ibach, Surf. Sci. Lett. [**223**]{}, L891 (1989). G. Held, J. Schuler, W. Sklarek, and H.-P. Steinrück, Surf. Sci. [**398**]{}, 154 (1998). L. Becker, S. Aminpirooz, B. Hillert, M. Pedio, J. Haase, and D. L. Adams, Phys. Rev. B [**47**]{}, 9710 (1993). L. D. Mapledoram, M. P. Bessent, A. Wander, and D. A. King, Chem. Phys. Lett. [**228**]{}, 527 (1994). P. T. Sprunger, F. Besenbacher, and I. Stensgaard, Chem. Phys. Lett. [**243**]{}, 439 (1995). J. T. Stuckless, N. Al-Sarraf, C. Wartnaby, and D. A. King, J. Chem. Phys. [**99**]{}, 2202 (1993). A. Grossmann, W. Erley, and H. Ibach, Surf. Sci. [**313**]{}, 209 (1994). O. L. J. Gijzeman, M. M. J. Zandvoort, F. Labohm, J. A. Vleigenthart, and G. Jongert, J. Chem. Soc. Faraday Trans. II [**80**]{}, 771 (1984). S. H. Payne, H. J. Kreuzer, K. A. Peterlinz, T. J. Curtiss, C. Uebing, and S. J. Sibener, Surf. Sci. [**272**]{}, 102 (1992).

ArXiv

--- abstract: 'This paper summarizes the observations which provide the best evidence for the presence of black holes in active galactic nuclei. This includes: X–ray variability; kinematical studies using optical emission lines as well as the distribution of megamaser spots; and the shape of the Fe K$\alpha$ X–ray emission line. It also presents the current status of our understanding of jet-dominated active galaxies (blazars), and briefly reviews the currently popular AGN “Unification Schemes” based on orientation effects. Finally, it reviews the observations of the X–ray and $\gamma$–ray continuum, which, at least for the radio-quiet objects, is likely to be the primary form of their radiative output, and summarizes the best current models for the radiative processes responsible for the high-energy electromagnetic emission in radio-quiet AGN, as well as in jet-dominated blazars.' author: - 'GREGM.MADEJSKI$^1$' title: 'Black holes in Active Galactic Nuclei: observations' --- **To appear in [*Theory of Black Hole Accretion Disks,*]{}** **M. Abramowicz, G. Bjornsson, & J. Pringle, Eds. (Cambridge University Press)** 0.4 cm 1. Introduction {#introduction .unnumbered} =============== Perhaps the most exciting astronomical observation leading to our current understanding of black holes has been the discovery of quasars. These celestial objects, originally found in the early sixties as point-like radio emitters, were identified with apparently stellar sources, possessing somewhat unusual spectra, with prominent emission lines. The identification by [@schm] of these lines as redshifted systems implied that quasars are distant and extremely luminous, commonly producing $10^{46}$ erg s$^{-1}$; this is a hundred times or more in excess of the total luminosity of all the stars in a galaxy. Sensitive imaging of the nebulosities which often surround them implied that quasars are nuclei of galaxies, and thus are higher-luminosity counterparts of the compact nuclei of Seyferts, studied some twenty years before the discovery of quasars as unusual emission line objects: hereafter, we assume that they are respectively the lower and higher luminosity end of the same population. A variety of scenarios were advanced to explain their nature, and this included multiple supernovae or massive spinning stars, but the proposal that quasars are powered by an accretion of surrounding matter onto a black hole, advanced in the mid-60s by [@salp] and [@zeld], became the paradigm that we are developing and testing today. While this is a viable and very attractive paradigm, only the last few years brought a solid evidence for it, allowing also to measure the mass of the central object. It is important to note here that quasars are much more numerous at a redshift $\sim$ 2 than they are locally, meaning that a substantial fraction of galaxies must have undergone the quasar phase. It is thus likely that many otherwise normal local galaxies harbor supermassive black holes, “dead quasars.” In fact, as we discuss below, there is a number of relatively anonymous galaxies that show no signs of nuclear activity, but [*do*]{} show evidence for such black holes. We present the observational evidence for black holes in AGN in Section 2; in Section 3, we discuss the effects of the orientation of the accretion disk surrounding the back hole on the appearance of the nucleus. In Section 4, we discuss the jet-dominated AGN known as blazars. In Section 5, we review the observations of AGN in X–rays and $\gamma$–rays, the bands that sample the regions closest to the black hole, and in Section 6, we review the radiation processes proposed to explain the emission in these bands. 2. Lines of Evidence for Presence of Black Holes in Active Galaxies {#lines-of-evidence-for-presence-of-black-holes-in-active-galaxies .unnumbered} =================================================================== There are two general lines of argument that are used to “prove” the existence of black holes in AGN. The first attempts to measure the total mass within a volume, and argues that no other form besides a black hole can have these parameters. This is done either via estimation of the volume from variability data (via the light travel time arguments) and mass from the luminosity (via the Eddington limit); alternatively, this can be determined by a measurement of velocity of matter at a specified distance from the central object, essentially using Kepler’s laws. The second method, discussed in more detail in Chapter 5.3 (by A. Fabian), relies on the distortion of the emission line shapes caused by strong gravity resulting from the presence of a black hole, and we cover it here only briefly. 2.1. X–ray Variability {#xray-variability .unnumbered} ---------------------- The variability of active galaxies generally shows the highest amplitude and the shortest time scales in the X–ray and $\gamma$–ray bands, which happen to be clearly separated from the optical / UV bands by the strong absorption of the interstellar medium in our own Galaxy. This rapid variability as well as other lines of argument indicate that the X–ray / $\gamma$–ray radiation arises the closest to the central source, and in many cases, is the primary source of energy in active galaxies. While the total bolometric luminosity of quasars is often dominated by the optical and UV flux (see, e. g., Laor et al. 1997; for a recent review, see Ulrich, Maraschi, & Urry 1997), the bulk of this flux probably arises in more distant regions from the central source than the X–rays and $\gamma$–rays. The optical and UV radiation arising in the innermost regions of the nucleus, on the other hand, is most likely a result of reprocessing of X–ray / $\gamma$–ray photons. This – as well as the author’s personal interest in X–rays – is the reason why this chapter focuses primarily on the high energy emission from quasars. In general, the X–ray variability of quasars is aperiodic. While a measurement of periodic variability would give us a clue to the circumnuclear environment and thus the nature of the black hole, besides the ill-fated NGC 6814 (cf. Madejski et al. 1993), no strict periodicity was reported for any AGN. However, there were two reports of quasi-periodic variation of flux of active galaxies: NGC 5548 (Papadakis & Lawrence 1993) and NGC 4051 (Papadakis & Lawrence 1995) inferred from the EXOSAT data, but the quality of the data is only modest, and these still need to be confirmed. More statistically significant is the quasi-periodic variability of of IRAS 18325-5926 by [@iwasawa], but this is inferred from only a few ($<$ 10) cycles, and requires confirmation via further monitoring before drawing any detailed conclusions. Nonetheless, we understand relatively little about the details of variability of active galaxies, although the recent light curves are sufficiently good to discriminate if the time series are linear or non-linear; this is discussed later in this chapter. In any case, this rapid variability implies a compact source size. This is of course the standard causality argument: no stationary source of isotropic radiation can vary faster than the time it takes for light to cross it. In the X–ray band, the power spectrum of variability generally is rather flat at long time scales, and above some characteristic frequency, it shows a power-law behavior, such that the variable power drops with decreasing time scale (see, e.g., McHardy 1989); the Fourier phases of these light curves show no coherence (cf. Krolik, Done, & Madejski 1993). As it was pointed by many authors, for this form of variability, the doubling time scale has no definitive meaning, but for the lack of better data, it suffices for the illustrative purposes: it is certainly valid as an order-of-magnitude relationship between the source radius $r$ and the time scale for doubling of the source flux $\Delta t$ such that $r < c$ $\Delta t$. Again, for quasars, this is particularly true in the X–ray band, where the variability is most rapid: even the early X–ray data gave us a clue that quasars are very compact. For example, an X–ray light curve for the Seyfert galaxy NGC 5506 (by no means an extreme object) as observed by the ME detector onboard the EXOSAT satellite shown on Fig. 1 illustrates it well. This, and other observations of it, with redshift $z = 0.007$ and a 2 - 10 keV X–ray flux of $\sim 4 \times 10^{-11}$ erg cm$^{-2}$ s$^{-1}$, imply a luminosity $L_{\rm X}$ of $\sim 10^{43}$ erg s$^{-1}$, corresponding to an Eddington mass of at least $\sim 10^5$ $M_{\rm o}$. This corresponds to a Schwarzschild radius $r_{\rm S}$ of $3 \times 10^{10}$ cm. The X–ray data show a doubling time of $\sim 10,000$ s, corresponding to $r < 3 \times 10^{14}$ cm, which, for a $10^5$ $M_{\rm o}$ black hole, would imply that the X–ray emission arises from a region of radius $r_{\rm X} \sim 10^4$ $r_{\rm S}$. This example is by no means extreme – we made a number of assumptions that are probably even too conservative: more realistic assumptions imply a mass of $10^6$ $M_{\rm o}$, and $r_{\rm X} \sim 10^3$ $r_{\rm S}$. A number of more extreme cases – including very luminous objects – were reported recently on the basis of the ROSAT data by [@boller]. These are generally for the so-called “narrow-line Seyfert 1s,” and we will return to those objects later. In brief, a doubling time scale of $\sim 1000$ s for a source with $L_{\rm X} \sim 10^{44}$ erg s$^{-1}$ is not uncommon, implying that the bulk of the X–ray emission arises around 10 - 100 $r_{\rm S}$. However, the use of the observed variability time scale does not provide an “airtight” argument for the size of the emitting region, since the observed emission may well be anisotropic, yielding an underestimate of the emitting volume, as is almost certainly the case for blazars. We discuss this in more detail later on. 2.2. Kinematic Studies of Active and “Normal” Galaxies Using Optical Emission Lines {#kinematic-studies-of-active-and-normal-galaxies-using-optical-emission-lines .unnumbered} ----------------------------------------------------------------------------------- The other line of evidence for the presence of black holes in galaxies (both active and “normal”) is the velocity field of the matter emitting closely to the nucleus. This kind of work has been recently reviewed by [@korm], and it dates back to the ground-based observations made in the late 70s, when W. Sargent and collaborators showed that the stellar velocity dispersion in the radio galaxy M 87 increases to 350 km s$^{-1}$ in the innermost 1.5$''$ from the nucleus. M 87 was in fact observed by the Planetary Camera by [@ford87] and the Faint Object Spectrograph by [@harms] onboard the repaired Hubble Space Telescope, and the images showed the presence of a disk-like structure of ionized gas in the innermost few arc seconds. The spectroscopy provided a measure of the velocity of the gas at an angular distance from the nucleus of 0.25$''$ (corresponding to $\sim 20$ pc, or $\sim 6 \times 10^{19}$ cm), showing that in the reference frame of the object, it recedes from us on one side, and approaches us on the other, with a velocity difference of $\sim 920$ km s$^{-1}$ (see Fig. 2). This implies a mass of the central object of $\sim 3 \times 10^{9}$ $M{\rm_o}$, and besides a black hole, we know of no other form of mass concentration that can “fit” inside this region. As an aside, it is worth noting that M 87 is known to have a relativistic jet perpendicular to the disk structure mentioned above, expanding with the bulk Lorentz factor $\Gamma_j$ of 4. As such, this object is probably just a blazar, with the jet oriented at an angle $\sim 40^{\rm o}$ to the line of sight, and thus is probably the closest to a hard evidence that blazars (which we discuss in more detail below) indeed [*do*]{} harbor black holes. Besides M 87, similar spectroscopy observations done with Hubble Space Telescope revealed high stellar velocities in the central regions of a number of normal galaxies which otherwise show no evidence for an active nucleus. These were summarized recently by [@ford], and include well-publicized observations (by L. Ferrarese, H. Ford, J. Kormendy and others) of NGC 6251, NGC 4261, NGC 4594, NGC 3115, but doubtless by now probably there are several new objects. Such high velocities in the innermost regions of these galaxies cannot be explained in any other way besides invoking the presence of massive ($10^{8} - 10^{9}$ $M_{\rm o}$) black holes in their centers. (Interestingly, several of these galaxies also show weak radio jets!) Even for our own Milky Way galaxy, the infrared data and velocity measurements – readily performed from the ground (cf. Eckart & Genzel 1997), at much higher resolution than the HST data for external galaxies – reveal a “modest” nuclear black hole with a mass of $\sim 3 \times 10^{6}$ $M_{\rm o}$. The evidence is building that supermassive black holes are quite common; recent estimates by [@ford] as well as [@ho] imply that they inhabit perhaps as many as half of all galaxies, and may well be the “dead quasars” of the past. Active galactic nuclei are thus probably only the “tip of the iceberg” of their population. 2.3. Megamasers in Active Galaxies {#megamasers-in-active-galaxies .unnumbered} ---------------------------------- Perhaps the most elegant observation showing the presence of a Keplerian disk around a black hole – and thus capable of measuring the mass of the hole independently of the otherwise uncertain estimates of its distance – was the Very Large Baseline Interferometry observation of megamasers in the vicinity of the nucleus of Seyfert 2 galaxy NGC 4258 reported by [@miyoshi]. The masing activity can only be observed along a line of sight where the velocity gradient is zero, meaning that these can be seen at locations with masers either between us and their source of energy, or locations at $\sim$ 90$^{\rm o}$ to the line of sight. Indeed, this is the spatial distribution of maser spots around the nucleus of NGC 4258, illustrated in Fig. 3. Specifically, these observations reveal individual masing spots revolving at distances ranging from $\sim 0.13$ pc to about twice that around the central object – presumably, again, a black hole – with a mass of $\sim 3.6 \times 10^{7}$ $M_{\rm o}$. What is truly remarkable about these data is the near-perfect Keplerian velocity distribution, in a slightly warped disk-like formation; this implies that almost all the mass is located well within the inner radius where the megamasers reside. It is not possible to have a cluster of distinct, dark, massive objects responsible for such gravitational potential; at least some of the objects would escape on a relatively short time scale, and form a potential well with a different shape, which would now force a departure of the megamaser–emitting material from pure Keplerian motion (cf. Maoz 1995). Since we know the central mass quite precisely, NGC 4258 has been a terrific laboratory to study the details of the accretion disk. In particular, this is a relatively low luminosity ($\sim 10^{42}$ erg s$^{-1}$) object, making it quite sub-Eddington, with $L/L_{\rm E} \sim 3 \times 10^{-4}$. Such sub-Eddington sources are likely to obey unique solutions of accretion disk structure (see, e.g., Ichimaru 1977; Narayan & Yi 1994; Abramowicz et al. 1995), where the accreting gas is optically thin and radiates inefficiently, and the accretion energy that is dissipated viscously, is advected with the accretion flow. With this, as was argued by [@lasota], the accretion disk in NGC 4258 can well be advection-dominated (but this does not [*have*]{} to be the case; see, e.g., Neufeld & Maloney 1995). However, it is important to note that such low Eddington rate cannot be universal among quasars; if most of them radiated at such low $L/L_{\rm E}$, the black hole masses of the most luminous sources would be much larger than expected on other grounds. However, an intriguing possibility (cf. Yi 1996) is that quasars in their “youth,” when the black hole masses were more modest, had standard, “cold” (Shakura - Sunyaev) accretion disks, and this is why they were so luminous in the past. The masses of the black holes grew with time, and even if the mass rate supplied for accretion remained constant, $L/L_{\rm E}$ actually decreased, and thus the the inner accretion disks switched from “cold” (bright, Shakura - Sunyaev) phase to “hot” (fainter, advection-dominated) phase even if the rate of mass supply [*did not*]{} decrease. This is perhaps why some of the yesterday’s bright quasars are today’s dormant, “dark” black holes, revealed only via the kinematical studies mentioned above. Nonetheless, NGC 4258 is probably [*not*]{} a unique object. We know of a large class of “low activity” active galaxies, known collectively as “Low Ionization Nuclear Emission Region” objects, or LINERs; they generally have low luminosity, coupled with the absence of the luminous inner disk as evidenced by the emission line ratios. However, unlike the [*bona fide*]{} Seyferts with low luminosity, which seem to vary relatively rapidly in X–rays, implying they are “scaled down” quasars with relatively low mass black holes – LINERs are known [*not*]{} to vary rapidly in any band, suggesting that the low activity is not due to a low black hole mass, but rather due to a low accretion rate (cf. Ptak 1997). As the nuclei of these objects are not very bright, the data are sparse, and thus the details of the radiative processes are poorly known; while workable models exist (see, e.g., Lasota et al. 1996), they still require more work on the details of the transition between the “standard” and advection-dominated regions of the disk. 2.4. Profile of the Fe K Emission Line {#profile-of-the-fe-k-emission-line .unnumbered} -------------------------------------- Perhaps the most convincing evidence that a strong gravitational field is present in active galactic nuclei comes from the recent measurements of the shape of the Fe K$\alpha$ fluorescence line, arising in a geometrically thin, but optically thick accretion disk. This is discussed in more detail by A. Fabian (Chapter 5.3), so what follows is a brief summary. The inner part of the disk is illuminated by X–rays. Because of the relative Cosmic abundances and the fluorescence yields of various elements, the strongest discrete spectral feature predicted from the disk is the 6.4 keV fluorescent Fe K$\alpha$ line; the strength (equivalent width) of the line of $\sim$ 150 eV, as measured by [@pounds], is in fact roughly consistent with predictions of [@george]. Since this line arises from matter in motion, its profile is a tracer of the velocity field of the accreting matter. The Asca observations of the X–ray bright Seyfert galaxy MCG-6-30-15 by Tanaka et al. (1995) indeed showed that the line (see Fig. 4) has a characteristic two-pronged shape expected to arise from matter flowing in a disk-like structure. The matter approaching us is responsible for the blue wing of the line, while that receding produces the red wing, with an additional redshift, since the photons are emitted in a strong gravitational field; the exact shape also depends on the inclination of the disk. A detailed spectral fitting of the line shape indicates that the emitted energy of the line is indeed 6.4 keV, while the bulk of its flux arises at $<$ 10 $r_{\rm S}$, implying in turn the presence of nearly neutral material very closely to the black hole. The analysis of a number of Seyfert spectra from the Asca archives by [@nandraline] suggests that many Seyferts indeed show the Fe K$\alpha$ line profiles that require an emission close to the black hole, but the quality of data is only modest. Fortunately, a number of more sensitive observatories – such as AXAF, Astro-E, XMM, and Constellation-X, will be launched in the next few years, providing ample opportunities for X–ray observations of effects of strong gravity. 3. Unifying Seyfert 1s and Seyfert 2s: the Orientation Effects {#unifying-seyfert-1s-and-seyfert-2s-the-orientation-effects .unnumbered} ============================================================== The megamaser source NGC 4258 is only the first of three active galaxies showing a spatial distribution of masing spots from which it is possible to measure the mass of the central object. The other two are the well-known NGC 1068 (Greenhill 1998), and NGC 4945 (Greenhill, Moran, & Herrnstein 1997). Both are also classified as Seyfert 2s, which from the observational side means that they show narrow emission lines, implying velocities on the order of 1,000 km s$^{-1}$; these lines show no variability, and thus it is generally accepted that they originate in a relatively large regions, on the order of 100 pc or more. Seyfert 1s as well as luminous quasars, on the other hand, exhibit generally very different spectra, with permitted emission lines, and these lines are usually broad, with velocities upwards of 1,500 km s$^{-1}$, often reaching 30,000 km s$^{-1}$. This, together with the variability of the lines that is commonly observed on time scales of weeks or months, implies that the broad line region is located much closer to the nucleus than the narrow line region. The likely relationship between the two classes of active galaxies was revealed by the seminal observation of the well-known Seyfert 2 galaxy NGC 1068 by Antonucci & Miller (1985). The spectropolarimetric study of the H$\beta$ line revealed that when observed in polarized light, the line is broad. They interpreted the polarization as due to electron scattering by material that is distributed preferentially along the symmetry axis of the system, and advanced the widely accepted scenario explaining the differences between the two types of Seyferts as an orientation effect. This is illustrated in Fig. 5; all Seyfert galaxies are surrounded by a geometrically and optically thick torus, with an inner radius of a fraction of a parsec. Such a torus can be, for instance, the outer regions of a severely warped disk, as recently suggested by M. Begelman and J. Pringle (see Chapter 8.1). When an object is viewed along the axis of the torus, it is a Seyfert 1, revealing all the ingredients of the nucleus: the broad line region and the unobscured X–ray source, both commonly varying on a short time scale. When viewed in or close to the plane of the torus, the broad line region is completely obscured, and the soft end of the X–ray spectrum is absorbed due to the photoelectric absorption by the material in the torus. The opening of the torus contains a “mirror” of ionized gas with free electrons, and these are responsible for the scattering of the broad line light into the line of sight, and hence the broad lines are seen only in polarized light. In the context of this “unification” model, it is thus not surprising that all three megamasers with measured rotation curves are Seyfert 2 objects: to even observe a megamaser, we have to be located in the plane of material which has large column density and presumably is roughly co-axial with the torus. Likewise, the X–rays traveling closely to the plane of the torus encounter a larger column density, resulting in the greater photoelectric absorption. This is in fact, observed; even early X–ray spectra of Seyferts showed preferentially larger absorption in sources showing only narrow lines (Seyfert 2s), with the broad line region obscured by the same material that absorbs soft X–rays. Note that this is different than the case of only narrow emission lines present in some low-luminosity radio galaxies, such as M 87 (collectively known as the FR-I objects): there, the broad lines cannot be obscured, as we see no evidence of X–ray absorption. Furthermore, while Seyfert 2s sometimes do show highly ionized, narrow permitted lines (in addition to the commonly detected forbidden lines), FR-I objects [*do not*]{}, implying an absence of the very strong isotropic ionizing UV continuum. We will return to this point below. 4. Anisotropy and Doppler Effects: the Case of Blazars {#anisotropy-and-doppler-effects-the-case-of-blazars .unnumbered} ====================================================== The use of the observed variability time scale does not, however, provide an “airtight” argument for the size of the emitting region, since the observed emission may well be anisotropic, yielding an underestimate of the emitting volume. As pointed out by [@rees], if the radiating matter moves at a relativistic speed towards the observer, the variability time scale appears shortened by the Doppler effect. Furthermore, the flux of radiation is boosted in the direction of motion of the emitting matter, and thus the total luminosity inferred under an assumption of isotropy is an overestimate. We are nearly certain that this is the case for blazars. This sub-class of quasars consists of sources that generally show strong radio emission and are highly variable and polarized in all observable wavelengths, and also includes members with very weak or absent emission lines, known as BL Lacertae – type objects. High angular resolution radio observations, performed since the late 60s – using Very Long Baseline Interferometry – implied compact radio emission regions, often associated with a rapid change of their structure. With the assumption that the redshifts are indeed cosmological, these structures appeared to expand at transverse speeds exceeding $c$, and in many cases, they had shapes of jets. If the speed of the emitting matter is very relativistic, pointing closely to our line of sight, the apparent superluminal motion is just a projection effect, providing further support for the Doppler-boosting scenario for blazars. In a number of aspects, blazars are the most extremely active galactic nuclei. Observations of many blazars by the EGRET instrument onboard the Compton Gamma-ray Observatory (CGRO) indicated that usually they are strong emitters in the GeV $\gamma$–ray band; we know of at least 50 such objects. The GeV emission can dominate the overall observed electromagnetic output of these sources, and this is illustrated in Fig. 6 for 3C 279, the first of the ERGET-discovered GeV blazars. In a few cases, the $\gamma$–ray emission has been observed with Cerenkov radiation telescopes to extend as far as the TeV range. This is particularly exciting, as TeV radiation can be observed from the ground, allowing a study of quite exotic phenomena occurring in quasars without the expense of space-borne platforms. Interestingly, we know of no “radio-quiet” MeV or GeV $\gamma$–ray emitting quasars, implying an association of the $\gamma$–rays with the compact radio source and thus a jet. In many cases, the $\gamma$–ray emission from blazars is variable on a time scale of a day, and recent, simultaneous multi-wavelength monitoring observations indicate that the flux swings are reasonably well correlated between various bands, with the amplitude of variability increasing with the increase of the energy of the observing band. If the emission in the lower energy bands arises from a region co-spatial with that where the $\gamma$–rays originate – and the tracking of light curves in various bands supports this – then the opacity to the $\gamma$-$\gamma$ pair production would imply large optical thickness. Similarly, even the early calculations indicated that if the radio sources are as luminous and as compact as the variability data imply, the cooling should be predominantly via the Compton process, with the radiating electrons losing all their energy via upscattering of the just–produced–photons, regardless of the radiation mechanism. Even the large $\gamma$–ray fluxes detected by EGRET are often orders of magnitude lower than the predictions of this self-Compton process. Both these difficulties, however, go away if the radio as well as the GeV emission are both Doppler-boosted, meaning that the true luminosity is lower, and source sizes are greater than inferred under an assumption of isotropy. In fact, we now believe that the entire broad-band continuum is produced in the relativistic jet (cf. Königl 1981; Ulrich et al. 1997), and statistical considerations imply that the bulk Lorentz factors $\Gamma_j$ of these jets are on the order of 10 (cf. Vermeulen & Cohen 1994). For the case of BL Lac – type blazars, the Doppler enhancement (going roughly as a square of $\Gamma_j$; see, e.g., Appendix A of Sikora et al. 1997) ) is so strong that the continuum completely outshines the emission lines; in some cases, however, it is possible that the emission lines are absent to begin with. In any case, the mechanism for an acceleration and collimation of these powerful jets to such high relativistic speeds is not known, but general considerations of jet energetics are strongly suggestive that this energy is ultimately tapped from an accretion onto a compact object: the best picture today involves the so-called Blandford-Znajek mechanism (Blandford & Znajek 1977), where the power of the jet derives from the spin of the black hole. 4.1. Emission Processes in Blazars {#emission-processes-in-blazars .unnumbered} ---------------------------------- The electromagnetic emission from jets in blazars – recently reviewed by [@sikora] and Ulrich et al. (1997) – is interesting in its own right, and the comparison of the broad-band spectra of blazars and the “radio-quiet” quasars shows it clearly. The strong GeV component which clearly dominates the spectra of the former (cf. Fig. 6) is entirely absent in the latter, and the relative strength of the GeV emission as compared to that in keV range is often different by a factor of $\sim 100$. The radio and optical emission shows polarization, and spectra are always non-thermal. All these facts argue that the radio-through-UV emission is produced by the synchrotron process, with ultrarelativistic electrons (with $\gamma_{\rm el} \sim 1000$ or more) radiating in magnetic field on the order of a Gauss. The Compton upscattering of lower energy photons mentioned above is indeed present - and most current models suggest that it is responsible for the radiation observed above a few keV. In the lineless BL Lac type objects, we have no evidence for any strong external radiation field, and the dominant “seed” photons for Compton upscattering are likely to be produced by the synchrotron process internally to the jet. As was shown by [@dermer] as well as by [@sikbegrees] and [@blandlev], in blazars with emission lines, the external, diffuse radiation dominates; since this radiation is isotropic in the stationary frame, it appears Doppler-boosted in the frame of the jet. This scenario predicts that there should be many objects with jets pointing farther away, up to the right angle to our line of sight. We now believe that for the quasar-type blazars, with strong emission lines, these are the giant, powerful radio galaxies. The difference between the quasar-type and BL Lac type blazars may well be related to the presence of the inner, extremely luminous accretion disk in the former. As suggested by M. Begelman, and, more recently [@reynolds], some radio galaxies (as, for instance, M 87) have massive black holes, yet relatively modest luminosity, and essentially no broad emission lines; this might argue for a very sub-Eddington accretion rate, with a likelihood of an advection-dominated inner accretion disk. An intriguing possibility is that these low-luminosity radio galaxies are the “misdirected” BL-Lac - type blazars: the absence of the inner, luminous disk would then be responsible for the absence of the strong ionizing radiation and thus broad emission lines. This in turn would mean that in BL Lac - type blazars, there are no external photons to act as “seeds” to be Comptonized by the relativistic electrons in the jet, and the only “seed” photons are internal, produced by the synchrotron process. 4.2. Origin of the Difference Between “Radio-Quiet” Active Galaxies and Blazars {#origin-of-the-difference-between-radio-quiet-active-galaxies-and-blazars .unnumbered} ------------------------------------------------------------------------------- We are nearly certain that there is a clear [*intrinsic*]{} distinction between the jet-dominated blazars, and the more common, radio-quiet quasars. However, the reason for the different behavior of these two subclasses is far from certain, and this is primarily because we do not have a good understanding of the formation and acceleration of relativistic jets. Even though we cannot use the causality arguments to infer the presence of black holes in blazars solely from their variability – since we do not know for sure as to what extent the true variability time scales are shortened by Doppler boosting – more indirect evidence implies that ultimately, accretion onto a black hole powers blazars as well. As to the difference between the two categories, perhaps the most appealing scenario, advanced by [@wilson], as well as by [@moder], is where the two classes differ by strong or weak spin of the black hole; again, the jet is formed and powered by tapping the rotational energy of the hole. This is somewhat similar to the distinction between the Galactic binary “microquasars” suggested by [@zhang], discussed in Chapter 2.1 (by P. Charles). Nonetheless, this is an area of very active research where no clear conclusions have been reached. In particular, if the shape of the fluorescence Fe K line seen in the Seyfert galaxy MCG-6-30-15 (see above) and possibly in other Seyferts (Nandra et al. 1997a) indeed requires that the cold material powering the nucleus flows via a disk-like structure, this implies that the fluorescence occurs at at a distance $r < 3$ $r_{\rm S}$. Since the last stable orbit in a non-rotating (Schwarzschild) black hole is at 3 $r_{\rm S}$ and beyond this, the matter is in a free fall, this would imply that the black hole is spinning (Kerr), where the last stable orbit can be substantially closer to the black hole, depending on its spin. However, we note that an alternative scenario, advanced by [@reynbeg] does [*not*]{} require a disk at $r < 3$ $r_{\rm S}$; nonetheless, this still requires X–ray emission at least at or just beyond 3 $r_{\rm S}$, merely eliminating the requirement of [*spinning*]{} black hole, but it still implies that at least a non-rotating black hole is present. 5. Zooming in on a Supermassive Black Hole: High Energy Spectra of Radio-Quiet Active Galaxies {#zooming-in-on-a-supermassive-black-hole-high-energy-spectra-of-radio-quiet-active-galaxies .unnumbered} ============================================================================================== As we mentioned above, active galaxies are generally variable, and the most rapid variability is observed in the X–ray and $\gamma$–ray bands; since we are interested in understanding the regions closest to the black hole, these bands deserve the most detailed study. A recent, excellent article by Mushotzky, Done, & Pounds (1993) reviewed the X–ray spectra of AGN; however, this covered the status of observations before the results from CGRO became available, while we do include these results here. The most conclusive results are gleaned by a study of the non-blazar active galaxies, to avoid any potential contamination by the jet. Since no radio-quiet (= jet-less) active galaxy was detected above several hundred keV, the most relevant spectral region is the X–ray and soft $\gamma$–ray bands. This covers nearly four decades in energy, from $\sim 0.2$ keV to $\sim 1$ MeV, and often requires observations with multiple satellites, which must be made simultaneously, since active galaxies are variable in all bands. The early X–ray observations of active galaxies by [@mush], [@halp], and [@roth] implied that, to the first order, X–ray spectra of active galaxies are power laws with the energy index $\alpha$ (defined such that the flux density $S_{\nu} \propto \nu^{-\alpha}$) of about 0.7, modified by photoelectric absorption at the low energy end. A major advance came from sensitive observations with the ROSAT and Ginga satellites (covering respectively the bands of 0.1 to 2, and 2 to $\sim$ 30 keV), revealing that the spectra are more complex, with a somewhat softer underlying continuum, with $\alpha \sim 0.9$. This continuum is modified by photoelectric absorption in the host galaxy of the quasar; the absorbing material can be either cold, or partially ionized, and this manifests itself as isolated edges, most notably of oxygen, seen in the ROSAT (and more recently, also in Asca) data. The Ginga spectra, reported by [@pounds], showed a strong emission line at the rest energy of $\sim 6.4$ keV, presumably due to fluorescence of the K shell of iron, and a hardening above $\sim 8$ keV. These last two features were interpreted as signatures of reprocessing (“reflection”) of the primary continuum in the cold material that is accreting onto the nucleus. The Fe K line is indeed the strongest expected, via the combination of the relatively high abundance of iron (decreasing with the atomic number $Z$), and fluorescence yield (increasing with $Z$); as pointed out by [@makis], the line equivalent width as measured by Ginga is too high to be produced in absorbing material of any column. Reprocessing by the matter that accretes onto the black hole is thus the most viable alternative. The hardening, predicted earlier in a seminal paper by Lightman & White (1988), arises as an additional spectral component, due to Compton reflection from cold matter. This is produced roughly at one Thomson depth, or when the equivalent hydrogen column density is $\sim 1.6 \times 10^{24}$ cm$^{-2}$. The cosmic abundances of elements are such that at low energies, the photoelectric absorption dominates, and unless the accreting matter is substantially ionized, this component emerges only above the last significant absorption edge, again from iron. The strengths of the reflection component and the line are in fact in agreement with theoretical predictions by [@george]. With the intensity of the incident X–rays greatest closely to the central source (as inferred from the variability of the continuum), its kinematic and gravitational Doppler shifts are powerful diagnostics of the immediate circumnuclear region. 5.1. A Working Template: Spectrum of Seyfert 1 IC 4329a {#a-working-template-spectrum-of-seyfert-1-ic4329a .unnumbered} ------------------------------------------------------- Perhaps the brightest [*bona fide*]{} Seyfert 1 on the sky is the luminous object IC 4329a, and it has been observed simultaneously with the ROSAT (0.2 - 2 keV) and CGRO OSSE (50 - 1000 keV) (Fabian et al. 1993; Madejski et al. 1995). This left a gap between $\sim 2$ and $\sim 50$ keV, so the observations were supplemented by non-simultaneous data obtained by Ginga a few years earlier; the Ginga data were renormalized to match the ROSAT flux at 2 keV, so this combined data set provided a good representation of a broad band high energy spectrum of a Seyfert 1, and is illustrated in Fig. 7. The data indeed showed a modest photoelectric absorption due to neutral material (most likely the ISM of the host galaxy), plus an additional modest column of ionized absorber. Beyond $\sim 2$ keV, the spectrum is the primary continuum, with $\alpha \sim 0.9$. At $\sim$ 6.4 keV, there is a strong Fe K line, which in the case of IC 4329a and many other Seyfert galaxies is broad, with $\sigma$ of at least 200 eV. The detailed studies of this line in another object, MCG-6-30-15 [@tanaka], show the characteristic two-pronged shape, implies relativistic motion in an accretion disk inclined to the line of sight as discussed above. The region beyond the Fe K line shows general hardening of the spectrum, accompanied by the pseudo-edge due to Fe K; this is a signature of Compton reflection, as discussed above. The intensity of this component is consistent with the reflector being a semi-infinite plane; it peaks at $\sim 30$ keV, and beyond this, it steepens gradually, most likely due to the Compton recoil as well as the Klein-Nishina effects. Of course the underlying power law also can steepen there – we only observe a sum of the primary and reflected spectrum. The detailed fits (spectral decomposition) of the data imply that the primary spectrum is consistent with the power law having an exponential cutoff at an e-folding energy of $\sim 200$ keV. The recent simultaneous RXTE and OSSE observations of this object generally confirm this picture. A very important constraint on any theoretical models for radiation processes in active galaxies is the [*absence*]{} of strong annihilation line at 511 keV. Is this high energy spectrum of IC 4329a unique? Given the variability of active galaxies, simultaneous observations in X–rays and soft $\gamma$–rays are required, but these are sparse; an alternative approach by [@zdziarsum] is to co-add many non-simultaneous observations. This in fact produced an average spectrum that is remarkably similar to the above picture, and thus any theoretical interpretation of the data for IC 4329a is probably valid for Seyfert nuclei in general. 5.2. The “Ultra-soft” Seyfert 1s: the Question of the “Soft Excess” {#the-ultra-soft-seyfert-1s-the-question-of-the-soft-excess .unnumbered} ------------------------------------------------------------------- It is important to note that luminous objects similar to IC 4329a are a majority of Seyfert 1s, but there is one important sub-class of Seyferts that shows decidedly distinct X–ray spectra, differing from the above description in soft X–rays, below $\sim 2$ keV. These are the so-called “ultra-soft” Seyferts. First observed in the HEAO data by S. Pravdo and collaborators, they were suggested to be a possibly distinct class of active galaxies by F. Cordova on the basis of the Einstein Observatory Imaging Proportional Counter data. The EXOSAT data analyzed by [@arnaud] as well as by [@turner] showed that in a number of active galaxies, the extrapolation of the hard power law towards low energies (taking properly into consideration the absorption from our own Galaxy) [*underpredicted*]{} the soft X–ray flux, indicating that there is an additional component of X–ray emission. This, the so-called “soft excess,” meant that in a given object, the ultra-soft component can co-exist with the hard power law, implying that active galaxies generally have two-component power law spectra, where either or both components are visible. It is important to note that this component appears in both low- and high-luminosity sources; for instance, it was observed in the Einstein Observatory data in the quasar PG 1211+143 by [@elvis]. For the sources where both components are present, they intersect at $\sim 1 - 2$ keV (but this may be an observational artifact); the spectrum of the soft component is very soft (steep), and can be described as a power law with an index $\alpha$ $\sim 2$ or even steeper (but a power law is usually not a unique model), while the hard component has just the “canonical” hard Seyfert 1 spectrum with $\alpha$ $\sim 1$. In the very few cases that the variability of both components has been measured, it appears that the two vary independently, showing no correlation between them (see, e.g., the case of Mkn 335 in Turner 1988). On the other hand, such a lack of correlated variability may be the result of a complex spectral deconvolution procedure, since only a tail of the soft excess component is observable due to intervening absorption in the ISM of the host, or our own Galaxy. In fact, a correlation of UV and EUV fluxes was clearly observed in the “soft excess” Seyfert NGC 5548 (Marshall et al. 1997). This implies some coupling of the soft excess component to the UV/X–ray reprocessing cycle discussed in more detail in Section 6 below. Similar correlation has been also detected in that source between the UV flux and soft X–ray residuals observed simultaneously in Ginga spectra (cf. Magdziarz et al. 1998). Since the EUV observations suggest much higher variability amplitude than the UV data, the lack of apparent correlation in fainter sources may be related to either variation of the cut off energy of the tail of the soft excess, or confusion with spectral index variations in the hard component. “Ultra-soft” Seyferts appear to be quite common in the ROSAT all-sky survey, which is not surprising, since ROSAT is a very sensitive instrument below 2 keV. Detailed follow-up studies by [@boller] showed a very interesting result: the optical/UV emission lines in these objects are generally quite narrow, with the widths on the order of 1,000 - 3,000 km s$^{-1}$, as compared to $>$ 5,000 km s$^{-1}$ for “normal” Seyfert 1s. These are [*not*]{} similar to Seyfert 2s at all; these are widths of permitted lines, while Seyfert 2s only rarely show permitted lines. Important clues to the nature of these soft components may be in the variability patterns, although only very few well-sampled light curves exist. The EXOSAT data were consistent with the hard X–ray variability that is aperiodic but linear, meaning that the time series can be described as uncorrelated noise (cf. Czerny & Lehto 1997); an example of a light curve for this is shown in Fig. 1. The variability of soft X–rays, on the other hand, is often “episodic,” with large flares (up to a factor of 100!) (see Fig. 8 for an example), and decidedly non-linear (cf. Boller et al. 1997). (In this case a “non-linear” behavior manifests itself qualitatively in an episodic, flare-like behavior such as that illustrated in Fig. 8; see, e.g., Vio et al. 1992. Quantitatively, a “non-linear” time series is a positive, definite one, which has the ratio of its standard deviation to its mean which is larger than unity; see, e.g., Green 1993.) This difference may imply different emission mechanisms in the hard and soft X–ray components; however two observational effects have to be considered before drawing any conclusions. First, the observational noise or/and presence of additional higher frequency variability in the hard component may effectively dissolve apparent signatures of non-linearity (cf. [@leighly]). Second, if the variability related to the energy reprocessing does indeed originate from variations of the soft excess component (e.g., Magdziarz et al. 1998), then the signatures of non-linearity should be suppressed in the UV and the hard continuum. Early modeling attempted to describe the “soft excess” as the tail end of the thermal, multi-blackbody emission from an accretion disk; however, this seems [*not*]{} to be the case, at least for the bright and well-studied NGC 5548. Magdziarz et al. (1998) have shown that the UV component in that object may be associated with rather cold disk continuum, with a temperature on the order of a few eV, while the soft excess requires a separate spectral component. We will return to this below. 5.3. High Energy Spectra of Seyfert 2s vs. Seyfert 1s {#high-energy-spectra-of-seyfert-2s-vs.-seyfert-1s .unnumbered} ----------------------------------------------------- As it was mentioned above, the popular “unification” picture explains the differences between the spectra of Seyfert 1s vs. Seyfert 2s as due to the orientation effects. To the first order, the only difference that should be seen in the X–ray spectra of the two classes is the amount of photoelectric absorption, while the underlying continuum should be the same. Just as in the case of Seyfert 1s, this requires simultaneous observations by multiple satellites, only the problem is more acute here, as the large amount of absorption leaves generally fewer soft X–ray photons to allow for a sensitive measurement of the continuum. Nonetheless, the observations with the Ginga satellite by [@awaki] revealed that X–ray spectra of Seyfert 2s are in fact equivalent to spectra of Seyfert 1s, absorbed by various column densities of cold gas. More detailed studies by [@smith] implied that the continua of Seyfert 2s may be somewhat harder, but only marginally so; a more conclusive results should be obtained from observations by the Rossi X–ray Timing Explorer, which features a broader bandpass, extending to 50 keV (or, for brighter sources, even to 100 keV). In any case, the inferred column densities in Seyfert 2s are $\sim 10^{22}$ cm$^{-2}$ or more. In a few cases – as, for instance, the well-studied NGC 1068 – we only know that no primary X–ray continuum is seen, so the absorber must be quite Thomson-thick, with the absorbing column greater than $\sim 10^{25}$ cm$^{-2}$. A good, illustrative example of a nearly-extreme Seyfert 2 – but, with the primary continuum still barely penetrating the absorber – is NGC 4945, a Seyfert 2 which also shows megamaser emission, and thus is most likely observed in the plane of the putative torus. In the Asca and Ginga ranges (below $\sim 10$ keV), the source is relatively faint, but above $\sim 10$ keV, the spectrum rises sharply ([@iwasawa4945]). OSSE observations of it by [@done] revealed that at 50 - 100 keV, this is the second brightest radio-quiet active galaxy in the sky (see Fig. 9). The absorbing column is large, $\sim 4 \times 10^{24}$ cm$^{-2}$, and while the observations were not simultaneous, a comparison of Fig. 7 and Fig. 9 reveals that the underlying continuum, to the first order, is consistent with that of a Seyfert 1. Another interesting aspect of this source is the fact that the megamaser distribution implies a mass of the black hole of $\sim 10^6$ M$_{\rm o}$ (Greenhill et al. 1997). With the bolometric luminosity of the nucleus of at least $\sim 10^{42}$ erg s$^{-1}$ (Iwasawa et al. 1993; Done et al. 1996), this source radiates at a few percent of Eddington luminosity, and thus is unlikely to be advection-dominated, as may be the case for NGC 4258. 5.4. High Energy Spectra of High Luminosity Sources {#high-energy-spectra-of-high-luminosity-sources .unnumbered} --------------------------------------------------- So far, we discussed primarily the relatively low-luminosity quasars, and an obvious question to be asked is: does this general picture hold for the higher luminosity counterparts? In general, the answer is yes, but with some modifications. It is important here to compare “apples to apples,” and in the case of quasars, this means selecting radio-quiet objects, as radio-loud quasars tend to have higher X–ray – to – optical flux ratios, which may be due to a contamination by a possible jet; unfortunately, this paucity of X–ray photons in the radio-quiet objects makes spectral studies somewhat more difficult. In general, the more luminous objects show a lower ratio of X–ray to optical luminosities (see, e.g., Kriss & Canizares 1985; Avni & Tananbaum 1986). The recent work by Laor et al. (1997) (using the ROSAT PSPC data) and [@nandraquas] (using Asca data) indicates that the continuum X–ray spectra of higher luminosity objects appear generally similar to those of the lower luminosity counterparts. Notable exception is an absence of the Fe K line and the Compton reflection component in quasars (see, e.g., Nandra et al. 1997b). With the more luminous central source, this may be the result of a nearly-complete ionization of the accreting material, such that the reflection component is present, but cannot be distinguished by its tell-tale Fe K line and spectral hardening above $\sim 8$ keV: for an entirely ionized reflector, the incident or emerging photons encounter no photoelectric absorption, and the reflection is (below $\sim 30$ keV) identical to the incident spectrum, with no Fe K line present. Unfortunately, those quasars are generally too faint to be studied in detail above $\sim 10$ keV, where the only instruments currently available are non-imaging proportional counters such as the RXTE, dominated by uncertainties of the instrumental as well as the Cosmic X–ray Background. As a result, any detailed studies must await X–ray reflective optics sensitive beyond 10 keV, already under development; such telescopes are essential for studies of these faint objects, as they permit subtraction of background from the same image as the source, and thus will yield the best quality data for luminous quasars. 6. Radiation Processes in Radio-Quiet Active Galaxies {#radiation-processes-in-radio-quiet-active-galaxies .unnumbered} ===================================================== The availability of good quality high energy spectra permits us to constrain the possible emission mechanisms that can operate in active galaxies. We briefly discussed the case of blazars above; observationally, the continuum high energy spectra of radio-quiet objects are also decidedly non-thermal, but these mechanisms are probably somewhat different than the synchrotron + Compton model discussed for the blazar jets. The early work by [@suntit], using the diffusion approximation, suggested that a power-law spectrum can be produced by a repeated Compton-upscattering of soft photons by a Compton-thick bath of hot electrons. A more general variant of this model is essentially what is used today to explain the primary, high energy spectrum in these objects as well as in the Galactic black hole candidates. In summary, a successful model has to explain a power law spectrum with an energy index $\alpha$ of $\sim$ 0.9 – 1, exponentially cutting off at $E_{\rm c} \sim 200$ keV: other spectral features are merely signatures of reprocessing. There are essentially two flavors of the Comptonization model that can be applied to the isotropic emission in quasars: the original thermal Comptonization version, and a non-thermal variant. The difference between the two is related to the distribution of electron energies. The non-thermal version, developed by A. Zdziarski, A. Lightman, P. Coppi, as well as by C. Done, R. Svensson, G. Ghisellini, and A. Fabian (Zdziarski et al. 1990; Zdziarski & Coppi 1991) involves acceleration of particles to relativistic energies, and a subsequent pair cascade; these particles Comptonize UV photons, believed to be produced in abundance by the accretion disk. The attractive feature of this model is that the pairs thermalize to relatively low temperatures ($\sim$ a few keV), providing the medium which, again, upscatters the UV photons to form the “soft excess” discussed above. One feature, however, predicted by this version of the model, is the annihilation line that should be present at $\sim 511$ keV. No spectrum of any active galaxy collected so far with the CGRO OSSE detector showed such a feature (Johnson et al. 1997), and thus the model is somewhat out of favor, despite its natural ability to provide the “soft excess.” Refinements to the thermal Comptonization model, primarily by [@pout], allowed the quasi-analytical calculation in the optically thin - to intermediate regime. This is the most viable current model for radio-quiet active galaxies, and is discussed in more detail in Chapter 5.3 by J. Poutanen. Roughly, for regime relevant to the hard X–ray spectra of Seyferts, the energy of the exponential cutoff $E_{c}$ determines the temperature of the Comptonizing plasma, such that $kT_{\rm plasma}$ $\simeq$ $E_{c}/1.6$, while the index of the power law, together with the cutoff, determine its optical depth $\tau_{\rm Th}$, such that $\tau_{\rm Th}$ $\simeq$ $0.16$ / $(\alpha \times (kT_{\rm plasma}/m_e c^2))$ (Pietrini & Krolik 1995; Poutanen, Krolik, & Ryde 1997). For Seyferts, the Comptonizing plasma has to have a temperature of $\sim 100$ keV, and optical depth $\tau \sim$ 1; the soft “seed” photons are available in abundance from the inner accretion disk, as evidenced by the presence of the reflection component (Zdziarski et al. 1997). However, a simple “sandwich” type structure (with a cold disk covered by a uniform corona) cannot work, as this would produce too many “seed” photons for Comptonization. Instead, an example of a good phenomenological model is a “patchy corona” above a surface of the disk, proposed by Haardt, Maraschi, & Ghisellini (1994); however, none of these models address the processes responsible for the particle acceleration. The issue of the “soft excess” remains unresolved in the context of the above models. Spectral fitting to the data for the well-studied NGC 5548 imply that the soft excess can be produced by a relatively cold ($kT \sim 200$ eV) but optically thick ($\tau > 10$) plasma, while the hard continuum requires $kT \sim 50$ keV, but $\tau \sim 2$ (Magdziarz et al. 1998). Coexistence of such two phases may be related to the disk structure and dynamics of possible multi-phase transition regions (see, e.g., Magdziarz & Blaes 1998). In fact, such multi-phase medium appears naturally in local solutions of the disk corona transition layer (see, e. g., [@rozanska]). However, without high quality observational data on the spectral shape and temporal correlation between both components, so far, we lack clear clues as to the time evolution of the plasma energetics. The author’s prejudice (based partially on the “episodic” nature of the soft light curves) is that we probably witness some form of a limit cycle operating in the inner disk, and thus the best avenue for this is a development of a more detailed theory for the structure of the inner accretion region, and in particular, the issue of stability of the transition region between the [*bona fide*]{} disk and the matter free-falling onto the black hole. However, any tests of theories require sensitive observations: especially needed are well-sampled light curves obtained simultaneously over a broad energy range, from the softest energies accessible ($\sim 0.1$ keV) up to the end of the observable spectrum, in the MeV range. The prospects for such observations are very good: with the impending launch of AXAF, XMM, Astro-E, Integral, and, eventually, Constellation-X, we should have the data for the more definitive modeling. 0.2 cm [**Acknowledgements:**]{} The author wishes to acknowledge helpful comments from Drs. J. Krolik, P. Magdziarz, C. Done, E. Boldt, and M. Sikora, and figures from Drs. W. Brandt, K. Nandra, and L. Greenhill. Abramowicz, M., Chen, X., Kato, S., Lasota, J. P., & Regev, O. 1995, ApJ, 438, L37 Antonucci, R. R. J, & Miller, J. S. 1985, ApJ, 297, 621 Arnaud, K. A., et al. 1985, MNRAS, 217, 105 Avni, Y., & Tananbaum, H. 1986, ApJ, 305, 83 Awaki, H., Koyama, K, Inoue, H., & Halpern, J. 1991, PASJ, 43, 195 Blandford, R. D., & Levinson, A. 1995, ApJ, 441, 79 Blandford, R. D., & Znajek, R. L. 1977, MNRAS, 179, 433 Boller, T., Brandt, W. N., & Fink, H. 1996, A&A, 305, 53 Boller, T., Brandt, W. N., Fabian, A. C., & Fink, H. 1997, MNRAS, 289, 393 Czerny, B., & Lehto, H. 1997, MNRAS, 285, 365 Dermer, C., Schlikheiser, R., & Mastichiadis, A. 1992, A&A, 256, L27 Done, C., Madejski, G. M., & Smith, D. 1996, ApJ, 463, L63 Eckart, A., & Genzel, R. 1997, MNRAS, 284, 576 Elvis, M., Wilkes, B., & Tananbaum, H. 1985, ApJ, 292, 357 Fabian, A., Nandra, K., Celotti, A., Rees, M., Grove, E., & Johnson, W. 1993, ApJ, 416, L57 Ford, H. C., et al. 1994, ApJ, 435, L27 Ford, H. C., Tsvetanov, Z. I., Ferrarese, L., & Jaffe, W. 1998, in [*The Central Regions of the Galaxy and Galaxies*]{}, proc. IAU Symp. 184, in press George, I. M., & Fabian, A. C. 1991, MNRAS, 249, 352 Green, A. 1993, PhD Thesis, University of Southampton, UK Greenhill, L. J. 1998, in [*Radio Emission from Galactic and Extragalactic Compact Sources*]{}, proc. IAU Coll. 164, eds. J. Zensus et al., ASP Conference Series, in press. Greenhill, L. J., Moran, J. M., & Herrnstein, J. R. 1997, ApJ, 481, L23 Haardt, F., Maraschi, L., & Ghisellini, G. 1994, ApJ, 432, L95 Halpern, J. 1982, PhD Thesis, Harvard University Harms, R., et al. 1994, ApJ, 435, L35 Ho, L. C., 1998, in [*The Central Regions of the Galaxy and Galaxies*]{}, proc. IAU Symp. 184, in press Ichimaru, S. 1987, ApJ, 214, 840 Iwasawa, K., et al. 1993, ApJ, 409, 155 Iwasawa, K., Fabian A. C., Brandt W. N., Kunieda H., Misaki K., Reynolds C. S., & Terashima Y. 1998, MNRAS, submitted Johnson, W. N., Zdziarski, A. A., Madejski, G. M., Paciesas, W. S., Steinle, H., & Lin, Y.-C. 1997, in [*Proceedings of the Fourth Compton Symposium*]{}, eds. C. D. Dermer, M. S. Strickman, & J. D. Kurfess (AIP: New York), AIP Conference Proceedings 410, p. 283 Königl, A. 1981, ApJ, 243, 700 Kormendy, J., & Richstone, D. 1995, ARA&A, 33, 581 Kriss, G. A., & Canizares, C. 1985, ApJ, 297, 177 Krolik, J., Done, C., & Madejski, G. 1993, ApJ, 402, 432 Laor, A., Fiore, F., Elvis, M., Wilkes, B., & McDowell, J. 1997, ApJ, 477, 93 Lasota, J.-P., Abramowicz, M., Chen, X., Krolik, J., Narayan, R., & Yi, I. 1996, ApJ, 462, 142 Leighly, K., & O’Brien, P. 1997, ApJ, 481, L15 Lightman, A. P., & White, T. 1988, ApJ, 335, 57 Madejski, G. M., et al. 1995, ApJ, 438, 672 Madejski, G. M., et al. 1993, Nature, 365, 626 Magdziarz, P., Blaes, O., Zdziarski, A., Johnson, W., & Smith, D. 1998, MNRAS, in press Magdziarz, P., & Blaes, O. 1998, in Proc. IAU Symp. 188, Kyoto, Japan, in press Makishima, K. 1986, in [*The Physics of Accretion onto Compact Objects*]{}, ed. K. Mason, M. Watson, & N. White (Springer-Verlag: Berlin), p. 249 Maoz, E. 1995, ApJ, 447, L91 Maraschi, L., et al. 1994, ApJ, 435, L91 Marshall, H. L., et al. 1997, ApJ, 479, 222 McHardy, I. 1989, in [*Two Topics in X–ray Astronomy*]{}, Proc. 23rd ESLAB Symp., eds. N. White, J. Hunt & B. Battrick, (ESA Publications: Paris), vol. SP-296, p. 1111 Miyoshi, M., et al. 1995, Nature, 373, 127 Moderski, R., Sikora, M., & Lasota, J.-P. 1997, in [*Relativistic Jets in AGNs*]{}, eds. M. Ostrowski et al. (Astronomical Observatiory of the Jagiellonian University: Krakow) p. 110 Mushotzky, R. F. 1980, Adv. Sp. Res., 3, 10 Mushotzky, R. F., Done, C., & Pounds, K. A. 1993, ARA&A, 31, 717 Nandra, K., George, I., Mushotzky, R. F., Turner, T. J., & Yaqoob, T. 1997a, ApJ, 477, 602 Nandra, K., Mushotzky, R. F., George, I., Turner, T. J., & Yaqoob, T. 1997b, ApJ, 488, L91 Narayan, R., & Yi, I. 1994, ApJ, 428, L13 Neufeld, D. A., & Maloney, P. R. 1995, ApJ, 447, L17 Papadakis, I. E., & Lawrence, A. 1995, MNRAS, 272, 161 Papadakis, I. E., & Lawrence, A. 1993, Nature, 361, 250 Pietrini, P., & Krolik, J. 1995, ApJ, 447, 526 Pounds, K., Nandra, K., Stewart, G., George, I., & Fabian, A. 1990, Nature, 344, 132 Poutanen, J., & Svensson, R. 1996, ApJ, 470, 249 Poutanen, J., Krolik, J., & Ryde, F. 1997, in [*Proceedings of the Fourth Compton Symposium*]{}, eds. C. D. Dermer, M. S. Strickman, & J. D. Kurfess (AIP: New York), AIP Conference Proceedings 410, p. 972 Ptak, A. 1997, PhD thesis, University of Maryland, College Park, MD Rees, M. J. 1967, MNRAS, 135, 345 Reynolds, C., DiMateo, T., Fabian, A., Hwang, U., & Canizares, C. 1996, MNRAS, 283, L111 Reynolds, C., & Begelman, M. C. 1998, ApJ, in press Rothschild, R. E., Mushotzky, R. F., Baity, W. A., Gruber, D. E., Matteson, J. L., & Peterson, L. E. 1983, ApJ, 269, 423 Różańska, A., 1998, MNRAS submitted Salpeter, E. E. 1964, ApJ, 140, 796 Schmidt, M. 1963, Nature, 197, 1040 Sikora, M., 1997, in [*Proceedings of the Fourth Compton Symposium*]{}, eds. C. D. Dermer, M. S. Strickman, & J. D. Kurfess (AIP: New York), AIP Conference Proceedings 410, p. 494 Sikora, M., Begelman, M. C., & Rees, M. 1994, ApJ, 421, 153 Sikora, M., Madejski, G. M., Moderski, R., & Poutanen, J. 1997, ApJ, 484, 108 Smith, D., & Done, C. 1996, MNRAS, 280, 355 Sunyaev, R., & Titarchuk, L. 1980, A&A, 86, 121 Tanaka, Y., et al. 1995, Nature, 375, 659 Turner, T. J. 1988, PhD Thesis, University of Leicester, UK Turner, T. J., & Pounds, K. 1989, MNRAS, 240, 833 Ulrich, M.-H., Maraschi, L., & Urry, C. M. 1997, ARA&A, 35, 445 Vermeulen, R. C., & Cohen, M. H. 1994, ApJ, 430, 467 Vio, R., Cristiani, S., Lessi, O., & Provenzale, A. 1992, ApJ, 391, 518 Wilson, A., & Colbert, E. 1995, ApJ, 438, 62 Yi, I. 1996, ApJ, 473, 645 Zdziarski, A. A., Ghisellini, G., George, I. M., Svensson, R., Fabian, A. C., & Done, C. 1990, ApJ, 363, L1 Zdziarski, A., & Coppi, P. 1991, ApJ, 376, 480 Zdziarski, A. A., Johnson, W. N., Done, C., Smith, D., & McNaron-Brown, K. 1995, ApJ, 438, L63 Zdziarski, A. A., Johnson, W. N., Poutanen, J., Magdziarz, P., & Gierlinski, M. 1997, in The Transparent Universe, eds. C. Winkler at al. (ESA: Paris), SP-382, 373 Zeldovich, Y., & Novikov, I. 1964, Sov. Phys. Dokl., 158, 811 Zhang, S. N., Cui, W., & Chen, W. 1997, ApJ, 482, L55

ArXiv

--- abstract: '*Truncated Backpropagation Through Time* (truncated BPTT, [@jaeger2002tutorial]) is a widespread method for learning recurrent computational graphs. Truncated BPTT keeps the computational benefits of *Backpropagation Through Time* (BPTT [@werbos:bptt]) while relieving the need for a complete backtrack through the whole data sequence at every step. However, truncation favors short-term dependencies: the gradient estimate of truncated BPTT is biased, so that it does not benefit from the convergence guarantees from stochastic gradient theory. We introduce *Anticipated Reweighted Truncated Backpropagation* (ARTBP), an algorithm that keeps the computational benefits of truncated BPTT, while providing unbiasedness. ARTBP works by using variable truncation lengths together with carefully chosen compensation factors in the backpropagation equation. We check the viability of ARTBP on two tasks. First, a simple synthetic task where careful balancing of temporal dependencies at different scales is needed: truncated BPTT displays unreliable performance, and in worst case scenarios, divergence, while ARTBP converges reliably. Second, on Penn Treebank character-level language modelling [@ptb_proc], ARTBP slightly outperforms truncated BPTT.' author: - 'Corentin Tallec, Yann Ollivier' bibliography: - 'artbp.bib' title: Unbiasing Truncated Backpropagation Through Time --- *Backpropagation Through Time* (BPTT) [@werbos:bptt] is the de facto standard for training recurrent neural networks. However, BPTT has shortcomings when it comes to learning from very long sequences: learning a recurrent network with BPTT requires unfolding the network through time for as many timesteps as there are in the sequence. For long sequences this represents a heavy computational and memory load. This shortcoming is often overcome heuristically, by arbitrarily splitting the initial sequence into subsequences, and only backpropagating on the subsequences. The resulting algorithm is often referred to as *Truncated Backpropagation Through Time* (truncated BPTT, see for instance [@jaeger2002tutorial]). This comes at the cost of losing long term dependencies. We introduce *Anticipated Reweighted Truncated BackPropagation* (ARTBP), a variation of truncated BPTT designed to provide an unbiased gradient estimate, accounting for long term dependencies. Like truncated BPTT, ARTBP splits the initial training sequence into subsequences, and only backpropagates on those subsequences. However, unlike truncated BPTT, ARTBP splits the training sequence into variable size subsequences, and suitably modifies the backpropagation equation to obtain unbiased gradients. Unbiasedness of gradient estimates is the key property that provides convergence to a local optimum in stochastic gradient descent procedures. Stochastic gradient descent with biased estimates, such as the one provided by truncated BPTT, can lead to divergence even in simple situations and even with large truncation lengths (Fig. \[fig:bpttf\]). ARTBP is experimentally compared to truncated BPTT. On truncated BPTT failure cases, typically when balancing of temporal dependencies is key, ARTBP achieves reliable convergence thanks to unbiasedness. On small-scale but real world data, ARTBP slightly outperforms truncated BPTT on the test case we examined. ARTBP formalizes the idea that, on a day-to-day basis, we can perform short term optimization, but must reflect on long-term effects once in a while; ARTBP turns this into a provably unbiased overall gradient estimate. Notably, the many short subsequences allow for quick adaptation to the data, while preserving overall balance. Related Work ============ BPTT [@werbos:bptt] and its truncated counterpart [@jaeger2002tutorial] are nearly uncontested in the recurrent learning field. Nevertheless, BPTT is hardly applicable to very long training sequences, as it requires storing and backpropagating through a network with as many layers as there are timesteps [@ilya-thesis]. Storage issues can be partially addressed as in [@constant-mem-bptt], but at an increased computational cost. Backpropagating through very long sequences also implies performing fewer gradient descent steps, which significantly slows down learning [@ilya-thesis]. Truncated BPTT heuristically solves BPTT deficiencies by chopping the initial sequence into evenly sized subsequences. Truncated BPTT truncates gradient flows between contiguous subsequences, but maintains the recurrent hidden state of the network. Truncation biases gradients, removing any theoretical convergence guarantee. Intuitively, truncated BPTT has trouble learning dependencies above the range of truncation. [^1] *NoBackTrack* [@nobacktrack] and *Unbiased Online Recurrent Optimization* (UORO) [@uoro] both scalably provide unbiased online recurrent learning algorithms. They take the more extreme point of view of requiring memorylessness, thus forbidding truncation schemes and any storage of past states. NoBackTrack and UORO’s fully online, streaming structure comes at the price of noise injection into the gradient estimates via a random rank-one reduction. ARTBP’s approach to unbiasedness is radically different: ARTBP is not memoryless but does not inject artificial noise into the gradients, instead, compensating for the truncations directly inside the backpropagation equation. Background on recurrent models {#sec:back} ============================== The goal of recurrent learning algorithms is to optimize a parametric dynamical system, so that its output sequence, or predictions, is as close as possible to some target sequence, known a priori. Formally, given a dynamical system with state $s$, inputs $x$, parameter $\theta$, and transition function $F$, $$s_{t+1} = F(x_{t+1}, s_t, \theta)$$ the aim is to find a $\theta$ minimizing a total loss with respect to target outputs ${o^\ast}_t$ at each time, $$\mathcal{L}_T = \sum\limits_{t=1}^T \ell_t = \sum\limits_{t=1}^T\ell(s_t, {o^\ast}_t).$$ A typical case is that of a standard recurrent neural network (RNN). In this case, $s_t = (o_t, h_t)$, where $o_t$ are the activations of the output layer (encoding the predictions), and $h_t$ are the activations of the hidden recurrent layer. For this simple RNN, the dynamical system takes the form $$\begin{aligned} h_{t+1} &= \tanh(W_x\,x_{t+1} + W_h\,h_t + b)\\ o_{t+1} &= W_o h_{t+1}\\ \ell_{t+1} &= \ell(o_{t+1}, {o^\ast}_{t+1})\end{aligned}$$ with parameters $\theta = (W_x, W_h, b)$. Commonly, $\theta$ is optimized via a gradient descent procedure, i.e. iterating $$\theta \leftarrow \theta - \eta \frac{\partial \mathcal{L}_T}{\partial \theta}$$ where $\eta$ is the learning rate. The focus is then to efficiently compute $\partial \mathcal{L}_T/\partial \theta$. *Backpropagation through time* is a method of choice to perform this computation. BPTT computes the gradient by unfolding the dynamical system through time and backpropagating through it, with each timestep corresponding to a layer. BPTT decomposes the gradient as a sum, over timesteps $t$, of the effect of a change of parameter at time $t$ on all subsequent losses. Formally, $$\frac{\partial \mathcal{L}_T}{\partial \theta} = \sum\limits_{t=1}^T {\olddelta \hspace{-0.3mm}}\ell_t \, \frac{\partial F}{\partial \theta}(x_t, s_{t-1}, \theta) \label{eq:gradient}$$ where ${\olddelta \hspace{-0.3mm}}\ell_t{\mathrel{\mathop:}=}\frac{\partial \mathcal{L}_T}{\partial s_t}$ is computed backward iteratively according to the backpropagation equation $$\begin{dcases} {\olddelta \hspace{-0.3mm}}\ell_T = \frac{\partial \ell}{\partial s}(s_T, {o^\ast}_T)\\ {\olddelta \hspace{-0.3mm}}\ell_{t} = {\olddelta \hspace{-0.3mm}}\ell_{t+1}\,\frac{\partial F}{\partial s}(x_{t+1}, s_t, \theta) + \frac{\partial \ell}{\partial s}(s_t, {o^\ast}_t). \label{eq:BPTT} \end{dcases}$$ These backpropagation equations extend the classical ones [@jaeger2002tutorial], which deal with the case of a simple RNN for $F$. Unfortunately, BPTT requires processing the full sequence both forward and backward. This requires maintaining the full unfolded network, or equivalently storing the full history of inputs and activations (though see [@constant-mem-bptt]). This is impractical when very long sequences are processed with large networks: processing the whole sequence at every gradient step slows down learning. Practically, this is alleviated by truncating gradient flows after a fixed number of timesteps, or equivalently, splitting the input sequence into subsequences of fixed length, and only backpropagating through those subsequences. [^2] This algorithm is referred to as *Truncated BPTT*. With truncation length $L< T$, the corresponding equations just drop the recurrent term ${\olddelta \hspace{-0.3mm}}{\ell}_{t+1}\,\frac{\partial F}{\partial s}(x_{t+1}, s_t, \theta)$ every $L$ time steps, namely, $$\begin{aligned} {\olddelta \hspace{-0.3mm}}\hat{\ell}_t &{\mathrel{\mathop:}=}\begin{dcases} \frac{\partial \ell}{\partial s}(s_t, {o^\ast}_t) &\text{if $t$ is a multiple of $L$}\\ {\olddelta \hspace{-0.3mm}}\hat{\ell}_{t+1}\,\frac{\partial F}{\partial s}(x_{t+1}, s_t, \theta) + \frac{\partial \ell}{\partial s}(s_t, {o^\ast}_t) & \text{otherwise.} \end{dcases}\end{aligned}$$ This also allows for online application: for instance, the gradient estimate from the first subsequence $t=1\ldots,L$ does not depend on anything at time $t>L$. However, this gradient estimation scheme is heuristic and provides biased gradient estimates. In general the resulting gradient estimate can be quite far from the true gradient even with large truncations $L$ (Section \[sec:exp\]). Undesired behavior, and, sometimes, divergence can follow when performing gradient descent with truncated BPTT (Fig. \[fig:bpttf\]). Anticipated Reweighted Backpropagation Through Time: unbiasedness through reweighted stochastic truncation lengths {#sec:artbp} ================================================================================================================== Like truncated BPTT, ARTBP splits the initial sequence into subsequences, and only performs backpropagation through time on subsequences. However, contrary to the latter, it does not split the sequence evenly. The length of each subsequence is sampled according to a specific probability distribution. Then the backpropagation equation is modified by introducing a suitable reweighting factor at every step to ensure unbiasedness. Figure \[fig:intuition\] demonstrates the difference between BPTT, truncated BPTT and ARTBP. [0.40]{} ![Graphical representation of BPTT, truncated BPTT and ARTBP. Blue arrows represent forward propagations, red arrows backpropagations. Dots represent either internal state resetting or gradient resetting.[]{data-label="fig:intuition"}](plots/graphs/bptt.pdf "fig:"){width="5cm"} [0.40]{} ![Graphical representation of BPTT, truncated BPTT and ARTBP. Blue arrows represent forward propagations, red arrows backpropagations. Dots represent either internal state resetting or gradient resetting.[]{data-label="fig:intuition"}](plots/graphs/tbptt.pdf "fig:"){width="5cm"} [0.40]{} ![Graphical representation of BPTT, truncated BPTT and ARTBP. Blue arrows represent forward propagations, red arrows backpropagations. Dots represent either internal state resetting or gradient resetting.[]{data-label="fig:intuition"}](plots/graphs/artbp.pdf "fig:"){width="5cm"} Simply sampling arbitrarily long truncation lengths does not provide unbiasedness. Intuitively, it still favors short term gradient terms over long term ones. When using full BPTT, gradient computations flow back [^3] from every timestep $t$ to every timestep $t'<t$. In truncated BPTT, gradients do not flow from $t$ to $t'$ if $t-t'$ exceeds the truncation length. In ARTBP, since random truncations are introduced, gradient computations flow from $t$ to $t'$ with a certain probability, decreasing with $t-t'$. To restore balance, ARTBP rescales gradient flows by their inverse probability. Informally, if a flow has a probability $p$ to occur, multiplication of the flow by $\frac{1}{p}$ restores balance on average. Formally, at each training epoch, ARTBP starts by sampling a random sequence of truncation points, that is $(X_t)_{1\leq t \leq T} \in \{0, 1\}^T$. A truncation will occur at all points $t$ such that $X_t=1$. Here $X_t$ may have a probability law that depends on $X_1, \ldots, X_{t-1}$, and also on the sequence of states $(s_t)_{1\leq t\leq T}$ of the system. The reweighting factors that ARTBP introduces in the backpropagation equation depend on these truncation probabilities. (Unbiasedness is not obtained just by global importance reweighting between the various truncated subsequences: indeed, the backpropagation equation inside each subsequence has to be modified at every time step, see .) The question of how to choose good probability distributions for the truncation points $X_t$ is postponed till Section \[sec:cchoice\]. Actually, unbiasedness holds for any choice of truncation probabilities (Prop \[prop:artbp\]), but different choices for $X_t$ lead to different variances for the resulting gradient estimates. \[prop:artbp\] Let $(X_t)_{t=1...T}$ be any sequence of binary random variables, chosen according to probabilities $$c_t {\mathrel{\mathop:}=}\mathbb{P}(X_t=1\mid X_{t-1}, \ldots, X_1)$$ and assume $c_t\neq 1$ for all $t$. Define ARTBP to be backpropagation through time with a truncation between $t$ and $t+1$ iff $X_t=1$, and a compensation factor $\frac{1}{1-c_t}$ when $X_t=0$, namely: $$\begin{aligned} \label{eq:artbp} {\olddelta \hspace{-0.3mm}}\tilde{\ell}_t &{\mathrel{\mathop:}=}\begin{dcases} \frac{\partial \ell}{\partial s}(s_t, {o^\ast}_t) &\text{ if $X_t=1$ or $t=T$}\\ \frac{1}{1-c_t}\, {\olddelta \hspace{-0.3mm}}\tilde{\ell}_{t+1}\,\frac{\partial F}{\partial s}(x_{t+1}, s_t, \theta) + \frac{\partial \ell}{\partial s}(s_t, {o^\ast}_t) &\text{ otherwise.} \end{dcases}\end{aligned}$$ Let $\tilde{g}$ be the gradient estimate obtained by using ${\olddelta \hspace{-0.3mm}}\tilde{\ell}_t$ instead of ${\olddelta \hspace{-0.3mm}}\ell_t$ in ordinary BPTT , namely $$\tilde{g}{\mathrel{\mathop:}=}\sum\limits_{t=1}^T {\olddelta \hspace{-0.3mm}}\tilde{\ell}_t \, \frac{\partial F}{\partial \theta}(x_t, s_{t-1}, \theta)$$ Then, on average over the ARTBP truncations, this is an unbiased gradient estimate of the total loss: $$\mathbb{E}_{X_1, \ldots, X_T}\left[\,\tilde{g}\,\right] = \frac{\partial \mathcal{L}_T}{\partial \theta}.$$ The core of the proof is as follows: With probability $c_t$ (truncation), ${\olddelta \hspace{-0.3mm}}\tilde\ell_{t+1}$ does not contribute to ${\olddelta \hspace{-0.3mm}}\tilde\ell_t$. With probability $1-c_t$ (no truncation), it contributes with a factor $\frac{1}{1-c_t}$. So on average, ${\olddelta \hspace{-0.3mm}}\tilde\ell_{t+1}$ contributes to ${\olddelta \hspace{-0.3mm}}\tilde\ell_t$ with a factor $1$, and ARTBP reduces to standard, non-truncated BPTT on average. The detailed proof is given in Section \[sec:proof\]. While the ARTBP gradient estimate above is unbiased, some noise is introduced due to stochasticity of the truncation points. It turns out that ARTBP trades off memory consumption (larger truncation lengths) for variance, as we now discuss. Choice of $c_t$ and memory/variance tradeoff {#sec:cchoice} ============================================ ARTBP requires specifying the probability $c_t$ of truncating at time $t$ given previous truncations. Intuitively the $c$’s regulate the average truncation lengths. For instance, with a constant $c_t\equiv c$, the lengths of the subsequences between two truncations follow a geometric distribution, with average truncation length $\frac{1}{c}$. Truncated BPTT with fixed truncation length $L$ and ARTBP with fixed $c=\frac{1}{L}$ are thus comparable memorywise. Small values of $c_t$ will lead to long subsequences and gradients closer to the exact value, while large values will lead to shorter subsequences but larger compensation factors $\frac{1}{1-c_t}$ and noisier estimates. In particular, the product of the $\frac{1}{1-c_t}$ factors inside a subsequence can grow quickly. For instance, a constant $c_t$ leads to exponential growth of the cumulated $\frac{1}{1-c_t}$ factors when iterating . To mitigate this effect, we suggest to set $c_t$ to values such that the probability to have a subsequence of length $L$ decreases like $L^{-\alpha}$. The variance of the lengths of the subsequences will be finite if $\alpha>3$. Moreover we might want to control the average truncation length $L_0$. This is achieved via $$c_t=\mathbb{P}(X_t=1\mid X_{t-1}, \ldots, X_1) = \frac{\alpha-1}{(\alpha-2) L_0 + {\olddelta \hspace{-0.3mm}}t} \label{eq:ct}$$ where ${\olddelta \hspace{-0.3mm}}t$ is the time elapsed since the last truncation, ${\olddelta \hspace{-0.3mm}}t=t-\sup \{s\mid s<t, X_s=1\}$. Intuitively, the more time spent without truncating, the lower the probability to truncate. This formula is chosen such that the average truncation length is approximately $L_0$, and the standard deviation from this average length is finite. The parameter $\alpha$ controls the regularity of the distribution of truncation lengths: all moments lower than $\alpha-1$ are finite, the others are infinite. With larger $\alpha$, large lengths will be less frequent, but the compensating factors $\frac{1}{1-c_t}$ will be larger. With this choice of $c_t$, the product of the $\frac1{1-c_t}$ factors incurred by backpropagation inside each subsequence grows polynomially like $L^{\alpha-1}$ in a subsequence of length $L$. If the dynamical system has geometrically decaying memory, i.e., if the operator norm of the transition operator $\frac{\partial F}{\partial s}$ is less than $1-{\varepsilon}$ most of the time, then the value of ${\olddelta \hspace{-0.3mm}}\tilde\ell_t$ will stay controlled, since $(1-{\varepsilon})^L\cdot L^\alpha$ stays bounded. On the other hand, using a constant $c_t\equiv c$ provides bounded ${\olddelta \hspace{-0.3mm}}\tilde\ell_t$ only for small values $c<{\varepsilon}$. In the experiments below, we use the $c_t$ from with $\alpha=4$ or $\alpha=6$. Online implementation ===================== Importantly, ARTBP can be directly applied online, thus providing unbiased gradient estimates for recurrent networks. Indeed, not all truncation points have to be drawn in advance: ARTBP can be applied by sampling the first truncation point, performing both forward and backward passes of BPTT up until this point, and applying a partial gradient descent update based on the resulting gradient on this subsequence. Then one moves to the next subsequence and the next truncation point, etc. (Fig. \[fig:artintuition\]). Experimental validation {#sec:exp} ======================= The experimental setup below aims both at illustrating the theoretical properties of ARTBP compared to truncated BPTT, and at testing the soundness of ARTBP on real world data. Influence balancing ------------------- The influence balancing experiment is a synthetic example demonstrating, in a very simple model, the importance of being unbiased. Intuitively, a parameter has a positive short term influence, but a negative long term one that surpasses the short term effect. Practically, we consider a row of agents, numbered from left to right from $1$ to $p+n$ who, at each time step, are provided with a signal depending on the parameter, and diffuse part of their current state to the agent directly to their left. The $p$ leftmost agents receive a positive signal at each time step, and the $n$ rightmost agents a negative signal. The training goal is to control the state of the leftmost agent. The first $p$ agents contribute positively to the first agent state, while the next $n$ contribute negatively. However, agent $1$ only feels the contribution from agent $k$ after $k$ timesteps. If optimization is blind to dependencies above $k$, the effect of $k$ is never felt. A typical instantiation of such a problem would be that of a drug whose effect varies after various delays; the parameter to be optimized is the quantity of drug to be used daily. Such a model can be formalized as [@uoro] $$s_{t+1} = A \, s_t + (\theta, \ldots, \theta, -\theta, \ldots, -\theta){^\top}$$ with $A$ a square matrix of size $p+n$ with $A_{k,k} = 1/2$, $A_{k,k+1}=1/2$, and $0$ elsewhere; $s_t^k$ corresponds to the state of the $k$-th agent. $\theta \in {\mathbb{R}}$ is a scalar parameter corresponding to the intensity of the signal observed at each time step. The right-hand-side has $p$ positive-$\theta$ entries and $n$ negative-$\theta$ entries. The loss considered is an arbitrary target on the leftmost agent $s^1$, $$\ell_t = {\textstyle \frac12} (s^1_t - 1)^2.$$ The dynamics is illustrated schematically in Figure \[fig:infl-bal\]. ![Influence balancing dynamics, $1$ positive influence, $3$ negative influences.[]{data-label="fig:infl-bal"}](influence-balancing.pdf){width="7cm"} Fixed-truncation BPTT is experimentally compared with ARTBP for this problem. The setting is online: starting at $t=1$, a first truncation length $L$ is selected (fixed for BPTT, variable for ARTBP), forward and backward passes are performed on the subsequence $t=1,\ldots,L$, a vanilla gradient step is performed with the resulting gradient estimate, then the procedure is repeated with the next subsequence starting at $t=L+1$, etc.. Our experiment uses $p=10$ and $n=13$, so that after $23$ steps the signal should have had time to travel through the network. Truncated BPTT is tested with various truncations $L=10,100,200$. (As the initial $\theta$ is fixed, truncated BPTT is deterministic in this experiment, thus we only provide a single run for each $L$.) ARTBP is tested with the probabilities using $L_0=16$ (average truncation length) and $\alpha=6$. ARTBP is stochastic: five random runs are provided to test reliability of convergence. The results are displayed in Fig. \[fig:bpttf\]. We used decreasing learning rates $\eta_t = \frac{\eta_0}{\sqrt{1 + t}}$ where $\eta_0=3\times 10^{-4}$ is the initial learning rate and $t$ is the timestep. We plot the average loss over timesteps $1$ to $t$, as a function of $t$. Truncated BPTT diverges even for truncation ranges largely above the intrinsic temporal scale of the system. This is an expected result: due to bias, truncated BPTT ill-balances temporal dependencies and estimates the overall gradient with a wrong sign. In particular, reducing the learning rate will *not* prevent divergence. On the other hand, ARTBP reliably converges on every run. Note that for the largest truncation $L=200$, truncated BPTT finally converges, and does so at a faster rate than ARTBP. This is because this particular problem is deterministic, so that a deterministic gradient scheme will converge (if it does converge) geometrically like $O(e^{-\lambda t})$, whereas ARTBP is stochastic due to randomization of truncations, and so will not converge faster than $O(t^{-1/2})$. This difference would disappear, for instance, with noisy targets or a noisy system. #### Character-level Penn Treebank language model. We compare ARTBP to truncated BPTT on the character-level version of the Penn Treebank dataset, a standard set of case-insensitive, punctuation-free English text [@ptb]. Character-level language modelling is a common benchmark for recurrent models. The dataset is split into training, validation and test sets following [@ptb_proc]. Both ARTBP and truncated BPTT are used to train an LSTM model [@lstm] with a softmax classifier on its hidden state, on the character prediction task. The training set is batched into $64$ subsets processed in parallel to increase computing speed. Before each full pass on the training set, the batched training sequences are split into subsequences: - for truncated BPTT, of fixed size $50$; - for ARTBP, at random following the scheme with $\alpha=4$ and $L_0=50$. Truncated BPTT and ARTBP process these subsequences sequentially, [^4] as in Fig. \[fig:intuition\]. The parameter is updated after each subsequence, using the Adam [@adam] stochastic gradient scheme, with learning rate $10^{-4}$. The biases of the LSTM unit forget gates are set to $2$, to prevent early vanishing gradients [@forget_init]. Results (in bits per character, bpc) are displayed in Fig. \[fig:ptb\]. Six randomly sampled runs are plotted, to test reliability. [0.47]{} [0.47]{} In this test, ARTBP slightly outperforms truncated BPTT in terms of validation and test error, while the reverse is true for the training error (Fig. \[fig:ptb\]). Even with ordinary truncated BPTT, we could not reproduce reported state of the art results, and do somewhat worse. We reach a test error of $1.43$ bpc with standard truncated BPTT and $1.40$ bpc with ARTBP, while reported values with similar LSTM models range from $1.38$ bpc [@bn_bench] to $1.26$ bpc [@graves_bench] (the latter with a different test/train split). This may be due to differences in the experimental setup: we have applied truncated BPTT without subsequence shuffling or gradient clipping [@graves_bench] (incidentally, both would break unbiasedness). Arguably, the numerical issues solved by gradient clipping are model specific, not algorithm specific, while the point here was to compare ARTBP to truncated BPTT for a given model. Conclusion ========== We have shown that the bias introduced by truncation in the backpropagation through time algorithm can be compensated by the simple mathematical trick of randomizing the truncation points and introducing compensation factors in the backpropagation equation. The algorithm is experimentally viable, and provides proper balancing of the effects of different time scales when training recurrent models. Proof of Proposition \[prop:artbp\] {#sec:proof} =================================== First, by backward induction, we show that for all $t\leq T$, for all $x_1,\ldots,x_{t-1}\in\{0,1\}$, $$\label{eq:induction} {\mathbb{E}}\left[ {\olddelta \hspace{-0.3mm}}\tilde\ell_t \mid X_{1:t-1}=x_{1:t-1} \right] = {\olddelta \hspace{-0.3mm}}\ell_t$$ where ${\olddelta \hspace{-0.3mm}}\ell_t$ is the value obtained by ordinary BPTT . Here $x_{1:k}$ is short for $(x_1,\ldots,x_k)$. For $t=T$, this holds by definition: ${\olddelta \hspace{-0.3mm}}\tilde{\ell}_T = \frac{\partial \ell}{\partial s}(s_T, {o^\ast}_T) = {\olddelta \hspace{-0.3mm}}\ell_T$. Assume that the induction hypothesis holds at time $t+1$. Note that the values $s_t$ do not depend on the random variables $X_t$, as they are computed during the forward pass of the algorithm. In particular, the various derivatives of $F$ and $\ell$ in do not depend on $X_{1:T}$. Thus $$\begin{aligned} {\mathbb{E}}& \left[ {\olddelta \hspace{-0.3mm}}\tilde\ell_t \mid X_{1:t-1}=x_{1:t-1} \right]=\nonumber\\ & {\mathbb{P}}(X_t=1\mid X_{1:t-1}=x_{1:t-1}) \,{\mathbb{E}}\left[ {\olddelta \hspace{-0.3mm}}\tilde\ell_t \mid X_{1:t-1}=x_{1:t-1}, X_t=1 \right] +\\&\quad {\mathbb{P}}(X_t=0\mid X_{1:t-1}=x_{1:t-1}) \,{\mathbb{E}}\left[ {\olddelta \hspace{-0.3mm}}\tilde\ell_t \mid X_{1:t-1}=x_{1:t-1},X_t=0 \right] \\ =&\;c_t \,{\mathbb{E}}\left[ {\olddelta \hspace{-0.3mm}}\tilde\ell_t \mid X_{1:t-1}=x_{1:t-1},X_t=1 \right] +(1-c_t)\, {\mathbb{E}}\left[ {\olddelta \hspace{-0.3mm}}\tilde\ell_t \mid X_{1:t-1}=x_{1:t-1},X_t=0 \right]\label{eq:conditioned}\end{aligned}$$ If $X_t=1$ then ${\olddelta \hspace{-0.3mm}}\tilde\ell_t=\frac{\partial \ell}{\partial s} (s_t, {o^\ast}_t)$. If $X_t=0$, then ${\olddelta \hspace{-0.3mm}}\tilde\ell_t= \frac{\partial \ell}{\partial s} (s_t, {o^\ast}_t) + \frac{1}{1-c_t}\,{\olddelta \hspace{-0.3mm}}\tilde{\ell}_{t+1}\,\frac{\partial F}{\partial s}(x_{t+1}, s_t, \theta)$. Therefore, substituting into , $$\begin{aligned} {\mathbb{E}}& \left[ {\olddelta \hspace{-0.3mm}}\tilde\ell_t \mid X_{1:t-1}=x_{1:t-1}\right] =\frac{\partial \ell}{\partial s} (s_t, {o^\ast}_t) + {\mathbb{E}}\left[ {\olddelta \hspace{-0.3mm}}\tilde{\ell}_{t+1} \mid X_{1:t-1}=x_{1:t-1},X_t=0 \right] \frac{\partial F}{\partial s}(x_{t+1}, s_t, \theta) $$ but by the induction hypothesis at time $t+1$, this is exactly $\frac{\partial \ell}{\partial s} (s_t, {o^\ast}_t)+{\olddelta \hspace{-0.3mm}}\ell_{t+1}\frac{\partial F}{\partial s}(x_{t+1}, s_t, \theta)$, which is ${\olddelta \hspace{-0.3mm}}\ell_t$. Therefore, ${\mathbb{E}}\left[{\olddelta \hspace{-0.3mm}}\tilde\ell_t\right]={\olddelta \hspace{-0.3mm}}\ell_t$ unconditionally. Plugging the ${\olddelta \hspace{-0.3mm}}\tilde{\ell}$’s into , and averaging $$\begin{aligned} \mathbb{E}_{X_1, \ldots, X_T} \left[\,\tilde{g}\,\right] &= \sum\limits_{t=1}^T \mathbb{E}_{X_t, \ldots, X_T} \left[{\olddelta \hspace{-0.3mm}}\tilde{\ell}_t\right]\frac{\partial F}{\partial \theta}(x_{t}, s_{t-1}, \theta)\\ &= \sum\limits_{t=1}^T {\olddelta \hspace{-0.3mm}}\ell_t\,\frac{\partial F}{\partial \theta}(x_{t}, s_{t-1}, \theta)\\ &= \frac{\partial \mathcal{L}_T}{\partial \theta} \end{aligned}$$ which ends the proof. [^1]: Still, as the hidden recurrent state is not reset between subsequences, it may contain hidden information about the distant past, which can be exploited [@ilya-thesis]. [^2]: Usually the internal state $s_t$ is maintained from one subsequence to the other, not reset to a default value. [^3]: Gradient flows between timesteps $t$ and $t'$ if there are no truncations occuring between $t$ and $t'$. [^4]: Subsequences are not shuffled, as we do not reset the internal state of the network between subsequences.

ArXiv

--- abstract: 'Recent observations with the Spitzer Space Telescope show clear evidence that star formation takes place in the surrounding of young massive O-type stars, which are shaping their environment due to their powerful radiation and stellar winds. In this work we investigate the effect of ionising radiation of massive stars on the ambient interstellar medium (ISM): In particular we want to examine whether the UV-radiation of O-type stars can lead to the observed pillar-like structures and can trigger star formation. We developed a new implementation, based on a parallel Smooth Particle Hydrodynamics code (called IVINE), that allows an efficient treatment of the effect of ionising radiation from massive stars on their turbulent gaseous environment. Here we present first results at very high resolution. We show that ionising radiation can trigger the collapse of an otherwise stable molecular cloud. The arising structures resemble observed structures (e.g. the pillars of creation in the Eagle Nebula (M16) or the Horsehead Nebula B33). Including the effect of gravitation we find small regions that can be identified as formation places of individual stars. We conclude that ionising radiation from massive stars alone can trigger substantial star formation in molecular clouds.' date: '?? and in revised form ??' --- Overview ======== In the surroundings of hot OB-Associations filamentary substructures on different scales are observed (see e.g. [@2002ApJ...565L..25S] and references therein). As observational resolution increases, more and more sub-millimeter sources, which could trace the birth of future stars are detected (e.g. [@2006MNRAS.369.1201W]). It has long been suggested that radiation driven implosion of molecular clouds can explain the morphology and the star formation in these regions (e.g. [@1995ApJ...451..675E]). Recent simulations (see e.g. [@2006ApJ...647..397M], [@2005MNRAS.358..291D], [@2003MNRAS.338..545K]) demonstrate the importance of massive stars for the subsequent evolution of their parental molecular clouds. The ionising radiation is a vital ingredient to understand the disruption of molecular clouds and their star formation efficiency. Our goal is to investigate the morphology of molecular clouds and the formation of protostars in much greater detail. To do so we use very high resolution simulations of a small region of a molecular cloud ionised by a massive nearby star. Numerical Method ================ We use the prescription for ionising UV-radiation of a young massive star proposed by [@2000MNRAS.315..713K]. The ionisation degree $x$ is related to the hydrodynamical quantities by an approximation for the resulting temperature of a partly ionised gas $$T = T_{ion} \cdot x + T_{cold} \cdot (1-x).$$ $ T_{cold} $ is the initial temperature of the cold, unionised gas and $ T_{ion} $ is the temperature of the ionised gas. To treat the hydrodynamic and gravitational evolution we use a parallel smoothed particle hydrodynamics (SPH) code called VINE (Wetzstein , in prep.). Its Lagrangian nature renders it extremely adept to cover several orders of magnitude in density and time, which is important to follow local gas collapse. We assume plane-parallel UV-irradiation of the simulated area, mimicking a radiation source sufficiently far away such that its distance is larger than the dimensions of the area of infall. To couple ionisation to hydrodynamics we use a flux conserving ray-shooting algorithm. A two dimensional grid is superimposed on the area of interest. Along each of the thereby created bins, the optical depth is calculated. The size of each bin, i.e. the grid resolution is defined by the volume each SPH-particle occupies. This guarantees that the density information given by the SPH-formalism is transformed to the calculation of radiation correctly. This implementation is fully parallelised. We call it IVINE (Ionisation+VINE). Numerical Tests =============== A standard test for numerical implementations of ionizing radiation has been proposed by . It deals with the steady propagation of an ionisation front: a box of constant density is exposed to a time-dependent ionising source. The initial conditions were chosen to be $n_0=100cm^{-3}$ and $T_{cold} = 100K$ to enable a direct comparison to their results. The flux increases linearly with time, starting at zero: $\mathrm{d}J/\mathrm{d}t = 5.07\cdot10^{-8}cm^{-2}s^{-2}$. The recombination parameter $\alpha_B$ is set to $\alpha_B= 2.7\cdot10^{-13}cm^3s^{-1}$. The ionised temperature is $T_{ion}=10^4K$. At the beginning a small fraction of the box is ionised. Due to the higher temperature of the ionised gas a shock front evolves. This shock front moves at a constant speed through the box. The analytical solution provides the exact position and speed of the front as well as of the ionised gas at any given time. We find a very good agreement with the analytical solution, the results are shown in Table \[leflochtbl\]. $n_c$ and $n_i$ denote the number density of the compressed layer and of the ionised gas, $v_i$ and $v_s$ are the velocities of the ionisation front and the shock front respectively. Analytical [@2000MNRAS.315..713K] IVINE ------------------ ------------ ------- ------------------------ ------- $n_c(cm^{-3})$ 159 169 155 156.8 $n_i(cm^{-3})$ 0.756 0.748 0.75 0.756 $v_i(km s^{-1})$ 3.48 3.36 3.43 3.41 $v_s(km s^{-1})$ 3.71 3.51 3.67 3.63 : Comparison of analytical and numerical results for the Lefloch test.\[leflochtbl\] First application: ionisation of a turbulent ISM ================================================ Initial conditions ------------------ The first high resolution simulations we performed with the new code address the effect of ionising radiation on a box of turbulent medium. We choose the initial conditions to mimick observed turbulence in the ISM. The cubic simulation domain with a volume of $(2pc)^3$ and a mean density of $\bar{n} = 100cm^{-3}$ is set up with a temperature of $T_{cold}=10K$. The turbulent velocity field is set up adapting a Gaussian random field with a steep power spectrum. The velocity field generates density fluctuations and after a dynamical timescale a turbulent medium with typical velocities of Mach 5 has been generated (see Fig. \[fig\_init\]). At this point the box is exposed to ionising radiation. The UV-radiation is impinging form the negative x-direction with a flux $J = 8.36\cdot10^8cm^{-2}s^{-1}$. This leads to a rapid ionisation of the first $\approx5\%$ of the cube before the medium reacts to the increased temperature of the ionised gas. The simulations were performed with 2 Million particles on a SGI Altix supercomputer. Results ------- The UV-radiation traces the turbulent density distribution, reaching further into the low density regions, and less far in the regions of high density. After a dynamical timescale the hydrodynamics react to the increase in temperature, shock fronts evolve and compress the gas while at the same time increasing the turbulent energy of the cold gas. During this phase a typical morphology evolves. The denser regions shadow regions behind them whereas in lower density regimes the radiation can propagate much further. After the first phase of maximum compression a more quiescent phase of evaporation sets in. The densities are not as high as before, but the structures become even more clear. This leads in the final stage to filamentary, pillar-like substructures, pointing towards the source of radiation as can be seen in Fig. \[fig\_end\]. The structures contain high density regions in their tips. These a very likely to become gravitational unstable after a free fall time. Conclusion ========== We developed a fully parallel treatment for the ionising radiation of young massive stars. Ionising Radiation alone is sufficient to explain the morphology observed in the surroundings of hot OB-clusters. In our simulations of turbulent ISM exposed to UV-radiation characteristic trunks similar to the ones observed in M16 evolve. Further studies including gravity will show whether the UV-radiation from young massive stars is sufficient to trigger gravitational collapse within these pillars. ![Initial density distribution after the turbulence has decayed to Mach 5. The box dimension are $2pc$ in each direction.[]{data-label="fig_init"}](./init_box_bw.eps){width="12cm"} ![Final stage of evolution after $t \approx 300kyrs$: the morphology has clearly evolved. The size of the most prominent pillar is roughly $1pc$ as it is observed in e.g. M16 []{data-label="fig_end"}](./end_box_bw.eps){width="12cm"} M. Gritschneder is supported by the [*Sonderforschungsbereich 375-95 Astro-Particle-Physics*]{} of the Deutsche Forschungsgemeinschaft. , 2005, *MNRAS* 358, 291 , 2000, *MNRAS* 315, 713 , 2003, *MNRAS* 338, 545 , 1994, *A&A* 289, 559 , 2006 *ApJ* 647, 397 *ApJ* 451, 675 , 2002 *ApJ* 565, L25 , 2006 *MNRAS* 369, 1201

ArXiv

--- abstract: 'We present an updated version (Nijm93) of the Nijmegen soft-core potential, which gives a much better description of the $np$ data than the older version (Nijm78). The $\chi^{2}$ per datum is 1.87. The configuration-space and momentum-space versions of this potential are exactly equivalent; a unique feature among meson-theoretical potentials. We also present three new $N\!N$ potential models: a non-local Reid-like Nijmegen potential (Nijm I), a local version (Nijm II), and an updated regularized version (Reid93) of the Reid soft-core potential. These three potentials all have a nearly optimal $\chi^{2}$ per datum and can therefore be considered as alternative partial-wave analyses. All potentials contain the proper charge-dependent one-pion-exchange tail.' address: - | School of Physical Sciences, The Flinders University of South Australia, Bedford Park, South Australia 5042, Australia [@perm]\ and Institute for Theoretical Physics, University of Nijmegen, Nijmegen, The Netherlands - 'Institute for Theoretical Physics [@email], University of Nijmegen, Nijmegen, The Netherlands' author: - 'V.G.J. Stoks' - 'R.A.M. Klomp, C.P.F. Terheggen, and J.J. de Swart' date: 'submitted to Phys. Rev. C' title: 'Construction of high-quality potential models' --- =10000 INTRODUCTION {#intro} ============ In the past many nucleon-nucleon ($N\!N$) potentials were constructed, which were supposed to fit the $N\!N$ scattering data available at the time of construction. The older models, from the 1950s and 1960s, are no longer suitable for describing the present set of more numerous and much more accurate data without refitting the parameters. Out of the various potential models constructed in the 1970s, the better ones fitted the data with $\chi^{2}/N_{\rm data}$ of about 2, where $N_{\rm data}$ denotes the number of $N\!N$ scattering data available at that time in the 0–350 MeV energy range. The potentials constructed in the 1980s have only slightly improved on this in the sense that, although they have been fitted to try to describe the newer and much more accurate data, these models still have $\chi^{2}/N_{\rm data}\approx2$. This number should be compared with $\chi^{2}_{\rm min}/N_{\rm data}=0.99$, obtained in the recently finished Nijmegen $N\!N$ multienergy partial-wave analysis [@St93b] (PWA93) of all $pp$ and $np$ scattering data below 350 MeV. On statistical grounds, $\chi^{2}_{\rm min}/N_{\rm data}\approx1$ is about the best one can expect to get in partial-wave analyses or for potential models. In a recent paper [@St93a], we investigated the quality with respect to the $pp$ scattering data below 350 MeV of a number of $N\!N$ potentials that had appeared in the literature. We found that only a few of the potential models we investigated are of a satisfactory quality. These models are the Reid soft-core potential [@Rei68] Reid68, the Nijmegen soft-core potential [@Nag78] Nijm78, and the new Bonn $pp$ potential [@Hai89] Bonn89. The latter is a readjustment of the momentum-space full Bonn potential [@Mac87], in order to fit the $pp$ data. If we do not consider the very low-energy (0–2 MeV) $pp$ data, also the parametrized Paris potential [@Lac80] Paris80 gives a satisfactory description of the data. The results of Ref. [@St93a] indicate that, at present, the best potential models have $\chi^{2}/N_{pp}\gtrsim1.9$, where $N_{pp}$ denotes the number of $pp$ scattering data. Moreover, only models which have explicitly included the $pp$ data in their fit belong to this category. Potential models which have been fitted only to the $np$ data often give a poor description of the $pp$ data, even after applying the necessary corrections for the Coulomb interaction. In Ref. [@St93a] we have demonstrated that most $np$ potentials unfortunately do not automatically fit the $pp$ data, a fact which has been generally overlooked. Any $N\!N$ potential should be fitted to the $pp$ data as well as to the $np$ data in order to be able to describe all $N\!N$ scattering data. Over the last decade the quality of the $np$ data has increased considerably. Consequently, the older potentials (Reid68, Nijm78, Paris80) do not fit these data very well. Also, the much newer Bonn potentials already needed revisions and updates [@Hai89; @Mac89]. In this paper we present updates of the Nijm78 and Reid68 potentials, denoted by Nijm93 and Reid93, respectively. Because our analysis of the $np$ data (and hence our careful scrutiny of the $np$ data) has only recently been finished [@St93b], we originally constructed an update (Nijm92$pp$) of the Nijm78 potential for the $pp$ data only. This $pp$ potential was used in our earlier preliminary $np$ analyses [@Klo91; @Klo92; @St93c] to parametrize the isovector partial waves. It has $\chi^{2}/N_{pp}=1.4$, which is not as good as the Nijmegen PWA93. One can wonder whether it is at all possible to construct a new class of potential models which fit the $N\!N$ data with the almost perfect $\chi^{2}/N_{\rm data}\approx1$. The answer turns out to be affirmative. This could already be surmised from the Nijmegen PWA93, because this analysis is in essence an energy-dependent potential fitted to the scattering data. (The reason for us using an energy-dependent potential is nothing more than just convenience.) In the partial-wave analysis we need 39 parameters to reach $\chi^{2}/N_{\rm data}=0.99$, whereas a conventional potential model typically has only 10–15 free parameters. It is therefore perhaps not surprising that the 15-parameter update (Nijm93) of the Nijm78 potential, which fits the $N\!N$ data with $\chi^{2}/N_{\rm data}=1.87$, cannot compete in quality with the Nijmegen PWA93. To obtain a high-quality potential we decided some years ago to follow a different approach. Because the Nijm92$pp$ potential already gives a reasonable description of the $pp$ data, this model forms the basis for the construction of a high-quality potential, which [*can*]{} compete with the Nijmegen PWA93. We adjust in each partial wave separately only a few of the parameters of this potential [@remark]. This way we will be able to construct a potential model which fits the data with $\chi^{2}/N_{\rm data}\approx1$. The resulting Reid-like potential Nijm I gives a very good fit to the data with $\chi^{2}/N_{\rm data}=1.03$. The Nijm I potential contains momentum-dependent terms (as do the Nijm78 and Nijm93 potentials), which in configuration space give rise to a non-local structure $(\Delta\varphi(r)+\varphi(r)\Delta)$ to the potential. We also constructed a purely local Nijm II potential, where these momentum-dependent terms were intentionally omitted. This local potential Nijm II gives an equally good fit to the data as the non-local potential Nijm I. Finally, we constructed a regularized update of the Reid68 potential [@Rei68], called Reid93. This Reid93 model is also a local potential and fits the scattering data very well. These latter three potential models are in a sense also alternative partial-wave analyses, because they have roughly the same number of fit parameters as our Nijmegen PWA93, these parameters were fitted to the same database, and the potential models achieve nearly the same values of $\chi^{2}_{\rm min}$ as the Nijmegen PWA93 (i.e., close to the expectation value). Hence, the differences between, e.g., the phase parameters of these models provide an indication for the systematic error in the Nijmegen partial-wave analyses. In Sec. \[outline\] we briefly discuss some general features of $N\!N$ potentials. In Sec. \[structure\] we give more details regarding the explicit form of the potentials used in this work. Two of these potentials are based on the original Nijm78 potential, whereas the third is a regularized update of the Reid68 potential in the sense that also in this new model each partial wave is parametrized by a number of Yukawa functions. In Sec. \[results\] we discuss the fitting procedure and the potentials are presented in more detail. GENERAL OUTLINE {#outline} =============== The $N\!N$ potential can be described in momentum space and in configuration space. Since it is difficult to solve the full four-dimensional scattering equation, it has become common practice first to make a reduction to a three-dimensional scattering equation. Various choices are possible, and it is important to note that the potential derived within the chosen reduction scheme should [*only*]{} be used in the scattering equation corresponding to that particular reduction scheme. These three-dimensional scattering equations can always be written in the form of the momentum-space version of the Lippmann-Schwinger equation. If the kinematics is treated relativistically, this is called the relativistic Lippmann-Schwinger equation. In configuration space, the differential form of this integral equation is the Schrödinger equation. The configuration-space potentials are to be used either in the nonrelativistic or the relativistic Schrödinger equation $$(\Delta+k^{2})\psi = 2M_{r}V\,\psi \ , \label{Schroed}$$ where $\Delta$ is the Laplacian, and where (non)relativistic refers to the kinematics. For nonrelativistic kinematics the relation between the center-of-mass energy $E$ and the center-of-mass momentum squared $k^{2}$ reads $E=k^{2}/2M_{r}$, whereas for relativistic kinematics it reads $E=\sqrt{k^{2}+M^{2}_{1}}+\sqrt{k^{2}+M^{2}_{2}}-M_{1}-M_{2}$. The earliest potential models were configuration-space potentials to be used in the nonrelativistic Schrödinger equation. They were phenomenological or semiphenomenological parameterizations, based on a general form for the potential. The potential must be invariant under rotations, reflections, and time reversal, and can be written [@Oku58] as the sum of 6 independent terms, $V=\sum_{i=1}^{6}V_{i}P_{i}$. A common choice for the 6 operators $P_{i}$ in configuration space is $$\begin{array}{l} P_{1}=1, \\[0.2cm] P_{2}=\mbox{\boldmath $\sigma$}_{1}\!\cdot\! \mbox{\boldmath $\sigma$}_{2}, \\[0.2cm] P_{3}=S_{12}=3(\mbox{\boldmath $\sigma$}_{1}\!\cdot\!\hat{\bf r}) (\mbox{\boldmath $\sigma$}_{2}\!\cdot\!\hat{\bf r}) -(\mbox{\boldmath $\sigma$}_{1}\!\cdot\! \mbox{\boldmath $\sigma$}_{2}), \\[0.2cm] P_{4}={\bf L\cdot S}, \\[0.2cm] P_{5}=Q_{12}={\textstyle\frac{1}{2}} [(\mbox{\boldmath $\sigma$}_{1}\!\cdot\!{\bf L}) (\mbox{\boldmath $\sigma$}_{2}\!\cdot\!{\bf L})+ (\mbox{\boldmath $\sigma$}_{2}\!\cdot\!{\bf L}) (\mbox{\boldmath $\sigma$}_{1}\!\cdot\!{\bf L})], \\[0.2cm] P_{6}={\textstyle\frac{1}{2}} (\mbox{\boldmath $\sigma$}_{1}-\mbox{\boldmath $\sigma$}_{2}) \cdot{\bf L} . \end{array} \label{Pconfig}$$ These operators are also frequently referred to as the central, spin-spin, tensor, spin-orbit, quadratic spin-orbit, and antisymmetric spin-orbit operators, respectively. For identical-particle scattering, the antisymmetric spin-orbit operator $P_{6}$ cannot contribute, whereas $V_{6}$ vanishes when charge independence is assumed (which is usually the case for $N\!N$ potential models). In general [@Oku58], each potential form $V_{i}$ in configuration space is a function of $r^{2}$, and of the operators $p^{2}$ and $L^{2}$. In most approaches one only keeps the dependence on $r^{2}$, while the $p^{2}$ dependence (when included) is often only present in a linear way in the central potential $V_{1}$. The inclusion of the $Q_{12}$ operator was found to be necessary, because otherwise it was impossible to describe simultaneously the $^{1}S_{0}$ and $^{1}D_{2}$ phase shifts using the same static potential. The presence of the operator $Q_{12}$ in the potential can to a certain extent be simulated by introducing non-local potentials [@Bry69]. In the expansion $\sum_{i=1}^{6}V_{i}P_{i}$, the potential forms $V_{i}$ are generally assumed to be the same in all partial waves. The potential differences between the partial waves are dictated by the differences in the expectation values of the operators $P_{i}$ in these partial waves. The Reid68 potential [@Rei68], however, is based on a quite different approach. Rather than having 6 potential forms $V_{i}$ which are the same for all partial waves, now each partial wave is parametrized separately. The potential forms $V_{i}$ therefore not only depend on $r^{2}$ and $L^{2}$, but also on $S^{2}$ and $J^{2}$. In this paper we present some potentials based on this approach that each partial wave is parametrized independently. We refer to these models as Reid-like models. With the discovery of the heavy mesons in the 1960s, it became common practice to write the potential as a sum over one-boson-exchange (OBE) potentials. The expressions for these OBE potentials are usually derived in momentum space. Introducing momentum vectors $${\bf k}={\bf p}_{f}-{\bf p}_{i}, \ \ {\bf q}={\textstyle\frac{1}{2}}({\bf p}_{f}+{\bf p}_{i}), \ \ {\bf n}={\bf q}\times{\bf k}\ ,$$ in terms of initial (${\bf p}_{i}$) and final (${\bf p}_{f}$) momenta, the equivalent in momentum space to Eq. (\[Pconfig\]) reads $$\begin{array}{l} P_{1}=1 \\[0.2cm] P_{2}=\mbox{\boldmath $\sigma$}_{1}\!\cdot\! \mbox{\boldmath $\sigma$}_{2}, \\[0.2cm] P_{3}=(\mbox{\boldmath $\sigma$}_{1}\!\cdot\!{\bf k}) (\mbox{\boldmath $\sigma$}_{2}\!\cdot\!{\bf k}) -{\textstyle\frac{1}{3}}{\bf k}^{2} (\mbox{\boldmath $\sigma$}_{1}\!\cdot\! \mbox{\boldmath $\sigma$}_{2}), \\[0.2cm] P_{4}={\textstyle\frac{i}{2}} (\mbox{\boldmath $\sigma$}_{1}+\mbox{\boldmath $\sigma$}_{2}) \cdot{\bf n}, \\[0.2cm] P_{5}=(\mbox{\boldmath $\sigma$}_{1}\!\cdot\!{\bf n}) (\mbox{\boldmath $\sigma$}_{2}\!\cdot\!{\bf n}), \\[0.2cm] P_{6}={\textstyle\frac{i}{2}} (\mbox{\boldmath $\sigma$}_{1}-\mbox{\boldmath $\sigma$}_{2}) \cdot{\bf n} . \end{array} \label{Pmom}$$ The potential forms $V_{i}$ in momentum space are functions of ${\bf k}$, ${\bf q}$, ${\bf n}$, and the energy. Although Eq. (\[Pmom\]) provides an adequate set of 6 linearly independent operators, the $Q_{12}$ operator in configuration space is [*not*]{} the exact Fourier transform of the $(\mbox{\boldmath $\sigma$}_{1}\!\cdot\!{\bf n}) (\mbox{\boldmath $\sigma$}_{2}\!\cdot\!{\bf n})$ operator in momentum space. This is of importance if we want both the momentum-space and the configuration-space versions to produce exactly the same phase shifts and bound states, which is only possible when the configuration-space version is the exact Fourier transform of the momentum-space version, and vice versa. This implies [@Rij91] that we have to use the inverse Fourier transform of the $Q_{12}$ operator; i.e., the potential contribution $(\mbox{\boldmath $\sigma$}_{1}\!\cdot\!{\bf n}) (\mbox{\boldmath $\sigma$}_{2}\!\cdot\!{\bf n})\,V_{5}({\bf k}^{2})$ is to be replaced by $$P_{5}V_{5}({\bf k}^{2}) - P'_{5}\int_{\infty}^{{\bf k}^{2}} \!d{\bf k}'^{2}V_{5}({\bf k}'^{2}) \ , \label{Q12replace}$$ where $$\begin{aligned} P'_{5} &=& [(\mbox{\boldmath $\sigma$}_{1}\!\cdot\!{\bf q}) (\mbox{\boldmath $\sigma$}_{2}\!\cdot\!{\bf q}) -{\bf q}^{2}(\mbox{\boldmath $\sigma$}_{1}\!\cdot\! \mbox{\boldmath $\sigma$}_{2})] \nonumber\\ & & -{\textstyle\frac{1}{4}} [(\mbox{\boldmath $\sigma$}_{1}\!\cdot\!{\bf k}) (\mbox{\boldmath $\sigma$}_{2}\!\cdot\!{\bf k}) -{\bf k}^{2}(\mbox{\boldmath $\sigma$}_{1}\!\cdot\! \mbox{\boldmath $\sigma$}_{2})] \ .\end{aligned}$$ Other restrictions imposed on the momentum-space potential forms $V_{i}$ in that case are that they should not depend on the energy, while the ${\bf q}$ dependence should be of second order at most (see also below). When the potentials are evaluated in momentum space and then Fourier transformed to configuration space, they are usually first regularized to remove the singularities at the origin. This can be achieved by introducing a form factor $F({\bf k}^{2})$. A typical Fourier transform, encountered in transforming the momentum-space potential to configuration space, then reads $$\begin{aligned} \int\frac{d^{3}k}{(2\pi)^{3}}\,\frac{e^{i{\bf k}\cdot{\bf r}}} {{\bf k}^{2}+m^{2}} \left({\bf k}^{2}\right)^{n}F({\bf k}^{2}) &\equiv&\frac{m}{4\pi}(-m^{2})^{n}\phi^{n}_{C}(r) \nonumber\\ &=&\frac{m}{4\pi}(-\mbox{\boldmath $\nabla$}^{2})^{n}\phi^{0}_{C}(r)\ . \label{phiC}\end{aligned}$$ The results for various frequently used choices are: (i) No form factor at all, $F({\bf k}^{2})=1$. This yields the familiar Yukawa function $$\phi^{0}_{C}(r)=e^{-mr}/mr \ , \label{Yukawa}$$ and the singularities at the origin are still present; (ii) Monopole form factor, $F({\bf k}^{2})=(\Lambda^{2}-m^{2})/ (\Lambda^{2}+{\bf k}^{2})$, normalized such that at the pole $F(-m^{2})=1$. This yields $$\phi^{0}_{C}(r)=\left[e^{-mr}-e^{-\Lambda r}\right]/mr \ ;$$ (iii) Dipole form factor, $F({\bf k}^{2})=(\Lambda^{2}-m^{2})^{2}/ (\Lambda^{2}+{\bf k}^{2})^{2}$, yielding $$\phi^{0}_{C}(r)=\left[e^{-mr}-e^{-\Lambda r} \left(1+\frac{\Lambda^{2}-m^{2}}{2\Lambda^{2}}\Lambda r\right) \right] \mbox{\LARGE /}mr \ ; \label{phidipole}$$ (iv) Exponential form factor, $F({\bf k}^{2})= e^{-{\bf k}^{2}/\Lambda^{2}}$, yielding $$\begin{aligned} \phi^{0}_{C}(r)=e^{m^{2}/\Lambda^{2}} \biggl[&&e^{-mr}{\rm erfc} \left(\frac{m}{\Lambda}-\frac{\Lambda r}{2}\right) \nonumber\\ && - e^{mr}{\rm erfc} \left(\frac{m}{\Lambda}+\frac{\Lambda r}{2}\right)\biggr] \mbox{\LARGE /}2mr \ ,\end{aligned}$$ where erfc$(x)$ is the complementary error function $${\rm erfc}(x)=\frac{2}{\sqrt{\pi}}\int_{x}^{\infty}\! dt\, e^{-t^{2}} \ .$$ We follow the normalization of Ref. [@Nag78]. This means that for the exponential form factor $F(0)=1$. Because ${\bf k}^{2}$ can be written as $({\bf k}^{2}+m^{2})-m^{2}$, we find that in the absence of a form factor $$\phi^{1}_{C}(r)=\phi^{0}_{C}(r)-4\pi\delta^{3}(m{\bf r}) \ , \label{phi1C}$$ When there is a form factor, this relation still holds, but the $\delta$-function contribution is smeared out. Using our definition (\[phiC\]), the Fourier transforms for the tensor and spin-orbit potentials can be simply expressed in terms of derivatives of the central function, i.e., $$\begin{aligned} \phi^{0}_{T}(r) &=& \frac{1}{3m^{2}}\,r\frac{d}{dr}\left(\frac{1}{r} \frac{d}{dr}\right)\, \phi^{0}_{C}(r) \ , \nonumber\\ \phi^{0}_{SO}(r)&=&-\frac{1}{m^{2}} \frac{1}{r}\frac{d}{dr}\, \phi^{0}_{C}(r) \ . \label{phiTSO}\end{aligned}$$ In order to ensure regularity at the origin for the tensor and spin-orbit functions, one must choose at least the dipole or exponential form factor. In that case, the tensor function also vanishes at the origin, as it should. The presence of explicit momentum-dependent terms in the momentum-space potential gives rise to non-local structures in the potential in configuration space. The ${\bf q}^{2}$ terms pose no difficulties for the configuration-space potential as long as they are linear in ${\bf q}^{2}$. The typical Fourier transform of such a term is given by $$\begin{aligned} \int\frac{d^{3}k}{(2\pi)^{3}}\,\frac{e^{i{\bf k}\cdot{\bf r}}} {{\bf k}^{2}+m^{2}} && \left({\bf q}^{2}+{\textstyle\frac{1}{4}} {\bf k}^{2}\right) F({\bf k}^{2}) \nonumber\\ =&& -\frac{m}{4\pi} \left[\Delta\frac{\varphi(r)}{2M_{r}}+ \frac{\varphi(r)}{2M_{r}}\Delta\right] \ , \label{nonlocal}\end{aligned}$$ where $\varphi(r)=M_{r}\phi^{0}_{C}(r)$. It is well known how to handle such a $(\Delta\varphi+\varphi\Delta)$ term [@Gre62]. The absence of ${\bf q}^{2}$ terms in the momentum-space potential will result in a radially local configuration-space potential. The three new potential models (Nijm93, Nijm I, and Nijm II) presented in this paper are based on the original Nijm78 potential with the exponential form factor, whereas the update of the Reid68 potential (Reid93) is regularized using a dipole form factor. STRUCTURE OF THE POTENTIALS {#structure} =========================== One-pion-exchange potential {#subsec:OPE} --------------------------- An important feature of the potential models presented in this paper is that in the one-pion-exchange (OPE) part of the potential, we explicitly distinguish between neutral-pion and charged-pion exchange. The pion masses are [@PDG90] $m_{\pi^{0}}=134.9739$ MeV and $m_{\pi^{\pm}}=139.5675$ MeV. Almost all other potentials that have appeared in the literature use a mean pion mass. In these other models the isovector $np$ phase parameters are [*larger*]{} in magnitude than the corresponding $pp$ phase parameters. By explicitly including the pion-mass differences exactly the opposite occurs: the isovector $np$ phase parameters are [*smaller*]{} than the corresponding $pp$ phase parameters. This is a unique feature of the potentials presented here. Defining $$V(m)\! = \!\left(\frac{m}{m_{\pi^{\pm}}}\right)^{2} \! m \! \left[\phi^{0}_{T}(m,r)S_{12}+{\textstyle\frac{1}{3}} \phi^{1}_{C}(m,r)(\mbox{\boldmath $\sigma$}_{1}\! \cdot\!\mbox{\boldmath $\sigma$}_{2})\right] , \label{VOPE}$$ the OPE potential for $pp$ scattering is given by $$V_{\rm OPE}(pp)=f_{\pi}^{2} V(m_{\pi^{0}}) \ , \label{Vpp}$$ whereas for $np$ scattering it reads $$V_{\rm OPE}(np)=-f_{\pi}^{2} V(m_{\pi^{0}}) \pm 2f_{\pi}^{2} V(m_{\pi^{\pm}}) \ , \label{Vnp}$$ where the plus (minus) sign corresponds to total isospin $I=1\, (0)$. The scaling mass $m_{\pi^{\pm}}$ in $V(m)$ is introduced in order to make the pseudovector coupling constant $f_{\pi}$ dimensionless. It is conventionally chosen to be equal to the charged-pion mass. The explicit distinction between neutral and charged pions implies that the isovector $np$ and $pp$ OPE potentials are different, and so charge independence is broken. In our present models we assume, however, that the pion-nucleon coupling constants obey charge independence. Nijmegen potential {#subsec:Nijm} ------------------ In this section we briefly discuss the structure of the Nijmegen potential. More details can be found in Refs. [@Nag78; @Mae89]. The basic functions are the one-boson-exchange (OBE) potential functions with momentum-dependent central terms and exponential form factors. The meson exchanges we include are those due to pseudoscalar mesons ($\pi, \eta, \eta'$), vector mesons ($\rho, \omega, \phi$), and scalar mesons ($a_{0}, f_{0}, \epsilon$). Here we use the modern nomenclature for the scalar mesons, i.e., $a_{0}(983)$ corresponds to the $\delta$ of Ref. [@Nag78], and $f_{0}(975)$ to the $S^{\star}$. The $\epsilon$ meson would correspond to an $f_{0}(760)$. No such meson is listed by the Particle Data Group [@PDG90]; however, a recent analysis of the $\pi N\rightarrow\pi^{+}\pi^{-}N$ reaction [@Sve92] provides evidence for a scalar–isoscalar resonant state $0^{++}(750)$. In the Nijmegen potentials the $\epsilon$ meson corresponds to a broad meson (see below) where the pole in its propagator is chosen to correspond to the pole position in the complex energy plane of the isoscalar $\pi\pi$ $S$ wave [@Pro73]. Here we will retain the name of $\epsilon$ meson. The aforementioned meson exchanges can be identified with the dominant parts of the lowest-lying meson trajectories in the complex $J$ plane. We furthermore include the dominant $J=0$ parts of the Pomeron, and of the $f_{2}, f'_{2}$, and $a_{2}$ tensor-meson trajectories. They give rise to Gaussian potentials. The meson propagators including the exponential form factor read $$\Delta({\bf k}^{2},m^{2},\Lambda^{2})=\frac{1}{{\bf k}^{2}+m^{2}} \ e^{-{\bf k}^{2}/\Lambda^{2}} \ .$$ For the Pomeron-type exchanges we have $$\Delta({\bf k}^{2},m^{2}_{p})=\frac{1}{M_{p}^{2}} \ e^{-{\bf k}^{2}/4m^{2}_{p}} \ ,$$ where $m_{p}$ has the dimension of a mass and will be called the Pomeron mass, and $M_{p}$ is a scaling mass, chosen to be the proton mass. The different potential forms are evaluated in momentum space and the resulting expressions are essentially those of Refs. [@Nag78] (save some misprints [@misprint]) with the following differences: (i) We explicitly account for the proton and neutron mass difference; (ii) the differences between the neutral and charged pion (see Sec. \[subsec:OPE\]), and between the neutral and charged $\rho$ meson are explicitly included; (iii) we have adjusted the quadratic spin-orbit operator of the potential in momentum space to include the $P'_{5}$ contribution as in Eq. (\[Q12replace\]). The effect of the first modification is obviously rather small. The second modification (as well as the first) implies that charge independence is broken in the non-OPE part of the potential as well. For $pp$ scattering the potential consists of only neutral-meson exchange, $V_{pp}=V({\rm neutral})$, whereas for $np$ scattering it consists of neutral-meson and charged-meson exchange, depending on the total isospin as in Eq. (\[Vnp\]), so $V_{np}=-V({\rm neutral})\pm2V({\rm charged})$. This distinction replaces the factor $(\mbox{\boldmath $\tau$}_{1} \!\cdot\!\mbox{\boldmath $\tau$}_{2})$ used in the old Nijm78 potential. Finally, the third modification means that we have constructed a potential which is exactly equivalent in both momentum space and configuration space, a unique feature of these Nijmegen potentials. For example, for the parametrized Paris potential [@Lac80] this is not true since it uses the same parameters in combination with the $Q_{12}$ operator in configuration space as it does with the $(\mbox{\boldmath $\sigma$}_{1}\!\cdot\!{\bf n}) (\mbox{\boldmath $\sigma$}_{2}\!\cdot\!{\bf n})$ operator in momentum space. Next we briefly discuss the coupling constants. For definitions and references we again refer to Refs. [@Nag78; @Mae89]. The coupling constants of the pseudoscalar mesons are related via SU(3) and singlet-octet mixing. The octet coupling $f_{\eta_{8}}$ is calculated using $\alpha_{P}=0.355$ for the $F/(F+D)$ ratio. For the singlet-octet mixing angle we use $\theta_{P}=-23^{\circ}$ to define the physical coupling constants $f_{\eta}$ and $f_{\eta'}$. This leaves the singlet coupling $f_{\eta_{1}}$ and the pion coupling $f_{\pi}$ as free input parameters. However, in our partial-wave analysis of the $pp$ scattering data [@Ber90], we found for the $pp\pi^{0}$ coupling constant $f_{p}^{2}=0.0749(7)$. This value was later confirmed in a combined partial-wave analysis of all $pp$ and $np$ scattering data, assuming charge independence for the pion-nucleon coupling constants [@Klo91; @St93c]. There the value $f_{\pi}^{2}=0.075$ is recommended for the pion-nucleon coupling constant at the pion pole. This is the value we adopt in our construction of the new Nijmegen potentials, and so it is not included as a free parameter. For the vector mesons we assume that the $\rho$ meson is universally coupled to the isospin current ($\alpha^{e}_{V}=1$) to define the octet coupling $g_{V_{8}}$. For the singlet-octet mixing angle we take $\theta_{V}=37.5^{\circ}$, which fixes the physical coupling constants $g_{\omega}$ and $g_{\phi}$ in terms of $g_{\rho}$ and $g_{V_{1}}$. The $\phi$ meson is assumed to have $f_{\phi}\equiv0$. The free parameters are now $g_{\rho}$, $g_{V_{1}}$, $f_{\rho}$, and $f_{\omega}$. For the scalar mesons we do not apply any constraints for the coupling constants, since the singlet-octet mixing angle for the scalar mesons is still an unsettled problem (see also Ref. [@Mae89]). The free parameters are the $a_{0}$, $f_{0}$, and $\epsilon$ coupling constants. For simplicity we take a single mass parameter $m_{p}$ for the Pomeron, and for the $J=0$ parts of the $f_{2}$, $f'_{2}$, and $a_{2}$ tensor-meson trajectories. We use two coupling constants: $g_{a_{2}}$ for the isovector $a_{2}$ meson and $g_{p}$ for the isoscalar and Pomeron exchanges. For each type of exchange we use an independent cutoff mass, so we have three cutoff parameters $\Lambda_{P}$, $\Lambda_{V}$, and $\Lambda_{S}$. This brings us to a total of 14 free parameters. We conclude this section with a discussion of the treatment of the broad $\rho$ and $\epsilon$ mesons. The width of a broad meson can be accounted for [@Sch71; @Bin71; @Swa78] by replacing the propagator $\Delta({\bf k}^{2})=1/({\bf k}^{2}+m^{2})$ of a stable meson by a dispersion integral $$\Delta({\bf k}^{2})=\int_{m_{t}^{2}}^{\infty} \frac{\rho(m^{\prime2})dm^{\prime2}}{{\bf k}^{2}+m^{\prime2}} \ ,$$ with mass distribution $$\rho(m^{\prime2})=\frac{1}{\pi}\frac{\gamma(m^{\prime2}- m_{t}^{2})^{n+1/2}} {(m^{\prime2}-m^{2})^{2}+\gamma^{2} \left(\frac{m^{\prime2}}{m^{2}}\right)^{2n} (m^{\prime2}-m_{t}^{2})^{2n+1}} \ ,$$ and where $$\gamma=m\Gamma\,(m^{2}-m_{t}^{2})^{-(n+1/2)} \ .$$ Here $\Gamma$ denotes the width and $n=0,1$ for spin-0 and spin-1 mesons, respectively [@Sch71]. The charged $\rho$ meson decays into a neutral and a charged pion and the threshold mass is $m_{t}=m_{\pi^{0}}+m_{\pi^{\pm}}$. The neutral $\rho$ meson cannot decay into two neutral pions and it decays into two charged pions, and so now the threshold mass is $m_{t}=2m_{\pi^{\pm}}$. The $\epsilon$ meson is an isoscalar meson which decays into both two neutral or two charged pions in the ratio 1:2. In our present models these distinctions have been explicitly accounted for, which is another extension of the old Nijm78 model. The configuration-space potential due to the exchange of a broad meson is calculated exactly. This exact potential is then approximated by the sum of two potentials of stable mesons [@Bin71] $$\begin{aligned} \int_{m_{t}}^{\infty}dm'2m'&&\rho(m^{\prime2}) m^{\prime} \phi^{0}_{C}(m',r) \nonumber\\ \approx && \beta_{1}m_{1}\phi^{0}_{C}(m_{1},r) + \beta_{2}m_{2}\phi^{0}_{C}(m_{2},r) \ . \label{Yukfit}\end{aligned}$$ Fitting from 0–2 fm yields the values as given in Table \[broad\]. Regularized Reid potential {#subsec:Reid} -------------------------- A disadvantage of the original Reid68 potential is that, at the time of its construction, the quality of the $np$ data was very poor. As a consequence, the Reid68 potential can no longer properly describe the numerous new and much more accurate $np$ data. Another disadvantage is that the Reid68 potential has an $r^{-1}$ singularity in all partial waves. Here we present an updated version of the Reid potential, where these singularities have been removed via the inclusion of a dipole form factor. With this choice, the tensor potential now also vanishes at the origin, as it should. As is the case for the original Reid68 potential, the OPE potential is explicitly included, while we now account for the neutral-pion and charged-pion mass differences as in Eqs. (\[VOPE\])–(\[Vnp\]). For the pion-nucleon coupling constant at the pion pole we take [@Klo91; @St93c] $f_{\pi}^{2}=0.075$, and for the dipole cutoff parameter we choose $\Lambda=8m_{\pi}$. In the OPE potential (\[VOPE\]) we use $\phi^{1}_{C}$ only for the $S$ waves. For all other partial waves, we found it more convenient to use $\phi_{C}^{0}$ instead of $\phi_{C}^{1}$. Note that $\phi_{C}^{1}$ equals $\phi_{C}^{0}$ up to a modified $\delta$ function \[see Eq. (\[phi1C\])\], and that this modified $\delta$ function is screened by the centrifugal barrier for all these other partial waves, except the $S$ waves. Starting with this OPE potential, the potential in each partial wave can now be extended by choosing a convenient combination of central, tensor, and spin-orbit functions with arbitrary masses and cutoff parameters. In the construction presented in the following we (more or less arbitrarily) settled for integer multiples of a mean pion mass $\overline{m}=(m_{\pi^{0}}+2m_{\pi^{\pm}})/3$, while the cutoff mass in the dipole form factor is chosen to be $\Lambda=8\overline{m}$ everywhere. For notational reasons we next define $$\begin{array}{l} Y(p)=p\overline{m}\,\phi^{0}_{C}(p\overline{m},r) , \\[0.2cm] Z(p)=p\overline{m}\,\phi^{0}_{T}(p\overline{m},r) , \\[0.2cm] W(p)=p\overline{m}\,\phi^{0}_{SO}(p\overline{m},r) , \end{array}$$ with $p$ an integer and $\phi^{0}_{X}$ given by Eqs. (\[phidipole\]) and (\[phiTSO\]). For the coefficients multiplying these functions, we use $A_{ip}$ for the isovector potentials, whereas the coefficients $B_{ip}$ are for the isoscalar and $np$ $^{1}S_{0}$ potentials. The index $i$ subsequently labels the different partial waves. For the total potential in a particular partial wave one should, of course, add the appropriate OPE potential as given by Eqs. (\[VOPE\])–(\[Vnp\]). For the non-OPE parts in the isovector singlet partial waves ($I=1$, $S=0$, $L=J$) we use $$\begin{aligned} V_{pp}(^{1}S_{0}) &=& A_{12}Y(2)+A_{13}Y(3)+A_{14}Y(4) \nonumber\\ & &+A_{15}Y(5)+A_{16}Y(6) \ , \nonumber\\ V_{np}(^{1}S_{0}) &=& B_{13}Y(3)+B_{14}Y(4)+B_{15}Y(5) +B_{16}Y(6) \ , \nonumber\\ V(^{1}D_{2}) &=& A_{24}Y(4)+A_{25}Y(5)+A_{26}Y(6) \ , \label{sin1}\\ V(^{1}G_{4}) &=& A_{33}Y(3) \ , \nonumber\\ V(^{1}J_{J}) &=& V_{pp}(^{1}S_{0})\ \ \ {\rm for}\ J\geq6 \ , \nonumber\end{aligned}$$ where the distinction between the $pp$ and $np$ ${^{1}S}_{0}$ potentials is necessary because of the well-known breaking of charge independence in the $pp$ and $np$ ${^{1}S}_{0}$ partial waves. The coefficients $A_{ip}$ and $B_{ip}$ are to be fitted. The presence of the two-pion range piece $A_{02}Y(2)$ in the $pp$ ${^{1}S}_{0}$ potential is purely coincidental, and was only included to improve the quality of the fit. A similar term in the $np$ ${^{1}S}_{0}$ was much less effective, and so we decided to leave it out. For the non-OPE parts in the isoscalar singlet partial waves ($I=0$, $S=0$, $L=J$) we use $$\begin{aligned} V(^{1}P_{1}) &=& B_{23}Y(3)+B_{24}Y(4)+B_{25}Y(5)+B_{26}Y(6) \ , \nonumber\\ V(^{1}F_{3}) &=& B_{33}Y(3)+B_{35}Y(5) \ , \label{sin0}\\ V(^{1}J_{J}) &=& V(^{1}P_{1})\ \ \ {\rm for}\ J\geq5 \ . \nonumber\end{aligned}$$ For the isovector triplet uncoupled partial waves ($I=1$, $S=1$, $L=J$) we use $$\begin{aligned} V(^{3}P_{0}) &=& A_{43}Y(3)+A_{45}Y(5)+A_{4z3}Z(3) \ , \nonumber\\ V(^{3}P_{1}) &=& A_{53}Y(3)+A_{55}Y(5)+A_{5z3}Z(3) \ , \label{trip1}\\ V(^{3}F_{3}) &=& A_{63}Y(3) \ , \nonumber\end{aligned}$$ and the isoscalar triplet uncoupled partial waves ($I=0$, $S=1$, $L=J$) are parametrized as $$\begin{aligned} V(^{3}D_{2}) &=& B_{43}Y(3)+B_{45}Y(5)+B_{4z3}Z(3) \ , \nonumber\\ V(^{3}G_{4}) &=& B_{53}Y(3) \ . \label{trip0}\end{aligned}$$ Following the parametrization of the original Reid68 potential, the non-OPE potential in the triplet coupled partial waves ($S=1$, $L=J\pm1$) is parametrized as $$V=V_{C}+V_{T}S_{12}+V_{SO}{\bf L\cdot S} \ , \label{copgen}$$ where the isovector ($I=1$) potentials are given by $$\begin{aligned} V_{C} &=& A_{73}Y(3)+A_{74}Y(4)+A_{75}Y(5)+A_{76}Y(6) \ , \nonumber\\ V_{T} &=& A_{7z4}Z(4)+A_{7z6}Z(6) \ , \nonumber\\ V_{SO}&=& A_{7w3}W(3)+A_{7w5}W(5)\ \ \ {\rm for}\ J=2 \ , \label{cop1}\\ V_{SO}&=& A_{8w3}W(3) \ \ \ {\rm for}\ J=4 \ , \nonumber\end{aligned}$$ and the isoscalar ($I=0$) potentials read $$\begin{aligned} V_{C} &= & B_{62}Y(2)+B_{63}Y(3)+B_{64}Y(4) \nonumber\\ & &+B_{65}Y(5)+B_{66}Y(6) \ , \nonumber\\ V_{T} &= & B_{6z4}Z(4)+B_{6z6}Z(6) \ , \label{cop0}\\ V_{SO}&= & B_{6w3}W(3)+B_{6w5}W(5)\ \ \ {\rm for}\ J=1 \ , \nonumber\\ V_{SO}&= & B_{7w3}W(3)+B_{7w5}W(5)\ \ \ {\rm for}\ J=3 \ . \nonumber\end{aligned}$$ Finally, for the triplet isovector partial waves ($I=1$, $S=1$) with $J\geq5$ we use Eq. (\[copgen\]) with the central and tensor potentials of Eq. (\[cop1\]), and the spin-orbit potential equal to zero. Similarly, for the triplet isoscalar partial waves ($I=0$, $S=1$) with $J\geq5$ we use the central and tensor potentials of Eq. (\[cop0\]). This choice is analogous to the extension of the Reid68 potential to the higher partial waves as given by Day [@Day81]. RESULTS ======= The parameters of the potential models are optimized by minimizing the $\chi^{2}$ in a direct fit to the data. Since the scattering data are spread over a large number of energies (about 200 different energies for the $pp$ data and almost 400 different energies for the $np$ data in the 0–350 MeV energy range) and the phase parameters need to be calculated up to at least $J\approx6$, the Schrödinger equation then has to be solved a very large number of times. This approach, therefore, is not very practical when the model parameters do not yet have reasonable values and, consequently, the $\chi^{2}$ is still very high. A more convenient approach is to start with the Nijmegen representation [@St93b] of the $\chi^{2}$ hypersurface of the scattering data. It is obtained from the 10 single-energy analyses and consists of 10 sets of phase parameters and the error matrix, each at a different energy. The error matrix is the inverse of half the second-derivative matrix of the $\chi^{2}$ hypersurface with respect to the phase parameters up to $J=4$ within the energy bin of the single-energy analysis. This $\chi^{2}$ hypersurface is, in principle, independent of the particular partial-wave analysis. In practice the representation we use is somewhat dependent on the Nijmegen multienergy analysis. The crucial point is, however, that it provides a very good and concise representation of the scattering data. For each change in the model parameters we need to solve the Schrödinger equation for all partial waves up to $J=4$ at only 10 different energies, which allows for a much quicker optimization for the parameters of the potential model. In the last stage of the fitting procedure the potential parameters have been further optimized in a very time-consuming direct fit to the data. In this case we use the potential model in all partial waves with $J\leq6$, whereas in the higher partial waves we include OPE only. The final $\chi_{\rm min}^{2}$ of the potential should be obtained from this direct comparison with the experimental data. Nijm92pp -------- Our first improvement of the Nijm78 potential [@Nag78] was already started several years ago, when we constructed an update to the $pp$ data of the Nijm78 potential. This potential has been used in the Nijmegen analyses [@Klo91; @Klo92; @St93c] to parametrize the $np$ isovector lower partial waves (except the $np$ $^{1}S_{0}$). In our latest analysis [@St93b] (PWA93) we refer to it as the Nijm92$pp$ potential. We found that a good fit to the $pp$ data could be obtained using one cutoff parameter $\Lambda=827.53$ MeV for all three types of meson exchanges. Some of the coupling constants were not refitted, but were kept at the values of the original Nijm78 potential. The reason is that, when we only fit to the $pp$ data, we cannot incorporate the isospin dependence of the isovector exchanges (there are only $I=1$ partial waves). A direct comparison of this Nijm92$pp$ potential with the $pp$ data yields $\chi^{2}=2487.1$ for 1787 data, which means $\chi^{2}/N_{pp}=1.4$. Nijm93 ------ Now that the Nijmegen analysis of the $np$ data is also finished [@St93b] and we have carefully scrutinized the $np$ database, we can extend the update of the Nijm78 potential to include the fit to the $np$ scattering data as well. This model we refer to as Nijm93. As already mentioned in Sec. \[subsec:Reid\], the $np$ $^{1}S_{0}$ partial wave has to be parametrized separately. The reason is that there is clear evidence for breaking of charge independence in the $^{1}S_{0}$ scattering lengths $a_{pp}$ and $a_{np}$. This difference in scattering lengths carries over into an approximately $2^{\circ}$ phase-shift difference between the $pp$ and $np$ $^{1}S_{0}$ phase shifts at higher energies. This difference cannot be explained as being only due to the difference between the $pp$ and $np$ OPE potentials. Allowing for a different value for the neutral-pion and charged-pion coupling constants does not help either, because the scattering lengths are very insensitive to these kind of changes. To accommodate the $pp$ and $np$ $^{1}S_{0}$ differences, we therefore introduce a purely phenomenological breaking of charge independence between the $\rho^{0}$ and $\rho^{\pm}$ coupling constants. This breaking of charge independence is [*only*]{} assumed in the $^{1}S_{0}$ partial wave; for all other partial waves the $\rho^{0}$ and $\rho^{\pm}$ coupling constants are taken to be the same. The parameters for the Nijm93 potential, rounded to four or five significant figures, can be found in Table \[parnym93\], where the meson masses are the masses as listed by the 1990 Particle Data Group [@PDG90]. The coupling constants are the values at ${\bf k}^{2}=0$. The pion coupling constants are fixed at $f^{2}=0.075$ at the corresponding pion pole ${\bf k}^{2}=-m^{2}_{\pi^{0}}$ or ${\bf k}^{2}=-m^{2}_{\pi^{\pm}}$; hence the different entries for neutral- and charged-pion exchange at ${\bf k}^{2}=0$. The $\epsilon$ and Pomeron coupling constants are rather large, whereas the $\omega$ coupling constant is reasonably small. For the $\rho$ coupling constants we find $(f/g)_{\rho}=4.094$, which is close to the value 3.7 from naive vector-meson dominance of the isovector electromagnetic form factors of the nucleon. For the Nijm93 potential we find $\chi^{2}(pp)=3175.6$ for 1787 $pp$ data and $\chi^{2}(np)=4848.4$ for 2514 $np$ data. So for the $pp$ data $\chi^{2}/N_{pp}=1.8$, for the $np$ data $\chi^{2}/N_{np}=1.9$, and for all $N\!N$ data $\chi^{2}/N_{\rm data}=1.87$. We find that this 15-parameter conventional meson-exchange potential cannot do better than $\chi^{2}/N_{\rm data}\approx1.87$, a result which is also found for similar potential models such as the Paris80 and all the Bonn potentials. Apparently, the conventional meson-exchange potentials cannot compete in quality with the Nijmegen PWA93. This indicates that these models lack some important physics. Nijm I and Nijm II ------------------ In order to be able to construct a potential model which is of almost the same quality as the Nijmegen PWA93 ($\chi^{2}/N_{\rm data}\approx1$), we follow a different approach and take advantage of the success of the Reid68 potential. We expect that, when we start with the Nijmegen potential (which already has a reasonable $\chi^{2}$ on the $pp$ data), we can construct a Reid-like potential where in each partial wave we probably need to adjust only a few parameters in order to arrive at $\chi^{2}/N_{\rm data}\approx1$. The potential forms are then given by a set of slightly adjusted Nijmegen potentials, each representing one particular partial wave. Starting with the parameters of the Nijm92$pp$ potential [@remark], we find that for most partial waves an adjustment of the $f_{\rho}$ and $g_{\epsilon}$ coupling constants already gives very good results. Counting all parameters which have been adjusted in the fit of each partial wave, we arrive at a total of 41 parameters. This should be compared with the 39 parameters used in the Nijmegen PWA93. In the last stage, the parameters of this Reid-like potential are optimized in a direct fit to the data. The potential is then used in all partial waves up to $J=6$ simultaneously. This model we refer to as Nijm I. It has $\chi^{2}(pp)=1795.8$ and $\chi^{2}(np)=2627.3$, and so $\chi^{2}/N_{\rm data}=1.03$ on all $pp$ and $np$ scattering data. We have also constructed a local Reid-like Nijmegen potential, where we leave out the explicit momentum-dependent terms which give rise to non-local contributions to the configuration-space potential as expressed in Eq. (\[nonlocal\]). We follow the same procedure as for the non-local Nijm I potential. First, the parameters are adjusted in a fit to the representation of the $\chi^{2}$ hypersurface, and then further optimized in a direct fit to the data. For this local potential, denoted by Nijm II, we use a total of 47 parameters and we find $\chi^{2}(pp)=1795.8$ and $\chi^{2}(np)=2625.7$, and so $\chi^{2}/N_{\rm data}=1.03$. Although these potentials are purely phenomenological (except for the correct OPE tail) and the coupling constants have no physical meaning, these potentials are the first to give an excellent description of the $N\!N$ scattering data. They have already been used successfully in three- and many-body calculations [@Fri93; @Zhe93]. Reid93 ------ Finally, we have constructed an updated Reid potential based on the original Reid68 potential. This regularized version, denoted by Reid93 and discussed in Sec. \[subsec:Reid\], gives an equally good description of the data as do the Nijm I and Nijm II potentials. The 50 phenomenological potential parameters $A_{ij}$ and $B_{ij}$ were fitted to the data, resulting in $\chi^{2}(pp)=1795.1$ and $\chi^{2}(np)=2624.6$, and so also for this potential $\chi^{2}/N_{\rm data}=1.03$. This Reid93 potential has been used in a triton calculation as well [@Fri93]. Comparison of the potentials ---------------------------- The results of the non-local Nijm I potential, the local Nijm II potential, and the local Reid93 potential are summarized in Table \[compare\]. The results of the Nijmegen PWA93 are shown for comparison. Although $\chi^{2}/N_{\rm data}$ for these three potentials is already very good, their description of the $np$ data is still not as good as the description of these $np$ data in the partial-wave analysis. Here we should also mention that we did not do a thorough investigation into the minimum number of parameters required to get these results (as we did for the partial-wave analysis). The reason is that in order to do this properly, all potential parameters have to be fitted simultaneously to all $N\!N$ data. But we found it more successful to do a large number of fits where in each separate run only a (completely arbitrary) subset of the potential parameters was optimized. Therefore, we cannot rule out the possibility that an equally good fit can be obtained with a few parameters less. As already mentioned in the Introduction, these new potentials (except Nijm93) are in a sense alternative partial-wave analyses. The differences between the phase parameters of the potentials and the phase parameters of the Nijmegen PWA93 are shown in Tables \[phspp\] and \[phsnp\]. These differences provide an indication for the systematic error on the results of the Nijmegen PWA93. For the $np$ phase parameters the differences at some energies (especially with the Reid93 potential) are relatively large. However, one has to bear in mind that $\chi^{2}(np)$ of the potential models is still substantially higher than that of the multienergy partial-wave analysis. On the other hand, the variation in the mixing parameter $\epsilon_{1}$ is small. It has often been claimed that this mixing parameter is ill determined and a wide range of values from potential models seemed acceptable (see, e.g., Ref. [@Mac87; @Mac89]). However, in our publication of the Nijmegen PWA93 we already argued that $\epsilon_{1}$ is in fact known very accurately. This is confirmed in Fig. \[epsilon\], where we note that the results of the Nijm I and Nijm II models lie essentially within the [*statistical*]{} uncertainty as obtained in the Nijmegen PWA93. Above 150 MeV, the result of the Reid93 model rises too strongly but is still within 2.5 standard deviations of the Nijmegen PWA93. The potentials between 0 and 2 fm for the singlet, triplet uncoupled, and triplet coupled $np$ partial waves are shown in Figs. \[potsin\], \[potunc\], and \[potcop\], respectively. For the non-local potential Nijm I we plot $V/(1+2\varphi)$, which more or less represents the effective potential when non-local terms are present (see Refs. [@Gre62; @Nag78]). For coupled channels, the potential is a $2\times2$ matrix and the $\varepsilon_{1}$ and $\varepsilon_{2}$ plots in Fig. \[potcop\] represent the off-diagonal elements of the potential. The main differences between the potentials show up in the inner region, i.e., for $r<1$ fm. In general, the non-local Nijm I potential is much softer than the local Nijm II and Reid93 potentials, while the Nijm II potential is again much softer than the Reid93 potential. The reason for this softness of the Nijmegen potentials is the exponential form factor. Finally, all potential models have been fitted to the deuteron binding energy $B=2.224\,575(9)$ MeV [@Leu82] using relativistic kinematics, i.e., $$B=M_{p}+M_{n}-\sqrt{M^{2}_{p}-\kappa^{2}}- \sqrt{M^{2}_{n}-\kappa^{2}} \ ,$$ rather than $B=\kappa^{2}/2M_{r}$. We also constructed versions of the Nijm I and Nijm II potentials to accommodate the latter nonrelativistic form. In any case the value $B=2.224\,575$ MeV is exactly reproduced to this accuracy. Some of the other deuteron parameters are listed in Table \[deuter\]. The different potential models all give very similar results. Because we consider the potentials Nijm I, Nijm II, and Reid93 as alternative partial wave analyses, the values of the deuteron parameter $\eta$, $A_{s}$, and $N^{2}$ given by these potentials are new experimental determinations of these quantities. For the $D/S$-state ratio $\eta$ we find $\eta=0.0252(1)$ in good agreement with the recent determination by Rodning and Knutson [@Rod90] of $\eta=0.0256(4)$. For the asymptotic $S$-state normalization $A_{S}$ we obtain $A_{s}=0.8843(10)$ fm$^{-1/2}$, which is in agreement with the determination by Kermode [*et al.*]{} [@Ker83] of $A_{S}=0.8883(44)$ fm$^{-1/2}$. This results in $N^{2}=A^{2}_{s}(1+\eta^{2})=0.7825(20)$ fm$^{-1}$. However, there have been other experimental determinations of these quantities which are not always in agreement with the values quoted above. For a more complete list of these experimental determinations and a discussion of the differences between them, we refer to Ref. [@Sto88], and references cited therein. A direct comparison of our value $Q_{d}=0.271(1)$ fm$^{2}$ of the deuteron quadrupole moment with the experimental value 0.2859(3) fm$^{2}$ of [@Bis79] is only possible after all possible corrections have been accounted for, which is outside the scope of this paper. However, we would like to turn the argument around and suggest that these corrections must obviously be about 0.015 fm$^{2}$. It is quite interesting to see that for our best potentials the $D$-state probability is $P_{d}=5.665(30) \%$. The deuteron parameters as well as the results for the scattering lengths for both the potential models and the Nijmegen PWA93 will be discussed in more detail elsewhere [@Ter93]. These potentials were used [@Fri93] in calculations of the triton binding energy. It turned out that all these potentials underbind the triton by roughly 800 keV, a result which can be expected from their $P_{d}$ values. CONCLUSIONS =========== In this paper we presented an update Nijm93 of the old Nijm78 $N\!N$ potential. It contains the correct OPE tail and has $\chi^{2}/N_{\rm data}=1.87$. Although it cannot compete in quality with the Nijmegen partial-wave analysis (a feature apparently all conventional meson-exchange potentials suffer from), the description of the $np$ data of the new Nijm93 model is substantially better than that of the original Nijm78 potential, which was fitted to the old 1969 Livermore database [@Mac69]. Here we would also like to point out that in the Nijm93 model we do not include two-meson-exchange contributions such as $\pi\pi$ and $\pi\rho$ exchange, and we still get a reasonable description of the $^{1}P_{1}$ and $^{3}D_{2}$ phase shifts. This in contrast to claims made in the literature [@Mac87; @Mac89] that this is impossible. However, this result is obtained at the cost of having a rather large value for the pseudoscalar (pion) cutoff mass of $\Lambda_{P}=1177.11$ MeV. We have also presented three new high-quality $N\!N$ potentials. The Nijm I potential is a non-local Reid-like potential where each of the lower partial waves up to $J=4$ is parametrized separately. For the higher partial waves we use the parameters of the Nijm92$pp$ potential, which is an update to the $pp$ data of the original Nijm78 potential. The Nijm II potential is a local Reid-like potential, and does not contain any explicit momentum-dependent terms. Both potentials fit the $N\!N$ scattering data with a nearly optimal $\chi^{2}_{\rm min}/N_{\rm data}=1.03$. A regularized update of the Reid potential, denoted by Reid93, gives the same excellent $\chi^{2}_{\rm min}/N_{\rm data}=1.03$. Computer codes for these Nijmegen potentials Nijm I, Nijm II, and Nijm93, and for the regularized Reid93 potential, in configuration space as well as in momentum space, can be readily obtained via anonymous FTP from thef-nym.sci.kun.nl. We would like to thank Dr. J.L. Friar and Dr. Th.A. Rijken for many helpful discussions. Part of this work was included in the research program of the Stichting voor Fundamenteel Onderzoek der Materie (FOM) with financial support from the Nederlandse Organisatie voor Wetenschappelijk Onderzoek (NWO). One of us (V.G.J.S.) would also like to thank the Australian Research Council for financial support. Permanent address. e-mail: thefalg[@]{}sci.kun.nl V.G.J. Stoks, R.A.M. Klomp, M.C.M. Rentmeester, and J.J. de Swart, Phys. Rev. C [**48**]{}, 792 (1993). V. Stoks and J.J. de Swart, Phys. Rev. C [**47**]{}, 761 (1993). R.V. Reid, Jr., Ann. Phys. (NY) [**50**]{}, 411 (1968). M.M. Nagels, T.A. Rijken, and J.J. de Swart, Phys. Rev. D [**17**]{}, 768 (1978). J. Haidenbauer and K. Holinde, Phys. Rev. C [**40**]{}, 2465 (1989). R. Machleidt, K. Holinde, and Ch. Elster, Phys. Rep. [**149**]{}, 1 (1987). M. Lacombe, B. Loiseau, J.M. Richard, R. Vinh Mau, J. Côté, P. Pirès, and R. de Tourreil, Phys. Rev. C [**21**]{}, 861 (1980). R. Machleidt, Adv. Nucl. Phys. [**19**]{}, 189 (1989). R.A.M. Klomp, V.G.J. Stoks, and J.J. de Swart, Phys. Rev. C [**44**]{}, R1258 (1991). R.A.M. Klomp, V.G.J. Stoks, and J.J. de Swart, Phys. Rev. C [**45**]{}, 2023 (1992). V. Stoks, R. Timmermans, and J.J. de Swart, Phys. Rev. C [**47**]{}, 512 (1993). At the time we first started to follow this approach, the analysis of the $np$ data was still in progress. This is the reason that the model we started with is the Nijm92$pp$ potential, rather than the (more obvious) Nijm93 potential. S. Okubo and R.E. Marshak, Ann. Phys. (NY) [**4**]{}, 166 (1958). R. Bryan and B.L. Scott, Phys. Rev. [**177**]{}, 1435 (1969). T.A. Rijken, R.A.M. Klomp, and J.J. de Swart, Nijmegen Report THEF-NYM-91.05 (unpublished). A.M. Green, Nucl. Phys. [**33**]{}, 218 (1962). M. Aguilar-Benitez [*et al.*]{}, (Particle Data Group), Phys. Lett. B [**239**]{}, 1 (1990). P.M.M. Maessen, Th.A. Rijken, and J.J. de Swart, Phys. Rev. C [**40**]{}, 2226 (1989). M. Svec, A. de Lesquen, and L. van Rossum, Phys. Rev. D [**46**]{}, 949 (1992). S.D. Protopopescu, M. Alston-Garnjost, A. Barbaro-Galtieri, S.M. Flatté, J.H. Friedman, T.A. Lasinski, G.R. Lynch, M.S. Rabin, and F.T. Solmitz, Phys. Rev. D [**7**]{}, 1279 (1973). There are several misprints in Ref. [@Nag78]: in Eq. (6) the second expression should read ${\cal V}_{2}^{(V)}=-{\textstyle\frac{2}{3}}{\bf k}^{2} {\cal V}_{3}^{(V)}$; the $m^{2}$ in front of $\phi^{1}_{C}(r)$ in Eq. (15) should be dropped; the $\nabla^{2}$ operator in Eq. (16) should operate on $S_{12}$ as well, as $[(-\nabla^{2})^{n}\phi^{0}_{T}(r)S_{12}]$; similarly, Eq. (19) should have $[(-\nabla^{2})^{n}\phi^{0}_{SO}(r){\bf L\cdot S}]$; and the coefficient in front of the $Q_{12}$ operator in Eq. (34) should be divided by 4. J.R. Bergervoet, P.C. van Campen, R.A.M. Klomp, J.-L. de Kok, T.A. Rijken, V.G.J. Stoks, and J.J. de Swart, Phys. Rev. C [**41**]{}, 1435 (1990). J. Schwinger, Phys. Rev. D [**3**]{}, 1967 (1971). J. Binstock and R.A. Bryan, Phys. Rev. D [**4**]{}, 1341 (1971). J.J. de Swart and M.M. Nagels, Fortschr. Phys. [**26**]{}, 215 (1978). B.D. Day, Phys. Rev. C [**24**]{}, 1203 (1981). J.L. Friar, G.L. Payne, V.G.J. Stoks, and J.J. de Swart, Phys. Lett. B [**311**]{}, 4 (1993). D.C. Zheng and B.R. Barrett, University of Arizona preprint, 1993. C. van der Leun and C. Alderliesten, Nucl. Phys. [**A380**]{}, 261 (1982). N.L. Rodning and L.D. Knutson, Phys. Rev. C [**41**]{}, 898 (1990). M.W. Kermode, S. Klarsfeld, D.W.L. Sprung, and J.P. McTavish, J. Phys. G [**9**]{}, 57 (1983). V.G.J. Stoks, P.C. van Campen, W. Spit, and J.J. de Swart, Phys. Rev. Lett. [**60**]{}, 1932 (1988). D.M. Bishop and L.M. Cheung, Phys. Rev. A [**20**]{}, 381 (1979). C.P.F. Terheggen [*et al.*]{}, in preparation. M.H. MacGregor, R.A. Arndt, and R.M. Wright, Phys. Rev. [**182**]{}, 1714 (1969). ------------- --------- --------- --------- $n$ 0 1 1 $m$ 760.0 768.7 768.3 $\Gamma$ 640.0 152.4 149.1 $\beta_{1}$ 0.16900 0.26552 0.38755 $m_{1}$ 487.818 645.377 674.152 $\beta_{2}$ 0.61302 0.56075 0.45083 $m_{2}$ 1021.14 878.367 929.974 ------------- --------- --------- --------- : Values for the parameters of Eq. (\[Yukfit\]) of the two-pole approximation for the broad $\epsilon$ meson and the broad neutral and charged $\rho$ mesons. Masses $m$ and widths $\Gamma$ are in MeV. \[broad\] 0.00000 -------------------------- ----------------------- --------- --------- $m$ (MeV) $g^{2}$ $f^{2}$ $\pi^{\pm}$ 139.5675 0.07395 $\pi^{0}$ 134.9739 0.07402 $\eta$ 548.8 0.01514 $\eta'$ 957.5 0.01466 $\Lambda_{P}$ 1177.11 $\rho^{\pm}$ 768.3, $\Gamma=149.1$ 0.8481 14.217 $\rho^{0}$ 768.7, $\Gamma=152.4$ 0.8481 14.217 $\omega$ 781.95 9.1765 0.3383 $\phi$ 1019.412 0.0985 0 $\Lambda_{V}$ 904.50 $a_{0}^{\pm}, a_{0}^{0}$ 983.3 1.9174 $\epsilon$ 760.0, $\Gamma=640.0$ 28.196 $f_{0}$ 975.6 12.142 $\Lambda_{S}$ 554.40 $a_{2}$ 208.16 0.0486 $P,f_{2},f'_{2}$ 208.16 27.339 -------------------------- ----------------------- --------- --------- : Masses and meson-nucleon coupling constants at ${\bf k}^{2}=0$ for the Nijm93 potential. For the $np$ $^{1}S_{0}$ partial wave the coupling constants of the charged-rho meson are increased by 4.371% (see text). $P$ in the last line denotes the Pomeron. Note that $(f/g)_{\rho}=4.094$. \[parnym93\] ------------------------- -------- -------- --------- -------- -------- PWA93 Nijm I Nijm II Reid93 Nijm93 $pp$ 1787.0 1795.8 1795.8 1795.1 3175.6 $np$ 2489.2 2627.3 2625.7 2624.6 4848.4 Total 4276.2 4423.1 4421.5 4419.7 8023.9 $N_{\rm par}$ 39 41 47 50 15 $\chi^{2}/N_{\rm data}$ 0.99 1.03 1.03 1.03 1.87 ------------------------- -------- -------- --------- -------- -------- : $\chi^{2}$ for the new potential models in comparison with the Nijmegen multienergy analysis [@St93b] PWA93. We also show the number of parameters ($N_{\rm par}$) for each model. \[compare\] [rddddddd]{} $T_{\rm lab}$ & $^{1}S_{0}$ & $^{1}D_{2}$ & $^{3}P_{0}$ & $^{3}P_{1}$ & $^{3}P_{2}$ & $\varepsilon_{2}$ & $^{3}F_{2}$\ 1 & 32.77 & 0.00 & 0.13 & –0.08 & 0.01 & –0.00 & 0.00\ & 32.79 & 0.00 & 0.13 & –0.08 & 0.01 & –0.00 & 0.00\ & 32.80 & 0.00 & 0.13 & –0.08 & 0.01 & –0.00 & 0.00\ & 32.79 & 0.00 & 0.13 & –0.08 & 0.01 & –0.00 & 0.00\ 5 & 54.85 & 0.04 & 1.58 & –0.90 & 0.21 & –0.05 & 0.00\ & 54.88 & 0.04 & 1.58 & –0.90 & 0.22 & –0.05 & 0.00\ & 54.91 & 0.04 & 1.57 & –0.89 & 0.22 & –0.05 & 0.00\ & 54.85 & 0.04 & 1.57 & –0.90 & 0.21 & –0.05 & 0.00\ 10 & 55.22 & 0.17 & 3.73 & –2.06 & 0.65 & –0.20 & 0.01\ & 55.25 & 0.17 & 3.73 & –2.05 & 0.66 & –0.20 & 0.01\ & 55.28 & 0.17 & 3.70 & –2.04 & 0.65 & –0.20 & 0.01\ & 55.22 & 0.16 & 3.71 & –2.05 & 0.65 & –0.20 & 0.01\ 25 & 48.66 & 0.70 & 8.58 & –4.93 & 2.49 & –0.81 & 0.10\ & 48.68 & 0.70 & 8.60 & –4.91 & 2.50 & –0.81 & 0.11\ & 48.72 & 0.70 & 8.52 & –4.89 & 2.49 & –0.81 & 0.11\ & 48.71 & 0.69 & 8.60 & –4.90 & 2.49 & –0.80 & 0.10\ 50 & 38.92 & 1.71 & 11.47 & –8.32 & 5.85 & –1.71 & 0.34\ & 38.91 & 1.71 & 11.55 & –8.31 & 5.85 & –1.70 & 0.34\ & 38.92 & 1.70 & 11.48 & –8.30 & 5.84 & –1.70 & 0.34\ & 39.03 & 1.68 & 11.67 & –8.30 & 5.83 & –1.69 & 0.34\ 100 & 24.98 & 3.79 & 9.45 &–13.26 & 11.01 & –2.66 & 0.82\ & 24.96 & 3.73 & 9.50 &–13.30 & 10.96 & –2.63 & 0.82\ & 24.91 & 3.75 & 9.55 &–13.33 & 10.97 & –2.64 & 0.83\ & 25.09 & 3.71 & 9.79 &–13.30 & 10.97 & –2.61 & 0.81\ 150 & 14.77 & 5.61 & 4.74 &–17.43 & 13.98 & –2.87 & 1.20\ & 14.79 & 5.60 & 4.63 &–17.51 & 13.94 & –2.87 & 1.19\ & 14.70 & 5.61 & 4.77 &–17.54 & 13.95 & –2.87 & 1.20\ & 14.83 & 5.55 & 4.97 &–17.49 & 13.96 & –2.85 & 1.16\ 200 & 6.57 & 7.06 & –0.37 &–21.25 & 15.63 & –2.76 & 1.42\ & 6.66 & 7.20 & –0.63 &–21.32 & 15.65 & –2.80 & 1.39\ & 6.56 & 7.15 & –0.47 &–21.29 & 15.63 & –2.79 & 1.42\ & 6.62 & 7.08 & –0.32 &–21.26 & 15.63 & –2.77 & 1.36\ 250 & –0.30 & 8.27 & –5.43 &–24.77 & 16.59 & –2.54 & 1.47\ & –0.20 & 8.50 & –5.72 &–24.81 & 16.65 & –2.62 & 1.39\ & –0.25 & 8.41 & –5.61 &–24.67 & 16.61 & –2.61 & 1.44\ & –0.23 & 8.35 & –5.45 &–24.71 & 16.59 & –2.55 & 1.39\ 300 & –6.14 & 9.42 &–10.39 &–27.99 & 17.17 & –2.34 & 1.34\ & –6.18 & 9.52 &–10.49 &–28.02 & 17.26 & –2.41 & 1.20\ & –6.12 & 9.42 &–10.49 &–27.71 & 17.21 & –2.41 & 1.29\ & –6.10 & 9.40 &–10.29 &–27.86 & 17.15 & –2.26 & 1.25\ 350 &–11.11 & 10.69 &–15.30 &–30.89 & 17.54 & –2.21 & 1.04\ &–11.51 & 10.28 &–14.94 &–30.98 & 17.63 & –2.23 & 0.86\ &–11.28 & 10.24 &–15.08 &–30.45 & 17.61 & –2.23 & 0.98\ &–11.22 & 10.30 &–14.80 &–30.77 & 17.49 & –1.95 & 0.95 \[phspp\] [rdddddddddddd]{} $T_{\rm lab}$ & $^{1}S_{0}$ & $^{3}P_{0}$ & $^{1}P_{1}$ & $^{3}P_{1}$ & $^{3}S_{1}$ & $\varepsilon_{1}$ & $^{3}D_{1}$ & $^{1}D_{2}$ & $^{3}D_{2}$ & $^{3}P_{2}$ & $\varepsilon_{2}$ & $^{3}F_{2}$\ 1 & 62.07 & 0.18 & –0.19 & –0.11 & 147.75 & 0.11 & –0.01 & 0.00 & 0.01 & 0.02 & –0.00 & 0.00\ & 62.11 & 0.18 & –0.19 & –0.11 & 147.76 & 0.10 & –0.01 & 0.00 & 0.01 & 0.02 & –0.00 & 0.00\ & 62.09 & 0.18 & –0.19 & –0.11 & 147.75 & 0.10 & –0.01 & 0.00 & 0.01 & 0.02 & –0.00 & 0.00\ & 61.89 & 0.18 & –0.19 & –0.11 & 147.73 & 0.10 & –0.01 & 0.00 & 0.01 & 0.02 & –0.00 & 0.00\ 5 & 63.63 & 1.63 & –1.49 & –0.94 & 118.18 & 0.67 & –0.18 & 0.04 & 0.22 & 0.25 & –0.05 & 0.00\ & 63.74 & 1.62 & –1.50 & –0.93 & 118.19 & 0.67 & –0.18 & 0.04 & 0.22 & 0.25 & –0.05 & 0.00\ & 63.66 & 1.60 & –1.52 & –0.93 & 118.17 & 0.66 & –0.18 & 0.04 & 0.22 & 0.25 & –0.05 & 0.00\ & 63.23 & 1.61 & –1.48 & –0.93 & 118.15 & 0.66 & –0.18 & 0.04 & 0.22 & 0.26 & –0.05 & 0.00\ 10 & 59.95 & 3.65 & –3.04 & –2.06 & 102.61 & 1.16 & –0.68 & 0.16 & 0.85 & 0.71 & –0.18 & 0.01\ & 60.10 & 3.64 & –3.08 & –2.05 & 102.62 & 1.15 & –0.68 & 0.16 & 0.85 & 0.71 & –0.18 & 0.01\ & 59.99 & 3.61 & –3.11 & –2.04 & 102.59 & 1.13 & –0.67 & 0.16 & 0.85 & 0.71 & –0.18 & 0.01\ & 59.46 & 3.64 & –3.04 & –2.05 & 102.59 & 1.14 & –0.67 & 0.16 & 0.85 & 0.72 & –0.18 & 0.01\ 25 & 50.90 & 8.13 & –6.31 & –4.88 & 80.63 & 1.79 & –2.80 & 0.68 & 3.71 & 2.56 & –0.76 & 0.09\ & 51.04 & 8.16 & –6.42 & –4.86 & 80.59 & 1.77 & –2.80 & 0.69 & 3.72 & 2.57 & –0.75 & 0.09\ & 50.88 & 8.09 & –6.51 & –4.84 & 80.56 & 1.73 & –2.80 & 0.68 & 3.72 & 2.56 & –0.75 & 0.09\ & 50.41 & 8.24 & –6.37 & –4.85 & 80.63 & 1.74 & –2.75 & 0.67 & 3.73 & 2.61 & –0.75 & 0.09\ 50 & 40.54 & 10.70 & –9.67 & –8.25 & 62.77 & 2.11 & –6.43 & 1.73 & 8.97 & 5.89 & –1.63 & 0.30\ & 40.56 & 10.81 & –9.80 & –8.25 & 62.64 & 2.09 & –6.45 & 1.72 & 8.98 & 5.88 & –1.61 & 0.31\ & 40.35 & 10.76 & –9.96 & –8.24 & 62.62 & 2.00 & –6.45 & 1.72 & 8.97 & 5.87 & –1.62 & 0.31\ & 40.18 & 11.13 & –9.89 & –8.26 & 62.78 & 2.03 & –6.31 & 1.69 & 9.00 & 6.00 & –1.60 & 0.30\ 100 & 26.78 & 8.46 &–14.52 &–13.24 & 43.23 & 2.42 &–12.24 & 3.90 & 17.27 & 10.94 & –2.58 & 0.76\ & 26.44 & 8.57 &–14.42 &–13.30 & 42.98 & 2.44 &–12.26 & 3.83 & 17.26 & 10.89 & –2.54 & 0.76\ & 26.18 & 8.65 &–14.59 &–13.33 & 42.95 & 2.25 &–12.31 & 3.85 & 17.22 & 10.88 & –2.54 & 0.77\ & 26.32 & 9.22 &–14.91 &–13.38 & 43.18 & 2.36 &–12.07 & 3.79 & 17.12 & 11.21 & –2.54 & 0.74\ 150 & 16.93 & 3.69 &–18.65 &–17.46 & 30.72 & 2.75 &–16.49 & 5.79 & 22.12 & 13.84 & –2.80 & 1.12\ & 16.33 & 3.67 &–18.23 &–17.56 & 30.47 & 2.83 &–16.45 & 5.77 & 22.15 & 13.80 & –2.79 & 1.11\ & 16.02 & 3.83 &–18.32 &–17.59 & 30.40 & 2.59 &–16.61 & 5.78 & 22.05 & 13.78 & –2.78 & 1.12\ & 16.11 & 4.47 &–18.91 &–17.68 & 30.67 & 2.82 &–16.30 & 5.68 & 21.72 & 14.27 & –2.82 & 1.08\ 200 & 8.94 & –1.44 &–22.18 &–21.30 & 21.22 & 3.13 &–19.71 & 7.29 & 24.50 & 15.46 & –2.70 & 1.33\ & 8.27 & –1.60 &–21.51 &–21.41 & 21.08 & 3.27 &–19.62 & 7.43 & 24.61 & 15.48 & –2.73 & 1.28\ & 7.92 & –1.42 &–21.52 &–21.38 & 20.98 & 3.03 &–19.85 & 7.38 & 24.49 & 15.42 & –2.71 & 1.31\ & 7.83 & –0.74 &–22.15 &–21.55 & 21.31 & 3.40 &–19.46 & 7.26 & 23.96 & 16.01 & –2.79 & 1.27\ 250 & 1.96 & –6.51 &–25.14 &–24.84 & 13.39 & 3.56 &–22.21 & 8.53 & 25.40 & 16.39 & –2.49 & 1.35\ & 1.48 & –6.69 &–24.28 &–24.93 & 13.46 & 3.70 &–22.13 & 8.78 & 25.48 & 16.47 & –2.56 & 1.26\ & 1.10 & –6.55 &–24.22 &–24.78 & 13.35 & 3.55 &–22.34 & 8.68 & 25.46 & 16.39 & –2.54 & 1.32\ & 0.86 & –5.80 &–24.71 &–25.08 & 13.81 & 4.05 &–21.84 & 8.55 & 24.81 & 17.02 & –2.61 & 1.28\ 300 & –4.46 &–11.47 &–27.58 &–28.07 & 6.60 & 4.03 &–24.14 & 9.69 & 25.45 & 16.95 & –2.30 & 1.19\ & –4.43 &–11.46 &–26.52 &–28.17 & 7.00 & 4.10 &–24.21 & 9.82 & 25.32 & 17.08 & –2.36 & 1.06\ & –4.84 &–11.43 &–26.43 &–27.85 & 6.90 & 4.12 &–24.26 & 9.73 & 25.52 & 16.99 & –2.35 & 1.14\ & –5.17 &–10.57 &–26.69 &–28.31 & 7.55 & 4.74 &–23.64 & 9.63 & 24.84 & 17.64 & –2.36 & 1.12\ 350 &–10.58 &–16.39 &–29.66 &–30.97 & 0.50 & 4.57 &–25.57 & 10.96 & 25.08 & 17.31 & –2.18 & 0.87\ & –9.70 &–15.90 &–28.25 &–31.14 & 1.36 & 4.47 &–26.00 & 10.60 & 24.48 & 17.44 & –2.19 & 0.69\ &–10.10 &–16.01 &–28.17 &–30.60 & 1.30 & 4.74 &–25.74 & 10.59 & 25.02 & 17.40 & –2.18 & 0.81\ &–10.47 &–15.02 &–28.15 &–31.27 & 2.20 & 5.45 &–25.00 & 10.55 & 24.40 & 18.03 & –2.09 & 0.82 \[phsnp\] --------- -------- --------- -------- -------- Nijm I Nijm II Reid93 Nijm93 $\eta$ 0.0253 0.0252 0.0251 0.0252 $A_{S}$ 0.8841 0.8845 0.8853 0.8842 $N^{2}$ 0.7821 0.7828 0.7843 0.7823 $P_{d}$ 5.664 5.635 5.699 5.755 $Q_{d}$ 0.2719 0.2707 0.2703 0.2706 --------- -------- --------- -------- -------- : Deuteron properties of the potential models: $D/S$-ratio $\eta$, asymptotic $S$-state normalization $A_{S}$ in ${\rm fm}^{-1/2}$, wave function normalization $N^{2}$ in ${\rm fm}^{-1}$, $D$-state probability $P_{d}$ in %, and quadrupole moment $Q_{d}$ in ${\rm fm}^{2}$. \[deuter\]

ArXiv

--- abstract: | We propose a stringent observational test on the formation of warm Jupiters (gas-giant planets with $10{{\,\rm d}}\lesssim P \lesssim 100{{\,\rm d}}$) by high-eccentricity (high-$e$) migration mechanisms. Unlike hot Jupiters, the majority of observed warm Jupiters have pericenter distances too large to allow efficient tidal dissipation to induce migration. To access the close pericenter required for migration during a Kozai-Lidov cycle, they must be accompanied by a strong enough perturber to overcome the precession caused by General Relativity (GR), placing a strong upper limit on the perturber’s separation. For a warm Jupiter at $a \sim 0.2{\rm AU}$, a Jupiter-mass (solar-mass) perturber is required to be $\lesssim3 {\rm AU}$ $(\lesssim 30 {\rm AU})$ and can be identified observationally. Among warm Jupiters detected by Radial Velocities (RV), $\gtrsim 50\%$ [(5 out of 9)]{} with large eccentricities ($e\gtrsim 0.4$) have known Jovian companions satisfying this necessary condition for high-$e$ migration. In contrast, $\lesssim 20\%$ [(3 out of 17)]{} of the low-$e$ ($e\lesssim 0.2$) warm Jupiters have detected additional Jovian companions, suggesting that high-$e$ migration with planetary perturbers may not be the dominant formation channel. Complete, long-term RV follow-ups of the warm-Jupiter population will allow a firm upper limit to be put on the fraction of these planets formed by high-$e$ migration. Transiting warm Jupiters showing spin-orbit misalignments will be interesting to apply our test. If the misalignments are solely due to high-$e$ migration as commonly suggested, we expect that the majority of warm Jupiters with low-$e$ ($e\lesssim0.2$) are not misaligned, in contrast with low-$e$ hot Jupiters. author: - 'Subo Dong, Boaz Katz, and Aristotle Socrates' title: 'Warm Jupiters Need Close “Friends” for High-Eccentricity Migration – A Stringent Upper Limit on the Perturber’s Separation' --- Introduction ============ The origin of warm Jupiters (gas giants with period $10{{\,\rm d}}<P<100{{\,\rm d}}$) presents a similar puzzle to that of hot Jupiters ($P\lesssim 10{{\,\rm d}}$) – neither populations can form [*in-situ*]{} according to popular theories of planet formation – yet much less attention has been paid to the former. Rossiter-Mclaughlin measurements reveal that a considerable fraction of transiting hot Jupiters have orbits misaligned with host star spin axes [e.g. @spin1; @spin2], which provide indirect support to high-eccentricity migration mechanisms [@fordrasio; @wu03; @ft07; @secularchaos; @supere]. These high-$e$ mechanisms involve the initial excitation of hot Jupiter progenitors at a few ${\rm AU}$ to very high eccentricity due to gravitational perturbations by additional objects in the system. The excitation is then followed by successive close pericenter passages ($r_p \lesssim 0.05 {\rm AU}$) that drain the orbital energy via tidal dissipation. The hot Jupiter progenitors eventually become hot Jupiters at $a<0.1 {\rm AU}$. The majority of known warm Jupiters are sufficiently distant from their hosts (${a_{\rm f}}= a(1-e^2)>0.1{\rm AU}$) to forbid efficient tidal dissipation, due to the strong distance dependence of tidal effects. However, if the orbital eccentricity of a warm Jupiter is experiencing Kozai-Lidov oscillations due to an external perturber [@holman; @takeda05], then it may be presently at the low-$e$ stage in the cycles and over a secular timescale, reach an eccentricity high enough for tidal dissipation to cause significant migration [e.g., @secularchaos]. A schematic illustration of such a high-$e$ migration scenario is shown in Fig 1 (red solid line). [Warm Jupiters detected by RV are shown in dots in Fig 1. within the black dashed lines. We define Jovian planets to have minimum mass $M_p\sin i > 0.3 M_{\rm Jup}$ and set an upper limit in semi-major axis of $0.5\,{\rm AU}$ for warm Jupiters. This upper bound is well below the theoretical “snow line” of [*in-situ*]{} core-accretion formation at about $2.5-3{\rm AU}$ for solar-type stars (e.g., @snowline) and the observed “jump” in the $a$ distribution of giant planets at $\sim 1{\rm AU}$ (e.g., @wright09). In planet-planet scattering, a Jupiter can migrate without tidal dissipation by factor of $\sim 2$ if another Jupiter is ejected [@fordrasio], and our upper bound in distance is set to disfavor such a process.]{} We discuss a stringent observational constraint on warm Jupiter formation via high-$e$ migration – they must be accompanied by close, easily observable perturbers. These close pertubers are strong enough to overcome the precession caused by General Relativity (GR) to reach close enough periapses for effective tidal dissipation within Kozai-Lidov cycles. In contrast, high-e migration for hot Jupiters does not subject to such a stringent constraint on perturbers. Hot Jupiter progenitors can be excited to close periapses at their initial, relatively large semi-major axes with distant perturbers, and throughout the subsequent migration, their periapses may keep close enough for tidal dissipation. Hot Jupiters formed by high-$e$ migration can thus have distant perturbers that are difficult to detect. Perturber Constraints on Warm-Jupiter High-$e$ Migration ======================================================== We derive below a lower limit on the perturber strength for warm-Jupiter in high-$e$ migration due to tidal dissipation. We adopt a conservative criterion that tidal dissipation may operate when a Jovian planet reaches ${a_{\rm f}}= a(1-e^2) < 0.1 {\rm AU}$. Observationally, the eccentricities of Jovian planets circularize at ${a_{\rm f}}\sim 0.06 {\rm AU}$ [e.g., @otherQ]. Given that tidal dissipation has strong dependence on planet-star separation, it is safe to assume that tidal dissipation ceases to be efficient when ${a_{\rm f}}> a_{\rm f, crit} = 0.1 {\rm AU}$. [We stress that the criterion presented below is a necessary but not an adequate condition for high-$e$ migration. Without satisfying the criterion, the migration cannot occur, while fulfilling this requirement does not guarantee migration.]{} Consider a warm Jupiter with mass $M_p$ at semi-major axis $a$ and eccentricity $e_0$ orbiting a star with mass $M$ accompanied by a perturber of mass $M_{{{\rm per}}}$ at $a_{{{\rm per}}}$ and $e_{{{\rm per}}}$. The criterion is to require the warm Jupiter to reach $a(1-e^2) < a_{\rm f, crit} = 0.1{\rm AU}$ during Kozai-Lidov oscillation (see Fig. 2 for an example). At a given $a$, the amplitude of Kozai-Lidov oscillation in eccentricity is limited by sources of precession other than those induced by the perturber and is insensitive to tidal dissipation. At ${a_{\rm f}}\sim 0.1{\rm AU}$, the precessions due to tides and the rotating bulge of the host are negligible compared to GR for typical hosts. Below we consider the Kozai-Lidov oscillation at the warm Jupiter’s current $a$ due to the gravitational perturbation and GR precession. We ignore tidal dissipation and precession. An analytical constraint is derived under the simplest assumptions: (1) quadrupole approximation in perturbing potential (2) the warm Jupiter treated as test particle (3) the equation of motion is averaged over outer and inner orbits (“double-averaging”). We show below with numerical simulations that these are excellent approximations in deriving this constraint. Under these approximations, the following is a constant: [e.g., @ft07], $$\label{eq:C_K_gr} e^2(2-5\sin^2i\sin^2{\omega})+\frac{{\epsilon}_{{{\rm GR}}}}{\sqrt{1-e^2}}={\rm const},$$ where $$\begin{aligned} \label{eq:ep_gr} {\epsilon}_{{{\rm GR}}} & = \frac{8GM^2b_{{{\rm per}}}^3}{c^2a^4M_{{{\rm per}}}} \cr &\approx 1.3 \big( \frac{M}{M_\odot} \big)^2 \big( \frac{a}{0.2 {\rm AU}} \big)^{-4} \big( \frac{M_{{{\rm per}}}}{M_\odot} \big)^{-1} \big( \frac{b_{{{\rm per}}}}{30 {\rm AU}} \big)^{3} \cr &\approx 1.4 \big( \frac{M}{M_\odot} \big)^2 \big( \frac{a}{0.2 {\rm AU}} \big)^{-4} \big( \frac{M_{{{\rm per}}}}{M_{\rm Jup}} \big)^{-1} \big( \frac{b_{{{\rm per}}}}{3 {\rm AU}} \big)^{3}\end{aligned}$$ represents the relative strength of GR compared to the perturber, $i$ is the planet-pertuber mutual inclination, $\omega$ is planet’s argument of pericenter, and $b_{{{\rm per}}} = a_{{{\rm per}}}(1-e_{{{\rm per}}}^2)^{1/2}$ is perturber’s semi-minor axis. From Eq. (\[eq:C\_K\_gr\]), to reach an eccentricity $e$ from $e_0$, the following criterion must be satisfied regardless of the values of $i$ and $\omega$, $$\label{eq:eminmax} {\epsilon}_{{{\rm GR}}} \left( \frac{1}{\sqrt{1-e^2}}- \frac{1}{\sqrt{1-e_0^2}} \right) < 2e_0^2 + 3e^2.$$ We then put a lower limit on the “strength” of the perturber to reach $a(1-e^2) < a_{\rm f,crit}$ (and an upper limit on the separation ratio between the perturber and the warm Jupiter), $$\begin{aligned} \label{eq:crit} &\frac{b_{{\rm per}}}{a}< \left(\frac{8GM}{c^2a}\right)^{-1/3} \left(\frac{M}{M_{{\rm per}}}\right)^{-1/3}\times\cr &\left[2e_0^2+3\left(1-\frac{a_{\rm f,crit}}{a}\right)\right]^{1/3} \left(\sqrt{\frac{a}{a_{\rm f,crit}}}-\frac{1}{\sqrt{1-e_0^2}}\right)^{-1/3}.\end{aligned}$$ Fig.3 shows the constraints on $b_{{{\rm per}}}$ for $M_{{{\rm per}}} = M_\odot$ and $M_{{{\rm per}}} = M_{\rm Jup}$ in the upper and lower panels respectively, derived from Eq. \[eq:crit\] for $a_{\rm f,crit} = 0.1{\rm AU}$. The blue lines from above to below are for $e_0 = 0.5, 0.3, 0.0$, respectively ($e_0=0.3$ in dashed lines while others in solid lines). Recently, it was realized that corrections due to various approximations above may lead to significant effects in several scenarios [e.g. @ford00; @naoznature; @katzoct; @lithoct; @katzdong]. We show that the analytic constraint given by Eq. 4 are not affected by the inaccuracies of the adopted approximations using numerical integrations without these approximations. The effects of the quadrupole and test particle approximations are studied by performing 20000 simulations (10000 for $M_{{\rm per}}=M_{\odot}$ and 10000 for $M_{{\rm per}}=M_{\rm Jup}$). These simulations employ the double averaged approximation but include the octupole term and are not restricted to the test particle approximation. The warm Jupiters have $e_0=0.3$ and $a$ uniformly distributed (randomly) between $0.15$ and $0.5$AU. The eccentricities of the outer perturbers are uniformly distributed within $0-0.5$. The ratio $b_{{\rm per}}/a$ are uniformly distributed within $100-300$ ($10-30$) for solar-mass (Jupiter-mass) perturbers. The orbital orientations of the outer and inner orbits are randomly distributed isotropically. All runs were integrated to $5\,\rm Gyrs$. The results are shown in Fig 3. The integrations in which the warm Jupiter reaches $a(1-e^2)<a_{\rm f,crit}=0.1\rm AU$ are plotted as red dots and others in black. The analytical constraint given by Eq. \[eq:crit\] for the appropriate eccentricity $e_0=0.3$ (dashed blue lines) accurately traces the border of required perturbers for achieving the required eccentricity. [Note that at the limit given in Eq.\[eq:crit\], the strength of the octupole , $\sim {\epsilon}_{\rm oct} = a/a_{{\rm per}}[e_{{{\rm per}}}/(1-e_{{{\rm per}}}^2)] \lesssim 1/10 (M_{{\rm per}}/M_{\rm Jup})^{-1/3}$, is negligible compared to GR, ${\epsilon}_{{{\rm GR}}} \sim 1$. While a small octupole can change the orbital inclination and lead to Kozai cycles with growing eccentricities, the eccentricity cannot surpass the limit Eq. , which is the maximal value for all mutual orientations.]{} [^1] For the scenarios considered, the Kozai-Lidov time scale is much longer than the outer (and inner) orbital time scales, justifying the double averaging assumption. To illustrate this, the results of a direct 3-body integration are compared to those of an double-averaging integration in Fig 2. The considered warm Jupiter is at $a=0.3\rm AU$ and $e_0=0.3$ and has a solar-mass perturber corresponding to the limit derived from Eq. \[eq:crit\] with $e_{{\rm per}}=0.5$ and $b_{{\rm per}}=68.8 \rm AU$. The initial inclination is at $90^\circ$. The results of two integrations are practically indistinguishable (black dashed: double-averaging; red solid: direct 3-body), validating the double-averaging approximation. The approximation is even better for a Jupiter-mass perturber satisfying the same constraint. This is because it has a shorter period and a similar Kozai-Lidov time scale. In the limit of $a\gg a_{\rm f,crit}$ and $e_0{\rightarrow}0$, the following useful approximation can be obtained using Eq. \[eq:crit\], $$\begin{aligned} \label{eq:minper} \frac{b_{{\rm per}}}{a} &< \left(\frac{8GM}{3c^2 \sqrt{a a_{\rm f,crit}}}\right)^{-1/3} \left(\frac{M}{M_{{\rm per}}}\right)^{-1/3} \cr &\approx 175 \left(\frac{M_{{\rm per}}}{M_\odot}\right)^{1/3} \left(\frac{M}{M_\odot}\right)^{-2/3} \left(\frac{a}{0.2{\rm AU}}\right)^{1/6} \left(\frac{a_{\rm f,crit}}{0.1{\rm AU}}\right)^{1/6}\cr &\approx 17 \left(\frac{M_{{\rm per}}}{M_{\rm Jup}}\right)^{1/3} \left(\frac{M}{M_\odot}\right)^{-2/3} \left(\frac{a}{0.2{\rm AU}}\right)^{1/6} \left(\frac{a_{\rm f,crit}}{0.1{\rm AU}}\right)^{1/6}.\end{aligned}$$ [We stress that this approximation should be used for order-of-magnitude estimates since it is only accurate in the limit $e\sim 0$ and $a\gg a_{\rm f,crit} = 0.1 {\rm AU}$.]{} For warm Jupiters with $a\sim 0.1{\rm AU}-0.5{\rm AU}$ with Jovian-planet perturbers, this constraint leads to an upper limit in orbital separation of $\sim 1.5-10 {\rm AU}$ (period $2-30 {\rm yr}$). The RV semi-amplitude is $\gtrsim 10 {\rm m/s}$, accessible to available high-precision RV instruments. The perturbers at the high end in period range ($\sim 20 - 30 {\rm yr}$) are more challenging since they may not have completed the full orbits yet during the monitoring projects. While for more massive perturbers, the upper limit in orbital separation implies much longer periods – ($P_{{{\rm per}}} \propto {M_{{\rm per}}}^{1/2}$), they can generally be identified from the easily detectable RV linear trends, $$\begin{aligned} |\dot{v_\perp}| &= |\frac{GM_{{\rm per}}}{r_{{\rm per}}^2} \sin i_{{\rm per}}\sin \theta_{{\rm per}}|\\ &\sim 200 {\,\rm m s^{-1} yr^{-1}} \big( \frac{M_{{{\rm per}}}}{1 M_\odot} \big) \big( \frac{r_{{{\rm per}}}}{30 {\rm AU}} \big)^{-2} \nonumber\\ &\sim 20 {\,\rm m s^{-1} yr^{-1}} \big( \frac{M_{{{\rm per}}}}{1 M_J} \big) \big( \frac{r_{{{\rm per}}}}{3 {\rm AU}} \big)^{-2} \nonumber\end{aligned}$$ where $i_{{\rm per}}$, $\theta_{{\rm per}}$ and $r_{{\rm per}}$ are the inclination, position angle with respect to the line of the node and orbital separation of the perturber, respectively. Observations & Discussion ========================= In RV surveys, close binaries are commonly excluded to avoid contamination of the spectra, making the bias for estimating stellar perturbers challenging. We focus on planetary perturbers. There are $34$ warm Jupiters discovered with RV at $a_{\rm f}>0.1{\rm AU}$ and $a<0.5{\rm AU}$ listed in the exoplanets.org database [@exoplanetsorg]. $10$ of them have additional Jovian planets at longer orbits (see Fig.1, in which planets with outer Jovian companions are plotted in cyan). All perturbers satisfy the constraint in Eq. \[eq:crit\]. Fig.4 shows the eccentricity distribution for all warm Jupiters (blue dashed) and for those with external Jovian perturbers (red solid). The fraction of warm Jupiters with detected Jovian perturbers appears to be a growing function of their eccentricities. This trend appears to be significant in spite of the uncertainties due to the small number statistics. If true, there are two interesting implications: (1) There seems to exist a connection between eccentricity and the existence of a planet perturber capable of exciting such eccentricity. This implies that the eccentricities for the eccentric warm Jupiters are likely excited by their perturbers. High-$e$ migration is therefore an attractive scenario for their formation. (2) The majority of low-$e$ ($e<0.2$) warm Jupiters lack strong perturbers necessary for high-$e$ migration (Eq. \[eq:crit\]). Given that for eccentric warm Jupiters, similar perturbers are indeed detected around a considerable fraction of the systems, this deficiency seems to be unlikely due to detection sensitivity. Moreover, out of the five warm Jupiter systems with $e<0.4$ with Jovian perturbers, three are in compact multiple planet systems with 3 or more planets (55 Cnc b, GJ 876 c and HIP 57274 c), which are challenging to explain with high-$e$ migration. In contrast, for the five eccentric warm Jupiters at $e>0.4$, there are no known additional planets in the system other than their Jovian perturbers, all of which are located further than $2\,{\rm AU}$ yet close enough to satisfy the constraint from Eq.\[eq:crit\]. This is indicative that the majority of low-$e$ warm Jupiters are unlikely due to high-$e$ migration induced by planet perturbers. It is worth noting that $M_p\sin i$ rather than $M_p$ is constrained from RV, so the above results are statistical. Note that the perturbers for all 5 eccentric warm Jupiters have larger $M_p\sin i$ than the inner planets, consistent with simple expectations from the planet-planet scattering scenario that less massive planets are easier to get excited into high-$e$ orbits. [We caution that the conclusions may be affected if the chance for detecting an outer perturber strongly depends on the eccentricity of the inner planet. The observing strategies in RV surveys can be complicated, especially for those involving multiple planets. For example, @wright09 pointed out that a system was observed more frequently after a planet was found, so the detection of a massive planet would likely facilitate the discovery of smaller planets. A similar selection effect may make the detection of perturbers of eccentric warm Jupiters easier if their eccentricities attract particular attentions. A comprehensive sensitivity study would be helpful. Additionally, there are a number of possible modeling degeneracies that may masquerade a double low-$e$ planet system as an eccentric warm Jupiter [@eccentric1; @eccentric2]. Systematic modeling efforts are possibly needed to evaluate such degeneracies. There might be other mechanisms that produce eccentric warm Jupiters associated with a perturber, including scattering followed by disk migration similar to @disk and scattering of 3 planets with the third planet being ejected (Petrovich et al., in prep).]{} [High-precision RV surveys $(\lesssim 5{\rm m/s})$ on thousands of stars have lasted for $\sim 15$ years [e.g. @mayor11; @wright09; @aat]. For a considerable fraction of their targets, they can detect Jupiters at $\lesssim 5{\rm AU}$ with full orbits ([though note that some discoveries from these surveys remain unpublished]{}). Given our constraint on axis ratio of $\sim 20$ for Jovian perturbers, this implies the present observational constraint on planet perturbers are likely relatively incomplete for warm Jupiters at $\gtrsim 0.3 {\rm AU}$. For these systems, a thorough analysis of incomplete orbits and trends in RV is required. Unlike close solar-type companions, low-mass stellar and brown dwarf companions are unlikely to be excluded from the RV samples to search for planets. The combined efforts of RV linear trends and high-contrast imaging will yield excellent constraints for such perturbers [e.g., @TREND].]{} Rossiter-Mclaughlin effects for transiting planets are an important diagnostic for high-$e$ migration, to which the spin-orbit misalignments have been commonly attributed. No ground-based surveys have so far detected transiting warm Jupiters $({a_{\rm f}}= a(1-e^2)>a_{\rm f, crit}=0.1{\rm AU})$.[^2] Yet it is interesting to note that, among the ground-based transiting planets with the longest period, possibly requiring eccentricity oscillations for tidal migration, several have known additional planet companions or have large RV linear trends (e.g., HAT-P-17b @hatp17b, WASP-8b @wasp8b, KELT-6b @kelt6b). Future ground-based surveys or space-based surveys targeting bright stars are likely to discover warm Jupiters suitable for spin-orbit alignment measurements (note a possible transiting warm Jupiter candidate with a strong perturber identified by @dawson). They will be particularly interesting candidates subject to our proposed observational test on perturbers. If the spin-orbit misalignments are solely due to high-$e$ migration, and given that the majority of low-$e$ warm Jupiters do not seem to have strong enough perturbers for high-$e$ migration, we expect that the majority of warm Jupiters with low-$e$ ($e\lesssim0.2$) will be found to be aligned with the spin axes of their hosts. [Finally, if the warm Jupiters are indeed migrating due to tidal dissipation at the high-$e$ stage during Kozai-Lidov oscillations, they should be tidally powered and luminous enough to be detected by the future high-contrast imaging facilities such as those to be installed at TMT, GMT and ELT [@direct]. Similar high-$e$ migration mechanisms have also been raised for the formation of close binary stars at $P\lesssim 10 {{\,\rm d}}$ [@ft07; @binary], and the constraint we derive in this work can also be applied to test the formation of binaries at $10 {{\,\rm d}}\lesssim P \lesssim 100 {{\,\rm d}}$ due to high-$e$ mechanisms.]{} We thank Andy Gould, Scott Tremaine and Cristobal Petrovich for discussions. We are grateful to the referee for a helpful report. S. D. was partly supported through a Ralph E. and Doris M. Hansmann Membership at the IAS and by NSF grant AST-0807444. B. K. is supported by NASA through the Einstein Postdoctoral Fellowship awarded by Chandra X-ray Center, which is operated by the Smithsonian Astrophysical Observatory for NASA under contract NAS8-03060. BK and AS acknowledges support from a John N. Bahcall Fellowship at the Institute for Advanced Study, Princeton. This research has made use of the Exoplanet Orbit Database and the Exoplanet Data Explorer at exoplanets.org. [99]{} Anglada-Escud[é]{}, G., L[ó]{}pez-Morales, M., & Chambers, J. E. 2010, , 709, 168 Collins, K. A., Eastman, J. D., Beatty, T. G., et al. 2013, arXiv:1308.2296 Crepp, J. R., Johnson, J. A., Howard, A. W., et al. 2012, , 761, 39 Dawson, R. I., Johnson, J. A., Morton, T. D., et al. 2012, , 761, 163 Dong, S., Katz, B., & Socrates, A. 2013, , 762, L26 Dong, S., Katz, B., & Socrates, A. 2013, , 763, L2 Holman, M., Touma, J., & Tremaine, S. 1997, , 386, 254 Howard, A. W., Bakos, G. [Á]{}., Hartman, J., et al. 2012, , 749, 134 Fabrycky, D., & Tremaine, S. 2007, , 669, 1298 Ford, E. B., Kozinsky, B., & Rasio, F. A. 2000, , 535, 385 Guillochon, J., Ramirez-Ruiz, E., & Lin, D. 2011, , 732, 74 Katz, B., Dong, S., & Malhotra, R. 2011, Physical Review Letters, 107, 181101 Katz, B., & Dong, S. 2012, arXiv:1211.4584 Kennedy, G. M., & Kenyon, S. J. 2008, , 673, 502 Lithwick, Y., & Naoz, S. 2011, , 742, 94 Mayor, M., Marmier, M., Lovis, C., et al. 2011, arXiv:1109.2497 Naoz, S., Farr, W. M., Lithwick, Y., Rasio, F. A., & Teyssandier, J. 2011, , 473, 187 Queloz, D., Anderson, D. R., Collier Cameron, A., et al. 2010, , 517, L1 Rasio, F. A., & Ford, E. B. 1996, Science, 274, 954 Rodigas, T. J., & Hinz, P. M. 2009, , 702, 716 Sanchis-Ojeda, R., Fabrycky, D. C., Winn, J. N., et al. 2012, , 487, 449 Socrates, A., Katz, B., Dong, S., & Tremaine, S. 2012, , 750, 106 Socrates, A., Katz, B., & Dong, S. 2012, arXiv:1209.5724 Takeda, G., & Rasio, F. A. 2005, , 627, 1001 Triaud, A. H. M. J., Collier Cameron, A., Queloz, D., et al. 2010, , 524, A25 Winn, J. N., Fabrycky, D., Albrecht, S., & Johnson, J. A. 2010, , 718, L145 Wittenmyer, R. A., Tinney, C. G., O’Toole, S. J., et al. 2011, , 727, 102 Wright, J. T., Upadhyay, S., Marcy, G. W., et al. 2009, , 693, 1084 Wright, J. T., Fakhouri, O., Marcy, G. W., et al. 2011, , 123, 412 Wu, Y., & Murray, N. 2003, , 589, 605 Wu, Y., & Lithwick, Y. 2011, , 735, 109 ![Schematic illustrations of high-$e$ migration on $a-e$ diagram. All Jovian planets discovered by RV are plotted in dots (those with additional known Jovian companions in cyan dots while others in black). Warm Jupiters are bounded by the black dashed lines. They are too close ($a\lesssim0.5{\rm AU}$) to be formed [*in situ*]{} and too distant to experience efficient tidal dissipation (${a_{\rm f}}= a(1-e^2)>0.1{\rm AU}$). The red solid line shows a possible evolution path to produce a warm Jupiter. The gravitational perturber is strong enough to overcome GR precession so that the planet has significant oscillations in eccentricity at $a \sim 0.3{\rm AU}$ to access ${a_{\rm f}}\lesssim 0.1{\rm AU}$. The observed warm Jupiters may be in the low-$e$ stages during such a migration from $a\gtrsim 1{\rm AU}$. The blue dotted-dashed line shows the high-$e$ migration due to a weak perturber that cannot compete with GR near $a \sim 0.3{\rm AU}$. Its eccentricity does not oscillate and ${a_{\rm f}}= a(1-e^2)$ is “frozen” to a low value, $<0.1 {\rm AU}$. \[fig:schematic\]](fig1.eps) ![ Numerical simulations of Kozai-Lidov oscillations with GR precession for a warm Jupiter. The planet is at $a=0.3{\rm AU}$, $e=0.3$, and $i=90^{\circ}$, and it has a solar-mass perturber at $e_{{{\rm per}}} = 0.5$ and $b_{{{\rm per}}} = a_{{{\rm per}}}(1-e_{{{\rm per}}}^2)^{1/2}=68.8 {\rm AU}$, which is at the limit derived from Eq. \[eq:crit\] to reach ${a_{\rm f}}= a(1-e^2) = a_{\rm f,crit}=0.1{\rm AU}$. At higher $e$ (lower ${a_{\rm f}}$), the tidal dissipation may be efficient. The eccentricity as a function of time from direct 3-body integration is shown in red line and that from double-averaging calculations (to octupole order in the perturbing potential) is shown in black dashed line. The two integrations show excellent agreement, validating the double-averaging approximation. As can be seen in the inset, the 3-body integration shows slight variations from the double-averaging calculations within each orbital period of the outer perturber. However, their impact on the long term evolution “averages out” to essentially zero, meaning that they play no role in the current study. ](fig2.eps){width="\textwidth"} ![The perturber constraint of warm Jupiters for high-$e$ migration. Upper and lower panels are for solar-mass and Jupiter-mass perturbers, respectively. The blue lines show the analytical upper limit (Eq. \[eq:crit\]) in the ratio between semi-minor axis of the perturber and the warm Jupiters’ semi-major axis $b_{{{\rm per}}}/a = (1-e_{{\rm per}}^2)^{1/2} a_{{\rm per}}/a$ as a function of $a$. The blue lines from above to below correspond to eccentricities of warm Jupiters of $e_0 = 0.5, 0.3, 0.0$, [which require increasing amount of oscillation amplitude to reach the required maximum eccentricity for decreasing initial eccentricities]{}. This is verified by 10000 numerical simulations with random initial orbital orientations that include the double-averaging octupole-order approximation and without neglecting the effect of the mass of the warm Jupiter. Each simulation is shown as a dot and the initial eccentricity of the planet is fixed at $0.3$. Red dots represent integrations in which the planet reaches ${a_{\rm f}}= a(1-e^2)<0.1{\rm AU}$ within $5{\,\rm Gyr}$ and black otherwise. The results are in excellent agreement with the corresponding analytical constraints shown in blue dashed lines ($e_0=0.3$). ](fig3.eps) ![The eccentricity distribution of warm Jupiters. The blue-dashed histogram is for known RV warm Jupiters ($M_p \sin i > 0.3 M_{\rm Jup}$, ${a_{\rm f}}>0.1{\rm AU}$, $a<0.5{\rm AU}$). The red solid histogram is for warm Jupiters with an external Jovian perturber. All satisfy the constraint in Eq. \[eq:crit\]. The fraction of warm Jupiters with detected Jovian perturbers is a growing function of eccentricities. $\gtrsim 50\%$ of the warm Jupiters with large eccentricities ($e\gtrsim 0.4$) have Jovian companions. A large fraction of low-$e$ warm Jupiters lack such perturbers. Out of the five warm Jupiter systems with $e<0.4$ with known additional Jovian companions, three are in compact multiple planet systems with 3 or more planet, which are difficult to be explained in high-$e$ migration. ](fig4.eps){width="\textwidth"} [^1]: See relevant discussion in “Maximal $e$ and General Relativity (GR) precession” of Katz, Dong & Malhotra., 2011, arXiv:1106.3340. [^2]: Note that the Kepler-30 system contains a warm Jupiter and the orbits of its three planets are shown to be aligned with the spin axis of the host [@kepler30]. The three planets are in a compact orbit configuration, and they are unlikely to be formed by high-$e$ migration.

ArXiv

Introduction ============ Following a long series of developments in the experimental techniques of atomic and optical physics, the Bose–Einstein condensation (BEC) of cold alkali atomic gases was realized in 1995(see, e.g., [@pethick_smith; @pitaevskii_stringari; @leggett_review] and references therein). The creation of this new state of quantum matter has opened up a new research field, the physics of ultracold atomic gases. The novelty of this system lies in its high controllability: various system parameters such as the dimensionality, the configuration of the external potentials, and the strength and the sign of the inter-atomic interaction can be manipulated dynamically as well as statically. In addition, this system has high measurability: since both the spatial and temporal microscopic scales of this system are relatively large, real time observation and direct imaging are possible. With these unique features, ultracold atomic gases serve as an unprecedented playground of the quantum world. Due to the emergence of the superfluid order parameter, BECs acquire a nonlinear character originating from the interparticle interaction. Here, nonlinearity means that the basic equation governing the state of the system depends on the state itself. A variety of phenomena caused by nonlinearity such as solitons [@burger99; @denschlag00; @strecker02; @khaykovich02] and matter-wave mixing [@deng99], *etc*. have been predicted and realized in BECs. Remarkably, the strength of the nonlinearity in BECs of cold atomic gases is controllable. This is because here $s$-wave scattering length $a_s$, the parameter characterizing the interatomic interaction, can be tuned using the Feshbach resonance. Therefore, the realization of BECs of cold atomic gases has opened up a new horizon for the study of nonlinear phenomena(see, e.g., [@kevrekidis]). With further development of technology and tools, superfluidity has been realized using cold fermionic atoms as well. It has been shown that, by increasing the interatomic attraction using Feshbach resonances, the state of an atomic Fermi gas can be varied in a controlled manner from a Bardeen–Cooper–Schrieffer (BCS) superfluid of delocalized Cooper pairs to a BEC of tightly-bound dimers [@zwierlein05]. Furthermore, these two limits are smoothly connected without a phase transition: the so-called BCS-BEC crossover [@eagles69; @leggett80; @giorgini_review; @bloch_review]. The experimental confirmation of the BCS-BEC crossover is one of the prime achievements in the field of cold atomic gases. Using BCS-BEC crossover, we can understand both the Bose and Fermi superfluids from a unified perspective. Another important development in the field of cold atomic gases is the realization of the external periodic potential called an “optical lattice”: Pairs of counter-propagating laser fields detuned from atomic transition frequencies, act as a free-of-defect, conservative potential for atoms via the optical dipole force. The realization of optical lattices has opened up the connection between the physics of cold atomic gases and solid state/condensed matter physics (see, e.g., [@bloch_review; @lattice; @yukalov_review; @jkps_review] for reviews), and especially enables the simulation of theoretical models of solid state physics using cold atoms. As a consequence of the competition and interplay between the effects of the periodic potential and nonlinearity, rich phenomena are expected to emerge in cold atomic gases in optical lattices. Especially, equipped with Feshbach resonance, a knob for controlling the strength of the nonlinearity, cold atomic gases in optical lattices allow us to enter a regime in which the effect of the nonlinearity is comparable to (or even dominates over) that of the periodic potential. Such a strongly nonlinear regime beyond the tight-binding approximation has not been well-explored in conventional solid state physics. For example, a loop structure called “swallowtail” in the Bloch energy band [@wu_st; @diakonov02] is a representative novel phenomenon emerging in this regime. In addition, using cold atomic gases, direct observations of the resulting nonlinear phenomena are possible, which is also a difficult task using solids. In this short review article, we discuss nonlinear phenomena of superfluid cold atomic gases in optical lattices. Especially, we consider extended states and focus on the following phenomena: the swallowtail band structure, Bloch states with multiple periods of the applied optical lattice potential called multiple period states, and those in nonlinear lattices, *i.e.*, systems with a periodically modulated interaction strength in space. This article is complementary to the existing review article on nonlinear phenomena in lattices [@malomed_review], which focuses mainly on localized states. Superfluidity is the most important macroscopic quantum phenomena and superfluid flow in a periodic potential is ubiquitous in many other systems, such as superconducting electrons in superconductors and even in astrophysical environments such as superfluid neutrons in “pasta” phases in neutron star crusts (see, e.g., [@pasta_review; @qmd_review] and references therein). Through the study of cold atomic gases in optical lattices, one may also expect to get deeper insights into these other systems. This article is organized as follows. In Section \[sec:framework\], we explain the setup of our system and basic theoretical formalism employed in the later discussions. In Sections \[sec:swallowtail\]–\[sec:nonlinlat\], we provide a comprehensive overview of the selected nonlinear phenomena in optical lattices starting with a simple physical explanation for each topic: swallowtail loops in Section \[sec:swallowtail\], multiple period states in Section \[sec:multiperiod\], and nonlinear lattices in Section \[sec:nonlinlat\]. Finally, summary and prospects are given in Section \[sec:conclusion\]. Theoretical Framework \[sec:framework\] ======================================= Setup of the System ------------------- In the present article, we discuss superfluid flows of either fermionic or bosonic atoms in the presence of the externally imposed periodic potentials. For the external periodicity, we mainly consider one of the most typical cases: one-dimensional (1D) sinusoidal potential of the form, $$V({\bf r}) = V(x) = s E_R \sin^2{q_Bx} \equiv V_0 \sin^2{q_Bx}\, , \label{eq:lat}$$ either in quasi-1D or 3D systems. Here, $E_R = \hbar^2q_B^2/2m$ is the recoil energy, $m$ is the mass of atoms, $q_B = \pi/d$ is the Bragg wave number (note that $q_B$ is different from the fundamental vector of a 1D reciprocal lattice, $2\pi/d$, by a factor of $2$), and $d$ is the lattice constant, $V_0 \equiv sE_R$ is the lattice height, and $s$ is the dimensionless parameter characterizing the lattice intensity in units of $E_R$. For simplicity, we also assume that the superflow is in the same direction as the periodic potential (*i.e.*, $x$ direction). Throughout the present article, we set the temperature $T = 0$. The systems which we discuss in this article consist of a large number of particles (the number of particles per site is also large) at temperatures close to absolute zero. One of the most convenient ways to deal with such many-body systems is to use the mean-field approximation. In this formalism, one focuses on a particular single particle, and the interactions produced by all the other particles are replaced by an averaged interaction described by the “mean field”. Thus the complicated many-body problem is effectively reduced to a far simpler one-body problem. The mean-field theory provides a minimal framework to study the nonlinear phenomena emerging from the presence of the superfluid order parameter. The mean-field theory enables us to predict novel nonlinear phenomena and obtain qualitative understanding of them although its validity is not always guaranteed. We resort to a mean-field description throughout as it readily fits our motivation in the present review—to provide a physical explanation of some selected, novel nonlinear phenomena of superfluids in periodic systems. In the rest of this section, we provide a brief explanation of the theoretical framework used in the discussions in the remaining part of this article. The main purpose of this section is to provide a minimal explanation and define the notation. Therefore, interested readers are encouraged to refer to other references (e.g., [@pethick_smith; @pitaevskii_stringari; @dalfovo_review; @giorgini_review]) for further details. Bosons ------ The mean-field theory describing Bose–Einstein Condensates (BECs) at zero temperature is given by the Gross–Pitaevskii (GP) equation [@pitaevskii61; @gross61; @gross63; @pethick_smith; @pitaevskii_stringari; @dalfovo_review]: $$\label{GP1} i\hbar \dfrac{\partial \psi({\bf r},t)}{\partial t}=\left[-\dfrac{\hbar^2}{2m}\nabla^2+ V({\bf r})+g|\psi({\bf r},t)|^2\right]\psi({\bf r},t)\, .$$ Here $\psi({\bf r},t)$ is the superfluid order parameter (or the condensate wave function) and $g$ is the effective coupling constant between two interacting bosons given by $$g=\dfrac{4 \pi \hbar^2 a_s}{m}\, ,$$ where $a_s$ is the $s$-wave scattering length. The average number density $n$ is $$n= \frac{N}{\mathcal V} = \frac{1}{\mathcal V}\int|\psi({\bf r})|^2\, d{\bf r}\, ,$$ where $N$ is the total number of particles and ${\mathcal V}$ is the volume of the system. Note that the GP equation can be viewed as the dynamical equation that results from a governing Hamiltonian known as the GP energy functional given by: $$E[\psi] = \int d {\mathbf{r}}\left(\frac{\hbar^2}{2m} \vert \nabla \psi \vert^2+ V({\mathbf{r}}) \vert \psi \vert^2 + \frac{g}{2} \vert \psi \vert^4 \right) \label{eq:GPerg}.$$ The stationary solution of Equation (\[GP1\]) is given by $$\label{GP2} \mu \psi({\bf r})=\left[-\dfrac{\hbar^2}{2m}\nabla^2 + V({\bf r})+g|\psi({\bf r})|^2\right] \psi({\bf r})\, ,$$ where $\mu$ is the chemical potential. Nonlinearity of the GP equation (the third term in the rhs of Equations (\[GP1\]) and (\[GP2\])) originates from the interaction between bosonic atoms. Many previous studies have shown that GP equation describes BECs of dilute, weakly interacting bosons at zero temperature quite successfully(see, e.g., [@dalfovo_review] and references therein). Fermions -------- A useful method for treating superfluid Fermi gases is the standard BCS mean-field theory of superconductivity. Such a mean-field theory for inhomogeneous systems is given by the Bogoliubov-de Gennes (BdG) equations [@degennes; @giorgini_review]: $$\begin{pmatrix} H'({\bf r}) & \Delta({\bf r})\\ \Delta^*({\bf r})& -{H'}({\bf r}) \end{pmatrix} \begin{pmatrix} u_i({\bf r})\\ v_i({\bf r}) \end{pmatrix} =\epsilon_i \begin{pmatrix} u_i({\bf r})\\ v_i({\bf r}) \end{pmatrix}\, . \label{eq:bdg}$$ Here $H'({\bf r}) = -\dfrac{\hbar^2}{2m}\nabla^2+V({\bf r})-\mu$. Also, $v_i({\bf r})$ and $u_i({\bf r})$ are the quasiparticle amplitudes, associated with the probability of occupation and unoccupation of a paired state denoted by an index $i$, while $\epsilon_i$ is the corresponding eigen-energy. The quasiparticle amplitudes $v_i({\bf r})$ and $u_i({\bf r})$ satisfy the normalization condition $\int d{\bf r}\, [u_i^*({\bf r})u_j({\bf r}) + v_i^*({\bf r})v_j({\bf r})] = \delta_{i,j}$. $\Delta$ is the order parameter (or the pairing field), which reduces to the pairing gap in the single quasiparticle spectrum in the region of $\mu>0$ for the uniform system. The pairing field $\Delta({\bf r})$ and the chemical potential $\mu$ in Equation (\[eq:bdg\]) are self-consistently determined from the gap equation, $$\Delta({\bf r}) = -g \sum_i u_i({\bf r})v^*_i({\bf r})\, ,\label{eq:gap}$$ and the average number density $$n= \frac{N}{\mathcal V} = \frac{1}{\mathcal V} \int n({\bf r})\, d{\bf r} = \frac{2}{\mathcal V}\sum_i\int|v_i({\bf r})|^2 d{\bf r}\, .$$ Since $\Delta$ depends on $\{u_i\}$ and $\{v_i\}$, the BdG equations (\[eq:bdg\]) are nonlinear for nonzero interatomic interaction parameter $g$. The superfluid Fermi systems bear a direct analogy with traditional superconducting systems, and likewise $g$, the contact interaction, plays similar role as the weakly attractive interaction term in the BCS-model. Only, now $g$ can be both small or large, and its value can be externally tuned using Feshbach resonances by applying a magnetic or an optical field. This controllability leads to a crossover between two ends: a weakly attractive BCS-like superfluid and a condensate of tightly bound bosonic molecules of a pair of fermionic atoms; popularly called the BCS-BECcrossover [@eagles69; @leggett80]. For contact interactions, the right-hand side of Equation (\[eq:gap\]) has an ultraviolet divergence, which has to be regularized by replacing the bare coupling constant $g$ with the two-body T-matrix related to the $s$-wave scattering length [@randeria]. A standard scheme [@randeria] is to introduce a cutoff energy $E_c \equiv \hbar^2k_c^2/2m$ in the sum over the BdG eigenstates and to replace $g$ by the following relation: $$\dfrac{1}{g} = \dfrac{m}{4 \pi \hbar^2 a_s}-\sum_{k<k_c}\dfrac{1}{2\epsilon_k^{(0)}}\, ,$$ with $\epsilon_k^{(0)}\equiv \hbar^2k^2/2m$. Discrete and Continuum Models ----------------------------- The systems of cold atomic gases can be studied by solving the GP equation (bosons) or the BdG equations (fermions) for the full continuum model. Let us, for the sake of simplicity, consider quasi-1D bosonic systems. So instead of $V({\bf r})$, we think of a potential in $x$ direction only: $V(x)$.If there is a periodicity in the form of $V(x)$, or, if $g$ itself is a periodic function of $x$, one approach is to try the Bloch solutions $\psi(x)=e^{ikx}\phi(x)$, where $\hbar k \equiv P$ is the quasimomentum of the superflow and $\phi(x)$ is a periodic function with the same periodicity as the externally imposed periodicity by $V(x)$ or $g(x)$. One can expand $\phi(x)$ in terms of plane waves to give the following form for the order parameter: $$\psi(x) = e^{ikx} \phi(x) = e^{ikx}\sum_{l=-l_{\rm max}}^{l_{\rm max}} a_l e^{i 2\pi l x/d}\, , \label{eq:Blochansatz}$$ to find the Bloch solutions. The normalization condition yields $\sum_l |a_l|^2 =1$. Instead of going for the full solution, one easier approach is to map the system to a discrete model, borrowed from the idea of tight-binding model in solid-state physics. In this approach the density of bosonic/fermionic atoms is assumed to be concentrated around the minima of the optical lattice potential. For example, in a quasi-1D periodic potential, the condensate wave function can be approximated by a superposition of wave functions $\phi_j(x)$ localized at the lattice sites, denoted by $j$, which are normalized as $\int |\phi_j(x)|^2 dx =1$. Thus, $\psi(x,t)= \sum_j\psi_j(t)\phi_j(x)$. The coefficient $\psi_j$ is dependent on the site index $j$. The Hamiltonian for such a discrete model is $$\label{hamilt} H = -K \sum_j (\psi_j^* \psi_{j+1}+ \psi_{j+1}^* \psi_j) +\dfrac{U}{2}\sum_{j}|\psi_j|^4\, . $$ In the case of the periodic solution with the same periodicity as that of the lattice (lattice constant $d$), the normalization condition is given by $$\int_{-d/2}^{d/2} |\psi(x)|^2 dx = \nu\, ,$$ where $\nu$ is the filling factor (number of particles per site) with $$\nu = |\psi_j|^2\, .\label{eq:normdiscrete2}$$ In evaluating the normalization condition, one neglects the overlap between $\phi_j$’s localized at different sites. The first term in the Hamiltonian describes hopping between the nearest-neighbor sites.The hopping parameter $K$ is given by $K = -\int \phi_j[-\frac{\hbar^2}{2m}\nabla^2 + V(x)] \phi_{j+1} dx$. The on-site interaction parameter $U$ characterizes the interaction energy between two atoms on the same site, and gives the nonlinear term. This $U$ is connected to the interatomic interaction parameter $g$ by $U=g\int|\phi_j(x)|^4 dx$. [In a manner similar to the continuum model one can also derive a dynamical equation for the amplitudes $\psi_j$ from an extermisation of the energy functional corresponding to Equation (\[hamilt\]). Such a dynamical equation is known in literature as the discrete nonlinear Schrödinger equation (analogous to Equation (\[GP1\]) for the continuous case) [@DNLS1; @DNLS2] and has served as an important tool to study ultracold atoms in optical lattices.]{} Energetic and Dynamical Stability\[subsec:thframeergstab\] ---------------------------------------------------------- Once one can solve for the system, either from the full continuum model or the discrete version, the next step is to study the energetic and dynamical stability of the stationary solutions. Energetic stability guarantees that the stationary states are a local energy minimum of the energy functional \[Equation (\[eq:GPerg\])\] and dynamical stability means that the time evolution of the system is stable with respect to small perturbations (this issue will be discussed in detail later in Figure \[fig:niustab1\]a in Section \[sec:loopstability\]). The standard approach in this context is the linear stability analysis described below [@pethick_smith; @wu01; @machholm03; @machholm04]. Let $\delta\phi_{q}(x)$ be the deviation from the stationary Bloch wave solution $\phi(x)$ for a given quasimomentum $\hbar k$ of the superflow. This can be written in the following form: $$\delta\phi_{q} =u(x,q) e^{i q x}+ {v}^*(x,q)e^{-i q x}\, .$$ Here $\hbar q$ is the quasimomentum of the perturbation. The energy deviation from the stationary states can be written in the following form $$\delta E=\int dx \begin{pmatrix} u^{*}& v^{*} \end{pmatrix} M(q) \begin{pmatrix} u\\ v\\ \end{pmatrix}\,. \label{eq:ergstabcondn}$$ The matrix $M(q)$ is Hermitian and gives the curvature of the energy landscape around the stationary solution. The system is energetically stable as long as all of the eigenvalues of $M(q)$ are positive. When even one of the eigenvalues is non-positive, the solution is no longer a local minimum and the system is energetically unstable. This is often termed as “Landau instability”. For the same perturbation $\delta\phi_q$, the time-evolution of the system for each $k$ is found to be: $$i \dfrac{\partial}{\partial t}\begin{pmatrix} u\\ v\\ \end{pmatrix}= \sigma_z M(q) \begin{pmatrix} u\\ v\\ \end{pmatrix}, \label{dynmat}$$ with $$\sigma_z = \begin{pmatrix} 1 & 0\\ 0 & -1 \end{pmatrix}.$$ Unlike $M(q)$, the matrix $\sigma_z M(q)$ is not Hermitian. If the eigenvalue of $\sigma_z M(q)$ is complex, the perturbation corresponding to an eigenvalue with positive imaginary part grows exponentially in time: in this case the stationary solution is dynamically unstable. If the imaginary part is zero, the stationary state remains stable (*i.e.*, dynamically stable). Therefore, by noting the eigenvalues of $M(q)$ and $\sigma_z M(q)$, one can learn whether the system belongs to an energetically stable region to start with, and how it evolves during the course of time. This is important because for BECs and superfluid Fermi gases, this stability translates to the sustenance of superfluidity in the system. Swallowtail Loops in Band Structure \[sec:swallowtail\] ======================================================= In this section we consider one unique manifestation of the nonlinearity (see Equation (\[GP1\])) governing the dynamics of BECs in optical lattices—the so-called swallowtail loops in band structures. In the first Section \[subs31\] we provide the basic physical idea and the context in which swallowtail structures arise in the energy dispersions. [Moreover, we discuss the key implication of having such swallowtail dispersions—breakdown of adiabaticity even at very slow driving.]{} The purpose of this subsection will be to communicate the central physical picture succinctly, hence we shall sacrifice chronology and use the most clear presentation (in our opinion) following [@niu00; @diakonov02; @machholm03; @mueller02]. In the second Section \[subs32\] we present a more detailed account of the various theoretical results on swallowtail band structures including their energetic and dynamical stability. In the third Section \[subsec:nonlinexp\] we present some experimental results that consider the effects of such nonlinear structures. In the following Section \[subs31\], we present a brief account of more recent developments that extend the swallowtail phenomena to situations beyond the standard $s$-wave interacting Bosons in optical lattices including dipole-dipole interactions, superfluid Fermi gases, *etc.* We conclude the section with some brief remarks indicating future prospects. Basic Physical Idea: The Nonlinear Landau–Zener Model and Variational Ansatz for Condensate Wavefunction in Optical Lattices {#subs31} ---------------------------------------------------------------------------------------------------------------------------- The GP equation (\[GP1\]) for the order parameter describing a BEC at zero temperature differs from the Schrödinger equation for a single particle in one key aspect — the presence of the interaction term $g \vert \psi (\mathbf{r}, t)\vert^2$ in addition to the externally applied potential term $V({\mathbf{r}})$. When the order parameter $\psi({\mathbf{r}})$ is expanded in terms of a given complete basis set of single particle wavefunctions, the nonlinear term in the GP equation essentially leads to an effective potential that is dependent on the occupation probability of the different single particle states. One may anticipate that in the limit that the nonlinearity is comparable to the external potential, *i.e.*, $g n \sim V({\mathbf{r}})$, the resulting dynamics as well as the stationary solutions of the GP equation (\[GP2\]) can be very different from the equivalent ones for the non-interacting $g=0$ case which are simply governed by the eigen-energies of the Schrödinger equation. The simplest model to see the structures that emerge in the limit of large nonlinearities is the nonlinear two-level system introduced in [@niu00]: $$\begin{aligned} i \frac{\partial}{\partial t} \begin{pmatrix} a \\ b \end{pmatrix} &= H(\gamma) \begin{pmatrix} a \\ b \end{pmatrix}\, ,\label{eq:nonlinLZ}\\ H(\gamma) &= \begin{pmatrix} \frac{\gamma}{2}+\frac{c}{2} \left(|b|^2-|a|^2 \right) & v/2 \\ v/2 & -\frac{\gamma}{2}-\frac{c}{2} \left(|b|^2-|a|^2 \right) \end{pmatrix}. \label{eq:nLZmat}\end{aligned}$$ Here $\gamma$ gives the level separation and $v/2$ is the coupling between the levels. , and $g=0$, the above Hamiltonian is the well known Landau–Zener (LZ) model [@lz32; @zn32].Equation (\[eq:nonlinLZ\]) represents an extension of the LZ model representing the dynamics of a BEC with an order parameter whose overlap with the two relevant single particle levels has the amplitude $a$ and $b$. Let us now consider the main results of such a model. In the limit of the small $c=0.1$, as shown inFigure \[fig:nLZ12\] left, the adiabatic energy levels (or the chemical potential $\mu$ in the language of our theoretical framework) is qualitatively similar to the linear case with the characteristic avoided crossing. When the nonlinearity exceeds the coupling strength $v$, the adiabatic energy levels are drastically changed with the development of the characteristic looped structures. Note that the shape of the loop for the chemical potential is not a swallowtail, the energy function has such a swallowtail structure. [While the modification of the energy level structure due to the strong nonlinearity is novel, what makes the looped energy structures particularly interesting from a physical point of view is the crucial implication of the looped structure for the transition probability between adiabatic energy levels.In the linear case ($c=0$), the transition probability for the LZ model is given by $r_0 = \exp (-\pi v^2/ 2 \alpha)$. In this case in the so-called adiabatic limit of $\alpha \rightarrow 0$, the transition probability vanishes. On the other hand in the nonlinear case, as shown in Figure \[fig:nLZ12\] right, for strong enough nonlinearity such that $c>v$ there is a finite transition probability even in the adiabatic limit. A simple explanation for this behavior is discernible from the looped energy structure in the third panel of Figure \[fig:nLZ12\] left—starting from the lower energy state initially, in the adiabatic limit there is very little tunneling as the system passes the point “X” and continues upwards along the loop to reach the final point “T”. At this point the system has to make a non-adiabatic transition to either the upper or lower level irrespective of how slow the sweep rate $\alpha$ is. This breakdown of adiabaticity is the key implication as well as an indication of the loop structure arising from the nonlinearity.]{} Having captured the basic setting in which loops arise via the nonlinear LZ model, we move on now to the system central to this review—BEC in an optical lattice. We are interested in the stationary states of the GP equation that satisfy Equation (\[GP2\]). In analogy to the LZ model, when the nonlinearity $g n$ is comparable to the applied lattice potential strength $V(x)$ we can expect interesting energy dispersions to arise. To be specific, henceforth we follow the treatment in [@machholm03; @diakonov02] and consider a one-dimensional potential of the form: $$\begin{aligned} V(x) = V_0 \cos(2\pi x/d), \label{eq:potl}\end{aligned}$$ and look for stationary states of the Bloch form, *i.e.*, $\psi(x) = e^{ik x} f(x)$ with $f(x+d) = f(x)$.Here $k$ denotes the quasimomentum of the condensate and is the proper quantum number in a periodic system. A key quantity of interest is the energy per unit volume given by $$\begin{aligned} \mathcal{E} = \frac{1}{d} \int_{-d/2}^{d/2} dx \, \left[ \frac{\hbar^2}{2m} \vert \nabla \psi \vert^2+ V_0 \cos(2\pi x/d) \vert \psi \vert^2 + \frac{g}{2} \vert \psi \vert^4 \right]. \label{eq:ergpervol}\end{aligned}$$ In the absence of interactions, *i.e.*, $g=0$, the energy per unit volume is arranged into the usual band structure which repeats after every reciprocal lattice momentum $k = 2 \pi/d$ and has gaps at the zone edges $k = r \pi/d$ with $r$ an odd integer in the extended zone representation of the bands (or at $k=\pm \pi/d$ in the reduced zone schemes). Equation (\[eq:Blochansatz\]) gives the full plane wave expansion for a Bloch state but a variational ansatz restricting to just three plane wave states, *i.e.*, $l_{\rm max} =1$ can already capture much of the physics as shown in [@machholm03]. Since the relative phases of the plane wave amplitudes $(a_0,a_1,a_{-1})$ do not change the energy, they can be chosen as real and using their normalization property restricted to the form: $$\begin{aligned} a_0 = \cos{\theta},\quad a_1 = \sin{\theta}\, \cos{\phi},\quad a_{-1} = \sin{\theta}\, \sin{\phi}\, . \label{eq:varansatz}\end{aligned}$$ The trial wavefunction (\[eq:Blochansatz\]) with the coefficients of the form (\[eq:varansatz\]) is inserted into the energy per unit volume expression (\[eq:ergpervol\]) and extremized with respect to the parameters $\theta$ and $\phi$. Also the recoil energy $E_0 = \hbar \pi^2/2md^2$ ($E_0=E_R$ in the notation of this review article) serves as a convenient unit for different energies in the problem. Let us first consider the situation close to the Brillouin zone edge $k= \pm \pi/d$. Using intuition from the nearly free particle models for periodic potentials, the ansatz (\[eq:varansatz\]) may further be simplified by taking $\phi = \{0 ,\pi/2\}$ [@diakonov02] giving: $$\begin{aligned} \psi(x) = \sqrt{n}\, e^{ikx} \left(\cos{\theta} + \sin{\theta}\, e^{-i2\pi x/d} \right) \label{eq:bandedgeansatz}\end{aligned}$$ with $n$ the average particle density. In fact it was shown in [@niu00] that such an ansatz can indeed be used to map the problem to that of the nonlinear LZ discussed earlier. At the zone edge $k=\pi/d$, upon extremization of energy density functional (\[eq:ergpervol\]) yields the solutions $\cos{2 \theta} = 0$ or $\sin{2 \theta} = V_0/(g n)$. When $g n < V_0$, the only possible solutions are $\theta = \pi/4$ or $\theta = 3 \pi/4$ representing the zone edge solutions of the lowest and first excited band and are qualitatively similar to their counterparts in the linear, $g=0$, limit. The condensate current density $J$, given by the derivative of energy $\mathcal{E}$ with respect to $k$, is zero for such solutions. When the interaction is strong enough such that $n g > V_0$, the other solution with $\theta = \sin^{-1}{\left[ V_0/(gn) \right]}/2$ is also allowed. Moreover this solution has no linear counterpart and has non-zero current even at the zone edge with: $$\begin{aligned} J = \pm \frac{\hbar \pi}{md}\sqrt{n^2-\frac{V_0^2}{g^2}}\, . \label{eq:loopcurrent}\end{aligned}$$ For values of $k$ away from the zone edge with $g>V_0/n$, the two solutions originally located at $\theta=\pi/4$ and $\sin^{-1}{\left[ V_0/(gn) \right]}/2$ approach each other and finally merge giving rise to the typical loop structure depicted in Figure \[fig:diakedgeloops\] top. Also note that the band-gap at $k=\pi/d$ in the weak lattice limit $V_0 \ll E_0$ is given by $V_0$ and the condition to have looped energy dispersion is that the interaction energy per particle $gn$ be greater than this band gap. Remarkably, the variational trial wavefunction (\[eq:bandedgeansatz\]), with $\theta$ values such that $\mathcal{E}$ is extremized, is identical with an exact analytical solution found in [@bronski01; @wu_st]. While this initially leads one to suspect that loops are somehow a restricted phenomenon subject to the availability of such exact solutions, further work in [@machholm03] dispelled this notion by demonstrating the possibility of loops even at the zone center, *i.e.*, with $k=0$, without any known exact solutions (see Figure \[fig:diakedgeloops\] bottom). Such solutions are also captured by the variational ansatz (\[eq:varansatz\]). In this context the key solution in the linear limit is the one corresponding to the top of the second band with $k=0$ and energy $\mathcal{E}/n =4 E_0$ per particle in the absence of interactions $g=0$ and $V_0 \ll E_0$. This solution with an equal admixture of $a_1$ and $a_{-1}$, and very small contribution from $a_0$ has $\theta = \pi/2$ and $\phi = 3 \pi/4$ (the solution with $\phi = \pi/4$ corresponds to the third band now) persists even when $g \neq 0$. However, it was shown in [@machholm03] that, for sufficiently large values of $g$, this solution becomes unstable and two new solutions emerge signaling the lower and upper point of the loop at $k=0$ depicted in Figure \[fig:diakedgeloops\] bottom. The condition for emergence of loops at the zone center is given by: $$\begin{aligned} g n > (16 E_0^2+V_0^2)-4E_0\, .\end{aligned}$$ In the weak lattice limit, the above condition simplifies to $g n > V_0^2/(8 E_0)$ which is precisely analogous to the condition that interactions exceed the band gap similar to the criterion to have loops at the zone edge. Figure \[fig:diakedgeloops\] bottom illustrates the zone-edge and zone-center loops computed using the ansatz (\[eq:varansatz\]). Furthermore it was found in [@machholm03] that the size of the loops, *i.e.*, their extent in $k$ increases monotonically as the ratio $gn/V_0$ is increased. Most interestingly in the limit of vanishing lattice potential $V_0 \rightarrow 0$, the swallowtail loops extend over the entire Brillouin zone and the upper edge of the swallowtail becomes degenerate with the states in the bands above. As we discuss more in detail in the next subsection, this behavior stems from the fact that the states in the upper edge of the swallowtail correspond to periodic soliton solutions of the GP equation in free space whose degeneracy is lifted by the application of a periodic potential. Finally, the swallowtail loop structures discussed so far may be intuitively viewed as generic features that arise in hysteretic systems as clarified in the work of [@mueller02]. Moreover the specific case of loops in the energy band structure of a BEC in an optical lattice can also be understood as a manifestation of superfluidity and extended to other analogous systems such as superfluids in annular rings that have been realized experimentally [@eckel14]. The key insight of [@mueller02] can be explained from Figure \[fig:muefig12\]. In Figure \[fig:muefig12\]a, the approximately sinusoidal energy dispersion (lowest band) of a quantum particle in a periodic potential is shown. The corresponding Bloch eigenstates at different quasimomentum $\hbar k$ have periodic probability distributions commensurate with applied lattice potential. When a uniform external force is applied, the particle adiabatically follows this energy dispersion and performs periodic Bloch oscillations. In contrast for a BEC with interactions obeying the GP equation, *i.e.*, a superfluid in a periodic potential, the interaction term tends to prefer uniform density distributions. Hence for strong enough interactions, as shown in Figure \[fig:muefig12\]b, the adiabatic band structure tends to be more similar to the quadratic free particle dispersion.This behavior has been referred to as the screening of the lattice potential by the superfluid in [@mueller02].As a result, the velocity of the superfluid does not go to zero at the zone edge leading to non-zero current as expressed by Equation (\[eq:loopcurrent\]). As shown in Figure \[fig:muefig12\]b, the velocity cannot increase indefinitely and the dispersion terminates (gray circle in Figure \[fig:muefig12\]b) when the velocity becomes comparable to the Landau critical velocity of the superfluid giving rise to the swallowtail structure. Clearly adiabatic evolution along such a trajectory has to breakdown beyond this terminal point and forcing the system across this point from left to right and back will not restore the initial state—which is a clear sign of hysteresis. Also looking at the vicinity of the zone edge, it is clear that there are two possible minima for the energy which naturally leads to a saddle point separating the two minima given by the upper branch of the loop (the dotted lines in Figure \[fig:muefig12\]b). The presence of multiple minima is a general characteristic of hysteretic systems. The number of minima for the superfluid in a periodic potential can be “controlled” by varying the quasimomentum $\hbar k$ or the interaction strength $g$ that sets the size of the swallowtail loop (remember that the swallowtail loops disappear if the $g$ is below some critical value). In the purview of catastrophe theory [@thombook], which is the study of singularities of gradient maps (any physical theory where a generating function is minimized to identify the stationary states of the system such as Fermat’s principle in optics or the extremization of the GP energy functional here are examples of gradient maps), the swallowtail structures of energy bands represents a cusp catastrophe where two control parameters ($g$ and $k$) may be varied to change the number of extrema of the energy by $2$. Swallowtail Loops Structures for Bosons in Optical Lattices {#subs32} ----------------------------------------------------------- Having provided a physical picture of swallowtail loops in the previous subsection we proceed now to a comprehensive overview of the main results. We will divide this subsection into two parts. In the first part we provide an account of the different results obtained regarding the occurrence of swallowtail loop structures for BECs in optical lattices and in the second part focus on the energetic and dynamical stability of the solutions. ### Occurrence of Loop Solutions The phenomenon of swallowtail loops in energy band structures arising from the GP equation were first investigated by Wu and Niu in [@niu00] for the nonlinear LZ introduced in detail in the previous subsection. In this work it was also pointed out that the nonlinear LZ model naturally arises near the zone edge in the dispersion for a BEC in a periodic optical lattice potential. The key results from this pioneering work are noted in Figure \[fig:nLZ12\]. Following this work, an exact solution for the GP equation in a periodic potential of the form $V(x) = -V_0 \mathrm{sn}^2(x,\kappa)$ with $\mathrm{sn}(x,\kappa)$ denoting the Jacobian elliptic sine function (with elliptic modulus $0\le \kappa\le 1$) was discovered in [@bronski01]. These solutions for $\kappa=0$ (Equation (2.1) of [@wu_st] or Equation (10) of [@bronski01] with elliptic modulus $k=0$ in their notation) take the following form in our notation: $$\begin{aligned} \psi_{\rm exact}(x) = \frac{\sqrt{c-v} + \sqrt{c+v}}{2\sqrt{c}} e^{i \pi x/d} + \frac{\sqrt{c-v} - \sqrt{c+v}}{2\sqrt{c}} e^{-i \pi x/d} \label{eq:carr}\end{aligned}$$ with $c = gn/(8 E_0)$ and $v = V_0/(8 E_0)$, which is of the Bloch wave form for the zone edge $k = k_L = \pi/d$. When $c>v$ (which is also the condition for loops to appear at the zone edge), Equation (\[eq:carr\]) gives the solution with finite current (see Equation (\[eq:loopcurrent\])). While on the one hand the exact solution lent more credibility [@wu_st] to the looped dispersions found from numerical solutions in [@niu00], they also led to the suspicion that such solutions may not be a general feature. This was quickly dispelled by a simple variational calculation for the loops at zone edge by Diakonov *et al.* [@diakonov02], followed by a more comprehensive analysis by demonstrating loops at the band center which we described in the previous subsection. also provided a thorough analysis of the width of the swallowtail loops, defined as the extent in quasimomentum space in an extended zone scheme, as the key parameters $gn$ and $V_0$ are varied for both the zone-edge loops (Figure \[fig:machloopwd12\]a) and zone-center loops (Figure \[fig:machloopwd12\]b). The results in Figure \[fig:machloopwd12\] were calculated by a numerical minimization of the GP energy functional (\[eq:GPerg\]) with the ansatz (\[eq:Blochansatz\]). An unexpected feature of the results in Figure \[fig:machloopwd12\] is that the width of the swallowtail loops remain non-zero even in the limit $V_0 \rightarrow 0$. In this limit the swallowtail solutions correspond to periodic soliton solutions of the GP equation. The zone edge solution with $k=\pi/d$ represents an equally spaced array of dark solitons (in one dimension, a dark soliton’s wave function vanishes at some point in space whereas for a gray soliton there is a density dip of non-zero value at some point in space) with one soliton per lattice spacing $d$. The wave number of the solution $k = \pi/d$ can then be justified as the average phase change per unit length giving the right change in phase of $\pi$ across the dark solitons in each period. When $V_0 = 0$, soliton array solutions with density dips located at $x = r d$ and $x=(r+1/2)d$ (with $r$ integer) are degenerate and correspond to the highest energy state in the first band and the one immediately above in the second band, respectively. The lattice potential breaks this degeneracy as the solution with soliton centers at potential maxima $x=r d$ have lower energy giving rise to the energy gap at the center of the swallowtail loop. The loops at $k=0$ can also be understood with an analogous argument. The solutions with $k \neq 0$ correspond to arrays of gray solitons. The phase change across a gray soliton is less than $\pi$ and they move with some finite velocity $v_{\mathrm{soliton}}$ in the absence of the potential. In order to create a stationary state from such solutions at finite $V_0$, one has to imagine boosting the condensate velocity by $-v_{\mathrm{\rm soliton}}$ giving a spatial dependent phase to the condensate wave function. The wave vector corresponding to such solutions now depends both on the density and the phase shift implied by the finite velocity boost. Moving away from $k =0$ or $\pi/d$, the minimum density and $v_{\mathrm{soliton}}$ increases eventually going to zero and the sound speed $(gn/m)^{1/2}$ respectively leading to the loop branch merging with the free particle dispersion as shown in Figure \[fig:diakedgeloops\] bottom. This manner of understanding the emergence of swallowtail loop structures from the soliton solutions provides a complementary physical picture of the phenomenon. The work by Mueller in [@mueller02] provides a general way to understand swallowtail loops from the point of view of hysteresis and superfluid response. A corollary of such an approach is that it is possible to extend the phenomena of loops to systems beyond the standard system in this review (BEC in a periodic potential). Amongst the various examples discussed in [@mueller02], the case of a BEC in an annular trap is of particular interest in the context of the experiment [@eckel14]. The Hamiltonian in the rotating frame describing a bosonic superfluid (mass $m$) in a 1D ring of length $L$, rotating at a frequency $\Omega$ is $$\begin{aligned} \frac{H}{\hbar^2/2mL^2} = \sum_{j} (2 \pi j + \Phi)^2 c_j^{\dagger} c_j + \frac{\tilde{g}}{2} \sum_{j+k=l+m} c_j^{\dagger}c_k^{\dagger}c_l c_m + \lambda \sum_{j} \left( c_j^{\dagger} c_{j-1}+c_j^{\dagger}c_{j+1} \right)\, . \label{eq:SFringham}\end{aligned}$$ In the above, $c_j$ stands for the annihilation operator for bosons with angular momentum $j \hbar$ in a quantized picture or the amplitude of occupation of the same mode within a mean-field GP-like picture, $\Phi = 2 mL^2 \Omega/\hbar$ is the rotation speed in a dimensionless form, $\tilde{g} = 4 \pi a_s L/d_{\perp}^2$ is the effective interaction for a trap perpendicular to the ring with harmonic oscillator length $d_{\perp}$, and $\lambda$ is an impurity term that breaks the rotational symmetry coupling different angular momenta. This can arise naturally due to imperfections in the container or be generated, for instance, by applying a laser potential externally. The key point is that there is a one-to-one correspondence between the Hamiltonian (\[eq:SFringham\]) and the GP energy functional (\[eq:GPerg\]) after substituting the Bloch ansatz (\[eq:Blochansatz\]) with $\Phi$ playing the role of quasi-wave number $k$ and $\lambda$ playing the role of the periodic potential strength leading to swallowtail loops as shown in Figure 11 in [@mueller02]. We will revisit Equation (\[eq:SFringham\]) in Section \[subsec:nonlinexp\]. [Seaman *et al.* [@seaman05a] and Dong and Wu [@dong07]]{} provided an interesting insight into the swallowtail loop structures for both repulsively ($g>0$) and attractively ($g<0$) interacting BECs for the special case of a Kronig–Penney periodic potential which is of the form, $$\begin{aligned} V(x) = V_0 \sum_{j=-\infty}^{\infty} \delta( x - j d). \label{eq:kronigpenney}\end{aligned}$$ The Bloch states in such a potential can be solved analytically. For the repulsive interactions case, the results in [@seaman05a] agreed qualitatively with the earlier numerical and approximate calculations but had unique features such as the fact that the critical interaction strength to have loops $g n > 2 V_0$ for *all* bands of the energy spectrum unlike the sinusoidal band case discussed in [@machholm03]. In the case of attractive interactions with $g<0$, they found the loop structures occurred in the upper branch at the band gaps, starting from the second band as shown in Figure \[fig:attloop\]. Moreover for the strongly attractive case $gn = -10 E_0$ shown in Figure \[fig:attloop\] the loop in the second band spans the entire Brillouin zone and splits from the original band. ### Stability of Loop Solutions\[sec:loopstability\] In the discussion of swallowtail loops so far, we have completely ignored the stability properties of the solutions. In what follows, we discuss results regarding the energetic and dynamical stability of steady state solutions of the Bloch form (of which swallowtail loops are special cases) for BECs in periodic potentials. The stability of the solutions we find by extremizing the GP energy functionals (\[eq:GPerg\]) are crucial as they will provide us clues as to whether in an experiment the system can reach such equilibrium solutions and if they do how long can they be stable. As a part of the the theoretical framework Section \[subsec:thframeergstab\], we provided the conditions for energetic stability but a physical picture of the two kinds of stability as exemplified in Figure \[fig:niustab1\]a is helpful—for energetic stability the equilibrium solution has to be a local minimum of the energy functional (\[eq:GPerg\]) whereas for dynamical stability perturbations about the equilibrium state should not grow with time when evolved according to the time-dependent GP equation (\[GP1\]). The stability of Bloch states in the lowest band excluding loops was discussed by Wu and Niu in [@wu01] and [@wu03]. Figure \[fig:niustab1\]b represents the results from a numerical calculation mapping out the stability of Bloch states with quasimomentum $\hbar k$ under perturbations of the Bloch wave form with quasimomentum $\hbar q$ for different values of the potential $v = V_0/(8 E_0)$ and nonlinearity $c = gn/(8E_0)$. Let us denote the stability matrix, Equation (\[eq:ergstabcondn\]), for a state with quasimomentum $\hbar k$ under a perturbation of quasimomentum $\hbar q$ as $M_k(q)$. For the special case of a free BEC with no lattice, *i.e.*, $v=0$, the eigenvalues of the matrix can be computed analytically and the requirement for positivity of eigenvalues leads to the well-known Landau criterion given by $$\begin{aligned} \vert k \vert \geq \sqrt{q^2/4+c} \label{eq:freeLandaucrit}.\end{aligned}$$ The shaded light and dark regions in Figure \[fig:niustab1\]b represent regions of energetic instability with $M_k(q)$ having negative eigenvalues. In the limit of small $v$ the equality in the expression (\[eq:freeLandaucrit\]) accurately reproduces the energetic stability region shown by triangles in the plot. Another key feature to note in Figure \[fig:niustab1\]b is that even at $v$ comparable to $c$, as the nonlinearity $c$ is increased, the BEC is energetically stable over an increasing area in the $k$-$q$ space, which can be anticipated from expression (\[eq:freeLandaucrit\]). In Figure \[fig:niustab1\]b, the system is dynamically unstable in the dark shaded regions. The first thing to notice is that energetic instability is a pre-requisite for dynamical instability and this can also be shown in general (see appendix of [@wu03]). Further insight into dynamical stability can be obtained by considering the structure of the eigenvalues of the matrix $\sigma_z M_k(q)$ since the requirement for dynamical stability is to have real eigenvalues for this matrix. The states represented by the eigenvectors of $\sigma_z M_k(q)$ can also be viewed as quasiparticle excitations in the BEC, *i.e.*, phonons [@wu03], with the positive eigenvalues giving the phonon spectrum. In general the eigenvalues of $\sigma_z M_k(q)$ can be complex but always occur in complex conjugate pairs [@machholm03; @wu03] owing to the real nature of the matrix in momentum representation. In the case of $v=0$, the eigenvalues of $\sigma_z M_k(q)$ are always real and given by $\epsilon_{\pm}(q) = kq \pm \sqrt{q^2 c + q^4/4}$ where $k$ and $q$ are measured in units of $4 \pi/d$. This implies that in free space BEC, flows are always dynamically stable. However, Figure \[fig:niustab1\]b shows that situations change significantly when the lattice potential is introduced as also evidenced in experiments [@burger01; @burger01comment; @burger01comment1]. In general for all parameter regimes there is a critical wave number $k_d$ beyond which Bloch waves are dynamically unstable. At $k=k_d$, dynamical stability always sets in for $q=\pi/d$, which interestingly also corresponds to a period doubling revealing a link further explored in Section \[sec:multiperiod\]. At the point where dynamical instability sets in, the eigenvalues of $\sigma_z M_k(q)$ change character from real to pairs of complex conjugates, *i.e.*, with equal real parts. Hence as explained in [@wu01; @wu03; @machholm03], dynamical instability can be viewed as arising from a lattice induced resonance between a pair of excitations that are degenerate in the $v=0$ limit. Thus when the instability sets in, two phonons with the sum of their momenta given by the primitive reciprocal lattice vectors $\pm G= \pm 2\pi/d$ are created from zero energy, *i.e.*, they satisfy $$\begin{aligned} \epsilon_{+}(q) + \epsilon_{+}(2 \pi/d-q) = 0\, . \label{eq:stabcondnfig}\end{aligned}$$ This can be used to clearly justify the observation (valid at small lattice depths $V_0$) that instability at the critical wave number $k_d$ always sets in with $q=G/2=\pi/d$ and the critical vector satisfies $\vert k_d \vert = (\pi/d)(gn/2E_0+1/4)^{1/2}$ agreeing with the numerical results in Figure \[fig:niustab1\]b. In the stability analysis of [@machholm03], in addition to standard Bloch states, also the ones corresponding to the loop solutions (lower branch) were considered. In Figure \[fig:pethergdynstab\] the results of this analysis is shown by plotting the largest Bloch wave vector $k$ for which the states are energetically stable as a function of $ng$ and $V_0$. States with $0\leq k \leq \pi/d$ correspond to states with the lowest energy and $k>\pi/d$ represents lower edge-loop states. In general it was found that the range of quasimomentum values at which the system is energetically stable increases with the interaction strength $gn$. They also found that the wave vector of the tip of the swallowtail sets a natural limit for the wave vector at which instability sets in for parameter regimes where swallowtail loops occur. Moreover they found, in agreement with [@wu01], with increasing $k$ the long wavelength perturbations with $q \rightarrow 0$ become energetically unstable first. Hence a hydrodynamic description can be constructed [@mamaladze66; @hakim97; @vc] and it also gives analytical predictions for the instability contour as a function of $V_0$ and $g n$ for the zone edge with $k=\pi/d$ which compares favorably with the numerical calculations [@vc]. Note that as expected the states on the upper edge of the loop are always energetically and dynamically unstable. Finally, the discussion so far was limited to only the linear stability of equilibrium solutions but the full response of the solutions of time dependent GP equation (\[GP1\]) to perturbations was also studied numerically in [@seaman05a]. Here the stable lifetime of a given initial equilibrium state was defined as the time taken for the variance of the Fourier spectrum of the time-dependent order parameter from the initial Fourier spectrum (normalized to the initial spectrum) to exceed the value $1/2$, *i.e.*, the time at which the order parameter becomes very different from the initial solution and taken as an indicative time for onset of dynamical instability. They found that for weakly attractive condensates the zero quasimomentum Bloch state in the first band is long-lived under white noise perturbations but highly unstable for time periodic perturbations. For both weak and strong attractive interactions the higher band Bloch states are unstable but the first band Bloch states with non-zero quasimomentum can be stable even to harmonic perturbations owing to their negative effective mass (defined as the inverse of the Band curvature). For weak repulsive interactions, Bloch states in the lowest band are stable as long as the quasi-wave number $k<\pi/d$, as at larger quasi-wave numbers the effective mass becomes negative, for the particular choice of Kronig–Penney potential (\[eq:kronigpenney\]) used in [@seaman05a]. However, in agreement with the linear stability analysis, as the interaction strength grows, larger parts of the energy band including the lower branch of the loops are stable. In a later publication the same authors [@danshita07; @sc07] showed that there is always a small part of the loop in the repulsive case that has negative effective mass but this area’s extent monotonically decreases as $gn$ is increased. [A clear discussion of the dynamical stability of attractive BECs in an optical lattice is provided in [@barontini07].A recent review [@zhu15] also provides detailed treatment of stability of BECs in optical lattices.]{} Experimental Realization {#subsec:nonlinexp} ------------------------ In the past decade and half there has been tremendous progress in the field of ultracold atoms in optical lattices [@bloch_review; @lattice] with a range of experiments tackling interesting many-body physics. In this light it must be said that specific experiments dealing with nonlinear energy dispersions in optical lattices have been few and far in between. The earliest relevant experiment concerns the instability of superfluids in optical lattices by the group at LENS, Florence reported in Burger [*et al.*]{} [@burger01; @burger01comment; @burger01comment1]. In this experiment a cigar shaped quasi-1D BEC was localized in an harmonic trap and an optical lattice potential was also turned on. Following this, the center of the trap was suddenly shifted, corresponding to a sudden change of the quasimomentum in our language. For small shifts $\Delta x$ the dynamics of the BEC was coherent but for shifts greater than a critical value $\Delta x > \Delta x_c$, the oscillations are disrupted and the dynamics became dissipative. The authors of the experiment attributed this to the energetic Landau instability of the condensate. But theoretical work from Wu and Niu [@burger01comment; @burger01comment1; @wu03] showed that it may be more appropriate to associate this behavior with dynamical instability especially considering that the experimental parameters fall in a regime ($c\sim0.02$, $v\sim0.2$) where dynamical instability is rampant (see Figure \[fig:niustab1\]b) and the critical displacement $\Delta x_c$ increases with decrease of lattice depth in the experiment (the energetic stability is essentially independent of lattice depth in the linearized stability treatment). A follow-up comment from the authors of the experiment [@burger01comment; @burger01comment1] was essentially inconclusive but hinted that even the GP equation may not be a valid description for some of their experimental results and beyond mean-field effects may have to be included. Following this, in [@modugno04] a thorough theoretical treatment of the problem including 3-dimensional GP equations and effective 1-dimensional GP equations taking into account the transverse degrees of freedom was undertaken. This theoretical treatment adapted to the experiment [@burger01] clearly showed that the onset of instability observed in the experiment was due to dynamical instability. There are two main reasons that led to some uncertainty regarding the conclusions in the experiments [@burger01]—the inability to distinguish experimentally between dynamical and energetic stability as both processes manifest as an enhanced loss of atom number from the condensate and the inability to accurately set the initial quasimomentum of the condensate as a mean displacement from the harmonic trap center can in general lead to a mixture of quasimomenta and band eigenstates. In the follow-up experiment from the LENS group [@dynstabexp; @dynergstabexp] both these issues were succesfully resolved. In [@dynstabexp], a moving optical lattice was implemented by frequency detuning the two laser beams creating the lattice. The ground state in a moving lattice is simply a state with a finite quasimomentum. By varying the amount of detuning, they were able to load the BEC adiabatically into states with a given quasimomentum both in the ground and excited bands. A subsequent measurement of the loss rates as a function of quasimomentum revealed a threshold for the onset of dynamical instability in very good agreement with the theoretical expectation for the lattice depth used. In [@dynergstabexp] they were able to also distinguish between the onset of energetic and dynamical instability by an ingenious use of a radio-frequency (RF) shield to selectively control the thermal fraction of the atomic cloud. The presence of the thermal cloud can effectively trigger energetic instability, which has generally a lower threshold in quasimomentum, by providing a dissipation channel. Hence when a large thermal fraction is present, the onset of dynamical instability is marred by energetic instability. On the other hand when the RF shield is turned on to remove the thermal fraction and experiments are performed with nearly pure condensate, the onset of dynamical instability stands out via a dramatic loss of atom number. It is important to emphasize at this point that experiments such as [@burger01] are performed at small lattice depths $v$. In this regime, as evident from Figure \[fig:niustab1\]b, there are significant regions of quasimomentum space that are energetically unstable but dynamically stable. On the other hand at large lattice depths (see Figure 1 of [@dynergstabexp]), the region of quasimomentum space that exhibits only energetic instability is tiny. Thus in this so called tight-binding lattice limit, there was unambiguous agreement between experiments [@dynstabexpdeep] and the theoretical [@wu03; @modinstabsmerzi] and numerical [@dynstabnumdeep; @dynstabnumdeep1] treatments that predicted dynamical instability.It was also shown in [@modinstabsmerzi] that the dynamical instability in this regime can also be viewed as a kind of modulational instability which is a general feature of nonlinear wave equations where a small perturbations of a carrier wave can exponentially grow as a result of interplay of dispersion and nonlinearity. In [@dynstabMISF], the behaviour of the critical quasimomentum upto which superfluidity persisits across the superfluid-Mott insulator (SF-MI) quantum phase transition was studied. Within a mean-field GP picuture, as we have discussed in Section \[sec:loopstability\], the stability of the superfluid is in general enhanced with increasing interaction and lattice depth. On the other hand, within the full quantum model, there is a critical interaction strength to tunneling ratio beyond which the system is no longer superfluid and goes into the Mott insulating phase where the critical quasimomentum is trivially equal to $0$. This study maps the behaviour of the critical quasimomentum as it goes from a finite value in the SF phase to zero in the MI phase giving an accurate determination of the phase boundary. In [@ferris08] beyond dynamical instability at the zone edge is investigated experimentally and theoretically within the truncated Wigner approach which can account for beyond GP effects including the thermal depletion of the condensate. Finally, in recent experiments [@dynstabSOC] dynamical instability of spin-orbit coupled (SOC) BECs in moving optical lattices was investigated and a manifestation of the breakdown of Galilean invariance predicted for SOC systems was evidenced by the difference in the strength of the dynamical instability (measured by atom loss rate) depending upon the direction of motion of the lattice. In the seminal experiment of the Bloch group [@chen11], some aspects of the nonlinear LZ tunneling phenomena originally considered in the theoretical work of Wu and Niu [@niu00] was explored.The experimental system consisted of an array of tubes of BECs in a 2D optical lattice potential.A superlattice potential along $x$ direction allows pair-wise coupling between tubes giving many copies of coupled double wells. In the experiment they effectively realize the nonlinear LZ energy function of the the form: $$\begin{aligned} E[\psi_R,\psi_L] \approx \frac{\Delta}{2} \left(\vert \psi_R \vert^2 -\vert\psi_L \vert^2 \right) - J (\psi_R^{*}\psi_L + \mbox{c.c.}) + \frac{U}{2} \langle \delta \hat{n}^2\rangle \left(\vert\psi_R \vert^4 + \vert \psi_L \vert^4 \right) \label{eq:blochexperg},\end{aligned}$$ where $\psi_L$ and $\psi_R$ represent the amplitude to occupy either the left or the right tube. $\Delta$, the energy detuning between the tubes, and $J$, the tunnel coupling between the tubes can be controlled by varying the relative phase and lattice depth respectively of the superlattice potential along $x$ direction. The experimental protocol consists of preparing all the atoms initially in the left tube with the initial detuning $\Delta_i$ either chosen to be negative (ground state) or positive (excited state) and sweeping the detuning at the linear rate $\alpha$ and finally measuring the number of atoms in the left and right tube. As already anticipated by Wu and Niu [@niu00], the model \[Equation (\[eq:blochexperg\])\] is exactly the same as the one introduced in Equation (\[eq:nonlinLZ\]) except for the interaction term’s sign is switched. As a result once the ratio of interaction to tunneling $\eta = U \langle \delta \hat{n}^2\rangle / J$ ($c/v$ in [@niu00]) is large, there is a loop in the upper branch as shown in lower panel of Figure \[fig:blochexp35\] left. In the experiment, for sweeps along the ground state branch (gray dots in Figure  \[fig:blochexp35\] left) there is no adiabaticity breakdown observed. In the sweep starting with the excited state (left tube at higher energy), there is a complete breakdown of adiabaticity even for reasonably small sweep rates $\alpha$ (red dots in Figure \[fig:blochexp35\] left). Moreover, for small sweep rates the transfer efficiency, given by number of atoms $n_R$ in the final state, decreases with decreasing sweep rate completely opposing the expected LZ behavior. The presence of the loop in the upper branch contributes to this adiabaticity breakdown for small sweep rates, as the atoms follow the upper branch and are *self-trapped* in the middle branch of the loop, (see lower panel of Figure \[fig:blochexp35\] left) which is a local maxima. They finally make a diabatic transition to the adiabatic ground state at the loop edge leading to the sharp breakdown of transfer efficiency near zero detuning seen in Figure \[fig:blochexp35\] left upper panel. A confirmation of the effect of loops on the adiabaticity breakdown is provided by the non-monotonic behavior of the transfer efficiency (number of atoms in the initially empty right tube at the end) at a given sweep rate and tunnel coupling as a function of the $z$-lattice depth shown in top panel of Figure \[fig:blochexp35\] right.At small $z$-lattice depths, an increase in lattice depth effectively increases the on-site interaction $U \langle \delta \hat{n}^2\rangle$, leading to larger loop sizes which leads to lower transfer efficiency but eventually beyond a certain lattice depth the suppression of on-site fluctuations $\langle \delta \hat{n}^2 \rangle$ dominates leading to a decrease in the effective interaction and increase of transfer efficiency restoring standard LZ behavior. Moreover as shown in the lower panel in Figure \[fig:blochexp35\] right, the experimentally determined minimum transfer efficiency agrees with a theoretical calculation for the position of the maximum loops size as a function of the $z$-lattice depth and tunnel coupling. In the experiment by Eckel [*et al.*]{} [@eckel14], a physical situation approximately corresponding to the Hamiltonian in Equation (\[eq:SFringham\]) was realized for a BEC of $^{23}$Na atoms confined in a ring shaped trap [@amico05]. The goal of this experiment was to observe hysteresis between quantized states of circulation of the superfluid BEC caused by the presence of swallowtail loops in the energy landscape of such a system [@mueller02; @baharian13]. In the experiment [@eckel14], they were concerned with the quantized circulation states with winding numbers $n=0$ and $n=1$ with frequency $n$ times the rotational quanta $\Omega_0 = \hbar/mR^2$ and drive transitions between these states by tuning the relative angular velocity between the trap and the superfluid $\Omega$ which can be controlled by applying a repulsive rotating laser potential. In the absence of coupling between the different circulation states, the energy landscape of the interacting superfluid forms a swallowtail loop shape as a function of the relative angular velocity $\Omega$. At a fixed value of $\Omega$ in the swallowtail region, $n=0$ (red line in Figure \[fig:eckelexp13\] left) and $n=1$ (blue line in Figure \[fig:eckelexp13\] left) states form the minima of a double-well energy landscape with a barrier state (green dashed line) separating them. If the system begins in the $n=0$ state and its angular velocity is increased, the flow will be stable as long as $\Omega < \Omega_{c+}$ when it reaches the edge of the swallowtail and after this it will make a transition to the lower energy $n=1$ state. Beginning with the $n=1$ state, a similar stable flow can exist as long as $\Omega > \Omega_{c-}$, leading to the hysteresis loop shown in lower panel of Figure \[fig:eckelexp13\] left. The rate at which the repulsive potential created using a blue detuned laser is rotated, controls the flow velocity $\Omega$ and the strength of the potential $U$ controls and drives transitions (via phase slips) [@wright13] between different circulation states. Comparing to the Hamiltonian (\[eq:SFringham\]) the repulsive potential actuates two of the terms namely the rotation frequency $\Phi = \Omega/\Omega_0$ and the “impurity” term $\lambda$ whose strength is controlled by $U$. In order to observe this hysteresis loop in the experiment, the BEC is prepared initially in either the $n=0$ or $n=1$ state in the trap and then this repulsive trap potential with a chosen strength $U_2$ and different rotation velocity $\Omega_2$ is applied for a fixed time of $2$ s, followed by a time of flight image to determine the final rotational state $n$. Due to the swallowtail loop structure, at a given strength $U_2$, as shown in upper panel of Figure \[fig:eckelexp13\] right, the transition from $0$ to $1$ and *vice-versa* clearly happen at different angular momenta $\Omega_{c\pm}$. Moreover as the strength of the applied potential $U_2$ is increased, $\Omega_{c \pm} \rightarrow \Omega_0/2$ and the swallowtail loop size decreases and eventually vanishes. The size of the loop as a function of $U_2$ was determined from the experiment and is plotted in the bottom panel of Figure \[fig:eckelexp13\] right. The discrepancy from theoretical calculations that include relaxation effects required to accomplish the non-adiabatic transitions indicate that further work may be required to understand some quantitative aspects of the experiment. A detailed treatment of modeling the relevant excitations leading to the dissipation by vortex-antivortex pairs was already provided by the authors in [@eckel14]. Other Extensions ---------------- Owing to the general nature of swallowtail shaped dispersions resulting from the interplay of atom-atom interactions and periodicity, they have been predicted to occur in a variety of systems different from the setting mainly considered in this review—namely that of BECs in an optical lattice with effectively 1D dynamics. In this subsection we catalogue these developments without going into the details owing to the restricted scope of the review. Lin [*et al.*]{} [@dipswallow] study BECs with magnetic dipole-dipole interactions in optical lattices. In this case the effective atom-atom interaction can be controlled by changing the alignment of the atomic dipoles to the optical lattice axis by applying magnetic fields. At strong enough interaction, they observe swallowtail loops whose sizes and stability may be controlled by modifying the magnetic dipole orientations. In [@cavloops], Venkatesh [*et al.*]{} study band-structures of atoms confined in optical lattices formed inside optical cavities that are continuously driven by an external laser. In the limit of very dilute gases, $s$-wave contact interactions do not play a role but the cavity-induced atomic interactions in the strong coupling regime of cavity quantum electrodynamics (QED) can lead to swallowtail loop structures in the atomic bands. Moreover this also corresponds to bistable solutions for the steady state photon number in the cavity. In the work of Watanabe and colleagues [@swallowtail], swallowtail band structures of superfluid Fermi gases in optical lattices in the BEC-BCS crossover are investigated. They find that typically the width of the swallowtail is largest at unitarity. In addition, they find that the microscopic mechanism of the emergence of the swallowtail in the BCS side of the crossover is very different in nature from that of the BEC case: a narrow band in the quasiparticle energy spectrum close to the chemical potential plays a crucial role for the appearance of the swallowtail in the BCS side. It is also pointed out that, as a consequence, the incompressibility experiences a profound dip. Chen and Wu extend the study of interplay between interactions and band structures of superfluid systems in optical lattices to two dimensions in [@diracpts] considering BECs in honeycomb shaped optical lattices. Such optical lattices serve as analogues to the structure of graphene and support Dirac points in their band close to which the energy dispersion is linear in 2D having characteristic conical shape. In [@diracpts] the authors show that even for arbitrarily small interaction strength, the Dirac point is extended into a closed curve and a tube like structure, a 2D version of the 1D swallowtail loop, arises around the original Dirac point. Moreover in work that followed closely thereafter Hui [*et al.*]{} showed that even in the case of 2D optical lattice with double-well superlattice like geometry along one direction, swallowtail loop structures emerge for any interaction strength [@twodloop]. Thus, the possibility of having swallowtail loops structures or analogues thereof for arbitrary small interactions seems to be a more ubiquitous feature in 2D as opposed to 1D where a the nonlinearity given by interactions has to be comparable to the lattice potential. [In [@impurity], a BEC trapped in a double-well potential with an additional degree of freedom given by a single bosonic impurity atom that interacts with the condensate is considered. In this setup, as the impurity-BEC interaction strength is tuned above tunneling energy of the bosons, swallowtail loops appear in the adiabatic energy dispersions as a function of the tilt of one of the wells relative to the other. The relation between swallowtail loops and self-trapping of the condensate in one well or the other as well as relation to the Dicke model are explored.]{} Future Prospects ---------------- Presently, experiments showing direct evidence for looped band structures for BECs in optical lattices have not been performed. One of the impediments preventing the experimental observation of loops in optical lattices is the requirement of large atom-atom interactions so that a large part of the loop solution is energetically stable [@danshita07; @sc07; @seaman05a]. The simplest way to ascertain breakdown of adiabaticity caused by the loop is to study Bloch oscillations in optical lattices. In this regard it is quite important to control and characterize other sources of loss of adiabaticity such as LZ tunneling to higher bands and distinguish them from the effect of the loops. Clearly for this a control of atom-atom interaction from very small to large enough to obtain loops is required. In this context, some recent experiments in the group of Nägerl with the ability to tune interactions using Feshbach resonances is promising [@haller10; @meinert14]. Also, the extended theoretical schemes for 2D optical lattices [@diracpts; @twodloop] may be easier to implement in an experiment as they do not require large atom-atom interactions to have looped energy dispersions. On the theoretical side, a clear understanding of the quantum mechanical underpinnings of the mean-field loops is already available for the case of repulsive and attractive interacting BECs in double-well potentials [@quantloops; @chen11]. An extension of such a study to optical lattices in two dimensions or for fermionic atoms can be interesting. Another interesting theoretical consideration would be examine the idea of shortcuts to adiabaticity [@shortcuts] that has received a lot of attention of late to systems where the underlying evolution equation is not linear and understand if one may conceive of protocols where the loss of adiabaticity predicted due to the loops could be avoided. Multiple Period States in Cold Atomic Gases in Optical lattices\[sec:multiperiod\] ================================================================================== Density structures and patterns caused by the interplay of competing effects are ubiquitous in nature. In the case of superfluids flowing in a periodic potential, non-trivial density patterns can emerge due to the interplay of spatial periodicity imposed by the external potential and the nonlinearity due to the superfluid order parameter. According to the conventional wisdom of the Bloch theorem, in the linear system described by the Schrödinger equation, the density pattern of the stationary solution in a lattice is periodic with periodicity coinciding with that of the lattice potential. However, nonlinearity can cause non-trivial density patterns with different periodicity. For BECs in a periodic potential, it has been found that nonlinearity of the interaction term can cause the appearance of stationary states whose period does not coincide with that of the lattice; instead, a multiple of it [@machholm04; @pethick_smith]. Such states are called multiple (or $n$-tuple) period states. In this section, we discuss multiple period states of superfluid atomic gases in optical lattices. In the Section \[subsec:basics\], we provide the basic physical idea of the emergence of the multiple period states due to nonlinearity. To provide a physical picture concisely, we take the BEC case as an simple example. In the Section \[subsec:pdbec\], we present an account of existing results of the multiple period states in BECs. In the Section \[subsec:pdfermi\], we present some theoretical results of the multiple period states in superfluid Fermi gases along the BCS-BEC crossover focusing on their unique features in contrast to the multiple period states in BECs. Basic Physical Idea: A Simple Explanation of the Emergence of Multiple Period States by a Discrete Model \[subsec:basics\] -------------------------------------------------------------------------------------------------------------------------- The emergence of the multiple period states in BECs can be explained by the discrete model (Equation (\[hamilt\])) in a simple manner [@machholm04]. In the following exposition, we follow the discussion given in the above cited paper. For clarity, here we focus on the period-2 states: states whose period is equal to twice of the lattice constant $d$. The stationary states with a fixed total number of particles $N$ can be obtained by the variation of $H' \equiv H - \mu N$ with respect to the amplitude $\psi_j^*$ at site $j$, where $\mu$ is the chemical potential and $N= \sum_j |\psi_j|^2$, $$\frac{\delta H'}{\delta \psi_j^*} = -K (\psi_{j+1} + \psi_{j-1}) + U |\psi_j|^2 \psi_j - \mu \psi_j = 0\, . \label{eq:dhdpsi}$$ We then separate from $\psi_j$ a plane wave part at site $j$, $e^{ikjd}$, as $\psi_j = e^{ikjd} g_j$ with $\hbar k$ being quasimomentum of the bulk superflow flowing in the same direction of the periodic potential and $g_j$ being the complex amplitude at site $j$. Equation (\[eq:dhdpsi\]) becomes $$-K (g_{j+1} e^{ikd} + g_{j-1} e^{-ikd}) + U |g_j|^2 g_j - \mu g_j = 0\, . \label{eq:statsolj}$$ Due to the boundary conditions of the period-2 states, we have $g_0 = g_2$ and $g_1 = g_3$. We solve combined two Equations (\[eq:statsolj\]) for $j=1$ and $2$ with these boundary conditions. Subtracting these two equations, we obtain $$-2K \cos{kd} \left(\frac{|g_2|}{|g_1|}\, e^{i(\phi_2-\phi_1)} - \frac{|g_1|}{|g_2|}\, e^{i(\phi_1-\phi_2)}\right) + U (|g_1|^2 - |g_2|^2) = 0\, , \label{eq:statsolper2}$$ with $g_j \equiv |g_j| e^{\phi_j}$. For the linear case ($U=0$), we can readily see that $|g_1| \ne |g_2|$ cannot satisfy Equation (\[eq:statsolper2\]) except at $kd = \pi/2$, which corresponds to a trivial solution of $g_1=g_2=0$. On the other hand, solutions with $|g_1| = |g_2|$ exist provided $\phi_2-\phi_1 = 0$ (modulus of $2\pi$): thus these solutions are normal period-1 states. For the nonlinear case ($U \neq 0$), nonzero contribution from the kinetic energy part (the first term in the left-hand side of Equation (\[eq:statsolper2\])) can be compensated by that from the interaction energy part (the second term in the left-hand side of Equation (\[eq:statsolper2\])) so that this equation can be satisfied. Therefore, the emergence of the period-2 states is a purely nonlinear phenomenon. Since the second term in the left-hand side is real, the phase difference $\phi_2 - \phi_1$ should be $0$ or $\pi$, namely: $$\pm 2 K \cos{kd}\, \left(\frac{|g_2|}{|g_1|} - \frac{|g_1|}{|g_2|}\right) = U (|g_1|^2 - |g_2|^2)\, .$$ Thus we obtain $$\pm \frac{\cos{kd}}{\frac{U\nu}{2K}} = \frac{|g_1| |g_2|}{\nu}\, , \label{eq:statsolfin}$$ where the filling factor $\nu \equiv (|g_1|^2 + |g_2|^2)/2$ is the average number of particles per cell (in the present case of the period-2 states, the cell consists of two lattice sites). Since the right-hand side of Equation (\[eq:statsolfin\]) takes $0 \le |g_1| |g_2|/\nu \le 1$, solutions with period $2d$ exist when $|\cos{kd}| \le U\nu/2K$ [@machholm04]. In Figure \[fig:pdtb\], we show the energy bands of the period-1 and period-2 states for $U\nu/2K=1/2$, $1$, and $2$. Note that, in the case of the Figure \[fig:pdtb\]a for $U\nu/2K=1/2$, the period-2 states exist in the limited region of $1/6 \le kd/2\pi \le 1/3$. At infinitesimally small $U\nu/2K$, the period-2 states exist only at $kd/2\pi=1/4$. As $U\nu/2K$ increases, the region in which the period-2 states exist increases and it finally extends over the whole Brillouin zone for $U\nu/2K \ge 1$ (see Figures \[fig:pdtb\]b and \[fig:pdtb\]c). Note that there is another class of period-2 states called the phase states [@machholm04]. From Equation (\[eq:statsolper2\]), we see that, at $|k|d = \pi/2$, $|g_1|=|g_2|$ is a solution for arbitrary phase difference $\phi_2-\phi_1$.The periodicity of the density distribution and the energy of the phase states are the same as the normal Bloch state at $|k|d = \pi/2$, but only the phase profile has the period $2d$ [@machholm03]. Multiple Period States in BECs\[subsec:pdbec\] ---------------------------------------------- Multiple period states in BECs in optical lattices were first predicted by Machholm [*et al.*]{} [@machholm04]. Using both (1) the simple discrete model within the tight-binding approximation to the mean-field GP equation and (2) the more general continuum GP equation, they studied BECs flowing along the 1-dimensional external periodic potential of the form given by Equation (\[eq:lat\]) (multiple period states of BECs in a Kronig–Penney potential (a periodic delta-function potential) were studied in [@li04]).They have shown the existence of the multiple period states as stationary states and have clarified that they emerge due to nonlinearity originating from superfluidity. Figure \[fig:pdcont\] shows the lowest energy bands obtained by solving the GP equation for the continuum model. A striking difference from the energy bands obtained from the discrete model discussed in Section \[subsec:basics\] is that the phase states form a band in the continuum model (see the lower thick solid lines in Figure \[fig:pdcont\]) and their density profiles have period $2d$. Figure \[fig:profiles\] shows the density profiles of the lower-energy and higher-energy period-2 states. By comparing with the external potential $V(x)$ shown in the lower panel, we can see that the periodicity of these states is indeed $2d$. Here, the quasi-wave number $k$ of the superflow is at $k=\pi/2d$ corresponding to the first Brillouin zone edge of the system with period $2d$, and thus the condensate wave function $\psi$ has a node in each period. The density profile shown by the solid (dashed) line, which has nodes at the potential maxima (minima), is that of the period-2 state in the lower (higher) energy branch (see the lower (upper) thick solid line in Figure \[fig:pdcont\]). According to the energy bands shown in Figure \[fig:pdcont\], we can also see that the lowest band of the period-2 states appears as an upper edge of the swallowtail for the period $2d$ system. Connection between the period-2 states and the swallowtail was studied in depth in [@seaman05b]. It was pointed out that the lowest band of the period-2 states is closely connected with the dynamical instability [@machholm04]. The linear stability analysis has shown that, with increasing $k$, the dynamical instability of the normal Bloch state sets on at the quasimomentum where the band of the normal Bloch states merges with the lowest band of the period-2 states. There, the dynamical instability is caused by the growing perturbation mode with wavelength $2d$ [@wu01; @machholm03; @wu03; @modugno04].The growth of the mode with wavelength $2d$ accompanying the dynamical instability has been observed experimentally as well [@gemelke05]. Since the lowest band of the period-2 states appears as the saddle of the swallowtail for the period $2d$ system and it forms the upper edge of the swallowtail, these period-2 states are dynamically unstable [@machholm03] while the upper branch of the period-2 states can be dynamically stable in some region of $k$ [@machholm04]. The lowest multiple period states can be dynamically stable by introducing long-range interactions. For example, it was demonstrated that, in dipolar BECs, multiple period states with period $2d$ and $3d$ can be dynamically stable even at $k=0$ provided the dipole-dipole interactions are repulsive and sufficiently strong [@maluckov12a; @maluckov12]. Multiple Period States in Superfluid Fermi Gases\[subsec:pdfermi\] ------------------------------------------------------------------ The emergence of the multiple period states in BECs in optical lattices is one of the novel nonlinear phenomena caused by the presence of the superfluid order parameter. However, in the normal repulsively interacting BECs without long-range interaction, the lowest multiple period states are higher in energy than the normal Bloch states and are dynamically unstable. Yoon [*et al.*]{}[@period_double] have shown that the situation is very different for the multiple period states of superfluid Fermi gases in the BCS regime. Figure \[fig:profilesfermi\] shows the profiles of the magnitude of the pairing field $|\Delta(x)|$ and the density $n(x)$ of the lowest period-2 states of superfluid Fermi gases in the BCS-BEC crossover obtained by solving the BdG equations for the continuum model (\[eq:bdg\]). The striking difference from the BEC case shown in Figure \[fig:profiles\] is that the feature of the period doubling shows up in the pairing field rather than the density. The difference between the regions of $-1<x/d\le0$ and $0<x/d\le1$ can be clearly seen in $|\Delta(x)|$ at any value of $1/k_F a_s$. On the other hand, the difference in $n(x)$ between these two regions is small in the deeper BCS side ($1/k_F a_s =-1$) (see the red line in Figure \[fig:profilesfermi\]b) and finally disappears in the BCS limit. Figure \[fig:efermi\] shows the energy bands of the lowest period-2 states in the BCS regime ($1/k_Fa_s=-1$) together with that of the normal Bloch states. In the region of small $P$, the line of the period-2 states coincides with that of the normal Bloch states, as they are equivalent in this region, the states with period 1 being just a subset of any multiple period states with integer periods. Unlike the period-2 states in BECs, which form the concave upper edge of the swallowtail (see Figure \[fig:pdcont\]), here the band of the period-2 states is convex upward. Remarkably, the lowest period-2 states are energetically more stable compared to the normal Bloch states around the Brillouin zone edge of the period-2 states ($P/P_{\rm edge} \sim 0.5$). The manner in which the energy of the lowest period-2 states relative to that of the normal Bloch states changes along the BCS-BEC crossover can be seen in Figure \[fig:defermi\]. As we have already seen, the period-2 states are energetically more stable (*i.e.*, $\Delta E<0$) in the deep BCS regime, where the band of the period-2 states is convex upward. With $1/k_{F}a_s$ increasing from the deep BCS regime, $\Delta E$ increases from a negative value and finally period-doubled states become higher in energy than the normal Bloch states (*i.e.*, $\Delta E>0$) in the BEC side, where the band of the period-2 states forms the concave upper edge of the swallowtail. The energetic stability of the period-2 states in the BCS regime can be physically understood as follows. Let us consider the different behavior of $\Delta(x)$ and $n(x)$ for a period-2 state and a normal Bloch state at $P=P_{\rm edge}/2$. In the case of the normal Bloch state, since $|\Delta(x)|$ is exponentially small in the BCS regime, we can distort the order parameter $\Delta(x)$ to produce a node, like the one in the period-2 state, with a small energy cost (per particle) up to the condensation energy $|E_{\rm cond}|/N \ll E_F$, where $E_{\rm cond} \equiv g^{-1} \int d^3r\, |\Delta({\bf r})|^2$. However, making a node in $\Delta(x)$ kills the supercurrent $j=V^{-1}\partial_P E$, which yields a large gain of kinetic energy (per particle) of the superfluid flow of order $\sim P_{\rm edge}^2/m \sim E_R$. Even if $\Delta(x)$ is distorted substantially to have a node, the original density distribution of the normal Bloch state is almost intact so that the increase of the kinetic energy and the potential energy due to the density variation is small. Therefore, the period-2 state is energetically more stable than the normal Bloch state in the BCS regime. In the above discussion, the key point is that $\Delta(x)$ and $n(x)$ can behave in a different way in the BCS regime. On the other hand, in the BEC limit, the density is directly connected to the order parameter as $n(x)\propto |\Delta(x)|^2$, and distorting the order parameter accompanies an increase of the kinetic and potential energies due to a large density variation. In the deep BCS regime, the period-2 states are not only energetically stable, but also they can be long-lived although dynamically unstable. The black solid line in Figure \[fig:survival\] shows the growth rate $\gamma$ of the fastest exponentially growing mode $|\eta(t)|=|\eta(0)|\, e^{\gamma t}$ of the deviation $|\Delta(x,t)|-|\Delta_0(x)|$ from the true stationary state $\Delta_0(x)$ for the period-2 states. We see that $\gamma$ is suppressed with decreasing $1/k_Fa_s$, which makes the period-2 states long-lived in the BCS regime. The growth rate $\gamma$ corresponds to the imaginary part of the complex eigenvalue for the fastest growing mode obtained by the linear stability analysis [@ring; @pethick_smith], which is an intrinsic property of the initial stationary state independent of the magnitude of the perturbation. On the other hand, the actual survival time $\tau_{\rm surv}$, the timescale for which the initial state is destroyed by the large-amplitude oscillations, depends on the accuracy of their initial preparation. The survival time can be estimated by $\tilde{\eta}(0) e^{\gamma t} \sim 1$, where $\tilde{\eta}(0)$ is the relative amplitude of the initial perturbation with respect to $|\Delta_0|$ for the fastest growing mode. In Figure \[fig:survival\], we show $\tau_{\rm surv}$ for various values of $\tilde{\eta}(0)$. This result suggests that if the initial stationary state is prepared within an accuracy of 10% or smaller, this state safely survives for time scales of the order of $100\hbar/E_R$ or more in the BCS side, corresponding to $\tau_{\rm surv}$ of more than the order of a few milliseconds for typical experimental parameters [@miller07]: for $E_{R, b}=2\pi\times 7.3 \mathrm{kHz}\times \hbar$ used in the experiment of [@miller07], $1 \hbar/E_R = 0.0109$ msec. In the deep BCS regime ($1/k_Fa_s \ll -1$), $\tau_{\rm surv}$ increases further and may become larger than the time scale of the experiments, so that the period-doubled states can be regarded as long-lived states and, in addition, they have lower energy than the normal Bloch states in a finite range of quasimomenta. Therefore, by (quasi-)adiabatically increasing the quasimomentum $P$ of the superflow starting from the ground state at $P=0$, multiple-period states such as the period-doubled states could be realized experimentally in the deep BCS regime. Nonlinear Lattices\[sec:nonlinlat\] =================================== This section deals with a special kind of optical lattices, called “nonlinear lattices”. Here the coupling constant of the nonlinear term itself (*i.e.*, the interatomic interaction strength or the scattering length) has a space-periodic dependence. This is quite different from the systems which we have discussed so far: unlike cold atomic gases in a linear external periodic potential, those in a nonlinear lattice with periodically modulated interaction in space can be designed to have no linear periodic counterpart at all. While, in the former, properties are determined by the competition between the linear periodic potential and a nonlinear term, in the latter, periodicity and nonlinearity are generated by a single term. This leads to unique stability properties of the superfluid, and imposes additional conditions on its survival [@zhang13; @nonlinlat]. In the first subsection, we provide a basic sketch of how the sustenance of superfluidity depends on the geometry (homogeneous$/$in a linear lattice$/$in a nonlinear lattice) of the BEC system. In the second subsection, we explain the stability properties of BECs in a nonlinear lattice in terms of a simple discrete model. The third subsection presents the results of studies on ultracold bosons and fermions in nonlinear lattices for various parameter regimes. Finally, in the last subsection, we present a short outline of the experimental setup that constructs such a space-periodic dependence of the interatomic interaction strength. Dynamical Stability of the Superfluid: Special Properties of Nonlinear Lattices ------------------------------------------------------------------------------- For uniform and homogeneous BECs, the dynamical stability of the superfluid is determined by the nature of the interatomic interaction. If it is repulsive (*i.e.*, the scattering length is positive), the superfluid remains dynamically stable for any value of the momentum $\hbar k$ of the superflow. On the other hand, if the interaction is attractive (*i.e.*, the scattering length is negative), the long-wavelength modes with $$q^2 < \frac{4mng}{\hbar^2}$$ grow or decay in time exponentially for any value of $\hbar k$, thus invoking an instability in the system (e.g., [@pethick_smith]). However, shorter-wavelength modes are stable because for them, the kinetic energy dominates over the interaction energy. The dynamical stability properties of the superfluid changes in the presence of an external periodic potential. The periodic nature of the system may lead to Bloch solutions of the form $\psi(x)=e^{ikx} \phi(x)$, where $\phi(x)$ is a periodic function with the same periodicity as that of the lattice. The quasimomentum of the superflow is given by $\hbar k$, $k$ being the corresponding Bloch wave number. Here, for simplicity, we have assumed that the superfluid flows in the same $x$ direction as the periodic potential. Unlike the homogeneous system, the system has a nonzero critical value of $k$ above which the Bloch states are dynamically unstable [@wu01] as has been seen in Section \[sec:loopstability\]: in other words, the $k=0$ state is always dynamically stable. For a nonlinear lattice, however, this picture changes still further. The coupling constant $g$ in the GP equation (\[GP1\]) for bosons and the BdG equations (\[eq:bdg\]) and (\[eq:gap\]) for fermions now depends on the space coordinate $x$. It can be thought of having a form as $$\label{gperiodic} g(x) = V_1 + V_2 \cos{2k_0x}\, ,$$ *i.e.*, $g(x)$ consists of one constant part and one sinusoidal component. $k_0$ is related to the period $d$ of the modulation by $k_0 = \pi/d$. If the nonlinear lattice is realized by an optical Feshbach resonance (details are given in the last part of this section), $k_0$ is equivalent to the wave number of the laser beam. This special type of periodic nonlinearity gives rise to a dynamical instability for the $k=0$ state [@zhang13], which is in contrast with the linear lattice case (*i.e.*, the case of an external periodic potential). We shall explain this in the following subsection. Basic Physical Idea: The Dynamical Stability of Nonlinear Lattices ------------------------------------------------------------------ Zhang *et al.* [@zhang13] and Dasgupta *et al.* [@nonlinlat] studied extended states of BECs in quasi-one-dimension with a periodically modulated interaction in space, *i.e.*, a nonlinear lattice with no periodic linear potential. It was observed that when the coefficient of the nonlinear term is purely sinusoidal (*i.e.*, $V_1=0$ and $V_2 \neq 0$), Bloch states at $k=0$ are dynamically (and energetically) unstable [@zhang13].In addition, even though $k=0$ state is dynamically unstable, states for nonzero $k$ could be dynamically stable in some region in $0.25\le k/k_0\le 0.5$ (this point will be seen in more detail later in Figure \[dyn\_p1\]) [@zhang13]. Why is the $k=0$ state unstable in the nonlinear lattice, in contrast to the stable $k=0$ state in the linear periodic potential? Why are the states with higher values of $k$ possibly dynamically stable even though the $k=0$ state is unstable? To explain these, we take resort to a simple discrete model [@nonlinlat].We map the 1D nonlinear lattice with $V_1=0$ and $V_2 \neq 0$ to a discrete model with the on-site interaction alternating between $U$ and $-U$ (with $U>0$): $$H = -K \sum_j (\psi_j^* \psi_{j+1} + \psi_{j+1}^* \psi_j) + \frac{U}{2} \left[\, \sum_{j={\rm even}}|\psi_j|^4 - \sum_{j={\rm odd}} |\psi_j|^4\, \right]\, .$$ So, if we denote the distance between two adjacent sites in the discrete model as $\tilde{d}$, the actual lattice constant $d$ of the unit cell in the original system is $2\tilde{d}$ (*i.e.*, the unit cell in the original system corresponds to a “supercell” with two sites in the discrete model). ![Density distributions in the lowest band of the normal Bloch states (*i.e.*, period-1 states whose period is one supercell) as functions of $k$ for different values of $U\nu/2K$. Panels (**a**) and (**b**): Populations of $|g_1|^2$ (attractive site) and $|g_2|^2$ (repulsive site) for $U\nu/2K=6$, respectively. Panels (**c**) and (**d**): Populations of $|g_1|^2$ and $|g_2|^2$ for $U\nu/2K=0.75$, respectively. This figure is taken from [@nonlinlat].[]{data-label="p1den"}](density_set4p1g1.pdf "fig:") ![Density distributions in the lowest band of the normal Bloch states (*i.e.*, period-1 states whose period is one supercell) as functions of $k$ for different values of $U\nu/2K$. Panels (**a**) and (**b**): Populations of $|g_1|^2$ (attractive site) and $|g_2|^2$ (repulsive site) for $U\nu/2K=6$, respectively. Panels (**c**) and (**d**): Populations of $|g_1|^2$ and $|g_2|^2$ for $U\nu/2K=0.75$, respectively. This figure is taken from [@nonlinlat].[]{data-label="p1den"}](density_set4p1g2.pdf "fig:")\ ![Density distributions in the lowest band of the normal Bloch states (*i.e.*, period-1 states whose period is one supercell) as functions of $k$ for different values of $U\nu/2K$. Panels (**a**) and (**b**): Populations of $|g_1|^2$ (attractive site) and $|g_2|^2$ (repulsive site) for $U\nu/2K=6$, respectively. Panels (**c**) and (**d**): Populations of $|g_1|^2$ and $|g_2|^2$ for $U\nu/2K=0.75$, respectively. This figure is taken from [@nonlinlat].[]{data-label="p1den"}](density_set3p1g1.pdf "fig:") ![Density distributions in the lowest band of the normal Bloch states (*i.e.*, period-1 states whose period is one supercell) as functions of $k$ for different values of $U\nu/2K$. Panels (**a**) and (**b**): Populations of $|g_1|^2$ (attractive site) and $|g_2|^2$ (repulsive site) for $U\nu/2K=6$, respectively. Panels (**c**) and (**d**): Populations of $|g_1|^2$ and $|g_2|^2$ for $U\nu/2K=0.75$, respectively. This figure is taken from [@nonlinlat].[]{data-label="p1den"}](density_set3p1g2.pdf "fig:") Assuming the state is in the Bloch form, we write the amplitude $\psi_j$ at site $j$ as $\psi_j=g_j e^{i kj\tilde{d}}$, where $\hbar k$ is quasimomentum of the superflow and $g_j$ is the complex amplitude at site $j$ with the periodic boundary conditions, $g_j = g_{j+2}$ (Note that the unit cell contains 2 sites.). In addition, the amplitudes $g_j$’s are subject to the normalization condition $|g_1|^2 + |g_2|^2 = \nu$, where $\nu$ is the total number of particles in a unit cell with two sites. One can obtain the stationary solutions of $g_1$ and $g_2$ by solving the combined equations of $\delta H/\delta \psi_1^* =0$ and $\delta H/\delta \psi_2^* =0$ in almost the same manner as in Section \[subsec:basics\]. Resulting populations $|g_1|^2$ and $|g_2|^2$ in the attractive and the repulsive sites, respectively, for the lowest Bloch band are $$\frac{|g_1|^2}{\nu} = n_+ \quad\mbox{and}\quad \frac{|g_2|^2}{\nu} = n_- \label{eq:pop}$$ with $$n_{\pm} = \frac{1}{2} \left\{1 \pm \left[ \left(\frac{\cos{k\tilde{d}}}{U\nu/2K}\right)^2+1 \right]^{-1/2} \right\}\, . \label{eq:npm}$$ The populations $|g_1|^2$ and $|g_2|^2$ for two different values of $U\nu/2K$ are shown in Figure \[p1den\] as functions of $k$ within the first Brillouin zone. As seen from Figure \[p1den\], the population difference between the adjacent sites is the smallest at the zone center $k=0$ while it increases as going toward the zone edge at which all the particles are accumulated in the attractive sites. To understand why $k=0$ state is dynamically unstable, it is instructive to see the interaction energy averaged over one unit cell (with 2 sites) [@zhang13].From Equations (\[eq:pop\]) and (\[eq:npm\]), we obtain the average interaction energy per particle for the lowest Bloch band as $$\frac{E_{\rm int}/K}{N} = -\frac{U}{2K\nu} (|g_1|^4 + |g_2|^4) = -\frac{U\nu}{2K} \left[\left(\frac{\cos{k\tilde{d}}}{U\nu/2K}\right)^2 +1 \right]^{-1/2} < 0\, .$$ Note that, in the lowest Bloch band, the average interaction energy per particle is negative for any value of $k$. So, roughly speaking, this situation resembles a BEC with attractive interparticle interaction, which is dynamically unstable as we have mentioned in the last subsection, and the dynamical instability of BECs in nonlinear lattices at $k=0$ could be understood as a consequence of the net attractive interaction energy [@zhang13]. ![Dynamical stability diagrams for the normal Bloch states (*i.e.*, period-1 states) for (**a**) and $0.75$ (**b**). Quasi-wave numbers $k$ and $q$ are in units of $\pi/2\tilde{d}$. are the dynamically unstable regions and the gray-shaded regions are the dynamically stable regions. growth rate of the fastest growing mode, *i.e.*, the largest maximum absolute value of the imaginary part of the eigenvalues of matrix $\sigma_z M(q)$ in Equation (\[dynmat\]) in units of $K$. [@nonlinlat]. []{data-label="dyn_p1"}](dynset4_p1.pdf "fig:") ![Dynamical stability diagrams for the normal Bloch states (*i.e.*, period-1 states) for (**a**) and $0.75$ (**b**). Quasi-wave numbers $k$ and $q$ are in units of $\pi/2\tilde{d}$. are the dynamically unstable regions and the gray-shaded regions are the dynamically stable regions. growth rate of the fastest growing mode, *i.e.*, the largest maximum absolute value of the imaginary part of the eigenvalues of matrix $\sigma_z M(q)$ in Equation (\[dynmat\]) in units of $K$. [@nonlinlat]. []{data-label="dyn_p1"}](dynset3_p1.pdf "fig:") Figure \[dyn\_p1\] shows the dynamical stability diagram of the stationary states in the lowest Bloch band in the $k$–$q$ plane, where $q$ is the quasi-wave number of the perturbation on the stationary states. It is noted that there is a region at larger values of $k$ in which the lowest Bloch states are dynamically stable (e.g., the gray-shaded region of $0.5 \lesssim 2k\tilde{d}/\pi \le 1$ in Figure \[dyn\_p1\]a) even though they are dynamically unstable at smaller values of $k$ including the zone center at $k=0$. To understand this somewhat counterintuitive fact, we shall take a closer look at the populations $|g_1|^2$ and $|g_2|^2$ shown in Figure \[p1den\]. As we have briefly mentioned before, we note that almost all the particles are accumulated in the attractive sites and thus the repulsive sites are almost empty near the zone edge at $2k\tilde{d}/\pi = 1$: at the zone edge, only alternate sites are occupied and fragments of the BEC in these alternate sites are isolated. Since the transition amplitude between the states with populations $\{|g_1|^2, |g_2|^2\}$ and $\{|g_1|^2\pm 1, |g_2|^2\mp 1\}$ is $\sim \sqrt{|g_1| |g_2|} K$, tunneling of particles between neighboring sites is suppressed (*i.e.*, the inter-site dynamics is frozen), and thus the dynamical instability is suppressed near the zone edge. Therefore, in nonlinear lattices, the lowest Bloch states can be dynamically stable at higher values of $k$ near the zone edge [@nonlinlat]. On the other hand, at the zone center $k=0$, the difference between the respective populations in adjacent sites is the smallest: No sites are empty and inter-site tunneling is non-negligible. Thus the suppression of the dynamical instability does not work around the zone center, rendering the system dynamically unstable due to the net attractive interaction mentioned before. Since the isolation of fragments of the BEC is a result of the attractive interaction in alternate sites and this new mechanism of dynamical stability is more effective for larger $U\nu/2K$ (see wider gray-shaded area in Figures \[dyn\_p1\]a than that in \[dyn\_p1\]b) resulting in larger net attractive interaction energy, this mechanism can be called “attraction-induced dynamical stability” [@nonlinlat]. We note that, in addition to the net attractive interaction energy and the suppression of the tunneling near the zone edge, there would be other factors to determine the dynamical stability of the nonlinear lattice system. When $U\nu/2K$ is sufficiently small, we observe that a dynamically unstable region appears near the zone edge (see, e.g., Figure \[dyn\_p1\]b) and the dynamically stable region is located at the intermediate values of $k$ ($0.65 \lesssim 2k\tilde{d}/\pi \lesssim 0.8$ in the case of Figure \[dyn\_p1\]b ). This non-trivial reentrant behavior suggests that there are several other factors that affect the stability, which are collectively responsible for the complicated stability diagram like Figure \[dyn\_p1\]b. As a final comment in this subsection, we mention that the discussion here is based on the discrete model, but the attraction-induced dynamical stability has been confirmed in the continuum model as well [@nonlinlat]. The main difference is that, in the continuum model, if the value of $V_2$ is increased beyond a certain point, the attractive interaction between intra-site particles becomes dominant and eventually leads to the collapse of fragments of the BEC. This intra-site dynamics cannot be accounted for by the discrete model, which does not include the intra-site degrees of freedom. Superfluid Cold Atomic Gases in Nonlinear Lattices -------------------------------------------------- Extended states of BECs in nonlinear lattices were first studied by Zhang *et al.* in [@zhang13] (Localized states such as solitons in nonlinear lattices were studied earlier in, e.g., [@sakaguchi05; @abdullaev07] (see also [@malomed_review] and references therein).). In this work, they considered quasi-1D BECs in nonlinear lattices described by the 1D version of the GP equation (\[GP1\]) with the periodically modulated interaction strength in space given by Equation (\[gperiodic\]): $g(x) = V_1 + V_2 \cos{2k_0x}$ with $V_1$ and $V_2 \ge 0$. They studied stationary Bloch states in nonlinear lattices and their energetic and dynamical stability summarized in Figure \[fig:zhang\].As we have discussed in the previous subsection, the key result is that, when $V_1=0$, $k=0$ state is dynamically (and energetically) unstable for any value of $V_2\ne 0$ (see black regions in the lower panels of Figure \[fig:zhang\] for $c_1=0$). This is in contrast to BECs in a linear external periodic potential whose dynamically unstable region is restricted in the domain of $1/4 < k/2k_0 \le 1/2$ (*i.e.*, the right half of each panel in Figure \[fig:niustab1\]b). In addition, states at higher values of $k$ near the zone edge can be dynamically stable even though $k=0$ state is dynamically unstable. It was pointed out that the dynamical instability of the $k=0$ state can be partially explained by the net attractive average interaction energy as we have discussed in the previous subsection. They also discussed the stability of the superfluidity due to the competition between the spatially modulated part ($V_2$ term) and the uniform, repulsive component ($V_1$ term); the latter tends superfluids to be stable. With increasing $V_1$ from zero for a fixed nonzero $V_2$, state at $k=0$ becomes dynamically and energetically stable(see Figure \[fig:zhang\]a,b), and by further increasing $V_1$ a swallowtail loop starts to appear at the zone edge when $V_1 > V_2$. Dasgupta [*et al.*]{} [@nonlinlat] studied multiple-period states of BECs in nonlinear lattices. They discussed stationary states with larger integer periodicity and their energetic and dynamical stability using the GP equation for both the discrete and continuum models. The main result of this work is that they found a new mechanism of dynamical stability called “attraction-induced dynamical stability” and provided the understanding of the dynamical stability around the Brillouin zone edge due to the isolation of fragments of the BEC in each attractive domain in the nonlinear lattice as discussed in the previous subsection. This attraction-induced dynamical stability is even better manifested for the period-2 case because the majority of the particles are stored in every second attractive domain (*i.e.*, every fourth site in the discrete model) making the fragments of the BEC more firmly separated, while for period-1 case they are stored in every attractive domain (*i.e.*, every second site in the discrete model). Figure \[dyn\_p2\] shows the dynamical stability diagrams for period-2 Bloch states calculated for the discrete model at the same values of $U\nu/2K$ as Figure \[dyn\_p1\]. It is noted that the growth rate of the fastest growing mode in Figure \[dyn\_p2\]a is much smaller than the period-1 counterpart shown in Figure \[dyn\_p1\]a; the dynamically stable region in Figure \[dyn\_p2\]b is larger than the period-1 counterpart shown in Figure \[dyn\_p1\]b. Superfluid Fermi gases in nonlinear lattices were studied by Yu [*et al.*]{} [@yu15]. They considered quasi-1D 2-component superfluid Fermi gases based on the 1D version of the BdG equation (\[eq:bdg\]) with spatially modulated interaction strength of the form $g(x) = V_1 + V_2 \cos{2k_0x}$ with $V_1<0$ and $V_2\ge 0$. Note that it has been assumed that the uniform component is attractive ($V_1 <0$) so that the system is in the superfluid phase. The properties of the Bloch states in this system for various parameter values of $V_1$ and $V_2$ are summarized in Figure \[fig:yu\]. ![The same as Figure \[dyn\_p1\], but for period-2 Bloch states. Taken from [@nonlinlat]. []{data-label="dyn_p2"}](dynset4_p2.pdf "fig:") ![The same as Figure \[dyn\_p1\], but for period-2 Bloch states. Taken from [@nonlinlat]. []{data-label="dyn_p2"}](dynset3_p2.pdf "fig:") The key point is the competition between the effects of the nonlinear uniform interaction by $V_1$ and the periodicity of the system induced by $V_2$. It was found that the former dominates over the latter for lager $|V_1|/V_2$ (the region denoted by “SW” in Figure \[fig:yu\]), the Bloch band has a swallowtail loop around the Brillouin zone edge. This situation is similar to superfluid Fermi gases in a periodic potential, where the swallowtail appears due to the effect of the nonlinear interaction dominating over the periodicity of the system induced by the external potential [@swallowtail]. On the other hand, for smaller $|V_1|/V_2$ (the region denoted by “FFLO-like” in Figure \[fig:yu\]), it was predicted that the state at the Brillouin zone edge with nonzero quasimomentum of the superflow (quasimomentum per atom $P=\hbar k_0/2$ in the notation of this review article) becomes the ground state of the system. They call this state as “FFLO-like state” because of the nonzero value of the quasimomentum of the superflow (Note that the current of this state, however, is zero.).The stability of the Bloch states of superfluid Fermi gases in nonlinear lattices has yet to be studied. Experimental Setup ------------------ Finally we give a brief sketch how nonlinear lattices can be created experimentally. With the aid of modern experimental techniques, it is now indeed possible to construct systems where the nonlinear term has an explicit spatial dependence. A very efficient way to do this is to employ optical Feshbach resonances (OFRs) [@fatemi; @chin]. Ultracold atoms offer an immense controllability over physical quantities like scattering length. It all started with magnetic Feshbach resonances, that use Zeeman shifts to make the scattering states resonate with a bound molecular state. It was later suggested [@fedichev96] that lasers can be used to induce the resonance optically. This Feshbach resonance via optical fields offers additional advantages over magnetic ones: the intensity as well as the frequency of the laser beams can be rapidly and precisely controlled. In addition, high resolution (submicron level) spatial control of the scattering length is possible by creating specially structured laser fields. In OFRs, the basic scheme is to use a laser beam (tuned near the photoassociation resonance) that couples an initial state of free atoms to a molecular bound state. The scattering length is accordingly modified. If the OFR is driven by a standing wave with a certain periodicity, the interparticle interaction derived from it has the same periodic nature. Thus, a nonlinear lattice is generated. Using OFRs, Yamazaki [*et al.*]{} [@yamazaki10] demonstrated rapid, spatial modulation of the scattering length periodically at the submicron level. In this particular setup, a pulsed optical standing wave was applied to a BEC of Ytterbium atoms. The resonant wavelength was 556 nm, and the optical pulse could generate a scattering length periodically modulated in space with wavelength 278 nm. Conclusions \[sec:conclusion\] ============================== Over the last two decades ultracold atoms in optical lattices [@bloch_review; @lattice; @yukalov_review; @jkps_review] have emerged as a key paradigm to study the *ideal* realizations of many important problems in highly controllable settings, ranging from single particle quantum mechanical effects such as Bloch oscillations to strongly correlated many-body effects such as the superfluid-Mott insulator quantum phase transition. In the recent past the focus in the field has shifted more towards observing equilibrium and non-equilibrium quantum many-body effects and topological phenomena [@lewenstbook]. In this light the subject of this review, namely, the interplay of mean-field atomic interactions and the periodicity of the applied external potential serves to re-emphasize the fact that there are unique and interesting phenomena even in a much simpler setting. Our hope is that such a review can rekindle some interest in this area especially from the experimental side as to date there has not been a clear observation of swallowtail loop structures or period-doubled solutions for optical lattices. Moreover, in striving to control complex quantum systems such as ultracold atoms to an ever greater degree in order to realize complex many-body ground states [@lewenstbook] it becomes very important to be aware of and understand fundamental limitations on state preparation due to unavoidable adiabaticity breaking imposed by phenomena such as swallowtail loop dispersions. In this review we focused on some interesting phenomena originating from the nonlinear, mean-field interactions of superfluid atomic gases in periodic potentials. We began with a summary of the basic theoretical description of Bose–Einstein condensates (BECs) and superfluid Fermi gases within the mean-field framework in Section \[sec:framework\]. In Section \[sec:swallowtail\] we provided a comprehensive overview of the phenomenon of swallowtail loops in the band-structure of superfluid atomic gases in an optical lattice, followed by the the discussion of Bloch states with multiple periods of the applied optical lattice potential in Section \[sec:multiperiod\], and Bloch states in nonlinear lattices, *i.e.*, situations in which the nonlinear interaction term is itself a periodic function in space in Section \[sec:nonlinlat\]. While we have covered a substantial portion of the various interesting phenomena that can arise due to the nonlinearity in the mean-field theory of superfluids, this is by no means complete. For instance, as mentioned in the introduction, we have made no attempt to describe localized soliton solutions to the Gross–Pitaevskii equation and suggest the excellent review [@malomed_review] on this topic for the interested reader. Although considerable amount of research work has already been accomplished in this field, it is still relatively young and has flourished only over the last two decades beginning with the experimental discovery of BECs. As a result we believe there are still a range of open problems that can be investigated. Some examples include, the stability of Bloch and swallowtail loop states of superfluid Fermi gases in nonlinear lattices, Bloch oscillation dynamics for both bosons and fermions in regimes with swallowtail loops, the study of quantum equivalents of mean-field nonlinear phenomena including solitons and loops, and nonlinear phenomena in more exotic systems such as BECs with spin-orbit interactions or spinor BECs *etc*. [999]{} Pethick, C.J.; Smith, H. [*Bose–Einstein Condensation in Dilute Gases*]{}, 2nd ed.; Cambridge University Press: Cambridge, UK, 2008. Pitaevskii, L.P.; Stringari, S. [*Bose–Einstein Condensation*]{}; Oxford University Press: Oxford, UK, 2003. Leggett, A.J. Bose–Einstein condensation in the alkali gases: Some fundamental concepts. [*Rev. Mod. Phys.*]{} [**2001**]{}, [*73*]{}, 307–356. Burger, S.; Bongs, K.; Dettmer, S.; Ertmer, W.; Sengstock, K.; Sanpera, A.; Shlyapnikov, G.V.; Lewenstein, M. Dark solitons in Bose–Einstein condensates. [*Phys. Rev. Lett.*]{} [**1999**]{}, [*83*]{}, 5198–5201. Denschlag, J.; Simsarian, J.E; Feder, D.L.; Clark, C.W.; Collins, L.A.; Cubizolles, J.; Deng, L.; Hagley, E.W.; Helmerson, K.; Reinhardt, W.P.; *et al*. Generating solitons by phase engineering of a Bose–Einstein condensate. [*Science*]{} [**2000**]{}, [*287*]{}, 97–101. Strecker, K.E.; Partridge, G.B.; Truscott, A.G.; Hulet, R.G. Formation and propagation of matter-wave soliton trains. [*Nature*]{} [**2002**]{}, [*417*]{}, 150–153. Khaykovich, L.; Schreck, F.; Ferrari, G.; Bourdel, T.; Cubizolles, J.; Carr, L.D.; Castin, Y.; Salomon, C. Formation of a matter-wave bright soliton. [*Science*]{} [**2002**]{}, [*296*]{}, 1290–1293. Deng, L.; Hagley, E.W.; Wen, J.; Trippenbach, M.; Band, Y.; Julienne, P.S.; Simsarian, J.E.; Helmerson, K.; Rolston, S.L.; Phillips, W.D. Four-wave mixing with matter waves. [*Nature*]{} [**1999**]{}, [*398*]{}, 218–220. Kevrekidis, P.G.; Frantzeskakis, D.J.; Carretero-González, R. (Eds.) [*Emergent Nonlinear Phenomena in Bose–Einstein Condensates: Theory and Experiment*]{}; Springer: Berlin/Heidelberg, Germany, 2008. Zwierlein, M.W.; Abo-Shaeer, J.R.; Schirotzek, A.; Schunck, C.H.; Ketterle, W. Vortices and superfluidity in a strongly interacting Fermi gas. [*Nature*]{} [**2005**]{}, [*435*]{}, 1047–1051. Eagles, D.M. Possible pairing without superconductivity at low carrier concentrations in bulk and thin-film superconducting semiconductors. [*Phys. Rev.*]{} [**1969**]{}, [*186*]{}, 456, doi:10.1103/PhysRev.186.456. Leggett, A.J. Diatomic Molecules and Cooper Pairs. In [*Modern Trends in the Theory of Condensed Matter*]{}; Pekalski, A., Przystawa, J., Eds.; Springer: Berlin/Heidelberg, Germany, 1980; Volume 115, pp. 13–27. Giorgini, S.; Pitaevskii, L.P.; Stringari, S. Theory of ultracold atomic Fermi gases. [*Rev. Mod. Phys.*]{} [**2008**]{}, [*80*]{}, 1215–1274. Bloch, I.; Dalibard, J.; Zwerger, W. Many-body physics with ultracold gases. [*Rev. Mod. Phys.*]{} [**2008**]{}, [*80*]{}, 885–964. Morsch, O.; Oberthaler, M. Dynamics of Bose–Einstein condensates in optical lattices. [*Rev. Mod. Phys.*]{} [**2006**]{}, [*78*]{}, 179–215. Yukalov, V.I. Cold bosons in optical lattices. [*Laser Phys.*]{} [**2009**]{}, [*19*]{}, 1–110, doi:10.1134/S1054660X09010010. Watanabe, G.; Yoon, S. Aspects of superfluid cold atomic gases in optical lattices. [*J. Korean Phys. Soc.*]{} [**2013**]{}, [*63*]{}, 839–857. Wu, B.; Diener, R.B.; Niu, Q. Bloch waves and bloch bands of Bose–Einstein condensates in optical lattices. [*Phys. Rev. A*]{} [**2002**]{}, [*65*]{}, 025601. Diakonov, D.; Jensen, L.M.; Pethick, C.J.; Smith, H. Loop structure of the lowest Bloch band for a Bose–Einstein condensate. [*Phys. Rev. A*]{} [**2002**]{}, [*66*]{}, 013604. Kartashov, Y.V.; Malomed, B.A.; Torner, L. Solitons in nonlinear lattices. [*Rev. Mod. Phys.*]{} [**2011**]{}, [*83*]{}, 247–305. Watanabe, G.; Maruyama, T. Nuclear pasta in supernovae and neutron stars. [**2012**]{}, arXiv:1109.3511. Maruyama, T.; Watanabe, G.; Chiba, S. Molecular dynamics for dense matter. [*Prog. Theor. Exp. Phys.*]{} [**2012**]{}, [*2012*]{}, 01A201. Dalfovo, F.; Giorgini, S.; Pitaevskii, L.P.; Stringari, S. Theory of Bose–Einstein condensation in trapped gases. [*Rev. Mod. Phys.*]{} [**1999**]{}, [*71*]{}, 463–512. Pitaevskii, L.P. Vortex lines in an imperfect Bose gas. [*Sov. Phys. JETP*]{} [**1961**]{}, [*13*]{}, 451–454. Gross, E.P. Structure of a quantized vortex in boson systems. [*Nuovo Cimento*]{} [**1961**]{}, [*20*]{}, 454–477. Gross, E.P. Hydrodynamics of a superfluid condensate. [*J. Math. Phys.*]{} [**1963**]{}, [*4*]{}, 195, doi:10.1063/1.1703944. De Gennes, P.G. [*Superconductivity of Metals and Alloys*]{}; Benjamin: New York, NY, USA, 1966. Randeria, M. Crossover from BCS Theory to Bose–Einstein Condensation. In [*Bose Einstein Condensation*]{}; Griffin, A., Snoke, D., Stringari, S., Eds.; Cambridge University Press: Cambridge, UK, 1995; pp. 355–392. Wu, B.; Niu, Q. Landau and dynamical instabilities of the superflow of Bose–Einstein condensates in optical lattices. [*Phys. Rev. A*]{} [**2001**]{}, [*64*]{}, 061603(R), doi:10.1103/PhysRevA.64.061603. Machholm, M.; Pethick, C.J.; Smith, H. Band structure, elementary excitations, and stability of a Bose–Einstein condensate in a periodic potential. [*Phys. Rev. A*]{} [**2003**]{}, [*67*]{}, 053613. Machholm, M.; Nicolin, A.; Pethick, C.J.; Smith, H. Spatial period doubling in Bose–Einstein condensates in an optical lattice. [*Phys. Rev. A*]{} [**2004**]{}, [*69*]{}, 043604. Wu, B.; Niu, Q. Nonlinear Landau–Zener tunneling. [*Phys. Rev. A*]{} [**2000**]{}, [*61*]{}, 023402. Mueller, E.J. Superfluidity and mean-field energy loops: Hysteretic behavior in Bose–Einstein condensates. [*Phys. Rev. A*]{} [**2002**]{}, [*66*]{}, 063603. Landau, L.D. Zur Theorie der Energieubertragung. II. [*Physikalische Zeitschrift der Sowjetunion*]{} **1932**, [*2*]{}, 46–51. (In German) Zener, C. Non-Adiabatic Crossing of Energy Levels. [*Proc. R. Soc. Lond. A*]{} **1932**, [*137*]{}, 696–702. Bronski, J.C.; Carr, L.D.; Deconinck, B.; Kutz, J.N. Bose–Einstein condensates in standing wave: The cubic nonlinear Schrödinger equation with a periodic potential. [*Phys. Rev. Lett.*]{} [**2001**]{}, [*86*]{}, 1402, doi:10.1103/ PhysRevLett.86.1402. Eckel, S.; Lee, J.G.; Jendrzejewski, F.; Murray, N.; Clark, C.W.; Lobb, C.J.; Phillips W.D.; Edwards, M.; Campbell, G.K. Hysteresis in a quantized superfluid ‘atomtronic’ circuit. [*Nature*]{} [**2014**]{}, [*506*]{}, 200–203. Thom, R. [*Structural Stability and Morphogenesis: An Outline of a General Theory of Models*]{}; W.A. Benjamin: Reading, MA, USA, 1975. Seaman, B.T.; Carr, L.D.; Holland, M.J. Nonlinear band structure in Bose–Einstein condensates: Nonlinear Schrödinger equation with a Kronig–Penney potential. [*Phys. Rev. A*]{} [**2005**]{}, [*71*]{}, 033622. Dong, X.; Wu, B. Instabilities and sound speed of a Bose–Einstein condensate in the Kronig–Penney potential. [*Laser Phys.*]{} [**2007**]{}, [*17*]{}, 190–197. Wu, B.; Niu, Q. Superfluidity of Bose–Einstein condensate in an optical lattice: Landau–Zener tunnelling and dynamical instability. [*New J. Phys.*]{} [**2003**]{}, [*5*]{}, 104, doi:10.1088/1367-2630/5/1/104. Burger, S.; Cataliotti, F.S.; Fort, C.; Minardi, F.; Inguscio, M.; Chiofalo, M.L.; Tosi, M.P. Superfluid and dissipative dynamics of a Bose-Einstein condensate in a periodic optical potential. [*Phys. Rev. Lett.*]{} [**2001**]{}, [*86*]{}, 4447–4450. Wu, B.; Niu, Q. Dynamical or Landau instability? [*Phys. Rev. Lett.*]{} **2002**, [*89*]{}, 088901. Burger, S; Cataliotti, F.S.; Fort, C.; Minardi, F.; Inguscio, M.; Chiofalo, M.L.; Tosi, M.P. [Burger [*et al.*]{} Reply]{}. [*Phys. Rev. Lett.*]{} **2002**, [*89*]{}, 088902. Mamaladze, Yu.G.; Cheĭshvili, O.D. Flow of a superfluid liquid in porous media. [*Sov. Phys. JETP*]{} [**1966**]{}, [*23*]{}, 112–117. Hakim, V. Nonlinear Schrödinger flow past an obstacle in one dimension. [*Phys. Rev. E*]{} [**1997**]{}, [*55*]{}, 2835–2845. Watanabe, G.; Dalfovo, F.; Piazza, F.; Pitaevskii, L.P.; Stringari, S. Critical velocity of superfluid flow through single-barrier and periodic potentials. [*Phys. Rev. A*]{} [**2009**]{}, [*80*]{}, 053602. Danshita, I.; Tsuchiya, S. Comment on “Nonlinear band structure in Bose–Einstein condensates: Nonlinear Schrödinger equation with a Kronig–Penney potential”. [*Phys. Rev. A*]{} [**2007**]{}, [*75*]{}, 033612. Seaman, B.T.; Carr, L.D.; Holland, M.J. Reply to “Comment on ‘Nonlinear band structure in Bose–Einstein condensates: Nonlinear Schrödinger equation with a Kronig–Penney potential’ ”. [*Phys. Rev. A*]{} [**2007**]{}, [*76*]{}, 017602. Barontini, G.; Modugno, M. Dynamical instability and dispersion management of an attractive condensate in an optical lattice. [*Phys. Rev. A*]{} [**2007**]{}, [*76*]{}, 041601(R). Zhu, Q.; Wu, B. Superfluidity of BEC in ultracold atomic gases. [*Chin. Phys. B*]{} [**2015**]{}, [*24*]{}, 050507. Modugno, M.; Tozzo, C.; Dalfovo, F. Role of transverse excitations in the instability of Bose–Einstein condensates moving in optical lattices. [*Phys. Rev. A*]{} [**2004**]{}, [*70*]{}, 043625. Fallani, L.; de Sarlo, L.; Lye, J.E.; Modugno, M.; Saers, R.; Fort, C.; Inguscio, M. Observation of Dynamical Instability for a Bose–Einstein Condensate in a Moving 1D Optical Lattice. [*Phys. Rev. Lett.*]{} [**2004**]{}, [*93*]{}, 140406. De Sarlo, L.; Fallani, L.; Lye, J.E.; Modugno, M.; Saers, R.; Fort, C.; Inguscio, M. Unstable regimes for a Bose–Einstein condensate in an optical lattice. [*Phys. Rev. A*]{} [**2005**]{}, [*72*]{}, 013603. Cataliotti, F.S.; Fallani, L.; Ferlaino, F.; Fort, C.; Maddaloni, P. Superfluid current disruption in a chain of weakly coupled Bose–Einstein condensates. [*New J. Phys.*]{} [**2003**]{}, [*5*]{}, 71, doi:10.1088/1367-2630/5/1/371. Smerzi, A.; Trombettoni, A.; Kevrekidis, P.G.; and Bishop, A.R. Dynamical Superfluid-Insulator Transition in a Chain of Weakly Coupled Bose–Einstein Condensates. [*Phys. Rev. Lett.*]{} [**2002**]{}, [*89*]{}, 170402. Adhikari, S.K. Expansion of a Bose–Einstein condensate formed on a joint harmonic and one-dimensional optical-lattice potential. [*J. Phys. B*]{} [**2003**]{}, [*36*]{}, 3951, doi:10.1088/0953-4075/36/19/006. Nesi, F.; Modugno, M. Loss and revival of phase coherence in a Bose–Einstein condensate moving through an optical lattice. [*J. Phys. B*]{} [**2004**]{}, [*37*]{}, S101, doi:10.1088/0953-4075/37/7/057. Mun, J.; Medley, P.; Campbell, G.K.; Marcassa, L.G.; Pritchard, D.E.; Ketterle, W. Phase Diagram for a Bose–Einstein Condensate Moving in an Optical Lattice. [*Phys. Rev. Lett.*]{} [**2007**]{}, [*99*]{}, 150604. Ferris, A.J.; Davis, M.J.; Geursen, R.W.; Blakie, P.B.; Wilson, A.C. Dynamical instabilities of Bose–Einstein condensates at the band edge in one-dimensional optical lattices. [*Phys. Rev. A*]{} [**2008**]{}, [*77*]{}, 012712. Hamner, C.; Zhang, Y.; Khamehchi, M.A.; Davis, M.J.; Engels, P. Spin-Orbit-Coupled Bose–Einstein Condensates in a One-Dimensional Optical Lattice. [*Phys. Rev. Lett*]{} [**2015**]{}, [*114*]{}, 070401. Chen, Y-A.; Huber, S.D.; Trotzky, S.; Bloch, I.; Altman, E. Many-body Landau–Zener dynamics in coupled one-dimensional Bose liquids. [*Nat. Phys.*]{} [**2011**]{}, [*7*]{}, 61–67. Amico, L.; Osterloh, A.; Cataliotti, F. Quantum Many Particle Systems in Ring-Shaped Optical Lattices. [*Phys. Rev. Lett.*]{} [**2005**]{}, [*95*]{}, 063201. Baharian, S.; Baym, G. Bose–Einstein condensates in toroidal traps: Instabilities, swallowtail loops, and self-trapping. [*Phys. Rev. A*]{} [**2013**]{}, [*87*]{}, 013619. Wright, K.C.; Blakestead, R.B.; Lobb, C.J.; Phillips, W.D.; Campbell, G.K. Driving phase slips in a superfluid atom circuit with a rotating weak link. [*Phys. Rev. Lett.*]{} [**2013**]{}, [*110*]{}, 025302. Lin, Y.Y.; Lee, R.-K.; Kao, Y.-M.; Jiang, T.-F. Band structures of a dipolar Bose–Einstein condensate in one-dimensional lattices. [*Phys. Rev. A*]{} [**2008**]{}, [*78*]{}, 023629. Venkatesh, B.P.; Larson, J.; O’Dell, D.H.J. Band-structure loops and multistability in cavity QED. [*Phys. Rev. A*]{} [**2011**]{}, [*83*]{}, 063606. Watanabe, G.; Yoon, S.; Dalfovo, F. Swallowtail Band Structure of the Superfluid Fermi Gas in an Optical Lattice. [*Phys. Rev. Lett.*]{} [**2011**]{}, [*107*]{}, 270404. Chen, Z.; Wu, B. Bose–Einstein Condensate in a Honeycomb Optical Lattice: Fingerprint of Superfluidity at the Dirac Point. [*Phys. Rev. Lett.*]{} [**2011**]{}, [*107*]{}, 065301. Hui, H.-Y.; Barnett, R.; Porto, J.V.; Sarma, S.D. Loop-structure stability of a double-well-lattice Bose–Einstein condensate. [*Phys. Rev. A*]{} [**2012**]{}, [*86*]{}, 063636. Mumford, J.; Larson, J.; O’Dell, D.H.J. Impurity in a bosonic Josephson junction: Swallowtail loops, chaos, self-trapping, and Dicke model. [*Phys. Rev. A*]{} [**2014**]{}, [*89*]{}, 023620. Haller, E.; Hart, R.; Mark, M.J.; Danzl, J.G.; Reichsöllner, L.; Nägerl, H.-C. Inducing Transport in a Dissipation-Free Lattice with Super Bloch Oscillations. [*Phys. Rev. Lett.*]{} [**2010**]{}, [*104*]{}, 200403. Meinert, F.; Mark, M.J.; Kirilov, E.; Lauber, K.; Weinmann, P.; Gröbner, M.; Nägerl, H.-C. Interaction-induced quantum phase revivals and evidence for the transition to the quantum chaotic regime in 1D atomic Bloch oscillations. [*Phys. Rev. Lett.*]{} [**2014**]{}, [*112*]{}, 193003. Karkuszewski, Z.P.; Sacha, K.; Smerzi, A. Mean field loops versus quantum anti-crossing nets in trapped Bose–Einstein condensates. [*Eur. Phys. J. D*]{} [**2002**]{}, [*21*]{}, 251–254. Torrontegui, E.; Ibáñez, S.; Martínez-Garaot, S.; Modugno, M.; del Campo, A.; Guéry-Odelin, D.; Ruschhaupt, A.; Chen, X.; Muga, J.G. Shortcuts to adiabaticity. [*Adv. At. Mol. Opt. Phys.*]{} [**2013**]{}, [*62*]{}, 117–169. Li, W.; Smerzi, A. Nonlinear Kronig–Penney model. [*Phys. Rev. E*]{} [**2004**]{}, [*70*]{}, 016605. Seaman, B.T.; Carr, L.D.; Holland, M.J. Period doubling, two-color lattices, and the growth of swallowtails in Bose–Einstein condensates. [*Phys. Rev. A*]{} [**2005**]{}, [*72*]{}, 033602. Gemelke, N.; Sarajlic, E.; Bidel, Y.; Hong, S.; Chu, S. Parametric Amplification of Matter Waves in Periodically Translated Optical Lattices. [*Phys. Rev. Lett.*]{} [**2005**]{}, [*95*]{}, 170404. Maluckov, A.; Gligorić, G.; Hadžievski, L. Long-lived double periodic patterns in dipolar cigar-shaped Bose–Einstein condensates in an optical lattice. [*Physica Scripta*]{} [**2012**]{}, [*T149*]{}, 014004. Maluckov, A.; Gligorić, G.; Hadžievski, L.; Malomed, B.A.; Pfau, T. Stable Periodic Density Waves in Dipolar Bose–Einstein Condensates Trapped in Optical Lattices. [*Phys. Rev. Lett.*]{} [**2012**]{}, [*108*]{}, 140402. Yoon, S.; Dalfovo, F.; Nakatsukasa, T.; Watanabe, G. Multiple period states of the superfluid Fermi gas in an optical lattice. [*New J. Phys.*]{} [**2015**]{}, [*18*]{}, 023011. Ring, P.; Schuck, P. [*The Nuclear Many-Body Problem*]{}; Springer: Berlin/Heidelberg, Germany, 1980. Miller, D.E.; Chin, J.K.; Stan, C.A.; Liu, Y.; Setiawan, W.; Sanner, C.; Ketterle, W. Critical Velocity for Superfluid Flow across the BEC-BCS Crossover. [*Phys. Rev. Lett.*]{} [**2007**]{}, [*99*]{}, 070402. Zhang, S.L.; Zhou, Z.W.; Wu, B. Superfluidity and stability of a Bose–Einstein condensate with periodically modulated interatomic interaction. [*Phys. Rev. A*]{} [**2013**]{}, [*87*]{}, 013633. Dasgupta, R.; Venkatesh, B.P.; Watanabe, G. Attraction-induced dynamical stability of a Bose–Einstein condensate in a nonlinear lattice. [**2016**]{} arXiv:1603.07486 \[cond-mat.quant-gas\]. Sakaguchi, H.; Malomed, B.A. Matter-wave solitons in nonlinear optical lattices. [*Phys. Rev. A*]{} [**2005**]{}, [*72*]{}, 046610. Abdullaev, F.; Abdumalikov, A.; Galimzyanov, R. Gap solitons in Bose–Einstein condensates in linear and nonlinear optical lattices. [*Phys. Lett. A*]{} [**2007**]{}, [*367*]{}, 149–155. Yu, D.; Yi, W.; Zhang, W. Swallowtail Structure in Fermi Superfluids with Periodically Modulated Interactions. [*Phys. Rev. A*]{} [**2015**]{}, [*92*]{}, 033623. Fatemi, F.K.; Jones, K.M.; Lett, P.D. Observation of optically induced Feshbach resonances in collisions of cold atoms. [*Phys. Rev. Lett.*]{} [**2000**]{}, [*85*]{}, 4462–4465. Chin, C.; Grimm, R.; Julienne, P.; Tiesinga, E. Feshbach resonances in ultracold gases. [*Rev. Mod. Phys.*]{} [**2010**]{}, [*82*]{}, 1225–1286. Fedichev, P.O.; Kagan, Yu.; Shlyapnikov, G.V.; Walraven, J.T.M. Influence of nearly resonant light on the scattering length in low-temperature atomic gases. [*Phys. Rev. Lett.*]{} [**1996**]{}, [*77*]{}, 2913–2916. Yamazaki, R.; Taie, S.; Sugawa, S.; Takahashi, Y. Submicron spatial modulation of an interatomic interaction in a Bose–Einstein condensate. [*Phys. Rev. Lett.*]{} [**2010**]{}, [*105*]{}, 050405. Lewenstein, M.; Sanpera, A.; Ahufinger, V. [*Ultracold Atoms in Optical Lattices: Simulating Quantum Many-body Systems*]{}; Oxford University Press: Oxford, UK, 2012.

ArXiv

--- abstract: | We obtained thermal equilibrium solutions for optically thin, two-temperature black hole accretion disks incorporating magnetic fields. The main objective of this study is to explain the bright/hard state observed during the bright/slow transition of galactic black hole candidates. We assume that the energy transfer from ions to electrons occurs via Coulomb collisions. Bremsstrahlung, synchrotron, and inverse Compton scattering are considered as the radiative cooling processes. In order to complete the set of basic equations, we specify the magnetic flux advection rate instead of $\beta = p_{\rm gas}/p_{\rm mag}$. We find magnetically supported (low-$\beta$), thermally stable solutions. In these solutions, the total amount of the heating via the dissipation of turbulent magnetic fields goes into electrons and balances the radiative cooling. The low-$\beta$ solutions extend to high mass accretion rates ($\gtrsim \alpha^{2} {\dot M}_{\rm Edd}$) and the electron temperature is moderately cool ($T_{\rm e} \sim 10^{8} - 10^{9.5} {\rm K}$). High luminosities ($ \gtrsim 0.1 L_{\rm Edd}$) and moderately high energy cutoffs in the X-ray spectrum ($ \sim 50 - 200 ~ {\rm keV}$) observed in the bright/hard state can be explained by the low-$\beta$ solutions. author: - 'H. Oda, M. Machida, K.E. Nakamura and R. Matsumoto' title: 'Thermal Equilibria of Optically Thin, Magnetically Supported, Two-Temperature, Black Hole Accretion Disks' --- Introduction ============ Galactic black hole candidates (BHCs) are known to exhibit transitions between different X-ray spectral states. Typically, a transient outburst begins in the low/hard state at a low luminosity. The X-ray spectrum in the low/hard state is roughly described by a hard power law with a high energy cutoff at $\sim 200 ~ {\rm keV}$. As the luminosity increases, these systems undergo a transition to the high/soft state (so-called a hard-to-soft transition). The X-ray spectrum in the high/soft state is dominated by the disk emission of characteristic temperature $\sim 1 ~ {\rm keV}$. Recently, two distinct types of hard-to-soft transitions, the bright/slow transition and the dark/fast transition, are reported [e.g., @bell06; @gier06]. The bright/slow transition occurs at $\sim 0.3 ~ L_{\rm Edd}$ and takes more than $30$ days. The system undergoes a transition from the low/hard state to the high/soft state via the “bright/hard” state and the very high/steep power law (VH/SPL) state during the bright/slow transition. The X-ray spectrum in the bright/hard state is described by a hard power-law with a (moderately) high energy cutoff at $\sim 50 - 200 ~ {\rm keV}$ and the luminosity is “brighter” than that in the low/hard state. The dark/fast transition occurs at less than $0.1 ~ L_{\rm Edd}$ and takes less than $15$ days. The system immediately switches from the low/hard state to the high/soft state during the dark/fast transition. Here, $L_{\rm Edd} = 4 \pi c G M / \kappa_{\rm es} \sim 1.47 \times 10^{39} \left(M /10 M_\sun \right) \left(\kappa_{\rm es} / 0.34 ~ {\rm cm}^2 ~ {\rm g}^{-1} \right)^{-1} {\rm erg} ~ {\rm s}^{-1}$ is the Eddington luminosity, $M$ is the black hole mass, and $\kappa_{\rm es}$ is the electron scattering opacity. [@miya08] analyzed the results of [*RXTE*]{} observations of the BHC GX 339-4 in the rising phases of the transient outbursts (this object showed the bright/slow transition in the 2002/2003 outburst and the dark/fast transition in the 2004 outburst). They found that the cutoff energy strongly anti-correlates with the luminosity and decreases from $\sim 200 ~ {\rm keV}$ to $\sim 50 ~ {\rm keV}$ in the bright/hard state, while the cutoff energy is roughly constant at $\sim 200 ~ {\rm keV}$ in the low/hard state. This suggests that the electron temperature of an accretion disk emitting hard X-rays decreases as the luminosity increases in the bright/hard state. Furthermore, the bright/hard state has been observed from $\sim 0.07 ~ L_{\rm Edd}$ up to $\sim 0.3 ~ L_{\rm Edd}$. They concluded that such anti-correlation is explained by the scenario that the heating rate from protons to electrons via the Coulomb collision balances the radiative cooling rate of inverse Compton scattering. The main purpose of this paper is to present a model explaining the bright/hard state. In the conventional theory of accretion disks, the concept of phenomenological $\alpha$-viscosity is introduced. In this framework, the $\varpi \varphi$-component of the stress tensor, which appears in the angular momentum equation and the viscous heating term, is assumed to be proportional to the gas pressure, $t_{\varpi \varphi} = -\alpha_{\rm SS} p_{\rm gas}$ (we ignore the radiation pressure in this paper because we focus on optically thin disks). Here $\alpha_{\rm SS}$ is the viscosity parameter introduced by [@shak73]. The magnetic field can be an origin of the $\alpha$-viscosity because the Maxwell stress generated by the magneto-rotational instability (MRI) efficiently transports angular momentum in accretion disks [e.g., @balb91] and the dissipation of magnetic energy contributes the disk heating [e.g., @hiro06]. Optically thin, hot accretion disks have been studied to explain hard X-rays from BHCs. [@thor75] proposed that hard X-rays from Cyg X-1 are produced in an inner optically thin hot disk. [@shib75] studied the structure and stability of optically thin hot accretion disks. We have to consider two-temperature plasma in such disks because the electron temperature is expected to become lower than the ion temperature in such a low density, high temperature region. The energy equations for ions and electrons are written in the form $$\begin{aligned} \rho_{\rm i} T_{\rm i} \frac{d S_{\rm i}}{d t} = \left(1 - \delta_{\rm heat} \right) q^{+} - q^{\rm ie} ~, \label{eq:org_ene_i}\end{aligned}$$ $$\begin{aligned} \rho_{\rm e} T_{\rm e} \frac{d S_{\rm e}}{d t} = \delta_{\rm heat} q^{+} + q^{\rm ie} - q_{\rm rad}^{-} ~, \label{eq:org_ene_e}\end{aligned}$$ where $q^{+}$ is the viscous heating rate, $q^{\rm ie}$ is the energy transfer rate from ions to electrons via Coulomb collisions, $q_{\rm rad}^{-}$ is the radiative cooling rate, and $\delta_{\rm heat}$ is the fraction of heating to electrons. The left hand-sides represent the heat advection terms for ions ($q_{\rm ad,i}$) and electrons ($q_{\rm ad,e}$). We note that early works on optically thin, hot, two-temperature accretion disks assumed that the viscous heating acts primarily on ions ($\delta_{\rm heat} \ll 1$). [@eard75] and @shap76 [hereafter SLE] constructed a model for optically thin two-temperature accretion disks. In the SLE solutions, the dissipated energy is transferred from ions to electrons via Coulomb collisions ($q^{+} \sim q^{\rm ie}$) and radiated away by electrons ($q^{\rm ie} \sim q_{\rm rad}^{-}$). Although the electron temperature is high enough to explain the X-ray spectrum in the low/hard state, the SLE solutions are thermally unstable in the framework of the $\alpha$-prescription of viscosity. [@ichi77] pointed out the importance of heat advection in hot, magnetized accretion flows, and obtained steady solutions of optically thin disks. Such geometrically thick, optically thin, advection-dominated accretion flows (ADAFs) or radiatively inefficient accretion flows (RIAFs) have been studied extensively by @nara94 [[-@nara95]] and [@abra95]. In the ADAF/RIAF solutions, a substantial fraction of the dissipated energy is stored in the gas as entropy and advected into the central object ($q^{+} \sim q_{\rm ad,i}$). Only a small fraction of the dissipated energy is transferred to electrons and radiated away. The ADAF/RIAF solutions are thermally stable. [@esin97] found that the maximum mass accretion rate of the ADAF/RIAF solutions is ${\dot M}_{\rm c,A} \sim 1.3 \alpha^2 ~ {\dot M}_{\rm Edd}$ which corresponds to $L \sim 0.4 \alpha^2 L_{\rm Edd}$, where ${\dot M}_{\rm Edd}$ is the Eddington mass accretion rate. [@esin98] showed that the electron temperature in the ADAF/RIAF solutions weakly anti-correlates with the luminosity and decreases to $\sim 10^{9.5} {\rm K}$. These features are consistent with the facts that the energy cutoff weakly anti-correlates with the luminosity around $\sim 200 ~ {\rm keV}$ in the low/hard state, and that these systems undergo a transition from the low/hard state to other X-ray spectral states (i.e., the high/soft state during the dark/fast transition, and the bright/hard state during the bright/slow transition) about at this maximum luminosity of the ADAF/RIAF solutions. However, the ADAF/RIAF solutions cannot explain the strong anti-correlation in the range of $L \gtrsim 0.1 ~ L_{\rm Edd}$ and $T_{\rm e} \lesssim 200 ~ {\rm keV}$ observed in the bright/hard state. The heat advection works as an effective cooling in the ADAF/RIAF solutions. [@yuan01; @yuan03a] presented a luminous hot accretion flow (LHAF) in which the heat advection for ions works as an effective heating. Above the maximum mass accretion rate of the ADAF/RIAF solutions, the heat advection overwhelms the viscous heating and balances the energy transfer from ions to electrons ($q_{\rm ad,i} \sim q^{\rm ie}$). The LHAFs are thermally unstable. However, [@yuan03b] concluded that the thermal instability will have no effect on the dynamics of the LHAFs because the accretion timescale is shorter than the timescale of growth of the local perturbation at such high mass accretion rate. The LHAF solutions also cannot explain the bright/hard state because the electron temperature is high and roughly constant at $\sim 10^{9.5} ~ {\rm K}$. In these models mentioned above, magnetic fields are not considered explicitly, and the ratio of the gas pressure to the magnetic pressure is assumed to be constant (typically, $\beta = p_{\rm gas} / p_{\rm mag} \sim 1$). [@shib90] suggested that an accretion disk evolves toward two types of disks, a high-$\beta$ disk and a low-$\beta$ disk, by carrying out two-dimensional magnetohydrodynamic (MHD) simulations of the buoyant escape of the magnetic flux owing to the Parker instability [@park66]. In the high-$\beta$ disk, the magnetic flux escapes from the disk owing to the Parker instability and $\beta$ inside the disk is maintained at a high value. Global three-dimensional MHD simulations of optically thin, radiatively inefficient accretion disks also indicate that the amplification of magnetic fields becomes saturated when $\beta \sim 10$ in a quasi-steady state except in the plunging region very close to the black hole [e.g., @hawl00; @mach00; @hawl01; @mach03; @mach04]. On the other hand, once a disk is dominated by the magnetic pressure, it can stay in the low-$\beta$ state because the strong magnetic tension suppresses the growth of the Parker instability. [@mach06] demonstrated that an optically thin, radiatively inefficient, hot, high-$\beta$ (ADAF/RIAF-like) disk undergoes transition to an optically thin, radiatively efficient, cool, low-$\beta$ disk except in the plunging region when the mass accretion rate exceeds the threshold for the onset of a cooling instability. During this transition, the magnetic flux $\langle B_{\varphi} \rangle H$ is almost conserved at each radius because the cooling timescale is shorter than that of the buoyant escape of the magnetic flux, where $\langle B_{\varphi} \rangle$ is the mean azimuthal magnetic field and $H$ is the half thickness of the disk. In this way, the magnetic pressure becomes dominant and supports the disk as the gas pressure decreases owing to the cooling instability. Eventually, the disk stays in a quasi-steady, cool, low-$\beta$ state. Because the MRI is not yet stabilized in this quasi-steady state, the magnetic field still remains turbulent and dominated by the azimuthal component. As a result, the heating owing to the dissipation of the turbulent magnetic field balances the radiative cooling. [@joha08] performed local three-dimensional MHD simulations of strongly magnetized, vertically stratified accretion disks in a Keplerian potential. They showed that strongly magnetized state is maintained near the equatorial plane because the buoyantly escaping magnetic flux is replenished by stretching of a radial field. The MRI feeds off both vertical and azimuthal fields and drives turbulence. The Maxwell and Reynolds stresses generated by the turbulence become significant. Therefore, they indicated that highly magnetized disks are astrophysically viable. We note that such low-$\beta$ disks are quite different from magnetically dominated accretion flows [MDAFs; @meie05] observed in the plunging region of optically thin accretion disks in global MHD and general relativistic MHD simulations [e.g., @frag09] in terms of their energy balance and configuration of magnetic fields. Outside the plunging region, the magnetic field becomes turbulent and dominated by the azimuthal component because the growth timescale of the MRI is shorter than the inflow timescale. As a result, the released gravitational energy is efficiently converted into the thermal energy via the dissipation of the turbulent magnetic field. On the other hand in the plunging region, the ratio of the timescales is reversed because the inflow velocity increases with decreasing the radius in such disks. Therefore, magnetic field lines are stretched out in the radial direction before turbulence is generated by the MRI and is dissipated. As a result, a substantial fraction of the gravitational energy is converted into the radial infall kinetic energy. Since the heating owing to the dissipation of turbulent magnetic fields becomes inefficient, the gas pressure becomes low, and the flow becomes magnetically dominated. Although both of the low-$\beta$ disk and the MDAF are cool and magnetic pressure dominant, they are essentially different. We focus on the low-$\beta$ disk in this paper. [@oda07] constructed a steady model of optically thin, one-temperature accretion disks incorporating magnetic fields on the basis of these results of three-dimensional MHD simulations of accretion disks. [@oda09] extended it to the optically thick regime. They assumed that the $\varpi \varphi$-component of the stress tensor is proportional to the total pressure. In order to complete the set of basic equations, they specified the advection rate of the azimuthal magnetic flux instead of $\beta$. They found a new thermally stable solution, a low-$\beta$ solution, which can explain the results by [@mach06]. In the low-$\beta$ solutions, the magnetic heating enhanced by the strong magnetic pressure balances the radiative cooling. The disk temperature is lower than that in the ADAF/RIAF solutions and strongly anti-correlates with the mass accretion rate. They also found that the low-$\beta$ solutions exist above the maximum mass accretion rate of the ADAF/RIAF solutions. Therefore, they concluded that the optically thin low-$\beta$ disk can qualitatively explain the bright/hard state. However, they considered one-temperature plasma and bremsstrahlung emission as a radiative cooling mechanism. It is expected that the electron temperature becomes lower than the ion temperature and that synchrotron cooling and/or inverse Compton scattering become effective in such disks. In this paper, we extend the model of optically thin, one-temperature disk to that of optically thin, two-temperature disks. We consider synchrotron emission and inverse Compton scattering as a cooling mechanism in addition to bremsstrahlung emission. We obtained the thermal equilibrium curves and found that the optically thin low-$\beta$ solutions can quantitatively explain the bright/hard state. The basic equations are presented in Section \[basic\_eq\]. In Section \[results\], we present the thermal equilibrium curves. Section \[discussion\] is devoted to a discussion. We summarize the paper in Section \[summary\]. Models and Assumptions {#basic_eq} ====================== Basic Equations --------------- We extended the basic equations for one-dimensional steady, optically thin, two-temperature black hole accretion flows [e.g., @kato08] incorporating magnetic fields. We adopt cylindrical coordinates $(\varpi,\varphi,z)$. General relativistic effects are simulated using the pseudo-Newtonian potential $\psi = -GM/(r-r_{\rm s})$ [@pacz80], where $G$ is the gravitational constant, $M$ is the black hole mass (we assume $M = 10 M_\sun$ in this paper), $r = (\varpi^2 + z^2)^{1/2}$, and $r_{\rm s} = 2GM/c^2 $ is the Schwarzschild radius. For simplicity, the gas is assumed to consist of protons (ions) and electrons. The number density of ions and electrons are equal by charge neutrality, $n = n_{\rm i} = n_{\rm e}$. We start with the resistive MHD equations $$\begin{aligned} \frac{\partial \rho}{\partial t} + \nabla \cdot \left( \rho \mbox{\boldmath{$v$}} \right) = 0 ~, \label{eq:vec_con}\end{aligned}$$ $$\begin{aligned} \rho \left[ \frac{\partial \mbox{\boldmath{$v$}}}{\partial t} + \left( \mbox{\boldmath{$v$}} \cdot \nabla \right) \mbox{\boldmath{$v$}} \right] = - \rho \nabla \psi - \nabla p_{\rm gas} + \frac{\mbox{\boldmath{$j$}} \times \mbox{\boldmath{$B$}}}{c} ~, \label{eq:vec_mom}\end{aligned}$$ $$\begin{aligned} \frac{\partial \left( \rho_{\rm i} \epsilon_{\rm i} \right)}{\partial t} + \nabla \cdot \left[ \left( \rho_{\rm i} \epsilon_{\rm i} + p_{\rm i} \right) \mbox{\boldmath{$v$}} \right] - \left( \mbox{\boldmath{$v$}} \cdot \nabla \right) p_{\rm i} \nonumber \\ = \left(1 - \delta_{\rm heat} \right) q^{+} - q^{\rm ie} ~, \label{eq:vec_ene_i}\end{aligned}$$ $$\begin{aligned} \frac{\partial \left( \rho_{\rm e} \epsilon_{\rm e} \right)}{\partial t} + \nabla \cdot \left[ \left( \rho_{\rm e} \epsilon_{\rm e} + p_{\rm e} \right) \mbox{\boldmath{$v$}} \right] - \left( \mbox{\boldmath{$v$}} \cdot \nabla \right) p_{\rm e} \nonumber \\ = \delta_{\rm heat} q^{+} + q^{\rm ie} - q_{\rm rad}^{-} ~, \label{eq:vec_ene_e}\end{aligned}$$ $$\begin{aligned} \frac{\partial \mbox{\boldmath{$B$}}}{\partial t} = \nabla \times \left( \mbox{\boldmath{$v$}} \times \mbox{\boldmath{$B$}} - \frac{4 \pi}{c}\eta_{\rm m} \mbox{\boldmath{$j$}} \right) ~, \label{eq:vec_ind}\end{aligned}$$ where $\rho = \rho_{\rm i} + \rho_{\rm e}$ is the density, $\rho_{\rm i} = m_{\rm i} n$ and $\rho_{\rm e} = m_{\rm e} n$ are the ion and electron densities, $m_{\rm i}$ and $m_{\rm e}$ are the ion and electron masses, $\mbox{\boldmath{$v$}}$ is the velocity, $\mbox{\boldmath{$B$}}$ is the magnetic field, $\mbox{\boldmath{$j$}} = c \nabla \times \mbox{\boldmath{$B$}} / 4 \pi$ is the current density, $p_{\rm gas} = p_{\rm i} + p_{\rm e} = n k \left(T_{\rm i} + T_{\rm e} \right)$ is the gas pressure, $p_{\rm i}$ and $p_{\rm e}$ are the ion and electron gas pressure, $T_{\rm i}$ and $T_{\rm e}$ are the ion and electron temperature, $k$ is the Boltzmann constant, $\epsilon_{\rm i} = \left( p_{\rm i} / \rho_{\rm i} \right) / \left( \gamma_{\rm i} -1 \right)$ and $\epsilon_{\rm e} = \left(p_{\rm e} / \rho_{\rm e}\right) / \left(\gamma_{\rm e} -1 \right)$ are the internal energy of ions and electrons. Here, $\gamma_{\rm i} = 5/3$ and $\gamma_{\rm e} = \gamma_{\rm e} \left(T_{\rm e}\right)$ are the specific heat ratio for ions and electrons. In the energy equations for ions (\[eq:vec\_ene\_i\]) and electrons (\[eq:vec\_ene\_e\]), $q^{+}$ is the heating rate, $q_{\rm rad}^{-}$ is the radiative cooling rate, and $q^{\rm ie}$ is the energy transfer rate from ions to electrons via Coulomb collisions. Here, $\delta_{\rm heat}$ represents the fraction of heating to electrons. In the induction equation (\[eq:vec\_ind\]), $\eta_{\rm m} \equiv c^2/4 \pi \sigma_{\rm c}$ is the magnetic diffusivity, where $\sigma_{\rm c}$ is the electric conductivity. ### Azimuthally Averaged Equations Three-dimensional global MHD and local radiation-MHD simulations of black hole accretion disks showed that magnetic fields inside the disk are turbulent and dominated by the azimuthal component in a quasi-steady state [e.g., @mach06; @hiro06]. On the basis of results of the simulations, we decomposed the magnetic fields into the mean fields $\mbox{\boldmath{$\bar{B}$}} = \left( 0, \bar{B_{\varphi}}, 0 \right)$ and fluctuating fields $\delta \mbox{\boldmath{$B$}} = \left( \delta B_{\varpi}, \delta B_{\varphi}, \delta B_{z} \right)$ and also decomposed the velocity into the mean velocity $\mbox{\boldmath{$\bar{v}$}} = (v_{\varpi}, v_{\varphi}, v_{z})$ and the fluctuating velocity $\delta \mbox{\boldmath{$v$}} = \left(\delta v_{\varpi}, \delta v_{\varphi}, \delta v_{z} \right)$. We assume that the fluctuating components vanish when azimuthally averaged, $\langle \delta \mbox{\boldmath{$v$}} \rangle = \langle \delta \mbox{\boldmath{$B$}} \rangle = 0$, and that the radial and vertical components of the magnetic fields are negligible compared with that of the azimuthal component, $|\bar{B_{\varphi}} + \delta B_{\varphi}| \gg |\delta B_{\varpi}|$, $|\delta B_{z}|$. Here $\langle ~~ \rangle$ denotes the azimuthal average. Let us derive the azimuthally averaged equations. We assume that the disk is in a steady state and in hydrostatic balance in the vertical direction. By azimuthally averaging equations (\[eq:vec\_con\]) - (\[eq:vec\_ind\]) and ignoring the second order terms of $\delta \mbox{\boldmath{$v$}}$, $\delta B_{\varpi}$, and $\delta B_{z}$, we obtain $$\label{eq:con} \frac{1}{\varpi} \frac{\partial}{\partial \varpi} \left( \varpi \rho v_{\varpi} \right) + \frac{\partial}{\partial z} \left( \rho v_{z} \right) = 0 ~,$$ $$\label{eq:mom_pi} \rho v_{\varpi} \frac{\partial v_{\varpi}}{\partial \varpi} + \rho v_{z} \frac{\partial v_{\varpi}}{\partial z} - \frac{\rho v_{\varphi}^{2}}{\varpi} = - \rho \frac{\partial \psi}{\partial \varpi} - \frac{\partial p_{\rm tot}}{\partial \varpi} - \frac{\langle B_{\varphi}^2 \rangle}{4 \pi \varpi } ~,$$ $$\begin{aligned} \label{eq:mom_phi} \rho v_{\varpi} \frac{\partial v_{\varphi}}{\partial \varpi} + \rho v_{z} \frac{\partial v_{\varphi}}{\partial z} + \frac{\rho v_{\varpi} v_{\varphi}} {\varpi} \nonumber \\ = \frac{1}{{\varpi}^{2}} \frac{\partial}{\partial \varpi} \left[ {\varpi}^{2} \frac{ \langle B_{\varpi} B_{\varphi} \rangle}{4\pi} \right] + \frac{\partial}{\partial z} \left( \frac{\langle B_{\varphi} B_{z} \rangle}{4 \pi}\right)~,\end{aligned}$$ $$\label{eq:mom_z} 0 = - \frac{\partial \psi}{\partial z} - \frac{1}{\rho} \frac{\partial p_{\rm tot}}{\partial z} ~,$$ $$\begin{aligned} \label{eq:ene_i} \frac{\partial}{\partial \varpi} \left[ \left( \rho_{\rm i} \epsilon_{\rm i} + p_{\rm i}\right) v_{\varpi}\right] + \frac{v_{\varpi}}{\varpi} \left( \rho_{\rm i} \epsilon_{\rm i} + p_{\rm i} \right) + \frac{\partial}{\partial z} \left[ \left( \rho_{\rm i} \epsilon_{\rm i} + p_{\rm i} \right) v_{z}\right] \nonumber \\ - v_{\varpi} \frac{\partial}{\partial \varpi} p_{\rm i} - v_{z} \frac{\partial}{\partial z} p_{\rm i} = \left( 1 - \delta_{\rm heat} \right) q^{+} - q^{\rm ie} ~,\end{aligned}$$ $$\begin{aligned} \label{eq:ene_e} \frac{\partial}{\partial \varpi} \left[ \left( \rho_{\rm e} \epsilon_{\rm e} + p_{\rm e}\right) v_{\varpi}\right] + \frac{v_{\varpi}}{\varpi} \left( \rho_{\rm e} \epsilon_{\rm e} + p_{\rm e} \right) + \frac{\partial}{\partial z} \left[ \left( \rho_{\rm e} \epsilon_{\rm e} + p_{\rm e} \right) v_{z}\right] \nonumber \\ - v_{\varpi} \frac{\partial}{\partial \varpi} p_{\rm e} - v_{z} \frac{\partial}{\partial z} p_{\rm e} = \delta_{\rm heat} q^{+} + q^{\rm ie} - q^{-}_{\rm rad} ~,\end{aligned}$$ $$\begin{aligned} \label{eq:ind_phi} 0 = -\frac{\partial}{\partial z} \left[ v_{z} \langle B_{\varphi} \rangle \right] -\frac{\partial}{\partial \varpi} \left[ v_{\varpi} \langle B_{\varphi} \rangle \right] \nonumber \\ + \{ \nabla \times \langle \delta \mbox{\boldmath{$v$}} \times \delta \mbox{\boldmath{$B$}} \rangle \}_{\varphi} - \{ \eta_{\rm m} \nabla \times \left( \nabla \times \mbox{\boldmath{$\bar{B}$}} \right) \}_{\varphi} ~,\end{aligned}$$ where $p_{\rm tot} = p_{\rm gas} + p_{\rm mag}$ is the total pressure and $ p_{\rm mag} = \langle B_{\varphi}^{2} \rangle /8 \pi$ is the azimuthally averaged magnetic pressure. The third and fourth terms on the right-hand side of Equation (\[eq:ind\_phi\]) represent the dynamo term and the magnetic diffusion term which we approximate later on the basis of the results of the numerical simulations. ### Vertically Integrated, Azimuthally Averaged Equations We assume that the radial velocity $v_{\varpi}$, the specific angular momentum $\ell = \varpi v_{\varphi}$, and $\beta \equiv p_{\rm gas} /p_{\rm mag}$ are independent of $z$, and that the disks are isothermal in the vertical direction. Under these assumptions, the surface density $\Sigma$, the vertically integrated total pressure $W_{\rm tot}$, and the half thickness of the disk $H$ are defined as $$\begin{aligned} \label{eq:si} \Sigma \equiv \int_{-\infty}^{\infty}\rho dz = \int_{-\infty}^{\infty}\rho_0\exp\left(-\frac{1}{2}\frac{z^2}{H^2}\right) dz = \sqrt{2 \pi} \rho_0 H ~,\end{aligned}$$ $$\begin{aligned} \label{eq:wtot} W_{\rm tot} &\equiv& \int_{-\infty}^{\infty} p_{\rm tot}dz \nonumber \\ &=& \int_{-\infty}^{\infty}p_{{\rm tot} 0} \exp \left(-\frac{1}{2}\frac{z^2}{H^2} \right) dz = \sqrt{2 \pi} p_{{\rm tot} 0} H ~, \end{aligned}$$ $$\begin{aligned} \label{eq:h2} \Omega_{{\rm K} 0}^2 H^2 = \frac{W_{\rm tot} }{\Sigma} ~,\end{aligned}$$ where $\Omega_{{\rm K}0}=(GM/\varpi)^{1/2}/(\varpi -r_{\rm s})$ is the Keplerian angular velocity. Here the subscript $0$ refers to quantities in the equatorial plane. Using the equation of state for the ideal gas, the vertically integrated total pressure is expressed as $$\begin{aligned} \label{eq:eos} W_{\rm tot} = W_{\rm gas} + W_{\rm mag} = \frac{k T_{\rm i} + k T_{\rm e}}{m_{\rm i} + m_{\rm e}} \Sigma \left(1 + \beta^{-1} \right) ~. \end{aligned}$$ Now we integrate the other basic equations in the vertical direction. We obtain $$\begin{aligned} \label{eq:con_int} \dot{M} = -2\pi\varpi\Sigma v_\varpi ~ ,\end{aligned}$$ $$\begin{aligned} \label{eq:mom_pi_int} v_{\varpi} \frac{\partial v_{\varpi}}{\partial \varpi} + \frac{1}{\Sigma} \frac{\partial W_{\rm tot}}{\partial \varpi} \nonumber \\ = \frac{{\ell}^2 - \ell_{\rm K 0}^2}{\varpi^3} - \frac{W_{\rm tot}}{\Sigma} \frac{\partial \ln \Omega_{\rm K 0}}{\partial \varpi} -\frac{2 \beta^{-1}}{1+\beta^{-1}} \frac{W_{\rm tot}}{\Sigma} \frac{1}{\varpi} ~,\end{aligned}$$ $$\begin{aligned} \label{eq:mom_phi_int} \dot M(\ell - \ell_{\rm in})= -2\pi \varpi^2 \int_{-\infty}^{\infty} \frac{\langle B_{\varpi} B_{\varphi} \rangle }{4 \pi} dz ~,\end{aligned}$$ $$\begin{aligned} \label{eq:ene_i_int} Q_{\rm ad,i}= \left( 1 - \delta_{\rm heat} \right) Q^{+} - Q^{\rm ie} ~,\end{aligned}$$ $$\begin{aligned} \label{eq:ene_e_int} Q_{\rm ad,e} = \delta_{\rm heat} Q^{+} + Q^{\rm ie} - Q_{\rm rad}^{-} ~,\end{aligned}$$ $$\begin{aligned} \label{eq:ind_int} \dot \Phi &\equiv& \int_{-\infty}^{\infty}v_{\varpi} \langle B_{\varphi} \rangle dz \nonumber \\ &=& \int_{\varpi}^{\varpi_{\rm out}} \int_{-\infty}^{\infty} [ \{ \nabla \times \langle \delta \mbox{\boldmath{$v$}} \times \delta \mbox{\boldmath{$B$}} \rangle \}_{\varphi} \nonumber \\ &-& \{ \eta_{\rm m} \nabla \times \left( \nabla \times \mbox{\boldmath{$\bar{B}$}} \right) \}_{\varphi} ] d\varpi dz + \mbox{const.}\end{aligned}$$ where $\dot M$ is the mass accretion rate, $\ell_{{\rm K} 0} = \varpi^2 \Omega_{{\rm K} 0}$ is the Keplerian angular momentum and $\ell_{\rm in}$ is the specific angular momentum swallowed by the black hole. In the energy equations, $Q_{\rm ad,i}$ and $Q_{\rm ad,e}$ are the vertically integrated heat advection terms for ions and electrons, $Q^{+}$, $Q_{\rm rad}^{-}$, and $Q^{\rm ie}$ are the vertically integrated heating rate, radiative cooling rate, and energy transfer rate from ions to electrons via Coulomb collisions. In Equation (\[eq:ind\_int\]), $\dot \Phi$ is the radial advection rate of the azimuthal magnetic flux (hereafter we call it the magnetic flux advection rate). In this paper, we assume that $\ell = \ell_{\rm K 0}$ instead of Equation (\[eq:mom\_pi\_int\]) because we focus on local thermal equilibrium solutions. $\alpha$-Prescription of the Maxwell Stress Tensor -------------------------------------------------- Global MHD simulations of radiatively inefficient, accretion flows [e.g., @hawl01; @mach06] showed that the ratio of the azimuthally averaged Maxwell stress to the sum of the azimuthally averaged gas pressure and magnetic pressure is nearly constant ($\alpha_{\rm B} \equiv - \langle B_{\varpi} B_{\varphi} / 4 \pi \rangle / \langle p_{\rm gas} + p_{\rm mag} \rangle \sim 0.05 - 0.1$) except in the plunging region. Following the simulation results, we assume that the azimuthally averaged $\varpi \varphi$-component of the Maxwell stress inside a disk is proportional to the total (gas and magnetic) pressure $$\begin{aligned} \label{eq:al} \frac{\langle B_{\varpi} B_{\varphi} \rangle }{4 \pi} = - \alpha p_{\rm tot}~.\end{aligned}$$ Integrating in the vertical direction, we obtain $$\begin{aligned} \label{eq:al_int} \int_{-H}^{H} \frac{\langle B_{\varpi} B_{\varphi} \rangle }{4 \pi} dz= - \alpha W_{\rm tot}~.\end{aligned}$$ This is one of the key assumptions in this paper. When the magnetic pressure is high, the stress level can be high even though the gas pressure is low. We can rewrite this relation in terms of the kinematic viscosity, $\nu$, as $$\begin{aligned} \label{eq2:nu} \nu = A_{\nu} \alpha \sqrt{{c_{{\rm s}0}}^2 + {c_{{\rm A}0}}^2} H\end{aligned}$$ where $$\begin{aligned} \label{eq2:anu} A_{\nu} \equiv - \left( \frac{\Omega}{\Omega_{{\rm K} 0}} \frac{d \ln \Omega}{d \ln \varpi}\right)^{-1} ~,\end{aligned}$$ $c_{{\rm s}0}=\sqrt{p_{{\rm gas}0}/ \rho_0}$ is the sound speed, and $c_{{\rm A}0} = \sqrt{2 p_{{\rm mag}0}/\rho_0}$ is the Alfvén speed. Prescription of the Magnetic Flux Advection Rate ------------------------------------------------ We complete the set of basic equations by specifying the radial distribution of the magnetic flux advection rate. If we perform the integration in the second term of the induction equation (\[eq:ind\_int\]), we obtain $$\begin{aligned} \label{eq:ind_int_ignore} \dot \Phi &\equiv& - v_{\varpi} B_0(\varpi) \sqrt{4 \pi} H \\ &=& \left[\mbox{dynamo and diffusion terms}\right] + \mbox{const.} \nonumber\end{aligned}$$ where $$\begin{aligned} \label{eq:b0} B_0(\varpi) = 2^{5/4} \pi^{1/4} \left(\frac{k T_{\rm i} + k T_{\rm e}}{m_{\rm i} + m_{\rm e}} \right)^{1/2}\Sigma^{1/2}H^{-1/2}\beta^{-1/2}\end{aligned}$$ is the mean azimuthal magnetic field in the equatorial plane. According to the result of the global three-dimensional MHD simulation by [@mach06], the magnetic flux advection rate at a radius is roughly unchanged before and after the transition from the ADAF/RIAF-like disk to the low-$\beta$ disk. Hence, we adopt the magnetic flux advection rate as the parameter in order to complete the set of the basic equations. The magnetic flux advection rate depends on various mechanisms such as the escape of magnetic fluxes due to the magnetic buoyancy, the regeneration of azimuthal magnetic fields by the shear motion, the generation of magnetic turbulence through the MRI, dissipation of magnetic fields due to the magnetic diffusivity, and magnetic reconnection. If the sum of the dynamo term and the magnetic diffusion term is zero in the whole region, the magnetic flux advection rate is constant in the radial direction. The global three-dimensional MHD simulation performed by [@mach06] indicated that the magnetic advection rate increases with decreasing radius, specifically, $\dot \Phi \propto \varpi^{-1}$, in the quasi steady state as a result of magnetic dynamo and diffusivity processes. Because it is hard to compute the dynamo term and the magnetic diffusion term explicitly from the local quantities, we parameterize the dependence of $\dot \Phi$ on $\varpi$ by introducing a parameter, $\zeta$, as follows. $$\begin{aligned} \label{eq:phidot} \dot \Phi(\varpi; \zeta, \dot M) \equiv {\dot \Phi}_{\rm out}(\dot M) \left( \frac{\varpi}{\varpi_{\rm out}}\right)^{-\zeta} ~,\end{aligned}$$ where ${\dot \Phi}_{\rm out}$ is the magnetic flux advection rate at the outer boundary $\varpi = \varpi_{\rm out}$. When $\zeta = 0$, the magnetic flux advection rate is constant in the radial direction. When $\zeta > 0$, the magnetic flux advection rate increases with decreasing radius [See also Figure 1 in @oda09]. Here we determine the parameter ${\dot \Phi}_{\rm out}$ by imposing the outer boundary condition, $T_{\rm i,out} = T_{\rm e,out} = T_{\rm virial} = \left[ \left(m_{\rm i} + m_{\rm e}\right) c^2 / 3 k \right] \left(\varpi_{\rm out} / r_{\rm s} \right)^{-1}$ and $\beta_{\rm out} = 10$ at $\varpi_{\rm out}=1000r_{\rm s}$. This leads that ${\dot \Phi}_{\rm out} \propto {\dot M}^{1/2}$. In the local model presented in this paper, the value of $\zeta$ just means the amount of the magnetic flux advection rate at a radius (e.g, $\dot \Phi (\varpi=5r_{\rm s}) = {\dot \Phi}_{\rm out}$, $4.9 ~ {\dot\Phi}_{\rm out}$, $24 ~ {\dot\Phi}_{\rm out}$ for $\zeta = 0$, $0.3$, $0.6$, respectively). Equation (\[eq:phidot\]) is the second key assumption in this paper. Specifying the magnetic flux advection rate enables the magnetic pressure to increase when the disk temperature decreases. By contrast, if we specified the plasma $\beta$ at each radius instead of the magnetic flux advection rate, the decrease in the temperature results in a decrease in magnetic pressure. This is inconsistent with the results of three-dimensional MHD simulations [e.g., @mach06]. We address the similarity between the concepts of the mass accretion rate and the magnetic flux advection rate in order to facilitate understanding of the concept of the magnetic flux advection rate. The right-hand side of equation (\[eq:con\_int\]) represents the mass flux crossing at $\varpi$ per unit time and we have denoted it by $\dot M$. If there is no mass loss or gain (e.g, due to inflows and/or outflows), $\dot M$ is constant in the radial direction. Otherwise $\dot M$ is a function of $\varpi$ (e.g., $\dot M \propto \varpi^s$, $s$ is a parameter). The right-hand side of equation (\[eq:ind\_int\_ignore\]) represents the azimuthal magnetic flux crossing at $\varpi$ per unit time and we have denoted it by $\dot \Phi$. If there is no loss or gain of azimuthal magnetic fields (e.g, due to the dynamo and/or the magnetic diffusion), $\dot \Phi$ is constant in the radial direction. Otherwise $\dot \Phi$ is a function of $\varpi$ and we prescribed it as equation (\[eq:phidot\]). That is, when $\zeta=0$, the magnetic flux is conserved in the radial direction, and when $\zeta>0$ (or $\zeta < 0$), the magnetic flux increases (or decreases) with a decreasing radius [see Figure 1(b) in @oda09]. We note that the azimuthal magnetic flux inside a disk can increase when the azimuthal flux of opposite polarity buoyantly escapes from the disk [e.g., @nish06]. Energy Equations ---------------- ### The Magnetic Heating Rate In the conventional theory, the viscous heating was expressed as $q^{+}_{\rm vis} = t_{\varpi \varphi} \varpi \left( d \Omega / d \varpi \right)$ and assumed to heat primarily ions ($\delta_{\rm heat} \sim m_{\rm e}/m_{\rm i} \sim 10^{-3}$), where $t_{\varpi \varphi}$ is the $\varpi \varphi$-component of the total stress and $\Omega$ is the angular velocity. The results of three-dimensional MHD simulations indicate that the dissipation of the magnetic energy dominates the total dissipative heating rate throughout a disk and is expressed as $q^{+} \sim \langle B_{\varpi} B_{\varphi} / 4 \pi \rangle \varpi \left( d \Omega / d \varpi \right)$ [e.g., @hiro06; @mach06; @krol07]. Hereafter, we refer to it as the magnetic heating rate. Following these simulation results, we employ magnetic heating as the heating mechanism inside a disk, and set the vertically integrated heating rate as follows: $$\begin{aligned} \label{eq:qmag} Q^{+} = \int^{\infty}_{-\infty} \left[ \frac{\langle B_{\varpi}B_{\varphi} \rangle}{4\pi} \varpi \frac{d \Omega}{d \varpi} \right] dz = - \alpha W_{\rm tot} \varpi \frac{d \Omega}{d \varpi} ~ ,\end{aligned}$$ where we have used equation (\[eq:al\]). We note that if the magnetic pressure is high, the heating rate can also be high even when the gas pressure is low. The eventual expression of the heating term is not at all unusual [e.g., @naka97; @manm97; @yuan01] except that we consider much lower values of $\beta$. We will discuss in Section [\[discussion\_main\]]{} a lower limit of $\beta$ below which the MRI is stabilized [e.g., @pess05] so that this expression is no longer valid. We note that the magnetic heating does not always heat primarily ions. [@yuan03b] suggested that $\delta_{\rm heat} \sim 0.1 - 0.5$ is required to fit the spectrum of Sgr $\rm A^{*}$ with RIAF models. [@shar07] carried out local shearing box simulations of the nonlinear evolution of the MRI in a collisionless plasma considering pressure anisotropy and showed that $\delta_{\rm heat} \sim (1+3\sqrt{T_{\rm i}/T_{\rm e}})^{-1}$ [note that the definition of $\delta_{\rm heat} = q_{\rm e}^{+}/q^{+}$ is different from that of $\delta_{\rm heat} = q_{\rm e}^{+}/q_{\rm i}^{+}$ in @shar07]. The exact value we choose for this parameter is not so important, in particular, for the low-$\beta$ solutions. We find that the low-$\beta$ solutions presented in this paper practically unchanged for any value of $\delta_{\rm heat} \gtrsim 0.1$. Therefore, we adopt $\delta_{\rm heat} = (1+3\sqrt{T_{\rm i}/T_{\rm e}})^{-1}$ as a fiducial value. ### The Energy Transfer Rate from Ions to Electrons by Coulomb Collisions If the ion temperature is higher than the electron temperature, Coulomb collisions transfer energy from ions to electrons. The energy transfer rate from ions to electrons per unit volume via Coulomb collisions is given by [@step83] $$\begin{aligned} q^{\rm ie} = -\frac{3}{2} \frac{m_{\rm e}}{m_{\rm i}} n^2 \sigma_T c \left(\ln \Lambda\right) \frac{k T_{\rm e} - k T_{\rm i}} {K_2(1/ \theta_{\rm e}) K_2(1/ \theta_{\rm i})} \nonumber \\ \left[ \frac{2(\theta_{\rm e} + \theta_{\rm i})^2 +1}{\theta_{\rm e} + \theta_{\rm i}} K_1 \left( \frac{\theta_{\rm e} + \theta_{\rm i}}{\theta_{\rm e} \theta_{\rm i}} \right) + 2 K_0 \left( \frac{\theta_{\rm e} + \theta_{\rm i}}{\theta_{\rm e} \theta_{\rm i}} \right)\right] ,\end{aligned}$$ where $\sigma_{\rm T}$ is the Thomson scattering cross section and $\ln \Lambda$ is the Coulomb logarithm (roughly $\ln \Lambda \sim 20$). $K_n$ are modified Bessel function of the second kind of the order $n$, respectively. The dimensionless electron and ion temperatures are defined by $$\begin{aligned} \theta_{\rm e} = \frac{k T_{\rm e}}{m_{\rm e} c^2} ~,~ \theta_{\rm i} = \frac{k T_{\rm i}}{m_{\rm i} c^2} . \end{aligned}$$ For technical reason, we use the following formula which uses no special functions, and is accurate to within a factor of 2 when $\theta_{\rm i} < 0.2$ [@derm91] $$\begin{aligned} q^{\rm ie} = -\frac{3}{2} \frac{m_{\rm e}}{m_{\rm i}} n^2 \sigma_{\rm T} c \ln \Lambda (k T_{\rm e} - k T_{\rm i}) \frac{(2 \pi)^{1/2} + (\theta_{\rm e} + \theta_{\rm i})^{1/2}}{(\theta_{\rm e} + \theta_{\rm i})} .\end{aligned}$$ Integrating in the vertical direction, we obtain $$\begin{aligned} Q^{\rm ie} & = & -\frac{3}{2} \frac{m_{\rm e}}{m_{\rm i}} \frac{\Sigma^2}{2 (m_{\rm i}+m_{\rm e})^2 \sqrt{\pi} H} \sigma_{\rm T} c \ln \Lambda \\ \nonumber \\ & \times & \left\{ \begin{array}{cc} \displaystyle \frac{k T_{\rm e} - k T_{\rm i}} {K_2(1/ \theta_{\rm e}) K_2(1/ \theta_{\rm i})} \times \nonumber \\ \left[ \frac{2(\theta_{\rm e} + \theta_{\rm i})^2 +1}{\theta_{\rm e} + \theta_{\rm i}} K_1 \left( \frac{\theta_{\rm e} + \theta_{\rm i}}{\theta_{\rm e} \theta_{\rm i}} \right) + 2 K_0 \left( \frac{\theta_{\rm e} + \theta_{\rm i}}{\theta_{\rm e} \theta_{\rm i}} \right)\right] & (\theta_{\rm i} > 0.2)\\ \\ \displaystyle (k T_{\rm e} - k T_{\rm i}) \frac{(2 \pi)^{1/2} + (\theta_{\rm e} + \theta_{\rm i})^{1/2}}{(\theta_{\rm e} + \theta_{\rm i})} & (\theta_{\rm i} < 0.2) \nonumber \end{array}\right. ~ .\end{aligned}$$ We note that $\theta_{\rm i} < 0.2$ in almost all solutions presented in this paper. ### Radiative Cooling Rate We assume that the radiative cooling occurs through electrons and consider bremsstrahlung, synchrotron, and Compton cooling by bremsstrahlung and synchrotron photons as cooling processes. The vertically integrated radiative cooling rate is expressed as $$\begin{aligned} Q_{\rm rad}^{-} = Q_{\rm br}^{-} + Q_{\rm sy}^{-} + Q_{\rm br,C}^{-} + Q_{\rm sy,C}^{-} ~.\end{aligned}$$ Following [@nara95] [see also @sven82; @step83], bremsstrahlung cooling rate per unit volume is $$\begin{aligned} q_{\rm br}^{-} = q_{\rm br,ei}^{-} + q_{\rm br,ee}^{-} = n^2 \sigma_{\rm T} c \alpha_{\rm f} m_{\rm e} c^2 \left[F_{\rm ei}(\theta_{\rm e}) + F_{\rm ee}(\theta_{\rm e}) \right]~,\end{aligned}$$ where the subscripts $\rm ei$ and $\rm ee$ denote the electron-ion and electron-electron bremsstrahlung cooling rates, $\alpha_{\rm f}$ is fine-structure constant, the function $F_{\rm ei}(\theta_{\rm e})$ and $F_{\rm ee}(\theta_{\rm e})$ have the approximate form $$\begin{aligned} F_{\rm ei}(\theta_{\rm e}) = \left\{ \begin{array}{cc} \displaystyle \frac{9 \theta_{\rm e}}{2 \pi} \left[ \ln ( 2 \eta \theta_{\rm e} + 0.48) + \frac{3}{2}\right] & (\theta_{\rm e} > 1) \\ \\ \displaystyle 4 \left( \frac{2 \theta_{\rm e}}{\pi^3}\right)^{1/2} \left[ 1 + 1.781 \theta_{\rm e}^{1.34}\right] & (\theta_{\rm e} < 1) \end{array} \right. ~,\end{aligned}$$ $$\begin{aligned} F_{\rm ee}(\theta_{\rm e}) = \left\{ \begin{array}{cc} \displaystyle \frac{9 \theta_{\rm e}}{\pi} ( \ln ( 2 \eta \theta_{\rm e}) + 1.28) & (\theta_{\rm e} > 1) \\ \\ \displaystyle \frac{5}{6 \pi^{3/2}} (44 - 3 \pi^2 )\theta_{\rm e}^{3/2} \times \nonumber \\ ( 1 + 1.1 \theta_{\rm e} + \theta_{\rm e}^2 - 1.25 \theta_{\rm e}^{5/2}) & (\theta_{\rm e} < 1) \end{array}\right. ~,\end{aligned}$$ $\eta_{\rm E} = \exp(- \gamma_{\rm E}) $ and $\gamma_{\rm E} \approx 0.5772$ is Euler’s number. Integrating in the vertical direction, we obtain $$\begin{aligned} Q_{\rm br}^{-} = \sigma_{\rm T} c \alpha_{\rm f} m_{\rm e} c^2 \frac{\Sigma^2}{\left(m_{\rm i} + m_{\rm e}\right)^2 \sqrt{\pi} H} \times \nonumber \\ ~ [F_{\rm ei}(\theta_{\rm e}) + F_{\rm ee}(\theta_{\rm e})] ~.\end{aligned}$$ The synchrotron emissivity of a relativistic Maxwellian distribution of electrons in the optically thin limit is given by [@pach70]: $$\begin{aligned} \epsilon_{\rm sy} d \nu = \frac{2 e^2 n_{\rm e}}{\sqrt{3} c} \frac{2 \pi \nu}{K_2 (1/ \theta_{\rm e})} I^{\prime}\left( x_{\rm M}\right) d \nu ~, \end{aligned}$$ where $$\begin{aligned} x_{\rm M} = \frac{2 \nu}{3 \nu_{\rm b} \theta_{\rm e}^2}, ~ \nu_{\rm b} = \frac{e B_0}{2 \pi m_{\rm e} c} ~ ,\end{aligned}$$ with the fitting function $I^{\prime}\left( x_{\rm M}\right)$ given by [@maha96] $$\begin{aligned} I^{\prime}\left( x_{\rm M}\right) = \frac{4.0505}{x_{\rm M}^{1/6}} \left( 1 +\frac{0.40}{x_{\rm M}^{1/4}} +\frac{0.5316}{x_{\rm M}^{1/2}} \right) \times \nonumber \\ \exp(-1.8899 x_{\rm M}^{1/3}) ~.\end{aligned}$$ Integrating in the vertical direction, we obtain $$\begin{aligned} \label{eq4:e_sy} E_{\rm sy} d \nu = \frac{2 e^2 }{\sqrt{3} c} \frac{\Sigma}{m_{\rm i} + m_{\rm e}} \frac{2 \pi \nu}{K_2 (1/ \theta_{\rm e})} I^{\prime}\left( x_{\rm M}\right) d \nu ~.\end{aligned}$$ We assumed that the emission below a critical frequency, $\nu_{\rm c}$, is completely self-absorbed so that the emissivity can be approximated by the blackbody emission from the surface of the disk. Following [@esin96], we estimate $\nu_{\rm c}$ as the frequency at which the synchrotron emission from the region $\varpi \sim \varpi + \Delta \varpi$ is equal to the blackbody emission (in the Rayleigh-Jeans limit) from the upper and lower surfaces of the region. This condition gives the equation $$\begin{aligned} \label{eq:nuc} (2 \pi \varpi \Delta \varpi) E_{\rm sy} d \nu = 2 (2 \pi \varpi \Delta \varpi) 2 \pi \frac{\nu_{\rm c}^2}{c^2} k T_{\rm e} d \nu ~ .\end{aligned}$$ We obtain $\nu_{\rm c}$ by solving this equation numerically. Integrating over frequency, we obtain $$\begin{aligned} Q_{\rm sy}^{-} &=& \displaystyle 2 \int_{0}^{\nu_{\rm c}} 2 \pi \frac{\nu_{\rm c}^2}{c^2} k T_{\rm e} d \nu + \int_{\nu_{\rm c}}^{\infty} E_{\rm sy} d \nu \nonumber \\ &=& \frac{4 \pi \nu_{\rm c}^3 k T_{\rm e}}{3 c^2} + \frac{2 e^2 }{\sqrt{3} c} \frac{\Sigma}{\left(m_{\rm i} + m_{\rm e}\right)} \frac{1}{K_2 (1/ \theta_{\rm e}) a_1^{1/6}} \nonumber \\ &\times& \left[ \frac{1}{a_4^{11/2}} \Gamma \left(\frac{11}{2}, a_4 \nu_{\rm c}^{1/3}\right) + \frac{a_2}{a_4^{19/4}} \Gamma \left(\frac{19}{4}, a_4 \nu_{\rm c}^{1/3}\right) \right. \nonumber \\ &+& \frac{a_3}{a_4^4}(a_4^3 \nu_{\rm c} + 3 a_4^2 \nu_{\rm c}^{2/3} \nonumber \\ &+& \left. 6 a_4 \nu_{\rm c}^{1/3} + 6) \exp(-a_4 \nu_{\rm c}^{1/3}) \right], \end{aligned}$$ where the parameters $a_1$, $a_2$, $a_3$, and $a_4$ are defined as $$\begin{aligned} a_1 = \frac{2}{3 \nu_{\rm b} \theta_{\rm e}^2} ,~ a_2 = \frac{0.4}{a_1^{1/4}} ,~ a_3 = \frac{0.5316}{a_1^{1/2}} ,~ a_4 = 1.8899 a_1^{1/3} ,\end{aligned}$$ and $\Gamma (a,x)$ is the incomplete gamma function $$\begin{aligned} \Gamma (a,x) = \int_{x}^{\infty} t^{a-1} \exp(-t) dt.\end{aligned}$$ We adopted the prescription for the Compton energy enhancement factor $\eta$ described by [@derm91], which is defined to be the average change in energy of a photon between injection and escape: $$\begin{aligned} \eta & = & 1 + \frac{P(A-1)}{(1 - PA)} \left[1- \left( \frac{x}{3 \theta_{\rm e}} \right)^{-1 -\ln P / \ln A}\right] \nonumber \\ & \equiv & 1 + \eta_1 -\eta_2 \left( \frac{x}{\theta_{\rm e}}\right)^{\eta_3} ~,\end{aligned}$$ where $$\begin{aligned} x & = & \frac{ h \nu}{m_{\rm e} c^2}~ , ~~~ \tau_{\rm es} = \frac{\kappa_{\rm es} \Sigma}{2}~, \nonumber \\ P & = & 1 - \exp (-\tau_{\rm es})~ , ~~~ A = 1 + 4 \theta_{\rm e} + 16 \theta_{\rm e}^2 ~, \nonumber \\ \eta_1 & = & \frac{P (A-1)}{1-PA}~ , ~~~ \eta_2 = 3^{-\eta_3} \eta_1 , \nonumber \\ \eta_3 & = &-1 - \ln P / \ln A ~.\end{aligned}$$ Here, $P$ is the probability that an escaping photon is scattered, $A$ is the mean amplification factor in the energy of a scattered photon when the scattering electrons have a Maxwellian velocity distribution of temperature $\theta_{\rm e}$. Following [@nara95], the vertically integrated Compton cooling by bremsstrahlung and synchrotron photons are given respectively by $$\begin{aligned} Q_{\rm br,C}^{-} = 3 \eta_1 Q_{\rm br}^{-} \times \nonumber \\ \left\{ \left( \frac{1}{3} - \frac{x_c}{3 \theta_{\rm e}}\right) - \frac{1}{\eta_3 + 1}\left[ \left( \frac{1}{3}\right)^{\eta_3 + 1} - \left( \frac{x_c}{3 \theta_{\rm e}}\right)^{\eta_3 + 1} \right]\right\} ~,\end{aligned}$$ $$\begin{aligned} Q_{\rm sy,C}^{-} = Q_{\rm sy}^{-} \left[ \eta_1 - \eta_2\left(\frac{x_c}{\theta_{\rm e}}\right)^{\eta_3}\right] ~, \end{aligned}$$ where $x_c = h \nu_{\rm c} / m_{\rm e} c^2$. ### Heat Advection Term The vertically integrated heat advection terms for ions and electrons are expressed as $$\begin{aligned} \label{eq:qadvi} Q_{\rm ad,i} = \frac{\dot M}{2\pi \varpi^2} \frac{k T_{\rm i}}{m_{\rm i} + m_{\rm e}} \xi_{\rm i}~,\end{aligned}$$ $$\begin{aligned} \label{eq:qadve} Q_{\rm ad,e} = \frac{\dot M}{2\pi \varpi^2} \frac{k T_{\rm e}}{m_{\rm i} + m_{\rm e}} \xi_{\rm e}~,\end{aligned}$$ where $$\begin{aligned} \label{eq:xi_i} \xi_{\rm i} = - a_{\rm i} \frac{\partial \ln T_{\rm i}}{\partial \ln \varpi} + \frac{\partial \ln \Sigma}{\partial \ln \varpi} - \frac{\partial \ln H}{\partial \ln \varpi} ~,\end{aligned}$$ $$\begin{aligned} \label{eq:xi_e} \xi_{\rm e} = - a_{\rm e} (T_{\rm e}) \left[ 1 + \frac{\partial \ln a_{\rm e} (T_{\rm e})}{\partial \ln T_{\rm e}}\right] \frac{\partial \ln T_{\rm e}}{\partial \ln \varpi} \nonumber \\ + \frac{\partial \ln \Sigma}{\partial \ln \varpi} - \frac{\partial \ln H}{\partial \ln \varpi}~, \end{aligned}$$ are dimensionless quantities of the order of unity (hereafter we call them the entropy gradient parameter), $a_{\rm i} = 1/(1-\gamma_{\rm i}) = 3/2$, $a_{\rm e}(T_{\rm e}) = 1/(1-\gamma_{\rm e} (T_{\rm e}))$, respectively. The positivity (negativity) of the entropy gradient parameter means that the heat advection term works as an effective cooling (heating). The entropy gradient parameter for ions, $\xi_{\rm i}$, has a positive value in ADAF/RIAF solutions, while it has a negative value in LHAF solutions. The global three-dimensional MHD simulations [e.g., @mach06] and steady, vertically integrated, one-dimensional transonic solutions [e.g., @naka96; @naka97; @oda07] of optically thin black hole accretion flows indicate that $\xi_{\rm i} \sim 1$. Following these results, we adopt $\xi_{\rm i} = 1$ as a fiducial value in this paper. We note that a value of $\xi_{\rm i}$ is not important in SLE and low-$\beta$ solutions because the heat advection term is negligible compared to the other terms. According to [@naka97], $\xi_{\rm e}$ can have a positive and negative value ($-0.5 \lesssim \xi_{\rm e} \lesssim 0.5$). However, they assumed that no viscous heat goes into electrons ($\delta_{\rm heat} = 0$). In the energy equation for electrons, when $\delta_{\rm heat} \gtrsim 0.1$, the dissipated magnetic energy term is typically greater than or comparable to the heat advection term in the inner region of the disk. Therefore the exact value we choose for this parameter is not so important, in particular, for SLE and low-$\beta$ solutions in which $\delta_{\rm heat} Q^{+}, Q^{\rm ie}, Q_{\rm rad}^{-} \gg Q_{\rm ad,e}$. We find that the results presented in this paper are practically unchanged for any value of $\xi_{\rm e}$ between $-0.5$ and $0.5$ when $\delta_{\rm heat} \gtrsim 0.1$. We show the results for $\xi_{\rm e} = 0.5$ in most part of this paper. Results ======= We solved the above basic equations at $\varpi = 5 r_{\rm s}$ for given parameters $\dot M$, $\alpha$, $\zeta$, $\delta_{\rm heat}$, $\xi_{\rm i}$, and $\xi_{\rm e}$. We obtained new thermal equilibrium solutions, low-$\beta$ solutions, in addition to the ADAF/RIAF (for positive $\xi_{\rm i}$), SLE, LHAF (for negative $\xi_{\rm i}$) solutions in the optically thin regime. Low-$\beta$ solutions {#results_main} --------------------- Figure \[al05ds3xp05sifoo\] shows the sequences of each thermal equilibrium solution in the $\Sigma$ versus $\dot M / {\dot M}_{\rm Edd}$, $T_{\rm i}$(thin line), $T_{\rm e}$(thick line), and $\beta$ plane. The disk parameters we adopted are $\alpha = 0.05$, $\xi_{\rm i} = 1$, $\xi_{\rm e} = 0.5$, $\delta_{\rm heat} = (1+3\sqrt{T_{\rm i}/T_{\rm e}})^{-1}$, $\zeta = 0.6$ (solid), $0.3$ (dashed), and $0$ (dotted), respectively. Here ${\dot M}_{\rm Edd} = L_{\rm Edd} / \eta_{\rm e} c^2 = 4 \pi GM / \left( \eta_{\rm e} \kappa_{\rm es} c \right)$ is the Eddington mass accretion rate, $\eta_{\rm e} = 0.1$ is the energy conversion efficiency, and $\kappa_{\rm es} = 0.40~ {\rm cm}^{2}~{\rm g}^{-1}$ is the electron scattering opacity. We obtain three types of solutions, ADAF/RIAF (for $\Sigma \lesssim 1 ~ {\rm g} ~ {\rm cm}^{-2}$ at this radius), SLE (for $\Sigma \sim 1 ~ {\rm g} ~ {\rm cm}^{-2}$), and low-$\beta$ solutions (for $\Sigma \gtrsim 1 ~ {\rm g}~ {\rm cm}^{-2}$). We find that the low-$\beta$ solutions exist above the maximum mass accretion rate of the ADAF/RIAF solutions, ${\dot M}_{\rm c, A} \sim 0.003 {\dot M}_{\rm Edd}$. This indicates that the disk initially staying in the ADAF/RIAF state undergoes transition to the low-$\beta$ state when the mass accretion rate exceeds ${\dot M}_{\rm c, A}$. Furthermore, the electron temperature in the low-$\beta$ solutions is lower ($T_{\rm e} \sim 10^{8}-10^{9.5} {\rm K}$) than that in the ADAF/RIAF solutions. The energy balance for ions and electrons is illustrated in figure \[al05ds3xp05sieie\]. The upper panel shows the ratio of the heat advection to the magnetic heating for ions, $Q_{\rm ad,i}/(1-\delta_{\rm heat})Q^{+}$ [this quantity is referred to as the advection factor $f$, e.g., @nara94; @nara95; @abra95; @yuan01]. The lower panel shows the ratio of the heat advection to the total heating for electrons (thin line) and the fraction of the energy transfer via Coulomb collisions to the total heating (thick line). The electrons receive the total amount of the magnetic heating in the low-$\beta$ solutions as well as in the SLE solutions even though we introduced the parameter $\delta_{\rm heat}$ which represents the fraction of heating to electrons. The fraction $(1-\delta_{\rm heat})$ of the magnetic heating goes into ions and the fraction $\delta_{\rm heat}$ of the magnetic heating goes into electrons. However, almost all the magnetic heating going into ions is transferred to electrons via Coulomb collisions in the low-$\beta$ solutions ($(1-\delta_{\rm heat})Q^{+} \sim Q^{\rm ie}$). Eventually, the total amount of the magnetic heating goes into electrons. This means that the parameter $\delta_{\rm heat}$ does not appear practically in the energy balance for electrons. The radiative cooling overwhelms the heat advection in the low-$\beta$ solutions. Therefore, the magnetic heating balances the radiative cooling in the low-$\beta$ solutions ($Q^{+} \sim Q_{\rm rad}^{-}$). The energy balance is essentially the same in both the SLE solutions and the low-$\beta$ solutions ($Q^{+} \sim Q_{\rm rad}^{-}$). The difference is which pressure dominates the magnetic heating. The magnetic heating, which is proportional to the total pressure in our model, is dominated by the gas pressure in the SLE solutions and the magnetic pressure in the low-$\beta$ solutions. Next we describe the main cooling mechanism in each solution. Figure \[al05ds3xp05sirr\] shows the vertically integrated bremsstrahlung (solid), bremsstrahlung-Compton (dotted), synchrotron (dashed), and synchrotron-Compton (dash-dotted) cooling rate for $\zeta = 0$ (bottom), $0.3$ (middle), and $0.6$ (top). When $\zeta = 0$, the synchrotron-Compton cooling is dominant in the low-$\beta$ solutions for lower mass accretion rate while the bremsstrahlung-Compton cooling is dominant for high mass accretion rates ($\dot M \gtrsim 10^{-3} {\dot M}_{\rm Edd}$, $\Sigma \gtrsim 1.6 ~ {\rm g} ~ {\rm cm}^{-2}$) even though $\beta < 1$. The synchrotron cooling is relatively ineffective for higher mass accretion rates because of the lower electron temperature. As $\zeta$ increases, the electron temperature become high because the large magnetic flux enhances not only the synchrotron cooling but also the magnetic heating. As a result, the synchrotron and synchrotron-Compton cooling become efficient. When $\zeta = 0.6$, the synchrotron-Compton is dominant in whole low-$\beta$ solutions. Now we show that $\dot M \propto \Sigma$, $T \propto \Sigma^{-2}$, $\beta \propto \Sigma^{-2}$ on the low-$\beta$ branch [see also @oda09], where $T = (T_{\rm i} + T_{\rm e})/2$ is the mean temperature. These relations depend on the dependence of ${\dot \Phi}_{\rm out}$ on $\dot M$ (our outer boundary condition leads ${\dot \Phi} \propto {\dot M}^{1/2}$). Here we introduce the parameter $s$, ${\dot \Phi}_{\rm out} \propto {\dot M}^{s}$, in order to leave this dependence explicitly. First, we derive the relations between $\dot M$ and $\Sigma$. Since $W_{\rm tot} \sim W_{\rm mag} \propto T \Sigma \beta^{-1}$ on the low-$\beta$ branch, equations (\[eq:h2\]), (\[eq:mom\_phi\_int\]), and (\[eq:al\_int\]) yield $H \propto T^{1/2} \beta^{-1/2}$ and $\dot M \propto T \Sigma \beta^{-1}$. Using equations (\[eq:con\_int\]), (\[eq:ind\_int\_ignore\]), and (\[eq:phidot\]), we find that $H \propto \Sigma^{-(1-2s)/(7-4s)}$, $\beta \propto \Sigma ^{2(1-2s)/(7-4s)} T$, and $\dot M \propto \Sigma^{1- 2(1-2s)/(7-4s)}$. Next, we derive the relation between $T$ and $\Sigma$ from the energy balance of the low-$\beta$ solutions ($Q^{+} \sim Q_{\rm rad}^{-}$). Here we roughly approximate the radiative cooling rate by $Q_{\rm rad}^{-} \propto \Sigma^{2} T^{1/2} H^{-1}$ for simplicity (this is the same dependence as non-relativistic bremsstrahlung cooling for single temperature plasma). Since $Q^{+} \propto W_{\rm tot} \propto \dot M$, we find that $T \propto \Sigma^{-2-6(1-2s)/(7-4s)}$. Therefore $\beta \propto \Sigma ^{2(1-2s)/(7-4s)} T \propto \Sigma^{-2-4(1-2s)/(7-4s)}$. We find that $\dot M \propto \Sigma$, $T \propto \Sigma^{-2}$, and $\beta \propto \Sigma^{-2}$ when $s=1/2$. We investigate the relation between the mass accretion rate and the electron temperature in order to explain the anti-correlation between the luminosity and the cutoff energy observed in the bright/hard state. In the low-$\beta$ solutions, we expect that the luminosity is roughly proportional to the mass accretion rate since $Q_{\rm rad}^{-} \sim Q^{+} \propto \dot M$, and that the electron temperature roughly represents the cutoff energy since the inverse-Compton scattering is the dominant radiative cooling mechanism. Therefore, the relation between the electron temperature and the mass accretion rate is useful for comparison with the observational data. Figure \[al05ds3xp05temd\] shows the relations between the electron temperature and the mass accretion rate for the same parameters as in Figure \[al05ds3xp05sifoo\]. We find that the electron temperature is typically $\lesssim 10^{9.5} ~ {\rm K}$ in the low-$\beta$ solutions while it is $\gtrsim 10^{9.5} ~ {\rm K}$ in the ADAF/RIAF solutions. Furthermore, the electron temperature in the low-$\beta$ solutions strongly anti-correlates with the mass accretion rate. Since $\dot M \propto \Sigma^{1- 2(1-2s)/(7-4s))}$ and $T \propto \Sigma^{-2-6(1-2s)/(7-4s)}$ in the low-$\beta$ solutions, we find that $T \propto {\dot M}^{-2 -2(1-2s)} \propto {\dot M}^{-2}$ for $s=1/2$. According to [@pess05], the MRI is stabilized for $v_{\rm A} \gtrsim \sqrt{c_{\rm s} v_{\rm K0}}$ (we plot this critical point (filled circle in Figure \[al05ds3xp05temd\]) at which $v_{\rm A} = \sqrt{c_{\rm s} v_{\rm K0}}$). Therefore, the low-$\beta$ solutions may not exist under the condition that $v_{\rm A} \gtrsim \sqrt{c_{\rm s} v_{\rm K0}}$. We discuss this issue in Section \[discussion\]. We also show the results for $\alpha = 0.2$ in Figure \[al20ds3xp05sifoo\] and \[al20ds3xp05temd\]. The maximum mass accretion rate of the ADAF/RIAF solution is around ${\dot M}_{\rm c, A} \sim 0.05 {\dot M}_{\rm Edd}$. This maximum mass accretion rate is higher than that for $\alpha = 0.05$. Here we investigate the dependence of ${\dot M}_{\rm c, A}$ on $\alpha$. Since $W_{\rm tot} \sim W_{\rm gas}$, $Q_{\rm ad,i}/(1-\delta_{\rm heat}) Q^{+} \sim 0.5$ and $Q_{\rm ad,e} \ll \delta_{\rm heat} Q^{+} + Q^{\rm ie}$ at $\dot M \sim {\dot M}_{\rm c, A}$, the energy equations yield (i) $Q_{\rm ad,i} \sim 0.5 (1-\delta_{\rm heat}) Q^{+}$, (ii) $Q^{\rm ie} \sim 0.5 (1-\delta_{\rm heat}) Q^{+}$, and (iii) $\delta_{\rm heat} Q^{+} + Q^{\rm ie} \sim [\delta_{\rm heat} + 0.5 (1-\delta_{\rm heat})] Q^{+} \sim Q_{\rm rad}^{-}$. Here we assume $\delta_{\rm heat}$ to be constant for simplicity. We find from Equation (i) that $T_{\rm i} \sim 10^{11} {\rm K}$. If $Q_{\rm rad}^{-}$ can be written in a simple form of $\Sigma^2/H \times f(T_{\rm e})$ like $Q_{\rm br}^{-}$, we find from Equations (ii) and (iii) that $T_{\rm e}$ at ${\dot M}_{\rm c,A}$ is independent of $\alpha$ and ${\dot M}_{\rm c, A}$ is proportional to $\alpha^2 {\dot M}_{\rm Edd}$ exactly. Although $Q_{\rm rad}^{-}$ has a complicated form in our model, our numerical results indicate that the $T_{\rm e}$ is roughly independent of $\alpha$ ($T_{\rm e} \sim 10^{9.5} {\rm K}$). Using this value of $T_{\rm e}$ instead of solving Equation (iii), we find from Equation (ii) that roughly ${\dot M}_{\rm c,A} \sim \alpha^2 {\dot M}_{\rm Edd}$. This result is consistent with ${\dot M}_{\rm c,A} \sim 1.3 √\alpha^{2} {\dot M}_{\rm Edd}$ by [@esin97]. Effect of $\delta_{\rm heat}$ ----------------------------- We investigated the dependence of the results on $\delta_{\rm heat}$ because $\delta_{\rm heat}$ is a poorly constrained parameter. We considered three cases: $\delta_{\rm heat} = 0.5$ as an example of model in which ions and electrons receive equal amounts of the dissipated magnetic energy, $\delta_{\rm heat} = 10^{-3}$ ($\sim m_{\rm e} / m_{\rm i}$) as an example of a conventional model in which ions receive a substantial amount of the energy, and $\delta_{\rm heat} = 0.2$ as an intermediate example. We note that $\delta_{\rm heat} = (1+3\sqrt{T_{\rm i}/T_{\rm e}})^{-1}$ lies between $\sim 0.05$ and $0.25$ in the solutions presented in this paper. Figure \[al05xp05sifoo\] shows the thermal equilibrium curves for the same parameters as in Figure \[al05ds3xp05sifoo\] but for $\delta_{\rm heat} = (1+3\sqrt{T_{\rm i}/T_{\rm e}})^{-1}$ (solid), $0.5$ (dashed), $0.2$ (dotted), and $10^{-3}$ (dot-dashed), respectively. The energy balance for ions and electrons is illustrated in Figure \[al05xp05sieie\]. The ion temperature decreases as $\delta_{\rm heat}$ increases because the magnetic heating for ions ($(1-\delta_{\rm heat}) Q^{+}$) which is the only heating source for ions decreases. ${\dot M}_{\rm c,A}$ also decreases because of the decrease in the magnetic heating for ions. The electron temperature increases as $\delta_{\rm heat}$ increases in the ADAF/RIAF solutions. By contrast, in the low-$\beta$ solutions, the electron temperature is almost independent of the value of $\delta_{\rm heat}$ because $\delta_{\rm heat}$ does not appear practically in the energy equation for electrons ($\delta_{\rm heat} Q^{+} + Q^{\rm ie} \sim Q^{+} \sim Q_{\rm rad}^{-}$). Therefore, we find that the exact value of $\delta_{\rm heat}$ is not so important for the low-$\beta$ solutions. Dependence on the entropy gradient parameters, $\xi_{\rm i}$ and $\xi_{\rm e}$ ------------------------------------------------------------------------------ Figure \[al05ds3xp05z000sifoo\] shows the thermal equilibrium curves for the same parameters as in Figure \[al05ds3xp05sifoo\] but for $\xi_{\rm i} = -1.0$ (dashed). We also plotted the thermal equilibrium curves for $\xi_{\rm i} = 1.0$ (solid) for comparison. We obtained the LHAF solutions in the high mass accretion rate and high surface density region. The heat advection works as a heating for ions and balances the energy transfer via the Coulomb collisions in the LHAF solutions. Above ${\dot M}_{\rm c,A}$, the heat advection term overwhelms the magnetic heating. Hence, the ion temperature becomes higher than that in the ADAF/RIAF solutions. In the LHAF solutions, the electrons receive a larger amount of heat from ions than that in the ADAF/RIAF solutions. Nonetheless, the electron temperature becomes lower than that in the ADAF/RIAF solutions (yet higher than that in the low-$\beta$ solutions) because the radiative cooling becomes more effective in such higher surface density region. We find that our results are practically unchanged for any value of $\xi_{\rm e}$ between $-0.5$ and $0.5$ because the heat advection term is negligible compared to the other terms in energy equation for electrons except when $\delta_{\rm heat} = 10^{-3}$. Furthermore, even when $\delta_{\rm heat} = 10^{-3}$, the other quantities except the electron temperature in the ADAF/RIAF solutions does not change practically. We show the thermal equilibrium curves plotted on the $\Sigma$ - $T_{\rm i}$ (thin) and $T_{\rm e}$ (thick) plane for $\delta_{\rm heat} = 10^{-3}$, $\xi_{\rm e} = 0.5$ (solid), $0$ (dashed), $-0.5$ (dotted), $\zeta = 0.6$ (left panel), and $0$ (right panel) in Figure \[al05dd3sitt\]. Discussion ========== Optically Thin, Magnetically Supported, Moderately Cool Disks {#discussion_main} ------------------------------------------------------------- We obtained thermal equilibrium solutions for an optically thin, two-temperature accretion disk incorporating magnetic fields. We included bremsstrahlung emission, synchrotron emission, and inverse Compton scattering as the radiative cooling mechanisms, and introduced the parameter $\delta_{\rm heat}$ which represents the fraction of the magnetic heating to electrons. We prescribed the $\varpi \varphi$-component of the azimuthally averaged Maxwell stress is proportional to the total pressure. In order to complete the set of basic equations, we specified the radial distribution of the magnetic flux advection rate by introducing a parameter $\zeta$. We found a branch of low-$\beta$ solutions in addition to the usual ADAF/RIAF (for $\xi_{\rm i} > 0$), SLE, and LHAF (for $\xi_{\rm i} < 0$) solutions. Here we remark why we can obtain the low-$\beta$ solutions. First, we prescribed that the $\varpi \varphi$-component of the Maxwell stress is proportional to the total pressure. Therefore, if the magnetic pressure is high, we can obtain the magnetic heating rate balancing the radiative cooling rate even in high surface density and low temperature region. Second, we specified the magnetic flux advection rate, $\dot \Phi$, in order to complete the set of the basic equations. In the conventional theory, $\beta$ is assumed to be constant (typically, $\beta \sim 1$), which means that the magnetic pressure is proportional to the gas pressure. This implies that $p_{\rm gas}$ is just multiplied by a constant ($(1 - \beta^{-1})p_{\rm gas}$). As a result, we cannot obtain the sequence of the low-$\beta$ solutions. On the other hand in our model, a decrease in temperature results in an increase in magnetic pressure (therefore an increase in the magnetic heating) under the conservation of the magnetic flux in the vertical direction. This is the reason why we can obtain the sequence of the low-$\beta$ solutions. We note that the exact values of $\xi_{\rm i}$, $\xi_{\rm e}$, and $\delta_{\rm heat}$ are not so important in the low-$\beta$ solutions. The magnetic heating enhanced by the high magnetic pressure balances the radiative cooling in the low-$\beta$ solutions. The heat advection terms including $\xi_{\rm i}$ and $\xi_{\rm e}$ are negligible for both ions and electrons. Furthermore, $\delta_{\rm heat}$ does not appear in the energy balance for electrons practically. Let us discuss the lower limit of $\beta$. Since $\beta \propto \Sigma^{-2}$ and $\Sigma \propto \dot M$ under our outer boundary condition ($s=1/2$), we find that $\beta \propto {\dot M}^{-2}$, that is, $\beta$ decreases as the mass accretion rate increases in the low-$\beta$ solutions. It has been confirmed by global MHD simulations [e.g., @mach06] that the MRI is not stabilized and hence the magnetic heating rate is expressed in the form of equation (\[eq:qmag\]), at least, when $\beta \gtrsim 0.1$. Local MHD simulations [e.g., @joha08] also indicated that the Maxwell and Reynolds stresses generated by magnetic turbulence are significant and yield an effective $\alpha$-viscosity ($\alpha \sim 0.1$) in highly magnetized disks ($\beta \sim 1$). However, the expression of the magnetic heating employed in our paper may no longer be valid for much lower values of $\beta$ because strong magnetic fields suppress the growth of the MRI. We implicitly assumed that the dissipation energy of the turbulent magnetic fields generated by the MRI is converted into the thermal energy of the disk gas. Therefore, such heating mechanism becomes ineffective if the MRI is stabilized. [@pess05] studied the evolution of the MRI in differentially rotating, magnetized flows beyond the weak-field limit, and showed that the MRI is stabilized for toroidal Alfv[é]{}n speeds, $v_{\rm A} = B_{\varphi} / \sqrt{4 \pi \rho_{0}}$, exceeding the geometric mean of the sound speed, $c_{\rm s}$, and the rotational speed, $v_{\rm K} = \varpi \Omega_{\rm K0}$, (i.e., $v_{\rm A} \gtrsim \sqrt{c_{\rm s} v_{\rm K0}}$, or equivalently $\beta \lesssim 2 c_{\rm s}/v_{\rm K0}$). Our results satisfy this condition when $\beta \lesssim 0.27 ~ (\zeta = 0.6)$, $0.09 ~ (\zeta = 0.3)$, $0.03 ~ (\zeta = 0)$ for $\alpha = 0.05$, and $\beta \lesssim 0.26 ~ (\zeta = 0.6)$, $0.08 ~ (\zeta = 0.3)$, and $0.02 ~ (\zeta = 0)$ for $\alpha = 0.2$, respectively. These critical points are denoted by filled circles in Figure \[al05ds3xp05temd\] and \[al20ds3xp05temd\] (see also Table \[tbl-1\]). When $\beta$ falls below this critical value, the low-$\beta$ solutions may not exist because there is no heating source balancing the radiative cooling. As a result, the disk may undergo a transition to other states (e.g, an MDAF-like disk or an optically thick disk). Thermal Stability ----------------- Let us discuss the thermal stability in this subsection. A general criterion concerning the thermal instability of disks can be expressed as [see @prin76; @kato08] $$\begin{aligned} \label{eq:criterion} \left[\frac{\partial}{\partial T} \left( -Q_{\rm ad} + Q^{+} - Q_{\rm rad}^{-}\right)\right]_{\Sigma} > 0 ~ .\end{aligned}$$ In the low-$\beta$ solutions, we ignore the heat advection term because $Q^{+} \sim Q_{\rm rad}^{-} \gg Q_{\rm ad}$. Once again we approximate the radiative cooling rate by $Q_{\rm rad}^{-} \propto \Sigma^{2} T^{1/2} H^{-1}$ for simplicity. We find from $Q^{+} \propto W_{\rm mag} \propto T \Sigma \beta^{-1} \propto \Sigma^{1-2(1-2s)/(7-4s)}$ and $Q_{\rm rad}^{-} \propto \Sigma^{2} T^{1/2} H^{-1} \propto \Sigma^{2+(1-2s)/(7-4s)} T^{1/2}$ (see Section \[results\_main\]) that the low-$\beta$ solutions do not satisfy the criterion (\[eq:criterion\]), that is, are thermally stable. We note that the thermal stability is independent of $s$. We also remark the thermal stability of the other solutions (ADAF/RIAF, SLE, and LHAF) in which the gas pressure is dominant ($W_{\rm tot} \sim W_{\rm gas} \propto \Sigma T$). Equations (\[eq:h2\]) and (\[eq:con\_int\]) yield $H \propto T^{1/2}$ and $\dot M \propto \Sigma T$. We find from $Q^{+} \propto W_{\rm gas} \propto \Sigma T $, $Q_{\rm rad}^{-} \propto \Sigma^{2} T^{1/2} H^{-1} \propto \Sigma^{2}$, $Q_{\rm ad} \propto \dot M T \xi \propto \Sigma T^{2} \xi$ that the ADAF/RIAF solutions ($Q^{+} \sim Q_{\rm ad}$) are thermally stable but the SLE ($Q^{+} \sim Q_{\rm rad}^{-}$) and LHAF ($Q_{\rm ad} \sim Q_{\rm rad}^{-}$) solutions are thermally unstable. A Candidate for the Bright/Hard State ------------------------------------- The main purpose of this paper is to explain the bright/hard state observed during the bright/slow transition in the rising phases of the transient outbursts of BHCs. In the low/hard state, the X-ray spectrum is described by a hard power law with a high energy cutoff at $\sim 200 ~ {\rm keV}$. When their luminosity exceed $\sim 0.1 L_{\rm Edd}$, these systems undergo a transition from the low/hard state to the bright/hard state. In the bright/hard state, the cutoff energy decreases from $\sim 200 ~ {\rm keV}$ to $\sim 50 ~ {\rm keV}$ as the luminosity increases from $\sim 0.1 L_{\rm Edd}$ to $\sim 0.3 L_{\rm Edd}$ [e.g., @miya08]. Beyond the bright/hard state, these systems undergo a transition to the high/soft state going through the VH/SPL state. The ADAF/RIAF solution explains the X-ray spectrum in the low/hard state because the electron temperature is high ($T_{\rm e} \gtrsim 10^{9.5} {\rm K}$). However, this solution does not exist at the high mass accretion rates and does not show the strong anti-correlation between the electron temperature and the mass accretion rate observed in the bright/hard state. The low-$\beta$ solution extends to such high mass accretion rates. Therefore, the disk initially staying in the ADAF/RIAF state undergoes transition to the low-$\beta$ state when the mass accretion rate exceeds ${\dot M}_{\rm c, A}$. On the low-$\beta$ branches, the electron temperature is low ($T_{\rm e} \sim 10^{8} - 10^{9.5} {\rm K}$ $\sim 10 - 300 ~ {\rm keV}$) and strongly anti-correlates with the mass accretion rate. These features are consistent with the anti-correlation between the luminosity and the energy cutoff observed in the bright/hard state. Therefore, the low-$\beta$ solution can explain the bright/hard state. In Section \[discussion\_main\], we have discussed the lower limit of $\beta$ below which the MRI is stabilized therefore the low-$\beta$ solutions may not exist. Since $\beta \propto {\dot M}^{-2}$ in the low-$\beta$ solutions, this means that the low-$\beta$ solution has a maximum mass accretion rate, ${\dot M}_{{\rm c,} \beta }$. The mass accretion rate and the electron temperature at the lower limit of $\beta$ are depicted by filled circles in Figure \[al05ds3xp05temd\] and \[al20ds3xp05temd\] (see also Table \[tbl-1\]). We found that when $\zeta \gtrsim 0.3$ (or equivalently, $\dot \Phi (\varpi = 5 r_{\rm s}) \gtrsim 4.9 {\dot \Phi}_{\rm out}$) ${\dot M}_{{\rm c,} \beta} > {\dot M}_{\rm c, A}$. This indicates that the disk staying in the ADAF/RIAF state undergoes transition to the low-$\beta$ disk when the mass accretion rate exceeds ${\dot M}_{\rm c, A}$, after that, the disk undergoes transition to the optically thick disk when the mass accretion rate exceeds ${\dot M}_{{\rm c,} \beta}$. In this case, the hard-to-soft transition occurs at ${\dot M}_{{\rm c,} \beta}$ (not at ${\dot M}_{\rm c, A}$). This can correspond to the bright/slow transition. Here we also remark on the dark/fast transition during which the system undergoes transition from the low/hard state to the high/soft state without going through the bright/hard state and the VH/SPL state. We found that when $\zeta \lesssim 0.3$ (or equivalently, $\dot \Phi (\varpi = 5 r_{\rm s}) \lesssim 4.9 {\dot \Phi}_{\rm out}$) ${\dot M}_{{\rm c,} \beta} < {\dot M}_{\rm c, A}$, that is, there is no optically thin, thermally stable solution above ${\dot M}_{\rm c, A}$. Therefore, the disk staying in the ADAF/RIAF state immediately undergoes transition to an optically thick disk without through the low-$\beta$ disk. This can correspond to the dark/fast transition. Accordingly, we conclude that the bright/slow transition occurs when $\dot \Phi$ has a large value and the dark/fast transition occurs when $\dot \Phi$ has a small value. Summary ======= We have obtained the low-$\beta$ solutions for optically thin, two-temperature accretion disks incorporating the mean azimuthal magnetic fields, and concluded that the low-$\beta$ solutions explain the bright/hard state observed during the bright/slow transition of BHCs. We assumed that the energy transfer from ions to electrons occurs through Coulomb collisions, and considered bremsstrahlung emission, synchrotron emission, and inverse Compton scattering as the radiative cooling processes. We prescribed that the Maxwell stress is proportional to the total (gas and magnetic) pressure. In order to complete the set of basic equations, we specified the radial distribution of the magnetic flux advection rate. Accordingly, a decrease in temperature can result in an increase in magnetic pressure under the conservation of the magnetic flux in the vertical direction. In the low-$\beta$ solutions, the magnetic heating is enhanced by the high magnetic pressure. The fraction $(1-\delta_{\rm heat})$ of the magnetic heating goes into ions and is transferred to electrons via Coulomb collisions. The fraction $\delta_{\rm heat}$ of the magnetic heating goes into electrons. Eventually, the total amount of the magnetic heating goes into electrons and balances the radiative cooling (Compton cooling by bremsstrahlung and/or synchrotron photons). The electron temperature is lower than that in the ADAF/RIAF solutions ($T_{\rm e} \sim 10^{8} - 10^{9.5} {\rm K} \sim 10 - 300 ~ {\rm keV}$) and strongly anti-correlates with the mass accretion rate in the low-$\beta$ solutions. These features are consistent with the X-ray spectrum observed in the bright/hard state. According to [@pess05], the MRI is stabilized for $\beta \lesssim 2 c_{\rm s}/v_{\rm K0}$. This indicates that the low-$\beta$ solutions disappear below this critical point. When the magnetic flux advection rate is high ($\zeta \gtrsim 0.3$), the critical mass accretion rate of the low-$\beta$ solutions, ${\dot M}_{{\rm c,} \beta}$, is above the maximum mass accretion rate of the ADAF/RIAF solutions, ${\dot M}_{\rm c, A}$. Therefore, the disk initially staying the ADAF/RIAF state undergoes transition to the low-$\beta$ state when the mass accretion rate exceeds ${\dot M}_{\rm c, A}$, after that, the disk undergoes transition to the optically thick disk when the mass accretion rate exceeds ${\dot M}_{{\rm c,} \beta}$. This corresponds to the bright/slow transition. On the other hand, when the magnetic flux advection rate is low ($\zeta \lesssim 0.3$), ${\dot M}_{{\rm c,} \beta}$ is below ${\dot M}_{\rm c, A}$. Therefore, the disk initially staying the ADAF/RIAF state immediately undergoes transition to the optically thick disk when the mass accretion rate exceeds ${\dot M}_{\rm c, A}$. This can correspond to the dark/fast transition. We are grateful to R. Narayan for valuable discussion and closely examining my draft. This work is supported by the Grant-in-Aid for Science Research of the Ministry of Education, Culture, Sports, Science and Technology (RM: 20340040) and Grant-in-Aid for JSPS Fellows (20.1842). Abramowicz, M. A., Chen, X., Kato, S., Lasota, J.-P., & Regev, O. 1995, , 438, L37 Balbus, S. A., & Hawley, J. F. 1991, , 376, 214 Belloni, T., et al. 2006, , 367, 1113 Dermer, C. D., Liang, E. P., & Canfield, E. 1991, , 369, 410 Eardley, D. M., Lightman, A. P., & Shapiro, S. L. 1975, , 199, L153 Esin, A. A., McClintock, J. E., & Narayan, R. 1997, , 489, 865 Esin, A. A., Narayan, R., Cui, W., Grove, J. E., & Zhang, S.-N. 1998, , 505, 854 Esin, A. A, Narayan, R., Ostriker, E., & Yi, I. 1996, , 465, 312 Fragile, P. C., & Meier, D. L. 2009, , 693, 771 Gierli[ń]{}ski, M., & Newton, J. 2006, , 370, 837 Hawley, J. F. 2000, , 528, 462 Hawley, J. F. & Krolik, J. H. 2001, , 548, 348 Hirose, S., Krolik, J. H., & Stone, J. M. 2006, , 640, 901 Ichimaru, S. 1977, , 214, 840 Johansen, A., & Levin, Y. 2008, , 490, 501 Kato, S., Fukue, J., & Mineshige, S. 2008, Black-Hole Accretion Disks: Towards a New Paradigm (Kyoto: Kyoto Univ. Press) Krolik, J. H., Hirose, S., & Blaes, O. 2007, , 664, 1045 Machida, M., Hayashi, M. R., & Matsumoto, R. 2000, , 532, L67 Machida, M., & Matsumoto, R. 2003, , 585, 429 Machida, M., Nakamura, K. E., & Matsumoto, R. 2004, , 56, 671 Machida, M., Nakamura, K. E., & Matsumoto, R. 2006, , 58, 193 Mahadevan, R., Narayan, R., & Yi, I. 1996, , 456, 327 Manmoto, T., Mineshige, S., & Kusunose, M. 1997, , 489, 791 Meier, D. L. 2005, , 300, 55 Miyakawa, T., Yamaoka, K., Homan, J., Saito, K., Dotani, T., Yoshida, A., & Inoue, H. 2008, , 60, 637 Nakamura, K. E., Kusunose, M., Matsumoto, R., & Kato, S. 1997, , 49, 503 Nakamura, K. E., Matsumoto, R., Kusunose, M., & Kato, S. 1996, , 48, 761 Narayan, R., & Yi, I. 1994, , 428, L13 Narayan, R. & Yi, I. 1995, , 452, 710 Nishikori, H, Machida, M, & Matsumoto, R , 2006, 641, 862 Oda, H., Machida, M., Nakamura, K. E., & Matsumoto, R. 2007, , 59, 457 Oda, H., Machida, M., Nakamura, K. E. & Matsumoto, R. 2009, , 697, 16 Pacholczyk, A. G. 1970, Radio Astrophysics (San Francisco, CA: Freeman) Paczyńsky, B. & Wiita, P. J. 1980, , 88, 23 Parker, E. N. 1966, , 145, 811 Pessah, M. E., & Psaltis, D. 2005, , 628, 879 Pringle, J. E. 1976, , 177, 65 Shakura, N. I., & Sunyaev, R. A. 1973, , 24, 337 Shapiro, S. L., Lightman, A. P., & Eardley, D. M. 1976, , 204, 187 Sharma, P., Quataert, E., Hammett, G. W., & Stone, J. M. 2007, , 667, 714 Shibata, K., Tajima, T., & Matsumoto, R. 1990, , 350, 295 Shibazaki, N., & H[= o]{}shi, R. 1975, Prog. Theor. Phys., 54, 706 Stepney, S., & Guilbert, P. W. 1983, , 204, 1269 Svensson, R. 1982, , 258, 335 Thorne, K. S., & Price, R. H. 1975, , 195, L101 Yuan, F. 2001, , 324, 119 Yuan, F. 2003, , 594, L99 Yuan, F., Quataert, E., & Narayan, R. 2003, , 598, 301 [llccc]{} $0.05$ & $0.6$ & $0.27$ & $1.3 \times 10^{-2}$ & $1.7 \times 10^9$\ & $0.3$ & $0.09$ & $6.4 \times 10^{-3}$ & $1.0 \times 10^9$\ & $0 $ & $0.03$ & $4.5 \times 10^{-3}$ & $1.3 \times 10^8$\ $0.2$ & $0.6$ & $0.26$ & $1.2 \times 10^{-1}$ & $1.1 \times 10^9$\ & $0.3$ & $0.08$ & $4.8 \times 10^{-2}$ & $7.4 \times 10^8$\ & $0$ & $0.02$ & $2.6 \times 10^{-2}$ & $1.3 \times 10^8$\

ArXiv